19. Internationales Stuttgarter Symposium: Automobil- und Motorentechnik [1. Aufl.] 978-3-658-25938-9;978-3-658-25939-6

Table of contents :
Front Matter ....Pages I-XXX
Motion control solutions for automated driving systems at BMW (Alexander Kron, Immanuel Schaffer, Jeffrey Tchai, Karl-Heinz Meitinger, Stefanie Schraufstetter)....Pages 1-13
Is a typical Mercedes-Benz Driving Character still necessary with an increasing number of driver assistance systems? (Stefan Botev, Horst Brauner, Ludger Dragon)....Pages 14-23
Reasons why customers will still buy premium (chassis) in the age of automated driving (Armin Schöpfel)....Pages 24-24
Technical scenarios for the decarbonization of road transport (Stephan Neugebauer)....Pages 25-25
Comparative evaluation of PtX processes for renewable fuel supply (Maximilian Heneka, Wolfgang Köppel)....Pages 26-40
Modelling of real fuels for an effective virtual engine development with focus on alternative fuel (Francesco Cupo, Marco Chiodi, Michael Bargende, Daniel Koch, Georg Wachtmeister, Donatus Wichelhaus)....Pages 41-59
Sustainable drive concepts for future Motorsports (Lea Schwarz, Michael Bargende, Stefan Dreyer, Ulrich Baretzky, Wolfgang Kotauschek, Florian Bach)....Pages 60-78
Automotive clusters in Germany and Baden-Württemberg (Albrecht Fridrich)....Pages 79-89
The role of clusters in supporting French automotive industry’s competitiveness and innovation (Thomas Röhr)....Pages 90-103
The potential of collaborative business model innovation in automotive eco-systems (Georg von der Ropp)....Pages 104-112
Denoxtronic 5.3 – A modular system for applications worldwide (Michael Raff, Erik Weingarten, Manuel Muslija)....Pages 113-127
Modular HD – Exhaust gas treatment system with autarcic thermal management for high urban NOx conversion (Klaus Schrewe, Bernd Maurer, Christoph Menne, Ingo Zirkwa)....Pages 128-141
New experimental insights in AdBlue-spray/wall interaction and its impacts on EGT system design (David Schweigert, Björn Damson, Hartmut Lüders, Carsten Becker, Olaf Deutschmann)....Pages 142-154
Environmental model extension for lane change prediction with neural networks (Martin Krüger, Anne Stockem Novo, Till Nattermann, Manoj Mohamed, Torsten Bertram)....Pages 155-170
Requirements & evaluation of friction information for the integration in vehicle systems (Staiger Sebastian, Nosrat Nezami, Dieter Schramm)....Pages 171-186
Kinetosis in autonomous driving (Carsten Lecon)....Pages 187-195
One-Stop-Test solutions for autonomous driving (Frank Heidemann)....Pages 196-196
Phenomenology and analysis of gas pressures at low-speed pre-ignitions (Christoph Beerens, Rainer Fischer, Christian Trabold)....Pages 197-213
Combustion stability improvement with turbulence control by air injection for a lean-burn SI engine (Takanori Suzuki, Bastian Lehrheuer, Tamara Ottenwälder, Max Mally, Stefan Pischinger)....Pages 214-228
Air intake temperature cooling thanks to pressure wave action and adapted air intake geometry (Vincent Raimbault, Jérôme Migaud, Heinz Bühl, Stéphane Guilain, David Chalet, Michael Bargende)....Pages 229-240
Future e-mobility and the change in system requirements (Lothar Schindele, David Schütz, Gaël Le Hen, Norbert Müller)....Pages 241-251
Active materials for electrical motors – Leverage for reducing costs and increasing performance (Moritz Kilper, Hristian Naumoski, Steffen Henzler)....Pages 252-265
Traction energy saving potentials for electric cars with gear shift (Oliver Zirn, Fabian Schmiel, Matthias Dellermann)....Pages 266-274
Impact of modulation strategies on the power losses of the dc-link capacitor of an inverter (Maximilian Weber, Michèle Hirsch, Helge Sprenger, Thomas Zeltwanger, Hans-Christian Reuss)....Pages 275-289
The six-step mode: Unwanted or rather the ideal voltage modulation method (Thomas Zeltwanger, Helge Sprenger, Mark Damson, Manabendra N. Gupta)....Pages 290-304
Design of a fail-operational powertrain for automated electric vehicles (Tunan Shen, Ahmet Kilic, Jochen Faßnacht, Christian Thulfaut, Hans-Christian Reuss)....Pages 305-318
A scalable approach for future vehicle electrification (Carsten Bünder)....Pages 319-319
Grid integration e-mobility – Developments and challenges (Ursel Willrett)....Pages 320-330
Validation of range estimation for electric vehicles based on recorded real-world driving data (Patrick Petersen, Jacob Langner, Stefan Otten, Eric Sax, Stefan Scheubner, Moritz Vaillant et al.)....Pages 331-344
Smart grids in mobile fleet operations (Dusko Mitrovic, Manuel Klein, Leonardo Uriona, Marius Klein, Kristian Binder, Edward Eichstetter et al.)....Pages 345-359
Infrared-based determination of the type and condition of the road surface (Lakshan Tharmakularajah, Jakob Döring, Karl-Ludwig Krieger)....Pages 360-370
Essential predictive information for high fuel efficiency and local emission free driving with PHEVs (Tobias Schürmann, Daniel Görke, Stefan Schmiedler, Tobias Gödecke, Kai André Böhm, Michael Bargende)....Pages 371-385
Analogy considerations for the design of hybrid drive trains (Michael Auerbach, Oliver Zirn)....Pages 386-396
Hybrid operating strategies in the trade-off between fuel consumption and emissions (Sven Eberts, H.-J. Berner, Michael Bargende)....Pages 397-397
External water management: A predictive challenge (Cameron Tropea, Johannes Feldmann, Daniel Rettenmaier, Patrick M. Seiler, Michael Ade, Daniel Demel)....Pages 398-411
Aerodynamic study on the vehicle rear shape parameters with respect to ground simulation (Chenyi Zhang, Daniel Stoll, Timo Kuthada, Jochen Wiedemann)....Pages 412-426
Development of an SUV reference model for aerodynamic research (Max Tanneberger, Chenyi Zhang, Timo Kuthada, Felix Wittmeier, Jochen Wiedemann, Juliane Nies)....Pages 427-442
Lamborghini Aventador SVJ Aerodynamics: Route to breaking the super sport car’s record (Ugo Riccio, A. Torluccio)....Pages 443-443
Attribute-based development of Advanced Driver Assistance Systems (Bernhard Schick, Florian Fuhr, Manuel Hoefer, Peter E. Pfeffer)....Pages 444-456
ITC – Integrated traction control for sports car applications (Lars König, Frieder Schindele, Andreas Zimmermann)....Pages 457-470
Computation time optimization of a model-based predictive roll stabilization by neuro-fuzzy systems (Philipp Maximilian Sieberg, Markus Schmid, Sebastian Reicherts, Dieter Schramm)....Pages 471-484
Integrated approach for the virtual development of vehicles equipped with brake control systems (Fabian Fontana, Jens Neubeck, Jochen Wiedemann, Ingo Scharfenbaum, Philippe Stegmann, Armin Ohletz et al.)....Pages 485-501
New entry OEM – A global phenomenon (David Ludwig)....Pages 502-507
Consistent application of systems engineering and simulation for cross-domain function integration (Marcus Boumans, Martin Johannaber, Ulrich Schulmeister)....Pages 508-522
Automotive megatrends and their impact on NVH (Georg Eisele, Michael Kauth, Christoph Steffens, Patrick Glusk)....Pages 523-539
Evaluation of the effects of trends on vehicle concepts based on a forecast of travel demand (Peter Lukas Peters, Rainer Demuth, Dieter Schramm)....Pages 540-556
Potentials of modular autonomous vehicles for variable scenarios of public transport (Dennis Wedler, Thomas Vietor)....Pages 557-570
Method for concept design and optimization of twist beam axles (Xiangfan Fang, Kanlun Tan, Jens Olschewski)....Pages 571-586
Technologies for a modular vehicle concept used in passenger and goods transport (Christian Ulrich, Horst E. Friedrich, Jürgen Weimer, Robert Hahn, Gerhard Kopp, Marco Münster)....Pages 587-598
Experimental investigation of Miller cycle combustion technology with water injection (Nils Neumann, Normann Freisinger, Guido Vent, Thomas Seeger)....Pages 599-611
Reduction of cold start emissions with microwave heated catalytic converters (Viola Papetti, Panayotis Dimopoulos Eggenschwiler, Daniel Schreiber)....Pages 612-623
Extended cylinder deactivation strategies to improve CO2 and pollutant emissions for light-duty diesel engine applications (Kai Deppenkemper, Markus Schönen, Bernd Lindemann, Mauro Scassa, Matthew Younkins, Robert Wang)....Pages 624-635
A new method for the objective assessment of ADAS based on multivariate time series classification (Uwe Moser, Nick Harmening, Dieter Schramm)....Pages 636-651
Test methods of visual distraction by interaction with driver information- and assistance systems (Thomas Stottan)....Pages 652-666
Feasibility study on the basis of a prototype for automated vehicle positioning in inductive charging process (Matthias Hisung, Dean Martinovic, Hans-Christian Reuss)....Pages 667-678
Continuous implementation of 0D/1D engine models in the development process of a race car engine (Simon Malcher, Michael Bargende, Stefan Dreyer, Ulrich Baretzky, Hartmut Diel, Wolfgang Kotauschek et al.)....Pages 679-698
Integrated flow model with combustion and emission model for VVT Diesel engine (Qirui Yang, Michael Bargende, Michael Grill)....Pages 699-718
A 1D co-simulation approach for the prediction of pollutant emissions of internal combustion engines (T. Cerri, G. D’Errico, G. Montenegro, A. Onorati, G. Koltsakis, Z. Samaras et al.)....Pages 719-736
Fully integrated design exploration for in-cylinder simulation (Warren Seeley, Simon Fischer)....Pages 737-756
Performance evaluation of an IC-engine with a novel partial admission turbocharger concept (Markus Schatz, Fabian Seeger, Damian M. Vogt, Sergej Koch, Denis Notheis, Uwe Wagner et al.)....Pages 757-771
The future bearing concept of BMTS Technology (Rüdiger Kleinschmidt, Steffen Schmitt, Frieder Stetter, Oliver Kuhne, Simon Nibler, Gunter Winkler)....Pages 772-783
Comparison of a chain driven vs. gear driven valve train in a commercial vehicle (Peter Bachmair, Christoph Biwer, Thomas Saupe, Thomas Fink)....Pages 784-797
Dynamic friction behavior of a gasoline engine in transient operation (Tobias Funk, Holger Ehnis, Reiner Künzel, Michael Bargende)....Pages 798-814
New vehicle concepts for future business model (Horst E. Friedrich, Christian Ulrich, Stephan Schmid)....Pages 815-829
UNICARagil – New architectures for disruptive vehicle concepts (Dan Keilhoff, Dennis Niedballa, Hans-Christian Reuss, Michael Buchholz, Fabian Gies, Klaus Dietmayer et al.)....Pages 830-842
New vehicle concepts for mobile vacation (Rüdiger Freimann, Udo Gillich, Gerhard Gumpoltsberger, Ria Kaiser)....Pages 843-854
RDE-thermal management – From road to rig (Christian Beidl, Johannes Hipp, Günter Hohenberg, Stefan Geneder)....Pages 855-875
Further optimization of NOx emissions under the EU 6d regulation (Michael Krüger, Stefan Bareiss, Andreas Kufferath, Dirk Naber, Daniel Ruff, Herbert Schumacher)....Pages 876-895
Developing GDI engines for minimum particle emissions in RDE test conditions (A. Hirsch, A. Hochnetz, M. Kortschak, E. Winklhofer)....Pages 896-908
Functional architecture and E/E-Architecture – A challenge for the automotive industry (Detlef Zerfowski, Andreas Lock)....Pages 909-920
Integrated avionics architectures (Reinhard Reichel)....Pages 921-921
Dedicated hybrid powertrain (DHP) – Hybrid concept based on a holistic system approach (Joerg Gindele, Manuel Diehl)....Pages 922-922
Experimental investigations on ICE direct start for hybrid powertrains (Thomas Pausch, Guido Vent, Normann Freisinger, Hardy Weymann, Roland Baar)....Pages 923-937
Extended engine-in-the-loop simulation for development of HEV energy management strategies (Bastian Beyfuss, Peter Hofmann, Bernhard Geringer, Philipp Grassl)....Pages 938-954
Using of an electrochemical compressor for hydrogen recirculation in fuel cell vehicles (Wilhelm Wiebe, Sven Schmitz)....Pages 955-963
Development of electric drive concepts for fuel cell vehicles for Germany and China (Katharina Bause, Adrian Braumandl, Alexander Stephan, Qiwen Xiao, Matthias Behrendt)....Pages 964-977
Test cell adaptation from engine to fuel cell development (Henning Münstermann, Jürgen Knust, Jörg Fischer)....Pages 978-987
Safety assessment of autonomous and connected vehicles by a model-based traffic simulation framework (Mustafa Saraoğlu, Andrey Morozov, Klaus Janschek)....Pages 988-1002
Identifying relevant traffic situations based on human decision making (Christoph Sippl, Florian Bock, Bernd Huber, Anatoli Djanatliev, Reinhard German)....Pages 1003-1017
Test in applications with regulatory requirements on the example of WLTP (Jan Daniel Jacob)....Pages 1018-1018
Chances of the digitalization in the test field for operations and product development (Roland Strixner)....Pages 1019-1028
New energies test facilities – Solutions for a sustainable future (Gregor Zemitzsch, Jörg Fischer, Edwin Heimberg, Thomas Ille, Stefan Zemitzsch)....Pages 1029-1036
The new Touareg – Innovision on 4 wheels (Stefan Gies)....Pages 1037-1051
The aerodynamics development of the new Mercedes-Benz GLE (Etienne Pudell)....Pages 1052-1066
Aerodynamics of the new Porsche 911 Carrera (Bernd Jachowski)....Pages 1067-1084
Comparing 48V mild hybrid concepts using a hybrid-simulation-toolkit (Anita Bongards, S. Mohon, D. Semenov, W. Wenzel)....Pages 1085-1100
From virtual to reality – How 48V systems and operating strategies improve Diesel emission (Hannes Wancura, Michael Weißbäck, Iaponaira Silva de Abreu, Tobias Schäfer, Steffen Lange, Bastian Unterberger et al.)....Pages 1101-1116
Safety and security – Basic vulnerabilities and solutions (Hubert B. Keller)....Pages 1117-1117
Tool-based development of efficient automotive multi-core systems (Patrick Friederich, Alexander Zeeb)....Pages 1118-1130
Architecture and independence controller for deep learning in safety critical applications (Ulrich Bodenhausen)....Pages 1131-1142
AI – Challenges in application with bus data in the automotive sector (Alexander Faul, Maria Floruß, Felix Pistorius)....Pages 1143-1153
Use of a criticality metric for assessment of critical traffic situations as part of SePIA (Matthias Lehmann, Maximilian Bäumler, Günther Prokop, Diana Hamelow)....Pages 1154-1167
Simulating surrounding traffic for interactive driving simulators (Michael Behrisch, Danny Behnecke, Jan Wegener, Robert Hilbrich)....Pages 1168-1174
A lifecycle model to support continuous component evolution in embedded automotive systems (Lukas Block, Oliver Riedel, Florian Herrmann)....Pages 1175-1189
Comparing current and future E/EArchitecture trends of commercial vehicles and passenger cars (Tenny Benckendorff, Andreas Lapp, Thomas Oexner, Thomas Thiel)....Pages 1190-1200
Evaluation of competition and virtual rear subframes by means of the data envelopment analysis (Martin Kundla, Thilo Heussner, Xianda Ye, Dieter Schramm)....Pages 1201-1217
Evaluation of the required accuracy of chassis models in the comfort relevant frequency range by intuitive switching of the level of detail in SimulationX (Tom Wiedemann, Claudia Belanger, Felix Kocksch, Kay Büttner)....Pages 1218-1230
Cause and effect chains analysis of rollover behavior with respect to chassis design (Fan Chang, Konrad Krauter, Sebastiaan van Putten, Jan Kubenz, Armin Ohletz, Günther Prokop)....Pages 1231-1243
AC-APU – A hydrogen based A/C-unit for electric vehicles (R. Hegner, C. Weckerle, I. Bürger, H. Dittus, M. Schier, H. E. Friedrich)....Pages 1244-1258
A complete digital engine cooling module catalog for balancing cooling and aerodynamics (Satheesh Kandasamy, C. Chang, T. Yasuda, Y. Yagi, S. Miura)....Pages 1259-1259
Thermal design of portable power tools with combustion engines and electric motors (Silke Kaminski, Gordon Groskopf)....Pages 1260-1276
Lightweight brake rotors with thermally sprayed ceramic coatings as friction surfaces (Rainer Gadow, S. Popa, A. Killinger)....Pages 1277-1277
Lightweight forging initiative III: Forging technology contribution to lightweight design (Hans-Willi Raedt, Thomas Wurm, Alexander Busse)....Pages 1278-1292
Investigation of interactions between fuels and fuel leading components of plug-in-hybrid electric vehicles (Sebastian Feldhoff, Simon Eiden, Jens Staufenbiel, Anja Singer)....Pages 1293-1307
Acoustic transmission loss in turbochargers (Hendrik Ruppert, Marco Günther, Stefan Pischinger)....Pages 1308-1322
Potential of air path variabilities for heavy duty Diesel engines (Marius Betz, Dávid Kovács, Peter Eilts)....Pages 1323-1338
Automated & cloud-based load profile-generation and evaluation of lithium ion batteries (Alexander Kohs, Thomas Freudenmann, Fabian Back, Mohannad El-Haji, Tobias Schilling, Praveen Kumar Kuppusamy)....Pages 1339-1351
Requirements for battery enclosures – Design considerations and practical examples (Jobst Kerspe, Michael Fischer)....Pages 1352-1367
Prediction of the lifetime of urban electric bus traction batteries in the context of the overall system design (Martin Ufert)....Pages 1368-1382
The impact of pass-by noise legislation on the design of exhaust systems (Jan Krüger, Peter Wink, Maike Werner)....Pages 1383-1396
Simulative research on the tire torsional vibration and its vehicle relevant influencing factors (Wenrui Han, Yi Guo, Günther Prokop, Thomas Roscher)....Pages 1397-1411

Citation preview

Proceedings

Michael Bargende · Hans-Christian Reuss Andreas Wagner · Jochen Wiedemann Hrsg.

19. Internationales Stuttgarter Symposium Automobil- und Motorentechnik

Proceedings

Ein stetig steigender Fundus an Informationen ist heute notwendig, um die immer komplexer werdende Technik heutiger Kraftfahrzeuge zu verstehen. Funktionen, Arbeitsweise, Komponenten und Systeme entwickeln sich rasant. In immer schnelleren Zyklen verbreitet sich aktuelles Wissen gerade aus Konferenzen, Tagungen und Symposien in die Fachwelt. Den raschen Zugriff auf diese Informationen bietet diese Reihe Proceedings, die sich zur Aufgabe gestellt hat, das zum Verständnis topaktueller Technik rund um das Automobil erforderliche spezielle Wissen in der Systematik aus Konferenzen und Tagungen zusammen zu stellen und als Buch in Springer.com wie auch elektronisch in Springer Link und Springer Professional bereit zu stellen. Die Reihe wendet sich an Fahrzeug- und Motoreningenieure sowie Studierende, die aktuelles Fachwissen im Zusammenhang mit Fragestellungen ihres Arbeitsfeldes suchen. Professoren und Dozenten an Universitäten und Hochschulen mit Schwerpunkt Kraftfahrzeug- und Motorentechnik finden hier die Zusammenstellung von Veranstaltungen, die sie selber nicht besuchen konnten. Gutachtern, Forschern und Entwicklungsingenieuren in der Automobil- und Zulieferindustrie sowie Dienstleistern können die Proceedings wertvolle Antworten auf topaktuelle Fragen geben. Today, a steadily growing store of information is called for in order to understand the increasingly complex technologies used in modern automobiles. Functions, modes of operation, components and systems are rapidly evolving, while at the same time the latest expertise is disseminated directly from conferences, congresses and symposia to the professional world in ever-faster cycles. This series of proceedings offers rapid access to this information, gathering the specific knowledge needed to keep up with cutting-edge advances in automotive technologies, employing the same systematic approach used at conferences and congresses and presenting it in print (available at Springer.com) and electronic (at Springer Link and Springer Professional) formats. The series addresses the needs of automotive engineers, motor design engineers and students looking for the latest expertise in connection with key questions in their field, while professors and instructors working in the areas of automotive and motor design engineering will also find summaries of industry events they weren’t able to attend. The proceedings also offer valuable answers to the topical questions that concern assessors, researchers and developmental engineers in the automotive and supplier industry, as well as service providers.

Weitere Bände in der Reihe http://www.springer.com/series/13360

Michael Bargende · Hans-Christian Reuss · Andreas Wagner · Jochen Wiedemann (Hrsg.)

19. Internationales Stuttgarter Symposium Automobil- und Motorentechnik

Hrsg. Prof. Dr. Michael Bargende Forschungsinstitut für Kraftfahrwesen und Fahrzeugmotoren Stuttgart – FKFS Stuttgart, Deutschland

Prof. Dr. Hans-Christian Reuss Forschungsinstitut für Kraftfahrwesen und Fahrzeugmotoren Stuttgart – FKFS Stuttgart, Deutschland

Prof. Dr. Andreas Wagner Fahrzeugmotoren Stuttgart – FKFS Forschungsinstitut für Kraftfahrwesen Stuttgart, Deutschland

Prof. Dr. Jochen Wiedemann Forschungsinstitut für Kraftfahrwesen und Fahrzeugmotoren Stuttgart – FKFS Stuttgart, Deutschland

ISSN 2198-7432 ISSN 2198-7440 (electronic) Proceedings ISBN 978-3-658-25938-9 ISBN 978-3-658-25939-6 (eBook) https://doi.org/10.1007/978-3-658-25939-6 Die Deutsche Nationalbibliothek verzeichnet diese Publikation in der Deutschen Nationalbibliografie; detaillierte bibliografische Daten sind im Internet über http://dnb.d-nb.de abrufbar. Springer Vieweg © Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 Das Werk einschließlich aller seiner Teile ist urheberrechtlich geschützt. Jede Verwertung, die nicht ausdrücklich vom Urheberrechtsgesetz zugelassen ist, bedarf der vorherigen Zustimmung des Verlags. Das gilt insbesondere für Vervielfältigungen, Bearbeitungen, Übersetzungen, Mikroverfilmungen und die Einspeicherung und Verarbeitung in elektronischen Systemen. Die Wiedergabe von allgemein beschreibenden Bezeichnungen, Marken, Unternehmensnamen etc. in diesem Werk bedeutet nicht, dass diese frei durch jedermann benutzt werden dürfen. Die Berechtigung zur Benutzung unterliegt, auch ohne gesonderten Hinweis hierzu, den Regeln des Markenrechts. Die Rechte des jeweiligen Zeicheninhabers sind zu beachten. Der Verlag, die Autoren und die Herausgeber gehen davon aus, dass die Angaben und Informationen in diesem Werk zum Zeitpunkt der Veröffentlichung vollständig und korrekt sind. Weder der Verlag, noch die Autoren oder die Herausgeber übernehmen, ausdrücklich oder implizit, Gewähr für den Inhalt des Werkes, etwaige Fehler oder Äußerungen. Der Verlag bleibt im Hinblick auf geografische Zuordnungen und Gebietsbezeichnungen in veröffentlichten Karten und Institutionsadressen neutral. Verantwortlich im Verlag: Markus Braun Springer Vieweg ist ein Imprint der eingetragenen Gesellschaft Springer Fachmedien Wiesbaden GmbH und ist ein Teil von Springer Nature Die Anschrift der Gesellschaft ist: Abraham-Lincoln-Str. 46, 65189 Wiesbaden, Germany

WELCOME IN STUTTGART  In March 2019, the Research Institute of Automotive Engineering and Vehicle Engines Stuttgart (FKFS) invites you to the 19th Stuttgart International Symposium »Automotive and Engine Technology«. I heartily welcome all participants to two fascinating congress days, dealing with nothing less than the future of our mobility. I am therefore delighted to be the patron of this event once again this year. FKFS has been demonstrating for decades that it is a reliable partner in many areas. The local automotive industry profits from a highly-regarded research institute in close proximity. Thanks to the cooperation with Stuttgart University, students are able to gain access to the regional automotive sector. As a hotbed of specialists and experts, FKFS provides a valuable source of young engineers who help keep Baden-Wuerttemberg one of the leading automotive locations in the world. The state government holds FKFS in high esteem as a long-term cooperation partner. This is underlined, for example, by the wind tunnels the State of Baden-Wuerttemberg set up at Stuttgart University, which FKFS still operates for the University today. In the »Baden-Wuerttemberg Automotive Industry Strategic Dialog« (which was also set up by the state government), FKFS has a permanent role in collaborative work on a concept for the future. The clock is ticking for new mobility solutions. Climate targets are looming ever closer and it is only possible to fulfill them with new, cleaner drive technologies. This is a huge challenge. However, we must not regard it as a burden, but as an opportunity we must take in order to actively shape the mobility of the future. Because those shaping the technology will remain the global leaders in it. Therefore, it is one of the state government’s key tasks to guide the car manufacturing state Baden-Wuerttemberg into a future which is both environmentally-friendly and successful. To achieve this, interlinking science, politics, business and civic society is essential. With this year’s Symposium motto, »The transformation of the automotive industry«, FKFS is tackling the central challenges for the future of the automotive industry headon. At the same time, it forms an important platform to bring various stakeholders together and enter intensive specialist discussions. I thank FKFS and all the participating companies for organizing and implementing the Symposium. I wish the participants interesting lectures, exciting discussions and new inspiration.

Winfried Kretschmann Prime Minister of the State of Baden-Wuerttemberg 

 

V

A WARM WELCOME  The transformation of the automotive industry The automotive industry is currently undergoing enormous and rapid changes. Demands upon research and development are constantly increasing, and new fields of research are arising as well. Both manufacturers and suppliers are developing new business models in order to remain competitive. The European Environment Ministers’ target to reduce vehicle CO2 emissions by a total of 35 per cent by 2030 (compared with the year 2020) highlights the necessity and urgency of a technological shift. Most discussions regarding potential solutions concern the electrification of drives, but the use of renewable fuels (or synthetic fuels) is also on the agenda. More than a year ago, the State of Baden-Wuerttemberg set up the »Automotive Industry Strategic Dialog«. This alliance of politics, business, science, trade unions, consumer organizations, environmental organizations and civic society is pursuing a holistic approach, tapping into innovation potential beyond traditional industry sector limits and aiming to drive the automotive industry in Baden-Wuerttemberg forward successfully. Drives and emissions, autonomous driving and networking, disruptive vehicle architectures, driving dynamics control systems and thermal management: the automotive industry has plenty to talk about. Experts from science and industry will be discussing these topics and many more at the 19th Stuttgart International Symposium »Automotive and Engine Technology« from 19 – 20 March 2019 at the Haus der Wirtschaft. The latest technology, future concepts and new research results will be presented in sessions running in six parallel strands, with more than a hundred lectures. There will be a podium discussion on the topic »The transformation of the automotive industry«, with prominent participants. We have also been able to attract very highprofile speakers for our plenary lectures. There will be many opportunities for exchanging ideas; not only during the breaks and at the trade exhibition, but at the special evening event too. We look forward to welcoming you to Stuttgart – the birthplace of the automobile – and wish you two highly-interesting days at the 19th Stuttgart International Symposium! Prof. Dr. Michael Bargende Prof. Dr. Hans-Christian Reuss Prof. Dr. Jochen Wiedemann

VI

INDEX – Volume 1 SECTION 1

AUTOMATED DRIVING VS. ATTRIBUTE BRANDING? Chairperson: Prof. Dr. Jochen Wiedemann Motion control solutions for automated driving systems at BMW Alexander Kron, I. Schaffer, J. Tchai, K.-H. Meitinger, S. Schraufstetter, BMW Group

1

Is a typical Mercedes-Benz Driving Character still necessary with an increasing number of driver assistance systems? Ludger Dragon, S. Botev, H. Brauner, Daimler AG

15

Reasons why customers will still buy premium (chassis) in the age of automated driving Armin Schöpfel, AUDI AG

25

POWERTRAIN CONCEPTS AND FUELS Chairperson: Prof. Dr. Michael Bargende Technical scenarios for the decarbonization of road transport Stephan Neugebauer, Europäische Technologieplattform für den Straßenverkehr (ERTRAC)/BMW Group

27

Comparative evaluation of PtX processes for renewable fuel supply Wolfgang Köppel, M. Heneka, DVGW-Forschungsstelle am EBI des KIT

29

Modelling of real fuels for an effective virtual engine development with focus on alternative fuels Marco Chiodi, F. Cupo, M. Bargende, FKFS; D. Koch, G. Wachtmeister, TU München; D. Wichelhaus, Volkswagen AG

45

Sustainable drive concepts for future motorsports Lea Schwarz, Universität Stuttgart, M. Bargende, FKFS; S. Dreyer, U. Baretzky, W. Kotauschek, F. Bach, AUDI AG

65

VII

INDEX – Volume 1 AUTOMOTIVE INDUSTRY – CLUSTERS AND BUSINESS MODELS Chairperson: Prof. Dr. Jochen Wiedemann Automotive clusters in Germany and Baden-Wuerttemberg Albrecht Fridrich, automotive-bw c./o. RKW Baden-Württemberg GmbH

85

The role of clusters in supporting french automotive industry’s competitiveness and innovation Thomas Röhr, Pôle Véhicule du Futur/ESTA School of Business and Technology

97

The potential of collaborative business model innovation in automotive eco-systems Georg von der Ropp, BMI Lab Deutschland GmbH

111

EMISSION (DENOX) Chairperson: Prof. Dr. Georg Wachtmeister Denoxtronic 5.3 – A modular system for applications worldwide Michael Raff, E. Weingarten, M. Muslija, Robert Bosch GmbH

121

Modular HD – Exhaust gas treatment system with autarcic thermal management for high urban NOx conversion Klaus Schrewe, B. Maurer, C. Menne, I. Zirkwa, HJS Emission Technology GmbH & Co. KG

137

New experimental insights in AdBlue-spray/wall interaction and its impacts on EGT system design David Schweigert, B. Damson, H. Lüders, C. Becker, Robert Bosch GmbH; O. Deutschmann, Karlsruher Institut für Technologie (KIT)

151

VIII

INDEX – Volume 1 AUTONOMOUS DRIVING I Chairperson: Prof. Dr. Dr. Michael Weyrich Environmental model extension for lane change prediction with neural networks Martin Krüger, A. Stockem Novo, T. Nattermann, M. Mohamed, ZF Group; T. Bertram, TU Dortmund

165

Requirements & evaluation of friction information for the integration in vehicle systems Sebastian Staiger, S. Nosrat Nezami, Dr. Ing. h.c. F. Porsche AG; D. Schramm, Universität Duisburg-Essen

181

Kinetosis at autonomous driving Carsten Lecon, Hochschule Aalen

197

One-Stop-Test solutions for autonomous driving Frank Heidemann, SET GmbH

207

SI-ENGINES Chairperson: Prof. Dr. Hermann Rottengruber Phenomenology and analysis of gas pressures at low-speed pre-ignitions Christoph Beerens, R. Fischer, C. Trabold, MAHLE GmbH

209

Combustion stability improvement with turbulence control by air injection for a lean-burn SI engine Takanori Suzuki, SOKEN Inc.; B. Lehrheuer, T. Ottenwälder, M. Mally, S. Pischinger, RWTH Aachen University

227

Air intake temperature cooling thanks to pressure wave action and adapted air intake geometry Vincent Raimbault, J. Migaud, MANN+HUMMEL France S.A.S.; B. Heinz, MANN+HUMMEL GmbH; S. Guilain, RENAULT Group; D. Chalet, École Centrale de Nantes LHEEA/TSM; M. Bargende, FKFS/IVK, Universität Stuttgart

243

IX

INDEX – Volume 1 SECTION 2

ELECTRIC MOBILITY I Chairperson: Prof. Dr. Nejila Parspour Future e-mobility and the change in system requirements Lothar Schindele, D. Schütz, G. Le Hen, N. Müller, Robert Bosch GmbH

255

Active materials for electrical motors – Leverage for reducing costs and increasing performance Moritz Kilper, H. Naumoski, S. Henzler, Daimler AG

267

Traction energy saving potentials for electric cars with gear shift Oliver Zirn, F. Schmiel, Hochschule Esslingen; M. Dellermann, Daimler AG

281

ELECTRIC MOBILITY II Chairperson: Franz Loogen Impact of modulation strategies on the power losses of the dc-link capacitor of an inverter Maximilian Weber, M. Hirsch, H. Sprenger, T. Zeltwanger, Robert Bosch GmbH; H.-C. Reuss, IVK, Universität Stuttgart

291

The six-step mode: Unwanted or rather the ideal voltage modulation method Thomas Zeltwanger, H. Sprenger, M. Damson, M. Gupta, Robert Bosch GmbH

307

Design of a fail-operational powertrain for automated electric vehicles Tunan Shen, A. Kilic, J. Faßnacht, Robert Bosch GmbH; H.-C. Reuss, IVK, Universität Stuttgart

323

A scalable approach for future vehicle electrification Carsten Bünder, Magna Powertrain

337

X

INDEX – Volume 1 ELECTRIC MOBILITY III Chairperson: Prof. Karl-Ernst Noreikat Grid integration e-mobility – Developments and challenges Ursel Willrett, IAV GmbH

339

Validation of range estimation for electric vehicles based on recorded real-world driving data Patrick Petersen, J. Langner, S. Otten, E. Sax, FZI Forschungszentrum Informatik; S. Scheubner, M. Vaillant, S. Fünfgeld, Dr. Ing. h.c. F. Porsche AG

351

Smart grids in mobile fleet operations Dusko Mitrovic, M. Klein, L. Urionia, M. Klein, K. Binder, E. Eichstetter, EKU Power Drives GmbH; M. Weyrich, IAS, Universität Stuttgart

365

Infrared-based determination of the type and condition of the road surface Lakshan Tharmakularajah, J. Döring, K.-L. Krieger, ITEM, Universität Bremen

381

HYBRID CONCEPTS Chairperson: Prof. Dr. Christian Beidl Essential predictive information for high fuel efficiency and local emission free driving with PHEVs Tobias Schürmann, D. Görke, S. Schmiedler, Daimler AG; T. Gödecke, K. Böhm, Hochschule Esslingen; M. Bargende, IVK, Universität Stuttgart

393

Analogy considerations for the design of hybrid drive trains Michael Auerbach, O. Zirn, Hochschule Esslingen

409

Hybrid operating strategies in the trade-off between fuel consumption and emissions Sven Eberts, H.-J. Berner, FKFS; M. Bargende, FKFS/IVK, Universität Stuttgart

421

XI

INDEX – Volume 1 AERODYNAMICS Chairperson: Prof. Dr. Lennart Löfdahl External water management: A predictive challenge Cameron Tropea, J. Feldmann, D. Rettenmaier, P. M. Seiler, TU Darmstadt; M. Ade, Daimler AG; D. Demel, BMW AG

423

Aerodynamic study on the vehicle rear shape parameters with respect to ground simulation Chenyi Zhang, IVK, Universität Stuttgart; D. Stoll, T. Kuthada, J. Wiedemann, FKFS

437

Development of an SUV reference model for aerodynamic research Max Tanneberger, C. Zhang, IVK, Universität Stuttgart, T. Kuthada, F. Wittmeier, J. Wiedemann, FKFS; J. Nies, Röchling Automotive SE & Co. KG

453

Lamborghini Aventador SVJ Aerodynamics: Route to breaking the super sport car’s record Ugo Riccio, A. Torluccio, Automobili Lamborghini S.p.A.

469

VEHICLE DYNAMICS – CONTROL SYSTEMS Chairperson: Prof. Dr. Oliver Sawodny Attribute-based development of Advanced Driver Assistance Systems Bernhard Schick, Hochschule Kempten; F. Fuhr, M. Hoefer, Dr. Ing. h.c. F. Porsche AG; P. Pfeffer, MdynamiX AG

471

ITC – Integrated traction control for sports car applications Lars König, F. Schindele, A. Zimmermann, Bosch Engineering GmbH

485

Computation time optimization of a model-based predictive roll stabilization by neuro-fuzzy systems Philipp Maximilian Sieberg, M. Schmid, S. Reicherts, D. Schramm, Universität Duisburg-Essen

499

Integrated approach for the virtual development of vehicles equipped with brake control systems Fabian Fontana, J. Neubeck, J. Wiedemann, FKFS/IVK, Universität Stuttgart; I. Scharfenbaum, P. Stegmann, A. Ohletz, U. Schaaf, AUDI AG

513

XII

INDEX – Volume 1 SECTION 3

AUTOMOTIVE TRENDS Chairperson: Prof. Dr. Hermann Winner New entry OEM – A global phenomenon David Ludwig, MAGNA STEYR Fahrzeugtechnik AG & Co. KG

531

Consistent application of systems engineering and simulation for cross-domain function integration Marcus Boumans, M. Johannaber, U. Schulmeister, Robert Bosch GmbH

537

Automotive megatrends and their impact on NVH Georg Eisele, M. Kauth, C. Steffens, FEV Europe GmbH; P. Glusk, FEV Consulting GmbH

553

VEHICLE CONCEPTS Chairperson: Prof. Dr. Thomas Maier Evaluation of the effects of trends on vehicle concepts based on a forecast of travel demand Peter Lukas Peters, R. Demuth, BMW Group; D. Schramm, Universität Duisburg-Essen

571

Potentials of modular autonomous vehicles for variable scenarios of public transport Dennis Wedler, T. Vietor, IK, TU Braunschweig

589

Method for concept design and optimization of twist beam axles Xiangfan Fang, K. Tan, J. Olschewski, Universität Siegen

603

Technologies for a modular vehicle concept used in passenger and goods transport Christian Ulrich, H. E. Friedrich, J. Weimer, R. Hahn, G. Kopp, M. Münster, Deutsches Zentrum für Luft- und Raumfahrt (DLR)

619

XIII

INDEX – Volume 1 WATER INJECTION AND EMISSION CONCEPTS Chairperson: Prof. Dr. Peter Eilts Experimental investigation of Miller cycle combustion technology with water injection Nils Neumann, N. Freisinger, G. Vent, Daimler AG; T. Seeger, Universität Siegen

631

Reduction of cold start emissions with microwave heated catalytic converters Viola Papetti, P. Dimopoulos Eggenschwiler, D. Schreiber, Empa

645

Extended cylinder deactivation strategies to improve CO2 and pollutant emissions for light-duty diesel engine applications Kai Deppenkemper, M. Schönen, B. Lindemann, FEV Europe GmbH; M. Scassa, FEV Italia s.r.l.; M. Younkins, R. Wang, Tula Technology Inc.

657

ADVANCED DRIVER ASSISTANCE SYSTEMS Chairperson: Prof. Dr. Klaus Dietmayer A new method for the objective assessment of ADAS based on multivariate time series classification Uwe Moser, N. Harmening, BMW Group; D. Schramm, Universität Duisburg-Essen

669

Test methods of visual distraction by interaction with driver information- and assistance systems Thomas Stottan, AUDIO MOBIL

685

Feasibility study on the basis of a prototype for automated vehicle positioning in inductive charging process Matthias Hisung; D. Martinovic, H.-C. Reuss, IVK, Universität Stuttgart

701

XIV

INDEX – Volume 1 SIMULATION COMBUSTION ENGINES Chairperson: Prof. Dr. Bernhard Geringer Continuous implementation of 0D/1D engine models in the development process of a race car engine Simon Malcher, Universität Stuttgart; M. Bargende, FKFS; S. Dreyer, U. Baretzky, H. Diel, W. Kotauschek, S. Wohlgemuth, AUDI AG

713

Integrated flow model with combustion and emission model for VVT Diesel engines Qirui Yang, M. Bargende, M. Grill, FKFS/IVK, Universität Stuttgart

733

A 1D co-simulation approach for the prediction of pollutant emissions of internal combustion engines Gianluca Montenegro, T. Cerri, G. D'Errico, A. Onorati, Politecnico Di Milano; G. Koltsakis, Z. Samaras, Aristotle University of Thessaloniki; V. Tziolas, N. Zingopis, K. Michos, Exothermia SA; J. Rojewski, V. Papetti, P. Dimopoulos Eggenschwiler, P. Soltic, Empa

753

Fully integrated design exploration for in-cylinder simulation Warren Seeley, S. Fischer, Siemens PLM

771

ENGINE MECHANICS AND CHARGING Chairperson: Prof. Dr. Wolfgang Thiemann Performance evaluation of an IC-engine with a novel partial admission turbocharger concept Markus Schatz, F. Seeger, D. M. Vogt, ITSM, Universität Stuttgart; S. Koch, D. Notheis, U. Wagner, T. Koch, Karlsruher Institut für Technologie (KIT)

791

The future bearing concept of BMTS Technology Rüdiger Kleinschmidt, S. Schmitt, F. Stetter, O. Kuhne, S. Nibler, G. Winkler, BMTS Technology GmbH & Co. KG

807

Comparison of a chain driven vs. gear driven valve train in a commercial vehicle Peter Bachmair, iwis motorsysteme GmbH & Co. KG; C. Biwer, T. Saupe, FEV Europe GmbH; T. Fink, Joh. Winklhofer Beteiligungs GmbH & Co. KG

819

Dynamic friction behavior of a gasoline engine in transient operation Tobias Funk, H. Ehnis, R. Künzel, MAHLE International GmbH; M. Bargende, IVK, Universität Stuttgart

833

XV

INDEX – Volume 2 SECTION 1

THE NEW WAY OF DRIVING Chairperson: Prof. Dr. Hans-Christian Reuss New vehicle concepts for future business models Horst E. Friedrich, C. Ulrich, S. Schmid, Deutsches Zentrum für Luft- und Raumfahrt (DLR)

1

UNICARagil – New architectures for disruptive vehicle concepts Dan Keilhoff, D. Niedballa, Universität Stuttgart; M. Buchholz, F. Gies, Universität Ulm; M. Lauer, Karlsruher Institut für Technologie (KIT); S. Ackermann, TU Darmstadt; A. Kampmann, B. Alrifaee, F. Klein, M. Struth, T. Woopen, RWTH Aachen University

17

New vehicle concepts for mobile vacation Rüdiger Freimann, Erwin Hymer Group SE; U. Gillich, G. Gumpoltsberger, ZF Friedrichshafen AG; R. Kaiser, TTT – The Team Technology

31

EMISSION (RDE) Chairperson: Prof. Dr. Thomas Koch RDE-thermal management – From road to rig Christian Beidl, J. Hipp, TU Darmstadt; G. Hohenberg, S. Geneder, IVD Prof. Hohenberg GmbH

43

Further optimization of NOx emissions under the EU 6d regulation Michael Krüger, S. Bareiss, A. Kufferath, D. Naber, D. Ruff, H. Schumacher, Robert Bosch GmbH

65

Developing GDI engines for minimum particle emissions in RDE test conditions Ernst Winklhofer, A. Hirsch, A. Hochnetz, M. Kortschak, AVL List GmbH

85

XVII

INDEX – Volume 2 SYSTEM ARCHITECTURE Chairperson: Prof. Dr. Dr. Gerhard Hettich Functional architecture and E/E-Architecture – A challenge for the automotive industry Detlef Zerfowski, ETAS GmbH; A. Lock, Robert Bosch GmbH Integrated avionics architectures Reinhard Reichel, ILS, Universität Stuttgart

99

111

HYBRID POWERTRAIN Chairperson: Prof. Dr. Günter Hohenberg Dedicated hybrid powertrain (DHP) – Hybrid concept based on a holistic system approach Jörg Gindele, M. Diehl, Magna Powertrain

113

Experimental investigations on ICE direct start for hybrid powertrains Thomas Pausch, G. Vent, N. Freisinger, H. Weymann, Daimler AG; R. Baar (posthumously), TU Berlin

115

Extended engine-in-the-loop simulation for development of HEV energy management strategies Bastian Beyfuss, P. Hofmann, B. Geringer, IFA, TU Wien; P. Grassl, Sohatex Engineering GmbH & Co KG

131

XVIII

INDEX – Volume 2 FUEL CELL Chairperson: Prof. Dr. Stefan Pischinger Using of an electrochemical compressor for hydrogen recirculation in fuel cell vehicles Wilhelm Wiebe, S. Schmitz, DHBW Mannheim

149

Development of electric drive concepts for fuel cell vehicles for Germany and China Katharina Bause, A. Braumandl, A. Stephan, Q. Xiao, M. Behrendt, IPEK, Karlsruher Institut für Technologie (KIT)

159

Test cell adaptation from engine to fuel cell development Henning Münstermann, J. Knust, J. Fischer, SBI Schreiber, Brand und Partner Ingenieurgesellschaft mbH

173

AUTONOMOUS DRIVING II Chairperson: Prof. Dr. Clemens Gühmann Safety assessment of autonomous and connected vehicles by a model-based traffic simulation framework Mustafa Saraoğlu, A. Morozov, K. Janschek, IfA, TU Dresden

183

Identifying relevant traffic situations based on human decision making Christoph Sippl, F. Bock, B. Huber, AUDI AG; A. Djanatliev, R. German, Friedrich-Alexander Universität Erlangen-Nürnberg

199

XIX

INDEX – Volume 2 SECTION 2

TEST BENCH TECHNOLOGY Chairperson: Prof. Dr. Karl-Ludwig Haken Test in applications with regulatory requirements on the example of WLTP Jan Daniel Jacob, Werum Software & Systems AG

215

Chances of the digitalization in the test field for operations and product development Roland Strixner, Kratzer Automation AG

217

New energies test facilities – Solutions for a sustainable future Gregor Zemitzsch, J. Fischer, E. Heimberg, T. Ille, S. Zemitzsch, SBI Schreiber, Brand und Partner Ingenieurgesellschaft mbH

227

NEW VEHICLES Chairperson: Prof. Dr. Thomas Vietor The new Touareg – Innovision on 4 wheels Stefan Gies, Volkswagen AG

235

The aerodynamics development of the new Mercedes-Benz GLE Etienne Pudell, Daimler AG

251

Aerodynamics of the new Porsche 911 Carrera Bernd Jachowski, Dr. Ing. h.c. F. Porsche AG

267

XX

INDEX – Volume 2 48 VOLT HYBRID Chairperson: Prof. Dr. Helmut Eichlseder Comparing 48V mild hybrid concepts using a hybrid-simulation-toolkit Anita Bongards, S. Mohon, D. Semenov, W. Wenzel, BorgWarner Inc.

285

From virtual to reality – How 48V systems and operating strategies improve Diesel emission Hannes Wancura, M. Weißbäck, AVL List GmbH; I. Silva de Abreu, T. Schäfer, AVL Schrick GmbH; S. Lange, AVL GmbH; B. Unterberger, S. Hoffmann, Hyundai Motor Europe Technical Center GmbH

301

SOFTWARE AND DEVELOPMENT METHODS I Chairperson: Prof. Dr. Eric Sax Safety and security – Basic vulnerabilities and solutions Hubert B. Keller, Karlsruher Institut für Technologie (KIT)

317

Tool-based development of efficient automotive multi-core systems Patrick Friederich, A. Zeeb, Vector Informatik GmbH

319

Architecture and independence controller for deep learning in safety critical applications Ulrich Bodenhausen, AI Coaching und Vector Consulting Services GmbH

333

XXI

INDEX – Volume 2 SOFTWARE AND DEVELOPMENT METHODS II Chairperson: Prof. Dr. Tobias Gerhard Flämig-Vetter AI – Challenges in application with bus data in the automotive sector Alexander Faul, M. Floruß, Vector Informatik GmbH; F. Pistorius, Karlsruher Institut für Technologie (KIT)

345

Use of a criticality metric for assessment of critical traffic situations as part of SePIA Matthias Lehmann, M. Bäumler, G. Prokop, D. Hamelow, TU Dresden

357

Simulating surrounding traffic for interactive driving simulators Michael Behrisch, D. Behnecke, J. Wegener, R. Hilbrich, Deutsches Zentrum für Luft- und Raumfahrt (DLR)

371

NETWORKING AND ARCHITECTURE Chairperson: Prof. Dr. Karl-Ludwig Krieger A lifecycle model to support continuous component evolution in embedded automotive systems Lukas Block, Universität Stuttgart; O. Riedel, F. Herrmann, Fraunhofer IAO

379

Comparing current and future E/E-Architecture trends of commercial vehicles and passenger cars Tenny Benckendorff, Bosch Engineering GmbH; A. Lapp, T. Oexner, T. Thiel, Robert Bosch GmbH

395

XXII

INDEX – Volume 2 SECTION 3

CHASSIS Chairperson: Prof. Dr. Xiangfan Fang Evaluation of competition and virtual rear subframes by means of the data envelopment analysis Martin Kundla, T. Heussner, X. Ye, BMW Group; D. Schramm, Universität Duisburg-Essen

407

Evaluation of the required accuracy of chassis models in the comfort relevant frequency range by intuitive switching of the level of detail in SimulationX Tom Wiedemann, C. Belanger, ESI ITI GmbH; F. Kocksch, K. Büttner, IAD, TU Dresden

425

Cause and effect chains analysis of rollover behavior with respect to chassis design Fan Chang, K. Krauter, J. Kubenz, G. Prokop, IAD, TU Dresden; S. van Putten, A. Ohletz, AUDI AG

439

THERMAL MANAGEMENT Chairperson: Prof. Dr. Stefan Böttinger AC-APU – A hydrogen based A/C-unit for electric vehicles Robert Hegner, C. Weckerle, I. Bürger, H. Dittus, M. Schier, H. E. Friedrich, Deutsches Zentrum für Luft- und Raumfahrt (DLR)

453

A complete digital engine cooling module catalog for balancing cooling and aerodynamics Satheesh Kandasamy, C. Chang, Dassault Systemes; T. Yasuda, Y. Yagi, S. Miura, Denso Corporation

469

Thermal design of portable power tools with combustion engines and electric motors Silke Kaminski, G. Groskopf, ANDREAS STIHL AG & Co. KG

471

XXIII

INDEX – Volume 2 LIGHTWEIGHT DESIGN Chairperson: Prof. Dr. Horst E. Friedrich Lightweight brake rotors with thermally sprayed ceramic coatings as friction surfaces Rainer Gadow, S. Popa, A. Killinger, Universität Stuttgart

489

Lightweight forging initiative III: Forging technology contribution to lightweight design Hans-Willi Raedt, Hirschvogel Automotive Group; T. Wurm, Georgsmarienhuette GmbH; A. Busse, RWTH Aachen University

491

REPORTS FROM FVV PROJECTS Chairperson: Dr. Karl Kollmann Investigation of interactions between fuels and fuel leading components of plug-in-hybrid electric vehicles Sebastian Feldhoff, S. Eiden, Oel-Waerme-Institut gGmbH (OWI); J. Staufenbiel, A. Singer, TAC, Hochschule Coburg

507

Acoustic transmission loss in turbochargers Hendrik Ruppert, M. Günther, S. Pischinger, RWTH Aachen University

523

Potential of air path variabilities for heavy duty Diesel engines Peter Eilts, D. Kovács, M. Betz, TU Braunschweig

539

XXIV

INDEX – Volume 2 BATTERIES Chairperson: Prof. Dr. Andreas Friedrich Automated & cloud-based load profile-generation and evaluation of lithium ion batteries Alexander Kohs, F. Back, T. Schilling, P. Kuppusamy, CTC cartech company GmbH; T. Freudenmann, M. El-Haji, EDI GmbH

555

Requirements for battery enclosures – Design considerations and practical examples Michael Fischer, GVI® by KÖNIG METALL GmbH & Co. KG; J. Kerspe, TEB Dr. Kerspe

569

Prediction of the lifetime of urban electric bus traction batteries in the context of the overall system design Martin Ufert, TU Dresden

585

NVH Chairperson: Prof. Dr. Frank Gauterin The impact of pass-by noise legislation on the design of exhaust systems Jan Krüger, P. Wink, M. Werner, Eberspächer Exhaust Technology GmbH & Co. KG

601

Simulative research on the tire torsional vibration and its vehicle relevant influencing factors Wenrui Han, Y. Guo, G. Prokop, IAD, TU Dresden; T. Roscher, AUDI AG

615

XXV

SPEAKERS, CHAIRPERSONS Prof. Dr. Michael Auerbach Hochschule Esslingen

Fan Chang TU Dresden

Peter Bachmair iwis motorsysteme GmbH & Co. KG

Dr. Marco Chiodi FKFS

Prof. Dr. Michael Bargende FKFS/IVK, Universität Stuttgart

Kai Deppenkemper FEV Europe GmbH

Katharina Bause Karlsruher Institut für Technologie (KIT)

Prof. Dr. Klaus Dietmayer Universität Ulm

Dr. Christoph Beerens MAHLE GmbH

Prof. Dr. Ludger Dragon Daimler AG

Dr. Michael Behrisch Deutsches Zentrum für Luft- und Raumfahrt (DLR)

Sven Eberts FKFS

Prof. Dr. Christian Beidl TU Darmstadt Tenny Benckendorff Bosch Engineering GmbH Bastian Beyfuss TU Wien Lukas Block Universität Stuttgart Dr. Ulrich Bodenhausen Ulrich Bodenhausen AI Coaching and Vector Consulting Services GmbH

Prof. Dr. Helmut Eichlseder TU Graz Prof. Peter Eilts TU Braunschweig Dr. Georg Eisele FEV Europe GmbH Univ.-Prof. Dr. Xiangfan Fang Universität Siegen Alexander Faul Vector Informatik GmbH Sebastian Feldhoff OWI Oel-Waerme-Institut gGmbH

Anita Bongards BorgWarner Inc.

Michael Fischer GVI® by KÖNIG METALL GmbH & Co. KG

Prof. Dr. Stefan Böttinger Universität Hohenheim

Prof. Dr. Tobias Gerhard Flämig-Vetter DHBW Stuttgart

Marcus Boumans Robert Bosch GmbH

Fabian Fontana FKFS/IVK, Universität Stuttgart

Dr. Carsten Bünder Magna Powertrain

Dr. Rüdiger Freimann Erwin Hymer Group SE

XXVI

SPEAKERS, CHAIRPERSONS

Dr. Albrecht Fridrich automotive-bw c./o. RKW Baden-Württemberg GmbH Patrick Friederich Vector Informatik GmbH Prof. Dr. Andreas Friedrich Deutsches Zentrum für Luft- und Raumfahrt (DLR) Prof. Dr. Horst E. Friedrich Deutsches Zentrum für Luft- und Raumfahrt (DLR)

Prof. Dr. Dr. Gerhard Hettich EAST Consulting Matthias Hisung Universität Stuttgart Dr. Nicole Hoffmeister-Kraut MdL Ministerium für Wirtschaft, Arbeit und Wohnungsbau Baden-Württemberg Prof. em. Dr. Günter Hohenberg IVD Prof. Hohenberg GmbH Bernd Jachowski Dr. Ing. h.c. F. Porsche AG

Tobias Funk MAHLE International GmbH

Dr. Jan Daniel Jacob Werum Software & Systems AG

Prof. Dr. Rainer Gadow Universität Stuttgart

Dr. Silke Kaminski ANDREAS STIHL AG & Co. KG

Prof. Dr. Frank Gauterin Karlsruher Institut für Technologie (KIT)

Satheesh Kandasamy Dassault Systemes

Prof. Dr. Bernhard Geringer TU Wien

Dr. Dan Keilhoff Universität Stuttgart

Prof. Stefan Gies Volkswagen AG

Dr. Hubert B. Keller Karlsruher Institut für Technologie (KIT)

Dr. Jörg Gindele Magna Powertrain

Gerald Killmann Toyota Motor Europe

Prof. Dr. Clemens Gühmann TU Berlin

Moritz Kilper Daimler AG

Prof. Dr. Karl-Ludwig Haken Hochschule Esslingen

Rüdiger Kleinschmidt BMTS Technology GmbH & Co. KG

Wenrui Han TU Dresden

Prof. Dr. Thomas Koch Karlsruher Institut für Technologie (KIT)

Robert Hegner Deutsches Zentrum für Luft- und Raumfahrt (DLR)

Alexander Kohs CTC cartech company GmbH

Frank Heidemann SET GmbH

Dr. Karl Kollmann

XXVII

SPEAKERS, CHAIRPERSONS

Dr. Lars König Bosch Engineering GmbH

Dr. Frank Mastiaux EnBW Energie Baden-Württemberg AG

Wolfgang Köppel DVGW-Forschungsstelle am EBI des KIT

Dusko Mitrovic EKU Power Drives GmbH

Prof. Dr. Karl-Ludwig Krieger Universität Bremen

Gianluca Montenegro Politecnico di Milano

Dr. Alexander Kron BMW Group

Uwe Moser BMW Group

Dr. Jan Krüger Eberspächer Exhaust Technology GmbH & Co. KG

Henning Münstermann SBI Schreiber, Brand und Partner Ingenieurgesellschaft mbH

Martin Krüger ZF Group

Dr. Stephan Neugebauer Europäische Technologieplattform für den Straßenverkehr (ERTRAC)/BMW Group

Michael Krüger Robert Bosch GmbH Martin Kundla BMW Group Prof. Dr. Carsten Lecon Hochschule Aalen Matthias Lehmann TU Dresden Prof. Dr. Lennart Löfdahl Chalmers University of Technology Franz Loogen e-mobil BW GmbH David Ludwig MAGNA STEYR Fahrzeugtechnik AG & Co. KG

Nils Neumann Daimler AG Prof. Karl Ernst Noreikat Viola Papetti Empa Prof. Dr. Nejila Parspour Universität Stuttgart Thomas Pausch Daimler AG Peter Lukas Peters BMW Group Patrick Petersen FZI Forschungszentrum Informatik

Prof. Dr. Thomas Maier Universität Stuttgart

Dr. Mathias Pillin Robert Bosch GmbH

Simon Malcher Universität Stuttgart

Prof. Dr. Stefan Pischinger RWTH Aachen University

Georges Massing Daimler AG

Etienne Pudell Daimler AG

XXVIII

SPEAKERS, CHAIRPERSONS

Dr. Hans-Willi Raedt Hirschvogel Automotive Group

Dr. Lothar Schindele Robert Bosch GmbH

Michael Raff Robert Bosch GmbH

Dr. Armin Schöpfel AUDI AG

Vincent Raimbault MANN+HUMMEL France S.A.S.

Klaus Schrewe HJS Emission Technology GmbH & Co. KG

Reinhard Reichel Universität Stuttgart Prof. Dr. Dr. Wolfram Ressel Universität Stuttgart Prof. Dr. Hans-Christian Reuss FKFS/IVK, Universität Stuttgart Ugo Riccio Automobili Lamborghini S.p.A. Thomas Röhr Pôle Véhicule du Futur/ESTA School of Business and Technology

Prof. Günther Schuh e.GO Mobile AG Tobias Schürmann Daimler AG Lea Schwarz Universität Stuttgart David Schweigert Robert Bosch GmbH Dr. Warren Seeley Siemens PLM

Prof. Dr. Hermann Rottengruber OvGU Magdeburg

Tunan Shen Robert Bosch GmbH

Hendrik Ruppert RWTH Aachen University

Philipp Maximilian Sieberg Universität Duisburg-Essen

Mustafa Saraoğlu TU Dresden

Christoph Sippl AUDI AG

Prof. Dr. Dr. Oliver Sawodny Universität Stuttgart

Sebastian Staiger Dr. Ing. h.c. F. Porsche AG

Prof. Dr. Eric Sax Karlsruher Institut für Technologie (KIT)

Thomas Stottan AUDIO MOBIL

Dr. Markus Schatz Universität Stuttgart

Roland Strixner Kratzer Automation AG

Wolf-Henning Scheider ZF Friedrichshafen AG

Takanori Suzuki SOKEN Inc.

Prof. Bernhard Schick Hochschule Kempten

Max Tanneberger Universität Stuttgart

XXIX

SPEAKERS, CHAIRPERSONS

Lakshan Tharmakularajah Universität Bremen

Prof. Dr. Jochen Wiedemann FKFS/IVK, Universität Stuttgart

Prof. Dr. Wolfgang Thiemann Helmut-Schmidt-Universität

Tom Wiedemann ESI ITI GmbH

Prof. Dr. Cameron Tropea TU Darmstadt

Ursel Willrett IAV GmbH

Martin Ufert TU Dresden

Dr. Ernst Winklhofer AVL List GmbH

Christian Ulrich Deutsches Zentrum für Luft- und Raumfahrt (DLR)

Prof. Dr. Hermann Winner TU Darmstadt

Prof. Dr. Thomas Vietor TU Braunschweig Georg von der Ropp BMI Lab Deutschland GmbH Prof. Dr. Georg Wachtmeister TU München

Johannes Winterhagen Redaktionsbüro delta eta Qirui Yang Universität Stuttgart Thomas Zeltwanger Robert Bosch GmbH

Hannes Wancura AVL List GmbH

Gregor Zemitzsch SBI Schreiber, Brand und Partner Ingenieurgesellschaft mbH

Maximilian Weber Robert Bosch GmbH

Dr. Detlef Zerfowski ETAS GmbH

Dennis Wedler TU Braunschweig

Chenyi Zhang Universität Stuttgart

Prof. Dr. Dr. Michael Weyrich Universtität Stuttgart

Prof. Dr. Oliver Zirn Hochschule Esslingen

Wilhelm Wiebe DHBW Mannheim  

XXX

Motion control solutions for automated driving systems at BMW Dr. Alexander Kron, Immanuel Schaffer, Jeffrey Tchai, Dr. Karl-Heinz Meitinger, Dr. Stefanie Schraufstetter BMW Group

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 M. Bargende et al. (Hrsg.), 19. Internationales Stuttgarter Symposium, Proceedings, https://doi.org/10.1007/978-3-658-25939-6_1

1

Motion control solutions for automated driving systems at BMW

1 Autonomous Driving to Shape the Future of Mobility by BMW Aside from technological trends, such as electrification and digitalization, automotive development is currently dominated by maturing the level of Autonomous Driving. Latest BMW models are already supporting the driver with an extensive portfolio of advanced driving assistance systems (ADAS) in the areas of driving comfort, active safety and parking [1, 2]. The maturity of these systems, and their level of automation can mainly be categorized as level-1 (driver assistance) and level-2 (partial automation) [3]. Partial automation, as depicted in Fig. 1, temporarily enables the driver to have hands and eyes off, however he or she is still responsible for the driving supervision task. The transition from level-2 to level-3 automation increases the system complexity extensively, as the driver is not “actively” required to monitor the environment, so called “hands and eyes off”. In this stage, he or she still requires to be ready to retake control at all times if notified by the vehicle. Level-4 functionality allows an even higher grade of automation in which the driver can drive the vehicle “mind and hands off”. In the last stage, level-5 (full automation), the vehicle is capable of performing all driving tasks under all environmental conditions without involving the driver at all.

Figure 1: System complexity in dependency of the level of automation.

The BMW Group has more than 10 years of experience in the development of advanced driver assistance systems [2], currently developing level-3 automated driving systems

2

Motion control solutions for automated driving systems at BMW in close cooperation with several partners. Managing the increasing system complexity requires a strong understanding of the overall system architecture. This paper describes the BMW approach to level-3 autonomous driving, which divides total functionality of the complete system in a chain of subsystems. In the following section, the system architecture is described, followed by proposing elected solutions for motion control as a subsystem for vehicle guidance for an automated driving system.

2 Architectural Overview of an Automated Driving System Fig. 2 depicts a simplified functional architecture of an automated driving system. The overall system is separated into subsystems which are observing, interpreting and acting. The observing is performed by several sensor components, such as LiDAR, radar, ultrasonic sensors, camera and surround view, which are all required for 360° degree environmental perception. These sensors are integrated on-board and detect environmental information.

Figure 2: System architecture of an automated driving system.

The next subsystem represents the environmental interpretation as part of on-board computer vision. Object and environmental data gathered from on-board sensors are fused with cloud data in the form of a high-definition map with several attributes, for instance the road friction layer. An overall environmental model is generated as a basis

3

Motion control solutions for automated driving systems at BMW for the driving strategy of automated driving. An essential sub chain is the calculation of trajectories. It masters the complex task of computing an optimal, safe and comfortable trajectory guiding the car on a collision-free local path increment to the global destination The calculated trajectory is used as input for the motion control sub chain, involving the central vehicle management as interface layer to different type of actuators, e.g. internal combustion engines, electrified engines, brake and steering systems. Moreover, motion control covers the measurement of vehicle motion denoted as “EgoMotion”. The actuators of the chassis, the engine and the HMI are again regarded to be autonomic elements of the total system. Development of such a complex coupled system requires efficient and structured collaboration within different departments. The BMW approach considers the overall complex system as a set of autonomous sub chains with precisely defined interfaces, enabling the decrease of system complexity and allowing for shared development between several departments. In the following sections, solutions are presented, providing a deeper understanding regarding the development of motion control as sub chain responsible for vehicle guidance.

3 Motion Control Approach for Autonomous Driving The BMW approach further subdivides on-board computer vision into two elements. Firstly, accountable for the driving strategy by answering the question “where” the vehicle is to be guided to. Secondly, motion control related to the vehicle guidance, by responding to the question “how” the car autonomously drives with regard to varying road conditions. Fig. 3 depicts the motion control architecture which comprises two submodules: “Central Vehicle Management” (CVM) and “EgoMotion”. CVM is responsible to compute the actuator inputs for engine, brake and steering, whereas “EgoMotion” provides the system with precise information about the vehicle states related to driving dynamics. As illustrated in Fig. 3 the functional architecture of CVM encompasses the coordination of the trajectory input, the subdivision of vehicle guidance into a lateral and longitudinal part and finally the coordination of computed output for the multi actuator setup. More detailed information to CVM is presented in Sec. 3.1.

4

Motion control solutions for automated driving systems at BMW

Figure 3: Motion control as a sub chain for vehicle guidance.

“EgoMotion” provides the overall system with vehicle states using post processed data from 6-DOF inertial acceleration sensors. In addition, an odometry module delivers precise information about the vehicle orientation and the change in position over time. Corresponding solutions are described in Sec. 3.2. Achieving high system availability for the costumer requires to cope with varying environmental conditions. Thus road friction has to be considered for vehicle guidance as illustrated in Sec. 3.3. Due to a peculiarity of a level-3 system, the driver is no longer required to immediately take over vehicle guidance in case of an occurring system error, it results in the need for fail operational motion control. In Sec. 3.4 an approach is outlined to upgrade current typical failsafe systems to include fail operational functionality.

3.1 Central Vehicle Management Handling the Vehicle Guidance The aim of the “Central Vehicle Management” is to provide both low level driver assistant systems and high level automated driving with a simple standardized interface allowing to move the vehicle on a given trajectory. CVM includes lateral and longitudinal vehicle guidance with high quality of control as illustrated in Fig. 4. As there is a wide variety of driver assistance systems, ranging from parking to highway piloting, the controllers must cover a wide field of system states with low latency and

5

Motion control solutions for automated driving systems at BMW high robustness. They have to be robust with regard to variable disturbances for example up- or downhill grades, side wind or potholes and need good follow-up performance while being smooth and predictable. The controllers for longitudinal and lateral control therefore include adaptive feed-forward pilot control and feedback control parts.

Figure 4: Central Vehicle Management sub chain.

Depending on the level of automation, CVM has to tradeoff conflicting objectives for quality of control and reliability. For low level of automation, where the driver is able to take over control any time, shutting down the driver assistance functionality in a failsafe way is preferred over poor control quality, e.g. in case input signals are unknown or otherwise degraded. In contrast, a high level automated driving system needs to be fully fail-operational, stressing the need for reliability. This requires degradation mechanisms to work in differently depending on the level of automation. Coordinating inputs of CVM need plausibility checks and provides the controllers with safe default values in case of an input signal is unknown or otherwise degraded. The level of controller performance is directly communicated to the driver assistant functions. In case the actuator performance is reduced, combined capability feedback is forwarded ensuring the driving strategy to be adjusted. Controller outputs are distributed to the relevant actuators: the lateral controller output to front- and rear-axle-steering systems and the longitudinal controller output to the engine and brake system. Moreover, the BMW approach also considers coordination of

6

Motion control solutions for automated driving systems at BMW additional signals from driving dynamic functions. This is especially necessary for partly automated driving when the driver acts “hands-on”.

3.2 EgoMotion: Measurement of Vehicle Motion For better understanding of autonomous driving, one may take the traditional manual driving technology as a starting point and extrapolate this information for automated system requirements. Humans process a great deal of information in order to validate and estimate the driving state of the vehicle. This may happen in a more or less unconscious manner. It includes combining visual ques of the vehicle motion as well as perceived translational and rotatory movements of the body through the vestibular system as shown in Fig. 5. The total evaluation of the driving situation is of course extended by further human-perceptible stimuli. Human fusion of sensory perception is an enabler to overcome short distractions, so called “sensor unavailability” or to extend the scope of sensory abilities in total. Beyond that, fusion increases the level of confidence with respect to the subjective overall assessment of the driving situation. The analogy with human perception reveals two major requirements for vehicle motion estimation with regard to automated driving, as the responsibility and the task itself moves from driver to machine.

Figure 5: Sensory fusion analogy for human perception and systems integration.

On the one hand, now-a-days reliability of the signals describing the vehicle in motion are to be increased. In the event of failure or degradation, there is no driver in the loop to fall back on. On the other hand, the current signals indicating vehicle motion are to

7

Motion control solutions for automated driving systems at BMW be more precise. In this case accuracy is not a necessity for describing instantaneous vehicle movement, but for being able to perform a long term movement tracking of the vehicle. So the traveled and the predicted trajectory can be determined with the desired exactness. Currently available sensors by themselves are not capable to provide sufficient accuracy. In order to overcome this challenge, BMW developed enhanced, model based calibration algorithms for detection and compensating of systematic sensor effects. In addition to that, fusion of multiple sensor sources like radar, LiDAR, GNSS and even camera systems also significantly enhances the achieved precision. To entirely fulfil the requirements algorithms including artificial intelligence and Kalman filtering are used to complete and observe the information gained by 6DOF inertial sensors and the wheel speed sensors. Besides the improvements on sensor hardware level like redundancy, fusion of physically independent principles also enhances reliability. One of the core challenges is to supply the risen safety or integrity level, while simultaneously accomplishing the increased claims in terms of accuracy. Position and orientation of the vehicle are described by the odometry, which is primarily derived from wheel speed and inertial sensors at this stage. It directly or indirectly contains relevant information like radius of curvature, sideslip angle and a precise velocity. A high quality odometry is essential for smooth autonomous driving. Both, the past trajectory and the prospective trajectory play an important role later on within the sub chain for environmental interpretation. It provides high definition movement data for a short-range that cannot be retrieved from environmental sensors alone. Odometry is an established common service within the BMW driving dynamics software since years. It needs to be extended to fulfil requirements for autonomous driving such that the fail operational philosophy can be accomplished as shown in Sec. 3.4.

3.3 Consideration of Friction by Use of Cloud Data As stated in Sec. 1, autonomous driving (AD) will force a shift in responsibility from the driver controlling the vehicle towards the vehicle taking decisions by itself. Decisions and tasks, which the driver today handles and processes naturally, will be replaced by algorithms relieving the driver in terms of observing, interpreting, operating and decision-taking. Observations show humans drive differently depending whether they drive on dry asphalt, rain, snow, or mud. Even the smartest autonomous and semi-autonomous cars will therefor need to adjust how they drive, too, in order to keep their passengers safe and comfortable.

8

Motion control solutions for automated driving systems at BMW The influence of weather-, road and vehicle condition on the human driving behavior can be represented by a single parameter, called road friction. Road friction is a substantial safety and performance related parameter, which physically represents the bandwidth between minimum and maximum potential in terms of stability and traction. During winter times, when the temperature decreases dramatically, friction level reduces substantially leading to an increase of car accidents. Systems put in place to avoid accidents, such as slip control, emergency braking, electronic stability, adaptive cruise control and roll-over avoidance are all direct benefiters from available friction information. Due to increasing maturity of ADAS technology, current onboard observers and algorithms are already supporting the driver by providing and interpreting information regarding the current vehicle state and pathing. In order to ensure passenger safety and to secure trust in automation, the next step is not only to describe the actual vehicle states, but also to provide the vehicle with a (connected) foresight for the future road segments. This can be realized by using vehicle-to-infrastructure data collected from vehicles and aggregating this information in the cloud, where it is then used to warn other cars in the same area, as depicted in Fig. 6. The information about the friction hazard is also be sent to the road administrator, which can help to better plan and execute winter road maintenance and quickly address changed conditions. Among other use cases, this type of use case will be vital for a self-driving car.

Figure 6: V2X-Architecture for transmitting road friction hazards by cloud communication.

9

Motion control solutions for automated driving systems at BMW V2X-communication provides the benefit to extend the information that is generated and provided within an EGO-vehicle environment, by including information provided by other agents. Which in turn results in a more complete and detailed environment model allowing for optimized autonomous decision-making. The cloud platform locally processes and utilizes fusion techniques to combine different information types, which are in turn geo-referenced on a high-definition map. The information contained in a high-definition map is represented as layers. Organizing the information in layers makes it easy to independently design, build, test, and release new information. The basic road network data offered by a map provider acts as the bottom layer. Each subsequent layer adds additional details to the map. Currently available on-line sources providing weather forecasts are not yet accurate and precise enough to be useful for operational AD guidance. Several public and private providers offer weather information services capable of delivering detailed forecasts down to geographical points or road segments, however the spacing and distribution of currently available ground based meteorological observation stations delivering “start data” for the forecast modelling used in these services are much too sparse. This makes them unable to predict local variations and road-specific conditions on the detailed scale necessary for AD vehicle operation. Real-time floating car data from connected AD vehicles delivering weather-related sensor information through the OEM is potentially a very useful source for significantly enhanced forecasting. Obtaining high-quality road friction measurements, without additional vehicle sensors, is highly complex. Significant developments have been made in tire-road friction coefficient estimation in recent decades. One of the challenges is that several of the methods require extensive maneuvering of the vehicle such as severe accelerating, decelerating, and steering. Aside from technical challenges, questions regarding privacy and legal issues need to be addressed if OEMs are to use large scale customer generated data.

3.4 From a Failsafe System to Fail Operational Functionality As mentioned in Sec. 1, up to automation level-2 the driver is still responsible for handling the driving tasks. Thus, in occurrence of a hardware or software fault, the safe state of a driving assistance function is to simply shutdown and let the driver overtake the driving task. This is known as a “failsafe system”. With the highway pilot, BMW will introduce its first level-3 autonomous driving function. Starting from level-3, the driver is no longer responsible for monitoring the driving environment at all time, thus is no longer available as “mechanical fallback”. Instead, it needs to be ensured that the vehicle is able to brake and steer in order to keep following a given driving trajectory for the period of time the automated driving function remains active. Hence, a fail-operational or fail-degraded system needs to be provided,

10

Motion control solutions for automated driving systems at BMW which ensures the availability of the motion control system including its actuators in the presence of hardware or software errors for a certain amount of time. In order to deal with the paradigm shift from fail-silent to fail-operational systems, BMW provides a new fail-operational system design which is based on a redundant design of safety-critical electrical control units (ECU) such as the ECUs controlling braking or steering. This design allows to define a nominal functional path which is used in case of a fault-free state and fallback paths in case of a hardware or software error. By sophisticated error detection and system switching algorithms, a fallback path will be activated with appropriately working ECUs in case of a fault-detection as depicted in Fig. 7. The system is designed in a way that common causes as well as cascading failures which may lead to a degradation in both the nominal path and the fallback path are avoided. By this, availability of the motion control system is ensured in case of an error, even if in a degraded manner, depending on the giving use cases and the level of automation.

Figure 7: Paradigm shift from failsafe to fail operational.

While the motion control system of the nominal path includes all common driving functions, the functionality of the fallback path may be reduced to safety-relevant functions and may depend on the level of automation. In case of automation level-3, the fallback path only requires to bridge a limited time. In contrast, for automation level-4, the fallback system also needs to support use cases where the vehicle is not stopped in case of a fault but still continues driving the trajectory with a form of a fail-degraded system.

11

Motion control solutions for automated driving systems at BMW Independent from the functionality of the fallback path, one of the main challenges is to prove that the designed system fulfills the functional safety requirements according to ISO 26262 [4] and that dependent failures, which affect the availability of the system, are avoided. This includes both faults in software as in hardware. To achieve this, extensive safety analyses are required, not only on component level, but also on system level. These analyses ensure especially that common causes as well as cascading failures are prevented by system architecture [5].

4 Prospects for Future Application of Motion Control Solutions The motion control solutions as proposed in Sec. 3 are essential to enable maturing autonomous driving systems, which will shape future of mobility. The BMW approach sees vehicle guidance as an autonomic architecture element, with precise interface definition facilitating motion control, as a sub chain in the sense of a functional service. Arbitrary applications requiring a vehicle guidance based on trajectory input are quickly designed around using this service. An example for future application of motion control is autonomous driving within the production process of plants. For that use case plant infrastructure establishes radio communication with manufactured cars. The corresponding road network is supervised by camera systems and the vehicle individual trajectories from line end to load cargo are calculated by a centralized computer vision off-board. Cars originally setup in plant mode during production process will receive their trajectory input through radio communication and the intrinsic motion control service on-board adopt the vehicle guidance. After load of the car, the plant specific communication path is skipped by setting transportation or customer mode regarding automotive security requirements. Furthermore on demand mobility will require level-5 systems, ensuring fully autonomous driving. Car sharing solutions handling a fleet of these cars necessitate the capability to perform teleoperated tasks, e.g. to navigate an electric vehicle to a charging station or in case of a system error to a service-workshop. The ability of teleoperation is fundamentally enabled by the motion control service, which receives trajectory input over telecommunication and guides the vehicle. BMW’s strategy “Number One Next” considers these prospects as part of premium mobility solutions for the future.

12

Motion control solutions for automated driving systems at BMW

Bibliography 1. Dorrer, Claus; Friedrich, Reiner; Reinisch, Philipp. “Fahrerassistenzsysteme – von der Assistenz zum Automatischen Fahren”. 3. Internationale ATZ-Tagung. Frankfurt, 26.04.2017. 2. Dorrer, Claus. „Automated Driving at BMW – solutions for today and tomorrow“. 18. Internationales Stuttgarter Symposium; 14.03.2018. 3. SAE (2014). Taxonomy and definitions for terms related to on-road motor vehicle automated driving systems. SAE Standard J3016, USA. 4. ISO 26262-Road vehicles-Functional safety FDIS, 2018. 5. Schmid, Tobias; Schraufstetter, Stefanie; Wagner, Stefan. „An Approach for Structuring a Highly Automated Driving Multiple Channel Vehicle System for Safety Analysis”. 3rd International Conference on System Reliability and Safety. Barcelona, 24.-26.11.2018.

13

Is a typical Mercedes-Benz Driving Character still necessary with an increasing number of driver assistance systems? Dr. Stefan Botev, Dipl.-Ing. Horst Brauner, Prof. Dr. Ludger Dragon Daimler AG, 71059 Sindelfingen, Germany

Is a typical Mercedes-Benz Driving Character still necessary with an increasing …

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 M. Bargende et al. (Hrsg.), 19. Internationales Stuttgarter Symposium, Proceedings, https://doi.org/10.1007/978-3-658-25939-6_2

15

Is a typical Mercedes-Benz Driving Character still necessary with an increasing …

1 Mercedes-Benz Driving Character If you read the German automotive press one can get the impression that the journalists know the character of the brand Mercedes-Benz. Expressions like “A Benz like it should be.” or “a Daimler with typical values …” underline the character of the Mercedes-Benz driving behavior. This expectation is also there for our customers. Expressions like “the Mercedes under the …” demonstrate this assessment. In former times this Driving Character of Mercedes-Benz cars has been worked out mainly by doing subjective evaluations. The car behavior has been improved by modifying real parts. Today Mercedes-Benz uses so called “digital prototypes” to develop the vehicle functions. The digital prototype has been established by Daimler at the beginning of this millennium and is indispensable in the Mercedes-Benz car development (ref. 1). Nowadays the digital prototype can be understood as a forerunner of the “digital transformation” process in the automotive industry. Once you have simulations which can predict the vehicle performance the development time becomes shorter and the development risk decreases. Even if your simulations are highly validated, that is to say that the simulations can predict the measured values of the real car, there is still the well-known challenge to know which measured or simulated value describe the driving character in a precise and sufficient way.

Figure 1: Connection between semantic descriptions of the Mercedes-Benz Driving Character and corresponding objective values

16

Is a typical Mercedes-Benz Driving Character still necessary with an increasing … This is the reason to establish the so called Mercedes-Benz Driving Character. There are two aspects which have to be fulfilled: a) a semantic description of the MercedesBenz Driving Character and b) the objective values, which have to correspond to the semantic expressions. Four disciplines are contributing to the Mercedes-Benz Driving Character by delivering the necessary features: ● Longitudinal dynamics ● Ride ● Handling ● Braking In a semantic way several distinct qualities describe the Mercedes-Benz Driving Character: ● Safety feeling: A safety feeling means that the driver always feels confident to control the car, also under difficult and changing conditions. ● Ride comfort: Ride comfort means, that the vehicle isolates the passenger against the street and carries them comfortably and capably to the goal. ● Sportiness: Sportiness is the possibility and the feeling to drive a vehicle fast with good control. ● Precision: Precision means precise and predictable controlling of the vehicle. ● … If necessary additional distinct qualities can be added.

Figure 2: Evaluation according to given characteristic values for the distinct quality “sportiness”

17

Is a typical Mercedes-Benz Driving Character still necessary with an increasing … The Mercedes-Benz Driving Character is defined by the claim of the distinct qualities. The claim for the distinct qualities safety feeling and ride comfort is to be the best whereas for the distinct qualities sportiness and precision the Mercedes-Benz Driving Character defines a position to be among the “top 3” of the corresponding car segment. In other words the Mercedes-Benz Driving Character is always defined for a segment of cars, i.e. segment A-class, C-class, E-class, S-class, … . The contributing disciplines deliver also which objective evaluation criteria have to be simulated or measured. For each of these criteria a corresponding maneuver is defined. At the next level objective values are created. For the maneuver “18 m slalom” it is the time needed for driving this maneuver and for driving over a single obstacle there are several possibilities to describe the car behavior. Thereby a set characteristic values and a semantic description of the Mercedes-Benz driving character is given (see Fehler! Verweisquelle konnte nicht gefunden werden.). Now a procedure is needed to calculate from the characteristic values the distinct qualities. As an example the evaluation according to the given characteristic values for the distinct quality “sportiness” is sketched in Figure 2. In the first out of 4 steps each objective value has to be transformed to a rating scale, which consists of nine units. 1 is the best rating. To set up a rating scale the experience of the different disciplines is mandatory. Some basic knowledges about the threshold of difference perceptions as well as the absolute range of reasonable characteristic values are helpful. For some disciplines these information’s can be worked out by the use of simulators (ref. 2).

Figure 3: The Mercedes-Benz Driving Character takes into account new technologies

After completing this transformation for all relevant characteristic values a weighted mean is calculated for each disciplines. In a third step the weighted mean values from each of the four disciplines is again used to create a weighted summation. For the distinct quality “sportiness” 45% are coming from the discipline acceleration, 15% from

18

Is a typical Mercedes-Benz Driving Character still necessary with an increasing … braking, 30% from handling and 10% from ride. The other four distinct qualities are calculated in an analogous way. For each distinct quality the amount of contribution from the four disciplines is varying. With this fixed procedure cars can be easily compared in all considered distinct qualities like safety feeling, ride comfort, sportiness, precision …

Figure 4: Description of driver task with no assistance system (EPS=electronic power steering, CPC=central powertrain controller, ESP=electronic stability program)

Now, it is a good point to check, whether the Mercedes-Benz Driving Character is prepared for the future, in other words for cars with new technologies like assistance systems. In order to do this we can look into the history of the E-class starting with the first one in 1953 until today (see Figure 3). Three examples demonstrate how the MercedesBenz Driving Character has improved over the years: ● A typical tire size for an E-class from 1975 was the dimension 195/70R14. Whereas today the E-class has a typical tire size of 245/40R18. ● Starting from 1998 ESP (electronic stability program) is a standard equipment and legally required in a lot of countries. Beside additional functions ESP helps to solve the classical conflict between the car performance in the maneuvers “high speed braking in the corner” and “braking on µ-split”. ● EPS (electronic power steering) started as a standard equipment in 2009. In former times the steering oscillations could be a major problem in the car development. Today EPS helps with an appropriate controller to reduce this problem or the steering wheel oscillations even vanishes with EPS.

19

Is a typical Mercedes-Benz Driving Character still necessary with an increasing … The methodology of the Mercedes-Benz Driving Character takes this improvements into account by always selecting a certain number of cars like predecessor, successor and recent competitor´s cars. Thereby the higher challenges for the next generation are considered and new technical solutions are covered.

2 Driver assistance systems and the Mercedes-Benz Driving Character: How do they interact?

Figure 5: New A-class: assistance systems use 19 sensors to get information’s about the environment

In order to describe the interaction of the assistance systems with the Mercedes-Benz Driving Character we start with the situation without assistance systems. Assistance systems means in this article those vehicle functions which take into account the environment, i.e. traffic and/or the road marking, … . The driver task is to steer, to brake and to accelerate the car by using the steering wheel, the brake pedal and the acceleration pedal. In addition to the vehicle responds of the driver input, there might be disturbances like crosswind, varying friction levels of the road or the traffic (cars, pedestrians, motorcyclists, bike-rider, …). With his senses the driver recognizes the driving situation und even more evaluates this situation as a whole. Especially he compares the actual track with the target track and afterwards he decides to do the next input to the car (see Figure 4), which is called the basic car in this article. The Mercedes-Benz Driving Character determines how the car reacts to the inputs and how robust the car is with respect to disturbances. As mentioned before ESP und EPS are standard options for all Mercedes-Benz passenger cars, so these features are included in the basic car.

20

Is a typical Mercedes-Benz Driving Character still necessary with an increasing … As an example the new A-class uses 19 sensors for assistance systems. The different sensor types are Multi Mode Radar, Stereo Multi Purpose Camera, Long Range Radar and Ultrasonic Sensors (see Figure 5). With these sensors the A-class “sees” the environment. After the evaluation of the driving situation has been done, the assistance systems check whether they will offer one of the following features: ● ● ● ● ● ●

Active distance assist DISTRONIC with steering assist Active speed limit assist Active emergency stop assist Active lane change assist Map based speed adaption …

Figure 6: A seamless interaction between both control loops contributes to the Mercedes-Benz Driving Character

If one of these assistance system function will be executed, digital information’s are given to the EPS (steering), ESP (braking) and CPC (accelerating). These systems from the basic car use their already existing connections to the vehicle hardware like producing hydraulic pressure at the brake discs in order to generate a brake force. In other words the assistance systems use the basic car as an intelligent actor and assumes that this intelligent actors has been already tuned to the Mercedes-Benz Driving Character. In Mercedes cars like the new A-class with assistance systems there are two control loops to realize the driving task. One loop is already known from Figure 4 and describes the driver tasks with no assistance systems. The driver interacts via steering wheel and

21

Is a typical Mercedes-Benz Driving Character still necessary with an increasing … pedals with the basic car. And the other loop describes the functionality of the assistance systems (see Figure 6). The assistance systems give their information’s directly to the different controller of the basic car. Thereby a seamless interaction of both control loops is realized and the assistance systems contribute to the Mercedes-Benz Driving Character.

Figure 7: Six levels are defined to describe the degree of automation. See also the similar versions from BAST and NHTSA

In the future the number of assistance systems will increase and their functionality will be extended. In the newest version of the standard SAE J3016 (ref. 3) six levels are defined to describe the degree of automation (see Figure 7). These levels are characterized by: ● ● ● ●

Execution of steering and de-/acceleration Monitoring of driving environment Fallback performance of dynamic driving task System capabilities

With respect to the cars in the market, we are today at the border to SAE level 3. The driver plays a role up to the SAE level 4. Hence the Mercedes-Benz Driving Character is mandatory up to this SAE level including the feedback from the steering wheel and the pedals.

22

Is a typical Mercedes-Benz Driving Character still necessary with an increasing …

3 Outlook for the Mercedes-Benz Driving Character with autonomous driving In the second half of 2019 Ride-Hailing will be tested in San José with an automated Sclass corresponding to SAE level 4/5. Daimler, Bosch and San José have signed a memorandum of understanding to pursue and finalize this activity. This self-driving car pilot is based on a modified S-class which fulfills the Mercedes-Benz Driving Character.

Figure 8: A seamless interaction between both control loops contributes to the Mercedes-Benz Driving Character

For those future cars which fulfill the SAE level 5 the Mercedes-Benz Driving Character has to be updated as it has been done in the past for technology milestones. It is expected that the distinct qualities safety feeling and ride comfort will be modified only slightly, i.e. it is obvious that the level of ride comfort shall be at least the same level independent whether the car possess a SAE level 2, 3, 4 or 5. The distinct qualities sportiness and precision seem to be only necessary if a car with SAE level 5 will also offer a driving mode with steering wheels and pedals. In other words driving a Mercedes-Benz car or been driven by a Mercedes-Benz car always displays the Mercedes-Benz Driving Character.

23

Is a typical Mercedes-Benz Driving Character still necessary with an increasing …

Bibliography 1. T. Breitling, L. Dragon, T. Grossmann, Digitale Prototypen – ein weiterer Meilenstein zur Verbesserung der Abläufe und Zusammenarbeit in der PKW-Entwicklung, VDI-Tagung Berechnung und Simulation in der Fahrzeugbau, VDI-Berichte 1967 (Würzburg, 2006), ISBN: 3-18-091967-1 Seite 315-326 siehe auch VDI-Jahrbuch 2006 Fahrzeug- und Verkehrstechnik, ISBN: 3-18-401655-2 2. Dipl.-Ing. J. Colditz, Dipl.-Ing. T. Grossmann, Dr. L. Dragon, Objektive und subjective Abstimmung der Fahrkultur digitaler Prototypen unter Einbeziehung von Simulatoren, 3. Nationale Tagung Humanschwingungen, 8.-9. Okt. 2007, Dresden 3. SAE International, J3016_201806: Taxonomy and Definitions for Terms Related to Driving Automation Systems for On-Road Motor Vehicles (Warrendale: SAE International, 15 June 2018), https://www.sae.org/standards/content/j3016_201806/

24

Reasons why customers will still buy premium (chassis) in the age of automated driving Armin Schöpfel AUDI AG

This manuscript is not available according to publishing restriction. Thank you for your understanding.

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 M. Bargende et al. (Hrsg.), 19. Internationales Stuttgarter Symposium, Proceedings, https://doi.org/10.1007/978-3-658-25939-6_3

25

Technical scenarios for the decarbonization of road transport Stephan Neugebauer European Road Transport Research Advisory Council (ERTRAC)

This manuscript is not available according to publishing restriction. Thank you for your understanding.

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 M. Bargende et al. (Hrsg.), 19. Internationales Stuttgarter Symposium, Proceedings, https://doi.org/10.1007/978-3-658-25939-6_4

27

Comparative evaluation of PtX processes for renewable fuel supply Maximilian Heneka, Wolfgang Köppel DVGW-Research Centre at Engler-Bunte-Institute (EBI) of Karlsruhe Institute of Technology (KIT)

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 M. Bargende et al. (Hrsg.), 19. Internationales Stuttgarter Symposium, Proceedings, https://doi.org/10.1007/978-3-658-25939-6_5

29

Comparative evaluation of PtX processes for renewable fuel supply

Motivation With the adoption of the Paris Climate Agreement on December 12, 2015, the international community committed itself to limiting global warming to below two degrees Celsius compared to the pre-industrial era. Germany's contribution to achieve the climate protection targets was ratified by the German government in November 2016 introducing a policy document called “Klimaschutzplan” 2050’ [1]. Within this document the German government intends to reduce the greenhouse gas (GHG) emissions by at least 55 % by 2030 and 80–95 % by 2050 compared to 1990 levels [2]. These climate targets pose major challenges to the transport sector. By 2030, transport-related GHG emissions must be reduced by at least 40–42 % [2]. Due to “unavoidable” emissions from industry and agriculture, a complete defossilization of the transport sector is necessary in order to achieve the 2050 targets. In 2017, the transport sector accounted for about 19 % of German GHG emissions. 96 % of these emissions were caused by road traffic (35 % trucks, 61 % cars)1 [2]. Future forecasts assume that GHG emissions will continue to rise due to increasing transport demand, especially in freight transport [3, 4, 5]. Considering the ambitious emission targets and the limited remaining CO2 budget2 until 2050 (cf. [6]) fast-acting approaches are needed to gradually but quickly reduce GHG emissions in Germany. In addition, legally enforced driving restrictions for diesel vehicles due to air pollution in many German cities increase the pressure on local governments and the automotive industry to provide sustainable, low-emission mobility concepts. Battery electric vehicles (BEV) or fuel cell electric vehicles (FCEV) are often mentioned as a suitable solution for reducing GHG emissions in transport. The advantage of these technologies lies in high powertrain efficiencies and in providing a (locally) zero-emission vehicle operation. However, BEV and FCEV still show little market relevance due to a narrow product range, high vehicle costs and unsolved challenges regarding the feasibility of a nation-wide refueling infrastructure. At the same time, it must be considered that the GHG reduction potential of these technologies depends mainly on the GHG footprint of the value chains and the electricity mix for production and (vehicle) operation. Even if some automobile manufacturers claim “the end of the combustion engine" to be approaching [7], a one-sided focus on e-mobility seems not expedient. Until BEV (or FCEV) show a measurable GHG-reducing effect on the transport sector, large parts of the remaining CO2 budget could already be depleted. Furthermore, Germany

1 2

Reference year: 2016. If emissions remain at the current level, the German CO2 budget will be depleted by 2025 [6].

30

Comparative evaluation of PtX processes for renewable fuel supply is at risk to miss its binding climate protection commitments for 2021–2030. The associated penalty could turn out to be a serious burden for the federal budget [8]. It seems reasonable to consider internal combustion engine vehicles (ICEV) for future mobility concepts in order to achieve a fast reduction in GHG emissions. Especially natural gas mobility offers a great potential for a gradual transition to a low-emission mobility sector. Modern natural gas vehicles could contribute to a significant reduction both in local emissions and in CO2 emissions compared to diesel vehicles (cf. Figure 1 and Figure 2). As of now, natural gas vehicles are economically viable in numerous traffic segments (see Figure 3). Obviously, the fuel used for natural gas vehicles does not necessarily have to be of fossil origin. Natural gas vehicles offer a high "greening potential" with the possibility of adding biomethane in gradually growing amounts (cf. Figure 2).

Emissions vs. Diesel Euro VI in %

100 90

100% = Euro VI emission standard

- 24 %

80

- 37 %

70 60

- 50 % - 54 %

Diesel LNG

50 40 30 - 95 %

20 10 0 NOx

Particulate matter

Traffic Noise

Figure 1: NOx-, Particulate matter emissions- and Traffic Noise of a Euro VI diesel truck (without AdBlue and particle filter) and a Euro VI LNG truck compared to the Euro VI emissions standard [9, 10]

31

Comparative evaluation of PtX processes for renewable fuel supply Well-to-Wheel emissions in g CO2-eq/km

Fuel and Engine Type

0

200

400

600

800

1000

Diesel (FQD)

1200

1400

1074

LNG (HPDI)

912

LNG (HPDI, 20% bioLNG)

791

CNG (SI)

908

CNG (SI, 20% bioCNG)

769

HPDI: High Pressure Direct Injection, SI: Spark Ignition, FQD: Standard Value from Fuel Quality Directive

Figure 2: Well-to-Wheel-GHG Emissions: Heavy Duty Vehicles in g CO2-eq/km [11]

Total Cost of Ownership in €/km

4,5 3,93

4,0 3,5 3,0

2,86

3,00

Emission Cost

2,58

2,54

2,5

Staff Cost

2,0

Fuel Cost

1,5

Maintenance

1,0

Capital Cost

0,5 0,0 Diesel

BEV (OC)

BEV (ON)

CNG

FCEV (H2)

BEV: Battery Electric Vehicle, FCEV: Fuel Cell Electric Vehicle, ON: Overnight Charging, OC: Opportunity Charging

Figure 3: Total cost of ownership (TCO) of city buses with different charging concepts in €/km

In the medium to long term, so-called Power-to-X (PtX) fuels allow for the establishment of a fully CO2-neutral mobility for ICEV. These electricity-based fuels can be domestically produced or they can be imported from countries with large amounts of renewable electricity capacity (e.g. Middle East, North Africa, Australia or South America).

32

Comparative evaluation of PtX processes for renewable fuel supply

Power-to-X fuels Definition and Classification PtX fuels are fuels whose production is based on the integration of renewable hydrogen. Depending on the type of the synthesis path, a distinction is made between Power-to-Gas (PtG) and Power-to-Liquid (PtL) (see Figure 4).

Figure 4: Overview of possible synthesis paths for PtX fuels

An important criterion for the GHG neutrality of PtX fuels is the type of carbon source used for the synthesis. For a complete defossilization of the transport sector, the carbon source must be CO2 neutral or at least unavoidable. If compatibility with existing infrastructure is to be maintained, only fuel syntheses providing liquid or gaseous hydrocarbon compounds (CxHyOz) can be used. This study focuses on the comparison of different PtX process chains (see Figure 5) on a uniform and detailed procedural basis. After modelling promising PtX process chains, the fuel production was evaluated in terms of efficiencies, fuel cost and well-to-wheel emissions. Only biogenic residues and CO2 from air were considered as carbon sources. Unavoidable CO2 emissions from industry were not investigated.

33

Water

Comparative evaluation of PtX processes for renewable fuel supply PtL-process route

H2

Electrolysis

PtG-process route H2 via electrolysis CRG: Compressed Renewable Gas LRG: Liquefied Renewable Gas

RWGS

Reverse Watergas Shift

CO Air

CO2

CO2

CO2-Capture

H2/CO/CO2/CH4 FB

Fluidized Bed Gasification

METH

Gas Gasaufbereitung Conditioning

Catalytic Methanation

Compression

CRG

Liquefaction

LRG

Biomass

H2 bioMETH

FER

Biological Methanation

CO2/CH4

Fermentation

H2 Pyrolysis + Entrained Flow Gasification

EF

H2 FT

RWGS

Reverse Watergas Shift

Fischer-TropschSynthesis

FT-Diesel H2

H2/CO/CO2

OME

Gas Conditioning

OMESynthesis

H2

OME

H2 DMEx2

MeOH

MethanolSynthesis

DMESynthesis

DME

H2 MtG

Naming convention for process chains: FT-Diesel-Synthesis via Entrained Flow Gasification and CO2-conversion:

„EF-RWGS-FT“ DME-Synthesis via CO2-Capture and MethanolSynthesis

„CO2-RWGS-DME x2“

MtGSynthesis

Gasoline

H2 DMEx1

DME-Synthesis

DME H2 DtG

DtGSynthesis

Figure 5: Overview of the PtX process chains considered in this study

34

Gasoline

Comparative evaluation of PtX processes for renewable fuel supply

Methodology For the comparative evaluation of the PtX fuels, each process chain was first evaluated in a standard configuration with minimum overall energy efficiency followed by the analysis of an optimized alternative with maximum overall energy efficiency. For the optimized process chains, a complete reintegration of usable waste heat was carried out. By-products, such as those produced by Fischer-Tropsch synthesis, were regarded as valuable products and credited to the overall process efficiency. For the hydrogen supply, a high-temperature electrolysis (SOEC) was considered, which allows an easier and more efficient integration of internal excess heat. Figure 6 schematically shows the methodology used to calculate the maximum process chain efficiency using an energy Sankey diagram.

Figure 6: Methodology for efficiency calculation of optimized PtX process chains

35

Comparative evaluation of PtX processes for renewable fuel supply

Comparison of process chains: Efficiency Figure 7 and Figure 8 provide an overview of the energetic efficiencies of the integrated process chains. High efficiencies can be achieved in methane-producing processes. For example, the production of CRG (Compressed Renewable Gas) via fluidized bed gasification of biomass and catalytic methanation (FB-METH) achieves a maximum efficiency of 75 % through the reintegration of waste heat. The efficiencies of PtL process chains generally decrease complexity. This can be observed particularly clearly for chains for the supply of OME and gasoline (MtG and FT-Diesel process chains using biomass achieve the (EF-MeOH: 58 %, EF-RWGS-FT: 57 %).

with increasing process the multi-stage process DtG). The MeOH and maximum efficiencies

If CO2 from air is used as carbon source (Figure 8), the efficiencies are significantly lower for all process chains due to the higher energetic expenditure of direct air capture (DAC) technologies. Maximum Process chain efficiency ηP,max 0,0

0,2

0,4

0,6

FER-METH

0,8

1,0

0,60

FER-bioMETH

PtG

0,57

Power-to-Gas

FB-METH

0,75

EF-FT

0,52

EF-RWGS-FT

0,57

EF-MeOH

0,58

EF-RWGS-MeOH

0,50

EF-DME-x1

PtL

0,48

EF-DME-x2

Power-to-Liquid

0,53

EF-MtG

0,41

EF-DtG

0,39

EF-OME

0,42

CRG

MeOH

Gasoline

FT-Diesel

DME

OME

Figure 7: Maximum efficiencies of the PtX process chains with biomass as carbon source

36

Comparative evaluation of PtX processes for renewable fuel supply Maximum Process chain efficiency ηP,max 0,0

0,2

0,4

0,6

CO2-RWGS-METH

0,8

CO2-RWGS-FT

1,0 PtG

0,57

Power-to-Gas

0,48

CO2-RWGS-MeOH

0,46

CO2-RWGS-DME-x1

0,33

CO2-RWGS-DME-x2

PtL

0,42

CO2-RWGS-MtG

Power-to-Liquid

0,35

CO2-RWGS-DtG

0,26

CO2-RWGS-OME

0,36

CRG

MeOH

Gasoline

FT-Diesel

DME

OME

Figure 8: Maximum efficiencies of the PtX process chains with direct air capture technologies

Comparison of process chains: Fuel cost To estimate the PtX fuel cost, first the net production cost were determined for the reference year 2015 using literature data and the annuity method described in VDI 6025 [12]. In a second step, the net production cost were calculated for 20503 considering learning curves. The biomass supply was taken into account as biomass cost including transportation. Other costs such as taxes or insurance cost were not considered. The by-products were integrated energetically into the overall process chain in accordance with the methodology shown in Figure 6 and credited to the process chain efficiency. The material and energy flows that were required to calculate the consumption-linked costs were taken from the material and energy balances of the examined process chains. In order to calculate the fuel cost at the filling station (excluding taxes), the cost for fuel transportation and the investment for the filling station were considered. The different powertrain efficiencies were taken into account using data from reference vehicles (passenger car, C-segment; fuel consumption see Table 1). Vehicle cost were neglected, since this study focuses on the comparative evaluation of PtX process chains.

3

“2050” represents an industrial scale PtX fuel production and well-established production sites.

37

Comparative evaluation of PtX processes for renewable fuel supply Table 1: Assumed fuel consumption of the reference car powertrains (C-segment) Powertrain

Fuel

ICEV-G ICEV-D ICEV-CH4

Gasoline, MeOH (FT-)Diesel CNG (CRG)

Fuel consumption in MJ/km 2015 2050 2,4 1,5 2,0 1,5 2,3 1,5

Figure 9 shows a schematic overview of the underlying calculation method.

Figure 9: Methodology for calculating fuel costs at the filling station (excluding taxes)

To investigate the influence of economies of scale on the PtX fuel cost, two different electrolysis capacities (5 MW and 500 MW) were considered. In the case of biogenic PtG process chains, the 5 MW segment is exclusively served by fermentation-based process chains (FER-METH, FER-bioMETH). In the 500 MW range, on the other hand, fluidized bed gasification is used.

38

Comparative evaluation of PtX processes for renewable fuel supply Further assumptions for the cost calculation are listed in Table 2: Table 2: Assumptions used to calculate the fuel cost Full load hours Depreciation period Biomass cost (gasification) Biomass cost (fermentation) Electricity cost Operation and maintenance

3500 h/a 15 a 100 €/t (dry matter) 130 €/t (dry matter) 3 ct/kWh 4 % of investment

Figure 10 shows the fuel cost for the year 2050 for selected PtX fuels as a function of the electrolysis performance. Minimum Fuel Cost (2050*) in €/100 km 5 10 15 20

500 MW

5 MW

500 MW

5 MW

0 FER-METH FER-bioMETH EF-RWGS-FT EF-RWGS-MeOH EF-MtG FB-METH EF-RWGS-FT EF-RWGS-MeOH EF-MtG CO2-METH CO2-RWGS-FT CO2-RWGS-MeOH CO2-RWGS-MtG CO2-METH CO2-RWGS-FT CO2-RWGS-MeOH CO2-RWGS-MtG

25

4,11 3,89 6,13 5,98 9,21

C from Biomass

2,58 3,31 3,21 4,63 6,27 8,78 8,68 13,41 CO2 from Air

3,77 4,56 4,77 6,74

*: 2050 represents an industrial scale PtX fuel production and well-established production sites.

CRG

MeOH

Gasoline

FT-Diesel

Figure 10: Fuel cost at filling station (without taxes) of selected PtX fuels broken down by carbon source and electrolysis capacity in MW

According to Figure 10, PtG processes represent the most cost-effective PtX fuels for all considered plant sizes. For the bio-based process chains the fuel cost in 2050 are 2,58–4,11 €/100 km and 3,77–6,27 €/100 km for the CO2-based process chains. The

39

Comparative evaluation of PtX processes for renewable fuel supply CRG fuel cost (industrial scale production, without taxes) thus are similar to current diesel cost (approx. 2,88–4,04 €/100 km; without taxes; C-segment ICEV-D; 2 MJ/km).

Comparison of process chains: Well-to-Wheel emissions The ecological evaluation of PtX fuels was carried out based on a Well-to-Wheel (WtW) approach. Therefore, all material and energy flows occurring along the fuel supply chain and the associated CO2 equivalent emissions (CO2-eq) are analyzed. Preliminary construction work, vehicle production and end-of-life expenses are not taken into account. Due to the different lower calorific values of the PtX fuels (compared to conventional Diesel) and the different powertrain efficiencies, the calculated GHG emissions were related to a reference vehicle. The reference vehicle and the associated fuel consumption values correspond to the data used in the fuel cost calculation for the year 2015 (see Table 1). Figure 11 provides a schematic overview of the considered WtW process chains.

Figure 11: Schematic structure of the considered Well-to-Wheel process chains for PtX and fossil fuels

In Figure 12, the WtW emissions of the individual process chains for the maximum process chain efficiencies are displayed. By assumption, wind power plants provide 100 % of the electrical energy that is required for the PtX fuel production. The other

40

Comparative evaluation of PtX processes for renewable fuel supply subsystems (filling station, biomass supply and transport) and the reference vehicle were based on the reference year 2015. According to the Renewable Energy Directive [13] and its recast for 2021–2030 (Renewable Energy Directive II) [14], the combustion of PtX fuels is considered carbon neutral, which means there are no Tank-toWheel CO2 emissions (closed CO2 cycle). Well-to-Wheel-Emissions in g CO2-eq/km 0

50

100

Diesel

150

200 177,4

Gasoline

209,8

CNG 15,9

FER-bioMETH

17,0

CRG

16,4

EF-FT

4,5

EF-RWGS-FT

3,9

EF-MeOH

6,3

EF-RWGS-MeOH

6,8

EF-MtG

5,9

CO2-METH

Fossil Reference

161,1

FER-METH FB-METH

Reference year: 2015

250

FT-Diesel

PtX-Fuels C from Biomass

MeOH Gasoline

14,5

CRG

CO2-RWGS-FT

1,0

FT-Diesel

CO2-RWGS-MeOH

2,7

MeOH

CO2-RWGS-MtG

1,3

PtX-Fuels CO2 from Air

Gasoline

Well-to-Tank

Tank-to-Wheel

Figure 12: Well-to-Wheel-GHG Emissions: Passenger car (C-segment, 2015) in g CO2-eq/km

The well-to-wheel analyses yield that PtX fuels offer a GHG savings potential of 90–99 % when using renewable power and green carbon for PtX production. The GHG emissions are subject to a continuous decrease, which is proportional to the market penetration of renewable fuels. With a distribution infrastructure based purely on renewable energy sources, the use of ICEVs based on PtX fuels can establish completely GHG-neutral transport sectors.

41

Comparative evaluation of PtX processes for renewable fuel supply

Conclusion In order to reach the climate protection targets, a fast and drastic reduction of GHG emissions in road transport is necessary. Considering the impending or already implemented driving restrictions in several German cities, the demand for sustainable mobility concepts limiting local air immissions is drastically increasing. In addition to battery electric vehicles or fuel cell electric vehicles, natural gas vehicles provide a high potential to reduce both local emissions and GHG emissions compared to gasoline and diesel vehicles. The advantage of methane-based mobility is its fast implementation. Both technology and infrastructure are available and can compete economically with modern diesel engines in numerous vehicle segments. PtX fuels offer a solution to the pretextual incompatibility between internal combustion engines and defossilization showing a GHG reduction potential of 90–99 %. Similar to e-mobility, GHG emissions are improving as the share of renewable energy sources in the transport and electricity sector increases. In the medium to long term, GHG emissions from ICEV can be reduced to almost zero with PtX fuels. PtX fuels offer advantages in terms of implementation time and cost since existing infrastructure and vehicles can be used with minor adaptions. It was shown that Powerto-Methane process chains offer a useful supplement for near future road traffic due to their high efficiencies of 60–75 %. Assuming well-established production sites or PtX fuel production at an industrial scale, the fuel cost for CRG (without taxes) is similar to current diesel cost (without taxes).

Bibliography 1. Bundesministerium für Umwelt, Naturschutz, Bau und Reaktorsicherheit (BMUB), „Klimaschutzplan 2050. Klimaschutzpolitische Grundsätze und Ziele der Bundesregierung“, Berlin, 2016. 2. Bundesministerium für Umwelt, Naturschutz und nukleare Sicherheit (BMU), „Klimaschutz in Zahlen. Fakten, Trends und Impulse deutscher Klimapolitik“, Berlin, 2018. 3. P. Kasten, M. Mottschall, W. Köppel, C. Degünther, M. Schmied und P. Wüthrich, „Erarbeitung einer fachlichen Strategie zur Energieversorgung des Verkehrs bis zum Jahr 2050“, Umweltbundesamt, Dessauz-Roßlau, 2016. 4. Öko-Institut e.V., „Klimafreundlicher Verkehr in Deutschland. Weichenstellungen bis 2050“, WWF Deutschland, 2014.

42

Comparative evaluation of PtX processes for renewable fuel supply 5. Öko Institut e.V.; Fraunhofer ISI, „Klimaschutzszenario 2050: 2. Endbericht“, Berlin, 2015. 6. Öko-Institut e.V.; Prognos, „Zukunft Stromsystem. Kohleausstieg 2035. Vom Ziel her denken“, WWF Deutschland, Berlin, 2017. 7. ZEIT Online; dpa, „VW kündigt Ende von Verbrennungsmotoren an“, ZEIT ONLINE, 2018. 8. Agora Energiewende; Agora Verkehrswende, „Die Kosten von unterlassenem Klimaschutz für den Bundeshaushalt. Die Klimaschutzverpflichtungen Deutschlands bei Verkehr, Gebäuden und Landwirtschaft nach der EU-Effort-Sharing Entscheidung und der EU-Climate-Action-Verordnung“, 2018. 9. S. Feldpausch-Jägers, M. Henel, J. Ruf und W. Köppel, „Potentialanalyse LNG Einsatz von LNG in der Mobilität, Schwerpunkte und Handlungsempfehlungen für die technische Umsetzung. Abschlussbericht“, DVGW-Förderzeichen G7/01/15, 2016. 10. K. Kröger und W. Köppel, „Wissenschaftliche Begleitung eines Demonstrationsprojektes zum Einsatz von LNG als Kraftstoff für LKW“, DVGW-Förderzeichen 1456/G20, 2018. 11. thinkstep AG, „Greenhouse Gas Intensity of Natural Gas“, NGVA Europe, 2017. 12. VDI Richtlinie 6025, Betriebswirtschaftliche Berechnungen für Investitionsgüter und Anlagen, November 2012 13. Richtlinie (EU) 2009/28/EG des Europäischen Parlaments und des Rates vom 23. April 2009 zur Förderung der Nutzung von Energie aus erneuerbaren Quellen und zur Änderung und anschließenden Aufhebung der Richtlinien 2001/77/EG und 2003/30/EG: RL (EU) 2009/28, 2009. 14. Council of the European Union, “Proposal for a DIRECTIVE OF THE EUROPEAN PARLIAMENT AND OF THE COUNCIL on the promotion of the use of energy from renewable sources (recast)”, 2016/0382 (COD), Brussels, 2018.

43

Modelling of real fuels for an effective virtual engine development with focus on alternative fuel M. Sc. Francesco Cupo, Dr.-Ing. Marco Chiodi, Prof. Dr.-Ing. Michael Bargende FKFS Dipl.-Ing. Daniel Koch, Prof. Dr.-Ing. Georg Wachtmeister Technische Universität München Dr.-Ing. Donatus Wichelhaus Volkswagen AG

Modelling of real fuels for an effective virtual engine development with focus on …

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 M. Bargende et al. (Hrsg.), 19. Internationales Stuttgarter Symposium, Proceedings, https://doi.org/10.1007/978-3-658-25939-6_6

45

Modelling of real fuels for an effective virtual engine development with focus on … -------------------------------------------------------------------------------------------------------This manuscript, according to publishing restrictions, is available only as a brief summary of the extensive simulation and experimental investigations. Thank you for your understanding. --------------------------------------------------------------------------------------------------------

1 Abstract The worldwide environmental issues are affecting the development processes in all industrial sectors. Among these, the automotive is probably the one facing the toughest challenges. The reduction of both harmful emissions (CO, HC, NOx, etc.) and gases responsible for greenhouse effects (especially CO2) are mandatory aspects to be considered in the development process of any kind of propulsion concept. A comprehensive well-to-wheel analysis – in comparison with the less inclusive yet very common tankto-wheel approach – is for sure the most appropriate way in order to effectively measure future progresses. All mobility scenarios until 2050 confirm that internal combustion engines (ICEs) will still play an important role (especially as hybrid-solutions) for passenger cars and even more for trucks and marine applications. Focusing on ICEs, the main development topics are today not only the reduction of harmful emissions, increase of thermodynamic efficiency, etc. but also the decarbonization of the fuels. The last point offers the highest potential for the reduction of CO2 emissions. Following this approach, the development of future ICEs is closely linked to the development of CO2 neutral fuels (e.g. biofuels and e-fuels). Accordingly, engines and fuels will be part of a common development process. This implies an increase in development complexity, which needs to be supported by engine simulations, now more than ever. In this work, the virtual modelling of real fuel behavior is addressed in order to improve current simulation capabilities in studying how a specific composition can affect engine performance. The goal is to create a series of models that allow to virtually investigate different fuels and to minimize as much as possible the costly and time-consuming experimental tests. The fuel investigation - conducted virtually at the FKFS of Stuttgart and experimentally at the engine laboratory of the Chair of Internal Combustion Engines at the Technical University of Munich with the support from Volkswagen Motorsport GmbH - was performed on a single cylinder research engine operating with the innovative SACI (Spark Assisted Compression Ignition) combustion concept [6,12,13].

46

Modelling of real fuels for an effective virtual engine development with focus on …

2 Introduction In recent times, we have witnessed a radical change in the automotive industry with many new technologies that aim principally to reduce the emission of greenhouse gases. More stringent regulations for corporate average CO2 emissions created a general skepticism towards the internal combustion engine, which moved the focus toward other forms of propulsion such as fuel cells and electric powertrains. Nevertheless, the combination of high-efficiency internal combustion engines and alternative fuels retains significant potential in reducing overall emissions. In particular, fuel flexibility towards bio-fuels and synthetic fuels (mainly e-fuels: power-to-liquid or power-to-gas fuels from renewable energy) support the strategy of turning the combustion engine into a carbon neutral powertrain, which combines the advantage of a reliable and cost-effective technology with a drastic reduction of well-to-wheel CO2 emissions (figure 1).

Figure 1: To effectively decrease emissions, the development of future ICEs will be closely linked to the development of CO2 neutral fuels.

Moreover, e-fuels contribute to grid stability, as a prerequisite for increasing the ratio of renewable electricity generation (storage of energy sur-plus), which is more and more important for an increasing share of electric vehicles. Consequently, alternative fuels will play an undeniably important role in the future not only for passenger cars but also for many other industrial sectors. During the history of ICEs, very often Motorsports played a central role in the development of innovative solutions. Highly motivated by competition, the solutions can be tested, implemented and validated within minimal development time, thus establishing

47

Modelling of real fuels for an effective virtual engine development with focus on … Motorsports as a genuine innovation driver. However, in the last years the strategies adopted to increase engine performance remarkably changed. Due to different recent regulations – e.g. the introduction of an air restrictor in the World Rally Championship (WRC) or the limitation of the fuel consumption imposed in the World Endurance championship (WEC) or the Formula One World Championship (F1) – engine power can be raised only by increasing engine efficiency. Accordingly, the development targets in Motorsports have changed. Thus, new motorsport solutions may become increasingly interesting for mass production engines in the future, too [1-4]. Following this approach, in the last two years, Volkswagen Motorsport GmbH (VWMS), the Technical University of Munich (TUM) and the Research Institute of Automotive Engineering in Stuttgart (FKFS) developed, in a joint research project, an innovative combustion concept (Spark Assisted Combustion Ignition) derived from the WRC configuration. This application shows very high-performance levels while remarkably reducing fuel consumption. In this study, a fuel investigation is conducted on this innovative engine with the goal to better understand how fuel properties affect the combustion and to develop, through a virtual fuel DoE, a new fuel composition that allows to additionally improve engine performance. For a better investigation of both innovative fuels and complex combustion processes, it is necessary to have an accurate description of real fuel combustion characteristics, which in many cases cannot be ensured by the commonly used PRF/TRF (Primary Reference Fuel and Toluene Reference Fuel, respectively) surrogates. A detailed fuel description allows, first of all, a better and more reliable validation with experimental measurements at the test bench, especially when several fuel batches with similar compositions are intensively tested (wide-range validation). Moreover, such chemical calculations can be used to support fuel investigations by selecting those compositions that better fit the requirements of the considered engine application. In the first part of this work, the 3D-CFD models adopted in QuickSim and the approaches used for the implementation of fuel specific properties, like the resistance to auto-ignition, are described. In the following section, the results of the experimental fuel investigation are discussed. In conclusion, a virtual fuel investigation is conducted, and, through a fuel DoE, a new fuel composition is proposed and tested at the test bench.

3 QuickSim: 3D-CFD tool for a reliable virtual engine development A combination of quantitatively fewer, but more targeted and high-quality investigations on the test bench, with comprehensive numerical simulations offers various

48

Modelling of real fuels for an effective virtual engine development with focus on … advantages. Since the observation and evaluation of internal engine processes is very limited, supplementary 3D-CFD simulations enable a more detailed investigation of spatially and temporally resolved phenomena within the engine. Moreover, significantly more virtual engine variants can be analyzed and compared within the same development time and, in many cases, also before the prototyping phase, i.e. with a preselection approach.

Figure 2: 3D-CFD tool: QuickSim virtual engine development process.

49

Modelling of real fuels for an effective virtual engine development with focus on … The numerical investigations within the presented project were carried out with the 3D-CFD tool QuickSim, which has been developed and constantly enhanced over the years at the FKFS. Using the commercial software STAR-CD as a solver, QuickSim is a fast response 3D-CFD tool, which is specifically designed for the simulation of ICEs. By using ICE-adapted and improved computational models, coarser meshes compared to traditional 3D-CFD approaches can be used without sacrificing the quality of the simulation results. Hereby, the time expense for a simulation is highly minimized (3-8 h for an operating cycle of a full engine) [5]. This not only allows the extension of the simulation domain to a full engine, but also the calculation of several successive operating cycles (up to transients) in a reasonable time frame. As a result, a stable flow field can evolve (depending on the operating point, usually after 5-10 cycles) and the influence of user-defined initial and boundary conditions can be reduced significantly [6]. In detail, there are no limitations for what concerns: ● Engine layout: any cylinder number, comb. chamber geometry, intake and exhaust system geometry, and any injection system (DI, SPI, MPI, etc.) with arbitrary injector geometry can be realized. ● Ignition type: spark ignition, compression ignition, homogeneous charge compression ignition (HCCI) and spark-assisted compression ignition SACI. The auto-ignition HCCI model was calibrated using a highly variable free piston linear generator) [7-8]. ● Fuel type: gasoline, Diesel, compressed natural gas (CNG), bio-fuels and e-fuels. ● Valve and piston motion. A schematic representation of the virtual engine development process is reported in figure 2.

4 Fuel Modelling in QuickSim The combustion process of a common gasoline fuel involves more than 7000 chemical species. Due to such complexity, reduced chemical mechanisms are commonly implemented for the simulation of reactions. The combustion is then described by simplified reaction schemes which include only a limited number of relevant species. By using a sufficiently detailed mechanism (with more than 100 species), reliable results can be obtained but with an increase in CPU-time. A different approach is used in QuickSim where the working fluid is described by few scalars which, in combination with lookup tables, thermodynamically represent fresh charge, EGR (burned gas of the previous cycle) and burned gas. More information can be found in [5]. The modelling of all engine processes (in particular the combustion) was adapted and accordingly optimized for this formulation. Thermodynamic properties of the mixture, laminar flame speed

50

Modelling of real fuels for an effective virtual engine development with focus on … and ignition timings are imported from comprehensive databases and not calculated during the simulation, leading to profoundly shorter simulation durations [5]. Chemical kinetics calculations are performed using Cantera, a suite of object-oriented software tools for problems involving chemical kinetics, thermodynamics and transport processes. Due to the molecular complexity of common fuels, experimental and computational investigations of fuel reaction kinetic during combustion is virtually impossible for practical applications. Therefore, fuels are usually represented by surrogates of simpler molecular composition. Different methodologies can be used to determine surrogate composition. The most widely used surrogates for gasoline fuels are a mixture of n-heptane and isooctane, commonly called primary reference fuel (PRF), or a ternary mixture of PRF plus toluene (i.e. TRF). As later proved, experimental researches show that these simple compositions are suitable to reproduce with sufficient accuracy properties like the laminar flame speed of commercial fuels but they show limitations in knock prediction. These simple surrogates cannot accurately describe the remarkable influence of small concentrations of chemical species like olefins and oxygenates which are typically used to increase the knock resistance of the fuel. For these reasons, it has been decided to use more complex surrogates which include at least one species for each relevant chemical family present in the fuel. For what concerns the chemical kinetic mechanisms, the one developed at the Lawrence Livermore National Laboratory (LLNL) [9] is used since it has been experimentally validated for both auto-ignition time and laminar flame speed in wide ranges of equivalence ratio, temperature and pressure which are relevant for internal combustion engines.

4.1 Auto-ignition modelling Since in QuickSim no detailed chemical reaction is directly implemented, the occurrence of auto-ignition must be described by a dedicated model. Most of the models developed in the past 50 years are generally based on the evaluation of an integral representing the pre-reaction state of the unburnt mixture. The formulation reported in equation 1 was originally proposed by Livengood and Wu [10] and they proved that the ignition delay time to auto-ignition of an air-fuel mixture in motored and firing SI engines can be estimated by evaluating an integral representing the degree of chemical reaction progress and thus the pre-reaction state of the mixture. 1

1 𝑑𝑡 𝜏

(1)

51

Modelling of real fuels for an effective virtual engine development with focus on … The integral calculation showed in equation (1) represents the process of concentration of chain carriers building up until a pre-defined critical concentration has been reached and exceeded, resulting in an auto-ignition of the air-fuel mixture. The value of the pre-defined critical concentration is constant and thus independent on changes in the boundary conditions, e.g. an engine speed increase. In this equation, t is the elapsed time [s], 𝑡 is the time at the end of the integration [s] and τ is the ignition delay to auto-ignition of the mixture at the current boundary conditions [s]. Hence, if the ignition delay times τ are known at every integration step, it is possible to predict when auto-ignition of a mixture in a firing engine will occur (end of integration te). Ignition delay times τ can be either measured in a rapid compression machine or calculated by means of detailed kinetic reaction mechanisms. The integral, according to its original formulation, is commonly calculated taking into account the average mixture conditions in the cylinder. However, such an approach has two main drawbacks. First, auto ignition occurs due the formation of radicals which depend on local charge conditions and significant approximations must be made if an average cylinder temperature is considered for the calculation of the integral. Secondly, phenomena like knock in SI engines occur only if a certain mass ratio of mixture auto ignites. Therefore, it is also important to calculate the quantity of charge in auto-ignition conditions so that knock occurrence can be correctly detected. To overcome these disadvantages, a local integral counter is implemented in QuickSim and this is updated in every cell for each time step. In this way, it is possible to take into account fluid mixing and local conditions and to calculate location and quantity of charge in auto-ignition conditions.

Figure 3: Percentage of unburned mass in auto-ignition conditions for two engine configurations.

52

Figure 4: Percentage of unburned mass in auto-ignition conditions for two engine operating points.

Modelling of real fuels for an effective virtual engine development with focus on … A practical example of engine knock analysis is shown in figures 3 and 4. Figure 3 shows two different injection strategies of the same engine are compared with constant operating parameters (IMEP and rpm). It is shown that configuration B is more knockcritical due to its higher amount of unburned mass in auto-ignition conditions. Figure 4 however, shows the increase in unburned mass in auto-ignition conditions followed by an advance of the ignition point.

5 SACI Operating Strategy and Experimental Setup The SACI operating strategy relies on a lean mixture by means of injecting fuel during the early intake stroke into the combustion chamber [11]. The mixture is compressed to a level safely below the critical auto-ignition threshold. An external ignition source is used to start a propagating flame front which further compresses the unburnt mixture inside of the cylinder. This initiates a compression ignition by crossing the auto-ignition threshold (Figure 5). The auto-ignition threshold is dependent upon air-fuel mixture, fuel type and amount of residual gas.

Figure 5: Technical principle of SACI combustion Figure 6: Pre-chamber spark plug used in the strategy [12]. combustion system for SACI operation [12].

This specific application uses a passive pre-chamber spark plug featuring a number of orifice holes and a separated chamber to ignite the highly diluted mixture (Figure 6). The ignition inside the pre-chamber spark plug results in a combustion of the trapped mixture, sending hot gaseous jets, radicals, and pressure waves towards the main combustion chamber. This way, the pre-chamber amplifies the effects of the spark plug. This effect is sufficient to exceed the auto-ignition threshold of the main charge and a rapid, stable, and almost complete combustion process even at high dilution levels is set in motion. This operation mode leads to substantial improvements in terms of indicated efficiency up to very high load operating conditions [13].

53

Modelling of real fuels for an effective virtual engine development with focus on … The experimental investigations on the SACI combustion strategy were conducted on a single cylinder research engine located at the engine laboratory of the Chair of Internal Combustion Engines at the Technical University of Munich. The research engine was originally built to carry out development work with regard to injection and combustion optimization and component testing. It was derived from Volkswagen Motorsports proven 1.6l I4 DI-SI WRC race engine used between 2013 and 2016 in the FIA World Rally Championship (WRC). The specifications of its single cylinder derivate used in this study are shown in Table 1. Table 1 Technical specifications of the single cylinder research engine. Displaced volume

400 cm3

Stroke

73.8 mm

Bore

83 mm

Compression ratio

From 12.0:1 to 18:1

Number of valves

4

Max engine speed

8000 rpm

Injection

Direct with one multi-hole injector

Max boost pressure

4 bar abs.

Max eff. power

> 75hp

Indicated efficiency

> 45% (with SACI)

6 Experimental fuel investigation The goal of this investigation was to increase engine performance (higher IMEP by the same fuel consumption in kg/h) by optimizing fuel composition. The analysis can be split in two parts: a first learning phase in which few existing fuels are compared to better understand how the engine reacts to different fuel characteristics (in terms of resistance to auto-ignition, laminar flame speed and LHV) and a second phase in which a new virtually developed composition is created and tested. No composition can perfectly fit every engine type. For instance, a pure HCCI operating condition is used, it may be desirable to have a fuel with rapid auto-ignition characteristics to auto ignite while for high performance engines, fast combustion and high resistance to knock are requested. The application here studied is of particularly interest because differently from conventional SI engines, a relevant part of the released energy comes from self-ignition of the mixture.

54

Modelling of real fuels for an effective virtual engine development with focus on …

6.1 First analysis – fuels comparison at the test bench The analysis at the test bench was conducted with constant engine parameters and merely the ignition point was adapted in each case to maximize power output in a knock limited region. More details are reported in table 2. The mixtures tested in this study are representative for high performance race fuels (RON higher 100). The properties of these fuels are summarized in table 3 and all the reported values – due to confidentiality reasons –are normalized to those of Fuel 1. Information shown in this table are those typically found in the technical datasheets available at the test bench. At a quick glance, these fuels appear to be very similar as they have the same RON number and a LHV difference lower than 0.5%. Table 2 Engine operating parameters used for fuel comparison Engine Speed

6000 rpm

Fuel pressure

200 bar

Fuel consumption

Constant

𝑝 and 𝑝 Delta between intake and exhaust back pressure IP

Constant 0.2 bar

Lambda

Constant - 1.5 < λ < 2

Adapted according to knock limit

Table 3 Composition and main properties of the analysed fuels. Fuel 1 is considered as reference (values are normalized due to confidentiality reasons). n-Alkanes

Fuel 1 -

Fuel 2

Fuel 3

Fuel 4

Fuel 5

+0.0% +0%

+0.0% +3%

+0.0% +3%

iso-Alkanes

-

+0.0% +10%

cyclo-Alkanes

-

-8%

+2%

+9%

+2%

Olefins

-

+0%

+0%

-10%

-5%

Aromatics

-

-2%

-2%

-2%

+0%

RON

-

+0.0%

+0.0%

+0.0%

+0.0%

MON

-

+1.5%

+0.0%

+0.7%

+0.8%

LHV

-

+0.4%

+0.2%

+0.2%

+0.0%

55

Modelling of real fuels for an effective virtual engine development with focus on … However, the experimental tests showed a completely different reality. Considering the indicated efficiency and the IMEP, which are respectively shown in figure 7 and 8, differences up to 5% were obtained. Fuel 1 is the one that performed best, and, for this reason, it is here considered as reference. To understand the cause of these differences it is necessary to take into consideration some of the fuel properties that affect the combustion process. Figure 9 and 10 compare the values of laminar flame speed and auto ignition time calculated with Cantera and, for better understanding, also the respective values for a common E10 fuel with a RON number of 95 are reported. Considering that the quantity of energy introduced is the same, since the difference in LHV is very small (lower than 0.5%) and that the fuel consumption is constant, it can be noticed that the fuel that performed better is the one with the highest resistance to auto ignition. This may be explained considering that the charge is in critical conditions after ignition of the spark plug and in order to have a more controlled knock-free combustion, the fuel must have a very high resistance to auto ignition. It is also interesting to notice that fuel 4, which has the highest laminar flame speed – a property beneficial for reducing the combustion duration - but the lowest resistance to auto ignition, is the one that obtained the lowest IMEP.

Figure 7: Variation in Indicated efficiency compared to Fuel 1.

56

Figure 8: Variation in IMEP compared to Fuel 1.

Modelling of real fuels for an effective virtual engine development with focus on …

Figure 9: Variation in Laminar flame speed compared to Fuel 1. Values calculated @ 750 K, 50 bar, lambda 1.

Figure 10: Variation in Auto ignition time compared to Fuel 1. Values calculated at @ 700 K, 50 bar, lambda 1.

6.2 Optimization of fuel composition – Fuel DoE The idea behind the fuel DoE is to exploit the chemical calculations done with Cantera to find a fuel composition that better satisfies engine requirements (for this case IMEP increase at limited fuel flow rates). This fuel is then tested virtually by means of engine simulations performed with QuickSim and, if significant advantages are obtained, it is finally tested at the test bench. Table 4 Composition and main properties of Fuel 6 (as the most promising compromise among 2000 virtual fuels tested) compared to Fuel 1.

n-Alkanes

Fuel 6 +0.00%

iso-Alkanes

-4.97%

cyclo-Alkanes

+0.33%

Olefins

+0.64%

Aromatics

+4.00%

RON

+0.0%

MON

-0.6%

LHV

-0.5%

57

Modelling of real fuels for an effective virtual engine development with focus on …

Figure 11: Variation in Laminar flame speed compared to Fuel 1. Values calculated @ 750 K, 50 bar, lambda 1.

Figure 12: Variation in Auto ignition time compared to Fuel 1. Values calculated at @ 700 K, 50 bar, lambda 1.

The Virtual DoE is performed by changing the mass fraction of the chemical species within predefined limits (that for example may come from technical regulations or supplier requirements). Afterwards, Cantera is used to calculate laminar flame speed and auto-ignition time of each composition. This process is also relatively fast as it gives the possibility to analyze more than 500 fuel compositions in one day using a common CPU. The results of these calculations are then filtered according to the desired constrains. In this case, the main limiting parameter was the LHV since the analysis at the test bench is based on a constant fuel consumption. Therefore, the goal of the DoE was to find a composition that would increase the resistance to auto ignition while keeping the LHV as high as possible. The composition and the properties of the chosen fuel (which is named Fuel 6) are reported in table 4 and they are expressed as variation compared to Fuel 1. This fuel was the most promising compromise among 2000 fuels tested. The most important differences in Fuel 6 are in the quantities of alkanes and aromatics. As is commonly known, aromatics have a good resistance to auto-ignition but low LHV and flame speed compared to alkanes and therefore a compromise is needed. As reported in figure 11 and 12, the new composition shows a significant increase in auto ignition resistance while keeping similar flame speed and LHV.

58

Modelling of real fuels for an effective virtual engine development with focus on …

Figure 13: Cylinder pressure obtained with fuel 1 and 6 for similar IMEP target.

Figure 14: Percentage of unburned mass in auto-ignition conditions obtained with fuel 1 and 6 for similar target IMEP.

Figure 15: Cylinder pressure obtained with fuel 1 and 6 at maximum IMEP under stable operating conditions.

Figure 16: Percentage of unburned mass in auto-ignition conditions obtained with fuel 1 and 6 at maximum IMEP under stable operating conditions.

59

Modelling of real fuels for an effective virtual engine development with focus on …

6.3 Fuel test through engine simulation QuickSim 3D-CFD-simulations of the engine operation confirm the expectations found with the virtual reactor in Cantera and the fuel selection criteria. Fuel 6, due to its higher knock resistance, gives a more stable combustion with consequently higher IMEP potential. Comparing the results of the simulations obtained at a similar cylinder pressure profile and IMEP (figure 13), fuel 6 shows a lower amount of unburned mass in autoignition conditions compared to fuel 1 (figure 14). Slightly higher combustion duration of Fuel 6 is to be expected given the lower flame laminar speed. This advantage in combustion stability and auto-ignition resistance can be then exploited when, for the same operating point, the maximum IMEP is reached as reported in figure 15 and 16. The results showed that with the new fuel composition it is possible to obtain a higher cylinder pressure while still having a lower amount of unburned mass in auto-ignition conditions.

6.4 Final validation at the test bench After this first virtual study, the new fuel was prepared and tested at the test bench. The same operating conditions previously introduced were used. Results are reported in figure 15 and 16. Compared to fuel 1, fuel 6 showed a significant increase in IMEP (optimization target) between 2.3% and 3% among different engine speeds. Even the indicated efficiency increased as well but to lower extent: a peak increase of 2.5% is obtained at 5000 and 7000 rpm while no significant difference was found at 4000 rpm.

Figure 17: Average variation in IMEP of fuel 6 compared to fuel 1.

60

Figure 18: Average variation in indicated efficiency of fuel 6 compared to fuel 1.

Modelling of real fuels for an effective virtual engine development with focus on …

7 Summary and Outlook The aim of this work was to create a detailed fuel modelling and to implement, in the 3D-CFD simulation tool QuickSim, a series of models that could support the increasing research activity on alternative fuels. It is also demonstrated that a detailed fuel description can, other than improve simulation accuracy, also be used to generate a virtual fuel DoE which may support the research of the fuel composition that better fits engine requirements. The investigation discussed in this study focuses on improving fuel composition to maximize power output of a high-performance single cylinder research engine operating with the innovative SACI (Spark Assisted Compression Ignition) combustion strategy. In the first part of the analysis, different high-performance fuels were tested to better understand engine behaviour and, even though the combustion is based on a partial auto-ignition of the charge, it appears that a more stable operation and higher IMEP are obtained with a fuel with lower tendency to auto-ignite. According to these findings, a virtual fuel DoE was used to search for a chemical composition that would increase resistance to auto-ignite without penalizing significantly laminar flame speed and LHV. The proposed fuel composition was first tested virtually with an engine simulation carried out in QuickSim and then at the test bench where increases of efficiency and IMEP higher than 2% were obtained compared to the reference case. With regard to the research on this engine, further work will be carried out to improve the interaction among combustion models in QuickSim with the goal to better reproduce and understand the SACI operation. This activity should also include an optical analysis at the test bench which would give a more detailed insight into the mixture formation and combustion processes.

Bibliography 1

Wentsch M., Chiodi M., Bargende M., Poetsch C. et al.: "Virtuelle Motorentwicklung als Erfolgsfaktor in der F.I.A. Rallye-Weltmeisterschaft (WRC)", 12th International Symposium on Combustion Diagnostics, Baden-Baden – Germany, May 2016.

2

Koch D., Wachtmeister G., Wentsch M., Chiodi M., Poetsch C., et al., "Investigation of the Mixture Formation Process with Combined Injection Strategies in High-Performance SI-Engines", 16th Stuttgart Symposium, Stuttgart – Germany. March 2016.

3

Chiodi M., Perrone A., Roberti P., Bargende M., Ferrari A., Wichelhaus D., "3D-CFD Virtual Engine Test Bench of a 1.6 Liter turbo-charged GDI-Race-Engine with Focus on Fuel Injection", SAENA Conference, Capri – Italy, SAE Technical Paper 2013-24-0149, September 2013.

61

Modelling of real fuels for an effective virtual engine development with focus on … 4

Wentsch M., Perrone A., Chiodi M., Bargende M. et al., "Enhanced Investigations of High-Performance SI-Engines by Means of 3D-CFD Simulations", SAE Technical Paper 2015-24-2469, 2015, doi:10.4271/2015-24-2469.

5

Chiodi, M., "An innovative 3D-CFD-Approach towards Virtual Development of Internal Combustion Engines", Dissertation (Ph.D. Thesis), Universität Stuttgart, Vieweg+Teubner Research Verlag, June 2010.

6

Chiodi M., Kaechele A., Bargende M., Koch D., Wachtmeister G., Wichelhaus, D., "Development of an innovative combustion process: Spark-assisted compression ignition", In: Bargende M., Reuss HC., Wiedemann J., 18. Internationales Stuttgarter Symposium, doi.org/10.1007/978-3-658-21194-3_22

7

Schneider S., Chiodi M., Friedrich H. and Bargende M., "Development and Experimental Investigation of a Two-Stroke Opposed-Piston Free-Piston Engine", SAE Technical Paper 2016-32-0046, 2016, doi:10.4271/2016-32-0046.

8

Schneider S., Chiodi M., Friedrich H. and Bargende M., "Analysis of SI and HCCI Combustion in a Two Stroke Opposed -Piston Free-Piston Engine. Small Engine" Technology Conference (SETC), Jakarta – Indonesia. SAE Technical Paper 201732-0037, 2017.

9

Mehl M., W.J. Pitz, C.K. Westbrook, H.J. Curran, "Kinetic modeling of gasoline surrogate components and mixtures under engine conditions", Proceedings of the Combustion Institute 33:193-200 (2011).

10 Livengood, J. C. and Wu, P. C., "Correlation of autoignition phenomena in internal combustion engines and rapid compression machines," Symp. Int. Combust, 1955, 5: 347–356. 11 Reuss, D., Kuo, T-W., Silvas, G., Natarajan, V., Sick, V., “Experimental metrics for identifying origins of combustion variability during spark-assisted compression ignition,” International Journal of Engine Research, Vol 9, Issue 5, pp. 409 – 434, First Published October 21, 2008, doi.org/10.1243/14680874JER01108 12 Koch, D., Berger, V., Bittel, A., Gschwandtner, M. et al., "Investigation of an Innovative Combustion Process for High-Performance Engines and its Impact on Emissions," SAE Technical Paper 2019-01-0039, 2019, https://doi.org/ 10.4271/2019-01-0039 13 Chiodi, M., Kaechele, A., Bargende, M., Wichelhaus, D. et al., "Development of an Innovative Combustion Process: Spark-Assisted Compression Ignition," SAE Int. J. Engines 10(5):2486-2499, 2017, doi.org/10.4271/2017-24-0147

62

Modelling of real fuels for an effective virtual engine development with focus on …

Abbreviations CFD

Computational Fluid Dynamics

DI

Direct Injection

DI-SI

Direct-Injected Spark-Ignited

DoE

Design of Experiments

FKFS

Forschungsinstitut für Kraftfahrwesen und Fahrzeugmotoren Stuttgart

FTDC

Firing Top Dead Center

HCCI

Homogeneous Charge Compression Ignition

ICE

Internal Combustion Engine

IMEP

Indicated Mean Effective Pressure

LHV

Lower Heating Value

LLNL

Lawrence Livermore National Laboratory

MON

Motor Octane Number

MPI

Manifold Port Injection

PRF

Primary Reference Fuel

RON

Research Octane Number

SACI

Spark Assisted Compression Ignition

SPI

Single-Point Injection

TRF

Toluene Reference Fuel

TUM

Technical University of Munich

WEC

World Endurance Championship

63

Sustainable drive concepts for future motorsports Lea Schwarz University of Stuttgart Prof. Michael Bargende FKFS Stefan Dreyer, Ulrich Baretzky, Wolfgang Kotauschek, Dr. Florian Bach AUDI AG

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 M. Bargende et al. (Hrsg.), 19. Internationales Stuttgarter Symposium, Proceedings, https://doi.org/10.1007/978-3-658-25939-6_7

65

Sustainable drive concepts for future motorsports

1 Abstract Global warming caused by greenhouse gas emissions is one of the major threats to humankind and challenges the automobile industry. Motorsport as a pioneer and a laboratory for automotive technologies has to face this issue and push the development of solutions. Nowadays motorsport is dominated by conventional propulsion systems powered by fossil fuels, predominantly by gasoline. First steps towards sustainability are encouraged by the introduction of regulations with a prime focus on efficiency realized via energy limitation as in FIA Formula One World Championship and the FIA World Endurance Championship. Moreover, alternative concepts as in FIA Formula E or the possible future FIA E-WRX and an intended hydrogen class at the 24 hours of Le Mans set a focus on alternative concepts. However, sustainability is not part of any motorsport regulations yet. There is no holistic evaluation of the most sustainable way of racing at an adequate performance to sustain the core of motorsports and simultaneously advance sustainable technologies. Thus, this study investigates sustainable highperformance concepts for endurance racing. The influence of various propulsion concepts and fuels on lap time performance is analyzed and extended by an ecological evaluation of the environmental impact via life cycle assessment. This innovative combination of sustainability and motorsports results in the methodical selection of a sustainable high-performance concept for endurance racing.

2 Introduction Global Warming resulting in climate change is considered as one of the major threats to humankind since the consequent damage to the environment and natural disasters are challenging people all over the world. The anthropogenic impact is caused by greenhouse gas emissions. Thus sustainability, sustainable development and, above all, the reduction of greenhouse gas emissions gained in importance during the last decades. International agreements as the Brundtland-Report “Our Common Future” (1987), the Report from the United Nations Sustainable Development Summit “Transforming Our World: the 2030 Agenda for Sustainable Development” (2015) or the Commitments by the United Nations Framework Convention on Climate Change ratified by 197 parties in Paris (2015) indicate the significance and the increasing urgency of this issue [4, 5, 6, 7]. Figure 1 shows the contribution of different sectors to the overall greenhouse gas emissions in Germany in 1990 and 2016. The total emissions have been reduced by 27 % over this period. In 2016 the transport sector was responsible for 18 % of the greenhouse gas emissions. This does not only indicate a 5 % higher share compared to 1990 but also the total emissions caused by transportation increased by 1,5 %, not least due to the increasing annual mileage and the demand for individual mobility.

66

Sustainable drive concepts for future motorsports

1990

6% 3% 8%

7%

2016

7% 1%

1%

3%

34% 18% 13% 15%

15%

37%

18% 14%

Energy Industries

Manufacturing Industries & Construction

Transport

Other Sectors **

Fugitive Emissions

Industrial Processes

Agriculture

Waste & Waste Water

Figure 1 – Greenhouse gas emissions in Germany [1]

The significant impact of the transport sector on greenhouse gas emissions and thus on climate change lead to strict emission legislation worldwide, especially in China, Europe, Japan and the United States of America. These regulations are constantly getting stricter increasing pressure on the automotive industry and leading towards a major focus on emissions within the development. As a result to this, two major topics characterize the automotive development activities at the moment. First, the introduction of alternative drive concepts as well as renewable fuels and second the focus on sustainability particularly via the reduction of tank-to-wheel emissions as these are regulated by law. Although legislation still focuses on local emissions at the exhaust pipe, the awareness for the global environmental impact gains in importance.

2.1 The DNA of Motorsports Motorsports as a pioneer regarding technologies and with the target to develop from “track to road” has to face the same problems as the development of production cars in order to keep its leading role for road relevant technologies [2, 3]. Nowadays motorsports is predominantly characterized by internal combustion engines and fossil fuels. The introduction of new drive concepts is already focused as can be seen by the hybrid systems within the FIA Formula One World Championship (Formula One) and the FIA World Endurance Championship (WEC) as well as by the FIA Formula E [8, 9, 10]. Moreover, alternative fuels have been introduced into a variety of racing series including Formula One and WEC [3]. These two series have an obligatory percentage of biological or sustainable fuel mixed into the fossil basis and their regulations focus on efficiency, resulting in the fact that the most efficient car is most likely to win [8, 9].

67

Sustainable drive concepts for future motorsports This shows that there is already a focus on alternative concepts and efficiency, but sustainability is not part of the regulations and the development targets within motorsports yet. Consequently, to keep the pioneering role for automotive development and to go one step further, a focus on global sustainability within motorsports is needed. In order to indicate the identity of motorsports it is analyzed from different perspectives. The motorsport world itself regards its role as caring, innovative, passionate, sustainable “global voice for a world in motion” [12]. The identity of the ideal motorsport from a fan perspective is shown in Figure 2 [11]. The most important characteristics for a motorsport fan are being close, to offer a lot of action and to be leading concerning technology. Less important but still relevant are the aspects of glamour, risk and sustainability. Leading Action

21%

27% The ideal Motorsport

Fan-sport

5,5%

Sustainability

4% Risk 4% Posh

38,5%

Figure 2 – The ideal Motorsport from the point of view of a fan [11]

Being close to the fan is defined by the concepts at the racing tracks and in the paddocks. The risk has rather to be reduced than to be encouraged and the glamour of racing series can hardly be influenced. As a result the focus of motorsports engineering has to be on action in the form of being loud, fast, unique and emotional, on innovations in order to be leading and technological progressive and on sustainability via resource-saving and eco-friendliness [11]. With a focus on the acceptance by society and the demand for being a pioneer by developing form track to road it is evident, that in the future the characteristic of being leading and innovative will be dominated by sustainability. This results in an increasing importance of sustainability also from a fan point of view. To sum up, the identity of future motorsports is characterized by being close and emotionally attractive to the fans as well as by gaining the acceptance and the pioneering role by realizing leading and sustainable innovations.

68

Sustainable drive concepts for future motorsports

3 Sustainability and Life Cycle Assessment The current legislation with regards to sustainability of vehicles is restricted to a limitation of exhaust pipe emissions [13, 14]. This tank-to-wheel approach neglects upstream and downstream emissions and focusses on a small extract of the product life cycle [15]. Moreover, it promotes locally emission-free vehicle concepts that might lead to a shift of emissions from the use phase to upstream processes. In order to confront a global challenge as the climate change, a holistic approach is required in order to reduce greenhouse gas emissions. This demands for an extension of the system boundaries from a tank-to-wheel analysis to a consideration of the whole life cycle. As demonstrated in Figure 3 this life cycle assessment analyses all environmental impacts throughout the life cycle of a product including, for example, the raw material extraction, production and transport regarding both, the fuel and the vehicle itself. The methodology and requirements concerning the realization of a life cycle assessment are defined in DIN ISO 14040 and set a primary focus on transparency and completeness [15, 16, 17, 18]. Cradle to Grave Raw material extraction

Life Cycle Assessment (LCA)

* Production Well-to-Wheel Raw material extraction

Tank-to-Wheel * *

Production

Well-to-Tank * Transport

*

Fuel station

Use phase * End-of-life/ Recycling

Figure 3 – System Boundaries and Life Cycle Assessment [15, 19]

If different drive concepts are compared with regards to their environmental impact, the definition of adequate system boundaries is highly significant. The tank-to-wheel analysis considers electric drive concepts as emission free. Whereas well-to-wheel and life cycle assessments indicate emissions during upstream and downstream processes that are also relevant for electric vehicles. Figure 4 shows the impact of system boundaries on the evaluation of greenhouse gas emissions for different drive concepts. The data is based on calculations by AUDI AG. The analysis are carried out for a compact car with a lifetime mileage of 200.000 km. For this “fossil” scenario the internal combustion

69

Sustainable drive concepts for future motorsports engines (ICE) are powered by fossil fuel and the battery electric vehicle (BEV) is charged with a European grid mix. The hydrogen for the fuel cell electric vehicle (FCEV) is produced by steam methane reforming. The Cradle-to-Grave emissions focus on the raw material extraction and the production of the vehicle.

Production vehicle - 200.000 km mileage Greenhouse gas emissions [gCO2-eq/km]

Cradle-to-Grave

Well-to-Tank "fossil"

Tank-to-Wheel "fossil"

180 150 120 90 60 30 0

ICE diesel ICE gasoline ICE methane *European grid mix

BEV*

FCEV

Figure 4 – Life Cycle Analysis of production vehicles with various propulsion systems

In addition to the system boundaries the assumptions, on which the calculations are based, are significant. Two examples for assumptions are the reference area, with a significant impact on the greenhouse gas intensity of the grid mix, and the lifetime mileage of the vehicle. For battery electric vehicles the lifetime mileage is closely linked to the battery lifetime as not only cyclical but also calendrical aging has to be considered. Figure 5 shows the analysis of the same vehicle concepts as shown in Figure 4 with a German grid mix considered for the recharging of the battery electric vehicle and with a drivetrain specific mileage. For vehicles with internal combustion engines this is slightly higher than the basis with 240.000 km for diesel and gaseous fueled cars and 210.000 km for cars with a gasoline engine [15, 20]. If a battery lifetime of 12 years and an annual mileage typical for battery electric vehicles are assumed, the lifetime mileage of a battery electric vehicle accounts for 138.800 km [15]. Regarding fuel cell electric vehicles, there is no data for the calculation of an average lifetime mileage available until now. Thus the mileage of 200.000 km is furthermore assumed.

70

Sustainable drive concepts for future motorsports

Production vehicle – specified lifetime mileage Greenhouse gas emissions [gCO2-eq/km]

Cradle-to-Grave

Well-to-Tank "fossil"

Tank-to-Wheel "fossil"

180 150 120 90 60 30 0

ICE diesel ICE gasoline ICE methane *German grid mix, battery lifetime 12 years

BEV*

FCEV

Figure 5 – Life Cycle Analysis of production vehicles with various propulsion systems and drivetrain specific mileages (except FCEV)

Another strong influence is achieved by the “green” scenario. Renewable fuels and renewable energy offer a huge potential to reduce the greenhouse gas emissions throughout the whole life cycle regarding all drive concepts. Compared to the fossil basis the reduction potential is up to 65 %. Figure 6 shows the assessments with exemplary renewable fuels, wind energy for the battery electric vehicle and hydrogen produced by electrolysis.

Figure 6 – Life Cycle Analysis of production vehicles with various propulsion systems using renewable fuels or electricity

71

Sustainable drive concepts for future motorsports The analysis show that specified assumptions as for the mileage have a significant impact on the life cycle assessment. This has to be taken into account when the certain use case of a race car is analyzed. Moreover, all drive concepts offer a significant greenhouse gas reduction potential if a renewable fuel or renewable energy is used. In consequence, with a focus on the whole life cycle all presented drive concepts have the potential for a sustainable solution.

4 Sustainable drive concepts for endurance racing Within the following section four potentially sustainable drive concepts are analyzed for the use within endurance racing, focusing on the Le Mans Prototype 1 with a hybrid system (LMP1-H) that is classified within the WEC and thus within the 24 hours of Le Mans. This racing series is characterized by a focus on technologies and regulations that encourage innovations and a variety of drive concepts within one class [9].

4.1 Internal combustion engine with liquid fuels The typical drive concept within the LMP1-H nowadays is a hybrid car with a highly efficient internal combustion engine and a hybrid system [9]. Until the season 2016 Audi was competing with a diesel engine, whereas the cars of Porsche and Toyota were powered by gasoline engines. The architecture of the drivetrain is shown in Figure 7.

Figure 7 – Internal combustion engine with hybrid system and liquid fuel

A sustainable concept with an internal combustion engine requires a renewable fuel and a high efficient engine. The regulations of the WEC and of the Formula One as well as of the Deutsche Tourenwagen Masters (DTM) are already focusing on efficiency by

72

Sustainable drive concepts for future motorsports limiting one or more of the following criteria, the maximum fuel mass flow, the amount of fuel per race or the amount of energy per lap regarding both the internal combustion engine and the hybrid system [8, 9]. Concerning the use of sustainable fuels, the Formula One demands for 5,75 % (m/m) of bio-fuel since 2008 and within the WEC diesel fuels contain 10 vol.-% bio-fuel and 20 vol.-% of the gasoline requires a bio basis [8, 9, 2, 3]. This regulations encouraged the development and use of various bio-components as cellulosic ethanol, gas-to-liquid (GtL), biomass-to-liquid (BtL) and Virent® bio-gasoline [2]. Moreover, other racing series as the VLN Endurance Championship Nürburgring, the IndyCar Series and the Formula Student encourage the use of alternative fuels [3, 21, 22, 23, 24]. The current state of technology offers a broad variety of renewable fuels as alternatives to fossil gasoline and diesel fuels. With regards to their feedstock they are divided into synthetic fuels and biological fuels. The latter are further divided into first, second and third generation of biological fuels. Table 1 gives an overview of these fuel types, their feedstock and exemplary fuels. All of these fuels are characterized by a closed carbon dioxide loop. The amount of carbon dioxide emitted by burning the fuel within the engine is the exact amount of carbon dioxide that is used to produce the fuel. Table 1 – Alternative and renewable fuels [3] Type 1. Generation biofuels

Examples for feedstock Rapeseed, corn, sugar cane, wheat, sunflower, … Waste/ residue from agriculture/forestry, waste cooking oil, organic waste, …

Examples for fuels Ethanol, fatty acids methyl esters* (FAME or Biodiesel) Cellulosic ethanol, hydrated vegetable oil* (HVO), paraffinic fuels* (BtL)

3. Generation biofuels

Algal biomass, algae oil, …

Ethanol, FAME*, HVO*

Synthetic fuels (e-fuels)

Carbon dioxide via direct air capturing, biogas plants, …

Synthetic gasoline, paraffinic fuels* (Power-to-liquid), oxymethylene ether* (OME), isooctane

2. Generation biofuels

* Diesel

fuel

First generation biofuels are critical and are not further investigated within this study as they are based on energy crops and compete with food. Moreover they are blamed to promote monocultures, deforestation as well as land use change [3, 25, 26]. Within the following, three type of gasoline like fuels and three types of diesel like fuels are investigated regarding their characteristics and their greenhouse gas reduction potential compared to the fossil basis. As alternative to fossil gasoline synthetic gasoline,

73

Sustainable drive concepts for future motorsports cellulosic ethanol and synthetic isooctane are analyzed. As shown in Table 2 these fuels offer a potential to reduce the greenhouse gas (GHG) emissions per MJ up to 88 %. Table 2 – Characteristics of fossil gasoline (Super) and sustainable alternatives [3, 27, 28] Fossil gasoline Lower calorific value [MJ/kg] 40,1-41,8 Stoichiometric air-to-fuel ratio [-] 14 Mixture calorific value [MJ/kg] 2,86-2,99 Density (ρ) [kg/m³] 720-775 Octane number (RON) [-] > 95 Heat of vaporization [kJ/kg] 420 GHG-reduction per MJ 0% Characteristic

Ethanol 26,26 8,98 2,92 790,35 109,6 879,6 - 59 %

Synthetic gasoline 42 14,7 2,86 755 94 420 - 88 %

Isooctane 44,51 14,99 2,97 694,98 100 310,2 - 78 %

Within the WEC the efficiency of the engine has a significant impact on the lap time performance [3]. Based on its characteristics an alternative fuel might offer a potential on the one hand to further reduce the environmental impact due to an increased efficiency and on the other hand to improve the performance. However, the impact of a reduced lower calorific value and thus an increased fuel mass and the subsequent impact on overall weight and lap time as well as further correlations have to be considered [3]. With regards to the replacement of fossil diesel the following investigations focus on paraffinic diesel as HVO or PtL, FAME based on algal biomass and OME. Table 3 shows the characteristics of these fuels as well as the greenhouse gas reduction potential. In particular, paraffinic fuels with a high lower calorific value, a high cetane number and a high greenhouse gas reduction potential offer a potential to be advantageous. Table 3 – Characteristics of fossil diesel (B7) and sustainable alternatives [3, 29] Characteristic Lower calorific value [MJ/kg] Stoichiometric air-to-fuel ratio [-] Mixture calorific value [MJ/kg] Density (ρ) [kg/m³] Cetane number (CN) [-] GHG-reduction per MJ

74

Fossil diesel 43 14,4 2,99 820-845 > 55 0%

Paraffinic fuels 44,13 14,82 2,98 772,52 75 - 85 %

OME

FAME

19,64 6,07 3,23 1.045,95 73 - 69 %

37,57 12,48 3,01 882,47 52 - 58 %

Sustainable drive concepts for future motorsports

4.2 Internal combustion engine with gaseous fuels A gasoline powered engine can also operate on gaseous fuels as methane or hydrogen. In the past there have been few race car concepts using bio methane, as for example the Volkswagen Scirocco R-Cup that has been carried out from 2010 until 2014 or the Welter Racing concept that was planned 2017 for Garage 56 in Le Mans [31, 32]. The vehicle concept of a methane powered car is characterized by a high efficient internal combustion engine and a hybrid system as shown in Figure 8. The gaseous fuel is stored in pressure tanks made out of a plastic liner fully wrapped by composite material [30].

Figure 8 – Internal combustion engine with hybrid system and gaseous fuel

As methane is used in gasoline engines, the characteristics of the gaseous fuel are compared to the characteristics of a fossil gasoline in Table 4 in order to discuss the potentials regarding efficiency and lap time performance. Table 4 – Characteristics of fossil gasoline (Super) and methane [3, 27, 28] Characteristic Lower calorific value (LCV) [MJ/kg] Stoichiometric air-to-fuel ratio (AFR) [-] Mixture calorific value (LCV/AFR) [MJ/kg] Density (ρ) [kg/m³] at 1,013 bar and 25 °C Density (ρ) [kg/m³] at 200 bar and 25 °C Density (ρ) [kg/m³] at 700 bar and 25 °C Octane number (RON) [-] Heat of vaporization [kJ/kg]

Fossil gasoline 40,1-41,8 14 2,86-2,99 720-775 > 95 420

Methane 49,73 17,20 2,89 0,72 157,7 257,96 130 -

75

Sustainable drive concepts for future motorsports The high lower calorific value of methane is compensated by the higher stoichiometric air-to-fuel ratio. However, methane has a high octane number and subsequently a reduced knock tendency allowing to increase the compression ratio and thus the thermal efficiency. The low density of methane, even at high pressures, challenges the packaging of a race car concept. Compared to average gasoline values, the volume per mega joule is 2,4 times higher for methane at a pressure of 700 bar. The volume and also weight of the fuel storage is further increased by the required cylindrical pressure tanks. Sustainable methane is either produced biologically based on second generation biomass or synthetically out of carbon dioxide and hydrogen via methanation of synthetic gas [3]. Both production routes offer a high greenhouse gas production potential. Synthetic methane reduces the greenhouse gas emissions by 82 % per mega joule compared to fossil compressed natural gas (CNG) and bio methane by 81 % [A3].

4.3 Battery electric vehicle In 2014 the FIA Formula E started as first battery electric racing series. However, the concept is limited by the characteristics of the battery, in particular due to the low energy density and the tradeoff between energy and power density. Focusing on the season of 2018/ 2019 the maximum power of the drivetrain is 250 kW during qualifying and 225 kW during race in attack mode. The useable amount of energy is 52 kWh for a race of 45 min and only the battery cells account for 31 % of the vehicle weight [10]. These conceptual challenges gain in significance for an endurance race car concept shown in Figure 9 as the current LMP1-H cars have up to 750 kW of peak power and the long distances require refueling or respectively recharging. Considering the drivetrain efficiencies and the characteristics of highly developed battery technologies 11 laps at Le Mans still require more than one ton of battery mass and about half an hour of charging with 350 kW. Furthermore, the increased weight reduces the achievable lap time and challenges the structure and tires of the car.

76

Sustainable drive concepts for future motorsports

Figure 9 – Battery electric drivetrain

4.4 Fuel cell electric vehicle Within motorsports fuel cell electric vehicles are not common as there is no racing series for this concept yet. However, there are concept cars as the Green GT H2 and the LMPH2G by Green GT and the Forze H2 of the Technical University of Delft [33, 34]. A fuel cell electric race car has to face challenges regarding the vehicle concept, particularly the cooling of the fuel cell and the hydrogen storage, and the race concept, especially regarding the refueling time and safety requirements. Figure 10 shows the architecture of a fuel cell electric drivetrain with an electric motor, including power electronics, gearbox and differential, at the front and at the rear axis to benefit from the high potential to recuperate at the front axis while maintaining the vehicle dynamics of a race car. For the recuperation a high voltage battery is needed and the integration of the fuel cell as second voltage source requires a DC/DC converter.

Figure 10 – Fuel cell electric drivetrain

77

Sustainable drive concepts for future motorsports The LMPH2G has a total mass of 1.420 kg plus 8,6 kg of hydrogen compared to 875 kg of the LMP1-H cars with an additional weight of 35,2 kg of fuel in 2018 [9, 34]. Its fuel cell has a power of 250 kW and the peak power of the car is 480 kW [34]. The high mass and the low power reveal the challenges influenced by the storage and the cooling system. If, based on the efficiency chain, a comparable powertrain efficiency is assumed 10,7 kg of hydrogen are required to run 11 laps with 116,9 MJ per lap in Le Mans [9]. The respective storage based on common energy densities has a volume of 397 l and a mass of 198 kg resulting in an extreme increase of mass and volume of the fuel storage compared to the conventional concept with liquid fuels [36]. High efficient engines within motorsports have efficiencies up to 50 % (Formula One). Most of the energy that is not converted into power output is transported out of the engine via exhaust gas. As for fuel cells, which are characterized by a comparable efficiency, the exhaust gas does not transport much energy out of the system, almost 50 % of the fuel energy is absorbed by the coolant requiring a significantly higher cooling power. The cooling power, limited by the boundary conditions and physical correlations, consequently limits the power of the fuel cell. The refueling of the LMP1-H cars lasts around 30 s and has to be done more than 30 times during the 24 hours of Le Mans. Refueling 10,7 kg of hydrogen takes 4 to 4,5 min [34, 35]. This difference regarding the refueling time has a significant impact on the race concept if both drivetrains run within one race. The hydrogen for the fuel cell is either produced via steam methane reforming, the predominantly used but fossil way, or via electrolysis, that can be realized onsite the fuel station or at a central plant. The well-to-wheel emissions of one mega joule of hydrogen produced via electrolysis are about 80 % lower than for hydrogen produced via steam methane reforming. However, this requires renewable energy as the electrolysis has a high demand for energy resulting in a high sensitivity of the greenhouse gas emissions with regards to the grid mix.

5 Life Cycle Assessment of drive concepts for endurance racing The previous section shows that there are various drivetrain concepts with the potential to be sustainable if a well-to-wheel or a life cycle perspective is applied. In the following the life cycle assessment of drive concepts for endurance racing is analyzed. A difference between race cars and production vehicles with a significant impact on the environmental analysis is the mileage. Figure 11 shows the relative greenhouse gas emissions of the production phase and the use phase for two drive concepts, a hybrid powertrain with a gasoline engine and a battery electric vehicle, for a production and a race car. The production car has 200.000 km of mileage and the race car has a mileage of 23.000 km for the basic car. The drivetrain is changed after a distance of 5.500 km

78

Sustainable drive concepts for future motorsports

Share of GHG-emissions

representing the race at Le Mans. It is shown that for race cars the emissions during the production phase gain in importance compared to the emission during the use phase, particularly when a sustainable fuel or renewable energy is used. 100%

100%

80%

80%

60%

60%

40%

40%

20%

20%

0%

0% ICE BEV ICE BEV ICE BEV ICE BEV gasoline gasoline* gasoline gasoline* (10MJ) Production car (10MJ) Race car Race car Production car *10MJ Hybrid use-phase "green" production use-phaseproduction "fossil" Figure 11 – Comparison of greenhouse gas emissions caused by use-phase and production for Production cars and LMP1-H concepts (ICE Hybrid and BEV)

Due to this shift from use phase to production phase, the materials and processes used to produce the car are of great significance. Figure 12 indicates that 29 % of the production emissions of the gasoline hybrid concept are caused by the production of the hybrid system and 63 % by the basic car, predominantly caused by the use of carbon fiber. A focus on sustainable or less carbon dioxide intensive materials has the potential to reduce these emissions significantly.

0%

20% Hybrid system

40% 60% ICE, gearbox, tank

80% Rest of the car

100%

Figure 12 – Distribution of the emissions caused during the production of a LMP1-H car with a gasoline ICE and a 10 MJ hybrid system

In order to reduce the greenhouse gas emissions during the use phase of the race car, two effective measures are identified. The first one is the use of alternative fuels and renewable energy. In relation to the use of fossil fuels and of energy grid mix the life cycle emissions are reduced by 29 % for the vehicle with a hybrid system and a gasoline engine and by 17 % for the battery electric LMP1 concept. The second measure is the

79

Sustainable drive concepts for future motorsports focus of the regulations on efficiency as within the WEC, the Formula One and the DTM starting in 2019.

6 Vision of a sustainable motorsport As shown in section 0, there are various potentially sustainable drive concepts including the used fuel or energy. Moreover, some steps towards sustainability are already implemented into motorsports as the use of sustainable fuels or fuel components or regulations focusing on efficiency. Two examples for this are the Formula One and the WEC. To realize the vision of a sustainable motorsport, it is essential to identify the hot spots of greenhouse gas emissions and derive effective measures based on these main emission sources. Due to a lower mileage the emissions caused during the production of a race car gain in importance. Thus the focus has to be on less carbon dioxide intensive materials and processes. In particular concerning hybrid systems, it has to be aimed at a balance between the increased emissions during the production and the benefits regarding recuperated energy as well as the influence on lap time during the race. Furthermore, the mileage is increased by a restriction of the number of drivetrains per season. With regards to the use phase, the main emphasis has to be on encouraging efficiency and implementing sustainable fuels or respectively renewable energy. To sum up, sustainable motorsports requires a focus on global sustainability within the regulations including the production of the vehicle and of the fuel as well as the use phase as shown in Figure 13. This implicates the potential to inspire sustainable technologies, to support scale up processes and to encourage track to road development.

Figure 13 – Measures to integrate sustainability into motorsports

80

Sustainable drive concepts for future motorsports

Bibliography 1. https://www.umweltbundesamt.de/themen/klima-energie/klimaschutz-energiepolitik-in-deutschland/treibhausgas-emissionen/emissionsquellen retrieved 03.12.2018. 2. Warnecke, W., Poulet, B., Landschof, J., Dreyer, A., et al., “Innovation from Track to Road: The Role Fuels can Play in Motorsport”, 37. Internationales Wiener Motorensymposium, 2016. 3. Schwarz, L., Bargende, M., Dreyer, S., Baretzky, U., Kotauschek, W., Wohlgemuth, S., Bach, F., “Methodical Selection of Sustainable Fuels for High Performance Racing Engines”, SAE Technical Paper 2018-01-1749, 2018, https://doi.org/10.4271/2018-01-1749 . 4. Hauff, V., „Unsere gemeinsame Zukunft: der Brundtland-Bericht der Weltkommission für Umwelt und Entwicklung.“, Greven, 1987. 5. United Nations, “Report of the World Commission on Environment and Development, Our Common Future”, 1987. 6. http://www.bmub.bund.de/themen/nachhaltigkeit-internationales/nachhaltigeentwicklung/2030-agenda/ retrieved 03.12.2018. 7. https://unfccc.int/process/the-convention/news-and-updates retrieved 03.12.2018. 8. Fédération Internationale de l’Automobile, F1 “Technical Regulations and Sporting Regulations”, 2018. 9. Fédération Internationale de l’Automobile Sport / Technical Department, “2017 Technical Regulations for LMP1 Prototype Hybrid”, 2017. 10. Fédération Internationale de l’Automobile Sport / Technical Department, “2018 – 2019, FIA Formula E Championship Technical Regulations and Sporting Regulations”, 2018. 11. Lucas, Ch., Woietschläger, D. M., „Motorsportstudie 2016 – Das Markenimage der Wettbewerbe im Vergleich“, Technische Universität Braunschweig, 2016. 12. Fédération Internationale de l’Automobile, “Brand Identity Guidelines 2014”, 2014. 13. Europäisches Parlament, Der Rat der europäischen Union, „Verordnung (EG) Nr. 443/2009 des Europäischen Parlaments und des Rates vom 23. April 2009 zur Festsetzung von Emissionsnormen für neue Personenkraftwagen im Rahmen des Gesamtkonzepts der Gemeinschaft zur Verringerung der CO2-Emissionen von Personenkraftwagen und leichten Nutzfahrzeugen“, Amtsblatt der Europäischen Union, 2009.

81

Sustainable drive concepts for future motorsports 14. Europäisches Parlament, Der Rat der europäischen Union, „Verordnung (EG) Nr. 715/2007 des Europäischen Parlaments und des Rates vom 20. Juni 2007 über die Typgenehmigung von Kraftfahrzeugen hinsichtlich der Emissionen von leichten Personenkraftwagen und Nutzfahrzeugen (Euro 5 und Euro 6) und über den Zugang zu Reparatur-und Wartungsinformationen für Fahrzeuge.“ Amtsblatt der Europäischen Union, 2007. 15. Schwarz, L., Bargende, M., Dreyer, S., Baretzky, U., Kotauschek, W., Bach, F., “The holistic life cycle assessment caught between development targets, usage profiles and methodology.”, Internationaler Motorenkongress 2018, Springer Vieweg, Wiesbaden, 2018, https://doi.org/10.1007/978-3-658-21015-1_26. 16. DIN, EN ISO, „Umweltmanagement–Ökobilanz–Grundsätze und Rahmenbedingungen (ISO 14040: 2006)“, 2006. 17. DIN, EN ISO, „Umweltmanagement–Ökobilanz–Grundsätze und Rahmenbedingungen (ISO 14044: 2006)“, 2006. 18. ISO/TR 14047, “Technical Report ISO/TR 14047 – Environmental management – Life cycle assessment – Illustrative examples on how to apply ISO 14044 to impact assessment situations”, 2012. 19. Schwarz, L., Baretzky, U., “The sustainable future of motorsport”, RACE TECH World Motorsport Symposium, 2018. 20. Weymar, E., Finkbeiner, M., “Statistical analysis of empirical lifetime mileage data for automotive LCA”, The International Journal of Life Cycle Assessment, 21. Jg., no. 2, 2016. 21. DMSB, „Ausschreibung/ Reglement VLN Langstreckenmeisterschaft Nürburgring 2018 preliminary“, 2018. 22. http://www.tuningakademie.de/wp-content/uploads/2014/11/Pressemitteilung_Tuning_ Akademie_20110830.pdf , retrieved 03.12.2018. 23. Ward, W., Race Engine Technology – Issue 063, pp. 64–72, 2012. 24. http://www.fstotal.com/history-2/, retrieved 03.12.2018. 25. Dunkelberg, E., „Umweltbewertung von Biokraftstoff-Systemen: eine kritische Analyse von Annahmen und Systemgrenzen“, in Feifel, Silke (ed.), “Ökobilanzierung 2009: Ansätze und Weiterentwicklungen zur Operationalisierung von Nachhaltigkeit“, Tagungsband Ökobilanz-Werkstatt 2009, KIT Scientific Publishing, 2010.

82

Sustainable drive concepts for future motorsports 26. Ausberg, L., Ciroth, A., Feifel, S., Franze, J., et al., „Lebenszyklusanalysen“ in Kaltschmitt,, M., Schiebek, L. (ed.), „Umweltbewertung für Ingenieure: Methoden und Verfahren“, Springer-Verlag, 2015, https://doi.org/10.1007/978-3-642-36989-6_5 . 27. DIN, EN, „228: Kraftstoffe für Kraftfahrzeuge. Unverbleite Ottokraftstoffe–Anforderungen und Prüfverfahren.“, Deutsche Fassung EN 228, 2000. 28. Reif, K. (ed.), „Ottomotor-Management im Überblick“, Springer Fachmedien Wiesbaden, 2015, https://doi.org/10.1007/978-3-658-09524-6. 29. DIN, EN, „590: Kraftstoffe für Kraftfahrzeuge–Dieselkraftstoff–Anforderungen und Prüfverfahren.“, Deutsche Fassung der EN 590, 2004. 30. Ravi, M., Muralidharan, A., Arun, S., “Composite Gas Cylinders for Automotive Vehicles – Current Status of Adoption of Technology and Way Forward.”, Symposium on International Automotive Technology 2013, SAE Technical Paper 201326-0074, 2013, https://doi.org/10.4271/2013-26-0074. 31. https://volkswagen-motorsport.com/index.php?id=2292, retrieved 03.12.2018. 32. Collins, S., “The return of Welter Racing”, Racecar Engineering, Le Mans Guide, http://www.racecar-engineering.com/articles/the-return-of-welter-racing/, retrieved 03.12.2018. 33. http://www.forze-delft.nl/forze-viii/, retrieved 04.12.2018. 34. http://greengt.com/en/tech-cat/the-lmph2g-the-first-hydrogen-racing-prototype/, retrieved 04.12.2018. 35. Society of Automotive Engineers (SAE), “Fueling Protocols for light duty and medium duty gaseous hydrogen surface vehicles (standard J2601_201612)”, 2016. 36. Klell, M., Eichelseder, H., Trattner,A., „Wasserstoff in der Fahrzeugtechnik“, Springer Fachmedien Wiesbaden, 2018, https://doi.org/10.1007/978-3-658-20447-1_5.

83

Automotive clusters in Germany and BadenWürttemberg Dr. Albrecht Fridrich automotive-bw c./o. RKW Baden-Württemberg

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 M. Bargende et al. (Hrsg.), 19. Internationales Stuttgarter Symposium, Proceedings, https://doi.org/10.1007/978-3-658-25939-6_8

85

Automotive clusters in Germany and Baden-Württemberg

1 The end game in the auto industry In January 2019, in an article about job cuts at Ford and JLR, Automobilwoche wrote that "the end game in the automotive industry has begun". The signs are unmistakable. At increasingly shorter intervals in the last few months, the article says, profit warnings have been issued and dire forecasts made. Now, it continued, things are also coming to a head in Europe: British auto maker JLR (which belongs to Tata, the Indian company), Ford and Kuka have announced job cuts in Europe. The list of problems with which manufacturers and parts suppliers are equally having to struggle is long, says the article. The trade conflict between China and the US, it goes on, has made many cars more expensive and led to a reluctance to buy, whereas growth in China, which is the world's most important market, has lessened considerably for the first time in many years. The stricter certification tests such as WLTP (Worldwide Harmonised Light Vehicle Test Procedure) are delaying deliveries and causing gaps in supplies, says the article, whereby stricter CO2 regulations, Brexit chaos and the diesel crisis are making life difficult for the manufacturers in Europe. Companies, it continues, need solid profits from conventional business in order to be able to come up with the billions required for new technologies such as electromobility and autonomous driving. For this reason, only those that avoid big mistakes and possess sufficient resources in order to adjust to the changes will come out of the transition phase in a stronger position, says Automobilwoche [1]. For doubtful cases and also for well-placed companies, the only possible survival strategy seems to be the formation of alliances: the example of Ford with VW, or premium manufacturer BMW with Daimler in their collaboration on the Car2go and DriveNow mobility services may have been the start of a move towards closer collaboration. To this end, they are bringing suppliers together with auto makers or with innovative parts suppliers that have already successfully made a move into the new world of mobility or are organising automotive clusters in Germany. Around two dozen auto-specific clusters are active in Germany at the present time Cluster [2].

2 Clusters: Possible drivers of innovation and helpers in the new "world of mobility" The term cluster describes a regional concentration of companies from one industry or field of technology or competence that may also compete with each other. They cooperate within the value adding chain and at the same time with technically relevant research, educational and knowledge transfer institutions and other public or semipublic institutions to achieve a higher overall benefit together. This requires a sufficient number of enterprises (critical mass).

86

Automotive clusters in Germany and Baden-Württemberg The most important point, however, is the cluster partners’ common goal of developing innovations and implementing innovative processes: So, the term cluster here must always be understood as an innovative economic cluster [3].. An important goal of clusters is to establish cooperative relations in order to bring about greater individual benefits for all players and an overall benefit greater than an individual company could achieve alone. The aim of this is to activate the potential for innovative cooperation in order to generate useful synergies and growth.

2.1 Automotive clusters in Germany support technology transfer and provide orientation The work of the automotive clusters is not spectacular and hardly ever causes a sensation. Nevertheless, the clusters in which automotive parts suppliers and, in some cases, manufacturers as well are allied, can offer advantages to their members and players. This particularly applies to small and medium-sized suppliers that are unable to maintain large research and development departments and do not have any capacity for the observation of trends or market developments. In this case, the network approach can ensure know-how transfer or know-how exchange that enables companies to make substantial progress with regard to concrete products or development of their business. The clusters network their members and players not only with external partners but also with each other [4]. The aim is always to promote contacts between the companies and bring together those that can be suitable partners for each other when it comes to cooperative ventures or the exchange of knowledge. Moreover, collaboration between companies and higher-education institutions or research institutes are usually an integral part of this. In addition, tailor-made further-training courses are often an important component of the activities of automotive clusters. In structural terms, there are two main types of cluster. Some have been initiated by ministries, agencies for economic promotion, or chambers of industry and commerce and offer support to companies in their region. In such cases, there is often no formal membership and rarely any membership fees. The second type of cluster is one that has been initiated by companies and is financed by fees. The clusters in Bavaria, Baden-Württemberg, Lower Saxony and Saxony - all being states that are involved in the auto industry - are especially active. But medium-sized companies and the industry as a whole are also networked in East Germany, Saarland, Hessen and Rheinland-Pfalz via their clusters. It is not easy for small and medium-sized automotive suppliers in particular to find their place in the mobility of the future with electric drives, new mobility services and autonomous driving. The further down the supply chain they are, the more restrictive

87

Automotive clusters in Germany and Baden-Württemberg is their view of the whole in many cases. And departments that are engaged in technology or business-model scouting do not usually have any small suppliers either. This is why the automotive clusters often step in in order to bridge these gaps. Automobilwoche has researched some examples [5]. In Baden-Württemberg, automotive-bw has initiated the promotional program "Transformation in Mobility" for medium-sized suppliers, whereby the suppliers are given grants if they allow themselves to be advised on their strategy for the future. The Mahreg cluster of Sachsen-Anhalt has ties to the competent regional players in the area of lightweight aluminium construction and, in conjunction with the state ministry of agriculture, has started a promotional project, among other things, in which the potential for combinations of the classic aluminium die-casting method with the new technologies of 3D printing are being explored. For the promotional project "Electromobility as the Key to the Mobility of the Future", the Automotive North-West cluster is working with the Oldenburg Energy cluster in order to register and network "regional competences" among other things and to generate synergetic effects of the project partners and the regions". The ITS Mobility cluster, together with Wolfsburg AG and other partners, has initiated the network "Alliance for Intelligent Mobility in Lower Saxony" in order to establish the federal state as a reference region for intelligent mobility. A project of the Bayern Innovativ automotive cluster, together with its partners, also concerns itself with electromobility. In the project, the mobility needs of an existing fleet of vehicles is being investigated. For this purpose, all trips over a period of around six weeks are being recorded and collated. In the next step, the recorded mobility pattern will be simulated with the help of models of energy requirements for a suitable charging infrastructure. In this way, it will be possible to reliably find out which vehicles can be replaced with electric vehicles without any loss of comfort. On the basis of the analysis of mobility requirements, recommendations for the placement of charging stations will be made and a solid and reliable decision-making basis for the possible purchase of electric cars will be provided by means of a mobility check. In the framework of the project, 150 vehicles could be replaced with electric vehicles without any loss of comfort, as a result of which CO2 emissions could reduced by over 400 metric tons.

3 Global players and local heroes: The automotive industry in Baden-Württemberg The car was invented in Baden-Württemberg more than 130 years ago – a fact that has left its imprint on the German south-west and is still doing so today. As a large network or cluster for the development and manufacture of high-quality, high-end vehicles and components, the automotive industry in Baden-Württemberg consists of over

88

Automotive clusters in Germany and Baden-Württemberg 330 companies with around 215,000 employees at the core of the cluster [6]. In addition to Porsche AG, Daimler AG and AUDI AG, numerous world-renowned parts suppliers such as Robert Bosch GmbH, Mahle International GmbH and ZF Friedrichshafen AG are part of it. However, the interaction of the big manufacturers (OEMs) and parts suppliers with innumerable small and medium-sized suppliers and specialised service providers is what makes the auto makers successful all over the world. This cooperation is intensive given that around 25% of the value added of the automobile, especially in Baden-Württemberg, comes from the large OEMs, with 75% coming from local companies. Many small and medium-sized suppliers, which often are counted among the "hidden champions" in the state, bear great responsibility in the value creation process and actively promote innovation in the industry. In Baden-Württemberg, there are almost twice as many owner-run and globally positioned "hidden champions" as in the whole of Germany. An additional success factor of the Baden-Württemberg automotive industry is that companies are strongly networked with science. With more than 70 universities and high-education institutes as well as over 100 non-university research institutes, Baden-Württemberg boasts an outstanding research and development environment. The entire bandwidth of application-oriented research is covered by numerous Fraunhofer, DLR and university research institutes in Karlsruhe and Stuttgart. And there are further higher-education institutes in Baden-Württemberg as a whole – all of which are development partners of the automotive industry and are in high demand. As one of the most innovative sectors, the automotive industry invests around six billion euros in the research and development of new technologies every year. This is more than half of all the R&D expenditure of Baden-Württemberg industrial companies. Additional key facts that underline the success of the automotive industry in Baden-Württemberg: In the last few years, it achieved an annual turnover of more than 115 billion euros, invested 1/3 of Baden-Württemberg's industrial volume and accounted for approximately 34% of the total exports of Baden-Württemberg's industry [7].

3.1 The greatest upheavals for and in the automotive industry at the present time The automotive industry is experiencing its greatest upheaval since the first years after the turn of the millennium. In future, it will have to face several challenges and, at the same time, take account of new requirements in its business models and solutions. One challenge that the automotive industry will have to master is its indispensable internationalisation. As production in this industry is a complex interaction of big

89

Automotive clusters in Germany and Baden-Württemberg OEMs as well as large and small suppliers and worldwide sourcing activities, permanent technological innovations are not the only things that exert an influence; the internationalisation of production is also doing so increasingly. At the moment, German auto makers manufacture 40% of their products in Germany and 60% abroad [8]. They are making greater use of the growth markets and opportunities in China, India and South America with Mexico. The proximity to the sales markets, favourable general conditions as well as the demand of national governments for high localisation rates are the drivers of this development. Large suppliers are therefore following the manufacturers abroad or are even playing a pioneering role. Small and medium-sized suppliers are also subject to this increasing internationalisation of production. The second big challenge, which has equally political and social origins and is buttressed by general legal conditions, is the striven-for reduction of CO2 emissions and, as a consequence thereof, the increasing electrification of the drive train in the context of electromobility and efficiency technologies. The conceptual underpinnings of drives, vehicles and materials are undergoing a fundamental change, as are also the business models of manufacturers and suppliers. Faster and faster model changes and the growing variety of vehicles are already compelling parts suppliers to be increasingly faster and more flexible. This mega-trend of individualisation will change the entire automotive industry and the associated manufacturing process. In future, production will have to be adaptable to enable series production of a wide variety of vehicles. This is a further challenge for the automotive industry in this regard. Each individual partner in the supply chain has to have the necessary relevant competence as only a maximum of 30 per cent of a vehicle is actually produced by the manufacturers; 70 per cent comes from somewhere else. Auto makers have the history, the brand, the market and often the ideas, and make sure that everything goes smoothly. But without the parts suppliers, this would not be possible. The suppliers therefore have a great deal of power. There are companies without whom a car manufacturer could hardly exist, examples being global suppliers such as Bosch, Conti, ZF and Mahle. But like the auto makers, these system suppliers depend on the small ones. For example, if a component or particular control unit does not fit, the assembly line comes to a standstill. An assembly line at a standstill is a terrible disaster. Recall actions – Why do they exist? Often because of some C parts that have been supplied. Small companies therefore also play an important role in the automotive industry, hence the emphasis on the need for these small but efficient enterprises. The car has to function properly. In our view, it is the most complex product that exists. And it is also the most innovative. There are the aircraft, space and medical engineering industries of course, all being sectors whose products are also quite complex, even more complex sometimes, very innovative and so on. But every year around the world, 73 million new vehicles come onto the market. In view of such a quantity, the yardstick for any other product is nowhere near as high. Other aspects are the high

90

Automotive clusters in Germany and Baden-Württemberg innovation rate, the very many new products and the large number of people employed in the industry who work on this product. In the framework of technological developments, automatic and autonomous driving and the associated spectrum of networked vehicle systems such as Car2Car and Car2X are placing new requirements on auto makers and, in the meantime, on potential new start-ups in the electronics industry and Internet technology sector, all in connection with digitisation relating to the product. Added to these economic and technological challenges are the changes in values and attitudes towards the car, especially among the young generation; possession is no longer relevant, use is what is important now.

3.2 Need for greater networking and cooperation in BadenWürttemberg In this context, the need for cooperation and networking between the individual parts suppliers, the small and medium-sized companies and the OEMs is increasing. This is how competitiveness can be maintained and how the ability of structures, organisations and technology to bring about innovation and change can be ensured on a permanent basis. But this networking has already been going on for many years, of course. It started in 1999 when RKW Baden-Württemberg, acting on behalf of the BadenWürttemberg Ministry of Economics, conducted the first automotive suppliers' day (Automobilzulieferertag). Since the middle of the 1980s, the state of BadenWürttemberg had already shown a strong interest in the automotive sector and supplier industry, and continually supports them, especially when it comes to innovation. In the years 2008 to 2010, society, the media and the industry itself expressed a desire for the promotion of different drive concepts, indeed there was a considerable increase in pressure to that effect from these sources. The idea was that drives based on fossil fuels should soon become a thing of the past. In this scenario, interest focussed strongly on the electric car and electromobility. The electric car was to be the future. This produced great uncertainty among suppliers, especially the small ones. At the time, there was already a question as to whether the very existence of small and medium-sized enterprises would be endangered by the disappearance of the entire drive train. If the entire drive train from the motor via the gearbox is dispensed with, the companies supplying the parts of the drive train could also become obsolete. This also indirectly affected the big global players of Baden-Württemberg's flagship industry. As a reaction to this, the government and state administration initiated the network automotive-bw and established e-mobil BW (the state agency for electromobility).

91

Automotive clusters in Germany and Baden-Württemberg

3.2.1 e-mobil BW GmbH, the state agency for electromobility e-mobil BW GmbH was established in order to strengthen and promote BadenWürttemberg as a location for industry and technology in the area of electromobility, including fuel-cell and hydrogen technologies. The aim was to enable BadenWürttemberg to position itself as a location for industry, research and science and as a leading market and provider in the area of electromobility and fuel-cell technology at home and abroad. This was to be based on coordinated concepts of location marketing, public relations and image building [9]. Today, e-mobil BW GmbH is the innovation agency and competence centre of the Federal State of Baden-Württemberg for new mobility solutions and automotive issues. In a network involving partners from industry, science and the public sector, it is shaping the move towards automated, networked and electric mobility within the framework of a type of energy system that can meet the needs of the future. Open to new technological developments, e-mobil BW GmbH promotes the industrialisation, rollout and application of sustainable, climate-friendly and locally emissions-free mobility solutions, the aim being to strengthen Baden-Württemberg as a location for industry and science [10].

3.2.2 automotive-bw, the state automotive network The different players have been networked with each other in the "automotive-bw" cluster initiative since July 2010. As a statewide network, automotive-bw represents a cross-section of vehicle manufacturers, parts/systems suppliers, regional networks and competence centres as well as highereducation institutions and research facilities active in the automotive area and other supporting organisations. The aim of the players in this cluster is to promote the sustainable further development and consolidation of Baden-Württemberg as a location for innovation and production in the automotive area. In order to secure the state's leading role in the increasingly complex area of automotive manufacturing, the activities of the players in the cluster are directed towards the specific deployment, bundling and augmentation of all existing capacities. RKW BadenWürttemberg coordinates the cooperation of the network partners, usually regional

92

Automotive clusters in Germany and Baden-Württemberg agencies for promotion of the economy or initiatives of the chambers of industry and commerce in the state (see map). It is the central support organisation and business office of automotive-bw and also organises state-wide functions and events. It collaborates closely with the state agencies e-mobil BW and Leichtbau BW (lightweight construction BW). automotive-bw has been awarded the quality label "ClusterExzellenz Baden-Württemberg" (Cluster Excellence Baden-Württemberg). In the beginning, the Baden-Württemberg Ministry of Economics sponsored the network with money from the European Regional Development Fund (ERDF) [11]. One important networking measure consists of the so-called TecNet-meetings, in which working groups concerned with the central auto-related issues of electromobility, lightweight construction, efficiency technologies and Connect with autonomous driving have been convening for several years in order to initiate innovations and cooperative ventures. As long-term work groups that focus on selected primary issues in the automotive industry, the TecNets enable the participants to exchange information and ideas with each other and give consideration to problems that, as a rule, relate to concrete technical and technological topics. The tested procedure that has been followed for the TecNet-meetings since the cluster was first established involves a factory tour with discussion, including brain-storming presentations on special issues relating to vehicles and innovations with (usually) three corporate experts, as well as individual talks on cooperation. In this way, experience can be exchanged, areas of potential improvement can be mutually explored, and concrete technical and technological problems can be dealt with. The technological needs of manufacturers and system suppliers where such needs can be compared with the competence of potential supplier companies are also considered. Well-known industrial and scientific representatives accompany the TecNet-groups as mentors. The open exchange of ideas and learning in the framework of collegial discussions are important components of the TecNets. In recent years, around ten TecNet-meetings with over 200 to 300 participants each have been conducted at OEMs, suppliers and higher-education institutions. Small suppliers in particular feel that the TecNets are very important because they can talk to manufacturers directly and get to know the research landscape better. Moreover, the parties sometimes make an agreement to engage in cooperative activities, which lead to innovations and patent applications. With regard to the second important driver of change in the industry, namely internationalisation and cooperation, regional future-oriented forums were conducted in 2018 in the framework of entrepreneur dialogues. The response to them was outstanding and they were held to be extremely useful. The format of these events consisted of a mixture of different presentations given by representatives of ministries and companies as well as by experts. Proposals for discussion and possibilities of networking were also looked at. At the same time, aspects of the industry's interna-

93

Automotive clusters in Germany and Baden-Württemberg tionalisation were focused on, one of these being that, in contrast to the situation 10 years ago when the discussion was about the relocation of production for reasons of costs and wages, the market and customers are at the centre of attention today. Internationalisation and the possible set-up of production abroad for market reasons were also central themes. The forums showed that especially small companies with fewer than 100 employees also need to become international in order to be able to maintain sufficient value-adding capacity at their production location in future. The idea that the increasing internationalisation of the automotive industry weakens or even threatens Germany or Baden-Württemberg as a production location was sometimes a controversial topic of discussion but in the end it was agreed that internationalisation was an opportunity and local companies and the location would tend to profit from such a change.

4 Networks: a small contribution towards the adaptability of the industry In the automotive networks of the industry such as automotive-bw, the main thing that happens is that information and experience are exchanged. This results in cooperation that leads to innovations and new ideas. In the automotive industry, the big companies and the OEMs call the shots but, without the many small enterprises, everything would come to a standstill. Networks enable small companies to talk to the bigger ones, they establish contact with the large corporations and bring the different players together. In Germany, Europe but also worldwide, these new forms of idea generation and cooperation via cluster structures are being tried out, often being initiated by governments. In a very narrow sense, networks are defined as the social interaction of several groups. If different groups come together in networks with the aim of working together and making progress on particular issues as is the case in the automotive networks with manufacturers, suppliers, researchers and troubleshooters, new ideas are generated as a matter of course. The ideas in this context relate to the future of the automotive industry. As they are new and need to be developed, innovation is often the focal point of the activities. Learning in groups from others also means increasing and - as a rule - improving one's own ability to innovate. A basic prerequisite for this is the unadulaterated trust of all participants in the network. If new ideas and innovations are promoted in groups, there is also the danger that ideas can be stolen. Mechanisms are therefore necessary in order to prevent this with non-disclosure agreements or the exclusion of direct competitors in the case of concrete projects. This is also a matter of trust. The nature of the industry and (r)evolutionary developments such as those relating to digitisation

94

Automotive clusters in Germany and Baden-Württemberg entail that networks have to change. They are becoming more international, other players are coming into the picture (see Apple and Google), and cooperation is becoming faster. For the networks, this means being open, and having and radiating trust [12]!

Bibliography 1. Michael Gerster, Das Endspiel in der Autoindustrie hat begonnen, Automobilwoche, 12 January 2019 2. https://www.clusterplattform.de/, Clustersuche automotive 3. Clever vernetzt!- Bausteine der Clusterstrategie Baden-Württembergs, Ministerium für Finanzen und Wirtschaft Baden-Württemberg,, dated November 2012, p.10 4. Gerd Scholz, Technologietransfer und Orientierung: Wo Automobilcluster in Deutschland helfen, Automobilwoche, 24 October 2018 5. Gerd Scholz, Automobil-Cluster: Coaching für die Mobilität der Zukunft, Automobilwoche, 12 October 2018 6. Strukturstudie BWe mobil 2015, e-mobil BW GmbH – Landesagentur für Elektromobilität und Brennstoffzellentechnologie Baden-Württemberg, 3. geänderte Auflage im Juni 20 7. https://www.statistik-bw.de/Industrie/Struktur/Produktionsdaten.jsp 8. http://www.oica.net/category/production-statistics/2017-statistics/, Paris 2017 9. Extract from entry in the commercial registry dated 2010-03-03 HRB 732997 10. https://www.clusterportal-bw.de/clusterdaten/clusterdatenbank/clusterdb/Clusterinitiative/show/clusterinitiative/e-mobil bw 11. Das Automobilcluster Baden-Württemberg, automotive-bw c/o RKW BadenWürttemberg GmbH , 2016 12. Albrecht Fridrich, „Wir bringen die Kleinen mit den Großen zusammen“, in Netzwerke der Zukunft - Zukunft der Netzwerke, Frankfurt 2016, p.36

95

The role of clusters in supporting French automotive industry’s competitiveness and innovation Thomas Röhr Pôle Véhicule du Futur, ESTA School of Business & Technology

The role of clusters in supporting French automotive industry’s competitiveness …

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 M. Bargende et al. (Hrsg.), 19. Internationales Stuttgarter Symposium, Proceedings, https://doi.org/10.1007/978-3-658-25939-6_9

97

The role of clusters in supporting French automotive industry’s competitiveness …

1 Introduction The European automotive industry is one of the most performant ones in the world, and the sector is crucial for Europe’s economy. The automotive sector directly and indirectly employs 13.3 million Europeans, representing 6.1% of total EU employment. The European automotive export totalled 5.9 million motor vehicles in 2017 worth a trade surplus of €90.3 billion. The sector is also a key driver of knowledge and innovation, representing Europe's largest private contributor to R&D, with nearly than €54 billion invested (2016) and more than 8 700 European patents granted to the sector (2017) [1]. Today, the European automotive industry faces multiple challenges regarding their product (e.g. need for electrification of cars, connected and autonomous vehicles), competition (with new arrivals: Google, Tesla...), changing behaviours (shared mobility), data becoming more important than the cars but also industrial modernisation and digitalisation. Therefore, new approaches for innovation and competiveness in products, processes, production and people’s future skills are necessary. Clusters bring together companies, research and training institutions, funding actors, economic development agencies and public authorities in a given geographical zone and in a selected domain for the benefit of their regional ecosystem. Via European and international inter-cluster cooperation the added-value for members can be even multiplied. Clusters are the key entry for the automotive (and mobility) industry to face the above-mentioned challenges. After a short presentation of the significance of the automotive industry for the French economy, the present article describes the two automotive cluster ecosystems present in France before focussing on the cluster Pole Vehicule du Futur (PVF) and the two European inter-cluster cooperation projects EACN and PAE, both co-funded by the European Union. The article closes with a conclusion.

2 The Automotive industry in France France is one of the leading automotive countries in Europe. In 2016, with 2.1 million units, France held the third place of vehicle production behind Germany and Spain in Europe, representing 14% of total western European vehicle production [2]. On a global level, with more than 10 million units, the Renault-Nissan Alliance reached the 3rd place in the 2017 ranking of OEM behind Volkswagen Group and Toyota, and the PSA Group, including Opel, was ranked 9th [3]. In addition, two French Tier 1 – Valeo SA and Faurecia – are also ranked within the top ten global OEM parts suppliers in 2017 [4].

98

The role of clusters in supporting French automotive industry’s competitiveness … The French automotive sector regroups nearly 4 000 companies employing 400 000 people and generating 155 B€ turnover [5]. The 28% of the so-called core companies – OEM, Tier 1 and engine and coachwork manufacturers – realised 64% of the total turnover. The automotive sector represented, 2012 figures, only 2% of the French companies of industrial goods but realised 16 % of the total turnover [6]. The automotive sector is also the second important sector for research and development in France representing a share of 13.3 %, behind the aeronautic and spatial sector (2014) [2]. Automotive related services total 139 000 more establishments with another 400 000 employees [5]. Geographically, OEM and Tier 1 are mainly concentrated in the Eastern and Northern parts of the country, Fig. 1. As for other countries, the financial crisis of 2008 had massive repercussions on the whole automotive value chain with companies at all levels getting in difficulties. In February 2009, the French president announced an Automotive Pact to support the suffering automotive industry especially by supporting the development and sales of electric vehicles, but also by according financial reductions or by improving the OEM-supplier relation [7].

Fig. 1: Geographic distribution of key automotive players in France [6], own translation

One result of this Pact is the launch of the Plateforme Automobile PFA in 2009 by two national federations, the Committee of French Automotive Constructers (Comité des Constructeurs Français d’Automobiles CCFA) and the Committee of Automotive suppliers (Comité de Liaison des Industries Fournisseurs de l'Automobile CLIFA). The PFA represents today 4 000 OEM, suppliers, sub-contractors and mobility actors and defines the national strategy of the automotive sector with regard to innovation, competitiveness, employment and competences [8]. In Mai 2018, the PFA signed a

99

The role of clusters in supporting French automotive industry’s competitiveness … ‘Strategy Contract for the Automotive Sector 2018 -2022’ with the French government [5]. The contract defines four structuring projects and the respective engagements around energetic transition, autonomous vehicles, future skills and competencies and competitiveness in response to the challenges of mobility and manufacturing industry of the future. The automotive industry is very capital driven for production means as well as product development. In answer to the crisis’ impact, the French government, alongside Renault and PSA, launched the Automotive Components Manufacturers Modernisation Fund (FMEA) aiming at investing in companies of strategic importance, particularly those capable of consolidating and modernising its leading sectors. Additionally, the French government launched the Tier 2 FMEA in partnership with Tier 1 auto component manufacturers [9]. In 2015, the FMEA became the Fonds d'Avenir Automobile (FAA, Future Automotive Fund) with two sections: One for Tier 1 suppliers with a capital of 600 M€ targeting investments of 5 to 60 M€, and a second one for Tier 2 (50 M€ capital, investments of 1 to 5 M€) [10]. Other financial means are available via the automotive related programmes of the New Industrial France programme such as e.g. the vehicle consuming less than 2l, autonomy and power of batteries, autonomous vehicles or production related topics (e.g. robotics, cybersecurity, factory of the future) [11], via tax reduction for R&D or via national and European R&D funding schemes.

3 Automotive clusters in France 3.1 Regional Automotive Industry Associations The Regional Associations of Automotive Industry (Associations Régionales de l’Industrie Automobile, ARIAs) are the regional representatives of the PFA. ARIAs have been created in the last 20 years mainly as associations of French law 1901 and cover the whole territory where automotive OEM and Tier 1 are located (see Fig. 1). A survey had been sent to the 11 existing ARIAs in December 2018 of which ten structures answered. In Nouvelle Aquitaine the regional development agency ADI-NA assumes the role of ARIA. They do not conduct automotive specific but crosssectorial actions and do not have a membership system. The numbers of members are very heterogenic, starting from 30 (ARIA Champagne-Ardenne) and going up to nearly 400 (PerfoEst). The three ARIAs that are part of a competitiveness cluster have the highest numbers of members, only the independent ARIA Haut-de-France and RAVI (Réseau Automobile et Véhicules en Ile-de-France) show similar values. The share of SME within all members is going from 24% (ARIA Auvergne Rhône-Alpes) to 89% (RAVI), Table 1.

100

The role of clusters in supporting French automotive industry’s competitiveness … ARIAs concentrate on industrial performance improvements. ARIA managers mentioned network animation, meaning organisation of exchanges, networking and other mutual benefits, and strengthening industrial performance as main missions. Thus, actions are realised to respond to these missions. Typical actions are networking events, thematic working groups, technical visits, organisation of and participation in conferences and exhibitions, offer and initiation of specific training and new educational programmes, information sharing, joint purchasing opportunities and others. Some ARIAs cooperate with pairs or competitiveness clusters to increase the addedvalue for their network. Benefits for members therefore are financial (access to potential partners and customers, expertise for free or at preferential fees, gains via mutual purchases …) as well as intellectual (filtered access to high quality information, exchange of experiences, common studies, training …). Table 1: Key data from French ARIAs (2018 data, own survey) Name

Year of launch 1996 2013 2012 1997

Supporting organisation (independent) CCI CCI PVF b)

Total members 185 31 94 395

of which SME 130 12 74 185

ARIA Hauts-de-France ARIA Champagne-Ardenne ARIA Lorraine PerfoEst (ARIA Grand Est – Bourgogne-Franche-Comté) a) ARIA Auvergne Rhône-Alpes CARA b) 210 50 Automotech Cluster (ARIA Oc2012 AD’OCC 75 55 citanie) Plateforme Automobile du Centre No data available (ARIA Centre) ARIA Pays-de-la-Loire - Bretagne a) iD4CAR b) 298 179 ARIA Normandie 2006 (independent) 110 75 RAVI (ARIA Ile-de-France) 2005 CCI 200 177 a) Merger from different former ARIAs b) For more information see chapter 4

3.2 Pôles de Compétitivité: the French competitiveness clusters The dynamic of competitiveness clusters has been initiated in 2004 by the French government [12]. The European Union defines clusters as “groupings of independent undertakings — innovative start-ups, small, medium and large organisations — operating in a particular sector and region and designed to stimulate innovative activity by

101

The role of clusters in supporting French automotive industry’s competitiveness … promoting intensive interactions, sharing of facilities and exchange of knowledge and expertise and by contributing effectively to technology transfer, networking and information dissemination among the undertakings in the cluster” [13]. Competitiveness clusters’ objectives are to strengthen competitiveness of the French economy by increasing innovation, to increase growth and employment on promising markets by consolidating activities with high technological content, to improve the attractiveness of France through enhanced international visibility, to increase economic returns of R&D projects and to support SME and mid-size companies via individual and collective assistance [14]. The main mission of French competitiveness clusters has evolved during time: ‘Initiating collaborative projects with companies and research institutions to develop innovative products able to be launched in five years’ at the beginning (Phase I, 2004), more general the stimulation of projects in Phase II (2009) and of products for the future in Phase III (2014). In the ongoing Phase VI (2019) the main focus is clearly set on European cooperation of clusters and their members. Funding bases generally on national, regional and local grants fixed in a contract as well as from private or third party’s incomes which shall count for at least half of the budget. Four clusters are active in the automotive/vehicle and the mobility sectors: Mov’eo, iD4CAR and PVF focus on light vehicles whereas CARA is concentrating on heavy vehicles and mass transport such as trucks and busses. All four clusters have been founded at the beginning of the Pôle de Compétitivité dynamic in France. They total 1270 members, of which 65% are SME, and labelled 1420 projects and more than 60 educational trainings with a seal of quality. Table 2 shows some key data gathered via a survey answered by all four clusters. Table 2: Key data of French automotive and mobility competitiveness clusters (own survey) Name

Year of Covered launch Regions

CARA 2006 (prior LUTB) iD4CAR 2005 Mov’eo

2005

PVF

2005

102

Total of which of which of which members SME bigger R&D enterpr. 210 50 25 15

AuvergneRhône-Alpes Pays-de-la-Loire 298 - Bretagne Normandy, Ile- 370 de-France Grand Est – 395 Bourgogne Franche-Comté

179

80

24

231

53

37

185

151

21

The role of clusters in supporting French automotive industry’s competitiveness … Where ARIAs put attention to industrial performance, clusters are focussing on innovation. Supporting and enabling members to develop new products, processes or services is their main overall objective. Missions that cluster manager consider as important are the promotion and moderation of their network (4 mentions), to boost innovation and to initiate collaborative R&D projects (3), to support and improve employment and skills, to improve the European and international development of members, to study and analyse future market and technology tendencies, to support members in launching new products/processes/services (2 each), and, for one cluster, to maintain and develop R&D activities on its territory. All clusters also develop strong partnerships with thematically similar or complementary clusters on European and/or international level. These cluster partnerships are the foundation for supporting members to access new markets or to find partners for collaborative R&D projects or for joint business activities. Since many years, the four clusters cooperate and coordinate actions. All are member in the French Association of Competitiveness Clusters (AFPC). They commonly organised, in 10 years, 42 international co-funded business travels for their members thanks to the cooperation with Business France. In 2018, all four clusters developed a common inter-cluster strategy ‘Automobile and Mobility 2019 – 2022’ [15]. The strategy describes common and individual actions with regard to six axes: Innovation, Network, Europe & International, Growth of companies, Skills and Performance & Industry of the future. The main multiplication factor is the coordination and concertation of activities; it will help out-performing in the next years. Clusters’ added value to members is the easy access to e.g. high-quality thematic information, to an in-depth expertise with regard to market environment, tendencies, technologies, the multiple national and European funding opportunities, to a strong network with other European and international clusters for finding potential research and business partners, to start-up advice and support.

3.3 The future of ARIAs and competitiveness clusters A strong concentration process started one decade before which has been boosted by the French territorial reform in 2016 and is still ongoing. Mergers take mainly place between one or more ARIAs and a competitiveness cluster as both address the same target companies and can value many synergies in work and results. As examples, PerfoEst (ARIA Alsace, Franche-Comté, Vosges) became a department of cluster Vehicule du Futur (2009), the Rhône-Alpes Automotive Cluster RAAC and the ARIA Auvergne (Automac) merged with CARA (2006, 2016) and the former ARIAs Paysde-la-Loire (IAM) and Bretagne (Autéo) merged with the iD4CAR (2017). Further mergers are planned within the next two years with the idea of an extensively fusion of ARIAs and competitiveness clusters.

103

The role of clusters in supporting French automotive industry’s competitiveness …

4 The example of the cluster Pole Vehicule du Futur PVF [16] is one of the first competitiveness clusters to be founded in 2005, a merger took place with the regional ARIA PerfoEst in 2008. Since that moment PVF has two departments covering innovation and industrial performance with two distinct managers. A team of about 15 persons supports the members in their development and in their projects. PVF successfully passed the European ‘Gold label for Cluster Management Excellence’, seal of quality of the team. PVF disposes also of the VeriSelect quality label for vocational training. With nearly 400 members, PVF is the automotive competitiveness cluster in France with the most members. As a result of the 2016 territorial reform, the covered geographic zone was nearly quadrupled in early 2018. It includes now the two newly formed Eastern regions Grand Est and Bourgogne Franche-Comté. In correlation with this territorial reform, the integration of the two ARIAs Lorraine and Champagne-Ardenne, located in the Grand Est Region, is in discussion for 2019. PVF offers companies perspectives over the medium and long term. This means making it a priority to anticipate the needs of the mobility market, whilst continuing to meet today's automotive industry’s needs. In concertation with the members, activities are built around five strategic sectors: Energy and Power trains, Materials & Composites, Industry of the Future, Infrastructures & Communication and Mobility services. They address five different markets: Automotive components, Battery electric vehicles, Hydrogen electric vehicles, Recycling and Intelligent Transport Systems (ITS). PVF offers a large service portfolio to its members including Technology Innovation Groups, support on R&D project building, follow-up support along projects, access to European projects, market and technological survey, international thematic travels or start-up consulting. Since the start of PVF in 2005 until 2018, 435 projects got a seal of quality after examination by an expert committee. 200 of them or 46% for a total amount of 683 M€ got a funding, including 34 projects with European funding. The success rate of proposals to European calls in the 2013–2018 period was 36%. Both success rates, 46% and 36%, are significantly higher as general success rates especially on European level and are the demonstration of the high quality of projects which got PVF’s seal of quality. PVF promotes fuel cell electric mobility since the very beginning and several companies established and started to work in this domain since. Various national and European collaborative projects, such as e.g. MobyPost [17], indicate the strong interest of PVF’s members for this topic. Another important issue is innovation in production, materials and processes as they promise increased industrial performance with rapid

104

The role of clusters in supporting French automotive industry’s competitiveness … return on invest. Therefore, more and more actions and projects are in line with industry of the future related topics. This topic mixes up with PerfoEst objectives; innovation meets industrial performance and is enhanced by corresponding actions regarding lean manufacturing and management, procurement pooling, vocational training or an annual performance survey of automotive companies. Thematic working groups on lean manufacturing, maintenance, logistics, project management take also place several times per year and allow participants to exchange ideas, best practices and experiences and prepare common actions to go further. More and more, national institutions or regional administrations entrust PVF with the coordination of the regional part of national projects or of regional initiatives such as ‘Factory of the future’ initiative from the Bourgogne Franche-Comté Region or the Grand Est initiative DYNAMHYSE aiming at setting up a hydrogen industry in the region. PVF also participates in national associations and initiatives. PVF is regional representative of the PFA and associated to its ‘Strategy Contract for the Automotive Sector 2018 -2022’. Other participations are e.g. the Alliance Industry of the Future [18] and La Nouvelle France Industrielle for industry of the future topics, the French Association of hydrogen and fuel cells AFHYPAC [19] and its Mobility France Hydrogen initiative for fuel cell electric mobility or the French Association for the Development of Electric Mobility (Avere France) [20] for electric mobility in general. These participations ensure a direct access to important information and the possibility to help shape the future. Close cooperation with cities and local and regional governments allow bringing members’ solutions to the market, testing and demonstrating, and by this improving mobility and user’s experiences. PVF is also active partner or even coordinator in European collaborative projects. Current ongoing projects are e-Moticon (Interreg VB Alpine Space), PAE (Interreg VA Grande Région/Großregion) and EACN for Joint Industrial Modernisation Investments (COSME ESCP-S3). Participating in European projects enables PVF to build a strong European network of partners, to study future mobility and automotive industry related topics on a European level and to accompany members to become more competitive.

105

The role of clusters in supporting French automotive industry’s competitiveness …

5 Increased added value by European inter-cluster cooperation: the examples of EACN and PAE 5.1 EACN for Joint Industrial Modernisation Investments In early 2017, on the initiative of PVF, nine European Automotive Clusters1 founded the European Automotive Cluster Network EACN aiming at collaborating and at commonly supporting especially their member SMEs in the field of industrial modernisation and digitalisation. SMEs, as part of the global automotive value network, are producing under difficult frame conditions: Suppliers of different OEMs and/or Tiers 1 using different software tools and management systems, and the need to be compatible with all of them, very low margins where one ‘wrong’ investment decision may put into danger the company or a frequent lack of in-house competences, information and time with regard to industrial modernisation and digitalisation possibilities, making it difficult to select the best solution. Presuming that there are companies in all European automotive regions being confronted to similar problems and being alone when looking for a solution, cooperation is a highly valuable help by two ways: (1) common developments of solutions (products, processes, production tools) means discussing problems with other potential users, improving by that the solution and sharing costs and risks, and (2) joint investments allow improving financial documents and getting easier access to funding and better purchasing conditions. Cooperation is a way to leave isolation when taking decisions. In 2018, the EACN for Joint Industrial Modernisation Investments project [21], submitted by six partners, was approved for co-funding by the COSME programme of the European Union. It runs from October 2018 to October 2020 and is built around two axes (Fig. 2): The first one includes cluster internal actions to strengthen cooperation and prepare further common actions of the project partners. The second axis targets SMEs and aims at initiating at least five joint R&D or investment projects. Actions addressing SMEs include thematic physical/virtual transnational workshops, workgroup building, matchmaking, joint needs identification and the development of common project ideas/topics. Teams of SMEs preparing common proposals will get a co-funding from the EACN project to pay an expert supporting the proposal writing. Four thematic areas will be addressed in the project: (1) Virtualisation and simulation, 1

Pôle Véhicule du Futur (coordinator, France), automotive-bw and Bayern Innovativ (Germany), Fundación Clúster de Empresas de Automoción de Galicia CEAGA and Cluster de la Industria d'Automocio de Catalunya CIAC (Spain), Silesia Automotive & Advanced Manufacturing Cluster SA&AM (Poland), Automotive Cluster Bulgaria and Automotive Cluster Serbia

106

The role of clusters in supporting French automotive industry’s competitiveness … (2) Robotics and artificial intelligence, (3) Elasticity of the production and (4) Skills and competencies.

Fig. 2: EACN for Joint Industrial Modernisation Investments’ project structure

Relationships built up during EACN project’s events between participants from different countries will be the starting point for further cooperation in new collaborative R&D projects or for joint businesses or investments.

5.2 Pôle Automobile Européen PAE Similar to EACN, PAE is also a cluster cooperation project. PAE brings together 10 project partners and three associated partners from Belgium, France, Germany and Luxembourg2. The project has been approved for co-funding by the European

2

Project partner: Belgium: Technifutur, Campus Automobile Spa-Franchorchamps, Université de Liège; France: Chambre de Commerce et d'Industrie du Grand-Est (coordinator), ARIA Lorraine, ARIA Champagne-Ardenne, Pole Vehicule du Futur; Germany: Fahrzeug Initiative Rheinland-Pfalz e.V., autoregion.eu; Luxembourg: LuxInnovation GIE plus three German Associated Partners: automotive-bw, Saaris, Commercial Vehicle Cluster CVC

107

The role of clusters in supporting French automotive industry’s competitiveness … Regional Development Fund ERDF via the Interreg VA Grande Région/Großregion (=Greater Region) programme. The basic problematic addressed in PAE is that many members of the project partners, especially Tier 2 and higher, are only locally active and often depend from only one or very few customers making them extremely vulnerable and putting them in a critical situation. The current evolutions in the automotive and mobility sectors therefore have a strong potential of negative impact on their economic viability. That’s why PAE aims at increasing their presence on new markets inside the Greater Region – Wallonia (Belgium), Luxembourg, Saar and parts of Rhineland-Pfalz (Germany) and Lorraine (France) – as well as abroad. The project also aims at increasing companies’ competitiveness by supporting them in discovering and understanding new technologies (Industry of the Future), by helping them to adapt skills and competencies to these new technologies, and to strengthen cluster cooperation between project partners. Foreseen actions are e.g. common stands on European and international exhibitions, technology site visits, the organisation of Greater Region internal events (AutomotiveDays, Congress of the Future), market and technology analysis, identification of potential new trainings and study programmes, but also the preparation of a strategy for the Greater Region automotive industry to face the upcoming future challenges. PAE will enable SMEs to get support to discover and enter to new markets and to prepare their company to address the future challenges of the sector.

6 Conclusion The European automotive industry currently faces several challenges such as a raising demand for zero emission cars, connected and autonomous vehicles, new competitors mainly interested in data created by cars and their users, raise of shared mobility or the need of industrial modernisation and digitalisation. These challenges will heavily impact many companies who must realign their business by entering new markets and developing new products and services. Cooperation is a good opportunity to prepare the future. Clusters represent an ecosystem where all actors of a defined area interested in a topic are coming together to share experiences, develop business, new products, processes or services and learn about future trends and technologies. The French automotive industry is supported by two types of clusters: The Regional Associations of the Automotive Industry (ARIAs) mainly addressing industrial performance issues and the competitiveness clusters (Pôles de compétitivité) more focussed on innovation and collaborative R&D projects. As both address the same companies, a merger of ARIAs and competitiveness clusters is currently ongoing.

108

The role of clusters in supporting French automotive industry’s competitiveness … With their actions, clusters have a real positive impact on their members as well on financial as on intellectual level. The steady increase of members joining the clusters is a clear proof for clusters’ added-value for companies.

Acknowledgment The author thanks the ARIAs and the automotive competitiveness clusters who shared data and information via the survey conducted in December 2018 and January 2019 as well as the European Union for co-funding the EACN for Joint Industrial Modernisation Investments project via the COSME programme and the PAE project via the European Regional Development Fund ERDF via the Interreg VA Grande Région/Großregion programme.

Bibliography 1. European Automobile Manufacturers’ Association (2018) The Automotive Industry Pocket Guide 2018-2019, Brussels 2. CCFA Comité des constructeurs français d’automobiles (2018) L'industrie automobile française: Analyse et statistiques 2017, Paris 3. Focus2Move (2018) Global Car Sales by Manufacturer: The 2017 final count. https://focus2move.com/world-car-brands-ranking/ 4. Crain Communication Inc. (ed) (2018) North America, Europe and the World Top Suppliers. Automotive News(June) 5. Conseil National de l'Industrie (2018) Contrat stratégique de la filière Automobile 2018-2022 6. Ministère de l'Economie, de l'Industrie et du Numérique, Direction Générale des Entreprises (2015) La filière industrielle de l’automobile: 4 400 sites de production sur une large étendue du territoire. Études économiques, Paris 7. Presidence of the French Republic (2009) Pacte Automobile, Palais de l’Elysée, Paris 8. PFA Filière Automobile et Mobilités Homepage. https://pfa-auto.fr/filiere-automobile-et-mobilites/. Accessed 09 Jan 2019 9. Agence des participations de l'Etat (2010) French State as a shareholder, Paris 10. bpifrance Le Fonds Avenir Automobile (FAA). https://www.bpifrance.fr/Quisommes-nous/Nos-metiers/Fonds-propres/Fonds-directs-Bpifrance/Capital-Developpement-Transmission-Generaliste/Mid-Large-Cap/Le-Fonds-Avenir-AutomobileFAA. Accessed 06 Jan 2019

109

The role of clusters in supporting French automotive industry’s competitiveness … 11. République Française (2014) La Nouvelle France Industrielle: Présentation des feuilles de route des 34 plans de la nouvelle France industrielle, Paris 12. La politique des pôles depuis 2005. http://competitivite.gouv.fr/politique-des-poles/ la-politique-des-poles-depuis-2005-472.html. Accessed 11 Jan 2019 13. Commission of the European Community (2008) The concept of clusters and cluster policies and their role for competitiveness and innovation: Main statistical results and lessons learned. SEC(2008) 2637 14. Direction Générale des Entreprises Les pôles de compétitivité. https://www.entreprises. gouv.fr/politique-et-enjeux/poles-competitivite. Accessed 14 Jan 2019 15. CARA, iD4CAR, Mov'eo, PVF (unpublished) L’inter-pôles « automobile et mobilités ». The French Automotive and & Mobility Innovation Network. 16.Pole Vehicule du Futur Homepage. http://www.vehiculedufutur.com/en/home.html. Accessed 14 Jan 2019 17. MobyPost project homepage. http://mobypost-project.eu/. Accessed 14 Jan 2019 18. Alliance Industrie du Futur Homepage. http://www.industrie-dufutur.org/. Accessed 14 Jan 2019 19. AFHYPAC Homepage. http://www.afhypac.org/. Accessed 14 Jan 2019 20. Avere France Homepage. http://www.avere-france.org/. Accessed 14 Jan 2019 21. European Cluster Collaboration Platform EACN project page. https://www.clustercollaboration.eu/escp-s3-profiles/eacn. Accessed 14 Jan 2019

110

The potential of collaborative business model innovation in automotive eco-systems Georg von der Ropp BMI Lab Deutschland GmbH

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 M. Bargende et al. (Hrsg.), 19. Internationales Stuttgarter Symposium, Proceedings, https://doi.org/10.1007/978-3-658-25939-6_10

111

The potential of collaborative business model innovation in automotive eco-systems

1 Current situation in automotive industry After decades of growth and success the automotive industry is facing heavy turbulences and upheavals. The era of Diesel engines for passenger cars seems to be coming to an end. Across the board, the future of all combustion engines is being questioned. In China, one of the most important automotive markets, sales of electric vehicles have increased by more than 50 % in 2018, whereas sales of cars powered by combustion engines have been falling steadily since 2nd half of 2018. State imposed minimum quota for EVs and further incentives, like an immediate car registration, will accelerate this development further. Trends and customer behavior are changing rapidly. While the demand for mobility continues to grow, the need for owning a vehicle is declining. The ongoing trend of urbanization leads to more and more traffic jams and congestions. This strengthens the demand for mobility as a service, e.g. by sharing vehicles or hailing rides. The trends predicted to shape the future of the automotive industry are car connectivity, autonomous or assisted driving, car sharing and electrified components (CASE). Although it is questionable when and to what extent these developments will prevail, the future for the automotive industry will change dramatically and these shifts will affect every actor.

2 Challenges for players in the automotive industry These changes present great uncertainties about the future for every player in the automotive industry. The value of previous investments and existing business relationships is in question. While incumbents are facing declining revenue streams, new entrants without legacy burdens and inherited obligations will enter the market. As cost cutting programs in order to maintain profitability will only lead to marginalization, company’s leaders have to find a new position for their company and create a new future within the CASE based mobility industry. While most of the OEMs and some of the major tier one suppliers have already started to invest in novel drive technologies and innovative mobility programs, many automotive suppliers are still hesitating to develop new strategies because the future is uncertain and volatile. Which technologies will be the base of their future products? When shall they start to invest and how can they prove to be on the right track? To find answers to these crucial questions it helps to use the conceptual base of business models and to aim for developing business model innovations.

112

The potential of collaborative business model innovation in automotive eco-systems

3 Rise of Business Model Innovation Thinking in business models as a holistic view of how companies create, deliver and capture value was introduced end of the last century. OR: Around the beginning of this millennium, business models were introduced as a holistic approach on how companies create, deliver and capture value. The breakthrough came around 2010 with the publications of Alex Osterwalder and Yves Pigneur (1). The business model concept allows us? to analyze, compare, design and describe how the magic of a business works based on its individual bits and pieces. The St.Gallen Business Model Navigator by Oliver Gassmann, Karolin Frankenberger and Michaela Csik (2) describes business models with the so called magic triangle, a conceptualization that consists of four central dimensions: the WHO, the WHAT, the HOW, and the VALUE. Due to the reduction to four dimensions the concept is easy to use, but, at the same time, exhaustive enough to provide a clear picture of the business model architecture: What?

Value Proposition

Revenue Model Value?

Who?

Value Chain How?

Figure 1: Business model definition – the magic triangle

The WHO dimension describes customers or customer groups and their needs. The WHAT dimension describes what is offered to the target customer, or, put differently, what the customer values. This notion is commonly referred to as the customer value proposition. The HOW dimension determines how the value proposition is built and distributed, i.e. which activities are necessary, which key partners are involved, and which technologies are used. The VALUE dimension explains why the business is

113

The potential of collaborative business model innovation in automotive eco-systems financially viable, how revenues are generated, what the main cost drivers are, and what the revenue mechanics are that enable profitable growth. Based on this concept we define business model innovation as innovation in which more than one dimensions are changed simultaneously to create a coherent and attractive new configuration of the business for the market. It allows companies to differentiate themselves without being compared primarily in terms of prices or features. The popularity of business model innovation can be attributed to a growing awareness that product innovation is not enough. Customers demand convenient services that meet their needs anytime, anywhere. This service orientation, which ideally offers a great customer experience, is a big challenge for product-focused companies as they must overcome the dominant logic of their traditional business.

4 How to successfully create a new business model By analyzing 250 business models in different industries the authors found that 90 % of all new business models have recombined already existing ideas, concepts and technologies. Consequently, they identified 55 patterns of business models which served as the base for new business models in the past. The RAZOR AND BLADE pattern, for example, goes back to Gillette’s 1904 move to give the base product (the razor) away for a low price and earn money through higher-priced consumables (the blades). This pattern, which defines the value proposition and revenue logic of a business model, has spread across many industries ever since. Examples include inkjet printers and cartridges, blood glucose meters and test stripes, or Nespresso’s coffee machines and capsules. When applied as an external stimulus to an existing business model they can spark new ideas, while at the same time allowing to prevent the not-invented-here syndrome (NIH). To successfully become a business model innovator a company has to be client-centric. They need to thoroughly understand their customers, their problems and unmet needs. As customers themselves are often not able to articulate their needs beyond better features and lower prices, an appropriate need finding methodology, e.g. design thinking can be applied. Developing and implementing an innovative business model means discovering new and uncertain territories, where customers, competitors and partners behave unpredictably because they are also looking for their future business.

114

The potential of collaborative business model innovation in automotive eco-systems The best way to reduce uncertainties is to explore options? based on critical assumptions for the future business model. With our approach you will first identify all assumptions and prioritize them according to their criticality. Then assumption by assumption is being tested. If test results confirm an assumption, a subsequent assumption can be formulated. If an assumption cannot be confirmed, new assumptions must be drafted, and the business model will be adjusted accordingly. Creating a great customer experience with the new value proposition usually implies opening up for new partners and new ways of co-operation. This allows you to make use of capabilities, technologies and experiences partners can bring to the table and speeds up the process of implementing a new business model. For companies with a traditional product-focused business, e.g. partners with data-science capabilities can be essential to deliver data-based, digital business models. The challenge is not to find such partners but to create a culture and mind-set of openness and external co-operation within the company.

5 Role of ecosystems Working with partners is not a new concept. All companies in the world collaborate with suppliers, maintain partnerships, e.g. with research institutions, and try to position themselves as partners of their customers. In practice however, most key partners are not treated at eye level. They are regarded as pure service providers and are exchanged if another provider can offer the corresponding service at a lower price. Due to Bernhard Lingens, Maximilian Böger, Steffen Gackstatter and Axell Lemaire the real benefits for companies will arise from business ecosystems consisting of partnerships of equals (3). A business ecosystem is a partnership between three or more companies that results in a service offering that none of the parties would be able to offer alone. These companies may belong to different segments like corporates, SMEs or startups. Based on their research at the Helvetia Innovation Lab, three elements are constitutive of a business ecosystem: ● The shared value proposition is the foundation on which all other projects and forms of cooperation are built. ● The fulfillment of this value proposition is based on modules that do not exist prior to the start of the project and must be provided by numerous partners. ● A so-called orchestrator ensures that the integration of the partners is promoted, and the desired value proposition achieved. To illustrate the definition and performance of ecosystems the authors explain the case of Tailored Fits, a startup that worked as an orchestrator with international partners to

115

The potential of collaborative business model innovation in automotive eco-systems offer individually 3D printed insoles. The customer journey begins at a retailer’s store where the customer’s foot is scanned with an iPad using a body scanning solution from Techmed 3D, Canada. The order to print the insoles is issued via an online platform, then printed with cutting-edge 3D Printing Technology supplied by Materialise N.V., Belgium. Upon delivery, the customer receives a custom-made insole that offers ultimate comfort and added value. The time-to-market for Tailored Fits and the ecosystem was 18 months, while Adidas spent several years trying unsuccessfully to bring a similar solution to market. These kinds of business ecosystems offer new growth opportunities, because they enable companies to accompany customers throughout the entire customer journey. This allows for the creating of a superior experience for the customer and hence offer significant competitive advantages for all partners in the ecosystem. At the same time, novel business ecosystems allow each party to profit from the different cultures and capabilities of the participating partners. Startups will bring in agility and speed. SMEs often contribute with deep knowledge on specific technologies, whereas corporates can contribute financial resources and global reach. Business models resulting from business ecosystems combine the BMI pattern “Open Business Model” and “Orchestrator”, where partners become a central source of value creation and the different parts of the value chain are actively coordinated. Often the pattern “Revenue Sharing” is also being applied in such environment in order to divide the revenues between partners. Similarly, all other patterns can be used to create an innovative business model based on the shared value proposition. What are the risks of this approach? One obvious risk is that the costs for cooperation including the costs of orchestrations exceed the additional benefits. To handle this risk is one of the key tasks of the orchestrator by creating clarity of each partner’s module and through transparency and close coordination. Thanks to digitization, transaction costs are reduced significantly. Many SMEs are traditionally focused on own resources and capabilities. Therefore, they are particularly concerned about the lack of control over the partners and the risk of lost investments when one of the partners drops out. This concern can only be covered through mutual trust and a growing identity based on the superior shared value proposition. How to find the right partners for a business ecosystem? Identifying the right partners is a key challenge in the creation of a viable business ecosystem. An example can illustrate how a company successfully scanned its competitive environment to identify potential partners: The Helvetia Innovation Lab, a cooperation of Helvetia Insurance and the Institute of Technology Management of the

116

The potential of collaborative business model innovation in automotive eco-systems University of St. Gallen, supported Helvetia in building a business ecosystem “Home” to offer new services to customers. In order to systematically explore potential ecosystem partners, they defined the customer journey “Home”. For each step of this journey they then identified startups with the ability to transform technology or to bring other capabilities into joint innovation. inform

purchase / build

finance

search decide (which house? rent or buy?)

rental agreement

design renovate secure live

insure move

Figure 2: Startups along the customer journey “Home”, Helvetia Innovation Lab, 2017

This ecosystem offers Helvetia opportunities to explore new revenue streams outside their core business and to build a service portfolio along this journey, which leads to a lock-in effect. The CEO of Helvetia, Philipp Gmür, sums up: "Value creation in tomorrow's world will take place in ecosystems, because they enable us to accompany the customer throughout the entire customer journey, which represents significant competitive edge."

117

The potential of collaborative business model innovation in automotive eco-systems

6 Conclusion for players in the automotive industry As we have seen, business ecosystems with partnerships of equals offer companies the possibility to create innovative and sustainable new business models. Emerging trends like urbanization, new customer behavior like sharing instead of owning, the increasing demand for convenience and availability of services always and everywhere combined with the possibilities of digitalization create great space for new business opportunities. To enter this new playing ground, players in the automotive industry need to consider themselves more than just technology-driven product or component suppliers. By redefining their mission e.g. as mobility providers, they will discover lots of new customer needs, which are currently not being met. To unlock this innovation potential, they need to determine their starting position. They have to identify their core competencies, to check their business innovation maturity and to describe their current stakeholder network. Based on these insights they can choose from two options to start developing a business ecosystem: ● Managing an own business ecosystem as an orchestrator: Creating a superior value proposition for a specific customer group with relevant and unmet customer needs. Reaching out for possible partners with complementary capabilities and sharing the value proposition. ● Becoming partner in business ecosystems of other orchestrators. In this case it is important to participate in different networks, e.g. with cross-industry partners or research institutes, to visit regularly innovation hubs or startup events and to maintain conversations with consumers. In order to manage risks, the authors of (3) recommend a portfolio approach where companies orchestrate their own business ecosystem, while participating as contributors in further business ecosystems, managed by other orchestrators. To establish an ecosystem, it is crucial to create alignment based on the shared value proposition, to create acceptance for the orchestrator and to build mutual trust. The partners in the ecosystem who are willing to deliver this shared value proposition have to ensure that this co-innovation approach will run smoothly. Crucial success factors are: ● Focus on the shared value proposition and clear goals for the project: Which customer needs and problems should be addressed? Which modules does each partner contribute and how will they fit together from a customer point of view? What are the timelines and milestones?

118

The potential of collaborative business model innovation in automotive eco-systems ● Create open mindset and transparency: Enable people to collaborate beyond company borders and overcome not-invented-here-syndrome. Being transparent towards partners as to the approach used and the results achieved. It is equally important to openly address challenges and problems met in the process. ● Commitment to the ecosystem approach and the project: Create mutual trust through visible management attention, allocate resources and foster ongoing exchange. ● Quick results and rapid success: Start with one specific customer segment, e.g. the one with the most urgent need. Validate assumptions in a lean way and share results with partners. Build minimal viable products (MVP) in order to learn rapidly. This approach ensures a short time-to-market and creates early feedback from real customers. After serving the first customer segment the partners will have learnt enough to decide which segment to address next. The approach of business model innovation combined with business ecosystems is not the new silver bullet. It will not guarantee success, but it allows a business to create new growth opportunities beyond the borders of their own enterprise. At the same time, the method accelerates customer centricity as it is focused on customer journeys. Since the automotive industry is facing tremendous changes, hardly any company can afford to ignore these opportunities. Especially for technology- and product-oriented companies it is impossible to bear the investments to develop all possible technology individually, which could become important for future mobility. The ecosystem approach allows companies to share investments and risks while bringing together core competencies from the various partners. As the shared value proposition only can be offered to the customers through the ecosystem, it also leads to differentiation and sustainable competitive advantages.

Bibliography 1. Osterwalder, A., & Pigneur, Y. (2010). Business Model Generation: A Handbook for Visionaries, Game Changers, and Challengers. Hoboken, NJ: John Wiley & Son 2. Gassmann, Oliver; Frankenberger, Karolin & Csik, Michaela: The St.Gallen Business Model Navigator, 2013 3. Lingens, Bernhard; Böger, Maximilian; Gackstatter, Steffen & Lemaire, Axelle: Business Ecosystems - Partnership of equals for corporates, SMEs and startups. (2019)

119

Denoxtronic 5.3 – A modular system for applications worldwide Dipl.-Ing. Michael Raff, Dipl.-Ing. Erik Weingarten, Dipl.-Ing. Manuel Muslija Robert Bosch GmbH

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 M. Bargende et al. (Hrsg.), 19. Internationales Stuttgarter Symposium, Proceedings, https://doi.org/10.1007/978-3-658-25939-6_11

121

Denoxtronic 5.3 – A modular system for applications worldwide

1 Summary Diesel engines will play an important part to reach future CO2-limits for on-road mobility in the passenger car (PC) and light duty (LD) segment. High efficiency and compliance with new emission legislations are drivers for further improvements of the technology for markets like Europe with EURO 6, enforced by introduction of European Real Driving Emissions in two steps (mid 2017 & beginning of 2020), but also China (CHINA 6) and India (Bharat Stage VI), where comparable legislations will be implemented in 2020. One major engineering target for the Diesel powertrains is to minimize nitrogen-oxide emissions. This is possible with selective catalytic reduction catalysts (SCR) and active dosing of aqueous urea solution (Diesel Exhaust Fluid DEF or AdBlue®). Configurations of exhaust gas treatment (EGT) – a combination of SCR with Diesel oxidation catalyst (DOC) / NOx storage catalyst (NSC) and Diesel particulate filter (DPF) – will vary depending on market and vehicle segment needs. This diversity of layouts has been taken into account during the development of the next generation AdBlue®-Dosing system of Bosch (Denoxtronic): A modularity of systems, software and a modular component approach enable a perfect fit from Indian 4-wheeler two-cylinder applications up to high-performance applications with double-injectionSCR systems. This next step in evolution of the 3rd generation Denoxtronic of Bosch including innovative system and software function aspects is presented.

2 Exhaust Gas Treatment: Market and trends PC/LD 2.1 Market situation PC/LD The worldwide combined average growth rate of passenger cars and light vehicles is still positive in the upcoming years. The volume of the overall PC/LD market in 2025 is pending between 105 and 111 million vehicles depending on the end consumer acceptance of car sharing and other new mobility solutions. The major powertrain will still be the internal combustion engine (ICE) with a share of market between 70 – 90% in 2025. If applied effectively in the respective vehicle segments (from small passenger cars up to large SUVs / light duty vehicles) ICE power trains in combination with a certain degree of hybridization are capable of fulfilling new emission and CO2 regulations.

122

Denoxtronic 5.3 – A modular system for applications worldwide The success of the battery electric vehicles (BEV) still depends on the availability of the necessary infrastructure, the acceptance of the end consumer to change their “driving behavior” and the availability of affordable electrical vehicles. Real driving fuel consumption together with CO2 emissions for the production of fuel or electricity to power the engines will increase transparency concerning CO2 emissions impact from different powertrain strategies and solutions.

2.2 Emissions Legislation worldwide1 In 2015 the emission standard EURO 6b based on the New European Driving Cycle (NEDC) has been introduced for the passenger car manufacturers in Europe. The NEDC introduced in 1996 was created to compare vehicles and their emissions: cycle time 20 minutes, distance 11,000 m, mean speed 34 km/h and maximum speed 120 km/h. The reduction of NOx limits from 180 mg/km to 80 mg/km in EURO 6 has been one additional driver for the introduction of more complex EGT systems for Diesel engines. At the same time, the CO2 PC fleet average was limited to 130 g/km (LD 175 g/km) in 2015, which was only reached by using the high efficient Diesel engine especially in the heavier vehicle segments. In September 2017 the worldwide-harmonized light-duty vehicles test procedure (WLTP, EURO 6c) with more dynamic and higher load areas replaced the NEDC test procedure: cycle time 30 minutes, distance 23,500 m, mean speed 46,6 km/h and maximum speed 131 km/h. The European Real-Driving-Emission (RDE) standard with a conformity factor of 2.1 will be obligatory from September 2019 (EURO 6d temp) on. The RDE regulation intends to cover significantly more than 90% of the real driving modes and ambient conditions. This regulation excludes only nonsensical driving modes, which may theoretically arise but are practically irrelevant, e.g. permanent idling or unrealistic or illegal vehicle dynamics. The second step of RDE regulation (EURO 6d final) will be introduced in effect from January 2020, in which the conformity factor will be further decreased to 1.5 (1 + 0.5 for tolerances of in service measuring devices). Also in terms of CO2 output, the limit for the OEM fleet will decrease to 95 g/km for PC and to 147 g/km for LD until 2021 and the test cycle – as a comparison base between different vehicles – will change to the WLTP. Due to this change the differences between consumption level in the test cycle and the real driving will decrease, but will still be remarkable for the end consumer. China has introduced CHINA 5 already in 2014 in Bejing and since 2018 nationwide. The emission limits are based on the European emission legislation (EURO 5). In 1

The condensed summary of elements of the emission legislations worldwide has been prepared and provided for informational purposes only and strictly on a non-reliance basis

123

Denoxtronic 5.3 – A modular system for applications worldwide July 2020 CHINA 6a will be introduced and also testing cycle will change over to the WLTP. This emission norm will be even more stringent than the EURO 6 limits de-pending on the vehicle classes, for example, NOx-limit for PC as well as LD Class I (reference mass (RM) < 1350 kg) is from 60 mg/km up to 82 mg/km for LD Class III (RM > 1760 kg). At the same time, a more stringent CO2 standard PHASE V (probable fleet target 95 g/km) is still under discussion to be phased in 2020. Introduction of RDE with a not yet defined conformity factor (CHINA 6b) is expected nationwide for July 2023. In parallel, the emission limits for the WLTP will be even decreased further (e.g. NOx 35 – 50 mg/km de-pending on LD Class).

Fig. 1: Emission and CO2 legislation worldwide

In India Bharat Stage (BS) IV has been introduced in 13 major cities in April 2010 and enforced in the entire country afterwards. Also Bharat Stage is based on the emission limits of the European legislation but with some modifications in terms of test conditions. In 2016, the Indian government decided to skip BS V and adopt directly BS VI by 2020. This decision was supported by the Supreme Court who banned the sales and registration of ICE vehicles not fulfilling BS VI in the entire country by April 2020. BS VI limits will still be verified on an adapted NEDC for India, which has a lower maximum speed than in Europe, but also new driving cycles, which are more representative for Indian city traffic, are under evaluation. Still under discussion is the introduction of RDE in 2023/2024 and the conformity factor for BS VI b. CO2 limits will be decreased to 113 g/km in 2022. However, due to a completely different vehicle segmentation with a high share of small vehicles in India the achievement of the CO2 limits shall not be so challenging for the Indian OEMs.

124

Denoxtronic 5.3 – A modular system for applications worldwide

3 Exhaust Gas Treatment Solutions All these changes in legislation are pushing the development of Diesel engine and especially of the EGT systems to new concepts to improve the efficiency of NOx reduction.

3.1 EGT in Europe, US & Japan The development of Exhaust Gas Treatment for Diesel vehicle started for the emission legislation EURO 4 with a Diesel Oxidation Catalyst (DOC), which transforms unburned hydrocarbons to CO2 and H2O. For EURO 5 a coated Diesel Particle Filter (cDPF) to reduce the particles of the modern Diesel engines became obligatory. With the evolution towards more efficient engines (higher combustion temperatures and higher raw NOx emissions), a Lean NOx trap (LNT) or NOx Selective Catalyst (NSC) was introduced and combined with the cDPF for the upcoming EURO 6a legislation and in some cases already an underfloor Selective Catalyst Reaction (ufSCR) with active AdBlue® dosing system was used. The last improvement and at the moment mainstream for EURO 6 b/c in the exhaust gas treatment is the closed-coupled SCR on DPF (SCRoF) in combination with a NSC and a passive SCR (pSCR) with Clean-UpCatalyst (CUC) to prevent NH3 slip in terms of AdBlue® over-dosing.

Fig. 2: Exhaust Gas Treatment evolution in the past years

These state-of-the-art SCR-based EGT systems currently available already offer a tremendous potential for NOx reduction in real driving conditions. Compact close-coupled systems make efficient use of the exhaust heat and can quickly reach the necessary light-off temperature of the close-coupled catalyst. This already results in a significant increase in efficiency, however an additional warm-up management is required and highly effective to limit the emissions during a cold start low-load operation.

125

Denoxtronic 5.3 – A modular system for applications worldwide EU6-Real driving emission-tests with this exhaust gas system in combination with a temperature management have been performed with the Bosch platform demonstrator vehicle on the “Bosch Stuttgart Route”.

Fig. 3: On road measurements “Bosch Stuttgart RDE track”

The values presented in Figure 3 are the mean values as measured without any weighting functions. For the sake of robustness, the test engineers deliberately tried to create extreme driving conditions that certainly exceed the regulation's requirements for "normal driving". The key findings are: All measurements are well below 80 mg/km. Further development addresses several topics e.g. further improved temperature management, optimization of turbo charger or enhanced system control functions up to model based predictive functions. Nevertheless cold start contributes the most to the overall emission result. To be compliant with EU6-RDE, low emissions during high load driving, a small Ki factor and a robust catalyst diagnostic are mandatory for series applications. One possibility to improve the benefit of the underfloor SCR catalyst is a second AdBlue® injector (Double Injection) in front of the underfloor SCR catalyst. This also brings opportunities for further improvements of NOx emissions during DPF re-generation, high engine load operation and for catalyst diagnostics. In a drive cycle without a particulate filter regeneration the benefits using Double Injection-SCR (DI-SCR) during WLTC appear (as expected) to be marginal. Singleinjection as well as double injection consume a similar amount of AdBlue®. Tailpipe emissions level out at the same magnitude. Caused by relatively low exhaust temperatures (< 400 °C) there is very low dosing share for the under floor injection position. The only benefit for such a low-load cycle is the minimized ammonia ingress into the Low-Pressure Exhaust-Gas-Recirculation (LP EGR).

126

Denoxtronic 5.3 – A modular system for applications worldwide Driving the same cycle plus an overlaid particulate filter regeneration in phase 2 and phase 3 increases the exhaust temperature beyond 600 °C. The thermal separation of SCR catalysts leads to significant improvement of total NOx conversion using Double Injection. Additionally the NOx reduction per mass of injected AdBlue® can be optimized by more than 40 % by controlling the second dosing position appropriately (see Figure 4).

Fig. 4: Results Double Injection SCR in WLTC with DPF regeneration

The integration of the second injection position for dynamic cycles with and without particulate-filter regeneration provides the leeway required to simultaneously meet the requirements “Low NOx”, “Low AdBlue® consumption” and “Low NH3 ingress to the LP-EGR”. Especially the possibility of an efficient use of AdBlue® with minimum NOx tailpipe emissions under high loads and the conversion benefits during particulate filter regeneration are worth mentioning.

3.2 EGT in EMM (China, India) In China, the EGT development for Diesel engines focusses the SUV, Pickup Truck, Van and Light Duty Segment. For CHINA 5 legislation, only the Light Duty vehicles (LDV) with engine certification (> 3.5 t) are equipped with a DOC-SCR combination with active AdBlue® Dosing. All other Diesel vehicles are using mainly a DOC-DPF exhaust system with a high-pressure exhaust gas recirculation (HP EGR). For the upcoming CHINA 6a regulation all these applications will change over to a NSC-DPF combination with an underfloor SCR or in some cases DOC with a closed coupled SCR on DPF system with active AdBlue© dosing to fulfill the more stringent NOx limits. The engine certified LDV will be equipped with a DPF in-between the DOC und the active SCR catalyst. Though the

127

Denoxtronic 5.3 – A modular system for applications worldwide conformity factor for CHINA 6b is not yet defined, it is expected that it will be comparable to Europe; therefore, Double Injection SCR systems will most probably be introduced.

Fig.5: Vehicle segmentation and EGT solutions in China

In China the vehicle segmentation and emission legislation is similar to Europe, therefore also the applied EGT technology is comparable. However, in India the situation is much more complex. The segmentation of the vehicle classes must be splitted into eight classes depending on the inertia, the engine size and the number of wheels. Today all these vehicle types are equipped with a DOC only or a combination of DOC and an open particle filter (POC), depending on the air and fuel injection system. Only the heavier LDVs are already in some cases equipped with an active SCR. Due to leapfrogging directly from BS IV to BS VI and the cost sensitive Indian market several studies were performed to choose the right exhaust gas treatment layout. The Low Price vehicles (LPV) with the highest inertia class (1020-1250 kg) for example were tested with different EGT systems. The first investigation was with a NSC-cDPFpSCR combination. With the highest possible payload of 1,080 kg and the typical Indian driving conditions (deep city driving-city-rural-highway), the exhaust temperature at EGT inlet got too hot and was in the rural and highway section outside of the active temperature range of the NSC (200 – 400°C). Only a decrease of the cargo load and limitation of the top vehicle speed enabled the system to stay in the temperature range of the NSC/pSCR system. To compare now the different EGT systems and the necessary measures a Total-Cost-of-Ownership (TCO) investigation between NSCcDPF-pSCR and DOC-cDPF-ufSCR w/ active DEF dosing system was performed. Additional component and AdBlue® refilling costs on the one hand and on the other hand additional fuel due to lower payload were taken into account and calculated over vehicle lifetime. Only after a few months of vehicle life, the SCR system with active AdBlue® dosing becomes more attractive for the end customer than the same vehicle with NSC and a lower payload.

128

Denoxtronic 5.3 – A modular system for applications worldwide

Fig. 6: Vehicle segmentation and EGT solutions in India

For the introduction of RDE with BS VIb also the other vehicle segments (except 3-wheeler) in India could require an active SCR-system depending on the conformity factor.

4 AdBlue® Dosing System Solutions To serve all these different markets, segments and exhaust gas treatment requirements the target of the AdBlue® Dosing System is to be as modular as possible. Therefore a requirements engineering together with the OEMs has been performed to develop a system fitting to all different needs. These requirements are broken down within the engineering process to define the adequate specifications for the components taking into account additional software functions on system level to meet the performance required, thus keeping the costs down. This process allows a clear traceability from requirement to test case and is in line with state of the art engineering methods and standards.

4.1 Denoxtronic System layout Depending on the vehicle class, the Denoxtronic System consists of a Supply Module (SM) welded into the AdBlue®-Tank, the Dosing Module (DM) – air cooled for underfloor applications and water-cooled for closed coupled – and a control unit. The hydraulic pressure of the system is at 6.0 bar, which can be sustained during start&stop phases without any pump actuation. Dosing Modules and pipes are partly emptied after a manual engine shut off only.

129

Denoxtronic 5.3 – A modular system for applications worldwide A separate Dosing Control Unit (DCU), which includes the heating power stages and control for the Supply Module and heated pressure lines, is connected via CAN to the Engine Control Unit (ECU). This solution allows full flexibility for the Denoxtronic System integration into the vehicle. Alternatively, an integration of the dosing-system controls into the Bosch Engine Control Unit can be a cost-optimized option. In this case a small Heater Control Unit (HCU) containing the power stages has to be added to the system. Additional tank heating devices, which can improve sustainability in cold conditions, depending on the tank geometry and the climatic conditions of the vehicle, can also be controlled either by DCU or by HCU.

Fig. 7: Denoxtronic System for Double-Injection

For Light Duty vehicles in China and in applications with sufficient mounting space, the Denoxtronic 6-5, a vehicle mounted supply module, with higher flow rates (up to 5.5 kg/h) and an exchangeable filter is used. Vibration, particle robustness and serviceability are some advantages of this chassis mounted supply module and are requested from Chinese OEMs in the heavier LD segment. For the Indian market, these Denoxtronic systems (in-tank as well as Chassis mounted) are developed without the integrated heater in the Supply Module (BaseLine Systems). Due to the climatic situation in India and the freezing point of DEF at -11°C, heating of the complete system is not required. This modularity helps to decrease the cost and to keep the vehicle costs in this price sensitive market as low as possible. The new Denoxtronic 5.3 – the 3rd generation of Bosch’s in-tank DEF dosing systems – consists besides new control software mainly of an evolutionary step of the supply module based on the proven technology of the current state of the art Denoxtronic 5.1 system.

130

Denoxtronic 5.3 – A modular system for applications worldwide Several sub-components were adapted to the new requirements derived from the new European legislation in 2020 but also looking into future requirements (RDE introduction) in China and India. The pump module, serviceable and integrated into the carrier, is upgraded with an icerobust pressure sensor (HighLine Version), improved main pump performance with a maximum delivery of 3.5 kg/h and a not acoustically perceptible backflow pump. Both pumps remain volumetric membrane pumps, which allow an improvement of the whole Denoxtronic system (see chapter 4.2) and are more robust over lifetime than rotatory pumps. The freeze-robust pump module is heated by the pump coils and heating plates, ensures a partially emptied AdBlue© system after engine shut off and enables a fast pressure build up after restart. For applications, with lower requirements on volume flow, the pressure sensor can be omitted (ValueLine Version). The system pressure in this case is controlled similar as in the current series product Denoxtronic 5.1 by a SW function, which uses the volumetric principle of the main pump. The self-limiting and fail-safe heating device is welded on the carrier and comprises an optimized connection of the PTCs (Positive Temperature Coefficient) to the aluminum heater core and an optimized material for the over-molding as well. The new design of the heater allows a faster defreezing of the filter and the Sensor Unit and improves the sustainability and readiness in cold conditions.

Fig. 8: Supply Module 5.3 HighLine / Value Line

Based on the experience of the current generation Denoxtronic 5.1 the filter and the Sensor Unit are installed at the lowest possible point of the supply module and thus of the tank. The Sensor Unit integrates the functions for temperature, level and concentration / quality measurements.

131

Denoxtronic 5.3 – A modular system for applications worldwide Instead of integrating the sensor unit directly into the carrier as done in the current supply module, the new sensor unit is welded onto the carrier. This improves robustness of the ultrasonic concentration sensing. The filter comprises the proven reservoir function, which avoids suction of air even at low fill levels and highly dynamic driving conditions. Furthermore, it is designed as a lifetime filter with regards to dirt load. In Europe the concentration sensing, introduced in 2017, has become already stateof-the-art and also in China with introduction of CHINA 6a this sensor will be implemented into the tank. For India, it is expected that with the introduction of RDE in 2023/2024, the concentration sensing will become like in Europe and China also obligatory. All these component improvements are supported by intelligent system functions, which keep the performance to cost ratio at an optimum.

4.2 Denoxtronic System functions Basis for the design of software strategies to actuate and control the Denoxtronic system is an understanding of the system behavior in real use, a definition of precise and adequate specifications for the components and a deep understanding of physical causeeffect relations between the functional elements of the system. In order to be compatible to various system-configurations – i.e. with single or with double injection, with or without pressure sensor, without or with several heater circuits – a new modular system architecture is created. This allows improvements of existing “proven in use” functional components, as well as new developments, while optimizing the sequence control state machines to get clear and consistent interfaces and transparent conditions for state changes. The complexity to adapt this new dosing control software to customer specific SW topologies will be lower and the effort to calibrate will be reduced significantly. Based on this architecture several intelligent functions are developed to improve system behavior in terms of sensing, heating and pressure control as well as valve actuation:

132

Denoxtronic 5.3 – A modular system for applications worldwide

Fig. 9: System Functions of Denoxtronic 5.3

The introduction of a pressure sensor into the pump module with an adaptive 2-point controller (A2C) improves the pressure regulation and increases at the same time the maximum delivery rate (> 3.5 kg/h depending on the operation point). The controller actuates a pump stroke when the pressure falls below a certain threshold and a software function determines an optimized actuation pattern. The internal memory of the pressure sensor (with SENT-interface) is used to record end-of-line test results of the displacement volume per pump stroke (EDM), eliminating part-to-part variations. For further improvement of the dosing tolerance, the controlled valve operation (CVO) function reduces the inaccuracy in the opening and closing time of the dosing module. Both the delivery rate of the pump and the dosing rate of both dosing valves are continuously compared. Inaccuracies can in turn be compensated by adaptation independently for each dosing valve. All these methods combined bring the system dosing accuracy up to levels unique in the market. By making use of the adaptation factors that are calculated for each valve and the system as a whole, it is possible not only to compensate for ageing effects. It is also possible to lower the threshold between effects caused by usual tolerances and erroneous behavior and thereby safely pinpoint to the component that needs to be diagnosed while keeping dosing accuracy at a maximum. The heating strategy was adapted (AHS) to react faster and to realize an earlier pressure build up by confirming that the AdBlue® in the supply module is completely thawed. By using independent tank state indicators like level response, tank temperature and time since last drive cycle the heating performance (defreezing vs. system readiness vs. sustainability) is optimized.

133

Denoxtronic 5.3 – A modular system for applications worldwide Level as well as concentration sensing are also supported by the system function “Adaptive Concentration/Level Sensing” (ACS). In case of a “fast” change in AdBlue quality, e.g. due to significant dilution with water a software function is used to trigger the warning and inducement system. A “slow” concentration change due to e.g. sensor drift will be compensated by the function. This keeps the tolerance of the sensor as well as the sensor price in an optimized range. The Level signal is also referenced to a calculated volume based on volumetric pump strokes and injected quantity. This enables the system to have stable level indication also during low fill level, strong inclinations or high slosh conditions.

5 Summary and Outlook The powertrains of passenger cars and light duty vehicles are in a transition phase from pure internal combustion engines to partly or fully electrified propulsions. Due to the technical and infrastructural issues, it is still unclear how long this transition phase will last. For this period, the market still needs combustion engines fulfilling emission limits in real driving conditions and at the same time keeping the CO2 output as low as possible. Bosch believes that Diesel technology, with its high efficiency, is an important contributor to achieve greenhouse gas reductions, especially for the heavier Passenger Cars, like SUVs, PickUps and LDVs and is worth to consider for further improvements as further steps to reduce nitrogen oxide emissions can be taken. With consequent system optimization e.g. of the interaction between the engine and the exhaust gas treatment compliance to EU6d final stage can be achieved and exceeded. Bosch sees that implementing and optimizing of AdBlue dosing technology with double injection and improved performance on system level as one key enabler to fulfill forthcoming EU real driving emissions standards and even beyond. 8%

27%

23%

30% 47%

2018

53%

Diesel w/o active SCR

2020

2025

62%

50%

Diesel w/ Mono-SCR

Diesel w/ DI-SCR

Fig. 10: Worldwide EGT Trends for PC/LD Diesel engines

134

Denoxtronic 5.3 – A modular system for applications worldwide The clear worldwide trend is to integrate EGT systems with SCR catalysts. In 2025 approximately 77% of all PC and LD Diesel applications will be equipped with an active Dosing system worldwide and even the Double Injection SCR systems will have a higher market share than systems w/o active AdBlue® Dosing. The new generation of AdBlue® dosing system, Denoxtronic 5.3, with the modularity of its components as well as with intelligent software functions combined in a unified architecture is able to serve all these different markets, vehicle segments and EGT solutions worldwide by fulfilling the requirements on the one hand and by keeping the cost targets on the other hand.

Bibliography 1. Erik Weingarten, Tobias Bayer, Marc Chaineux, Dr. Hartmut Lüders, Dr. Stefan Bareiss Robert Bosch GmbH “Bosch AdBlue Dosing Technology for EU6 RDE and Beyond”, Berlin 2018 2. Tobias Bayer, Dirk Samuelsen, Stefan Bareiss, Marc Chaineux, Robert Bosch GmbH Stuttgart: “Double Injection SCR – Bosch’s development for future emission regulations”, FKFS Stuttgart 2018 3. Official Journal of the European Union: "Regulation No 83 of the Economic Commission for Europe of the United Nations (UN/ECE)" Revision 3, 2006

135

Modular HD – Exhaust gas treatment system with autarcic thermal management for high urban NOx conversion Klaus Schrewe, Dr. Bernd Maurer, Dr.-Ing. Christoph Menne, Ingo Zirkwa HJS Emission Technology GmbH & Co. KG

Modular HD – Exhaust gas treatment system with autarcic thermal management for …

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 M. Bargende et al. (Hrsg.), 19. Internationales Stuttgarter Symposium, Proceedings, https://doi.org/10.1007/978-3-658-25939-6_12

137

Modular HD – Exhaust gas treatment system with autarcic thermal management for …

Abstract With the European Directive 2008/50 EC on Ambient Air Quality [1], the European Commission is pursuing the goal of ensuring that the annual average ambient air quality limit of 40 µg/m3 for NO2, which has been in force since 2010, is ensured. Compliance with this directive is a major challenge for many municipalities and federal states. Especially in urban areas, NO2 limit values are often exceeded. Transient low-load operation with low exhaust temperatures and sometimes high NOX raw emissions during the acceleration phases also pose special challenges for the complex exhaust aftertreatment systems of buses and commercial vehicles. With regards to the problem of Real Driving Emissions (RDE) in the urban environment and in light of the expected future exhaust emission legislation, HJS has further developed the well-known SCRT technology and added important technology components. In addition to highly efficient catalytic converters and optimized NH3 mixture preparation, the focus was placed on autonomous thermal management technologies and corresponding control algorithms in order to achieve very high NOX conversion rates even in demanding urban operation. The first application of this modular HJS heavy duty exhaust aftertreatment platform is in city buses. Diesel particulate filters are well established to reduce particulate emissions from diesel engines and are being used in most new applications already. The guidelines for retrofitting diesel buses as part of the "Sofortprogramm Saubere Luft 2017 – 2020" have defined emission limits and testing boundary conditions which even go beyond the IUC regulations for Euro VI buses, especially in inner-city operation. In order to meet the emission requirements of this directive in real city bus operation, a thermal management is required in addition to an optimised EURO VI exhaust gas aftertreatment with DOC, DPF and SCR in order to reliable operate the SCR system in an efficiency-optimised temperature window even at the lowest speed profiles. The modular heavy duty exhaust aftertreatment system with independent thermal management presented in this paper aims to significantly increase the efficiency of the SCR system in transient inner-city operation. The exhaust gas temperature in the low load range is raised by up to 50 °C for this purpose. Central elements for this are a quick, electrically operated exhaust flap along with an electrically heated diesel oxidation catalyst. By the exhaust flap, the exhaust back pressure is slightly increased in certain low load driving situations. The electrically heated catalyst arranged in front of the particulate filter increases the exhaust gas temperature and the engine load via the additional load on the alternator.

138

Modular HD – Exhaust gas treatment system with autarcic thermal management for … The HJS aftertreatment control unit ACU processes the measured values such as NOX, temperatures, pressures, exhaust gas mass flow, status of the electrical system, etc. and controls the actuators in such a way that the best possible NOX reduction is achieved with minimum fuel consumption penalty. The system is also fully diagnosable. The system contributes immediately to the reduction of NOX emissions and thus to improved air quality in urban areas. The modular concept and the smart CAN bus and sensor-based control also qualifies the HJS system for other commercial vehicle applications beyond city bus applications and EURO VI legislation. For the first time, engine heating measures can be enhanced by engine operating point independent thermal management measures to help reliably meet future RDE-focused legislation in critical city operation. In the following, the performance of the system will be demonstrated as an example for retrofitting a city bus application.

1 Technical requirements for DeNOX retrofit systems for city buses When retrofitting Euro IV, V and EEV city buses to Euro VI level, the real operating conditions must be taken into account. City bus operation is characterized by frequent stops, relatively low engine loads and correspondingly low exhaust gas temperatures. In Germany, there has been a corresponding guideline since mid-2018 that takes these demanding requirements into account. The following boundary conditions were defined by legislation [3]: ● The test run has to consider inner-city and representative public transport driving including bus stops (15s opening of the door). ● The average speed of each measurement run shall be between 10 and 30 km/h. ● Measurements with cold and warm exhaust aftertreatment systems must be taken into account. During the certification measurements, the raw gas emissions of the engine and the tailpipe emissions downstream the exhaust aftertreatment system are measured. As the use of a diesel particulate filter is mandatory and effectiveness is guaranteed by manufacturer's declaration and measurements from existing certifications, the focus of the evaluation is on nitrogen oxide emissions (NOX emissions).

139

Modular HD – Exhaust gas treatment system with autarcic thermal management for …

Figure 1: Classification of route-related NOX emissions and certification limit [2]

Two conditions are essential for a successful test: ● NOX reduction rate, comparing raw to tail pipe emissions is exceeding 85%. ● The classified route-related emissions are adhered to (Fig. 1)

2 Concept development 2.1 Design The Euro IV, V and EEV bus types on the market are equipped with partial or full-flow particulate filters, SCR systems or SCRT systems as standard exhaust after-treatment. The existing exhaust system is always replaced by a combination system (SCRT with particulate filter or SMF®) (Fig. 2), which must always be integrated into the given installation space without vehicle adaptations (Fig. 3).

140

Modular HD – Exhaust gas treatment system with autarcic thermal management for …

Figure 2: SCRT with SMF®

Figure 3 : Integration into the vehicle

The above example shows how compact the system is. Due to an optimized design it was possible to integrate the additional functions of dosing and mixing AdBlue®, SCR module as well as the required sensors and actuators into the installation space of the original CRT system (Fig. 4). The volume of the CRT system (DOC and SMF®) was not reduced. The SCR assembly was designed with two SCR substrates in parallel. As a result of extensive 3D CFD flow simulations, both minimal pressure loss and excellent NH3 uniformity were achieved.

Figure 4: Components of the SCRT system

The monitoring of the CRT system is performed by the engine control unit. For the additional components, in particular the AdBlue® dosing system, the control is taken

141

Modular HD – Exhaust gas treatment system with autarcic thermal management for … over by a control unit developed exclusively by HJS. The adaptive functions developed by HJS enable primarily NOX sensor-based dosing with high accuracy. The space velocity of the SCR catalytic converter is calculated from the exhaust gas mass flow signal. In addition, a NOX mass flow is determined from the signal of the NOX raw gas sensor. The required dosage quantity is determined as a function of room velocity / exhaust gas temperature and NH3 loading condition of the catalytic converter.

2.2 Integration of the system Depending on the serial exhaust gas aftertreatment concept of the bus mainly two different types of retrofit systems have to be integrated. OE emission concepts using a DOC-DPF solution are supplemented with an autarcic SCR system keeping all the diagnosis of the CRT system in the engine ECU. Functionality of the SCR system is completely controlled by the HJS ACU, including system diagnostics. If the bus is serially equipped with a SCR or SCRT system, the higher NOX reduction of the new system has to be integrated into the existing diagnosis and inducement of the OE system. This guarantees, if the serial Adblue dosing components can be maintained that all diagnosis of the original bus are still the same at a much higher NOX reduction. Main difference between an autarcic and an integrated SCR(T) system is not given by the exhaust gas streamed componets of the EGT system but much more by the control layout and strategy. Examples and the range of application of these two types of retrofit SCR(T) systems are shown in Figure 5.

Figure 5: Different types of retrofit SCR(T) systems for city busses

142

Modular HD – Exhaust gas treatment system with autarcic thermal management for … To further improve NOX reduction of the retrofitted systems they include an electrical heated DOC and, if not part of the OE exhaust gas treatment concept, an electronically controlled flap for thermal management.

2.3 Thermal management The evaluation of real operating data shows exhaust gas temperatures in the range of 220°C to 260°C upstream of the SCR system (Fig. 6). In this temperature range, the SCR system achieves a sufficiently high NOX reduction. However, the data also shows operating ranges with temperatures below 200 °C, especially during the warm-up phase but also at low ambient temperatures. In order to achieve the best possible NOX conversion for these boundary conditions, the DOC was designed with a heating element. This converts electrical power from the vehicle's electrical system into heat and thus increases the exhaust gas temperature. Additionally the heating catalyst minimizes the cooling of the SCR system. Fig. 6 clearly shows that the heat up time is shortened when the heating element is switched on and that the exhaust gas temperature of 200 °C is reached about 10 minutes earlier. Furthermore, activation of the heating element prevents the SCR system from cooling down, so that the tailpipe gas temperature is stabilized at a level of approx. 250 °C.

Figure 6: Temperature curve with and without activation of the heated catalytic converter vs. time [sec.]

As a further thermal management measure, an optional exhaust gas flap can be integrated into the system. The cooling down of the SCRT system can be greatly reduced in overrun phases. Throttling in the lower load range of driving operation, increases the gas exchange losses and the resulting higher engine load also raises the exhaust gas

143

Modular HD – Exhaust gas treatment system with autarcic thermal management for … temperature. The overall configuration of such a SCRT system with thermal management is shown in Fig. 7.

Figure 7: SCRT system with two-stage active thermal management

Intelligent, integrated control of the heating element and the exhaust flap enables a CO2optimised increase in NOX reduction. The electrically adjustable exhaust flap for increasing the exhaust back pressure is located downstream of the SCR system. The exhaust gas flap is used to regulate the exhaust gas back pressure to an increased value, e.g. 130 mbar. The back pressure increase results in an increase in the exhaust gas temperature of 15 to 20 °C. Increasing the exhaust gas temperature is not sensible or necessary in all driving situations. Therefore the control of the exhaust flap depends on certain criteria. The heated DOC (eDOC) is also only switched on under certain boundary conditions. The most important variable for controlling the components is the exhaust gas temperature, whereby both the temperature upstream and downstream of the SCR catalytic converter is taken into account. As shown in Fig. 8, separate switch-on and switch-off thresholds for the temperature are defined for the exhaust gas flap and the heating catalytic converter. The evaluation of the temperatures before and after SCR allows the differentiation between cold and warm SCR systems, to enable different operating modes of the system [4].

144

Modular HD – Exhaust gas treatment system with autarcic thermal management for …

Figure 8: Control logic for exhaust flap and electrically heated catalyst

Engine operating states such as idling or highly transient accelerations additionally influence the control of the system. The voltage level of the electrical system is permanently monitored to prevent overloading due to excessively long activation of the heating element. Just like the heated catalyst, the function of the exhaust flap is continuously monitored. In addition to electrical diagnostics, numerous plausibility checks and function checks are carried out to check all actuators and sensors. The control of electrical components with high power consumption has been integrated in the HJS control unit for many years and has been proven in the field by numerous SMF-AR® systems [4]. However, these systems only activate the components for a relatively short time (approx. 3 min) with a long delay until they are switched on again (approx. 2 h). For the modular HJS exhaust aftertreatment system with active thermal management, a significantly higher availability is required, as shown in Fig. 6. For this reason, the heating element control was implemented using additional components (Fig. 9).

145

Modular HD – Exhaust gas treatment system with autarcic thermal management for …

Figure 9: Control structure for electrically heated catalyst

The electrically heated catalyst consumes up to 1.6 kW electrical power at 28 V. For this reason, great importance was given to a reliable and robust diagnosis that immediately deactivates the heating element in the event of relevant faults and thus puts the system in a safe operating state. After checking the overall system status at engine start, the safety relay is activated and then, if necessary, the operating relay enables the corresponding operating status in order to increase the exhaust gas temperature. The system is continuously monitored. To avoid unnecessary load on the electrical system, the operating relay is switched off again when an exhaust gas temperature of approx. 250 °C is reached.

3 Functional Testing 3.1 RDE measurements An exemplary RDE measurement was carried out with an MAN solobus with a 206 kW engine. After reaching the required operating temperature, AdBlue® is dosed according to the NOX raw gas emissions (Fig. 10).

146

Modular HD – Exhaust gas treatment system with autarcic thermal management for …

Figure 10: Temperature and dosing profiles during certification test

With the HJS system, a NOX conversion rate of 92 % could be achieved in the context of the RDE test drive with a city bus, which is originally equipped with DOC and particle filter only. The distance-related NOX emissions are a maximum of 1 g/km even at low driving speeds or in stop-and-go operation (Fig. 11). The results achieved clearly exceed the requirements of the public transport retrofit legislation and are thus on a par with buses with EURO VI original equipment or better.

Figure 11: Classification of measured route-related NOx emissions [g/km] from certification measurements (dashed blue) vs. limit (red)

Within the framework of the RDE test runs and the approval tests in real city bus operation, a portable emissions measurement system (PEMS) has been used to measure NOX reduction rates between 87% and 98% with various engine sizes from 6.4 l to 12 l displacement.

147

Modular HD – Exhaust gas treatment system with autarcic thermal management for …

3.2 Deposit testing HJS's many years of extensive experience with SCR systems highlight that one of the biggest challenges for low-maintenance, robust operation with high conversion rates is the prevention of deposits from AdBlue®. During the development of SCR systems, extensive flow simulations for excellent uniformity and AdBlue® mixture preparation were carried out. Thus, numerous problem areas of deposits are avoided early in the development process and the number of complex hardware iterations are minimized. However, not all boundary conditions, such as installation situation or ambient temperature, can always be considered accurately in simulation runs. It is therefore necessary to regularly inspect the system for deposits during the development phase. As an example, the relevant exhaust gas routing in the exhaust system of two test vehicles is shown; both vehicles were used in low-load operation with low exhaust gas temperatures and reflect a critical operating profile. Even after six months of critical operation, only minor deposits in the form of non-critical, light streaks are visible on the pipe wall. These are due to the swirl of the exhaust gas (Fig. 12).

Figure 12: Analysis of AdBlue® treatment field test vehicle [6]

3.3 Field experience The field test vehicles are equipped with measurement data acquisition systems that continuously record and transmit the operating parameters. With the help of this data, the necessary application parameters can be adjusted and optimized promptly if necessary. Fig. 13 shows an example of an evaluated measurement run. In this measurement, the exhaust gas temperature without a heating catalyst before SCR (TusSCR) exceeds

148

Modular HD – Exhaust gas treatment system with autarcic thermal management for … 200 °C after 15 minutes. In the standard version, the temperature rise is significantly faster with an active heating element, then a temperature of 200 °C in front of the SCR is reached after a period of less than 10 minutes (as shown in Fig. 6). Once the minimal temperature thresholds have been exceeded, AdBlue® dosing is started. The dosing quantity is determined with the available sensors (as described above). The current NOX conversion rate is then calculated with the additional signal from the NOx sensor after the SCR catalytic converter, which is not required for the system functionality of the exhaust aftertreatment system. The cumulated conversion rate is based on the previous current NOx conversion rates and results in the integral NOX reduction rate at the end of the test run, in this test run with cold start of 90 %. Since the system does not cool down significantly in an idle phase (approx. 5 min), the dosing is still active in this range and the NOX conversion rate is very high with > 95% during idling, marked with a grey bar in Figure 13.

Figure 13: Measurement run (cold start) over approx. 60 min

4 Summary HJS has developed a modular exhaust aftertreatment kit for commercial vehicles. The HJS modular system is characterized by optimized catalyst coatings, elaborately optimized AdBlue® mixture preparation and, in particular, stand -alone, active thermal management. The results presented were achieved in a critical city bus application, which poses special challenges for exhaust aftertreatment due to the highly transient

149

Modular HD – Exhaust gas treatment system with autarcic thermal management for … Stop and Go operation and the low exhaust temperature levels. NOX conversion efficiencies greater than 90% could also be realized for cold started cycles and also at low ambient temperatures. The extremely compact HJS system can be integrated into the existing installation space of a EURO IV or EURO V bus designed for DPF or SCR only. The additional components for AdBlue® dosing are controlled by a HJS control unit. Since operating conditions in public transport are often characterized by the fact that the exhaust gas temperature is too low for a high NOX conversion, the DOC was designed with an electrical heating element as an eDOC. This eDOC is controlled under the aspect of operational safety and the load limits of the vehicle electrical system. With the help of the eDOC the heat up time is shortened so that NOX conversion can start much earlier. By using a fast electrically operated exhaust gas flap, the exhaust gas temperature can be raised by a further 15 to 20°C and the temperature loss of the SCRT system in overrun phases can be significantly reduced. The RDE measurements achieved with a solo bus are well above the legal requirements both in the averaged NOx reduction rate and in the classified route-related emissions. The combination of heat exchanger and exhaust flap enables high NOX reduction of more than 85% even at very low engine loads and ambient temperatures. The optical findings of the various field test vehicles show that the system has no critical deposits. Evaluations of measurement data from these vehicles confirm the high NOX conversion properties of the exhaust system.

References 1. 96/62/EC and 2008/50/EC 2. NOX reduction system with increased reduction capacity: 2018-07-26 Information event Berlin “LoMo DkV DiBusse” 3. BAnz AT 28.03.2018 B6 4. Dr. B. Maurer, K. Schrewe, I. Zirkwa, HJS Emission Technology GmbH & Co. KG: "TMT - Optimization of NOX Emissions from Vehicles with SCR Systems in Urban Areas" 13th FAD Conference 5. D. Lamotte, K. Schrewe, J. Witte: "Exhaust Flow Sensor for Load Detection of Parallel DPF Systems" 10th FAD Conference 6. D. Lamotte, K. Schrewe, I. Zirkwa: "Upgrade of EU IV / V city buses to EU VI emission level under real operating conditions" 16th FAD Conference

150

New experimental insights in AdBlue-spray/ wall interaction and its impacts on EGT system design David Schweigert, Björn Damson, Hartmut Lüders, Carsten Becker Robert Bosch GmbH, Powertrain Solutions, Advanced Engineering Exhaust Systems, Wernerstraße 51, 70469 Stuttgart, Germany Olaf Deutschmann Karlsruhe Institute of Technology, Institute for Chemical Technology and Polymer Chemistry, Engesserstraße 18/20, 76131 Karlsruhe, Germany

New experimental insights in AdBlue-spray/ wall interaction and its impacts on …

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 M. Bargende et al. (Hrsg.), 19. Internationales Stuttgarter Symposium, Proceedings, https://doi.org/10.1007/978-3-658-25939-6_13

151

New experimental insights in AdBlue-spray/ wall interaction and its impacts on …

Abstract With the introduction of new emission legislation (e.g. EU6-RDE, LEVIII/SULEV) and related nitrogen oxides (NOx) emission limitations, the automotive industry is facing new challenges concerning efficient and robust exhaust gas treatment (EGT). For conversion of NOx, selective catalytic reduction (SCR) via AdBlue has prevailed in Diesel exhaust gas treatment. The preferably complete reduction of NOx in EGT systems still remains one of the most important and challenging tasks of today’s powertrain development. For fast warm-up and early catalyst light-off, for quite some time now SCR systems are positioned close to the engine, leading to a very compact EGT design with strongly confined space. Simultaneously, fuel-efficient engine calibration (with respect to a CO2 optimization) results in higher AdBlue dosing amounts and demands for new innovative approaches to prevent excessive wall wetting and, thus, formation of undesired solid AdBlue by-products. While the optimization of injector and compact exhaust/mixing geometries is commonly considered, the role of the AdBlue-impinged wall has not been investigated sufficiently. The wall material-relevant requirements and impacts on EGT system design are not fully understood and deserve further studies. Through the build-up of a basic laboratory test bench and advanced experimental approaches, the impact of surface and material properties on spray/wall interaction phenomena is comprehensively investigated. Remarkable interdependencies were found and described quantitatively for the very first time. These new spray/wall interaction findings reveal high potential regarding the optimization of SCR system robustness and efficiency.

1 Introduction Future nitrogen oxides (NOx) limitations of Diesel engines demand for new innovative technical approaches to undercut the legislative emission limits (EU6-RDE, LEVIII/SULEV). As already demonstrated in FKFS proceedings of previous years, through the smart combination of internal engine measures and advanced exhaust gas treatment NOx emissions can effectively be reduced [1-3]. These days the use of a selective catalytic reduction (SCR)-based exhaust gas treatment (EGT) system is indispensable and has prevailed in Diesel applications across all vehicle segments. AdBlue – a 32.5% urea water solution (UWS) – is injected in front of the SCR catalyst. The appropriate preparation of ammonia (NH3) through evaporation, thermal decomposition and subsequent hydrolysis of the injected reducing agent is a challenge of increasing difficulty. Fuel-efficient engine calibration demands for higher AdBlue dosing rates, whereby exhaust treatment systems positioned close to the engine

152

New experimental insights in AdBlue-spray/ wall interaction and its impacts on … additionally impose restrictions on available preparation and mixing space. To fulfill these challenging tasks, Bosch supports EGT system engineering and acts as an integration partner for vehicle manufacturers. For further reduction of NOx, topological considerations on system design level, like adding a second dosing device is becoming increasingly popular. The double injection (DI)-SCR system, which is shown in Fig. 1, further enhances the EGT system’s capability to reduce NOx tailpipe emissions. The separate diagnosis of the two SCR catalysts enables proper dosing [4]. Furthermore, spray surface load (impinging mass per area) can potentially be reduced by distributing the needed dosing amount at distinct positions. The injection of UWS into the mixing section of the exhaust pipe and the related physicochemical effects are schematically shown in Fig. 1. Birkhold summarized these fundamental processes and implemented his findings in a simulation environment [5].

Figure 1: DI-SCR system and related physicochemical effects in SCR mixing section (adapted from Birkhold [5])

For evaporation and subsequent thermolysis of liquid droplets in the hot exhaust gas, time-scales of evaporation and thermolysis are usually too long to avoid the impingement of droplets on the wall. Spray/wall contact is inevitable, especially at cold start conditions and highly transient engine loads. Consequently, droplets adhere to the wall (deposit regime) and eventually cumulate to a liquid wall film. While exhaust gas flow pass, the film is transported into regions with higher wall temperatures. Dissolved isocyanic acid (a reaction product of urea thermolysis) may be concentrated in the wall film, which potentially leads to polymerization. Hereby, high-temperature resistant species, like cyanuric acid and other undesired reaction products (ammeline, ammelide, melamine), form [6, 7]. The accumulation of these undesired deposits potentially harm

153

New experimental insights in AdBlue-spray/ wall interaction and its impacts on … system performance (e.g. the decrease of NOx reduction or the increase of ammonia slip) and can lead to a rise of back pressure up to a blockage of the exhaust pipe. In experiments on a hot gas test bench, wall film and deposit formation on a generic wall has been visually observed, as shown in Fig. 2. In these first experiments, mixer wall material shows strong influence on tolerable maximum dosing amount until wall film formation and subsequent deposit accumulation.

Figure 2: Wall film formation and deposit build-up on a generic mixer wall at Bosch flow laboratory

This finding encouraged us to investigate the heat transfer characteristics upon spray impingement and their dependency on material related effects in detail. Liao et al. [8] already performed first investigations on SCR spray heat transfer. In our studies, we extend these important findings, with respect to how heat transfer alters with changes of injector operating conditions and wall material. With the described experimental setup and the presented results, the fundamental understanding of the underlying physics during spray/wall interaction is enhanced. The acquired knowledge can accompany EGT development and can be considered for the optimization of SCR system design.

154

New experimental insights in AdBlue-spray/ wall interaction and its impacts on …

2 Experimental setup For the investigation of the evolving heat transfer characteristics, a basic experimental setup has been built-up as shown in Fig. 3. The spray impingement of a Bosch serial 3hole injector on hot plates at different wall temperatures is captured via infrared thermography (IRT) and simultaneous optical high-speed measurements. The injector op5 𝑏𝑎𝑟 and an injection duration of 𝑡 5 𝑚𝑠. erates at an injection pressure 𝑝 The plates can be exchanged for the ease of investigating different plate materials (𝑑 0.5 𝑚𝑚). The sample plates are heated in an oven and are rapidly transported into the test cell. The injection is triggered, as soon as wall temperature falls below the initial of the plate can be calcuwall temperature 𝑇 . The spray-induced heat loss Δ𝑄 lated with the information about the transient temperature field from the IRT recordings. Information all over the relevant temperature range can be obtained by measuring the heat loss at distinct wall temperatures (temperature steps Δ𝑇 10 𝐾).

Figure 3: Experimental setup

Via introduction of an evaporated mass fraction 𝑓 , the influence of injector operating conditions and the change of wall material, can be quantified: 𝑓

𝛥𝑄

/ 𝑚

𝛥ℎ

,

𝑐 , 𝛥𝑇

(1)

The heat loss induced by the impinging spray 𝛥𝑄 is normalized by the needed energy for complete evaporation of the injected mass 𝑚 . The corresponding evaporation enthalpy is given by 𝛥ℎ , , heat capacity 𝑐 , of the liquid 𝑙 and the temperature difference for heat up from injection temperature until saturation temperature of the liquid.

155

New experimental insights in AdBlue-spray/ wall interaction and its impacts on …

3 Results and discussion In this chapter, the results of the measured heat transfer characteristics during spray/wall interaction are given. Furthermore, the initial findings concerning the influence of material on heat transfer are discussed.

3.1 Boiling heat transfer during spray/wall contact In Fig. 4a and Fig. 4b the wall film formation after spray impingement and the corresponding temperature field is shown. The IRT picture is mirrored, in order that it corresponds with the orientation of the high-speed recordings. Even through the 0.5 𝑚𝑚 thick steel plate, the droplet and wall film evaporation is clearly visible. Note, that the imaging at the rear is blurred according to the plate thickness.

Figure 4a: Visualization of wall film formation captured via high-speed camera (water spray impingement on steel at 𝑇 180 °𝐶)

Figure 4b: Field of temperature drop Δ𝑇 /𝐾 captured via IRT (water spray impingement on steel at 𝑇 180 °𝐶)

The transient temperature field, which is exemplarily shown in Fig. 4b, is post-processed to calculate the heat loss during spray/wall impingement. Measurement results of heat loss and corresponding evaporated mass fraction for spray impingement on ferritic steel are given in Fig. 5. For better physical interpretation pool boiling regimes are additionally depicted. Furthermore, kinetic effects of the impinging droplets contribute to the characteristics of the boiling curve. These are typically classified into different droplet/wall interaction regimes (deposition, splash, rebound, thermal breakup).

156

New experimental insights in AdBlue-spray/ wall interaction and its impacts on …

Figure 5: Spray/wall interaction heat loss and evaporated mass fraction with corresponding boiling regimes (UWS on ferritic steel)

These two superposing effects finally lead to the boiling curve specifically valid for urea water solution (UWS) spray impingement shown in Fig. 5. The boiling curve can be divided into four boiling regimes, namely convective boiling (I), nucleate boiling (II), transition boiling (III) and film boiling (IV). The regime boundaries can only be estimated; different regimes occur simultaneously because of the polydisperse nature of the spray. A characteristic temperature which marks the regime boundary between nucleate and transition boiling is given by the temperature of critical heat flux 𝑇 . Even in this point, the impinging mass is not evaporated completely. The maximum 𝑇 60% , indicating, that part of the mass reevaporated mass fraction is 𝑓 mains deposited or is expelled without being evaporated on the wall. Another characteristic quantity is the so-called Leidenfrost temperature 𝑇 , marking the regime boundaries between III and IV. The Leidenfrost point is usually defined for sessile droplets above hot walls. It describes the wall temperature, where a droplet initially levitates on a vapour cushion without actual wall contact. The vapour phase between the hot wall and the liquid acts as a thermally insulating layer and drastically reduces heat transfer from the wall to the liquid. As it can be seen from the measurement depicted in Fig. 5 only approximately 10% of the impinging mass is evaporated through the wall heat flux, whereas the rest of the mass is expelled from the prompt formed vapour cushion. Concerning the operation of UWS injection in mixing sections of SCR systems, the Leidenfrost effect could be useful in real applications. With the knowledge about these characteristics, two discrete wall temperatures give the optimum for SCR dosing with regard to spray/wall heat transfer:

157

New experimental insights in AdBlue-spray/ wall interaction and its impacts on … – Wall temperatures at the temperature of critical heat flux 𝑇 𝑇 ) describe the temperature optimum, where the highest amount of injected mass can be evaporated. Adversely, through the high cooling rates of the injected mass, wall is cooled fastest, wherefore sufficient heat input through convection of hot exhaust gas or wall heat conduction from hotter areas has to be ensured. 𝑇 ) delay the wetting – Wall temperatures above the Leidenfrost temperature (𝑇 of the wall and deposit formation is potentially mitigated. If it is possible to keep the wall temperature (mixer, tailpipe) above the Leidenfrost point, either by increasing the heat transfer from the exhaust gas or by selecting a suitable material, deposits at the mixing plates can be avoided. Additionally, the mixer acts as a source for secondary atomization, tending to lead to smaller droplets after the impingement. The expelled mass is distributed to downstream-located areas, ensuring an increase of available area for evaporation. Furthermore, the residence time of the injected droplets in the exhaust stream is longer, providing additional time and exhaust enthalpy for evaporation and thermolysis. These two characteristic temperatures - marking the regime boundaries - are very sensitive to the operating conditions of the injector and the wall material.

3.2 Influence of material upon heat transfer during spray/wall interaction With the change of the material, literature reveals partially significant changes in heat transfer. Because of the lack of knowledge, concerning which material property is the most affecting parameter on heat transfer, we start to investigate different materials upon their thermal behavior during spray/wall interaction. Varying the thermal properties in a broad range potentially affects heat transfer characteristics in a preferable way. The resulting boiling curves at spray impingement with urea water solution are determined with the experimental setup described in chapter 2. The materials under investigation are given in Table 1, together with their associated Leidenfrost temperatures, which are extracted from the measured boiling curves. Table 1: Measured Leidenfrost temperatures 𝑇 of urea water solution spray impingement on different wall materials Material

𝑇 /°𝐶

158

Ferritic steel (1.4509) 322

Austenitic steel (1.4301) 328

Iron (Fe)

Tungsten (W)

Nickel (Ni)

Silver (Ag)

359

320

295

296

New experimental insights in AdBlue-spray/ wall interaction and its impacts on … Taking into account the corresponding thermal conductivity of the investigated materials, their dependency is given in Fig. 6. The dashed line approximates the assumed linear dependency of the thermal conductivity and the measured Leidenfrost temperature. It is clearly notable, that thermal conductivity is not the sole parameter determining the emerging Leidenfrost temperatures. Measurements on iron and nickel show significant different Leidenfrost temperatures than expected from the linear assumption. It still has to be determined, which material property is the main influencing parameter during spray impingement. Further potential influencing properties can be classified into: – surface chemistry (contaminants, oxide layer), – surface energy (contact angle), – surface topography (roughness, structuring), – thermal properties (heat capacity, thermal conductivity, thermal effusivity) and – other bulk properties (density, porosity, etc.).

Figure 6: Leidenfrost temperature dependent on thermal conductivity for spray/wall interaction with UWS on various materials

The measurements show, that the Leidenfrost temperature is strongly influenced by the choice of material and varies in a broad range. Finding a suitable material, which lowers the Leidenfrost temperature could lead to a remarkable improvement concerning system robustness. All the aforementioned properties are likely to influence the thermal and kinetic characteristics during spray impingement. The harsh conditions in the exhaust system additionally complicate the choice of appropriate materials (resistant against high temperatures, thermal shock, corrosion, etc.). Future studies are dedicated to further understand the underlying physics and their strong interaction with changing wall material.

159

New experimental insights in AdBlue-spray/ wall interaction and its impacts on …

4 Impacts on EGT system design With the focus of improving NOx reduction capabilities especially at cold start conditions, exhaust system developers tend to design very compact EGT systems close to the engine. The confined space for ammonia preparation limits the available impingement areas and adversely affect NH3 homogenization in front of the SCR catalyst. Additionally, the risk of wall film and deposit formation is higher. With the basic findings of chapter 3 – the heat transfer characteristics and their sensitivity on injector operating conditions as well as wall material - system robustness can potentially be improved. Since the experimental test bench operates at ambient conditions, spray/wall interaction is hardly affected by gas flow or evaporation in cold, quiescent air. Almost exclusively all of the injected mass impinges on the plate target, which enables the quantified description of the wall evaporated mass. Classic droplet/wall regime maps, e.g. introduced by Bai and Gosman or Kuhnke [9, 10], show distinct boundaries according to the wall temperature and the droplet properties (droplet diameter, droplet velocity). For the investigated AdBlue spray impingement a gradual transition between the regimes can be stated. The emerging heat loss during spray/wall interaction is highly dependent on wall temperature. The measurement results can serve as a data basis for implementation of heat transfer modeling and should be taken into account for better simulation accuracy. Exemplary considering a DI-SCR system, as shown in Fig. 1, the engine close mixing section (e.g. Bosch Helix Mixing Section concept (HMS), as shown in Fig. 7) usually experiences other interaction regimes than the downstream located mixing areas.

Figure 7: Schematic illustration of Bosch Helix Mixing Section concept

160

New experimental insights in AdBlue-spray/ wall interaction and its impacts on … At higher engine loads with exhaust gas temperatures above Leidenfrost point, with 10 % of the impinging mass, a considwall evaporation of only approximately 𝑓 erable amount of droplets rebound at an instantly formed vapour layer. Consequently, NH3 distribution in front of the SCR catalyst and the corresponding uniformity index (UI) can decrease. Mixing sections are likewise too short for complete thermal decomposition, wherefore liquid urea water solution may hit the SCR catalyst. This results in ammonia hotspots and potential ammonia slip. Henceforth, with the introduced evapoa comprehensible explanation can be given. With the rated mass fraction 𝑓 knowledge about the aforementioned heat transfer characteristics, the mixing section design has to be reconsidered concerning an optimized usage of the available exhaust enthalpy for evaporation. 𝑐𝑜𝑛𝑠𝑡 For example, changing the injection frequency for a given dosing rate 𝑚 shows significant influence on evaporated mass fraction, especially at lower wall temperatures. Results indicate faster wall cooling (as well as a higher evaporated mass fraction) with higher injection frequencies. Reversely, at lower frequencies a higher amount of the injected mass is expelled (splash, thermal breakup). This finding could make new injection strategy applications possible. Furthermore, it provides explanatory approaches, when selecting injectors with different static mass flow for the specific needs of a certain SCR mixing section. Changing the wall material could be another effective way to positively alter heat transfer characteristics. Lower the Leidenfrost temperature by finding a suitable wall material or coating could further enhance the system robustness against deposits and would allow higher dosing amounts. First results indicate great potential, although appropriate materials are still to be found.

5 Conclusions The selective catalytic reduction of NOx via AdBlue as a precursor for ammonia has prevailed in Diesel exhaust gas treatment. The distribution and evaporation of AdBlue in exhaust systems can be challenging especially at high NOx engine-out emissions, which are typical for fuel-efficient engine calibration. For better understanding of the related spray/wall interaction phenomena in SCR mixing sections, a basic laboratory test bench has been built-up. Single spray impingement of urea water solution is investigated by infrared thermography and simultaneous optical high-speed recordings. Through introduction of an evaporated mass fraction, heat transfer can be characterized. The results reveal that heat loss is strongly related to the wall temperature. Heat transfer is reduced by almost one order of magnitude in the film boiling regime (above the Leidenfrost temperature 𝑇 ) compared to the heat transfer around the temperature of critical heat flux 𝑇 .

161

New experimental insights in AdBlue-spray/ wall interaction and its impacts on … Changing the material of the impingement target lead to significant shifts of the boiling regimes and indicate high potential regarding improvement of system robustness and efficiency. Leidenfrost temperature is not governed solely by the thermal conductivity. The influence of other surface and material properties would deserve further studies to further understand the underlying physics and derive additional consequences for EGT system design. The basic description of the heat transfer characteristics during spray impingement enables improvement of the system knowledge and supports EGT system developers in finding new innovative approaches for the optimization of AdBlue preparation.

Bibliography 1. Naber, D., Kufferath, A., Krüger, M., Scherer, S., Schumacher, H., & Strobel, M. (2016). Solutions to fulfill “Real Driving Emission (RDE)” with diesel passenger cars. In 16. Internationales Stuttgarter Symposium, 285–285. Springer, Wiesbaden. 2. Naber, D., Kufferath, A., Krüger, M., Maier, R., Scherer, S., & Schumacher, H. (2017). Measures to fulfill “real driving emission (RDE)” with Diesel passenger cars. In 17. Internationales Stuttgarter Symposium, 423–446. Springer Vieweg, Wiesbaden. 3. Naber, D., Bareiss, S., Kufferath, A., Krüger, M., & Schumacher, H. (2018). Measures to Fulfill "Real Driving Emission (RDE)" with Diesel Passenger Cars. In 18. Internationales Stuttgarter Symposium 257-278. Springer Vieweg, Wiesbaden. 4. Bayer, T., Samuelsen, D., Bareiss, S., & Chaineux, M. (2018). Double injection SCR–Bosch’s development for future emission regulations. In 18. Internationales Stuttgarter Symposium, 579–593. Springer Vieweg, Wiesbaden. 5. Birkhold, F. (2007). Selektive katalytische Reduktion von Stickoxiden in Kraftfahrzeugen: Untersuchung der Einspritzung von Harnstoffwasserlösung. PhD thesis. Universität Karlsruhe (TH). Shaker. 6. Brack, W., Heine, B., Birkhold, F., Kruse, M., Schoch, G., Tischer, S., & Deutschmann, O. (2014). Kinetic modeling of urea decomposition based on systematic thermogravimetric analyses of urea and its most important by-products. Chemical Engineering Science, 106, 1–8. 7. Börnhorst, M., Langheck, S., Weickenmeier, H., Dem, C., Suntz, R., & Deutschmann, O. (2018). Characterization of solid deposits from urea water solution injected into a hot gas test rig. Chemical Engineering Journal.

162

New experimental insights in AdBlue-spray/ wall interaction and its impacts on … 8. Liao, Y., Eggenschwiler, P. D., Furrer, R., Wang, M., & Boulouchos, K. (2018). Heat transfer characteristics of urea-water spray impingement on hot surfaces. International Journal of Heat and Mass Transfer, 117, 447–457. 9. Bai, C., & Gosman, A. D. (1995). Development of methodology for spray impingement simulation. SAE transactions, 550–568. 10. Kuhnke, D. (2004). Spray/wall interaction modelling by dimensionless data analysis. PhD thesis. Technische Universität Darmstadt. Shaker.

163

Environmental model extension for lane change prediction with neural networks Martin Krüger, Anne Stockem Novo, Till Nattermann, Manoj Mohamed ZF Group, Automated Driving & Integral Cognitive Safety, 40547 Düsseldorf, Germany Torsten Bertram TU Dortmund University, Institute of Control Theory and Systems Engineering, 44227 Dortmund, Germany

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 M. Bargende et al. (Hrsg.), 19. Internationales Stuttgarter Symposium, Proceedings, https://doi.org/10.1007/978-3-658-25939-6_14

165

Environmental model extension for lane change prediction with neural networks

1 Introduction Driven by the desire of improving traffic safety, traffic efficiency and a better utilization of the time people spend in traffic, the development proceeds from Advanced Driver Assistance Systems (ADAS) towards fully automated systems. ADAS systems are categorized as level 2 systems according to the Society of Automotive Engineers (SAE) [1] definition, on a classification scheme from level 0 to 5 where level 5 is a fully automated system. With increasing automation level the complexity rises, due to the piecewise shift of responsibility towards the automated driving system. In accordance with Tas et al. [2] any structure of an automated driving system is composed of three basic blocks, detection and perception, situation analysis and motion planning, and control. First, an environmental sensor system, that may be composed of different sensor types perceives the surrounding of the automated vehicle and fuses the information to build an environmental model. The subsequent planning module further processes the information of the environmental model to evaluate the surrounding traffic participants behavior and to plan the motion of the ego vehicle (EV). Finally, the motion that was planned for the EV is executed and controlled. For the motion planning of the EV, the evaluation of all the surrounding vehicles is important. To plan the motion of the EV on a trajectory and a strategical level, an estimation of the future state of the impending vehicles is of special interest. Since any vehicle is particularly influenced by vehicles on its ego lane and the neighboring lanes, the three closest impending vehicles on these lanes are considered for this study. For these three vehicles mentioned above, the current and future driving maneuver is determined. The maneuvers that are distinguished within this work are lane keeping (LK), lane change left (LCL) and lane change right (LCR). Since a lane change (LC) maneuver is defined as the entire process when the target vehicle (TV) moves all the way from its initial position until the actual crossing of the lane marking, the early detection of a LC maneuver means predicting the LC in advance. The advances in deep learning have led to the introduction of these tools to LC prediction e.g., Lee et al. [3] and Mänttäri et al. [4]. The trend of using models that become larger continuously resulting in an increasing number of parameters and regularly used datasets that provide complete knowledge of the environment has led to the present work. Here it is investigated how LCs can be predicted as early as possible from data taken with a test vehicle equipped with an appropriate 360° sensor setup. This data contains occlusion and measurement noise which leads naturally to an incomplete knowledge about the environment. The presented neural network-based approach extends the set of input features by two classes of information. Common approaches like Krüger et al. [5] use motion-based and situational features which are extended in this

166

Environmental model extension for lane change prediction with neural networks work by contextual and uncertainty-related features. Compared to the situational features the contextual ones describe the scenario on a more abstract level which reflects a wider time scale. The contextual features aim at strengthening or weakening the evidence for a specific maneuver by adding further information regarding the traffic scene. For example, by considering information about speed limits the longitudinal motion of a vehicle can be explained probabilistically excluding a LC. The uncertainty-related features are intended to create a correlation between the uncertainty of the measurements from the perception system and the predicted output. This information is usually related to the distance a specific measured object is away. In this way, the semantic description is improved. This extension is supposed to improve the classification performance by overcoming the conflict of early predictions with a low false positive (FP) rate. While the motion-based features recognize a LC only when it is already very close, situational features take into account the traffic situation and indicate a possible LC quite early. When LCs are prohibited by traffic rules or seem not meaningful due to other restrictions the proposed contextual features should detect that and prevent FPs. The remainder of this paper is structured as follows. In section 2 related works are discussed. Section 3 introduces the proposed new features. Results according to these features and further investigations are presented and discussed in section 4. Section 5 concludes the paper.

2 Related Work The work of Dagli et al. [6] was one of the first that was concerned with detecting LCs of surrounding traffic participants focusing on the improvement of Adaptive Cruise Control (ACC) systems. This approach uses Bayesian networks to predict cut-in maneuvers based on the motion of the TV. Bayesian networks allow the modeling of probabilistic relations in the form of graphical models and enable the possibility to introduce expert knowledge about the domain by the design of the graph. The concept was further improved to object-oriented Bayesian networks introduced in Kasper et al. [7]. There the introduction of a more hierarchical and modular structure of the networks enable an easier extension to further maneuver classes that should be recognized. In Dogan et al. [8] LCs of the EV based on driving simulator data are considered. The authors compare different machine learning techniques (Support Vector Machines – SVMs, Feed Forward Neural Networks – FFNNs and Recurrent Neural Networks – RNNs) and different sets of input features against each other. The SVM has been identified as the best classifier, detecting LCs up to one second earlier than FFNNs with fewer false predictions than RNNs. However, it has been found that the steering angle is an important input feature. These results are only partially transferable to our approach since the steering angle of other vehicles cannot be determined. Wissing et al. [9] explicitly distinguish

167

Environmental model extension for lane change prediction with neural networks between motion-based and situational features. For both kinds of features, different approaches have been used to determine whether a TV is performing a LC. Finally, the probability determined by the SVM for the motion-based features and the probability of a probabilistic model for the situational features have been combined. The results indicate that LCs can be predicted earlier with fewer false prediction applying this combination compared to the single subsystems. The work is based on simulated data conducted in a driving study with several participants. A neural network-based approach for predicting LCs in real-world driving data is presented in Krüger et al. [5]. There, using a fixed network setup and a fixed set of features, the influence of the imbalanced nature of the LC prediction application is studied. It has been shown that balancing techniques yield to significant performance improvements, of up to around 17% for the class-wise averaged accuracy, compared to a baseline. A simple and deterministic approach that could be used as a baseline to compare against is developed in Nilsson et al. [10]. The authors suggest a set of logical tests and inequations for checking longitudinal and lateral motion-related features. If the handcrafted thresholds for the features are exceeded or not reached, respectively, a LC is indicated. The system was developed on the Interstate 80 dataset Colyar and Halkias [11]. Lee et al. [3] and Mänttäri et al. [4] use Convolutional Neural Networks (CNNs) on an occupancy grid-like representation of the data. While CNNs have reached impressing improvements in detecting objects in natural images, the conversion of environmental data into a grid leads to images that have sparser information than natural images. The large number of parameters in relation to the sparsity of the data yield to the problem of overfitting, especially if only a few training samples are available. Lee et al. [3] argue that to overcome this issue a horizontal flipping of the images could be used for data augmentation. In Germany, however, it is obligated for slower cars to drive on the right lane. Since symmetry cannot be guaranteed, this method for data augmentation could not be applied in an arbitrary country. Mänttäri et al. [4] on the other hand use the Interstate 80 data. Due to a birds-eye perspective, the environment of a specific vehicle is exactly known with complete information about surrounding vehicles without any occlusion. Therefore, approaches developed on such datasets are not guaranteed to perform comparably well implemented in vehicles. Instead of using complex models with many parameters and datasets with complete knowledge about the environment the focus should be on elaborating simpler models using data recorded from the perspective of the vehicle. Simultaneously, it is tried to put as much knowledge as possible about the environment into the model, respecting the limitations of occlusion and sensor noise. Therefore, classical feature engineering is applied which leads to the two new input feature classes: contextual and uncertaintyrelated features.

168

Environmental model extension for lane change prediction with neural networks

3 Environmental Model and Feature Generation The combination of different features and the deduction of the best set of features has been considered in the literature before. Schlechtriemen et al. [12] have investigated how strong single features are correlated with LCs based on a Naïve Bayes classifier. The work mentioned and also Wissing et al. [9] found that based on the lateral motion of the TV relative to its lane LCs can be predicted accurately but only a short time before the actual LC occurs. Situational features, on the other hand, are able to indicate future LCs early at the cost of an increasing FP rate. The exact partition of feature classes and their meaning are investigated further in the following sections.

3.1 Basic Environmental Model and Related Features The standard environmental model described in Krüger et al. [5] will be used in this paper, see Figure 1. There, for a specific TV the six closest vehicles within well-defined areas are considered. The areas are either on the Left, Center or Right lane relative to the TV and Behind or Ahead.

Figure 1: Representation of the environmental model with the TV in the center surrounded by up to six neighboring vehicles. For the TV the LC prediction is evaluated. Therefore, the dashed line represents the center of the lane and the bent arrow line indicates the intent of the TV to change lanes. The highlighted vehicle in lighter (green) color represents the EV. Based on Krüger et al. [5].

The investigation is started with a summary of the standard features for LC prediction, known from Kasper et al. [7] and Krüger et al. [5]. Table I summarizes these features. The feature categorization in Wissing et al. [9] is extended by two new classes: contextual and uncertainty-related features: ● Motion-based features describe the motion of the TV absolute or relative to static environmental objects, such as lane marks ● Situational features describe the current dynamical traffic situation the TV is located in

169

Environmental model extension for lane change prediction with neural networks ● Contextual features describe the current traffic situation the TV is located in from a more static, abstract and semantic describable aspect ● Uncertainty-related features introduce measures for the uncertainty about the other input features caused by the distance-dependent measurement errors from the perception system In the following sections the newly introduced features are discussed in detail together with their intent. Table I: General input feature set used as baseline for the investigations in this paper. Input Feature Description 𝑑 𝑣 𝑣 𝑑 𝑑 𝑑 𝑑 𝑑 𝑑 𝛥𝑣

, , , , , , ,

𝛥𝑣

,

𝛥𝑣

,

𝛥𝑣

,

𝛥𝑣

,

𝛥𝑣

,

170

Lateral distance between the center of the TV and the center of its current lane [m] Lateral velocity of the TV [m/s] Longitudinal velocity of the TV [m/s] Distance between the TV and the closest vehicle in the LA area [m] Distance between the TV and the closest vehicle in the CA area [m] Distance between the TV and the closest vehicle in the RA area [m] Distance between the TV and the closest vehicle in the LB area [m] Distance between the TV and the closest vehicle in the CB area [m] Distance between the TV and the closest vehicle in the RB area [m] Relative velocity between the TV and the closest vehicle in the LA area [m/s] Relative velocity between the TV and the closest vehicle in the CA area [m/s] Relative velocity between the TV and the closest vehicle in the RA area [m/s] Relative velocity between the TV and the closest vehicle in the LB area [m/s] Relative velocity between the TV and the closest vehicle in the CB area [m/s] Relative velocity between the TV and the closest vehicle in the RB area [m/s]

Environmental model extension for lane change prediction with neural networks

3.2 Normalized Lateral Distance The normalized lateral distance 𝑑 , is calculated as the difference between the mean of the lateral position of the vehicle over a specific time window and its current lateral position. Figure 2 illustrates the determination of this feature. While the exact size of this time window could be a parameter that may be optimized in its own right, it has been chosen to be 2s which corresponds to 𝐾 50 discrete time steps, using a sample time of 0.04s: 𝑑

,

∑

𝑑

𝑑

(1)

(dashed line, behind the vehicle) represents the lateral offset with respect to Figure 2: 𝑑 , an averaged offset of 𝑑 over a certain time window from the past trajectory (dash-dotted line, behind the vehicle).

There are typical situations in real-world traffic when a large distance of a vehicle to the center of its lane does not directly indicate a LC. These large distances can often last for several seconds. Some drivers feel uncomfortable driving next to a truck. So, they decide to keep their vehicle further away from the lane marking of the side of the truck and closer to the other side. Highway exits are another example. At a low traffic density, drivers often choose to drive on the right-most lane even if the exit is still some distance ahead. But in preparation of the maneuver of leaving the highway, they already drive very close to the lane marking that separates the hard shoulder. In both mentioned cases a usual pattern recognizable is that the vehicle is maneuvered close to the lane marking with a significant offset from the center of the lane without the immediate intent to change the lane. Once the vehicle has

171

Environmental model extension for lane change prediction with neural networks reached the desired offset from the center of the lane, this distance is often kept constant. The introduction of this feature intends to get independent of this offset which is no indicator for a LC.

3.3 Speed Limit Information The introduction of the current speed limit 𝑐𝑠𝑙 aims at suppressing FPs for LCs in any direction by indicating the maximum allowed velocity on a given highway section. Additionally, the specification of the difference 𝛥𝑠𝑙 between the current speed limit and the previous one enables the network to correlate the acceleration behavior of the TV and the change of the speed limit value. Since the time between the current sample and the last change in the speed limit 𝑡 is used too, the network may additionally learn to correlate the acceleration behavior of the TV with the speed limit only in a certain time window shortly after a new speed limit sign was detected. between the Schlechtriemen et al. [12] figured out that the relative velocity 𝛥𝑣 , TV and its impending vehicle is the single most important feature to predict LCs correctly. Therefore, the longitudinal motion of the TV is of special interest. A relative velocity of about 0m/s can be forced by traffic rules through speed limits and may indicate LK in order to obey the traffic rules. A slow impending vehicle at a high traffic density or an equal desired velocity several vehicles share may lead to small relative velocities, too. Nevertheless, a LC might be possible in these situations without violating the traffic rules, if a speed limit is not set or much higher than the current velocity. These considerations are of probabilistic nature based on the observation that most drivers obey the traffic rules.

3.4 Distance to the Ego Vehicle The distance between the TV and the EV 𝑑 can be used in two different ways: , directly as an input feature or to weight the samples during training according to their relevance and accuracy. According to Figure 1, the EV is always one of the six surrounding vehicles of the TV, more precisely one of the three vehicles behind the TV. The feature is not explicitly considered within the feature vector yet but has a major influence on the feature quality and thus on the prediction accuracy. Hence, it could be beneficial to consider the distance explicitly. While adding the feature to the feature vector is straight forward another way to utilize the information is by weighting the training samples accordingly. Since it is of special interest that the LC prediction is correct for close TVs, the accuracy can be relaxed for TVs further away which is in loose consistency with the limitation of the relevant area up to 50m ahead of the EV in Lee et al. [3]. Since the bounds for the sample weights, 𝑠𝑤 are naturally restricted to the interval [0,1] and the TV is always ahead of the EV which results in

172

Environmental model extension for lane change prediction with neural networks a positive value for 𝑑 tial function: 𝑠𝑤

e

,

, scaling the samples could be done using the exponen-

,

(2)

Using these sample weights for training, a smaller emphasis is put on samples where the TV is far away, reducing their contribution to the total loss. By taking the norm , the sample weighting could also be done for TVs behind the EV. 𝑑 , The feature is of special interest for two reasons. First, if the TV is further away then it is less relevant due to the remaining time to react. Second, if the TV is further away then the measurement of this vehicle gets more inaccurate. This is due to the sensor characteristics on the one hand and the perception and tracking algorithms on the other hand. Typically used approaches for object tracking such as Kalman filters estimate the uncertainty of their tracks. This uncertainty usually increases with the distance to the measured object. Another kind of uncertainty is introduced due to the limited view range of the camera that determines the lane makers. A third order polynomial is fitted for each of the detected lane markings to provide a compact representation of the course of the lane marking. Beyond the view range of the polynomial, the fitting does not extrapolate accurately. Due to the two mentioned sources of uncertainty, the distance from the TV to the EV should be regarded as a potentially useful input feature.

3.5 Lane Marking Type The feature lane marker type 𝑙𝑚𝑡 is a discrete variable encoding whether the lane marking is solid, dashed or has a different style. The feature lane marker type is controversially discussed. While lane marking information is used for LC prediction in some works like Schlechtriemen et al. [12], it was not in the reference work Krüger et al. [5] or Wissing et al. [9]. The intention is, similar to the speed limit related features, to reduce the number of FPs by trying to implement a correlation between solid lane markings and LK. Since solid lane markings in Germany prohibit crossing them, the prediction of a LC would then result in forecasting an illegal driving maneuver which is expected to be quite rare.

3.6 Categorization of Features Table II summarizes all features considered within this paper and assigns them to one of the four feature classes introduced in section 3.1. Note that the differentiation between situational and contextual features could be also be seen as not directly related to the traffic rules and directly related to the traffic rules, respectively.

173

Environmental model extension for lane change prediction with neural networks Table II: Summary of features and allocation to feature classes. Motion-based 𝑑 𝑣

,𝑑 ,𝑣

𝑑 𝑑 𝛥𝑣 𝛥𝑣 𝛥𝑣

,

,

Contextual

Situational , , , , ,

,𝑑 ,𝑑 , 𝛥𝑣 , 𝛥𝑣 , 𝛥𝑣

,𝑑 ,𝑑

, , , , ,

, ,

,

, , ,𝑡

,

𝑐𝑠𝑙, 𝛥𝑠𝑙, 𝑙𝑚𝑡

Uncertaintyrelated 𝑑 ,

4 Experimental Results 4.1 Experimental Setup Since the focus of this paper is not on the methodology of the LC prediction method itself but instead of finding a good representation of the environment by means of input features 𝒙, a summary about neural networks and their training process should be given here. The Multilayer Perceptron (MLP), a special form of the FFNN, is characterized by stacking multiple layers of neurons. An input layer, having the same number of neurons as there are input features and an output layer having the same number as there are elements in the output 𝒚 are required. For LC prediction the dimension of the output vector 𝒚, 𝑁 is three due to the set of distinct classes LCL; LCR; LK . So-called hidden layers between the input and the output layer are optional. The number of contained neurons is arbitrary. In a MLP all neurons of one layer are connected to all neurons of the next layer. The network is called fully connected. The connections are parametrized by weights 𝒘. All the weights and all the bias values 𝒃 are summarized to the parameter vector 𝜣 of the network. During the training procedure 𝜣 is adapted in a way that a prediction error, also called loss, is minimized. 𝜣 is the vector of optimal parameters: 𝜣

arg min 𝐿

𝒚, 𝒚

𝜣

(3)

with 𝒚

𝑓 𝒙, 𝜣

so that

∑

𝑦

1

(4)

and 𝐿

174

∑

𝑦

log 𝑦

(5)

Environmental model extension for lane change prediction with neural networks 𝐿 is the categorical cross entropy loss function that expresses how different the netis from the desired network output 𝒚, also called label. The network work output 𝒚 parameters 𝜣 are than optimized during the training procedure. Since the motion-based and the situational features are highly dynamical in a way that they change very fast, not only their current value is considered but also an entire history vector of each feature for the past second sampled in 0.1s time intervals.

4.2 Experimental Results and Evaluation While the use of the past values comprised to the feature history is obvious for the standard features, some of the newly introduced features only change on a wider time scale and so whether to use a history for them or not is questionable. For this reason, all of these features are evaluated twice, once with their current value only and once with a history. To assure good generalization performance the speed limit related features were used and evaluated altogether. Data collection with a test vehicle on a highway results naturally in an imbalance between the three distinguished classes LCL, LCR and LK. In Krüger et al. [5] different methods to deal with this imbalance were discussed. In this paper, the imbalance is to give 𝒚 during Receiver considered by adapting the threshold for binarizing 𝒚 Operating Characteristics (ROC) analysis, according to Schisterman et al. [13]. The difference between the standard used (arg max) function for discretization and the threshold selected to minimize the distance to the optimal point in ROC space [0,1], is illustrated in Figure 3 for the baseline configuration used in this paper, see section 3.1. The more distinct main diagonal of the confusion matrix based on the ROC threshold indicates the better classification performance of this configuration with respect to the balanced accuracy. Table III and IV visualize the main characteristics of all the trained classifiers based on the formerly specified feature sets. To generate these results the ROC threshold method was used.

Figure 3: Comparison of the relative confusion matrices for the baseline classifier but with different discretization methods. (Left) arg max, (Right) minimal distance to [0,1] in ROC space.

175

Environmental model extension for lane change prediction with neural networks The main finding from the numbers in Table III and IV is that the number of FPs in an event-based evaluation is the most sensitive indicator to display the different classification performances of all the trained classifiers. Based on these results, the networks trained with the additional information about the speed limits with its history values, the lane marker type, and the usage of all new features can be considered significantly better than the baseline. Interestingly, the improvements for LCL are about 50% while for LCR only about 20% fewer FPs are predicted. Compared to the baseline almost all classifiers reach a lower number of FPs. This confirms the intent and hypothesis from the introduction section that the presented features mainly improve the classification performance by reducing the number of FPs. Note, that indeed the numbers for FPs are high but since most of these FPs only last for a few consecutive samples (up to three), they can easily be filtered out without losing much recognition time 𝛥𝑡 / . FPs are important since they are expected to contribute significantly to the acceptance of such systems by human drivers. To prevent too much intervening control actions caused by falsely predicted LCs, this measure is of special interest. Although there might be traffic situations in which a TV does not perform a LC, a warning caused by a predicted LC would be desired. This kind of situations are not considered in the evaluation yet. The recognition time 𝛥𝑡 / determines how much time a LC is predicted before the actual crossing of the lane marking. Compared to the baseline, some of the classifiers detect LCs (in either direction) earlier while some recognize them later. However, the recognition times do not vary much and lie in a range of about 1.75 to 1.95s for LCL and 2.10 to 2.50s for LCR. Since the motion-based features are the main indicators for LCs and they were kept throughout all experiments, the number of correctly detected LCs does not differ much, between all the experiments. Depending on the actual classifier between approximately 95 to 100% of the LCs in both directions are correctly predicted. The number of FP events is considered the main performance indicator since the other measures, the number of correctly predicted LCs and the recognition time do not allow for a good differentiation regarding the classification performance due to similar values. Table III: Results of the event-based evaluation. Bold values indicate the best values while the italic and underlined mark the worst ones. Description

#LCL (of 38)

#LCR (of 43)

#FP LCL

#FP LCR

𝛥𝑡 [s]

𝛥𝑡 [s]

Baseline feature set + 𝑐𝑠𝑙, 𝛥𝑠𝑙, 𝑡 + 𝑐𝑠𝑙, 𝛥𝑠𝑙, 𝑡 history + 𝑙𝑚𝑡 + 𝑙𝑚𝑡 history

37 38 37

42 43 41

1501 1225 853

2188 1949 1767

1.86 1.92 1.78

2.25 2.43 2.22

36 36

42 43

752 857

1827 1962

1.55 1.81

2.32 2.52

176

Environmental model extension for lane change prediction with neural networks +𝑑 , +𝑑 , history +𝑑 , +𝑑 history , + all new features + all new features history 𝑠𝑤 (𝑑 , -based)

36 37 38 36 36 37

42 43 43 41 41 42

1056 1097 1322 899 794 1509

1997 2111 2076 2137 1924 2247

1.94 1.86 1.74 1.82 1.82 2.09

2.21 2.12 2.24 2.26 2.20 2.12

36

41

1664

2426

1.78

2.40

The frame-based evaluation results show comparable behavior to the number of correctly predicted LCs and the recognition time in the way that they do not differ very much. Since mainly only several hundred FPs, each merely lasting a few samples (out of several thousand samples entirely), could be reduced, this translates to small changes in the frame-based metrics. These metrics are classic measures of classification performance that ignore the temporal coupling of subsequent samples. While the accuracy measures the classification performance overall and does not account for data imbalance, the balanced accuracy does. Since the imbalance is considered from the beginning on, the gap between accuracy and balanced accuracy could be reduced significantly compared to Krüger et al. [5]. The numbers for the frame-based results indicate that indeed the accuracy and the F1 score correlate stronger with the event-based results than the balanced accuracy. Although the deviations are small, a relation is indicated. Table IV: Results of the frame-based evaluation. Bold values indicate the best values while the italic and underlined mark the worst ones. Description

Accuracy [%]

Balanced Accuracy [%]

F1 Score [1]

Baseline feature set + 𝑐𝑠𝑙, 𝛥𝑠𝑙, 𝑡 + 𝑐𝑠𝑙, 𝛥𝑠𝑙, 𝑡 history + 𝑙𝑚𝑡 + 𝑙𝑚𝑡 history +𝑑 , +𝑑 , history +𝑑 , +𝑑 history , + all new features + all new features history 𝑠𝑤 𝑑 , -based)

86.11 86.60 87.83 87.19 86.98 85.71 86.85 86.18 86.72 87.40 85.43

82.06 83.68 83.13 82.25 82.81 83.54 83.48 82.34 83.57 81.17 81.99

0.39 0.40 0.42 0.42 0.41 0.40 0.40 0.39 0.41 0.42 0.39

83.39

82.67

0.38

177

Environmental model extension for lane change prediction with neural networks The weighting of the samples during training according to the feature 𝑑 , , does not result in any improvement at all. Rather, the number indicates a worse performance for most of the metrics compared to the baseline. While the idea to provide some information about the uncertainty of the samples seems still meaningful, the results indicate a performance degradation instead of an improvement. This may mean that the feature alone does not capture the necessary information about the measurement 𝑑 , uncertainty and further indicators are needed. While the speed limit related features and the lane marker type lead to a performance improvement, the significance of the gain for the normalized lateral distance and the distance between EV and TV is smaller. Since also the assumption that the history for the speed limit related features should not be of great importance, was not proven by the results, this indicates that a detailed investigation of the optimal design of the feature vector including its history component is needed. By doing so, the full potential of the extended environmental model may be utilized, while this study presented a first indication of how improvements can be reached. The identification of features that benefit from a history and the ideal history length is of great importance because the MLP architecture suffers heavily from non-informative features due to the fully connected structure. If a feature does not provide new information, many parameters are newly added to the network without any benefit during the optimization which quickly leads to overfitting and a performance drop. This relation may explain why some of the improvements are not as significant as expected, based on theoretical considerations.

5 Conclusion In this paper several aspects for extending the environmental model and thereby the input feature vector for a neural network-based LC prediction were proposed. It was shown that this extension could improve the performance. Information about the speed limit and the lane marking type have been proven to be of great significance by reducing the number of FPs for LC events in both directions remarkably. The assumption that they can explain away specific hypothesis in certain situations was proven by experimental results. However, other proposed features showed a less clear indication for performance improvements. The categorization of the input features in different classes has been extended by the proposed features, too. The newly proposed class contextual has shown a greater impact than the uncertainty-related class. This highlights the importance of the correct framing of the problem. It could be shown that feature engineering makes sense in this context and that simple network architectures do not have to be inferior compared to complex ones, dependent on the problem. The exact composition of an ideal input feature vector remains open for further work. Additionally, it needs to be checked if network architectures that especially benefit from

178

Environmental model extension for lane change prediction with neural networks temporal structured data such as 1D CNNs and Long Short-Term Memory (LSTM) lead to further improvements given the extended environmental model. Investigations if the environmental model should be extended even further seem meaningful, too. The information about the type of the neighboring lanes instead of only lane marker type and the type of the surrounding vehicles might be beneficial, too.

Bibliography 1. SAE International (2018); Taxonomy and Definitions for Terms Related to Driving Automation Systems for On-Road Motor Vehicles; Standard J3016_201806 2. Tas, Ö.S.; Kunht, F.; Zöllner, J.M.; Stiller, C. (2016); Functional System Architectures Towards Fully Automated Driving. In: 2016 Intel. Veh. Symp. (IV), 304-–309 3. Lee, D.; Kwon, Y.P.; McMains, S.; Hedrick, J.K. (2017); Convolution Neural Network-Based Lane Change Intention Prediction of Surrounding Vehicles for ACC. In: 2017 Int. Conf. on Intel. Transp. Sys. (ITSC), 1–6 4. Mänttäri, J.; Folkesson, J.; Ward, E. (2018); Learning to Predict Lane Changes in Highway Scenarios Using Dynamic Filters on a Generic Traffic Representation. In: 2018 Intel. Veh. Symp. (IV), 1385–1392 5. Krüger, M.; Stockem Novo, A.; Nattermann, T.; Glander, K.H.; Bertram, T. (2018); Lane Change Prediction Using Neural Networks Considering Classwise Non-Uniformly Distributed Data. In: 2018 Proceedings Automotive meets Electronics, 1–6 6. Dagli, I.; Breuel, G.; Schittenhelm, H.; Schanz, A. (2004); Cutting-in Vehicle Recognition for ACC Systems – Towards Feasible Situation Analysis Methodologies. In: 2004 Intel. Veh. Symp. (IV), 925–930 7. Kasper, D.; Weidl, G.; Dang, T.; Breuel, G.; Tamke, A.; Rosenstiel, W. (2011); Object-Oriented Bayesian Networks for Detection of Lane Change Maneuvers. In: 2011 Intel. Veh. Sysmp. (IV), 673–678 8. Dogan, Ü.; Edelbrunner, J.; Iossifidis, I. (2011); Autonomous Driving: A Comparison of Machine Learning Techniques by Means of the Prediction of Lane Change Behavior. In: 2011 Int. Conf. on Robotics and Biomimetrics (ROBIO), 1837–1843 9. Wissing, C.; Nattermann, T.; Glander, K.H.; Seewald, A.; Bertram, T. (2017); Lane Change Prediction by Combining Movement and Situation based Probabilities. In: 2017 IFAC-PapersOnLine 50(1), 3554–3559 10. Nilsson, J.; Fredriksson, J.; Coelingh, E. (2015); Rule-Based Highway Maneuver Intention Recognition. In 2015 Int. Conf. on Intel. Transp. Sys. (ITSC), 950–955

179

Environmental model extension for lane change prediction with neural networks 11. Colyar, J.; Halkias, J. (2006); Interstate 80 Freeway Dataset. In: 2006 Fed. Highway Admin. (FHWA) Tech. Report FHWA HRT-06-137 12. Schlechtriemen, J.; Wedel, A.; Hillenbrand, J.; Breuel, G.; Kuhnert, K.D. (2014); A Lane Change Detection Approach using Feature Ranking with Maximized Predictive Power. In: 2014 Intel. Veh. Symp. (IV), 108–114 13. Schisterman, E.; Perkins, N.; Liu, A.; Bondell, H. (2005); Optimal Cut-Point and its Corresponding Youden Index to Discriminate Individuals using Pooled Blood Samples. In: 2005 Epidemiology 16(1), 73–81

180

Requirements & evaluation of friction information for the integration in vehicle systems M. Eng. Staiger, Sebastian, M. Eng. Nosrat Nezami, Siamak Dr. Ing. h.c. F. Porsche AG Prof. Dr.-Ing. Dieter Schramm Lehrstuhl für Mechatronik, Universität Duisburg-Essen

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 M. Bargende et al. (Hrsg.), 19. Internationales Stuttgarter Symposium, Proceedings, https://doi.org/10.1007/978-3-658-25939-6_15

181

Requirements & evaluation of friction information for the integration in vehicle systems

1 Introduction and Description of the problem Increasing levels of automation in driver assistance systems (DAS) require safeguarding the steer and brake intervention in order to further prevent accidents. Special attention must be given to the interaction between tires and road surface. The friction coefficient between road (µ) and tires has a significant impact on driving performance and safety. Figure 1 illustrates that a very high percentage of the accidents happen in wet, snowy or icy road conditions. 100% 80% 60% 40% 20% 0% 2011

2012

2013

2014 other accidents

rain

2015

2016

2017

snow and ice

Figure 1: Accidents due to road conditions

Therefore, a substantial challenge for the development of new DAS is the environmental and road condition detection, which was so far detected, interpreted and transferred to a driving command by the driver. For future DAS like the highway pilot, the car must detect the road condition with respect to the road surface, as well as the occurrence of an intermediate medium, in order to be able to react to possible scenarios such as rain, snow or icy road conditions. The integration of the corresponding friction coefficient into driving assistance opens the possibility to prevent critical driving situations as a precautionary measure. Due to various approaches, such as the estimation method with sensor fusion the current friction coefficient of the road can be determined during the drive. Before a partially autonomous or autonomous intervention of the vehicle is possible, the methods and the resulting friction coefficient have to be evaluated regarding sufficient reliability and quality. Accordingly, the proposed paper presents a method to develop the requirements identification, evaluation and optimization of a present friction coefficient for automated driving functions.

1.1 State of the art This article relates to the signal handling and evaluation as well as the potential of the friction coefficient and its processing. Multiple articles were already released with respect to this topic. In [Lex, Shao, Hackl, & Eichelberger, 2016] the available friction coefficient is described as essential for automated driving performance. The approach

182

Requirements & evaluation of friction information for the integration in vehicle systems to determine the coefficient is based on standard built-in automotive sensor systems. For this purpose, the vehicle longitudinal motion is estimated using recursive Bayesian estimation and across the vehicle road the friction coefficient is estimated with the least square method. In [Singh, Arat, & Taheri, 2012] an intelligent tire sensor, which detects the current slip by means of the vibrations of the tread, is used. Taking this information, the efficiency of the ABS-controller can be increased. In the article of [Gustafsson, 1997], a Kalman filter is used to estimate the parameters of the slip curve thus determining a current friction coefficient. This approach is picked up by [Müller, Uchanski, & Hedrick, 2003], making it more robust by combining road information from ABS controlled braking processes with the help of the traction circle. A procedure that estimates the linear range of the slip curve using artificial neuronal networks is introduced in [Matusko, Petrovic, & Peric, 2007] . In addition, a condition observer is implemented, which validates the plausibility. Concluding, the paper from [Halgamuge & Herpel, 1993] can be mentioned, which combines multiple of the mentioned methods. A sensor detects the deformation of the tires. Then, together with a fuzzy logic, the road condition is estimated. Additionally, an artificial neuronal network is used to create a control algorithm in order to be independent from expertise knowledge and estimations. In summary, the articles mentioned above describe several methods to identify the grip of the road, but do not address the requirements of the signal. This means that the requirements have to be addressed and a friction coefficient must be assessed.

2 Identification of the requirements In order to maintain the link between customer perspective, specific business targets and technical requirements, the Use-Case is taken as a tool of requirements engineering. The description is made by the Epic, which describes the customer use case. The technical requirements are described in the Use-Case consideration systematically below. Finally, the requirements are broken down to development perspectives and architectures. For clarification, the approach is illustrated by the integration of the friction coefficient. According to assumptions of the Epic are considered in relation to common language, reduction of the technical details as well as focusing on presentation the additional value. In terms of content the Epic presents a uniform procedure. The target is the optimization of the vehicle and the added value for the customer, for example integration of bad weather conditions in the control of the vehicle with the help of the friction coefficient. The system visualization is shown schematically according to (Wirdemann, 2011) in figure 2.

183

Requirements & evaluation of friction information for the integration in vehicle systems

Figure 2: Procedure of the Epic, based on Wirdemann 2011

As an example, the increase of the distance between two vehicles under wet road conditions can be mentioned. One method to describe the braking process is the visualizadescribes the distance a vehicle needs to tion of the braking distance. The value 𝑠 or the vehicle standdecelerate from an initial speed 𝑠 to a defined target speed 𝑠 0. The negative acceleration 𝑎 , represents the strength of the interstill 𝑠 vention. The braking process is controlled via the anti-lock braking system ABS, which . The braking controls the present braking slip to the range of maximum friction 𝜇 distance results in formula (1) [Wollf, 2017]. 𝑠

1 2𝑎

∗𝜇

∗ 𝑣

𝑣

²

(1)

The reaction time of the driver is neglected, as an automated braking should react under optimal condition simultaneous to the occurrence of the incident. Figure 3 emphasizes the necessary distance to a object depending on speed and friction coefficient.

184

Requirements & evaluation of friction information for the integration in vehicle systems  300,00

25 Km/h 50 Km/h

 250,00

75 Km/h 100 Km/h 125 Km/h

Distance to stationary vehicle

150 Km/h  200,00

175 Km/h 200 Km/h

 150,00

 100,00

 50,00

 ‐ snow/ice

wet

0,1

0,2

0,3

0,4

0,5

dry 0,6

0,7

0,8

0,9

1,0

1,1

1,2

Friction

Figure 3: Impact of road condition on the braking distance

Assistance systems in vehicles can benefit from the knowledge of the friction coefficient. Especially the safety of the occupants can be increased by using the knowledge of the friction coefficient, e.g. for early emergency braking. Thus, the number of accidents on wet, snowy or icy roads can be reduced. Following the requirements on a friction coefficient are shown using an example of the highway pilot. The technical boundary conditions are the maximum speed of 130 km/h and the detection distance of 150m. The prerequisite here is that the vehicle has sufficient braking capability (i.e., emergency braking approximately equaling gravitational acceleration) to be able to reach standstill within the detection range. Applied to formula (2), it results in: 𝜇

𝑣² 2 ∗ 0.9𝑔 ∗ 𝑠

130 𝑚 3.6 𝑠 ² 𝑚 2 ∗ 0.9 ∗ 9.81 ∗ 150𝑚 𝑠

0.49

(2)

In order to take into consideration wear of relevant components, like tires or brakes and further tolerances, a reserve factor of 25% is added to the value: 𝜇

1.25 ∗ 𝜇

𝟎. 𝟔

(3)

The result 𝜇 0.6 is therefore the current threshold at which the customer can use and activate a road condition adaptive driver assistance. Additionally, for the actual implementation, e.g. increase in the vehicle distance, a time delay has to be considered. The input signal must therefore be of high quality and represent the actual condition to

185

Requirements & evaluation of friction information for the integration in vehicle systems achieve a reliable behavior of the driving function. Thus, the fundamental Quality criteria are linearity and plausibility.

2.1 Signal processing A signal is a physical value, in most cases an electrical voltage, which are recorded over a certain period, the sampling period. The main purpose of signal processing, except from gaining information of the measured value, is to verify the data quality in terms of plausibility and the preparation of the signals. Plausibility reflects that the values have to be within a certain range. All other information is not relevant. Another distinctive feature is the deterministic or stochastic signal. The stochastic signal is of importance for this paper. Due to the multiple influencing factors, like temperature, weather or road condition, the signal has a random characteristic. Therefore, the signals are not predicative [Böhme, 1998]. Filtering technologies: Modern signal processing is digitally configured, nevertheless it is essential to consider analog filters. Analog low-pass filters are used to convers analog to digital signals and serve as the basis for modern filter systems. In Figure 4 a filter is added to the system. As a result, the undesired parts 𝑥 𝑡 are eliminated. The useable part 𝑦 𝑡 remains. Input signal

Feed distrubance

𝑥 𝑡

𝑥 𝑡

𝑥 𝑡

Use Signal

Filter

𝑦 𝑡

𝑦 𝑡

Filtered 0

𝑦 𝑡

Figure 4: System with filter (based on (Meyer, 2017))

In [Meyer, 2017] two different categories of the digital filtering are described. Based on the feedback, recursive and non-recursive filters are distinguished. Recursion describes the loopback of the output value to the input value. This has effects on the impulse response of the systems, which describes the reaction of the input value to an impulse. Systems with feedback have an infinite impulse response and are therefore named as IIR (Infinite Impulse Response) - filters. Non-recursive systems have a finite impulse response and are called FIR (Finite Impulse Response) - filters. The recursive IIR-filter is based on pure analog filters. With different transformation methods analog models are transferred into digital form. As already mentioned, this means that the initial point is always an analog low-pass, which is converted by different frequency transformations to one of the named filters. Due to the missing feedback the synthesis of a

186

Requirements & evaluation of friction information for the integration in vehicle systems filter system deviates with FIR-systems. Instead of transferring analog filters by transformation, the target is to try to approach the behavior of these filters with the help of approximation. The advantage of this approach is that consistently stable and faulttolerant behavior can be set. This is the prerequisite for the integration in adaptive systems and puts the increased calculation effort in to perspective compared with FIRfilters. Due to these positive characteristics it leads to a preferred usage and higher spread of this category in all areas of signal processing [Meyer, 2017].

2.2 Increase of the signal quality An increase of the quality is achieved, when signal output was checked for plausibility with an appropriate filter concept so that pure useable data remains. In order to reduce the noise component and peaks, a moving average filter, which is often named as first order FIR low-pass filter, is helpful [Runkler, 2015]. For this application the spectrum has to be looked at. Choosing a too high value, it can result in negligence of valid data points and therefore in a loss of information. Figure 5 shows the implementation of a wet road sensor and the result after filtering.

Figure 5: Implementation of the moving average to a signal of the sensors for wet conditions

187

Requirements & evaluation of friction information for the integration in vehicle systems An improvement due to reducing the noise component and the straightening of the signal is notable. To the fact that the signal processing has a defined sampling rate, this straightening is reasonable. In the following the friction coefficient is described with the help of the traction circle. As the determined friction coefficient does only represent the currently used value, a filter is needed which measures the maximum of the values and outlines constantly the maximum values over a defined period for further processing. Curve sketching is used to calculate the maximum values. Considering the zero points of the gradient 𝑚 from the curve tangents, resulting from the linear function of formula (4), the max. value can be concluded by zeroing the derivation of the curve equation (formula (5)). 𝑦 𝜇

𝑚𝑥

𝑡

𝑓 𝜇

(4) 0

(5)

With the definition of an appropriate sampling rate it results in a filtering of the used friction coefficient 𝜇 from the traction circle, see figure 6. A sampling rate of around 20 seconds was chosen for the visualization.

Figure 6: Maximum value filter

188

Requirements & evaluation of friction information for the integration in vehicle systems After linearization the signal is present in a form that describes the maximum of the used friction coefficient and is therefore useable for further identification methods. The state of the art showed that the approach of the fuzzy logic is suitable for determination of complex quantities with multiple influencing parameters and dependencies. In using the fuzzy logic, according to the theory of fuzzy quantities, a specific numerical value is received after some steps [Zimmermann, Angstenberger, Lieven, & Weber, 1993]. The abstracted theory according to [Biewer, 1997], [Müller P. D., WS18], [Borgelt, Klawonn, Kruse, & Nauck, 2003] und [Zacher & Reuter, 2017] is derived as follows. A fuzzy quantity is defined to a quantity 𝑒 of a set 𝐴 in the interval between zero and one by the membership function. Formula (6) underlines the binary characteristic of a fuzzy quantity, which only enables the explicit allocation of a value: 0 𝑓𝑎𝑙𝑙𝑠 𝑒 ∉ 𝐴 1 𝑓𝑎𝑙𝑙𝑠 𝑒 ∈ 𝐴

𝜇 𝑒

(6)

As fuzzy quantities are used, fluent transitions between values zero and one are possible. These are characterized with the fuzzy membership function: 0

𝜇 𝑋

1

(7)

As an example, the temperatures in linguistic terms are named as cold, mild or hot instead of specific values in °C. Reversely it is possible to define quantities, which can be described by a word and processed by computers. This operation is called fuzzification. 𝑐𝑜𝑙𝑑

20, 19, … 5 °𝐶 𝑚𝑖𝑙𝑑

5,6, … 20 °𝐶 ℎ𝑜𝑡

20,21, … 45 °𝐶

As the areas are affected by uncertainties and inaccuracies, the specification is made by fuzzy quantities. With the description of the affiliation in trapezium or triangle functions, tolerance range 𝑡𝑜𝑙 𝑥 and influence width 𝑚 𝑥 can be added. By cutting the individual quantities the failure tolerance of the fuzzification is enlarged. This combination of several quantities is called linguistic variable. 𝑚 𝑥 𝑡𝑜𝑙 𝑥

𝑥 𝜇 𝑥 𝑚 ,𝑚

0 𝑥|𝜇 𝑥

(8) 1

(9)

If a system have multiple input values, the fuzzificated quantities have to be combined, with the help of logical operators. An intersection is processed with the logical AND as well as the minimum operator and generates the smallest value of the two membership functions. Analog the logical OR and the maximum operator are building the union of both data and create the maximum values. Rules are used to set the input and output values in relation. All established rules are in total building the rule base. This part defines the logic of the controller. For the implementation of the interference the condition with the logical IF, as well as the “conclusion” with the logical THEN are

189

Requirements & evaluation of friction information for the integration in vehicle systems described. The conditions consist of the respective input values, which are connected with the logical AND/OR. Hence depending on the fulfillment degree, if a respective statement is true (>0) or false (=0), a conclusion can be made. If a statement is true, the rules of the respective membership functions of the input value are utilized. The MIN/MAX operators are results of the transfer from the input value to the linguistic variables. Figure 7 shows an example with two rules and two input values. In this figure the specific values of wet condition and temperature are fuzzificated to the affiliation of the input values and transferred in the linguistic variables of the output value by the “conclusion”. 1.) if temperature = cold 2.) if temperature = cold

AND AND

wetness = high wetness = low

result friction = low result friction = medium

Assumtion cold

low

high

1

𝜇

𝑮

𝜇

𝑀𝐼𝑁 𝜇 , 𝜇

0,8

0,8

𝑑𝐵

𝜇

0,9

𝜇

1

low

low

high

𝑀𝐼𝑁 𝜇 , 𝜇

0,4

𝜇 𝜇

𝑇

medium

1 𝑮𝟐

𝜇

𝜇

𝑤𝑒𝑡𝑛𝑒𝑠𝑠

𝜇 cold

high

𝜇

°𝐶

𝑇

𝜇

medium

1

1

0,9

𝜇

1

Conclusion

𝜇

𝜇

𝜇

°𝐶

0,4

𝑑𝐵

𝜇

Figure 7: Example of the membership function dependent on the input parameter

The result of the degrees of fulfillment is again a fuzzy quantity. Different methods allow the conclusion to a specific numerical value. This paper introduces the finale step of defuzzification of the balance point COG of the conclusion. With this point, it is from the conclusion. Merging the possible to derive a specific actuating variable 𝑦 individual quantities 𝜇 resulting from the degree of fulfillment 𝐺 , a total area of the conclusions is created, whose balance point is generated according to formula (2-9) and figure 8:

190

Requirements & evaluation of friction information for the integration in vehicle systems 𝐶𝑜𝐺

∑

𝑦

∑

𝐺 ∗𝑦 𝐺

(10)

The result is a specific numerical value, which is suitable for an output value of a controller. The advantage of this method is the possibility to model and simule complex systems from simple basis approaches. The result of the fuzzy logic is robust as well as reproducible. This means, it enables stable behavior despite of stochastic input signals and is suitable for the evaluation of the present signals [Zacher & Reuter, 2017]. 𝜇

𝜇 low 𝜇

medium

high

1

0,8

𝑦

𝜇

low

medium

high

1

0,2

low medium

hoch

1 𝜇

0,4

𝑦

𝑦

0,4

0,27

𝜇

Figure 8: Defuzzification

3 Implementation Currently the limit value estimator is the up to date technology in use. The term arises from the input variables used, which are based on the engagement of the electronic stability control (ESP) and anti-lock braking system (ABS). The system measures the vehicle acceleration in both the transverse and longitudinal directions. Transferring the values into the friction circle with the results in an active global maximum coefficient . In order to obtain a usable friction coefficient signal, they use of friction 𝜇 as an impulse for a recalibration of a high assumed signal the parameter 𝜇 𝜇_𝐿𝑒𝑎𝑟𝑛𝑒𝑑. This represents the usable coefficient of friction. After an event in the border area (i.e. such as high accelerations across or along the vehicle), a significant deviation of the exploited value from the learned signal can occur. In this case, the signal is lowered to the approximate value of the maximum utilized friction coefficient (see figure 9).

191

Requirements & evaluation of friction information for the integration in vehicle systems

Figure 9: State of the art friction estimator

To be able to represent the compiled friction coefficient in a plausible and accurate way, the fuzzy logic will be used from now onwards. Figure 10 illustrates the structure of this approach. Combining several input variables and criteria of the logic make it possible to only process those values of the input variables which can be correlated and plausibilized in comparison with other sensors. First, a linguistic variable is put in place to describe the output value. This makes it possible to derive the exact numerical variable of the friction coefficient. Input parameter

Filter

Fuzzification

Inference

Defuzzification

Result

Rules

Maximum value filter

1

µused

Moving average filter

1

Sensor Waterfilm

IF… AND … AND 1

Sensor Temperature

… THEN …

Speed windscreen wiper

Moving average filter

1

Figure 10: Model design new Fuzzy Logic

192

Correcting Variable

Requirements & evaluation of friction information for the integration in vehicle systems 𝜇

𝑣𝑒𝑟𝑦 𝑝𝑜𝑜𝑟, 𝑝𝑜𝑜𝑟, 𝑚𝑒𝑑𝑖𝑢𝑚, 𝑔𝑜𝑜𝑑, 𝑣𝑒𝑟𝑦 𝑔𝑜𝑜𝑑

No tolerance range is added to the amount "medium" in order to be able to detect the defined threshold value of 0,6. Due to the large range of influence of the surrounding quantities, the results will mostly either be good or bad. All of this prevents excursive behaviour for value 0.6. Figure 11 describes the conversion of the previously created criteria into linguistic variables. Fuzzification

Input parameter Friction [µused] Temperature [°C] Sensore Waterfilm [dB] Speed windscreen wiper [1/min]

very low 0 - 0,25 very cold -20 - 0 dry 0 - 25 slow

low medium 0,2 - 0,6 0,4 - 0,8 cold warm -5 - 10 5 - 25 damp wet 20 - 65 60 - 110 medium

0 - 25

15 - 45

high 0,6 - 1 hot 15 - 40 very wet 100 - 150

very high 0,9 - 1,2

Aquaplaning 140 - 180 fast

35 - 65

Figure 11: Fuzzification and input parameter

It is possible to achieve a suitable fuzzification of the input quantities and classification in quantities based on the perceptions. Quality criteria are also the base for the inference. The correlation of the coefficient of friction is now converted into conditions. The conditions get connected to further input variables of the rule base. Through this approach, the reliability of the logic increases along with redundancy and cooperation. The resulting coefficient of friction signal can be compared with the input quantities to achieve a first evaluation. Scenarios such as increasing wetness of the road or a significantly lower temperature should affect the available potential road liability. The illustrated model focuses only on distinct dependencies of the quality criteria. Input variables that have a significant influence on the friction coefficient are weighted more through reducing the connection with other variables. The best result was with 40 inferences, regarding to accuracy and inertia. The simulation shows the result shown in Figure 12. The preconditions were initial temperature of 14 °C (decreasing tendency) and multiple changes in the wetness of the road. It also represents the impact of the partially wet road surface as well as the associated fluctuating values of road grip. The simulation reveals the impact of the escalated wetness of the road at 1,400 seconds: the simultaneous dropping of the available coefficient of friction.

193

Requirements & evaluation of friction information for the integration in vehicle systems To confirm the findings, another simulation with different preconditions was operated. It tested the performance in freezing and dry conditions. Due to the temperature, no high friction coefficients are expected. The temperature fluctuated around 2 °C. the input value of the wet sensor indicated a dry road surface. Accordingly, a constant value of the coefficient of friction was expected. Another simulation confirmed these expectations. Both simulations show an increase in the quality of friction coefficient information.

Figure 12: Simulation Model with 40 inferences as rule base

The fused signal is continuous and noise free. It does not show any devious outlier in value. The friction coefficient does not change abruptly but by almost linear slopes in the signal. This enables a reliable detection of the threshold value of 0,6.

4 Conclusion This paper handles the analysis, evaluation and optimization of a friction coefficient signal. It starts with the presentation of the added value of the integration of a friction coefficient through the customer perspective in the form of various use-cases. The resulting requirements for the vehicle and the quality of the signal follow. The quality and shape of the friction coefficient signal is shown on the basis of the highway pilot. The paper shows that the use of a fuzzy controller is purposeful for the implementation. This results in good outcomes, which were checked by two steps of the verification and thus the functionality of the fuzzy controller was confirmed. The optimization of the present friction signal has been achieved and can be followed up. For continuation it is recommended to validate the shown method in road trials. The target should be to get a certain amount of statistical data in order to assess the availability of the controller. In addition the controller should be complemented by external data like weather information from meteorological service.

194

Requirements & evaluation of friction information for the integration in vehicle systems

Bibliography 1. Biewer, B. (1997). Fuzzy-Methoden – Praxisrelevante Rechenmodelle und FuzzyProgrammiersprache. Berlin: Springer Vieweg. 2. Böhme, J. F. (1998). Stochastische Signale – Eine Einführung in Modelle, Systemtheorie und Statistik mit Übungen und einem Matlab Praktikum. Stuttgart: Teubner. 3. Borgelt, C., Klawonn, F., Kruse, R., & Nauck, D. (2003). Neuro-Fuzzy-Systeme Von den Grundlagen künstlicher Neuronaler Netze zur Kopplung mit Fuzzy-Systemen. Wiesbaden: Friedr. Vieweg & Sohn Verlagsgesellschaft/GWV Fachverlage GmbH. 4. Gustafsson, F. (1997). Slip-based Tire-Road Friction Estimation. England: Elsevier. 5. Halgamuge, S., & Herpel, H.-J. G. (1993). Echtzeit Fahrbahnzustandserkennung mit Fuzzy-Neuronalen Netzen. In B. Reusch, Fuzzy Logic. Dortmund : Springer Verlag. 6. Lex, C., Shao, L., Hackl, A., & Eichelberger, A. (2016). Einfluss von Reifen- und Fahrbahneigenschaften auf die Funktionalität von Fahrerassistenzsystemen. In E. Sucky, R. Kolke, N. Biethahn, J. Werner, & G. Koch, Mobility in a Globalised World 2016. Bamberg: University of Bamberg Press. 7. Matusko, J., Petrovic, I., & Peric, N. (2007). Neural network based tire/road friction force estimation. Zagreb (Kroatien): Elsevier. 8. Meyer, M. (2017). Signalverarbeitung – Analoge und digitale Signale, Systeme und Filter . Wiesbaden: Springer Fachmedien Wiesbaden GmbH. 9. Müller, P. D. (WS18). Handgeschriebenes Skript Adapative Systeme - Fuzzy-Logik. Ingolstadt: Technische Hochschule Ingolstadt. 10. Müller, S., Uchanski, M., & Hedrick, K. (2003). Estimation of the Maximum TireRoad Friction Coefficient. Journal of Dynamic Systems, Measurement, and Control (Vol. 125). 11. Runkler, T. A. (2015). Data Mining - Modelle und Algorithmen intellegenter Datenanalyse. Wiesbaden: Springer Fachmedien Wiesbaden. 12. Singh, K.-B., Arat, M.-A., & Taheri, S. (2012). Enhancement of Collision Mitigation Braking System Performance Through Real-Time Estimation of Tire-road Friction Coefficient by Means of Smart Tires. Virginia Tech : SAE International . 13. Wirdemann, R. (2011). Scrum mit User Stories. München: Carl Hanser Verlag München Wien.

195

Requirements & evaluation of friction information for the integration in vehicle systems 14. Wollf, K. (2017). Grundlegendes zum Bremsvorgang. In B. Breuer, & K. H. Bill, Bremsenhandbuch – Grundlagen, Komponenten, Systeme, Fahrdynamik. Wiesbaden: Springer Vieweg. 15. Zacher, S., & Reuter, M. (2017). Regelungstechnik für Ingenieure – Analyse, Simulation und Entwurf von Regelkreisen. Wiesbaden: Springer Fachmedien Wiesbaden GmbH. 16. Zimmermann, H.-J., Angstenberger, J., Lieven, K., & Weber, R. (1993). Fuzzy Technologien – Prinzipien, Werkzeuge, Potentiale. Düsseldorf: VDI-Verlag GmbH.

196

Kinetosis in autonomous driving Carsten Lecon, Carsten, [email protected] Aalen University of Applied Sciences, Anton-Huber-Str. 25, 73430 Aalen

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 M. Bargende et al. (Hrsg.), 19. Internationales Stuttgarter Symposium, Proceedings, https://doi.org/10.1007/978-3-658-25939-6_16

197

Kinetosis in autonomous driving

1 Motivation Autonomous driving is a current research and development topic. Mostly, the focus is on the technical implementation including artificial intelligence. However, further aspects like ‘social and ethic aspects’, ‘legal aspects’ and ‘ergonomics and physiological requirements’ should be taken into consideration. The latter point includes the kinetosis (travel sickness), which is often neglected, but plays an import role in autonomous driving: While the car is driving autonomously, passengers in the car are often sitting sideways or backwards, which can cause travel sickness (= motion sickness). This concerns persons in particular, who are usually sitting behind the steering wheel.This factor may even influence the buying behavior of the potential customer and could therefore be minimized by the developers of autonomous cars. Kinetosis could be one of the limiting factors of mobility concepts in the future. The user in future autonomous cars will get information from different visual sources, especially from (traditional) monitors, but – probably – also by Augment Reality glasses or Head Mounted Displays (VR glasses) for a full immersion (virtual reality).

2 Definitions The term ‚kinetosis‘ is derived from the (old) Greek word , which literally means „move“. However, if this term is used in the context of (autonomous) driving or in the context of virtual reality applications, the (negative) effects of ‘moving’ is meant: motion sickness or travel sickness. For example, this effect can often be observed at passengers in a car or on a boat. In this article, the terms ‘kinetosis’ and ‘motion sickness’ are used synonymously. The Virtual Reality Sickness (VR Sickness) is a special case of the motion sickness as well as the simulator sickness. If it is clear from the context, the term ‘motion sickness’ will sometimes be used instead of the term ‘VR sickness’. In this way, we try to minimize the word plurality.

3 Causes and Symptoms of Kinetosis 3.1 Causes of motion sickness (kinetosis) The causes of the occurrence of motion sickness are not yet fully understood. However, there exist several explanation attempts. Mostly, the explanation of motion sickness is due to the major organs of the vestibular system: the semi-circular canals and the otoliths. In general, the discrepancy between perceived locomotion and physical

198

Kinetosis in autonomous driving steadiness can cause sickness [Hettinger L J, Riccio G E (1992)]. The vestibular organ is responsible for the sense of balance. Both he macula organs (macula saccule and macula utriculi) in the inner ear are orthogonal. Is the body idle (= not moving), the macula saccule is vertical, and the macula utriculi is horizontal. These organs notify the linear acceleration. The inner ear uses three semi-circular canals (lat. Ductus semicirularis) for the notifying of the angular acceleration. Depending on the direction of movement, one canal will be stimulated more than the others. More specifically, typical theories to explain motion sickness can be found in the metastudy of Previc [Previc F H (2018)]: ● Sensory conflict theory: This is the most widely held theory. It posits that a variety of sensory conflicts create nausea and other symptoms. The most important conflict is the Coriolis (cross-coupling) effect that occurs after a head tilt during sustained rotation. For example, this effect can occur, if there exits an insufficient accordance between the visual information of the vision system and information of movement of the (simulator) platform. Cause of the occurrence of a kinetosis are the simultaneous sensor stimuli of two sensor organs. This happens, if during an optokinetic stimulus (for example viewing out of a side window of a car) a vestibular stimulus (for example a rolling turn of the car) is triggered. ● Postural instability: Conditions of postural instability are especially nauseogenic [Ricco G Stoffregen Th (1991)]. The motion sickness is not caused by a sensor conflict, but by a long lasting instability of the posture control. An abrupt or heavy changing of the environment, so that the posture control will be lost, can cause the effect. For example, this occurs if a person slips on the ice. ● Subjective verticular mismatch: This theory argues that only conflicts that lead to a misperception of the vertical relative to previous experience create motion sickness. ● Poison theory: Unusual movements of the head indicate that a poison had been ingested and, therefore, are designed to induce the vomiting in response to some vestibular lesions affect toxins. However, the effects of such lesions were selective for mainly centrally acting substances (e.g., nicotine, dopamine) that may interact with vestibular responses centrally [Money K E, Lackner J R, Cheung R S (1996)]. Beside these theories, there exist the Velocity storage theory [Laurens, J, and Angelaki, D.E. 2011], and the Otolith asymmetry theory [Baumgarten R J, Thumler R (1997)] and more.

3.2 Categories and symptoms of motion sickness Llorach, Evans, and Blat describe several factors in VR and simulator environments, which can be a trigger for VR sickness [Llorach G, Evans A, Blat J (2014)]. The factors are extensive duration, field of vision (FOV), interpupillary distance, positon tracking

199

Kinetosis in autonomous driving error, refresh rate, lag, and scene complexity. Neukum and Grattenthaler divide these factors into three categories: Individual factors, simulator factors, task factors [Neukum A, Grattenthaler H (2006)]. Regarding all these categories, it should be the target to create an environment that minimizes the probability of kinetosis. Individual factors (personal information like age, gender, experience with simulations/ VR environments, etc.): The sensitivity of kinetosis decreases with the experience in acting in virtual worlds. Ill or tired persons for example are more vulnerable to motion sickness. Also, the age plays a role: Head mounted displays should not be used by children under thirteen (children under two years feel no motion sickness – apart from the fact, that they should not use head mounted displays at all …). It is important to stop a VR session immediately, if one feels any kind of indisposition. Even the user manual of the HTC Vive says: ‘Stop using Vive if you experience discomfort.’ Furthermore, women are more likely to show kinetosis symptoms rather than men [Neukum A, Grattenthaler H (2006)]. There exist a difference between gender and the age regarding the incidence of motion sickness. Younger (less than 50 years) and female persons are more prone to motion sickness. Simulator factors (calibration, field of vision, content of the scenario, frame rate, flickering of objects, brightness, screen resolution, and many others): In order to avoid motion sickness, an adequate frame rate is important. A frame rate of at least 90 Hz for both eyes is recommended. For example, HTC Vive, as well Oculus Rift use a frame rate of 90 Hz for the rendering. This can only be achieved, if the computer hardware is equipped accordingly, regarding the processor, the graphic card and the main memory in particular. Furthermore, it should not be forgotten that the software can influence the rendering of the virtual world. One potential bottleneck is the update method, which is called very often (in order to achieve the recommended frame rate of 90 Hz), even if this method contains complex algorithms. At worst, so called frame drops result – which increases the risk of motion sickness. If the pupil distance is too low or too large, the pictures generated from the VR glasses are not matched, so that a blurred overall picture appears. Most of the head mounted displays allow this adjustment of the individual pupil distance already. Furthermore, motion sickness can result if the latency is too low (see also [Waltemate Th, Senna I, Hülsmann F, Rohde M, Kopp St, Ernst M, Botsch M (2016)]). This applies to the movements of the user, for example head movement, controller movement or moving of an avatar in the virtual environment. The latency should be less than 20 milliseconds. Actually, this is achieved by the HTC Vive (less than 11 milliseconds) and the Oculus Rift (about 13 milliseconds), the Playstation is VR slightly slower (18 milliseconds). Regarding latency the efficiency of the code plays a significant role, too. Task factors (degree of control mechanism, head-movements, height above ground, kind of movement, and many others): Especially if a user is unexperienced with regard

200

Kinetosis in autonomous driving to VR applications, she or he should take several breaks – initially at least every 30 to 45 minutes. The duration of a break depends on obvious indications of motion sickness (for example wobbling, breaking out into a sweat): The break should at least last until all symptoms are not longer recognizable. On the occurrence of symptoms of motion sickness during a session, the session should be stopped immediately; otherwise, it is possible, that the kinetosis may stay for longer time. This could be a serious problem, because the person might not be willing to use VR environments in the future. Motion in VR rooms can be done while sitting or standing. If the movement in the virtual world is intensive (for example a roller coaster), a sitting position is of advantage. If the movement in the virtual world is restricted to a stationary room, real walking (in the real room) is possible. At the beginning, one should start with VR experiences without movements. In the game ‘I Expect You To Die’ [SchellGames (2019)] all interactions take place in a sitting position: Not the player is moving to the object, but the objects are moving to the player. Slow user movements upwards or downwards are not critical. However, movements to the sides or backwards are critical. On acceleration – independent of the direction –, the body reacts very sensitive. Furthermore, rotations are often critical. Due to minimize motion sickness, sometimes black pictures are integrated in a rotation animation, similar to a teleportation. This effect can be combined with a rotation, where the viewing direction is successively changed in different degree adjustments, for example 0°, 15°, 30°, 45°, etc. In Figure 1, the different forms of possible motion sickness for linear movement (left side) and rotation (right side) are illustrated.

Figure 1: Linear movement and rotation1

1

Source: https://www.heise.de/newsticker/meldung/VR-Brille-Oculus-Rift-Hilfe-gegen-dieSimulatorkrankheit-3157108.html?hg=1&hgi=0&hgf=false

201

Kinetosis in autonomous driving Typical symptoms of kinetosis are dizziness, headache, nausea, stomach (feel different), general malaise, eructation, sweating, increased salivation, stressed eyes, blurred vision.

4 Implications for displays in autonomous cars Future kinds of visualization in cars will probably be: ● Traditional display units (lamps, digital pointers, etc.) ● Augmented Reality (AR): ● Blending in additional information (for example pointing to dangerous areas) in the front shield of the car. Here, no special device is necessary. ● Using AR glasses (for example Microsoft Hololens or Google Glass) to enrich the real view by additional information. In this case, the passenger can move the head in all directions – the computer-generated information will be adjusted accordingly. ● Using VR glasses (for example Oculus Rift or HTC Vive) for full immersion (visual and aural); the view consists of virtual generated information exclusively. ● Using MR (Mixed Reality) glasses (for example Lenovo Explorer Headset): The real world – recorded by integrated cameras – is enriched by computer-generated content. ● Voice Control systems, for example Amazon´s Alexa, Apple´s Siri, Microsoft´s Cortona), independent of the used visual devices.

The passengers in the cars will use these artificial displays with computer generated – mostly in animated 3D – content. Hereby the danger of motion sickness is given. There already exist some (well-tried) possibilities to minimize motion sickness when providing additional visual information in an animated three-dimensional shape. These Counteractions are: Visual distraction (videos, etc.) minimizes the occurrence of motion sickness. It is presumed, that auditory distraction (music, spoken text, noises) has a similar effect. Doing a specific task as a distraction minimizes motion sickness as well. In one of our laboratory experiments, the user has to answer quickly a question (distraction task) while moving through the virtual 3D world. At the DLR (German Aerospace Center), the astronauts-to-be have to perform some tasks (for example solving math problems at

202

Kinetosis in autonomous driving a computer) during their parabolic flight. It turned out, that this distraction – as well as physiacl activities (like biking) - reduced the frequency of occurrence of motion sickness. Like in traditional cars or on boats, where the person tries to find a stable reference point to look at, in a virtual 3D environment a fix point is of advantage as well [Whittinghall D, Ziegler B, Case T, Moore B (2015)]. A fix point can be a virtual nose, a cockpit or the brim of a helmet. We confirmed this effect by verifying the so-called rest in frame theory [Berg G (2014)]: We observed by eye-tracking that persons who reported kinetosis in a subsequent questionnaire showed almost nervous motion of the eyes in order to find a fixpoint, in contrast to persons without kinetosis symptoms (see Figure 2: Left: Subject without kinetosis symptoms, right: Subject with kinetosis symptoms) .

Figure 2: Heatmap from eye tracking [Deuser F, Schieber H, Lecon C (2019)]

Further counteractions are: Instead of moving through the virtual world, teleportation possibilities should be available. In addition, subtle particle effects (snow, rain, falling leaves, etc.) seem to reduce motion sickness. Moreover, one can accustom oneself to acting in VR worlds – to a certain extend.

5 Conclusion and Outlook Passengers in autonomous driving vehicles are faced with the problem of kinetosis (motion sickness). Still, that problem is often neglected. This concerns in particular drivers of cars suffering now from motion sickness, because they are not sitting at the

203

Kinetosis in autonomous driving steering wheel, but sideways or backwards. Furthermore, often – during the ride – passengers look to displays, which offer information or serve as entertainment devices. The discrepancy between the optokinetic stimulus (visible motion and acceleration) and the vestibular stimulus which (true motion and acceleration) can lead to motion sickness. In order to minimize motion sickness we have presented some methods how to design a computer generated visual output. Some of these counteractions have been verified in laboratory experiments. We will continue with our laboratory experiments, which are primarily designed for kinetosis research in VR (learning) environments ([Lecon C (2018)]). The results may have effects on the design of visual devices in real cars later on.

Abbreviations AR – Augmented Reality MR – Mixed Reality VR – Virtual Reality

Acknowledgments Part of this work was funded through the ‘HUMUS’2 program of the Federal State of Baden Wuerttemberg (Germany). We would like to thank Fabian Deuser, Hannah Schieber, Benjamin Engel and Lukas Schneider for the implementation of VR application and the performing of evaluations.

Bibliography 1. Baumgarten R J, Thumler R (1997). A model for vestibular function in altered gravitational states. Life Sci Space Res. 1979; 17:161-170 2. Berg G (2014). Das Vehicle in the Loop - Ein Werkzeug für die Entwicklung und Evaluation von sicherheitskritischen Fahrerassitenzsystemen. Ph.D. dissertation, Universität der Bundeswehr München iIn German) 3. Deuser F, Schieber H, Lecon C (2019). Kinetosis analyzation of symptoms occurrence in combination with eye tracking. 15th International Conference on Computer Science and Education. Athens (Greece), 20 -23 May 2019 (accepted for publication)

2

HUMUS: Hochschuldidaktisch- und methodisch unterstützte Selbstinitiierung von Lernprozessen an Hochschulen für Angewandte Wissenschaften in Baden-Württemberg (03-12 2018).

204

Kinetosis in autonomous driving 4. Hettinger, L J, Riccio G E (1992). Visually induced motion sickness in virtual environments. Presence: Teleoperators and Virtual Environments. Volume 1, Issue 3, summer 1992, pages 306-310 5. Laurens J, Angelaki D E (2011). The functional significance of velocity storage and its dependence on gravity. Exp. Brain Res. 2011; 210(3-4):407–422 6. Lecon C (2018) Motion Sickness in VR Learning Environments (2018) 14th International Conference on Computer Science and Education. Athens (Greece), 21-24 May 2018. ATINER Conference Series, No. COM2018-2514, pp 1–16 7. Llorach G, Evans A, Blat J (2014). Simulator sickness and presence using HMDs: comparing use of game controller and position estimate system. VRST’14: Proceedings of the 20th ACM Symposium on Virtual Reality – Software and Technology. Edinburgh (Scotland), November 11-13, pages 137–140. 2015 8. Money K E, Lackner J R, Cheung, R S (1996). The autonomic nervous system and motion sickness. In Yates, B.J, Miller, A.D. (eds.). Vestibular autonomic regulation. Boca Ragon (FL): CRC, 1996:147–173 9. Neukum A, Grattenthaler H (2006). Kinetose in der Fahrsimulation. ‘Simulation von Einsatzfahrten’ im Auftrag des Präsidiums der Bayerischen Bereitschaftspolizei; Abschlussbericht. Würzburg: Interdisziplinäres Zentrum für Verkehrswissenschaften an der Universität Würzburg. 2006 (in German) 10. Previc F H (2018). Intravestibular Balance and Motion Sickness. 2018. Aerospace Medicine and Human Performance, Vol. 89, No. 2, February 2018 11. Ricco G, Stoffregen Th (1991). An Ecological Theory of Motion Sickness and Postural Instability. Ecology Psychology 3 (3): 195-240. September 1991 12. SchellGames (2019). https://www.schellgames.com/games/i-expect-you-to-die. Access 20 January 2019 13. Waltemate Th, Senna I, Hülsmann F, Rohde M, Kopp St, Ernst M, Botsch M (2016). The Impact of Latency on Perceptual Judgement and Motor Performance in Closedloop Interaction in Virtual Reality. VRST’16: Proceedings of the 22th ACM Symposium on Virtual Reality – Software and Technology. Munich (Germany), November 2-4, pages 27–35. 2016 14. Whittinghall, D, Ziegler B, Case T, Moore, B (2015). Nasum Virtualis: A Simple Technique for Reducing Simulator Sickness. In Games Developers Conference (GDC). 2015

205

One-Stop-Test solutions for autonomous driving Frank Heidemann SET GmbH

This manuscript is not available according to publishing restriction. Thank you for your understanding.

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 M. Bargende et al. (Hrsg.), 19. Internationales Stuttgarter Symposium, Proceedings, https://doi.org/10.1007/978-3-658-25939-6_17

207

Phenomenology and analysis of gas pressures at low-speed pre-ignitions Dr.-Ing. Christoph Beerens, Dipl.-Ing. Rainer Fischer, Dipl.-Ing. Christian Trabold MAHLE GmbH, Pragstr. 26-46, 70376 Stuttgart

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 M. Bargende et al. (Hrsg.), 19. Internationales Stuttgarter Symposium, Proceedings, https://doi.org/10.1007/978-3-658-25939-6_18

209

Phenomenology and analysis of gas pressures at low-speed pre-ignitions

Abstract The phenomenon of low-speed pre-ignition (LSPI) has gained widespread significance with the growing implementation of downsizing concepts in gasoline engines. Failures of piston top land, second land or skirt are inducing catastrophic engine failures both in testing and field applications. In order to assess the influence of LSPI on the components’ stability the application of realistic load cases is crucial. Classical approaches are not applicable here any more. In this publication a new method is presented, which employs measurement and simulation in order to derive realistic load cases for LSPI influence on pistons. This is done by correlation of gas dynamics measurements and simulations in order to model the propagation of pressure waves during LSPI events. Problems still exist due to the stochastic nature of LSPI pressure amplitudes and subsequent damage accumulations leading to a piston failure. In spite of that, conclusive evidence found about the damage mechanisms allows for a much improved understanding of actual engine failures.

Abstract (in German) Bei Ottomotoren hat das Phänomen der Vorentflammung mit zunehmendem Downsizing an Bedeutung gewonnen. Brüche des ersten Ringsteges oder des Feuersteges können bereits nach wenigen Betriebsstunden zu Motorschäden führen, leider nicht nur in der Erprobung. Für die Auslegung von Kolben gegen das Versagen unter Vorentflammung ist die Anwendung realitätsnaher Lastfälle entscheidend, aber klassische Verfahren greifen hier nicht mehr. In diesem Beitrag wird eine neue Methode der Messung und Simulation zur Ableitung realistischer Druckrandbedingungen bei Vorentflammung am Kolben vorgestellt. Diese Anwendung der Gasdynamik bei der Ausbreitung von Druckwellen aus der Vorentflammung führt zu einer guten Korrelation von Versuch und Simulation. Mit der FEM lassen sich durch diese neue Methode typische Schäden am Kolben gut erklären. Probleme bestehen weiterhin in der stochastischen Variation der Druckamplituden bei den realen Vorentflammungen im Motor und der nachfolgenden Schadensakkumulation. Dennoch erlauben die gefundenen Phänomenologien der Druckwellenausbreitung schon eine schlüssige Ursachenanalyse der in der Praxis gefundenen Schäden.

210

Phenomenology and analysis of gas pressures at low-speed pre-ignitions

1 Description of field problems With the advent of gasoline direct-injection and downsizing concepts in modern engines low-speed pre-ignition (LSPI) has gained attention in the industry. Self-ignition before spark-ignition generates excessive pressures inside of the combustion chamber and induces mechanical failures of the pistons, the rings and the conrods as well. Downsized gasoline engines are prone to LSPI in the low-end torque region predominantly, at low engine speeds and full load.

Figure 1: First and second land LSPI piston failures

This paper will focus on the pistons' rather complicated pressure and failure situation, omitting conrod aspects. As there are many research activities currently working on LSPI avoidance, we will limit the scope here to the investigation of the piston failure modes, assuming there simply are LSPI events, not caring about their origins and root causes. The aim thus is merely a development of LSPI tolerant pistons.

211

Phenomenology and analysis of gas pressures at low-speed pre-ignitions In order to gain more LSPI tolerance, the phenomenology of the events' pressure distributions was investigated in order to improve stress-strain loading prediction and modelling as well as mechanical failure modes of the piston materials. Some of the actual field failures could not be explained by classical approaches in simulations up to now. Top land failures with crack initiations directed upwards e.g. were so far inaccessible for explanation and simulation. This study of LSPI phenomenology was done a) in fired engines, b) in combustion chamber models (with explosives) and c) by gas dynamics simulations, as will be described in the following chapters.

2 Pressure measurements in fired engines Looking at actual measurements of LSPI pressure distributions with standard single-gauge indication of the respective cylinders one can find excessive pressure amplitudes at very high frequencies, for example here in Figure 2 on cylinder 6. This is completely different to the normal combustion event, as can be seen in the other graphs for cylinder 1-5. 160 bar 140 bar 120 bar 100 bar 80 bar 60 bar 40 bar 20 bar ‐60°CA

‐30°CA

0 bar 0°CA

30°CA

60°CA

90°CA

120°CA

Figure 2: Engine testing, standard piëzos only, LSPI on cyl. 6

Surely such events with extremely high pressure gradients cannot be modelled as uniform pressure distributions in the combustion chamber anymore; they clearly constitute L000

212

Phenomenology and analysis of gas pressures at low-speed pre-ignitions gas-dynamics events. Thus cylinders of an engine with LSPI issues were equipped with a multitude of pressure sensors in order to gain insight into the time- and space-dependency of the pressures inside of the cylinders.

L090

L000

L180 Figure 3: Sensors in cylinder head (CH1-CH6) and liner (L000, L090, L180)

he engine then was operated in the relevant speed range of 1700-2000 rpm at full load and pressures were recorded only during non-conformal events: High pressure peaks in combination with very high frequencies. During these LSPI events the pressure distributions were found to be extremely nonuniform, in terms of amplitudes and timings both, see Figure 4 and Table 1 below. The standard piëzo sensor is plotted in Figure 4 as sensor “CH1” with a dashed line, clearly giving no indication of the absolute maximum pressures in the liner area. The 5 different LSPI events in Table 1 are typically non-uniform in both timing and amplitude. There are some similarities, but one can make the assumption that in all 5 cases the spatial origin of the LSPI events was a different one.

213

Phenomenology and analysis of gas pressures at low-speed pre-ignitions

Figure 4: LSPI event from a fired engine test, one cylinder only (event #2)

Table 1: Comparison of five sequential LSPI events Event # 1 2 3 4 5

Timing CH 1 CH 2 CH 3 CH 4 CH 5 CH 6 L000 L090 L180 618.3° 145 148 407 287 305 220 279 248 289 607.5° 130 150 156 154 162 154 178 178 161 623.5° 85 89 86 92 89 90 97 93 105 614.0° 115 118 115 123 121 125 133 130 130 95 106 102 104 108 104 131 107 107 620.0°

Key finding of all these measurements was the location of maximum pressures during all of the recorded events: The gap position between the liner and the piston top land. While there were a lot of non-uniform events with different peak locations recorded, those maxima always occurred in the top-land to liner gap, not in the combustion chamber itself. This was considered crucial for the understanding of actual piston failures, as it delivered a completely new pressure load distribution model. Another crucial point in this discovery is that the standard pressure gauges employed for indication of cylinder pressures never pick up the actual load on the piston lands. In these standard measurements of combustion chamber pressures there is no key to the actual timings and amplitudes acting on the piston’s top land and top ring.

214

Phenomenology and analysis of gas pressures at low-speed pre-ignitions

3 Measurements using explosives in an experiment with combustion chamber geometry In order to better understand gas dynamics inside of the combustion chamber during the LSPI events, a test rig model with combustion chamber geometry was created, identical to the engine geometry employed for the fired measurements. Here the piston position was also defined identically to the “start of LSPI event” position in the engine. Cylinder head and liner portion were machined in one solid part. Pistons and ring pack were assembled together with a pin into this test setup. Inside of this test rig we applied pressure gauges in the standard piëzo location and also in the gap between liner and top land. The LSPI pressure event was then physically simulated by a small explosive charge (“bomb” in Figure 4 Figure 5) in the center of the combustion chamber. The mass of this charge was properly scaled in order to match the pressure levels inside of the fired engine.

Figure 5: Experimental model, half-view of liner, piston and cylinder head geometry (rings and pin not plotted here, but installed in test)

215

Phenomenology and analysis of gas pressures at low-speed pre-ignitions Surely every change of the ignition position will change the spatial distribution of the pressure waves, but the effect of relatively higher pressures inside of the gap is unaffected by this change. Highly transient pressure waves during explosion were recorded just as in the real fired engine, see Figure 6. Maximum values were 129bar at the standard piëzo position and 237bar in the gap. This is a relation very similar to the fired engine’s pressure distribution. Typically again, maximum pressures in the gap occur earlier than those in the standard measurement position. CH1 14322 L090 14322

250 bar

237 bar

CH1 14322 L090 14322

200 bar

150 bar

129 bar

100 bar

50 bar

0 bar

-50 bar

-100 bar -5

4.0·10 s

-5

8.0·10 s

-4

1.2·10 s

-4

1.6·10 s

-4

2.0·10 s

-4

2.4·10 s

-4

2.8·10 s

Figure 6: Experimental transient pressures CH1 and L090

Thus a mere gas-dynamic event via explosion without gasoline fuel, oil droplets etc. does generate such an unexpected location of maximum pressures on the piston. This greatly simplifies realistic modelling of the pressure distributions inside of the cylinder. Now with gas-dynamics simulation alone details of the pressure distributions can be modelled and studied. All specifics about oil formulations, droplet size and temperature of oil and fuel, soot particles from combustion residuals etc. can be neglected in the specific investigation of correlations between pressure distributions and actual mechanical LSPI piston failures.

216

Phenomenology and analysis of gas pressures at low-speed pre-ignitions

4 Root cause analysis via gas dynamics analysis of the combustion chamber In a further step of investigations the experiment with explosive charges inside of the combustion chamber geometry was modelled by a gas dynamics simulation code. The explosive charge from the experiment was modelled as a source of high pressure gas, released at charge ignition time. #1: Std. piëzo pos.

Explosive charge pos.

Figure 7: Numerical model with pressure gauges

Numerical pressure gauges now were applied in the same locations as in the experimental setup and also in the fired engine. Transient simulation pressures were numerically recorded and gauge signals were plotted on top of each other.

217

G01 Std [Pa] G02 Ring [Pa] G17 Edge [Pa]

Phenomenology and analysis of gas pressures at low-speed pre-ignitions 4·10

7

3·10

7

2·10

7

2·10

7

8·10

6

0·10

0

G01 Std [Pa] G02 Ring [Pa] G17 Edge [Pa]

0

0.2

0.4

0.6

Time [ms]

Figure 8: Transient pressures on 3 positions in 3D gas dynamics simulation

Again, as in the fired engine and in the explosive experimental setup, pressures inside of the gap between liner and top land were higher than in the combustion chamber itself. The pressure level at the standard gauge position in the cylinder head again did not yield any prediction of the actual top land peak pressures’ amplitude or timing on the piston.

5 2D-simulation gas dynamics of top land pressures

liner

In order to study details of the pressure situation at top land and first ring land, a simplified 2D-geometry was created, which describes the gas-dynamics around top land, liner and top ring, Figure 9.

G1 G2 G3

105 bar top land top ring

140 bar piston crown

Figure 9: Numerical 2D-model of top land gas-dynamics at t=0sec

218

Phenomenology and analysis of gas pressures at low-speed pre-ignitions Here the effect of a shock wave impinging vertically on a wall (i.e. the liner here) with a small channel (i.e. the gap here) branching off at 90° was all the simulation model required. Numerical pressure gauges were applied at the piston crown edge, in the channel at 50% height and at the first ring position (G1-G2-G3 respectively). Resolution of the pressure measurements in a gas dynamics simulation is much higher, naturally, as one can monitor pressures in a single point, while the physical pressure gauges applied in both the engine and the experiment here had a diaphragm diameter of about 3mm. 7

2·10

7

2·10

7

G1 [Pa] G2 [Pa] G3 [Pa]

3·10

10

G1 [Pa] G2 [Pa] G3 [Pa]

7

6·10

6

0·10

0

0

0.06

0.12

0.18

0.24

Time [ms]

Figure 10: Pressure transients on 2D-Model for gauges G1, G2, G3

Again maximum pressures were located at the bottom of the gap (gauge G3), like in all testing and simulation described here before, Figure 10. Comparison of the pressures derived from the theory of pressure waves impinging vertically on a hard wall and branching off into a relatively narrow channel correlates better with all these results the farther off from the edge of the piston crown we are. This is due to the quite complicated physics of a slightly supersonic pressure front bending around the piston crown edge and interacting with the liner wall. At G2-G3 there is a uniform 1D-channel flow again, fitting much more closely to theory.

219

liner

Phenomenology and analysis of gas pressures at low-speed pre-ignitions

G1 G2 G3

105 bar top land

140 bar piston crown

top ring gap

G4 G5 Figure 11: 2nd channel and G4 and G5 gauges added (not to scale!), model at t=0sec

G1 [Pa] G2 [Pa] G3 [Pa] G4 [Pa] G5 [Pa]

A further step of 2D-modelling then adds the small gap between top land and first ring into the previous model, yielding two more numerical pressure gauges G4 and G5, Figure 11. The model now contains a second channel of reduced width, again branching off orthogonally. This second event thus is added serially and adding further pressures into the top-ring to top-land gap. The resulting pressure transients are plotted in Figure 12, based on these two serial gas dynamics events. 24000000

G1 G2 G3 G4 G5

20000000

16000000

12000000

8000000

4000000

0.08

0.12

0.16

0.2 Time [ms]

Figure 12: Gauge pressures 1-5 in extended gap model

220

[Pa] [Pa] [Pa] [Pa] [Pa]

Phenomenology and analysis of gas pressures at low-speed pre-ignitions This further pressure increase cannot be measured in an experiment, even less so in an engine. There simply are no microscopic pressure sensors giving access to these small gaps. But the total forces onto first ring and top land derived from these pressure transients (and pressure differences) can easily explain the first two common failure modes of the piston: 1. Second land failure due to downward pressure force onto the top ring, which subsequently hits onto second land 2. Top land failure due to pressures in both liner and ring gaps acting upwards By transfer of the gas-dynamics pressures onto an explicit FEA model of the piston (i.e. only top land, first groove and second land model), the pressures yield local notch stresses and strains in the first ring groove’s inner radiae.

Top land

first ring groove

Second land Figure 13: Stresses in the first ring groove, total maxima during LSPI event

These stresses are of a magnitude where crack initiation towards top land or second land can easily be explained. By slight changes in the relation of upper and lower radiae, failure can easily be shifted up- or downwards. In effect this model yields conclusive explanation for current testing experience. An open issue in terms of material characteristics is the definition of the relevant material properties. Standard piston FEA is employing high cycle fatigue (HCF) limits at relatively low stress levels required for engine lifetime. LSPI creates failures at low cycle numbers, typically 50 to 1500 cycles (i.e. LSPI events) after relatively short

221

Phenomenology and analysis of gas pressures at low-speed pre-ignitions runtime. This is more similar to low cycle fatigue (LCF) behaviour, but due to extremely high strain rates on the other hand very much unlike traditional LCF. Selection and generation of the relevant material characteristics and selection of the most LSPI tolerant piston materials is a new task, which is already in progress.

6 Piston skirt failure conditions Another failure mode is the piston skirt fracture due to LSPI events, as depicted here in Figure 14.

Figure 14: Piston skirt LSPI failures, 2 examples

In excellent correlation for this case, a gas dynamic study into the torque on the piston due to transient gas pressure distributions shows quite a hard tilting moment of about 420 Nm around the pin axis, Figure 15. The impulse onto the piston is rather short, but it will definitely affect secondary motions during LSPI events nonetheless.

222

Mtot [Nm]

Phenomenology and analysis of gas pressures at low-speed pre-ignitions 200

0

-200

-400

-600

0

0.05

0.1

0.15

0.2

0.25

Zeit [ms]

Figure 15: Piston torque due to transient pressures

The tilting movement of the piston is finally limited by the liner and consequently creating high stresses in the skirt area. These gas dynamic pressure boundary conditions make the mechanical explanation of piston skirt failure modes quite straightforward now. An assessment of local effects can now be done based on realistic pressure boundary conditions, leaving only the proper material failure model open.

7 Conclusions By application of experimental and numerical gas dynamics combined with measurements in a fired engine, transient LSPI pressure loading inside of the combustion chamber was analysed. Especially the critical loadings just above the first ring or inside of the first ring groove were found to be in excellent correlation with the observed LSPI failure modes of pistons. Employing this gas dynamics approach the actual mechanical loading on the piston material is well understood now and will be further applied in improving the piston’s LSPI tolerance. This will be done by investigation of the required material properties and failure models in future.

223

Phenomenology and analysis of gas pressures at low-speed pre-ignitions Still the stochastic nature of LSPI events does not allow exact predictions of where failures may occur, but as the failure mechanisms themselves are completely understood. Piston LSPI tolerance can now be increased based on knowledge rather than trials, which is already in progress (Q1, 2019). Damage models for LSPI-typical strain amplitudes and strain rates will be investigated, tested, evaluated and applied in future.

Bibliography 1. Klomfass, A., Stolz, A., Hiermaier, S., 2016, Improved explosion consequence analysis with combined CFD and damage models, Chemical Engineering Transactions, Vol. 48, 109-114 2. M. Amann, T. Alger, und D. Mehta, The Effect of EGR on Low-Speed Pre-Ignition in Boosted SI Engines, In: SAE 2011 World Congress & Exhibition, Detroit, Nr. 2011-01-0339, 2011 3. W. Chester, The quasicylindrical shock tube, phil. Mag., vol. 45, pp. 1293, 1954 4. F. Chisnell, The motion of Shock wave in a channel with application to cylindrical and spherical shock waves, J. Fluid Mech., vol. 2, pp. 286, 1957 5. C. Dahnz, K-M. Han, und M. Magar, Vorentflammung bei Ottomotoren, Untersuchung des Auftretens und der Ursache von Selbstzündung vor Zündungseinleitung bei aufgeladenen Motoren mit hohem Verdichtungsverhältnis, FVV-Vorhaben Abschlussbericht, vol. 907, pp 1–58, 2010 6. T. Döge. Zur Reflexion von Luftstoßwellen an nachgiebigen Materialien und Baustrukturen, Dissertation Universität der Bundeswehr München, pp. 60–85, 2012 7. F. A. Ettner, effiziente numerische Simulation des Deflagrations-Detonations-Übergangs, Dissertation, Technische Universität München, pp. 19, 2012 8. A. V. Fedorov, Yu. V. Kratova, T. A. Khmel, Numerical study of shock wave diffraction in variable section in gas suspensions, Combustion, Explosion and Shock Waves, Vol. 44, No 1, pp. 76–85, 2008 9. H. Grönig, Dämpfung von Stoßwellen in verzweigten Rohrsystemen, Forschungsberichte des Landes Nordrhein-Westfalen, Nr. 2793, Westdeutscher Verlag, pp 17–18, und 49, 1978 10. S. Schultz, Über die Ausbreitung von Stoßwellen im abgeknickten und verzweigten Rohren, Forschungsberichte des Landes Nordrhein-Westfalen, Nr. 2119, Westdeutscher Verlag Köln und Opladen, pp 5–6 und 49, 1970 11. S. Schultz, Eine theoretische und experimentelle Untersuchung zur Beugung von Stoßwellen, Dissertation Technische Hochschule Aachen, pp. 7, 1971

224

Phenomenology and analysis of gas pressures at low-speed pre-ignitions 5. C. Dahnz, K-M. Han, und M. Magar, Vorentflammung bei Ottomotoren, Untersuchung des Auftretens und der Ursache von Selbstzündung vor Zündungseinleitung bei aufgeladenen Motoren mit hohem Verdichtungsverhältnis, FVV-Vorhaben Abschlussbericht, vol. 907, pp 1–58, 2010 6. T. Döge. Zur Reflexion von Luftstoßwellen an nachgiebigen Materialien und Baustrukturen, Dissertation Universität der Bundeswehr München, pp. 60–85, 2012 7. F. A. Ettner, effiziente numerische Simulation des Deflagrations-Detonations-Übergangs, Dissertation, Technische Universität München, pp. 19, 2012 8. A. V. Fedorov, Yu. V. Kratova, T. A. Khmel, Numerical study of shock wave diffraction in variable section in gas suspensions, Combustion, Explosion and Shock Waves, Vol. 44, No 1, pp. 76–85, 2008 9. H. Grönig, Dämpfung von Stoßwellen in verzweigten Rohrsystemen, Forschungsberichte des Landes Nordrhein-Westfalen, Nr. 2793, Westdeutscher Verlag, pp 17–18, und 49, 1978 10. S. Schultz, Über die Ausbreitung von Stoßwellen im abgeknickten und verzweigten Rohren, Forschungsberichte des Landes Nordrhein-Westfalen, Nr. 2119, Westdeutscher Verlag Köln und Opladen, pp 5–6 und 49, 1970 11. S. Schultz, Eine theoretische und experimentelle Untersuchung zur Beugung von Stoßwellen, Dissertation Technische Hochschule Aachen, pp. 7, 1971

225

Combustion stability improvement with turbulence control by air injection for a lean-burn SI engine Takanori Suzuki SOKEN Inc. Bastian Lehrheuer, Tamara Ottenwälder, Max Mally, Stefan Pischinger Institute for Combustion Engines / RWTH Aachen University

Combustion stability improvement with turbulence control by air injection for …

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 M. Bargende et al. (Hrsg.), 19. Internationales Stuttgarter Symposium, Proceedings, https://doi.org/10.1007/978-3-658-25939-6_19

227

Combustion stability improvement with turbulence control by air injection for …

1 Introduction Minimizing crude oil consumption is one of the major challenges for automotive industry to keep a sustainable society. Especially, transportation is one of the major sectors of energy consumption [1] and therefore improvement of thermal efficiency is one of the important roles for automotive engine systems. Improvement of thermal efficiency of gasoline engines is necessary as the main propulsion system of passenger cars. Lean combustion diluted with EGR (Exhaust Gas Recirculation) or intake air is one of the evolutional strategies to improve efficiency. Though, one of the issues is an increase of combustion instability. To address this issue, the stratified combustion concept is being developed using piezoelectric-actuated injectors [2] and corona ignition system [3]. In this paper, an alternative concept was investigated to improve combustion stability by controlling in-cylinder air flow.

2 Concept of Turbulence Enhancement 2.1 Combustion Fluctuation Combustion fluctuation has to be reduced below a certain level. Combustion fluctuation behaviour affected by excess air ratio is evaluated. The investigations were carried out on the single cylinder research engine of the Institute for Combustion Engines at RWTH Aachen University with the engine specification shown in Table 1. The operation condition is shown in Table 2. Result is shown in Figure 1. Left figure shows the relationship between IMEP (Indicated Mean Effective Pressure) and the period of main combustion (10-90 % heat release ratio). Lean mixture (excess air ratio: 1.8) leads to larger combustion fluctuation than stoichiometric mixture due to main combustion fluctuation. As seen in the right figure, main combustion fluctuation depends on initial combustion (period of 0-10 % heat release ratio) fluctuation. Especially for very lean conditions it appears that the stable initial combustion leads to the stable main combustion. Table 1: Specification of the single cylinder research engine Bore Stroke Displacement Volume Compression Ratio Fuel Injection Type

228

75 / mm 82.5 / mm 364 / cm3 7 – 13.5 Port Fuel Injection, Side-mounted Direct Injection, Centre-mounted Direct Injection

Combustion stability improvement with turbulence control by air injection for … Table 2: Engine operation condition of combustion fluctuation Engine Speed Load (IMEP) Compression Ratio Fuel Injection Type Fuel Fuel Pressure Excess Air Ratio

1500 / (1/min) 6.8 / bar 13.0 Centre-mounted Direct Injection E100 200 / bar 1.0, 1.8

Figure 1: Relation between initial combustion and main combustion

2.2 Period of Initial Combustion and Turbulence There is a well-known relationship between the initial combustion fluctuation and the period of the initial combustion; the fluctuation of the initial combustion depends on the period of the initial combustion [4]. P. G. Hill explained this in his theory by indicating that the period of the initial combustion (𝑇) depends on the turbulence intensity (𝑢′) and the turbulence length scale (𝐿) as shown in the equation (1) [5]. 𝑇

(1)

Here, 𝑢 is the laminar burning velocity which is determined by the fuel properties, and λ is Taylor microscale.

229

Combustion stability improvement with turbulence control by air injection for … This indicates an importance of reducing initial combustion fluctuation for increasing the turbulence intensity 𝑢′ and decreasing the turbulence length scale L in the combustion chamber. This can also lead to shorter initial combustion period by earlier transition from laminar to turbulence combustion and therefore enhance the flame kernel growth [6].

2.3 Concept of Turbulence Enhancement by Direct Air Injection Based on previous survey results, the turbulence enhancement concept was developed as shown in Figure 2. The air injector is connected to the high pressure air source and mounted close to spark plug (left figure). Small scale turbulence is generated at the spark plug gap by sharing force between the ambient air and the injected air. The injected air has strong kinetic energy due to high injection pressure and therefore the turbulence intensity is high at the area of the injected air. Thus an optimized turbulence can be prepared for reducing the fluctuation of the initial combustion period, shorten the combustion period resulting the improvement of the combustion fluctuation.

Figure 2: Turbulence enhancement concept

3 Optimization of Air Injection Parameters 3.1 Parameters of Air Injection A formation process of turbulence in a combustion chamber was evaluated with Converge [7] of 3D-CFD (Computational Fluid Dynamics) software in order to optimize parameters of the air injector. Converge automatically resizes the mesh with the embedded AMR (Adaptive Mesh Refinement) function based on the calculated physical quantities such as turbulence scale in each time-step. This enables easier compromise

230

Combustion stability improvement with turbulence control by air injection for … of the accuracy and the cost of the calculation. Standard k-ε model of RANS (RaynoldsAveraged Navier-Storks) was deployed as the calculation scheme of turbulence. Effects of parameters in Figure 3 on turbulence intensity, scale and excess air ratio were evaluated. The evaluation scheme of nozzle direction (θ) is described in chapter 3.3. Hereafter, θ means the angle between air injection direction and centre of spark plug gap. Compressed air pressure was set to 80 bar to inject into high pressure ambient at TDC (Top Dead Centre).

Figure 3: Parameters of air injection

3.2 Condition of Turbulence Enhancement Concept The single cylinder research engine (Table 1) was deployed to evaluate the turbulence enhancement concept. Evaluation condition is shown in Table 3. Table 3: Engine operation condition of turbulence enhancement concept Engine Speed Load (IMEP) Compression Ratio Fuel Injection Type Fuel Fuel Pressure Excess Air Ratio

1500 / (1/min) 6.8 / bar 12.0 Side-mounted Direct Injection Gasoline E10 200 / bar 1.6

231

Combustion stability improvement with turbulence control by air injection for …

3.3 Direction of Air Injection Turbulence intensity, turbulence length scale and the excess air ratio after the air injection into combustion chamber were calculated by CFD to evaluate the effects of the air injection. Results are shown in Figure 4. Compared with the baseline, small-scale turbulence is spread toward the spark plug from the air injector with air injection. The area of small-scale turbulence is wider than the area of strong turbulence in any case. This suggests that the small-scale turbulence is generated by shearing force due to difference of velocity between ambient gas and injected air. However, excess air ratio is higher than the baseline at the area of injected air in any case with air injection. Over-lean mixture prevents flame kernel growth therefore air injection timing and air injection direction (θ) have to be properly adjusted with consideration of ignition timing.

Figure 4: Effects of nozzle direction

Figure 5 shows calculated turbulence intensity, turbulence length scale and excess air ratio evaluated in a spherical region of 10 mm diameter at the centre of the spark plug. The end of air injection was set to 26° CA bTDC (before Top Dead Centre) as the mixture at the spark plug gap will not be over-lean at the ignition timing, and the period of air injection was set to 1ms. Strong and small-scale turbulence is generated after the air injection; however, the turbulence is dissipated afterwards. The turbulence intensity behaviour does not depend on nozzle direction within 0 to 14°. However, turbulence length scale depends on nozzle

232

Combustion stability improvement with turbulence control by air injection for … direction and has a minimum value at θ = 8.5°. A decrease of turbulence intensity is quicker than that of turbulence length scale; small-scale turbulence induced by the air injection keeps longer effect than turbulence intensity. Excess air ratio at the spark plug gap increases during air injection and rapidly decreases after the injection. Effect of the injected air on the air fuel ratio is limited up to about 6° CA, so it will be more suitable to set the ignition timing a short interval of time after the air injection, not during or right after the air injection.

Figure 5: Turbulence behaviour after air Injection at spark plug gap

Effect of air injection direction (θ) on the period of initial combustion was evaluated with the prediction model of initial and main combustion which was developed at RWTH-Aachen University [6]. This model can estimate an initial combustion period (0-5 %) with 10 % accuracy. Result is shown in Figure 6. The period of initial combustion becomes shorter than the baseline in any case with air injection of which ignition timing was set 6° CA after the end of air injection. Especially the period of initial combustion becomes shorter by 6.3° CA between 0 to 8.5° of the nozzle direction. The optimized nozzle direction (θ) is determined as 8.5° from centre of spark plug gap based on this result.

233

Combustion stability improvement with turbulence control by air injection for …

Figure 6: Calculated periods of initial combustion and main combustion

3.4 Optimized Specification of Air Injection Air injection specification was determined as shown in Table 4 on the basis of the CFD evaluation. Nozzle direction (θ) is set to 8.5° away from spark plug centre to prepare optimized turbulence field, and nozzle hole diameter (D) is set to ϕ0.2 mm of straight single hole. Compressed air pressure (𝑃𝑐 ) is set to 80 bar. Table 4: Air injector specification Nozzle direction (θ) Nozzle hole diameter (𝐷) Compressed air pressure (𝑃 )

8.5° away from centre of spark plug gap 𝜙0.2 mm, single hole of straight hole shape 80 / bar

4 Experimental Setup An improvement of the period and deviation of initial combustion was confirmed with an engine. Therefore an optical analysis in addition to a heat release was evaluated. Furthermore, the improvement of lean limit was clarified. The single cylinder research engine of Table 1 was applied.

4.1 Air Injection System Benefit of air injection was clarified. Figure 7 shows the experimental apparatus. A highpressure air was supplied from an air vessel to the air injector with 80 bar. The air pressure and flow quantity were regulated. Two accumulators were set between the air vessel and the injector to reduce pressure fluctuation due to intermittent air injection. Moreover, temperature of air supply line was controlled to reduce quantity deviation of injected air.

234

Combustion stability improvement with turbulence control by air injection for … Air injector, fuel injector, ignition and other equipment were controlled with a prototyping ECU (Engine Control Unit).

Figure 7: Experimental apparatus

4.2 Optical Analysis The optical access was prepared at side of pent-roof between intake and exhaust [8]. Figure 8 shows the observation area in the combustion chamber. A digital high-speed camera (Photron [9]) is used with a high-speed image intensifier (LaVision [10]), in order to enhance the sensitivity. The measurements are conducted with a frame rate of 18 kHz, corresponding to a resolution of 0.5°CA. The exposure time of each single image is approximately equal to this resolution.

Figure 8: Observation area of the flame kernel

235

Combustion stability improvement with turbulence control by air injection for …

5 Results and Discussion 5.1 Improvement of Initial Combustion Period with Air Injection An effect of air injection on the period of initial combustion was clarified. The heat release ratio shown in Figure 9 was evaluated on n = 1500 1/min, IMEP = 6.8 bar, excess air ratio=1.55. The start timing of the heat release was compared in both cases of the air injection and the baseline. The heat release occurs earlier by applying the air injection. This result shows that the air injection concept promotes the flame kernel growth.

Figure 9: Effects of air injection (the period of initial combustion)

5.2 Improvement of Initial Combustion Fluctuation with Air Injection Effects of the air injection on the fluctuation of initial combustion were evaluated. Variation of initial combustion speed was optically investigated as the burned region at n = 1500 1/min, IMEP = 6.8 bar, excess air ratio=1.55. Figure 10 shows outlines of burned region of each cycle at the timing of 5% heat release ratio calculated by the pressure in the combustion chamber. Fluctuation of initial combustion speed was improved with the air injection compared with that of the baseline.

236

Combustion stability improvement with turbulence control by air injection for …

Figure 10: Outlines of burned region at 5% heat release ratio (overdrawn 100 cycles)

Fluctuation of the gravity centre of the initial combustion was also evaluated. Figure 11 indicates the distribution of the gravity centre of 100 cycles and the distribution area with the double of standard deviation (2σ). The area of distribution with the air injection is smaller than that of the baseline. This implies that the injected air flow is more robust and strongly affects to the initial flame propagation than in-cylinder flow generated in intake process. These results show that the air injection improves robustness of the initial combustion speed and the gravity centre of the initial combustion region.

Figure 11: Distribution of gravity centre of initial combustion region

237

Combustion stability improvement with turbulence control by air injection for …

5.3 Improvement of Lean Limit with Air Injection Effects of air injection on combustion period were clarified. Period of initial combustion (period of 0-5 % heat release ratio) was evaluated under different excess air ratio condition on n = 1500 1/min, IMEP = 6.8 bar. Results are shown in Figure 12. The period of initial combustion was improved by 7 to 10° CA. This improvement corresponds to 0.15 to 0.2 point difference of excess air ratio at the spark plug gap with the same initial combustion period. Based on this result, it is clarified that the air injection can generate stronger and smaller scale turbulence, which is suitable to promote an initial flame kernel growth and improve the period of initial combustion.

Figure 12: Improvement of the period of initial combustion

Benefits of air injection on ISFC (Indicated Specific Fuel Consumption) by extended lean limit are evaluated. In this paper, lean limit is defined as excess air ratio at the point where standard deviation of IMEP reaches 0.15 bar. Evaluation results are shown in Figure 13. Air injection ensures less S.D. (standard deviation) of IMEP, and extends lean limit by 0.15 point of excess air ratio (1.5 to 1.65) (left figure). This effect of lean limit extension corresponds to an improvement of initial combustion process in Figure 12. As a result, ISFC was improved by 1.4 % due to increase of thermal efficiency by lean limit extension and better combustion process by high-pressure air injection (right figure).

238

Combustion stability improvement with turbulence control by air injection for …

Figure 13: Benefits of air injection on lean limit and ISFC

Injected air quantity of each cycle is 1.4 mg under the lean limit condition. This quantity accounts for only 0.37 % of intake air into the combustion chamber. Therefore, considerable ISFC improvement of 1.1% is still possible even taking account of compression work for the air injection (Figure 14). Thus, the air injection can generate suitable turbulence field around the spark plug despite small quantity. This turbulence promotes an initial flame kernel growth, which leads to an improvement of the initial combustion period. Due to this improvement, main combustion process corresponding to the combustion fluctuation is also improved. Consequently, ISFC is improved by lean limit extension.

Figure 14: Benefit of air injection on ISFC including air compression work

239

Combustion stability improvement with turbulence control by air injection for … Figure 15 shows benefits of the air injection on exhaust emissions. There is no specific difference in emission behaviour between with and without air injection. However, in the case of emission under lean limit condition, NOx emission decreased 790 ppm to 280 ppm with air injection. And CO2 also decreased due to ISFC improvement. Other emission (HC and CO) amount is quite similar between with and without air injection.

Figure 15: Benefits of air injection on exhaust emissions

6 Conclusion In this paper, effects of an in-cylinder turbulence control concept were clarified using compressed air injection system. High pressure air was injected into the combustion chamber to generate a strong and small scale turbulence. Parameters of the air injection such as injection timing and direction were optimized with 3-D CFD to enhance a flame kernel growth after ignition resulting a minimum initial combustion period. The benefits of the new concept were confirmed with a single cylinder research engine. The period of initial combustion shortened by 7 to 10° CA with n = 1500 1/min, IMEP = 6.8 bar. This effect corresponds to an improvement of lean limit by 0.15 point of excess air ratio and 1.4 % of ISFC with the same combustion stability. NOx emission decreased by 510 ppm (790 to 280 ppm) at the lean limit. Under the lean limit condition,

240

Combustion stability improvement with turbulence control by air injection for … the quantity of the injected air accounts for only 0.37 % of intake air into combustion chamber and considerable ISFC improvement of 1.1 % is still possible even taking account of compression work for the air injection. The promotion effect of flame kernel growth was also clarified with optical analysis. The stability of flame propagation was improved and any major negative effect was not observed. Moreover, it was clarified that generation of suitable turbulence field generation is important to promote an initial flame kernel growth, which leads to a better process of lean-burn.

Bibliography 1. IEA (International Energy Agency), World Energy Balances – 2016 edition – excerpt – Key World Energy Trends, http://www.iea.org/, 2016 2. Hermann B., Anton W., Tilo L., Guido P., Lean-burn Stratified Combustion at Gasoline Engines, MTZ Vol. 74, May, 2013 3. John B., Jim L., Kristapher M., Corona Ignition System for Highly Efficient Gasoline Engies, MTZ Vol. 74, June, 2013 4. Philip. G. Hill., Cyclic Variations and Turbulence Structure in Spark-ignition Engines, Combustion and Flame, Vol. 72, pp.73-89, 1988 5. Tennekes, H. , Simple Model for the Small‐Scale Structure of Turbulence, The Physics of Fluids, Vol.11, No.3, 1968 6. W. Wiese, S. Pischinger, P. Adomeit, J. Ewald, Prediction of Combustion Delay and – Duration of Homogeneous Charge Gasoline Engines based on In-Cylinder Flow Simulation, SAE Paper 2009-01-1796, 2009 7. Convergent Science, https://convergecfd.com/ 8. Hülser, T., Grünefeld, G., Brands, T., Günther, M., Pischinger S., Optical Investigation on the Origin of Pre-Ignition in a Highly Boosted SI Engine Using Bio-Fuels, SAE Technical Paper 2013-01-1636, 2013 9. Photron, https://photron.com/ 10. LaVision, https://www.lavision.de/de/

241

Air intake temperature cooling thanks to pressure wave action and adapted air intake geometry Vincent Raimbault, Jérôme Migaud MANN+HUMMEL France S.A.S Heinz Bühl MANN+HUMMEL GMBH Stéphane Guilain RENAULT GROUP David Chalet Ecole Centrale de Nantes LHEEA/TSM Michael Bargende FKFS/IVK, Universität Stuttgart

Air intake temperature cooling thanks to pressure wave action and adapted air …

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 M. Bargende et al. (Hrsg.), 19. Internationales Stuttgarter Symposium, Proceedings, https://doi.org/10.1007/978-3-658-25939-6_20

243

Air intake temperature cooling thanks to pressure wave action and adapted air …

Abstract As powertrain electrification gains popularity, gasoline engines must work to keep up with new performance demands and strict emission regulations. As such, a new gasoline engine must be engineered for high performance and high efficiency, as well as for acceptable CO2 levels and the European Union’s Real Driving Emission (RDE). To maintain this tradeoff, for new GTDI engines, lambda 1 upstream of the 3 way catalyst needs to be ensured. Thus, a new method of cooling the exhaust components needs to be found at full load without fuel enrichment. A solution was explored using a combined action between pressure waves occurring at inlet valves, and the air intake geometry. In the past, a focus on pressure wave’s action to increase the air amount inside cylinder was developed, and the air temperature was ignored. This research investigates the pressure wave action as a method to reduce the air temperature at inlet valve closing with associated gas expansion. Even if it could reduce the pressure accordingly to the gas laws, this can be compensated by controlling the waste gate. Among other concepts such as early intake valve closing, water injection, and low pressure EGR, this new approach does not require any complex control strategy. Design, system simulation, and engine tests have been conducted to prove the concept, quantify the benefit, and integrate it in a real engine compartment.

Introduction Increasing pressure to restrict engine fuel consumption has conducted toward the downsizing trend, pushing engine manufacturers to reduce engine displacement volumes. This shift has permitted engines to run the standard driving cycle with higher opening angle throttle valves, and thus better brake output specific fuel consumption. Nevertheless, engines have to extend their performance to higher torque and power. This has been possible with a wide use of boosting systems such as turbochargers. The output power is not only limited by the air loop, but also by the combustion process. At high boost pressure, the auto-ignition in the cylinder, known as knocking, hinders the combustion to occur at the optimum timing in the cycle. At the maximum power operating point, retarding the ignition is not always enough to control the knocking. One solution is to increase the fuel to air ratio above the stoichiometric conditions. In that case, part of the fuel does not directly participate in the combustion process, but helps to reduce the combustion temperature and thus exhaust temperature. This technique is useful to increase output torque while protecting the exhaust components and the piston, but it significantly increases the fuel consumption and hinders the three-way catalyst conversion for HydroCarbon (HC) and Carbon monoxide (CO). The next Euro 7 regulation and the application of the real drive emission estimation are likely to make this enrichment unacceptable anymore.

244

Air intake temperature cooling thanks to pressure wave action and adapted air … Keeping a stoichiometric mixture during all engine operating conditions would reduce pollutant emissions by optimizing the three-way catalyst conversion. Reducing enrichment leads also to fuel savings. The risk of knocking and exhaust temperature exceeding the exhaust line should be address without changing the air to fuel ratio. The water injection or the Millerization through early intake valve closing. The proposed concept uses pressure waves to cool down the intake air, thus reducing the knocking tendency and the high temperature. The effect of the wave action on temperature at the intake and later in the cylinder is modeled with simulation. Then, the results of in-cylinder pressure are evaluated. Through 1D simulation, a concept allowing the wave action to provide the air cooling at the intake will be presented. A prototype was built according to the simulation and tested on an engine test bench. Finally, constraints on the available space in the engine compartment were taken into account.

1 Acoustic wave effect on in cylinder temperature The acoustic waves have been used on naturally aspirated engines to improve their volumetric efficiency. Volumetric efficiency can be enhanced when the pressure wave reaches a maximum at the valve closing [1][2]. In such a case, the pressure in the cylinder can be increased, thus increasing the air density and output performance. In the last version of the turbocharged gasoline engine, the turbocharger is able to provide high boost pressure, mitigating the issue of volumetric efficiency. The limits lies on the knocking risk. The last generation of engines are built with high compression ratios. Thus, with high boost pressure, the in-cylinder pressure and temperature of these engines rises. In this condition the auto ignition can occur. The knock can be controlled through a late combustion, but the combustion occurs at high temperature and the reduced expansion in the cylinder leads to high temperature in the exhaust. The exhaust manifold and the turbine have to withstand high temperature. The current technologies of turbine limits the exhaust temperature to approximately 1250K. Finally, the turbocharger is controlled in such a way that this temperature is not exceeded. If the exhaust temperature is too high, the bypass duct of the turbine (waste gate) opens and the outlet pressure of the compressor reduces. Another way to reduce the knock risk and the combustion temperature is to reduce the intake air temperature. This temperature is bounded by the ambient temperature which cannot be controlled. On the other hand, it has been observed that pressure waves influence the temperature. The variation of the pressure leads to the variation of the instantaneous temperature. Then with the appropriate phasing of the wave, the in-cylinder pressure can be reduced as depicted by Theilemann [3].

245

Air intake temperature cooling thanks to pressure wave action and adapted air … The MANN+HUMMEL concept uses the expansion created by the movement of the piston during the intake stroke as presented in Figure 1. Indeed, the expansion created at the cylinder propagates into the intake system with the sound speed. If the correct length between cylinders is chosen, the expansion can arrive just as the valve closes, thus creating the expected temperature reduction. Figure 1 shows how an air flow shot at cylinder 1 creates the expansion which propagates until the position of the cylinder 2.

Figure 1: Pressure response to air flow pulse

The concept is based upon connecting the intake of the cylinders which are not directly succeeded in the firing order. For example, a 4 cylinder engine with a firing order of 1-3-4-2 indicates that the intake of cylinder of 1 and 4 are connected with the shortest duct possible without a change in cross section. The same logic is applied for the intake of the cylinder 2 and 3. Then, the new blocks are connected to each other with a duct, without change in cross section to prelude to acoustic attenuation during the wave propagation. Figure 2 gives a schematic overview of the concept. The so called expansion length is defined by the wave propagation distance from one cylinder to the following one in the firing order.

246

Air intake temperature cooling thanks to pressure wave action and adapted air …

Figure 2: Schematic of the concept

As the wave are propagated at the sound speed, it will be defined by the delay where the expansion created at one cylinder expands at the other location. This length will change with the engine speed as the time between two consecutive intake events change. The distance 1000mm is suitable for a 4 cylinders engine at 4500RPM. The pressure pulsation can then reach high amplitude as depicted in the Figure 3. The amplitude can reach 1.5 bar peak to peak, and the minimum occurs when the valve closes.

Figure 3: Intake pressure at 4500RPM with 1000mm expansion length

When compared to the base line, Figure 4 shows the increase of the pulsation amplitude at 4500RPM. The base line does not present much pulsation level compared to the expansion device.

247

Air intake temperature cooling thanks to pressure wave action and adapted air …

Figure 4: Base line and Expansion device intake pressure comparison at 4500 RPM

The expansion in the intake manifold at the intake valve closing results in a temperature decrease in the cylinder as presented in Figure 5. It shows the 15K reduction with the expansion device compared to the base line version.

Figure 5: Base line and Expansion device intake pressure comparison at 4500 RPM

After the temperature decrease was demonstrated, the benefit for the engine output was evaluated.

248

Air intake temperature cooling thanks to pressure wave action and adapted air …

2 Air intake cooling benefit on engine output The engine simulation was carried out with an up-to-date turbo charged direct injection 4 cylinder engine, as described in Table 1. Table 1: Engine characteristics [4] Aspiration Injection Fuel Intake Valve Exhaust Valve Displacement Max exhaust temperature Lambda Load Compression Ratio Bore Stroke

Turbo Charged Direct Gasoline Variable timing Variable timing 1.3 L 1250 K 1 Full Load - WOT 10.5 72 mm 81 mm

The 1D simulation was carried out with GT Power. The cylinders were modeled with UserCylinder from FKFS [5]. The knock evaluation was based on values of the prereaction state [6] Ik defined in the equation 1. The ignition delayed 𝜏 is calculated with equation 2 which is an Arrhenius equation where the figures Ci depends on the actual unburned zone composition (lambda, EGR, water, etc.). 𝑑𝛼

𝐼 𝜏

𝐶𝑝 𝑒

(1) (2)

Ik value defines the knock risk. The limit was set with the base line configuration. The spark timing was then controlled to ensure the limit was not exceeded. The injection ensured a stoichiometric mixture in cylinder. The boost pressure was controlled through the turbine bypass circuit flow control with the waste gate. The control was set to ensure the exhaust temperature could not exceed the 1250K. The gain was then directly observable with the output engine performance as the power presented in the Figure 6.

249

Air intake temperature cooling thanks to pressure wave action and adapted air …

Figure 6: Base line and Expansion device output power comparison at full load

The temperature reduction, as shown in Figure 4, leads to an increase in output power as depicted in Figure 6. This simulation shows an increase of 11 kW at 5000 RPM which is about a 13% increase in output power. The effect on the combustion timing and exhaust temperature require more analysis. Indeed, the reduction of the temperature leads to an earlier combustion which helps to provide more torque but also to increase the expansion in the cylinder. The expansion increase lowers the exhaust gas temperature. The lower temperature gives room to further close the waste gate and thus increase the energy recovered at the turbine. This energy is spent at the compressor side to rise the boost pressure. This increase in the boost pressure then leads high pressure in the cylinder which a change in the combustion timing. Finally the new equilibrium between all those effects is found and provides the power increase. The area around 3500RPM shows a drawback of the expansion device which does not allow the production of the same output power of the base line. This issue will be addressed with a deactivation device. Those simulations have demonstrated a significant potential, and the layout of the expansion device has been defined. The next step was to produce a prototype and to confirm the benefit of this intake air system geometry.

3 Validation on engine test bench A prototype using stainless steel duct and elbow allowing to change the length was produced and mounted on the engine equivalent to the simulated one, as shown in Figure 7.

250

Air intake temperature cooling thanks to pressure wave action and adapted air …

Figure 7: Prototype mounted on engine

This first prototype had a length of 1200 mm and can be placed at the same position as the base line. The base line air intake manifold is replaced by the prototype. The charge air cooler and all upstream parts remained the same. The intake temperature was regulated to keep the same charge air cooler outlet temperature for both configurations. The knocking intensity was evaluated with high frequency in-cylinder pressure signal analysis. Figure 8 represents the full load results with the expansion device and with the base line. The Bypass configuration resulted in a deactivation mode where the expansion length is bypass with another duct.

Figure 8: Base line and Expansion device output power comparison at full load

251

Air intake temperature cooling thanks to pressure wave action and adapted air … The measurement confirmed the possibility of the expansion to increase the power and the bypass configuration allows to almost completely retrieve the baseline configuration performance. The benefit is a bit lower than exhibited with the simulation with 5% increase. An investigation on the intake and exhaust valve timing has been launched to find out if further increase is possible. One reason for these differences is that the intake manifold temperature is different between simulation and measurement. For the measurement the cooler charge air outlet temperature is maintained constant whereas the simulations left the cooler efficiency constant. The difference of duct raw material between the base line and the prototype could also affect the heat transfer.

Figure 9: Intake and Exhaust valve timing variation at 4500RPM

With the same controlled parameters than previous full load evaluation, Figure 9 shows the valve timing variation which allows 4.5 % more power which gives a total 9.5% increase compared to the base line. This first part demonstrated the power increase potential of the expansion device. The simulations have permitted to layout the device. The expansion length of 1 to 1.2 m is still challenging to implement in a real engine compartment. In the last years the engine compartment volume was reduced. At the same time, the number of components have increased significantly making it challenging to find a route for the current turbo duct and thus so making it even more challenging to implement the expansion length.

252

Air intake temperature cooling thanks to pressure wave action and adapted air …

4 Integration in engine compartment Using the CAD model of an engine compartment a more compact concept has been proposed. It consists of a folded duct to reduce the apparent duct length. The plastic production process makes it possible. Figure 10 shows a possible design to integrate the expansion length in an engine compartment.

Figure 10: Design with folded duct for expansion length

Folding the duct as depicted in the figure 11 gives the opportunity for package saving volume. The depth of the part is higher than the base line but does not clash with any other component.

Figure 11: Folded structure with plastic process

The plastic process allows smooth elbows and should limit the increased pressure drop.

253

Air intake temperature cooling thanks to pressure wave action and adapted air …

5 Conclusions and Outlook The future regulation (Euro7) as well as the emission evaluation methodology (WLTC and RDE) have pushed the OEMs to find solution to maintain higher power output without enrichment. The expansion device is a solution to increase the output power while remaining at stoichiometric mixture. The measurement has shown approximately 10% increase in maximum output power. The deactivation methodology has demonstrated its ability to damp the drawback at lower engine speed. A plastic design of the device in engine compartment environment has been initiated. The possibility to use it in an actual engine compartment has been demonstrated. Further design activities will be launched and a version of a geometry taking into account the packaging volume and production process will be produced and tested. The possibility to use the expansion device in addition to the water injection or early intake valve closing will be investigated.

Bibliography [1]

M. F. Harrison and A. Dunkley, “The acoustics of racing engine intake systems,” J. Sound Vib., vol. 271, no. 3–5, pp. 959–984, Apr. 2004.

[2]

J. B. Heywood, Internal Combustion Engine Fundementals, vol. 21. 1988.

[3]

L. Theilemann, “INTERNAL COMBUSTION ENGINE,” EP 2 179 153 B1, 2007.

[4]

D. H. Schnüpke, D. T. Maass, and D. S. Zimmer, “Modern , Compact and Efficient : M 282 – The New 1 . 4-Liter Gasoline Engine from,” pp. 875–894, 2017.

[5]

M. T. Keskin, M. Bargende, and M. Grill, “Fast predictive burn rate model for Gasoline- HCCI,” W. Springer, Ed. 2016, pp. 155–169.

[6]

A. Fandakov, M. Grill, M. Bargende, and A. C. Kulzer, “Two-Stage Ignition Occurrence in the End Gas and Modeling Its Influence on Engine Knock,” SAE Int. J. Engines, vol. 10, no. 4, pp. 2017-24-0001, 2017.

254

Future e-mobility and the change in system requirements Dr. Lothar Schindele, Dr. David Schütz, Dr. Gaël Le Hen, Dr. Norbert Müller Robert Bosch GmbH

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 M. Bargende et al. (Hrsg.), 19. Internationales Stuttgarter Symposium, Proceedings, https://doi.org/10.1007/978-3-658-25939-6_21

255

Future e-mobility and the change in system requirements

1 Motivation Mobility of tomorrow is undergoing a radical change. Against the background of increasing electrification, automated driving, the connectivity of vehicles and ever stricter CO2 regulations, not only the powertrain technologies (e.g. BEV, FCEV, ...) but also its usage profiles (e.g. robo-taxi, car sharing, ...) and thus the requirements of the vehicle components will change considerably. The segment of fleet operated cars will gain more importance in the mobility system of the future, and fleet operators will define vehicle requirements more and more. Their increased focus on total cost of ownership (TCO) for a given and specific use case requires flexible and scalable vehicle architectures. In order to address diverging and specific use cases, new vehicle development and optimization tools have to be applied for optimized solutions on both, a vehicle architectural level, and on a component level. The importance of an optimal component dimensioning and design becomes obvious considering the battery sizing of an EV, where under-dimensioning may prohibit the usage for a certain application and where over-dimensioning prohibits an attractive cost situation. As an exemplary application-focused vehicle design, that entered the market in 2016, StreetScooter GmbH’s motivation was to design an EV platform specifically optimized for the purpose of parcel delivery [1]. Optimizing a vehicle for a specific use case like the aforementioned parcel delivery will typically lead to other vehicle architectures and component designs than optimizing a vehicle for an “average” usage. An efficiency measure that is effective during a worldwide “average” vehicle usage, as it is replicated by a WLTP test condition, might show less benefit in a specific use case like parcel delivery, or vice versa. Bosch has developed a holistic simulation tool chain, which allows to optimize an EV’s powertrain design and vehicle architecture for a specific vehicle usage. For this purpose, usage profiles (also called “use cases”) are considered in detail, which describe the usage of the respective vehicle over the entire year. All energy flows within the vehicle (e.g. mechanical, electrical and in particular thermal) are taken into account, not only during driving, but also during parking or charging phases, in order to determine annual energy consumption. This development tool chain allows to compare a wide variety of vehicle and powertrain topologies, to assess the effectiveness of efficiency measures for a specific use case, and it allows to derive component load profiles for all relevant components. This paper describes the methodology of the development tool chain and gives some exemplary assessments of vehicle thermal topologies and efficiency measures for a range of vehicle use cases.

256

Future e-mobility and the change in system requirements

2 EV use cases In the following chapter, the term “use case” is defined for the considered context, and exemplary use cases of today’s and future vehicle usages are given.

2.1 Definition of use cases In the context of system engineering, the term “use case” describes a list of event steps, defining the interactions between an actor (e.g. the driver or traveling person) and a system (e.g. the vehicle) to achieve a goal (e.g. drive from A to B). In the given context of mobility, a use case also describes a specific context of use (e.g. traffic and environmental conditions). For the presented study, the following use cases are considered: Table 1: Considered use cases and their characteristics Use case Short Distance Commuter

Transport missions Family care, shopping

European Privately Owned Vehicle

Family care, shopping, commuting, business trips All (but restricted to a metropolitan area)

Taxi (human driven)

Robo Taxi (fully autonomous)

All (but restricted to a metropolitan area)

Key parameters Mileage: 9.000 km/year Average speed: 20 km/h Starts: 7/day Mileage: 15.000 km/year Average speed: 46 km/h Starts: 4/day Mileage: 87.000 km/year Average speed: 33 km/h Starts: 7/day Mileage: 87.000 km/year Average speed: 39 km/h Starts: 7/day

In order to determine a vehicle’s energy consumption for such a use case, or in order to assess the vehicle components’ load profiles for such a use case, a detailed description of operation is required. This description does not only consist of vehicle driving conditions, but also of vehicle states like parking. Table 2 shows some of the relevant vehicle states. Accordingly, an analysis of energy flows in the vehicle needs to consider not only the energy consumption and energy flows during driving, but also during all other vehicle states, e.g. thermal losses during parking when the battery cools down, or thermal losses during charging. In addition to the drivetrain’s energy consumption, the energy for heating, ventilation and air-conditioning (HVAC) of the passenger compartment is of major relevance for

257

Future e-mobility and the change in system requirements the vehicle’s overall energy consumption and therefore for the TCO assessment of a BEV. The energy consumption for HVAC depends on both, climatic zones and seasons, which means that the vehicle location has to be considered. Table 2: Exemplary EV states States Driving

Parking w/ infrastructure Parking w/ infrastructure, Fast/DC charging Parking w/o infrastructure Maintenance & cleaning Parking w/ infrastructure; Preconditioning

Definition The vehicle is driving according to a standardized or specific driving cycle in order to fulfill a transport mission The vehicle is parked and connected to a charging station (e.g. 7 kW) and is able to charge the battery The vehicle is parked and connected to a DC charging station (e.g. 50 kW) and is able to fast charge the battery The vehicle is parked and not connected to a charging station The vehicle can be maintained and cleaned if required Infrastructure allows thermal preconditioning of cabin, battery and other components

These states are used to describe a daily vehicle usage profile. Figure 1 shows a description of typical usage days (for one week) for the use case Short Distance Commuter.

Figure 1: Vehicle usage profile of the use case Short Distance Commuter

258

Future e-mobility and the change in system requirements

2.2 Today’s versus future vehicle use cases A detailed description of today’s vehicle use cases is preferably based on measured vehicle and mobility data. However, a strategic component portfolio management has to consider future vehicle use cases as well, which might not be on the market until now, e.g. autonomous driving with shared vehicles. Therefore, in a first development step it is required to assess and describe in detail the usage of future vehicles or mobility concepts, i.e. of future use cases. Based on this, an appropriate development tool chain is to be used in a following step in order to derive new system and component requirements for such new use cases. Based on the use case “human driven Taxi”, a use case for a fully autonomous Robo Taxi has been derived in [2] by taking into account traffic simulation, lower acceleration rates and the interaction of the vehicle with other vehicles and the traffic infrastructure. This results in an increase of the average speed and a reduction of the peak battery discharge power. Furthermore, the energy consumption of the automated driving kit (e.g. 1kW) has to be taken into account. Both effects lead to a higher overall consumption and thus to a reduction of the vehicle range (see Table 3). Table 3: Differences between the use cases Taxi and Robo Taxi max. acceleration [m/s^2], [2] driving time [h/day], [2] idle time [h/day], [2] mean vehicle speed [km/h], [2] mean battery discharge power [kW] peak battery discharge power [kW]

Taxi to Robo Taxi - 67 % - 15 % + 11 % + 19 % + 22 % - 73 %

3 Methodology and Tools For a use case dependent technology assessment on a vehicle level, a 0/1D simulation model of an electric vehicle has been developed using the CAE system simulation software GT-SUITE [3]. It captures all relevant mechanical, electrical and thermal effects and consists of the following elements: – Dynamic models of the complete powertrain and HVAC system (esp. electrical and thermal models of all components like powertrain, battery, charger, etc.; fluid/air circuits, passenger compartment model, etc.).

259

Future e-mobility and the change in system requirements – Control strategies for different vehicle conditions such as driving, charging (with AC or DC), preconditioning and parking. – Vehicle environment model for the daily profiles, esp. driven route including topology, and climatic data (temperature, humidity, solar radiation …). – Pre and post processing tools (software: Matlab [4]). A special attention has to be put on an optimized simulation time which allows to calculate use cases for instance for a complete week within a reasonable computing time.

The workflow for a use case optimized vehicle design consists of several layers and optimization loops which are described in Figure 2. Initially, the use case is described according to section 2. This includes the definition of the vehicle class as well as the climatic data of the considered location, especially the temperature data, humidity data and solar radiation need to be taken into account. For the calculation of the average annual energy consumption, a 7 day calculation is carried out, one of it during summer conditions, one during winter conditions and one in the transitional month (weighted two times). A second optimization layer considers vehicle architecture, powertrain topology and thermal topology including operating strategies. A third optimization layer consists of the dimensioning of all relevant components, especially powertrain components including battery and onboard charger, as well as thermal system components. A post-processing tool in Matlab allows to analyze all the energy flows in the vehicle, to assess the optimization process and the operating strategy. After the optimization process it also allows to derive use case specific requirements for subsystems and components, e.g. load profiles. The optimization itself is done in an iterative way using a newly developed simulation and optimization framework.

260

Future e-mobility and the change in system requirements

Figure 2: System optimization workflow

4 Results The methodology and tool chain described in the previous chapter has been applied to study the use cases described in section 2.1, assuming climatic conditions of Shanghai/China for the Short Distance Communter (SDC) and the Taxi use case. The energy consumption at WLTP and NEDC conditions are at climatic standard conditions (23°C). Figure 3 shows the annual energy consumptions for the use cases Short Distance Commuter and Taxi in relation to this vehicle’s energy consumption during the WLTP test procedure. In addition, the relative energy consumption during the NEDC test procedure is given as a reference. The considered vehicle is a compact class EV, equipped with a 80 kWh lithium ion battery and a thermal system with heat pump and waste heat recovery as shown in Figure 4.

261

Future e-mobility and the change in system requirements

Figure 3: Annual energy consumption per km for use cases Short Distance Commuter (SDC) and Taxi (both for climate conditions Shanghai/China), as well as for the NEDC test procedure; values in [%] and in relation to the energy consumption during WLTP

The considered uses cases and ambient conditions result in an additional energy consumption per km of 69 % for the Short Distance Commuter and 15 % for the Taxi, in relation to the WTLP test procedure. The difference in fuel consumption between the Short Distance Commuter and the Taxi use case is mainly due to the different daily mileage and the number of starts per day. For the use case Short Distance Commuter, the vehicle must be conditioned several times per day, which is why energy consumption is considerably high during hot and cold months. For the use case Taxi on the other hand, the high daily mileage provides for sufficient powertrain waste heat, such that only little additional energy is required for vehicle conditioning during cold months. The influence of different thermal topologies on vehicle energy consumption is shown in Figure 4 for three different thermal topologies. The first topology ("Baseline topology") uses a PTC for cabin heating. The second topology "+ Waste Heat Utilization" allows to utilize the cooling system’s waste heat for cabin heating, along with a PTC. The third topology "+ Heat Pump" performs the cabin heating function using waste heat recovery from the cooling circuit in combination with a heat pump and a PTC. For these three thermal topologies, the energy consumption has been calculated for the vehicle described above for the use cases Short Distance Commuter, Privately Owned Vehicle and Robo Taxi (see Table 1), in this case at a constant ambient temperature of 9 °C. For the use case Robo Taxi, the additional energy consumption of the autonomous

262

Future e-mobility and the change in system requirements driving kit is not taken into account for better comparability. The WLTP consumption is set as reference. In the WLTP data there is no influence of the cabin HVAC system on the vehicle’s energy consumption, due to the definition of the test procedure.

Figure 4: Influence of thermal topologies on vehicle consumption at constant ambient temperature 9 °C for the use cases “short distance commuter” privately owned vehicle” and “robo taxi” (without AD-Kit)

Even though the results shown in Figure 4 reflect special conditions and depend for instance heavily on ambient temperature, the following conclusions can be drawn: – The WLTP has been designed such that it reflects worldwide average usage under worldwide average climate conditions, and accordingly are design targets of most vehicles today. However, in case of a fleet operator for instance, who focuses on a specific use case in a specific location and climate zone, a customized vehicle design offers significant efficiency and therefore TCO benefits. – Especially an EV’s thermal and HVAC system has a significant impact on vehicle efficiency, offering multiple degrees of freedom to optimize a vehicle for a specific use case and climate condition. – It is worthwhile to design an EV’s thermal system such that waste heat recovery, e.g. of powertrain components, is possible. – For use cases with a small annual mileage (e.g. use cases like Short Distance Commuter and Privately Owned Vehicle), a heat pump can be highly beneficial.

263

Future e-mobility and the change in system requirements – For use cases with a high annual mileage, (e.g. use cases like Taxi or Robo Taxi), an appropriate waste heat recovery system is typically much more attractive to support cabin heating than a heat pump.

5 Conclusions The definition, description and analysis of use cases is key to develop optimized powertrain and vehicle solutions, and to derive optimized component requirements of current and future eMobility solutions. The use cases described in this study consist of commercial applications (Taxi and Robo Taxi) and privately owned vehicles (Short Distance Commuter and Privately Owned Vehicle). Relevant vehicle conditions have to consider not only driving conditions, but also vehicle states like parking and charging. Special focus has to be on the use case location and climate, since heating, ventilation and air conditioning has a significant impact on energy consumption. Automated driving as referred to by the use case Robo Taxi has a significant impact on vehicle design, for instance only about one third of today’s powertrains peak power will be required any more. The thermal vehicle architecture offers a large degree of freedom for an application specific design optimization. For use cases with a low annual mileage, a heat pump can be highly beneficial. For use cases with a high annual mileage, an appropriate waste heat recovery system is typically even more attractive for efficient cabin heating. The presented methodology for vehicle optimization consists of three layers: – Description of use case, location, climate, – Optimization of vehicle architecture and operation strategies, – Dimensioning of all relevant subsystems and their components, e.g. powertrain, battery, thermal system. The optimization itself is done in an iterative way using a newly developed simulation and optimization framework. Future work of the developed methodology and optimization tool chain will take into account additional vehicle classes such as trucks and buses, as well as fuel cell vehicles.

264

Future e-mobility and the change in system requirements

Abbreviations BEV

Battery Electric Vehicle

EV

Electric Vehicle

FCEV

Fuel-Cell Electric Vehicle

HVAC Heating, Ventilation and Air Conditioning NEDC

New European Driving Cycle

PTC

Positive Temperature Coefficient

TCO

Total Cost of Ownership

WLTP

Worldwide harmonized Light vehicles Test Procedure

Bibliography 1. Achim Kampker, Jürgen Gerdes, Günther Schuh, „Think Big, Start Small. Streetscooter die e-mobile Erfolgsstory: Innovationsprozesse radikal effizienter.“ SpringerVerlag GmbH. 2. Jochen Schwarzer, Christian Thulfaut, et al., "Automated Vehicles – Powertrain Challenges and Concepts", 27th Aachen Colloquium Automobile and Engine Technology 2018. 3. Gamma Technologies, Inc., www.gtisoft.com 4. MathWorks, www.mathworks.com

265

Active materials for electrical motors – Leverage for reducing costs and increasing performance Moritz Kilper M.Sc., Dipl.-Ing. Hristian Naumoski, Dr.-Ing. Steffen Henzler Daimler AG

Active materials for electrical motors – Leverage for reducing costs and increasing …

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 M. Bargende et al. (Hrsg.), 19. Internationales Stuttgarter Symposium, Proceedings, https://doi.org/10.1007/978-3-658-25939-6_22

267

Active materials for electrical motors – Leverage for reducing costs and increasing …

1 Introduction 1.1 Facing new Challenges Driven by new social, environmental and economic challenges the automotive industry has to face those challenges in order to offer fascinating products for the costumer. The so-called CASE trends will lead to a fundamental change in the automotive industry.

Figure 1: CASE as trends in the automotive industry

One of the main aspects mentioned is the fact, that the view on the powertrain is transforming starting from a single solution world with an internal combustion engine (ICE) towards a world with diverse powertrain solutions [1]. In December 2018, the EU Parliament set new targets for the European fleet-wide average CO2 emissions of new passenger cars and vans. The average CO2 emissions have to be lower 15% in 2025 and 37,5% lower in 2030 compared to the limits in 2021 [2]. Also outside of Europe CO2 emissions play an important role in the development of the automotive sector. In China, mentioned as an example, some cities are already banning fossil fuel vehicles out of their town centers [3]. Besides measures for the ICE powertrain, like efficient high-tech engines and efficient transmissions, the electrification of the powertrain is a necessary solution.

1.2 Our way to emission free mobility “Daimler is still pushing ahead with an electric vehicle offensive.” Above 130 electrified alternatives will be offered to the costumer by 2022 [4].

268

Active materials for electrical motors – Leverage for reducing costs and increasing …

Figure 2: Electrification of the powertrain

The introduction of the 48 V system enables reduced fuel consumption, a more agile response of the powertrain and an increase in driving comfort. The integrated starter generator (ISG) takes over some hybrid function such as boosting during the acceleration or recuperation while the vehicle is slowing down. A further step to reach CO2 targets is the Hybrid offensive. One example for an outstanding performance is the Mercedes Benz S-Class 560 e Plug-In with an all-electric range of around 50 kilometers. One of the major advantages of plug-in hybrid vehicles is combining the benefits of the internal combustion engine with those of the electric motor. Daimler expects a battery electric vehicle (BEV) share between 15% and 25% until 2025 [4]. Therefore, more than 10 BEV vehicle types will launch. The new founded brand EQ underlines the importance of the electric mobility. What all those vehicles have in common is that at least they are all powered by electrical drive trains.

2 Electrical motor system In the next section a short description of the main components of a BEV will be given, to provide the basics for the following discussion on the materials of an electrical motor.

2.1 Electric has now a Mercedes The main role of the lithium ion battery is to provide the energy needed to operate the vehicle. However, it is not just about the capacity of the battery, but also the lifetime and operational safety. Lithium as cathode material is particularly suitable for a vehicle battery because it is has the most negative electrochemical potential in the electrochemical series. Due to this, the results are a high electrical capacity and high cell voltage. The intelligent onboard charger (OBC) makes it possible to charge the battery at any conventional household power socket.

269

Active materials for electrical motors – Leverage for reducing costs and increasing …

Figure 3: Driving components of the new introduced Mercedes Benz EQC

The OBC converts AC to DC and adjusts the DC voltage to the levels required by the battery. A quicker way of charging is the charging e.g. via CCS (Combined Charging Systems) in Europe and the US. In order to provide crashworthiness the battery is surrounded by deformable crash elements between the car body frame and battery pack. These elements absorb an additional energy during the impact.

2.2 Key to performance and efficiency – The electric drive module The electric drive unit consist of the converter, a cooling system, gearbox and the electrical motor. The task of the converter is to transform the direct voltage of the battery into a 3-phase alternating current, which is necessary to feed the electrical motor. The transistors of the converter generate a sinusoidal current using a pulse width modulation with short pules at a high frequency. The cooling system is essential to increase the continuous power and to avoid any demagnetization of the magnets caused by heat impact. To adjust the rotational speed and torque of the electric motor a gearbox is installed next to the output shaft. The operating principle of electrical machines is based on the interaction between magnetic field and the current flowing in the windings of machines [5]. Figure 4 gives an overview of different motor types and emphasizes the best choice of the motor type for a vehicle application. Considering the requirements of a BEV the main demand is a

270

Active materials for electrical motors – Leverage for reducing costs and increasing … high driving range and the reduction of battery size. Therefore, a high power density within in a small space is required.

Figure 4: Motor types considering costs, space and power density

The components of an electrical motor can be distinguished due to their contribution in the generation of torque [6]. The active parts of the motor are responsible for transferring electrical energy into mechanical rotational energy and vice versa. The non-active components provide the support of the motors torque. Bellow table 1 gives an overview of selected parts and their material assignment. Table 1: Overview motor parts and their material

Active part

Inactive Parts

Component

Material

Magnets Iron Core Windings Insulation Housing Bearing Shield Shaft

NdFeB FeSi Cu Dielectric material e.g. Aluminum e.g. Aluminum e.g. Medium carbon steel

3 Material Choice and Consequences for the Motor This paper focuses now on two main components of the electrical motor in more de-tail. In the following a detailed overview for the magnets and the iron core is given.

271

Active materials for electrical motors – Leverage for reducing costs and increasing …

3.1 Magnets

Figure 5: Hysteresis graph for the characterization of permanent magnets

For the characterization of permanent magnets is the hysteresis graph an appropriate measurement. The second-quadrant of a hysteresis loop is shown in figure 5. The following characteristics regarding the description of magnetic material are explained briefly and can be determined by the hysteresis graph: ● Coercive field strength 𝐻 : The capability of a material to resist an external magnetic field without getting demagnetized; in short coercivity. ● Remanence 𝐵 : The magnetization of a material after an external magnetic field is decreased to zero; for anisotropic and very hard magnetic materials this value is near the saturation polarization. The alloy of 𝑁𝑑𝐹𝑒𝐵 shows not only a high remanence and coercivity but also a highenergy product, which turns out to be an appropriate choice for permanent magnets of a synchronous motor. The extrinsic properties of a magnet are combined out of the material´s intrinsic magnetic properties as well as by the microstructure of the magnetic object. The intrinsic magnet properties, in turn, depend on the chemical composition. The chemical composition consists beside the main alloying element Neodymium of heavy rare earth elements (HRE) like Dysprosium and Terbium. These elements increase the coercivity caused by their higher anisotropy constants. However, these elements also decrease the saturation polarization of the magnet. The microstructure defines the extrinsic properties of the material, more precisely by the grain size and the inter-grain phase. The main phase 𝑁𝑑 𝐹𝑒 𝐵 is responsible for the magnetic properties of the magnet. Additionally an 𝑁𝑑-rich phase occurs at the grain boundary during the sinter process within the production of the magnet. The

272

Active materials for electrical motors – Leverage for reducing costs and increasing … emerged phase has paramagnetic properties and therefore interrupts the exchange interaction between the grains. The production process has a large effect on the described quantities. In general, the production process includes the following steps: Powder production  Compacting of the powder  Sintering  Post-treatment For instance, the powder production determines the grain size of the sintered magnets. Influence of Magnet Properties for the Motor Figure 6 demonstrates the influence of 𝐵 on the torque at a given revolution speed for a motor and the temperature dependency of the coercivity.

Figure 6: Temperature dependence of the magnetic properties (left hand side) and effect on the motors torque (right hand side)

A higher remanence of the magnet increases the torque. But on the other hand also the maximum revolution speed of the motor is limited by different values for the remanence. The cause of this effect is that a magnet with a high remanence induces a voltage in the windings of the motor, which is called the back-electromotive force. The backemf acts against the applied voltage of the inverter and therefore the current flowing into the motor decreases. One risk in permanent magnet machines is the demagnetization of the magnet during the operation of a motor. In general, temperature and an opposing field could cause this failure. The left chart of figure 6 shows the temperature dependency of the demagnetization curve. A higher temperature leads to a lower coercivity and vice versa.

273

Active materials for electrical motors – Leverage for reducing costs and increasing …

3.2 Iron Core

Figure 7: Typical hysteresis graph of electrical steel

The main task of the electrical steel is to conduct the magnetic flux. Therefore, only special materials fulfill the requirements of high permeability and low power losses. A typical alloy for electrical steel contains silicon and aluminum as well as a low amount of manganese. The silicon and aluminum decrease the coercivity and increases the electrical resistivity. As consequences, the dynamic eddy current losses as well as the static hysteresis losses are reduced. The characterization of electrical steel is done in the same, as the magnets. The hysteresis graph shown in figure 7 displays the emerging magnetic flux due to the magnetic field strength. These losses are discussed later in more detail. A way of increasing the permeability of the steel sheet is the grain orientation due to the rolling process. However, since the magnetic flux in a motor is permanently changing, this option is not practical for this use case and consequently the steel sheet is nonoriented. Next to the magnetic performance, the mechanical properties like the yield strength are important aspects for the decision of the right material. This is due to high centrifugal forces, which affect the rotor. Looking closer to the motor performance, low losses of the steel sheet are important for overall efficiency of the motor. In order to describe iron losses several different models exist [7] and [8]. The model of Bertotti [9] is one example: 𝑊 𝐵, 𝑓

𝑝

𝑝

𝑝

The total losses are combined of the individual loss parts.

274

(1)

Active materials for electrical motors – Leverage for reducing costs and increasing … Iron losses can be divided into: ● Hysteresis Losses 𝑝

:

The magnetic field is constantly changing in the operation of the motor. Due to the magnetic hysteresis the magnetization is constantly changing, which requires energy. This energy is proportional to the area of the hysteresis curve. ● Eddy Current Losses 𝑝

:

A changing magnetic field induces a voltage due to the electromagnetic induction. This voltage leads to loops of current flow called eddy currents. These losses are proportional to the electrical resistivity. ● Excess Losses 𝑝

:

The losses occur in magnetic materials because of induced eddy currents by moving domain walls. The moving domain walls are again a consequence of the changing magnetic field. Influence of Electrical Steel Properties for the Motor

Figure 8: Influence of different alloys on the permeability and motor efficiency

The left part of figure 8 shows that the higher the alloying content the lower the losses but also the lower the polarization of the electrical steel. The higher losses correspond as already mentioned to a higher alloying conent. A compromise between reduced losses and a high permeability has therefore to be found. The right part shows the effect on the motor efficiency. Only the alloying content is changed for this comparison, the topology of the motor and the operation conditions are the same. The overall motor efficiency is lower for a motor with higher losses through the complete revolution speed range.

275

Active materials for electrical motors – Leverage for reducing costs and increasing …

4 Challenges in the Material Choice Strategic as well as technological challenges occur in the choice for the right active material. The strategic challenges dominate for the magnets whereas technological issues do for the electrical steel. Both aspects will be reflected in the following. As mentioned before HRE are needed to increase the resistivity against demagnetization at higher temperature. The demand estimated for magnets is about 18.500 metric tons for the year 2020 and even higher with 65.500 metric tons for 2025 [10]. Considering the limited production capacity of the magnet manufactures a latent risk for the automotive sector is possible. At this point, the magnet manufactures are in charge to increase their production volume. While looking at the reserves of HRE and the annual production another issue points out. From 2013 through 2017, China was responsible for 98% of the annual production of Dysprosium [11]. It is clear that this leads to a strong dependency on only one source. Besides China, there are only Vietnam, Brazil and China that are close to have the same amount of HRE reserves [12]. Strongly changing prizes for HRE also affect the global magnet market negative. In order to summarize the listed points, one can say, that an approach to solve the challenges for the usage of magnets in motors is the reduction of HRE. Switching to the electrical steel and looking at Bertotti’s equation, a way to reduce the losses in the iron core becomes obvious. Additionally to the alloying content the iron losses are strongly affected by the thickness of the electrical steel. Consequently, one way to reduce the losses is to use electrical steel with small thickness. The positive effect of reduced losses comes along mainly with technological challenges in the manufacturing of the motor components [13]. The influence of stresses caused by the manufacturing process of the steel is much higher for thinner sheets. In general, a thin steel sheet increases the effort for the manufacturer due to higher demands for the cutting tools. Added together this factors result in higher costs for the electrical steel.

5 Possibilities to face the Challenges One way of reducing HRE is to reduce the magnet mass itself. Figure 9 shows some measures done by Daimler to reduce the magnet mass in electrical motors for hybrid cars. If the weight of the rotor is reduced, the less magnet mass is required caused by a lower inertia. This can be done by design the motor in such way that in the areas with less stress, the material can be removed. Another way is to increase the cooling capability of the motor which results in an overall better motor performance and the option to reduce magnet mass for the same amount of motor power

276

Active materials for electrical motors – Leverage for reducing costs and increasing … The before listed measures are done by Daimler, the following ways of reducing HRE are research topics where Daimler acts as a research partner or is member at the advisory board e.g. [14].

Figure 9: Ways to reduce magnet mass for a hybrid motor

A further possibility is the substitution of HRE with other magnetic materials. In a first step, material systems have to be found, which have suitable intrinsic magnetic properties. Machine Learning is one tool for this task [15]. However, as mentioned before only the intrinsic properties are not enough, therefore in a next step the materials have to be synthesized to a stable phase. After this, the validation of the phase and the determination of the extrinsic magnetic properties for the new found material system is up to follow. Another method of increasing coercivity field strength is the precise application of the so-called grain boundary diffusion, where Dysprosium or Terbium is only used at the grain boundary [16]. Further reduction of the grain size is done by new technologies like the grain size tuning through spark plasma sintering and rapidly solidifying of metallic melts [17].

277

Active materials for electrical motors – Leverage for reducing costs and increasing …

Figure 10: Comparison between different steel thicknesses

Figure 10 shows different aspects a chosen sheet thickness. A thin sheet has low losses but on the other hand high costs and vice versa for a thicker sheet. The costs and the manufacturing effort are related to each other. More manufacturing steps e.g. the rolling process are necessary for a thin steel sheet and as a results the costs are increasing. As mentioned before the cooling of an electrical motor is an important issue, the thermal conductivity of the complete steel sheet is higher for a thicker and lower for thinner steel sheet. To decide which aspects are important for the choice of the electrical steel’s thickness the individual application is important. For instance, the Mercedes - AMG Petronas Formula 1 team has other requirements then the development team of the EQ-C has.

6 Conclusion Starting with the challenges the automotive sector has to face, the paper demonstrates the importance of the right material choice for the electrical motor. In a first step, the reasons why a permanent synchronous motor has advantages are demonstrated by listing different motor types. The requirements of a vehicle electrical motor are in best agreement with permanent synchronous motor. Two components are picked out and discussed in more detail. The properties that describe and determine the behavior of magnets are discussed. The influence of the remanence on the motor’s torque and the consequences for maximum revolution speed are explained. It is presented in detail that the need for HRE is the main challenge in the usage of magnets in motors. Possibilities to face those challenges like reduction of the magnet mass through advanced motor design or the substitution of HRE with new material systems are demonstrated.

278

Active materials for electrical motors – Leverage for reducing costs and increasing … Discussing the properties of electrical steel sheets for electrical motors, it points out that the thickness is the best leverage to reduce the iron losses for electrical steel during the operation of an electrical motor. However, the reduced thickness is also increasing the costs and the effort in the manufacturing of iron core components like stator and rotor. The choice of the thickness therefore depends strongly on the requirements of the individual application.

Bibliography 1. Lüdiger Thomas: “Electrification of the Powertrain – Impact on the Machinery Industry and Component Suppliers” 10. E-Motive Expertenforum für elektrische Antriebe 2018 2. Reuters: „EU legt strengere CO2-Grenzwerte für Neuwagen fest“. Web. 20.12.2018. https://de.reuters.com/article/eu-autos-co2-grenzwerte-idDEKBN1OH0G5 3. Volk Frank: „Peking verbannt CO2-Schleudern von den Straßen“. Web. 10.12.2018. https://www.automobil-produktion.de/maerkte/peking-verbannt-co2-schleudernvon-den-strassen-105.html 4. Daimler: “Corporate Presentation Q3 2018”. Web 10.01.2019. https://www.daimler.com/dokumente/investoren/praesentationen/daimler-ir-corporatepresentationq3-2018.pdf 5. Pyrhonen, Juha, Tapani Jokinen, and Valeria Hrabovcova. “Design of rotating electrical machines.” John Wiley & Sons, 2013. 6. Müller, Germar, and Bernd Ponick. „Grundlagen elektrischer Maschinen.“ John Wiley & Sons, 2012. 7. Steinmetz, Chas P. “On the law of hysteresis. ” Proceedings of the IEEE 72.2 (1984): 197-221. 8. Jordan, Heinz. „Die ferromagnetischen Konstanten für schwache Wechselfelder.“ Elektr. Nach. Techn 1 (1924): 8. 9. Bertotti, Giorgio. “General properties of power losses in soft ferromagnetic materials.” IEEE Transactions on magnetics 24.1 (1988): 621-630. 10. Statista: “Demand for rare earths in automotive permanent magnets worldwide from 2010 to 2025”. Web. 10.12.2018 https://www.statista.com/statistics/693357/automotive-permanent-magnet-rare-earth-demand-worldwide/ 11. Adams Intelligence Research: “Spotlight on Dysprosium” 2018

279

Active materials for electrical motors – Leverage for reducing costs and increasing … 12. Statista: “Rare earth reserves worldwide as of 2017, by country”. Web. 10.12.2018. https://www.statista.com/statistics/277268/rare-earth-reserves-by-country/ 13. Tietz Marco, Dorner D., Teger K. and Basteck A. „Nichtkornorientiertes (NO-) Elektroband zur Herstellung von elektrischen Antrieben bei Kraftfahrzeugen“, 17. Aachener Kolloquium Fahrzeug- und Motorentechnik, 2008 14. Fraunhofer: „Das Fraunhofer-Leitprojekt Kritikalität Seltener Erden stellt sich vor.“ Web. 10.12.2018. https://www.seltene-erden.fraunhofer.de/ 15. J. J. Möller, W. Körner, G. Krugel, D. F. Urban, C. Elsässer. “Optimizing hardmagnetic phases with machine learning.” Acta Materialia 153, S. 53-61. 10.1016/j.actamat.2018.03.051. 16. Sepehri-Amin, H., T. Ohkubo, and K. Hono. “The mechanism of coercivity enhancement by the grain boundary diffusion process of Nd–Fe–B sintered magnets.” Acta Materialia 61.6 (2013): 1982-1990. 17. Sawatzki, Simon, et al. “Electrical and magnetic properties of hot-deformed Nd-Fe-B magnets with different DyF3 additions.” Journal of Applied Physics 114.13 (2013): 133902.

280

Traction energy saving potentials for electric cars with gear shift Oliver Zirn, Fabian Schmiel, Hochschule Esslingen Matthias Dellermann, Daimler AG

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 M. Bargende et al. (Hrsg.), 19. Internationales Stuttgarter Symposium, Proceedings, https://doi.org/10.1007/978-3-658-25939-6_23

281

Traction energy saving potentials for electric cars with gear shift

1 Introduction The efficiency of electrical traction drive chains is quite good (i.e. above 80%) in a wide operation range, while for low speed or small torque operation, efficiency falls significantly. In realistic load cycles, electric drives often have to work with efficiency levels below 60%. This is elucidated in Fig. 1 for the experimental vehicle shown in Fig. 2. In this contribution, the motor characteristics with efficiency mapping are given for the complete drive chain (battery – current converter – motor – gearboxes – wheel). ASM E-Boxter 44kW - efficiency mapping 150

100

0.5 0.7

0.5 0.7 0.75

0.7 0.75

9 0.7

200

300

9 0.7 79 0.

75 0.

400

Angular velocity in rad/s

75 0.

500

600

79 0.

-100

0. 75

0. 75

-150

0.5 0.7

0.5 0.7 0.75

-50

0.75

0.5

0.5

9 0.7

79 0.

-100

0.7

0.7

0

0.7 5

0.75

0.75

0.5

100

5

Strategy IV (automatic)

79 0.

0.7

0.5

0

0.7

50

0.75 0.7

-50

79

50

0.

0.7 5

79 0.

100

0.57

57 0.

150

0

ASM E-Boxter 44kW - efficiency mapping

-150

0

100

200

300

75 0.

400

500

600

Angular velocity in rad/s

Figure 1: Operation points for manual and automatic gear shift strategies (experimental car 1, WLTP)

With the actual electrical motor types asynchronous machine (ASM) and permanent magnet synchronous machine (PMSM) it is not possible to reach both, high velocity as well as high gradability together with high efficiency at low speed and small motor torque.

2 Experimental electrical vehicles with gearshift For the study presented in this contribution two converted experimental electrical vehicles with gear shift were available. Fig. 2 shows the vehicle data and characteristics for an electrified roadster with adequate power and manual six gear change. Fig. 6 shows a compact city car with moderate motorization and manual five gear change.

282

F in N

Traction energy saving potentials for electric cars with gear shift

Figure 2: Experimental car 1 (Porsche Boxter) with adequate electrical motorization (ASM 44kW, 24 kWh LiFePO-Battery, 150km range, 150km/h top speed)

In city traffic the roadster could operate exclusively in the 3rd speed. Maximum velocity and (moderate) acceleration are completely satisfying. Although, the ordinary driver will shift into the 2nd speed to start or rank in ascending slopes in order to feel enough reserve. Above 80 km/h the common driver will change into the 4th speed to reach a smaller engine angular velocity. This gear shift strategy derived from the typical combustion engine driven car routine (strategy I in section 3) results in the load cycle operation points shown in Fig. 1, that that are almost not situated in the optimum efficiency regions of the drive chain. For the experimental cars with gear shift we implemented the longitudinal dynamic vehicle model [2] shown in Fig. 3 and validated the models with real load cycles in the Stuttgart metropolitan area (SMA). These models yield a reliable simulation of mechanical, electrical and thermal vehicle states as the exemplary comparison with the measured energy consumption (SOC – State of Charge) and motor temperature shows in Fig. 4. Beneath realistic on-board consumption (e.g. electric heating) the variable gear ratio with variable gear efficiency is considered for gear shift strategy simulation. Although the load cycle in Fig. 4 is quite realistic for the challenging montainous Stuttgart region, we also investigated standardized load cycles – like the worldwide harmonized light vehicles test procedure (WLTP) – particulary with regard to the suitable gear-shift strategy. The exemplary operating point distribution in the efficiency mapping in Fig. 1 are shown for the WLTP. Although most of the operating points lay

283

Traction energy saving potentials for electric cars with gear shift outside the optimum efficiency region, mainly the operating points with high traction power influence the traction energy demand of the vehicle.

Figure 3: Vehicle longitudinal dynamic model implemented in MATLAB/Simulink

3 Gearshift strategy and power rated efficiency To compare different load cycles and gear shift strategies, the pure battery energy consumption is less meaningful than the average efficiency over a complete cycle. As the contribution of operating points with small brake- or drive power to the complete traction energy consumption is quite small, the power rated average efficiency cyc turned out to be a suitable reference number:

ηcyc =   η(i)  P(i)  /  P(i) i i with  - drive chain efficiency of the operating point in the load cycle

(1)

P - absolute value of the drive- or brake power at the motor shaft Table 1 compares the energy consumption for the SMA load cycle in Fig. 4 for different gear shift strategies. The high-motor-speed strategy II (augmented usage of the 2nd and 1st speed) compared to the low-motor-speed strategy II (preferred usage of the 4th speed) elucidate the reliable representation of the differences in traction energy consumption by the power rated average efficiency cyc.

284

Traction energy saving potentials for electric cars with gear shift E-Boxter Stuttgart metropolitan area load cycle, 23,8 km

100

Velocity Motortemp. Measurement

50

B10

0 0 Start 7:03 450

5

Botnanger Sattel

400

10 Frauenkopf 7:33 Schloßplatz

15

20

HE Geb.13 8:01

Rohracker Frauenkopf

350 300

Hedelfingen

ES-Pliens

250 0

5

10

15

40

20

Recuperation

20 0 -20 -40 -60

Driving power 0

5

10

15

20

20 Simulation Measurement

19 18 17 16 15

0

5

10

15

20

x in km

Figure 4: Real load cycle in the Stuttgart metropolitan area (SMA) with measured traction energy and motor temperature

285

Traction energy saving potentials for electric cars with gear shift Table 1: Gear shift strategy , energy consumption and power rated average efficiency for the SMA load cycle Strategy I (2nd – 4th gear) II (esp. 2nd gear) III (only 3rd/4th gear) IV (1st-5th gear opt.) V (2nd-4th gear opt.)

Traction energy in kWh/100km 19,32 19,22 19,49 17,44 17,98

cyc 0,707 0,710 0,691 0,764 0,748

Manual gear-shift strategies result in quite small energy saving potential due to the shifting time and the shifting laziness of the common driver. Here the additional costs for a gear shift cannot be compensated e.g. by battery size reduction. Although a fast shifting automatic gearbox, that keeps the operating point in the high-efficiency region (strategy IV in Table 1), yields a significant energy saving potential of about 10%. If the gearbox is reduced from five to three speeds (strategy IV in Table 1), the traction energy consumption saving potential remains significant. This elucidates, that an automatic three speed gearbox that is able to shift within less than a second (e.g. with a double clutch) enhances both, vehicle dynamics and efficiency. Fig. 5 shows the dynamic requirements for such an effective gear-shift strategy. While manual gear shift speed changes remain for dozens of seconds to minutes, the energy effective strategies require quite higher speed change frequencies:

Figure 5: Exemplary shifting sequence for strategy V and WLTP (automatic with double clutch)

For the implementation of a gear shift strategy, the jittering (first 10 seconds in Fig.5) or the very short acceleration caused changes into the 1st speed should be suppressed. The energy saving effect for the WLTP shown in Table 2 is visible too, but quite smaller, as only the kinetic recuperation energy of the vehicle can be used efficiency optimization. Here, the slow manual gear shift strategies I-III have negligible influence on the traction energy consumption.

286

Traction energy saving potentials for electric cars with gear shift Table 2: Gear-shift strategy , energy consumption and power rated average efficiency for the WLTP Strategy I-III IV (1st-5th gear opt.) V (2nd-4th gear opt.)

Traction energy in kWh/100km 20,38 19,49 19,78

cyc 0,727 0,764 0,752

4 City vehicle with 48V traction battery High-voltage traction drives reach any desirable vehicle performance, but make electric cars more expensive and less maintenance friendly. Low priced repair at the auto shop down the street is impossible for the lack of an electrical expert. A smaller city car with a safety-low voltage traction drive (< 60 V) would be an attractive and economic key element for emission-free metropolitan individual traffic. Actual light vehicles are not safe enough for the high traffic density [2] and often that ugly that they had been no alternative up to now. Vehicle characteristic diagram E-Golf IV 22kW 5-speed

2500

1st speed

10%

2000

2nd speed

1500

5%

3rd speed

1000

4th speed

0%

500

0

0

20

40

60

80

100

v in km/h

Figure 6: Experimental car 2 (VW Golf IV) with moderate electrical motorization (ASM 22kW, 16 kWh LiFePO-Battery, 100km range, 100km/h top speed)

A much better acceptance can be expected for city cars with 1-1.5 tons, enough space (4 seats), modern vehicle security features (airbag, crash zone, brake assistant) and 48V traction voltage. The actual realizable and affordable current converters reach rated currents up to 500 A and thus traction power limited to 20-25 kW. Without gear

287

Traction energy saving potentials for electric cars with gear shift shift, this motorization will result in a slowcoach with very limited maximum velocity or gradability that is as unattractive as the former ugly light vehicles. Also tourist or sharing vehicles would not be accepted with such limitations. Table 3: Gear-shift strategies , energy consumption and power rated average efficiency for the experimental car 2 (Fig. 6), SMA load cycle cyc

0.85

0.8 5 0.8

0.8

0.7

0. 8

0. 8

6

6

6 0.8 5 0.8

0.8

0.7

0.8 5 0.8

0.8

0.7

0. 8

6 0.8

0.85

0.8

0,822 0,852 0,851

6 0.8

6 0.8

0.85

0.8

I-III IV (1st-5th gear opt.) V (2nd-4th gear opt.)

Traction energy in kWh/100km 17,40 16,77 16,79

0.8

Strategy

Figure 7: Operation points for manual and automatic gear shift strategies (experimental car 2, SMA-cycle)

Supplying a 48V city car with a fast shifting two or three speed gearbox results in an acceptable maximum speed for metropolitan areas and a suitable gradability resp. manoeuvrability. The experimental car 2 that is shown in Fig. 6 reaches such an attractive combination with adequate (not overwhelming) vehicle dynamics superior to the heavy lead-acid-battery powered CityStromer-prototypes based on the VW Golf III in the mid 90’s. A fast shifting double clutchgearbox includes the possibility to optimize the traction chain efficiency as Table 3 shows for the SMA load cycle. The energy saving for the WLTP is even smaller.

288

Traction energy saving potentials for electric cars with gear shift

5 Summary and Outlook Fast shifting gearboxes with two or three speeds increase the power rated average efficiency of electric vehicles in moutainous regions up to 10%. For the WLTP load cycle, the benefit is smaller. For limited traction power due to safety low voltage (e.g. 48V) or high vehicle mass (commercial vehicles) a 2-3 speed gearbox is an important precondition to achieve gradability and maneuvrability in combination with adequate maximum velocity. Also for these city- and commercial vehicles the suitable gear shift strategy can save traction energy and thus battery size and costs. The optimization for given motor and vehicle parameters as well as required gradability and maximum velocity including the suitable gear ratio and punishing terms for jitter prevention is part of the ongoing research.

Acknowledgement The research results presented in this contribution had been worked out by the endowed professorship “Electrified Commercial Vehicles”, that is supported since 2015 by the Stifterverband für die Deutsche Wissenschaft e.V., Essen, as well as the Advanced Engineering Truck department of the Daimler AG in Untertürkheim.

Bibliography 1. Zirn, O.: Elektrifizierung in der Fahrzeugtechnik - Grundlagen und Anwendungen, Hanser-Verlag, Leipzig, 2017, ISBN 978-3-446-45094-3 2. Frei, P. et. al.: Vehicle Structural Crashworthiness with respect to Compatibility in Collisions. Working Group on Accident Mechanics at the Universities of Zürich, Switzerland, October 1999. 3. Döring, B. et. Al.: Gearbox Design via Mixed-Integer Programming. Proceedings of the VII European Congress on Computational Methods in Applied Sciences and Engineering, Crete Island, Greece, 5.–10. June 2016.

289

Impact of modulation strategies on the power losses of the dc-link capacitor of an inverter Maximilian Weber, Michèle Hirsch, Helge Sprenger, Thomas Zeltwanger Robert Bosch GmbH, Schwieberdingen, Germany Hans-Christian Reuss Institute for Internal Combustion Engines and Automotive Engineering, University of Stuttgart, Germany

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 M. Bargende et al. (Hrsg.), 19. Internationales Stuttgarter Symposium, Proceedings, https://doi.org/10.1007/978-3-658-25939-6_24

291

Impact of modulation strategies on the power losses of the dc-link capacitor of an inverter

Abstract Power losses within the powertrain of electric vehicles are crucial for the efficiency as well as the thermal load of its main components. Due to the limited robustness with regard to thermal degradation, the dc-link capacitor of a traction inverter requires good knowledge of its power dissipation. The harmonics occurring in the dc-link current are of vital importance considering the thermal stress on components of the dc-link circuit. Using different modulation strategies, various quantities of the powertrain can be positively influenced and are accompanied by a change of the harmonic characteristics. The effect of the voltage modulation needs to be considered appropriately in a loss model. Therefore, in this paper a new generic approach for the calculation of the current and the related power loss of the dc-link capacitor is presented. The validity of the algorithm was verified by simulation applying two different types of voltage modulation: space vector pulse width modulation (SVPWM) and six-step mode (SSM). The analyses indicate a high accuracy of the loss model with a maximum error of less than 2 %. Using SSM a reduction of the power dissipation of the capacitor of nearly 97 % against the standard modulation strategy SVPWM could be proven. As a result, the outlined methodology may be a basis for real-time computation of the power losses and a thermal protection function of a dc-link capacitor using a broad spectrum of modulation strategies.

1 Introduction Research on electric vehicles (EV) has recently intensified due to global warming, constraints on energy resources and more strict regulations on emissions [1, 2]. In EVs a common powertrain topology, as shown in figure 1, consists of a high-voltage battery pack, a two-level inverter and a three-phase electric machine. As one key component within the electric drive, the inverter is considered e. g. with aspects of compact design [3], low cost [4], high reliability [5] and long lifetime [6]. Apart from the actual power module, the dc-link capacitor is an important part of the inverter and contributes to cost, size and efficiency in a considerable scale [7]. Within automotive applications, the capacitor essentially provides for the compensation of power difference between source and load, the minimization of voltage variation and current harmonics in the dc-link circuit, as well as the supply of transient power peaks [8, 9]. Due to increasing requirements regarding reliability, lifetime, safety and performance the dc-link capacitor has to deal with these challenges. Generally, capacitors are not the elements with high sensitivity in respect to short power overload conditions, but they often limit the continuous performance of the power electronics due to their large thermal time constants. Because of high performance requirements, the dc-link capacitor is often stressed by thermal heating. Regarding modified

292

Impact of modulation strategies on the power losses of the dc-link capacitor of an inverter modulation strategies, the electrical and thermal load of the capacitor is affected. Besides the classic SVPWM with its adaptions, like injection of third harmonics [10, 11], other types of voltage modulation are applied to a higher extent to use various system advantages in different operating points, e. g. SSM at high rotational speeds [12, 13]. The power losses inside the capacitor and the resulting corresponding component temperatures are of high importance for the reasons mentioned above. For real-time applications, the analytical approach presented in [8] is a common way to calculate the current stress and power dissipation of a dc-link capacitor for SVPWM with a static equivalent series resistance (𝐸𝑆𝑅). The current flowing through the 𝐸𝑆𝑅 of the capacitor causes power dissipation, where the 𝐸𝑆𝑅 is strongly frequency-dependent. The spectral content of the capacitor current is closely related to the applied modulation strategy. In this paper, a methodology for real-time capable, frequency-dependent determination of the current and power loss of the dc-link capacitor without limitation on a certain modulation strategy is presented. In addition, the power losses resulting from SVPWM and SSM are compared.

Figure 1: Equivalent circuit diagram of the electric drive consisting high-voltage battery, two-level inverter and three-phase electric machine.

2 Basic Considerations As a first step the analytical calculation of the capacitor current and power loss within the dc-link capacitor, a brief review of some fundamentals of the inverter is presented.

2.1 B6 Bridge (Power Module) The B6 bridge is the main component of a three-phase inverter that includes the power semiconductors. As shown in figure 2, one phase leg consists of two transistors (T) and two corresponding freewheeling diodes (D). One pair assigned to the high side (HS) and one to the low side (LS) according to their voltage potential.

293

Impact of modulation strategies on the power losses of the dc-link capacitor of an inverter

Figure 2: Equivalent circuit diagram of the two-level three-phase inverter including single half bridges respectively with one pair of transistors (T) and diodes (D) on high side and low side.

The switching pattern is chosen to provide a three-phase voltage at the inverter output towards the machine. The typical operational mode deals with one switch on and one switch off in a single phase leg (freewheeling or active short circuit are not considered). A common way to describe the switching state of every phase leg uses the switching function 𝑠 [10] given by 𝑠

1, if HS-switch is closed and LS-switch is open 1, if LS-switch is closed and HS-switch is open

𝑥 ∈ 𝑈, 𝑉, 𝑊 .

(1)

1, the terminal point 𝑈 is connected to the positive supply rail For instance, if 𝑠 1, the terand current flows through the upper leg of the inverter. Contrary, if 𝑠 minal point 𝑈 is connected to the negative supply rail and current flows through the lower leg. Therefore, the inverter output voltage 𝑢 (with respect to the neutral point) is equivalent to the switching state multiplied with half the dc-link voltage ½ 𝑢 . The phase voltage 𝑢 (over one phase winding) can take five different voltage levels. For 8 different switching states and voltage the full B6 bridge a total number of 𝑧 2 vectors 𝑢⃗ [14] result as listed in table 1. The active pointers lead to vectors (𝑢⃗ … 𝑢⃗ ) with a length of ⅔ 𝑢 , while both zero pointers (𝑢⃗ , 𝑢⃗ have zero length. With each phase leg contributing to the total inverter current, the dc-link current 𝑖 is expressed by the switching functions and the phase currents 𝑖 , 𝑖 , 𝑖 of each phase leg 𝑖

∙ 𝑠 𝑖

𝑠 𝑖

𝑠 𝑖

.

(2)

For a star-connected machine, the sum of the three load currents equals zero. Therefore, no inverter current is formed during zero pointer and during active pointer the dc-link current is determined by that phase which provides a unique connection to either the positive or negative rail.

294

Impact of modulation strategies on the power losses of the dc-link capacitor of an inverter Table 1: Switching states and inverter/load voltages (normalized to dc-link voltage) of the two-level three-phase inverter. 𝒛

𝒔𝑼

𝒔𝑽

𝒔𝑾

𝒖𝑼𝟎

𝒖𝑽𝟎

𝒖𝑾𝟎

𝒖𝑼

𝒖𝑽

𝒖𝑾

𝒖⃗

0 1 2 3 4 5 6 7

1 1 1 1 1 1 1 1

1 1 1 1 1 1 1 1

1 1 1 1 1 1 1 1

½ ½ ½ ½ ½ ½ ½ ½

½ ½ ½ ½ ½ ½ ½ ½

½ ½ ½ ½ ½ ½ ½ ½

0 ⅔ ⅓ ⅓ ⅔ ⅓ ⅓ 0

0 ⅓ ⅓ ⅔ ⅓ ⅓ ⅔ 0

0 ⅓ ⅔ ⅓ ⅓ ⅔ ⅓ 0

0 1 2 3 4 5 6 7

2.2 DC-Link Capacitor Besides the B6 bridge the dc-link capacitor is the other essential component in the inverter, see figure 1. The equivalent circuit diagram of a real capacitor consisting of an equivalent series resistance (ESR), a capacitance (C) and an equivalent series inductance (ESL) is given in figure 3.

Figure 3: Equivalent circuit diagram of the dc-link capacitor comprising series resistance (ESR), capacitance (C) and series inductance (ESL).

The series resistance represents the ohmic losses of the terminals and contacts as well as dielectric losses due to dielectric absorption and molecular polarization [7-9] and is mainly responsible for power dissipation and heating in the capacitor.

2.2.1 Power Losses Clearly, the impedance of a non-ideal capacitor and thus the series resistant is dependent on frequency. As an example, figure 4 shows the characteristic of the 𝐸𝑆𝑅 of a dc-link capacitor for an automotive application, where a high peak within a small frequency range from about 200 400 𝑘𝐻𝑧 can be observed. This peak is caused by resonance which results from the distributed capacitance, resistance and inductance inside the capacitor itself. In general the capacitor’s behavior can be split into three parts: At low

295

Impact of modulation strategies on the power losses of the dc-link capacitor of an inverter frequencies the capacity is dominant, whereas at high frequencies the inductivity is dominating. The resonance frequency represents the transition of both areas. In the context of power losses, the frequency content of the capacitor is therefore important. Obviously, high current amplitudes in a frequency range of likewise high resistance have to be avoided. In terms of the reduction of power losses, on the one hand the characteristic of the impedance can be influenced by the capacitor’s design, on the other hand the frequency response of the current depends on the type of voltage modulation (switching patterns) and therefore is a degree of freedom in the operation of the traction inverter.

Figure 4: Series resistance of a dc-link capacitor of a traction inverter.

Caused by a static 𝐸𝑆𝑅 and the effective value of the capacitor current 𝐼 power loss of the dc-link capacitor 𝑃,

𝐸𝑆𝑅 ∙ 𝐼

, the total (3)

can be expressed by equation 3, where Kolar [8] proposed an approximation of the effective value of the capacitor current 𝐼

𝐼

2𝑀

√

cos 𝜑

√

𝑀

(4)

taking the effective value of the phase current 𝐼 , the modulation index 𝑀 and the power factor 𝑐𝑜𝑠𝜑 into account. Here, the modulation index is defined as ratio of the fundamental phase voltage to half the dc-link voltage.

2.2.2 Modulation Strategies The referenced description is often quoted (> 200 times) and a common way for comprehensive and real-time evaluation of the power losses of an aluminum electrolytic capacitor, but comes along with two significant disadvantages: First, the analytical calculation is based on the assumption of SVPWM and may show discrepancy in case of

296

Impact of modulation strategies on the power losses of the dc-link capacitor of an inverter other modulation strategies. Second, the frequency dependence of the 𝐸𝑆𝑅 depends strongly on the type of capacitor. For instance, film capacitors (widely used for automotive traction applications) reveal an entirely different characteristic compared to the electrolytic capacitors addressed in [8], where the frequency dependence can be assumed to be constant above 10 𝑘𝐻𝑧. In addition, the frequency content of the capacitor current is strongly affected by the voltage modulation itself, for example continuous modulation (represented by SVPWM), discontinuous modulation or synchronous switching (represented by SSM) [15]. With respect to an accurate and generic determination of the capacitor’s power losses, a new algorithm has to be developed that overcomes some of the restrictions.

3 Capacitor Power Loss Algorithm Based on the previous considerations, this section presents a methodology for a generic loss modeling of the dc-link capacitor. Like [8], the presented model is based on a few premises. Following model assumptions have been made: – Battery: The battery voltage is constant. – Inverter: The switching operations are ideal; switching delays and finite switching slopes are not considered. Without loss of generality the number of switching operations is integer, i. e. switching sequences are repeated periodically after one electric period. – E-Machine: The load currents are sinusoidal; harmonic components and current distortion are neglected. One essential step for the presented methodology is a modified description of the B6 bridge switching states allowing for a better understanding and simplified formulation of the dc-link current. Therefore, a modified switching function 𝑠 ∗ is defined, where the value not only depends on the phase leg itself, but additionally on the state of the other switches. The following equation 5 describes 𝑠 ∗ for phase leg 𝑈: 𝑠∗

1, if 𝑠 0, if 𝑠 1, if 𝑠

𝑠

1 and 𝑠 𝑠 or 𝑠 𝑠 1 and 𝑠 𝑠

𝑠 (5) 𝑠 .

1, the terThe other phase legs are determined by permutation. For instance, if 𝑠 ∗ minal point 𝑈 is connected to the positive supply rail, while the terminal points 𝑉 1, the terminal and 𝑊 are connected to the negative supply rail. Contrary, if 𝑠 ∗ point 𝑈 is connected to the negative supply rail, while the terminal points 𝑉 and 𝑊 are connected to the positive supply rail. Therefore, the modified switching function describes the contribution of this phase leg to the dc-link current. Last, if 𝑠 ∗ 0, the

297

Impact of modulation strategies on the power losses of the dc-link capacitor of an inverter considered phase leg has no contribution to the dc-link current. The relation between the switching function of equation 1 and 5 is demonstrated exemplarily in figure 5.

Figure 5: Relation between the switching functions explained by a switching example.

The most important feature in terms of the algorithm is that the proposed function 𝑠 ∗ leads to a specific sequence of rectangular pulses, which are characteristic for a given type of voltage modulation.

3.1 DC-Link Current With the proposed description of the switching operations, the dc-link current 𝑖 be expressed with 𝑖

𝑠∗ 𝑖

𝑠∗ 𝑖

𝑠∗ 𝑖 .

can (6)

In contrast to equation 2 this formulation yields the advantage that for each active pointer exactly one of the three summands describes the 𝑖 exclusively. Focusing e. g. on phase leg 𝑈 the product 𝑠 ∗ 𝑖 accumulated over all switching events in one electric period provides the share of the 𝑈-phase current to 𝑖 , as illustrated in figure 6.

Figure 6: Contribution of phase current (left) and modified switching function 𝑠 ∗ (middle) to the dc-link current (right) presented over an electric period.

In order to receive a frequency-dependent description of the dc-link current, the subsignals, that is the sinusoidal phase current and the rectangular modified switching function, are first transferred to the frequency domain by means of Fourier and finally combined by convolution. Because of the simple Fourier spectrum of the sine wave and a

298

Impact of modulation strategies on the power losses of the dc-link capacitor of an inverter rectangular pulse [16], the convolution algorithm can be handled with comparatively little effort and may even be applicable for modeling within the inverter software in real-time.

3.2 Capacitor Current Based on the derived dc-link-current, it is possible to determine the capacitor current 𝑖 . Considering a non-ideal battery model (see figure 7), resonance effects may occur. To receive additional information about the behavior of the dc-link circuit with respect to the battery current 𝑖 and the dc-link voltage 𝑢 , a deeper analysis of the dc-link circuit is helpful. The battery voltage 𝑢 represents the voltage source, whereas the dc-link-current 𝑖 serves as a current source. The components 𝑅, 𝐿 and 𝐶 build a resonant circuit in which energy circulates between magnetic and electric field. In comparison to [8], a more realistic representation of the load mechanisms of the powertrain can be examined.

Figure 7: Equivalent circuit diagram of the dc-link circuit containing high-voltage battery, dc-link capacitor and simulated load.

As a result from Kirchhoff’s current/voltage law and the constitutive relations for the circuit elements, the following second order differential equation is valid 𝑖

𝑅𝐶 ∙

𝐿𝐶 ∙

𝑖

0.

(7)

For solving, equation 7 is transferred into a state space model using dynamic matrix 𝑨, input matrix 𝑩, output matrix 𝑪 and feedthrough matrix 𝑫. Equation 8 and 9 outline the dynamic and output equation 𝒙

𝑨𝒙

𝑩𝒖

(8)

𝒚

𝑪𝒙

𝑫𝒖

(9)

299

Impact of modulation strategies on the power losses of the dc-link capacitor of an inverter with the state vector 𝒙, the input vector 𝒖 and the output vector 𝒚 given as: 𝑢 𝑢 𝑢 𝑖 𝑖 (10) 𝒙 (11) 𝒚 . 𝒖 𝑖 𝑖

(12)

For establishing the dynamic matrix 𝑨 in equation 13 and the input matrix 𝑩 in equation 14, the correlation between the quantities of the derived state vector 𝒙 and state vector 𝒙 and input vector 𝒖 have to be found

𝑨

⎡0 ⎢ ⎢ ⎢ ⎣0

0 ⎤ ⎥ 0 ⎥ ⎥ ⎦

𝑩

(13)

⎡0 ⎢ ⎢ ⎢ ⎣0

⎤ ⎥ 0 ⎥. ⎥ ⎦

(14)

In an analogous manner, the output matrix 𝑪 in equation 15 and the feedthrough matrix 𝑫 in equation 16 are determined to characterize the quantities in the output vector 𝒚 𝑪

1 0 0 0 1 0 0 1 0

𝑫

(15)

0 0 0

0 0 . 1

(16)

The equations 8 and 9 are transferred in the frequency domain using Laplace. The transfer function 𝑮 reads as: 𝑮

𝑪 𝑠𝑰

𝑨

𝑩

𝑫

(17)

and leads to the capacitor current 𝑖

𝑖

∙ 1

𝑠𝑅𝐶

𝑠 𝐿𝐶

𝑖

.

(18)

With respect to a real-time application, the determined transfer function can be performed either for every single switching period with respectively changing frequency response or as a whole for an entire electric period. In order to minimize computational costs, the latter offers certain potential.

3.3 Power Losses The ESR of a film capacitor (see figure 4) shows a strong frequency-dependent behavior with frequency ranges with strongly varying ESR. In fact, each frequency 𝑓 contributes as partial power loss 𝑃,

300

𝑓

∙ 𝐸𝑆𝑅 𝑓 ∙ 𝑖

𝑓

(19)

Impact of modulation strategies on the power losses of the dc-link capacitor of an inverter to the total power loss which is the sum of the individual frequency components. It is important to note, that the scaling factor ½ results from converting single-sided amplitude spectrum to single-sided power spectrum [17]. As a result, significant power dissipation occurs in cases of large resistance or reasonable amount of current amplitude at matching frequencies.

4 Results The real-time capable method is applied to both types of voltage modulation strategies SVPWM and SSM. Regarding the impact of the different modulation strategies on the dc-link stress, the capacitor current and the power losses are evaluated by simulation. The calculation of the power dissipation is performed using the 𝐸𝑆𝑅 that was addressed previously (see figure 4). The input parameters used for the simulation are listed in table 2. As can be seen, the number of switching operations, described by the ratio of the switching frequency to the electric frequency, is integer and the amount of switching pulses are equal in each phase leg. Furthermore, due to voltage phasor length limitation, the given modulation index cannot be reached during SVPWM. The actual modulation index has the value of 1.05. Table 2: Input quantities of the simulative investigations during motor operation of the electric drive. quantity

description

value

𝑢 𝚤̂ 𝑓 𝑓 𝑀 𝜑

battery voltage phase current electric frequency switching frequency modulation index phase angle

300 𝑉 150 𝐴 833. 3 𝐻𝑧 10 𝑘𝐻𝑧 4/𝜋 𝜋/9

As a first consideration, the current spectrum resulting from the proposed algorithm was compared to that receiving from Fast Fourier Transform (FFT) of the time signal. In this context, a maximum deviation from less than 2 % could be detected for both SVPWM and SSM. In a second analysis, the effective value of the capacitor current was compared to that receiving from [8], see equation 4. For SVPWM good agreement with a discrepancy of about 4 % could be proven. For SSM a reasonable discrepancy of approximately 28 % could be detected, that is because of the limitation of Kolar’s analytical expression on SVPWM.

301

Impact of modulation strategies on the power losses of the dc-link capacitor of an inverter Focusing on the impact of the modulation strategies on the dc-link stress, figure 8 and 9 demonstrate the capacitor current resulting from operating with SVPWM and SSM under the aforementioned boundary conditions. Beginning with figure 8, high current amplitude is detectable at twice the switching frequency. Other distinctive spectral lines can be found predominantly at even multiples of the switching frequency, especially at 4, 6, 8 and 10 times the switching frequency. Because of advanced SVPWM and injection of third harmonics, current peaks at odd multiples of the switching frequency show only a weak occurrence in the higher frequency range. In the lower frequency range particularly the spectral lines in the zone of 1, 3, 5 and 7 times the switching frequency are observable, where the highest amplitude can be found at the switching frequency. In general, relevant amplitudes appear at 2𝑛 1 𝑓 3𝑓 ∀ 𝑛 ∈ ℕ and 2𝑛𝑓 ∀ 𝑛 ∈ ℕ. For SSM, figure 9 reveals a completely different characteristic due to synchronous switching with the electric frequency. Spectral lines only occur at 6𝑛𝑓 ∀ 𝑛 ∈ ℕ with descending magnitude, where the highest peaks can be observed at six times the electric frequency. Evidently, the current amplitudes in figure 9 are substantially smaller compared to those in figure 8 and stay in a much narrower frequency range.

Figure 8: Capacitor current at SVPWM during motor operation of the electric drive.

Figure 9: Capacitor current at SSM during motor operation of the electric drive.

Considering the goal of a low power dissipation, the modulation strategy plays an important role. In this context, figure 10 and 11 outline the cumulated power losses of the dc-link capacitor for SVPWM and SSM. Starting at figure 10, it can be observed that the current amplitudes at the low frequency range essentially contribute to the total power loss with a share of about 35 %, where a fifth can be assigned to eight times the switching frequency. With an amount of nearly 45 % the zone of the resonance frequency (200 400 𝑘𝐻𝑧) has a substantial impact on the overall value. Even though the current amplitudes are comparatively small, the resonance peak of the 𝐸𝑆𝑅 leads to considerable power losses in this frequency range. The remainder of the power loss is

302

Impact of modulation strategies on the power losses of the dc-link capacitor of an inverter caused by approximately 20 % in the high frequency range. A highly reduced stress of the capacitor is shown in figure 11 for SSM, where the entire power loss is only 3 % of the power loss at SVPWM. The most of the power dissipation at SSM can be addressed of nearly 65 % to the lower frequencies, where almost a half can be related to six times the electric frequency. The range of the resonance frequency still contributes with a share of about 25 %. The remainder of the power loss is caused by approximately 10 % at the high frequency range. In fact, the power losses within the dc-link capacitor depend on the common occurrence of the series resistance and the capacitor current. Therefore, high power dissipation occurs when high values of resistance coincide with a significant amount of magnitude of the capacitor current at same frequencies. By simulation, it could be shown that SSM yields lower amplitudes of current in a narrower frequency range compared to SVPWM and therefore allows a reduction of the power dissipation of 97 %.

Figure 10: Capacitor power losses at SVPWM during motor operation of the electric drive.

Figure 11: Capacitor power losses at SSM during motor operation of the electric drive.

5 Conclusion This paper presents a generic approach for the computation of the capacitor’s current spectrum and its impact on the power dissipation of the dc-link capacitor in a two-level three-phase inverter for automotive applications. Based on a modified description of the power module’s switching states, the dc-link current was determined in the frequency domain without applying FFT (high computational effort) in the time domain. With the idea of real-time processing in mind, the sinusoidal load currents and single rectangular switching pulses were calculated separately in a Fourier synthesis, so that the frequency-dependent sub-signals could be easily merged by convolution. The capacitor current was computed using a dc-link model consisting of a simple RLC-circuit

303

Impact of modulation strategies on the power losses of the dc-link capacitor of an inverter to allow for a comprehensive understanding of the behavior in the dc-link circuit. As a final step, the power losses within the capacitor were derived from the 𝐸𝑆𝑅 and the capacitor’s current spectrum. The validity of the given algorithm was proven by simulation. Additionally, the impact of two different modulation strategies on the power loss formation were analyzed. It could be shown that the presented way of calculating the capacitor current in terms of energy content, is only deviating by roughly 2 % from the FFT spectrum of the same time domain signal. Furthermore, the frequency spectrum of the capacitor current, which is heavily dependent on the type of used voltage modulation, was discussed. For the given operating conditions, the power loss of the capacitor could be reduced by approximately 97 % using SSM. The generic approach may provide several advantages compared to [8]. First, it can be applied to any modulation strategy and is not limited to SVPWM. Second, power losses can be calculated with a high accuracy, because the spectral components in the frequency domain are considered. Third, resonance effects in the dc-link circuit and their impact on the electric powertrain components are taken into account as well. Therefore, the analytical model outlines a valuable methodology for the calculation of the power loss inside the dc-link capacitor, which allows for the real-time computation of the capacitor current for all possible modulation strategies.

Acknowledgment The authors thank Jan Allgeier from the Robert Bosch GmbH for providing the measured characteristic of the equivalent series resistance of the given film capacitor.

Bibliography 1. Chan, C.: The State of the Art of Electric, Hybrid, and Fuel Cell Vehicles. In: Proc. of the IEEE, 2007, pp. 704-718. 2. Casals, L.; Martinez-Laserna, M.; García, B.; Nieto, N.: Sustainability Analysis of the Electric Vehicle use in Europe for CO2 Emissions Reduction. In: Journ. of Cleaner Prod., 2016, pp. 425-437. 3. Namuduri, C.; Murty, B.: High-Power-Density Electric Drive for a Hybrid Electric Vehicle. In: Conf. Proc. 123th Annu. APEC, 1998, pp. 34-40. 4. Shen, Z.; Robb, F; Robb, S.; Briggs, D.: Reducing Voltage Rating and Cost of Vehicle Power Systems with a New Transient Voltage Suppression Technology. In: IEEE Trans. Veh. Technol., 2009, pp. 3198-3215.

304

Impact of modulation strategies on the power losses of the dc-link capacitor of an inverter 5. Bryant, A.; Mawby, P.; Palmer, P.; Santi, E.; Hudgins, J.: Exploration of Power Device Reliability using Compact Device Models and Fast Electrothermal Simulation. In: IEEE Trans. Ind. Appl., 2008, pp. 894–903. 6. Hirschmann, D.; Tissen, D.; Schroder, S.; De Doncker, R.: Reliability Prediction for Inverters in Hybrid Electrical Vehicles. In: IEEE Trans. Power Electron., 2007, pp. 2511–2517 7. Wang, H.; Blaabjerg, F.: Reliability of Capacitors for DC-Link Applications in Power Electronic Converters – An Overview. In: IEEE Trans. on Ind. Appl., 2014, pp. 3569–3578. 8. Kolar, J.; Round, S.: Analytical Calculation of the RMS Current Stress on the DCLink Capacitor of Voltage-PWM Converter Systems. In: IEE Proc.-Electric Power Appl., 2006, pp. 535–543. 9. Li, Q.; Jiang, D.: DC-Link Current Analysis of the Three-Phase 2L-VSI considering AC Current Ripple. In: IET Power Electron., 2018, pp. 202–211. 10. Jenni, F.; Wueest, D.: The Optimization Parameters of Space Vector Modulation. In: 1993 Fifth European Conference on Power Electron. and Appl. IET, 1993, pp. 376–381. 11. Holtz, J.: Pulsewidth Modulation – A Survey. In: Power Electron. Specialists Conference IEEE, 1992, pp. 11–18. 12. Kwon, Y.; Kim, S., Sul, S.: Six-Step Operation of PMSM with Instantaneous Current Control. IEEE Trans. on Ind. Appl., 2014, pp. 2614–2625. 13. Holtz, J.; Lotzkat, W.; Khambadkone, A.: On Continuous Control of PWM Inverters in the Overmodulation Range including the Six-Step Mode. In: IEEE Trans. on Power Electron., 1993, pp. 546–553. 14. Holmes, D.; Lipo, T.: Pulse Width Modulation for Power Converters – Principles and Practice. John Wiley & Sons, 2003. 15. Bruckner, T.; Holmes, D.: Optimal Pulse-Width Modulation for Three-Level Inverters. In: IEEE Trans. on Power Electron.s, 2005, pp. 82–89. 16. Oppenheim, A.; Schafer, R.: Discrete-time Signal Processing. Pearson Education, 2014. 17. Cerna, M.; Harvey, F.: The Fundamental of FFT-Based Signal Analysis and Measurement, National Instrument Corporation, 2000.

305

The six-step mode: Unwanted or rather the ideal voltage modulation method M. Eng. Thomas Zeltwanger, Dr.-Ing. Helge Sprenger, Dipl.-Ing. Mark Damson, Dipl.-Ing. Manabendra N. Gupta Robert Bosch GmbH, Schwieberdingen, Germany

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 M. Bargende et al. (Hrsg.), 19. Internationales Stuttgarter Symposium, Proceedings, https://doi.org/10.1007/978-3-658-25939-6_25

307

The six-step mode: Unwanted or rather the ideal voltage modulation method

1 Abstract In today’s electric and (plug-in) hybrid vehicles, properties such as the overall driving efficiency are an essential design goal next to its pure drivability characteristics. Component costs and power rating are additional factors to determine the commercial success of the product. Different voltage modulation methods and their configuration may be considered in design decisions to further optimize the electric drive train system. Currently, the classic SVPWM (Space Vector Pulse Width Modulation) [4] is used as the standard voltage modulation method. Other modulation methods are hardly applied to the powertrain of electric vehicles. Reasons for that could be: No sufficient practical experience, retention in terms of doubts concerning unknown or disadvantageous system interactions such as NVH (Noise-Vibration-Harshness) or restricted controllability. As a representative for the possibilities of different other voltage modulation, the “classic” six-step mode (SSM or ger. Blockbetrieb), also known as fundamental frequency clocking [1], with its significant system advantages at high machine speeds will be introduced and discussed. Measurements and simulations cannot confirm supposed disadvantages. In addition, it was possible to implement a series-ready solution on an electronic control unit (ECU) with the help of the synchronous switching method (ger. Synchrone Taktung [2] or Synchrone Steuerverfahren [1]) to solve classic controllability problems of the six-step mode. Only the voltage modulation method was enhanced, the fieldorientated current control is the same as for SVPWM.

2 Introduction The typical topology (Figure 1) of a hybrid and electric vehicle powertrain consists of an inverter with three half bridges (HB) - each half bridge has two power-electronic switches – and a connected three-phase electric machine (EM). The energy for the drive train is provided by a high voltage battery pack, which is connected via a DC-link to the inverter. The battery pack in hybrid vehicles may be smaller, because the energy shall only be buffered for a shorter period of time. If not described differently, the paper refers more specifically to the following characteristics of the electric powertrain: ● a permanent magnet synchronous machine (PMSM), ● insulated-gate bipolar transistor (IGBT) with separate diode as power electronic switch ● and a DC-link voltage 𝑢

of about 400 V.

Such an electric drive train system is typical nowadays.

308

The six-step mode: Unwanted or rather the ideal voltage modulation method DC-Link

Inverter

XDC

HBW

HBHigh

HBHigh

uDc

HBHigh

HBLow

HBLow

HBLow

uInvW

U

uInvU

uBat

Electric Machine HBV

HBU

uInvV

Battery

uPhaU

Figure 1: Typical topology of an electric drive system, with electric machine, inverter, DC-link and battery.

The essential design characteristic of such a drive train system are a maximum efficiency under all operating conditions as well as a maximum performance with an overall minimum of space consumption/box size. Table 1: A fictitious but realistic cost distribution example of an electric drive train. Traction system 100% 8.600€

Battery 81% 7000 €

Inverter 7% 600€

E-Machine 6% 500€

Transmission 6% 500€

Cost of the different drive train components is currently the main competition factor in the market. A fictitious but realistic example of the cost distribution of the drive system components is shown in Table 1. The battery pack costs dominate very much. If it would be possible to exceed the overall system efficiency goals through optimization of one or more components and meet the cost target and space requirements, then the battery pack’s capacity could be minimized and thus be designed smaller and cheaper. The saved costs on the battery side can be credited to the component with the efficiency increasing changes. The leverage effect is quite significant due to the existing imbalance of the component costs. In addition, a higher efficiency on the single component leads to lower losses and therefore less heating up of the components. This might lead to smaller component size, less costs or more performance. If we assume a component with 100 kW power rating and

309

The six-step mode: Unwanted or rather the ideal voltage modulation method an efficiency of 97.6%, the power loss is still comparable with the rating of a usual water kettle of 2400 W. The applied voltage modulation method of the inverter is one possible design element to optimize the overall system efficiency. Currently, the classic SVPWM [4] is used as the standard voltage modulation method and partially exclusively required by some customers (OEM). Other modulation methods are hardly applied. The goal of this paper is it to elaborate on the high potential of other voltage modulation methods, using the comparison of the six-step mode and the SVPWM method as an example. Besides a basic comparison of both methods, the characteristics of voltage yield, efficiency and losses as well as the controllability are analyzed more closely. Chapter 7 will shortly discuss further supposedly negative side effects to the system and will refute these.

3 Voltage modulation

Figure 2: General partition of the voltage modulation method (abstract). Overmodulation can be used with all PWM-methods.

The voltage modulation’s main objective is to actuate the desired voltage phasor 𝑈 of the current controller as accurate as possible on the machine phases. Thereby the machine phase currents can be controlled for a desired torque. The voltage modulation method (Figure 2) differ in their characteristics as followed: The switching pattern or , the voltage yield 𝑚 and the controllaits effective switching frequency 𝑓𝑟𝑞 bility and its interaction to the overall system.

310

The six-step mode: Unwanted or rather the ideal voltage modulation method

jβ

V (010) (000) (111)

-U (011)

ul

HBU

U1 ur

α U (100)

-V (101)

W (001)

tiPWM

-W (110)

HBV HBW 000

111 100 100 110 110 000

Figure 3: Left) Voltage phasor plain with actuation limits (secants), active voltage phasors (xxx) and zero phasors (000 or 111); exemplary desired voltage phasor 𝑈 ; Right) Typical switching pattern of one switching period at SVPWM which reproduces the voltage phasor 𝑈 .

3.1 Six-Step Mode The six-step mode (Figure 4) is an angle synchronous method and switches therefore synchronously depending on the angle position to the electric frequency or machine speed. Only every 60° (electric) one half bridge switches and changes hereby to the next active voltage phasor. Zero pointers are not switched. The effective switching fre, the real executed switching events per second, is the smallest of all quency 𝑓𝑟𝑞 . voltage modulation methods and proportional to the electric frequency 𝑓𝑟𝑞 𝑓𝑟𝑞

𝑓𝑟𝑞

(1)

is required to work. For the six-step mode, a minimum electric frequency 𝑓𝑟𝑞 This minimum electric frequency is defined by considering the ripple requirements in the system and its current controllability. Hence, standstill and low switching frequenis theoretically not cies are not possible. The maximum electric frequency 𝑓𝑟𝑞 limited. The six-step mode has amongst all possible voltage modulation methods and given topology the maximum voltage yield on the machine in respect to the fundamental wave of the phase voltage 𝑈 . Therefore, it is suitable to normalize the modulation index 𝑚 to the voltage of the six-step mode (SSM) [3]. 𝑚

| |

| |

(2)

311

tqEm

The six-step mode: Unwanted or rather the ideal voltage modulation method

typical range restricted range e. g. frqElecMin

frqElec

Figure 4: The six–step mode with its switching pattern (left) and its actuation range (right). The bright and dark blue range marks together the field weakening range.

The voltage phasor length with 𝑚 1 is constant and cannot be varied. Thus, the six-step mode is only interesting for operation points in the field weakening range. However, this is typically a wide machine speed range (see Figure 4, right). The six-step mode restricts the degree of freedom of the current controller by the constant voltage phasor length |𝑈 |. Only the angel 𝜗 of the phasor 𝑈 are freely adjustable.

3.2 SVPWM The SVPWM belongs to the PWM-methods. The desired voltage phasor is modulated asynchronously to the angle position by a continuous and time synchronous switching . Hence, the effective switching with the carrier switching frequency 𝑓𝑟𝑞 frequency is 𝑓𝑟𝑞

𝑓𝑟𝑞

(3)

The voltage phasor length can be adjusted between zero and theoretically 𝑚 𝜋/ 2 ⋅ √3 0.907. This is possible due to the use of zero pointers (Figure 3). More voltage yield can be achieved by overmodulation [3]. However, overmodulation is not discussed in this paper, because customers (OEM) do not often request this method. Due to the carrier switching frequency, fix voltage pointers 𝑈 can be modulated. These fix voltage pointers are necessary for standstill of the machine. The actuation range over the electric frequency is only limited by the upper range. An empirical formula defines the frequency ratio 𝑞

312

(4)

The six-step mode: Unwanted or rather the ideal voltage modulation method

tqEm

to a minimum of 𝑞 6. Amongst others, the carrier switching frequency is chosen to comply the ripple requirements in the system.

typical range restricted range

frqElec

Figure 5: The SVPWM switching pattern with 𝑞

50 (left) and actuation range (right).

The SVPWM does not restrict the degree of freedom of the current controller. Angle 𝜗 and length |𝑈 | of the voltage phasor are freely adjustable within the actuation limits (Figure 3, left). and phase current 𝑖 Concluding, Figure 6 shows a simulated phase voltage 𝑢 the both modulation methods in a typical operation point while field weakening.

of

Figure 6: Resulting simulated phase voltage 𝑢 and phase current 𝑖 of six-step mode (SSM) 10 𝑘𝐻𝑧, at same operation point with 50 Nm and compared to SVPWM, with 𝑓𝑟𝑞 1000 Hz electric frequency. Due to the better voltage yield, 𝑐𝑜𝑠 𝜑 is different in field weakening and the phase current is about 15% lesser at six-step mode. Optically evaluated, the current ripples have the same amplitudes but with other frequencies.

313

The six-step mode: Unwanted or rather the ideal voltage modulation method

4 Voltage yield The theoretical voltage yield with six-step mode is defined as the maximum with 1. SVPWM without overmodulation has its theoretical maximum at 𝑚 𝑚 0.907 [3]. The maximum of SVPWM equals the maximum possible circle – which fits into the hexagon - in the voltage phasor plain (Fehler! Verweisquelle konnte nicht gefunden werden., left). Longer phasors over one electric period result in a circle with secants. These secants define the range of overmodulation. The voltage yield, especially with SVPWM, is significantly smaller in application. The real voltage yield can be defined by 𝑚

𝑚

⋅𝜂

⋅𝜂

𝐶

.

(5)

The voltage yield restriction, due to the hardware 𝜂 , is always 1. This factor describes the real voltage yield depending on the minimum pulse width 𝑡𝑖 and lock (see Figure 7) at a given switching frequency 𝑓𝑟𝑞 and voltage modtime 𝑡𝑖 . With a neutral current (0 A), we get: ulation method 𝑠𝑡 𝜂

|

𝜂

𝑓𝑟𝑞

, 𝑡𝑖

, 𝑡𝑖

, 𝑠𝑡

1.

(6)

SVPWM switches continuously and often. The minimum pulse width is kept by limiting the minimum and maximum duty cycle on each half bridge (Figure 7). Therefore, the restricting effect is active twice. With typical values, the voltage yield can quickly 0.92. Modulation methods similar to the principle of flat-top, have only reach 𝜂 a single side restricting effect. The minimum pulse will just not be switched. The 1. The effect is independent of the machine six-step mode has no restriction 𝜂 speed.

tiPwm >tiPlsMin HBHigh

tiLock

HBLow >tiPlsMin/2 Figure 7: One half bridge signals over one switching period 𝑡𝑖 with description of the hardware conditioned switching times. If the pulse is switched it follows; solid) minimum duty cycle; dashed) maximum duty cycle.

314

The six-step mode: Unwanted or rather the ideal voltage modulation method The machine speed dependent voltage yield 𝜂 is also always 1. It considers and electric mainly the ratio 𝑞 between carrier switching frequency 𝑓𝑟𝑞 , as well as the voltage modulation method 𝑠𝑡 . frequency in the stator 𝑓𝑟𝑞 𝜂

𝜂

𝑞

𝑓𝑟𝑞

/𝑓𝑟𝑞

, 𝑠𝑡

1

(7)

the The ratio 𝑞 is high at low electric frequencies. Over one electric period 1/𝑓𝑟𝑞 voltage phasor 𝑈 shapes almost a circle at the voltage phasor plain. At higher speeds, the ratio is low. It results in a polygon with 𝑞-corners due to the actuated voltage phasors. At first, the voltage yield decreases hardly, but with higher speeds it decreases prothese are approximately 1.5% or gressively. At SVPWM and 𝑞 10 0.985. At 𝑞 6 it could reach 𝜂 0.956. Those values have been verified 𝜂 by simulation and confirm experiences from real measurements. The function of 𝜂 behaves noticeably equal to the ratio of a 𝑞-corners polygon and a circle. Within the 1 because 𝑞 6 stays constant, six-step mode the factor keeps also constant at 𝜂 as well. considers the effects of the electric The error voltage dependent correction term 𝐶 current and its orientation to the voltage by 𝑐𝑜𝑠 𝜑 . 𝐶

≅

𝐴 ⋅ cos 𝜑 𝑤𝑖𝑡ℎ 𝐴

𝐴 𝑡𝑖

, 𝑓𝑟𝑞

,𝑖

,𝑢

(8)

The amplitude 𝐴 is dependent on the cause of the error voltage [5] in which the effective and the lock time 𝑡𝑖 have a dominant part. Phase switching frequency 𝑓𝑟𝑞 and DC-link voltage 𝑢 have at high absolute values only a low dependcurrent 𝑖 ency. With 𝑐𝑜𝑠 𝜑 0 (motor mode), the error voltage reduces the voltage yield. With 𝑐𝑜𝑠 𝜑 0 (generator mode), the voltage yield is higher. SVPWM with 𝑓𝑟𝑞 , can result in 𝐴 0.029. This is another 2.9% 10 𝑘𝐻𝑧 and typical lock times 𝑡𝑖 loss of voltage yield in motor mode. The six-step mode switches much lesser 𝑓𝑟𝑞 𝑓𝑟𝑞 1000 𝐻𝑧. At same lock time, it results in 𝐴 0.008. This behavior could be proven in simulation and via measurements. An error voltage compensation [5] helps significantly to reduce the correction sumfor SVPWM. The amplitude 𝐴 can be reduced to the value 0.005. Note, mand 𝐶 that there is always a remainder, because it is not possible to compensate completely close to the secants of the actuation range. For the six-step mode a compensation is up to now not familiar and not necessary.

315

The six-step mode: Unwanted or rather the ideal voltage modulation method

Figure 8: Comparison of the real voltage yield between SVPWM and six-step mode. Motor 10 𝑘𝐻𝑧. Hatched would be the not mode and error voltage compensated with 𝑓𝑟𝑞 compensated part.

Figure 8 illustrates the real voltage yield with the reducing shares. At 𝑓𝑟𝑞 1000 𝐻𝑧 it results in a different maximum voltage ∆𝑚 of about 17.5% between SVPWM and six-step mode. That is the significant advantage of the six-step mode. Through the higher maximum voltage, we get: ● Proportional more maximum performance or torque for a PMSM not only through the smaller electric current (Figure 6) for the field weakening, but through higher available magnetic flux, which depends on the maximum voltage. ● Through the higher apparent power, for same requirements, the electric machine can be designed smaller and cheaper. For example, by shortening the motor length. ● Under same operating conditions, we get a higher machine and system efficiency by the smaller field weakening current. See next section.

5 Losses, efficiency and performance Measurements on a test bench of the electric machine show in Figure 9 (left) the power loss 𝑃 compared between six-step mode and SVPWM. Main reason is the smaller in the field weakphase current 𝑖 , to proportionally higher voltage yield 𝑚 ening (Figure 6). Another aspect is the spectrum of the current, especially through the synchronous switching pattern subharmonic oscillations are avoided. The power loss reduction does not only happen in the stator, but also in the rotor. That is mainly because of smaller phase currents. Thus, the possible continuous power increases by using the six-step mode compared to the SVPWM. In the inverter, the power loss is reduced (Figure 9, right) mainly due to the low effec. Nevertheless, even the lower phase current 𝑖 tive switching frequency 𝑓𝑟𝑞 reduces the power loss, as well.

316

The six-step mode: Unwanted or rather the ideal voltage modulation method tqEmDes/Nm Δ𝑝 150 22,3% 100 11,6% 50 1,7% 0 -3,2% n/rpm 8000 533 frqElec /Hz

𝑃

𝑃 𝑃

16,3% 8,8% 7,8% 10000 667

31,5% 11,6% 10,8% 12000 800

⋅ 100% 𝑢 320 𝑉 11,0% 14,3% 12,3% 10,8% 14000 16000 933 1067

tqEmDes/Nm Δ𝑝 150 27,7% 100 25,1% 50 32,3% 0 59,4% 8000 n/rpm 533 frqElec /Hz

𝑃

𝑃

⋅ 100%

𝑃 25,5% 33,2% 48,9% 10000 667

32,9% 31,1% 39,3% 12000 800

𝑢 24,7% 34,1% 14000 933

320 𝑉 24,7% 32,8% 16000 1067

Figure 9: Reduction of the power loss 𝑃 by the six-step mode compared to the SVPWM on 320 𝑉. the electric machine (left) and the inverter (right) at 𝑢

Due to the choice of the voltage modulation method are no significant power loss changes in the battery expected. Through smaller losses in both components, we achieve: ● More performance through lower thermal load of the components. ● Less expensive components and smaller box volumes (space) for the machine and lesser cooling power for the same requirements. The inverter has its maximum thermal load at the corner point. It is the point with the maximum torque and maximum speed on the transition to field weakening. In this case, the conditions for applying six-step mode are not completely fulfilled. Therefore, inverter optimization is not possible yet. The reduction of the power loss results in an improvement of the electric drive train efficiency 𝜂 (without battery) see Figure 10. In the illustrated operation points, the efficiency is already above 90% with SVPWM. Through six-step mode, the results show mostly an efficiency increase of at least 1%. tqEm Des/Nm Δ𝜂 𝜂 𝜂 150 1,27% 𝑢 320𝑉 100 0,78% 1,19% 2,97% 50 0,57% 0,96% 1,15% 1,30% 1,47% n/rpm 8000 10000 12000 14000 16000 frqElec /Hz 533 667 800 933 1067

Figure 10: Different of the electric drive (ED) efficiency (without battery) between six-step 320 𝑉. mode and SVPWM at 𝑢

The six-step mode shows its advantages for long and energy consuming trips in highspeed ranges, such as highway trips. The higher efficiency reduces the consumption of a trip significantly and enhances the range distance. As reverse conclusion, with same requirements, the necessary battery capacity and thus the cost of it can be reduced.

317

The six-step mode: Unwanted or rather the ideal voltage modulation method

6 Controllability 90°

0°

180° 180°-α2

0° 270° 180°-αf 360°-αf 360°-α2

uInv

uDc/2 -uDc/2 α1

α2 α3

αf 180°-αf 180°-α2 180°-α3 Voltage on the half bride output Six-Step Mode Synchronous Switching

180°-α3

360°-α3

Fundamental voltage amplitude Six-Step Mode Synchronous Switching

Figure 11: Switching pattern on a half bridge. The extension of the six-step modes by a more complex synchronous modulation method (ger. Synchrone Taktung) to reduce if necessary the voltage phasor length |𝑈 |.

A conventional field orientated current controller has two degrees of freedom to output a desired voltage phasor. The desired voltage phasor can be represented in Cartesian with 𝑢 and 𝑢 or in polar with length/amplitude |𝑈 | and angle 𝜗 . As mentioned in section 3.1 the voltage phasor length |𝑈 | is constant while using six-step mode. Only the angle 𝜗 can be changed for actuation to react on disturbances and desired value changes. The current controller loses one of its two degrees of freedom. This leads to a restricted controllability and robustness. Known solutions for this problem change the closed loop control strategies, e.g. use of the angle as sole controller output, which results in a more complex software structure. An alternative solution is to enhance the six-step mode in that way that the voltage phasor length |𝑈 | can be adjusted dynamically. This can be achieved for modulation indexes 𝑚 1 by introducing a more complex synchronous switching method [1]. By introducing of specific additional voltage pulses in the voltage, block of the six-step mode. It results in a shorter voltage phasor length or modulation index as follows [2] 𝑚 𝜶

2⋅

∑

1 ⋅ cos 𝛼

(9)

The basic goal is still to deploy the classic six-step mode in steady state, with its low effective switching frequency and a modulation index of 𝑚 1. The enhancement has been implemented for series deployment and showed the expected more robust current control during tests on the test bench. Runs, maintaining a minimum electric frequency of about 700 Hz, showed equal results compared to SVPWM for the dynamic of the current control with six-step mode. Oscillations on the signals could be reduced and software structures were optimized significantly. Due to this adaption, it is also possible

318

The six-step mode: Unwanted or rather the ideal voltage modulation method to smooth now the voltage step during transition from SVPWM to six-step mode and back. Unwanted torque steps are avoided now.

7 Other system interactions With the introduction of an alternative voltage modulation method, several system interactions have to be considered. One primary concern is the DC-link voltage ripple and the stress it causes onto the DC-link capacitor. The design of the capacitor is driven itself by component costs and size (space requirements). In simulations, it could be shown, that the six-step mode even less stresses the DC-link capacitor when operating at a high enough electric frequency. This is because the six-step mode does not set any zero pointers, in contrast to SVPWM. For high modulation indexes, zero pointers cause high charge/discharge-currents in the capacitor. For electromagnetic compatibility (EMC), no additional emissions are expected. The lower effective switching frequency limits its influence, thus cost intensive corrective measures are not necessary. Especially for the psychoacoustic perception, the acoustics or NVH behavior needs to be considered. First tests did not reveal any essential changes compared to SVPWM. The sound of the PWM frequency disappears completely, in contrast to SVPWM. In the higher harmonic bands of the electric frequency no significant negative effects could be detected. Proband tests in a vehicle confirmed no negative acoustic effects.

8 Conclusion This paper discussed the six-step mode in comparison with the standard voltage modulation method SVPWM. The essential characteristic of the six-step mode, the extremely high voltage yield has been illustrated and compared, considering the real operating effects of the voltage modulation. Due to the high voltage yield and not at least due to the low number of synchronous switching events, significant less losses occur in the electric machine and inverter and thus result in a higher system efficiency. Disadvantages in the controllability of the six-step mode could be solved with the help of the more complex synchronous switching methods in an efficient and ready-forseries way. Other system interactions on NVH, EMC and DC-link have been analyzed and didn’t show a negative impact. Table 2 sums all mentioned effects in comparison with a subjective rating.

319

The six-step mode: Unwanted or rather the ideal voltage modulation method Table 2: Summary effect comparison of the modulation methods within operation points which fulfills the conditions for six-step mode. Effect / Goal Efficiency / more Voltage yield / more Effective switching frequency / less Phase current / less Max. peak performance / more Max. continuous performance / more DC-link capacitor stress / less EMC emissions / less Acoustic/NVH emission / less Cooling power / less Controllability / better Machine design* / smaller / cheaper Battery design* / smaller / cheaper

Six-step mode in comparison to SVPWM ++ ++ ++ + ++ + 0 + 0 + 0 (improved) ++ + *with same requirements ++/+ better; 0 equal; --/- worse

When the electric drive system operates in the field weakening range and its minimum is exceeded, then the conditions to apply the six-step electric frequency 𝑓𝑟𝑞 mode are fulfilled. This is true at least in the upper third of the machine speed range. During assumed long and fast highway trips, these conditions are continuously fulfilled and the system could improve the maximum driving range due to the higher drive train efficiency. If now the overall drive system design considers the six-step mode from scratch, then it is possible to save costs and space requirements of the drive system under equal performance requirements with an optimized electric machine, inverter and battery pack. However, SVPWM is often explicitly and exclusively requested from customers (OEM), which prevents the use of other voltage modulation methods. If the drive system is already designed exclusively for SVPWM, it is still possible to upgrade the system by a pure software solution for the power electronics’ micro controller. It would then be a little more limited in system optimization for the six-step mode or other voltage modulation methods. In this case, more performance and a higher system efficiency for a higher maximum driving range with the same system components could be in focus.

320

The six-step mode: Unwanted or rather the ideal voltage modulation method To conclude, this paper strongly suggests to additionally consider other voltage modulation methods, next to SVPWM, in the design process of electric powertrain topologies to optimize the overall system efficiency, component costs and space/box volume requirements.

References 1. Schröder; Elektrische Antriebe – Regelung von Antriebsystemen; 3. Auflage, Springer, 2009 2. Jenni; Wuest; Steuerverfahren für selbstgeführte Stromrichter; vdf/Teubner, 1995 3. Holtz; On Continuous Control of PWM Inverters in Overmodulation Range Including the Six-Step Mode; IEEE; 1993 4. Handley; Boys; Harmonic analysis of space vector modulated PWM waveforms; IEEE; 1990 5. Weichbold; Einfluß nichtidealer Wechselrichtereigenschaften auf das Betriebsverhalten von Pulsweitenmodulierten Spannungszwischenkreisumrichtern; Technische Universität Graz, 2001

321

Design of a fail-operational powertrain for automated electric vehicles M.Sc. Tunan Shen, Dr.-Ing. Ahmet Kilic, Dr.-Ing. Jochen Faßnacht, Dr.-Ing. Christian Thulfaut Robert Bosch GmbH, Germany Prof. Dr. Hans-Christian Reuss Institute of Internal Combustion Engines and Automotive Engineering, University of Stuttgart, Germany

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 M. Bargende et al. (Hrsg.), 19. Internationales Stuttgarter Symposium, Proceedings, https://doi.org/10.1007/978-3-658-25939-6_26

323

Design of a fail-operational powertrain for automated electric vehicles

1 Introduction Automated vehicles will change the role of privately owned cars in the future. New mobility solutions like Urban Automated Shuttles (UAS) and Urban Automated Taxis (UAT) can provide mobility as a service with a lower price. As a result, most urban and suburban residents will be able to be mobile without owning a car. According to a forecast by Roland Berger the kilometers driven with privately owned passenger cars will decrease by approx. 28% in 2030 compared to 2015, while on the other hand the kilometers driven with automated vehicles will increase to 27% of the kilometers driven worldwide [1]. Figure 1 presents an internal analysis of Robert Bosch GmbH regarding the future mobility scenarios [2]. It can be seen that the market volume of UAS and UAT will increase dramatically in the next 10 to 15 years and automated vehicles will take a key role in the future mobility scenarios.

Figure 1: Market volume of UAS and UAT regarding an internal analysis of Bosch.

Nowadays, automated shuttles have been deployed in different restricted areas e.g. campus, parking lots and industrial sites [3]. However, the operation still relies on a safety steward who is responsible for transitioning the vehicle to a safe stop at the ego lane and ensuring the safety of the passengers in case of failure. To enable the shuttle operation without a safety steward, the automated vehicle must be capable of achieving a minimal risk condition by itself in case of a failure [4]. To fulfill this requirement, several relevant systems have been designed with redundancy to enhance the reliability of systems, such as sensors, computing, braking, steering, and power supply systems [5], [6], [7]. However, if the powertrain system of an automated driving vehicle fails during

324

Design of a fail-operational powertrain for automated electric vehicles a trip, the vehicle might stop in the middle of a traffic lane without adequate measures to guarantee sufficient safety. The passengers in the vehicle may stay in danger in some situations. Therefore, a higher reliability of the powertrain system is recommended for a safe automated driving vehicle [8]. This paper presents a fail-operational powertrain solution, which is able to achieve a safe stop transition to the next parking area in case of a failure in the powertrain system. This paper is organized as follows. In Section 2, related requirements for the design of the powertrain system are discussed. Section 3 presents the development of a failoperational powertrain topology and subsequent redundant components. Section 4 introduces a concept to avoid critical failures in the powertrain by predictive diagnosis. Section 5 briefly summaries the main results and concludes this paper.

2 Requirements In this section, the requirements of automated vehicles are analyzed from a legal perspective. The current legal regulations in Germany (StVO, StVZO) require a driver to secure the vehicle and safeguard the passengers and other traffic participants in case of a vehicle breakdown [9]. Some automated vehicles (SAE level 3) may be designed to allow for driver operation and the driver will take over the responsibility. However, if there are only not trained passengers, e.g. children, in the vehicles (SAE Level 4), a vehicle breakdown is unsafe for the passengers as well as others traffic participants e.g. a vehicle stops in the middle of an active traffic lane. Therefore, a higher reliability of the powertrain system is recommended. Moreover, different safe stop maneuvers after occurring a system failure are discussed to match the requirements of different usecases and customer needs. An overview of seven safe stop levels (SSL) is depicted in Figure 2. The main distinguishing parameter of these levels is the intended safe stop location or destination, starting with today’s emergency braking (SSL G) and ending up with driving home to complete the mission (SSL A). Besides the safe stop location, the table above also illustrates the required functionality of the three subsystems driving, steering and braking involved into vehicle motion. Also the needed energy demand to assure each level is characterized schematically. From SSL D (emergency lane) to SSL A (driving home) a fail-operational powertrain system (drive) is needed [10]. Regarding different SSL, requirements for three different reliability scenarios for the fail-operational powertrain system are derived. In case of SSL D, a degraded mode enabling a driving functionality for short time and low power is assumed. In this scenario, the automated vehicle would be able to reach the emergency lane on highways in case of a first failure. In other scenarios, e.g. SSL A or B, the degraded propulsion functionality for a long trip with even high torque and power are required to enable a transition to a safe state for example in mountainous areas. Furthermore, the required failure rate

325

Design of a fail-operational powertrain for automated electric vehicles

Figure 2: Safe state definition of automated vehicles using safe stop levels (SSL).

for avoiding loss of minimum traction power or torque must be defined for a given use case [2]. Other important perspectives are the cost and weight of the system. As a powertrain system is expensive, a simple duplication is usually costly. The additional weight and demand of mountinge space have to be taken into account. This paper focusses on the powertrain system of an electric vehicle, which consists of the power supply system (e.g. battery system, fuel cell system), inverter(s), electrical motor(s), gearbox(es) and other mechanical parts. The efficiency, noise, torque ripple and torque accuracy of a fail-operational e-powertrain should be comparable to standard e-powertrains in normal operation. Last but not the least, the compromise between safety and remaining useful life (RUL) has to be considered. For example, under a detected over temperature condition a component may be degraded or deactivated to avoid over load and reach the designed RUL. However, due to the fail-operability, this overheated component may be required to operate continuously in order to reach a safe state.

3 Design of Fail-Operational Powertrain The powertrain system of an electric vehicle consists of battery system(s), inverter(s), electric motor(s), gearbox(es) and other mechanical parts. The classical powertrain will be shut down if any subsystem or component fails. In this section, the typical failure mechanism of each subsystem and component will be analyzed firstly. The fail-operational design and proposed solution will be explained in detail.

326

Design of a fail-operational powertrain for automated electric vehicles

3.1 Gearbox Gearbox is an important mechanical component in the electric drive systems. Failure occurrence in the gearbox will results in performance degradation, which affects the efficiency of the powertrain and increases energy consumption. In critical cases, the gearbox can block the wheel shaft which would lead to severe consequences [11]. A mechanical clutch may avoid the block of wheel, but it will result in increased complexity, cost and an additional component, which must be fail-operational as well. The gearbox and bearings are proven-in-use components, whose lifetime can be well calculated and designed. Therefore, a simple design of the mechanical system without additional electromechanical or hydraulic actuations and clutches is preferred. To avoid a critical failure in those mechanical system like bearings, shafts, gearwheels and housing, a sufficient design or even overdesign of lifetime is required. Furthermore, with the new diagnostic technology, the wear and potential damages can be predicted [12]. In the following section, the predictive diagnosis will be introduced.

3.2 Inverter and electric machine In this paper, only induction and synchronous electric machines are considered. The main reason is to fulfill the requirements in normal as well as failure conditions. Additionally, these two kinds of motor are mostly used in today’s industry and automotive applications. We can profit from high production volumes and production technologies of standard series traction drives. To reduce the complexity and avoid additional hardware, which could fail, the separately excited synchronous machine is not considered in the following despite the advantages like good efficiency and the reducible excitation in case of a failure. There are a lot of different fail-op e-drive concepts mentioned in literature for aviation [13], [14] and for industrial applications in [15], [16]. The aviation concepts demand higher reliability compared to automotive traction drives, but weight and mounting space is much more relevant and cost is not so dominant because of lower production volumes. Therefore, here mainly one multiphase machine concepts feed by redundant power electronics from individual electrical power sources are used. Figure 3 presents one possible fail-operational topology for aviation applications. In case of one failure, only one phase is lost which means the possible torque and power is only slightly reduced which decreases the demand for necessary overdesign. But the automotive application cannot afford the high cost due to the multiple redundancy e.g. of powernets, control units and cabling.

327

Design of a fail-operational powertrain for automated electric vehicles

Figure 3: Example of fail-op drive for aviation applications [13].

In industrial applications, an interrupted production caused by a failure in the e-drive system (inverter and elec. machine) can result in high loss of profit or a lot of scrap material and a damaged production line. Therefore, the requirements of availability are higher than in aviation or automotive applications, but the safety requirements e.g. Fitrates, are much lower. Redundancy is often integrated into the power electronics by additional bridges, fuses and additional switches to decouple damaged power semiconductors and activate redundant ones [15], [16] and the DC power supply is supposed to be sufficiently reliable. A typical fail-operational design in industry is shown in Figure 4. However, these additional fuses and switches in the current path while normal operation are not preferred in electric vehicles because they cause additional losses and reduce the range of electric vehicles. Neither the concepts from aviation nor the concepts for industrial applications are therefore suitable for automotive applications. For the design of a suitable e-drive concept used in automotive fail-operational powertrains different measures depicted in Figure 5 increasing the reliability in general are proposed. In addition, also standard process measures avoiding production and design failures are necessary for the increased reliability. Since the reliability of components with distinct wear mechanisms reduces significantly at the end of their lifetime, also service measures like preventive change of components are required. Other measures like thermal overdesign of the e-drive, resulting in higher component cost, weight and mounting space demand, should be avoided if possible. Nevertheless, the derating of fail-operational e-powertrains has to be taken into account, since loss of traction might become safety relevant depending on the use case and should be avoided by design or appropriate control strategy. Unlimited derating for component protection as used in classical powertrains will be not possible any more.

328

Design of a fail-operational powertrain for automated electric vehicles

Figure 4: Example of fail-op e-drive for industrial applications [15].

A design consisting of two galvanically separated power distribution systems as depicted in Figure 6 is proposed for automotive powertrains in this work. Each distribution system has an individual power supply and inverter and feeds three of six phases of an induction or a synchronous electric machine. In order to avoid a single point or common cause failures, both redundant systems should have no common components, meaning that each inverter and battery must have its individual control unit. Each of the two galvanically separated electrical systems should be able to provide the necessary torque and power to reach the required degraded minimum performance. In this case, a defect system can be switched off without violating the requirement for fail-operability, solving so also the contradiction of high voltage safety or prevention of hot corrosion and the provision of minimum torque limits. In the case of a necessary minimum power and torque limit of 50%, no or only slight overdesign of the components of the two systems is necessary. If higher torque or power limits must be reached in case of a single failure, significant overdesign may be necessary. If nearly 100% of torque and power are demanded in case of a single failure, smaller galvanically isolated and completely independent systems like the one in aviation systems become interesting, but the increased granularity will also increase the control unit and cabling effort significantly.

329

Design of a fail-operational powertrain for automated electric vehicles

Figure 5: Possible measures to increase reliability of systems.

Figure 6: Proposed fail-operational e-powertrain topology.

330

Design of a fail-operational powertrain for automated electric vehicles For enabling a degraded vehicle motion in case of a failure for at least 30min, an appropriate control strategy keeping the state of charge of all redundant energy storages or supplies reliable and sufficiently high is necessary. Such a control strategy is described in for example [8]. With the presented redundant topology and design measures, it is possible to reach the requirements for a fail-operational powertrain.

4 Predictive Diagnosis An alternative way to achieve sufficient reliability is to achieve predictive fault detection. If a failure can be detected in an early stage, it is able to limit, reconfigure the system as well as repair or replace the component before it breaks down. With this approach, some hardware redundancy may be avoid. In this section, two kinds of failures in the electric machine will be considered and the diagnostic concept are introduced.

4.1 Wear out Failures Many failures in the electric machine are wear out failures, such as pitting of bearing or gear due to overloading [18] and drift of resistor due to thermal stress or corrosion [19]. The failure rate of these kinds of failures always increase over the use time. If a failure is predicted in advance or at the beginning of degradation process, the risk of system breakdown can be prevented. For example, the diagnostic concept of bearing faults in wind turbines and industries are developed in [20], [17]. In [17], the data sets of run-to-failure tests from the Center of Intelligent Maintenance System (IMS), Universitiy of Cincinnati [21] were used. Figure 7 shows the measured signal and filtered signal from a run-to-failure test due to a defect on the outer raceway. The de-noised signal shows a gradual increase of vibration magnitude while day 4 and a significant increase and fluctuation more than two days before the break down of the bearing at day 7 [17]. With proper modifications, this method can also be used in the automotive applications. Another solution is to use specific bearing, which can still work after bearing failure for short time [22].

4.2 Contribution of Predictive Diagnosis in Fail-op Topology In the fail-op topology proposed in section 3, a predictive diagnostic concept can also help to reduce the cost. In case of a common failure in the electric machine, e.g. interturn short circuit, a short circuited phase cannot be shut down. If this failure is not detected in time, a very large fault current will be induced and produce excessive local heat, which results in a rapid deterioration of the neighboring winding’s insulation. In the critical case, the electric machine can be damaged. If this failure can be detected

331

Design of a fail-operational powertrain for automated electric vehicles earlier, the development of the failure can be avoid by operating at degradation mode and the whole system can still operate. To detect the interturn shot circuit failure, an

Figure 7: The original and de-noised signals [17].

analytical model is built in accordance with [23] and the equivalent circuit of the failure is shown in Figure 8. The parallel resistor 𝑅 represents the insulation resistance between two turns of the stator. 𝐼 is the current flow on the short-circuited path, is the leakage current flowing through the winding insulation layer. The whereas 𝐼 ratio of these two short circuited turns is denoted by 𝜂. By varying two failure parameters 𝑅 and 𝜂, different stages of failure can be simulated. Some simulation results of the model in different conditions are presented in Figure 9. The phase currents and output torque of the machine in healthy condition is shown in Figure 9(a), in which the insulation resistance is 10 MΩ. In this condition, there is no and the amplitudes of the three phase current are almost the fault current 𝐼 same. The deviation of the output torque from the reference value is also very low. In Figure 9(b), 𝑅 deteriorates to 1Ω, which results in a small fault current. While 𝑅 reduces to 0.1Ω in Figure 9(c), a large fault current is induced, which can damage the electric machine. A fault detection method based on four detection features is developed in [23]. With this method, an incipient failure such as the condition in Figure 9(b) can be detected effectively with the consideration of measurement error.

332

Design of a fail-operational powertrain for automated electric vehicles

Figure 8: Equivalent circuit of interturn short circuit [23].

Figure 9: (a) Current and torque of machine in healthy condition at 100Nm, 5000rpm; 1Ω and 𝜂 10% at 100Nm, 5000rpm; (b) Current and torque of machine with 𝑅 (c) Current and torque of machine with 𝑅 0.1Ω and 𝜂 10% at 100Nm, 5000rpm.

333

Design of a fail-operational powertrain for automated electric vehicles

5 Summary This paper introduces a way to design a powertrain system for future mobility solutions. Firstly, the requirements of the powertrain system are derived from different scenarios of future transportation systems. To reach a more comfortable and safer situation in case of a system failure of autonomous driving vehicles, a fail-operational powertrain is needed. After analysis of the concepts from industry and aviation, a cost optimized fail-op concept for automotive applications is presented. To reduce the hardware redundancy and save cost, a predictive diagnosis concept is introduced in this paper. The contribution of predictive diagnosis in the fail-operational design is explained with two examples of bearing fault and interturn short circuit in the stator of an electric machine.

Bibliography 1. Roland Berger GmbH, „A CEO agenda for the (r)evolution of the automotive ecosystem,“ Roland Berger GmbH, Munich, Germany, 2016. 2. A. Kilic, J. Faßnacht, T. Shen and C. Thulfaut, „Fail-operational powertrain for future mobility,“ Springer, Frankfurt, 2018. 3. Easymile, Easymile, 28 11 2018. [Online]. Available: http://www.easymile.com/. [access on 28 11 2018]. 4. SAE International, J3016 SEP2016: Taxonomy and Definitions for Terms Related to Driving Automation Systems for On-Road Motor Vehicles, United States: SAE International, 2016. 5. Maymo, „Waymo Safety Report, On the road to fully self-driving“ Maymo, 2017. 6. A. Kilic, T. Shen and K. Gorelik, Development of fail-operational power net for automated driving, 8. VDI/VDE Fachtagung, Autoreg 2017 automatisiertes Fahren und vernetzte Welt, Berlin, 2017. 7. T. Shen, A. Kilic and K. Gorelik, Dimensioning of Power Net for Automated Driving, The 30th Interna-tional Electric Vehicle Symposium & Exhibition, Stuttgart, 09-11.Okt. 2017. 8. K. Gorelik, R. Obermaisser and A. Kilic, Optimal, Adaptive and Predictive RealTime Control of Fail-Operational Powertrain for Automated Electric Vehicles, Long Beach, USA, 2018.

334

Design of a fail-operational powertrain for automated electric vehicles 9. UNECE, CONVENTION ON ROAD TRAFFIC, UNECE, VIENNA, 1968. 10. K. Gorelik, A. Kilic and R. Obermaisser, Energy Management System for Automated Driving - Optimal and Adaptive Control Strategy for Normal and Failure Case Operation, in 11th Annual IEEE International Systems Conference, Montreal, Canada, 2017. 11. T. Shen, A. Kilic and H.-C. Reuss, A New System Diagnostic Method for Powertrain of Automated Electric Vehicles, in The 21st IEEE International Conference on Intelligent Transportation Systems, Maui, 2018. 12. FAG, Wälzlagerschäden, Publ.-Nr. WL 82 102/2 DA, , Stand 2000. 13. F. Baudart, Design and control of fault-tolerant permanent magnet drives, Diss. Universite Catholique de Louvain, 2012. 14. I. Bolvashenko and a. et., Research on Reliability and Fault Tolerance of MultiPhase Traction Elektric Motors Based on Markov Models for Multi-State Systems, Institute of Energy Conversion, TUM, 2016. 15. R. Errabelli, Inverter and Controller for Highly Available Permanent Magnet Synchronous Drives, Diss. SRT, TU Darmstadt, 2012. 16. B. Welchko and e. al., Fault Tolerant Three-Phase AC Motor Drive Topologies: A Comparison of Features, Cost and Limitations, IEEE Transactions on Power Electronics, Vol. 19, No. 4;, July 2004. 17. A. J. Cardoso, H. V. Khang, K. G. Robbersmyr and A. J. M., Bearing fault detection for drivetrains using adaptive filters based wavelet transform, in 2017 20th International Conference on Electrical Machines and Systems (ICEMS), 2017. 18. A. J. Bazurto, E. C. Quispe and R. C. Mendoza, Causes and failures classification of industrial electric motor, Arequipa, 2016. 19. Bohrer, C. W. Lewis and J. J., Physics of Resistor Failure, in First Annual Symposium on the Physics of Failure in Electronics, 1962. 20. Ramachandran, T. Praveenkumar, M. Saimurugan, P. Krishnakumar and K.I, Fault Diagnosis of Automobile Gearbox Based on Machine Learning Techniques, Procedia Engineering, Bd. 97, pp. 2092-2098, 2014.

335

Design of a fail-operational powertrain for automated electric vehicles 21. J. Lee, H. Qiu, G. Yu, J. Lin and R. T. Services, IMS, University of Cincinnati in Bearing Data Set, NASA Ames Prognostics Data Repository (http://ti.arc.nasa.gov/project/prognostic-data-repository), NASA, Ames Research Center, Moffett Field, CA, 2007. 22. Vogel, www.maschinenmarkt.vogel.de, Vogel, [Online]. Available: https://www.maschinenmarkt.vogel.de/index.cfm?pid=7502&pk=552862&fk=10 83085&type=article. [access on 19 12 2018]. 23. D. Yan, T. Shen, A. Kilic and H.-C. Reuss, New Fault Detection Method for Interturn Short Circuit on Stator Winding of IPMSM for Highly Automated Driving Applications, in 2019 IEEE/SICE International Symposium on System Integration, Paris, 2019.

336

A scalable approach for future vehicle electrification Dr. Carsten Bünder Director Product Management, Magna Powertrain, Transmission Systems

This manuscript is not available according to publishing restriction. Thank you for your understanding.

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 M. Bargende et al. (Hrsg.), 19. Internationales Stuttgarter Symposium, Proceedings, https://doi.org/10.1007/978-3-658-25939-6_27

337

Grid integration e-mobility – Developments and challenges Ursel Willrett IAV GmbH, Sindelfingen, Germany

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 M. Bargende et al. (Hrsg.), 19. Internationales Stuttgarter Symposium, Proceedings, https://doi.org/10.1007/978-3-658-25939-6_28

339

Grid integration e-mobility – Developments and challenges

1 Introduction The charging interface between grid and electric vehicles is new. To integrate the “two worlds” power supply and vehicle for many challenges solutions have to be found and introduced. Many different requirements for charging are considered: electric vehicles are charged in private environments, at work place or at public charging infrastructures. Therefore adequate power supply access is necessary to charge the batteries safely and reliably within the desired charging duration. For charging in public areas a communication system is required which connects all involved actors and supports grid integration of e-mobility. Most important functions are authentication, billing including e-roaming facilities, load management with different charging power including support of high power charging > 300 kW. For communication between electric vehicle and charging stations to support the functions mentioned above the international standard ISO 15118 is available. Due to the transfer of personal data, secure communication procedures are requested. Therefore transparent end-to-end communication concepts have to be implemented with clearly specified interfaces. One of the challenges regarding communication is the proof of interoperability of electric vehicles and charging infrastructure. Electric consumers (i.e. electric vehicles) need on the one hand a stable power supply, they expect grid quality according to the given standards. Otherwise every consumer adds distortion or noise into the grid. During the charging process of already one electric vehicle perturbations are visible. For proper grid integration of all components in a system respective procedures have to be specified which are accepted and supported by all actors: automotive industry, power suppliers, producers of charging infrastructure and users. The system “E-Mobility” provides the integration of further applications and system solutions, for example applications for intelligent mobility, traffic control, grid integration with renewable energies or value added services for users. Major actual projects in the development and introduction of e-mobility are standardization and implementation of high power charging, bi-directional energy transfer, load management, ensuring grid quality and transparent and easy-to use charging procedures for the user [1].

2 High Power Charging Charging systems for electric vehicles, which are capable to re-charge high-voltage batteries in short load times become more and more important for introduction and acceptance of e-mobility. Therefore in the industry the effort had been increased to specify and develop DC (Direct Current) fast charging systems of the new generation, especially with increased charging power and decreased load times.

340

Grid integration e-mobility – Developments and challenges Installed DC Charging systems today based on CCS (Combined Charging System) or CHAdeMO (CHArge de Move, DC charging standard, Japan) provide usually 50kW charging power. The roll-out of the Supercharger of Tesla shows that fast charging systems with charging power above 100kW are feasible and usable by normal users. According to the existing international standards (IEC 61851 [2],[6], IEC 62196 [3]) charging systems based on CCS with a maximum charging power of 170 kW (850V, 200A) can be developed. The necessary technology to perform charging systems with charging power above 170kW is available. Existing standards can be extended respectively. For implementation an agreement is required regarding the new extended limits above 170kW for maximum charging power, charging voltage and current. The international standardization teams are working on an extension for a maximum charging power of 400 kW (1000V, 400A). Research projects are working on first solutions for charging power >500 kW. Based on the agreed maximum values all involved companies (automotive industry, charging station manufacturer, energy provider) can contribute properly to the solution. For example the power supply access will be a medium voltage access in the future for fast charging systems with higher charging power. Charging with higher charging voltages and currents require new solutions in charging stations, load cables and electric vehicles. Some of the system components will become larger and heavier. Power loss in the charging circuit increases with increasing charging power and causes higher temperatures. Built-in methods to perform electrical safety are mandatory to avoid damage of persons or of property. DC fast charging systems with charging power far beyond the installed systems today are feasible using available technology. Beside the challenges to provide the technical solutions for the fast charging systems additional enhancements in the international standards have to be included to ensure international market introduction.

3 Grid Quality Electric consumers (i.e. electric vehicles) need on the one hand a stable power supply, they expect grid quality according to the given standards. Otherwise every consumer adds distortion or noise into the grid. Today only few electric vehicles exist. During the charging process of already one electric vehicle perturbations are visible. These nonlinear distortions are caused mainly by semiconductors used in charging rectifiers either in the charging station (DC charging) or in electric vehicles (AC charging). Predictions of distortions for many electric vehicles cannot be determined, it is known that the results cannot be calculated by a linear approximation. Reductions of noise and distortions is performed by adequate filters and compensation circuits.

341

Grid integration e-mobility – Developments and challenges Grid distortions may cause interruptions in power supply systems. Unexpected breakoffs of the charging process may be caused by grid distortions. Devices and equipment connected at the same power distribution may be damaged or destroyed, fuses may fire.

U(t) and i(t) during charging process of one electric vehicle [IAV]

Voltage quality and electromagnetic compatibility (EMC) are linked. Definition of grid quality according to IEC 61000 [5]: „Characteristics of electricity at a certain location in an electrical system, the characteristics is rated based on a selection of technical reference parameters.“ This definition is location independent and includes EMC compliance. The characteristics for grid quality from the view of the power network is specified in EN 50160 [4]. The standard contains the tolerance thresholds for permitted distortions caused by connected consumers. Examples for parameters according to EN 50160: total harmonic distortion (THD) or active power factor cos 𝜑 .

4 Communication and Interoperability 4.1 Communication ISO 15118 ([7],[8]) is the specified protocol stack on Power Line Communication (PLC) between electric vehicles and charging stations modulated on a basic pulse-width communication (PWM [6]). In available charging stations and electric vehicles, the communication is implemented according to the standard DIN SPEC 70121 [13]. This standard is a derivate of the ISO 15118 with a reduced set of functions. It is used for DC-Charging only to support especially charging control. Data security is not part of this standard.

342

Grid integration e-mobility – Developments and challenges For data which are transferred between electric vehicles and the backend, the charging station acts as a gateway. Between charging station and the backend, adequate protocols are applicable (i.e. OCPP [12]). Several protocol stacks are currently used, a worldwide standard is actually developed. The system consists of many parties. The electric vehicles and charging stations are primary actors. All parties beyond the charging station from the view of the user of an electric vehicle are called “secondary actors”. Secondary actors are electricity providers, clearing house, e-mobility operator, meter operator, fleet operator, e-mobility operator clearing house, distribution system operator and original equipment manufacturer. The communication between the primary actors (charging station, electric vehicle) is performed according to ISO 15118. The communication between charging station and secondary actors in the backend uses respective network protocols. Examples for standardized protocols for public charging infrastructure are OCPP, IEC 61850 and IEC 63110. Standardized protocols for home energy management systems (HEMS) are SPINE ([9],[10],[11]), SEP 2.0 or Echonet Lite.

4.2 Interoperability Communication between charging station and electric vehicle is performed according to ISO 15118 protocol stack. A protocol stack is in general the conceptual architecture of a set of communication protocols according to the OSI-Reference model (OSI – open system for interconnection). The protocols are sorted in seven layers starting from the physical layer (layer 1) to the application layer (layer 7), s. figure below:

Protocol stack of ISO 15118 [IAV]

343

Grid integration e-mobility – Developments and challenges ISO 15118 describes all sequences, messages and parameters to provide the required functions. Data security methods are specified, i.e. encryption methods, format and contents of certificates. It provides controlled charging communication which includes connection and disconnection, authentication, selection of services, charging control, load management, data security methods and the support of value added service. The communication within a protocol is performed using specified messages and parameters. Each protocol includes the definition of syntax, semantics and timing constraints. Examples for important parameters are charging voltage, current, charging power, desired start or stop of charging process, tariff tables. Manufacturers of charging stations and electric vehicles deploy products which include the communication modules. If the communication of one or both parties is not performed, according to the standard either charging is not even started or the process is interrupted before completion. It is required to prove interoperability before deployment. To prove interoperability the communication modules of charging stations and electric vehicles are tested using a protocol test system. The test system acts as a charging station if an electric vehicle is tested. Therefore, a programmable power supply is required to support tests in the charging loop. The test system acts as an electric vehicle if a charging station is tested. In this case, an electrical load is part of the system. All requirements specified in the ISO 15118 are tested using test cases. Using a test system which is capable to run the test cases in real time conditions electric vehicles and charging stations are tested separately. After successful test conformity of the standard has been proven. They will interact reliably with all other components which meet the standards. Electric vehicles can charge at all charging stations. There will be no problems caused by communication failures. Manufacturers of charging stations and electric vehicles can focus on their products only. Service requests by customers will decrease significantly.

5 Data Security Data security covers three main properties: Authenticity Confidentiality Integrity

344

The communication parties are really those which they claim to be. The contents of a message can be read only by intended recipients, not by unauthorized third parties. Unauthorized modification of the sent message must be avoided or at least detected.

Grid integration e-mobility – Developments and challenges

5.1 Data security methods standardized in ISO 15118 The protocol stack of the ISO 15118 provides encryption in two of the layers. In layer four, the transport layer security protocol (TLS) is specified. TLS supports encryption of the data between electric vehicle and charging station. In layer seven XML security is available with encryption of data between electric vehicle and secondary actors. Asymmetric and symmetric encryption algorithms are used for secure data transmission. Encryption covers confidentiality of the two communication entities. It is a two-stage concept: lower layer encryption between electric vehicle and charging station and higher layer encryption between electric vehicle and secondary actor (end-to-end encryption). It is therefore not possible to read personal data in the charging station. The methods for handling of certificates and signatures including the respective data formats are also described in ISO 15118 ([7],[8]). Creation and verification of digital XML based signatures is the method to cover authenticity and integrity. In the ISO 15118, seven types for certificates are specified used for different purposes which are handled by different instances: V2G root certificate, charge point operator certificate, mobility operator root certificate, contract certificate, OEM root certificate, OEM provisioning certificate, private operate root certificate. According to ISO 15118 the chain for the certificates consists of at most three elements deviated by the root certificate. The lower layer certificates are signed by the higher layer certificate. For validation of authenticity of the certificates, private and public keys are used.

5.2 Challenges for implementation of data security methods Methods are available to provide secured data communication between electric vehicle, charging station and all secondary actors in the backed. In the ISO 15118, the communication between electric vehicle and charging station is specified including the description of the adequate encryption methods, handling of signatures and certificates. All specified methods comply with modern public key infrastructure systems. The methods are specified, but the standards do not describe the process how to handle data security in the system. Because of the e-mobility system is built and maintained by several parties (i.e. automotive industry, charging station manufacturers, energy suppliers, providers, further secondary actors, root instances) the definition and establishment of a process is required with clearly defined roles. Some examples of topics to be specified are: ● World-wide standard of communication between charging station and backend ● Administration of various certificates (i.e. OEM, charge point operator CPO, provider) ● Encryption ley handling and updates, secured storage of private keys and certificates ● Administration of several contracts, e-roaming

345

Grid integration e-mobility – Developments and challenges A reasonable specification, implementation and maintenance of secure communication is performed by IT specialists. Cooperation and agreement of all contributing parties of the e-mobility system is required.

6 Bi-directional Energy Transfer Many houses are already equipped with an own photovoltaics system on the roof which generates the required energy for the household. Over produced energy is fed back into the grid, the home owner is payed for each kWh with a preassigned price. In case of residents require more energy than locally available the grid serves the household with energy from the network. Goal of the energy transition is a balance between renewable energy production and demand of users. Energy production is performed centralized using wind farms or solar energy plants but also de-centralized by PV systems installed on private houses. An intelligent control (smart grid) optimizes the balance between demand and availability of electrical energy. Locally in house the same function is provided by smart home systems.

6.1 Requirements for smart charging Electric vehicles are capable to be used for bi-directional energy transfer. Not immediately used energy generated in the PV system may be used to charge the traction battery in the vehicle. During peak load phase the traction battery can be used to stabilize the smart home system internally without the need to access the grid via smart meter. Important requirements for smart charging are: ● Transfer of tariff tables ● Scheduling of charging times cooperatively between energy provider and electric vehicles ● Modification of scheduling ● Control of energy feedback using respective tariff tables for feedback ● Control of changes during the charging process, i.e. reduction of charging power, stop of charging, resume of charging process.

6.2 Potential for electric vehicles in a smart home system With the introduction of e-mobility more and more households own an electric vehicle with a battery storage (capacity > 25kWh). Statistically an average consumption of a household is determined 3500 kWh per year, approximately 10 kWh per day are required. Also according to statistical data privately used passenger cars park 23 hours per day. In the busy hour (5-8 pm), when the price to get energy from the grid will be

346

Grid integration e-mobility – Developments and challenges high, it is attractive to use the energy from the battery of the electric vehicles. Re-charging can be done either during night in times of cheaper energy price from the grid or during daytime using the own photovoltaic system. Of course the decision how the electric vehicle is used is controlled by the home owner. Calculation of costs for the home owner with different scenarios (electric vehicle versus combustion engine car, fixed battery storage for HEMS, cost to receive energy from the grid) it is already today remunerative to use electric vehicles for mobility and stabilization in the home energy system. An electric vehicle can be used for mobility and as a buffer for electrical energy. It is therefore a key component to stabilize the load management in the house locally. For a proper solution the respective communication standards are used: ISO 15118 ([7],[8]) and standardized HEMS protocols (i.e. SPINE [9],[10],[11]).

6.3 System integration Smart home concepts are an application of Internet of Things. A smart home system connects all devices (i.e. PV system, thermostats, light, heating) using an intelligent control to support local load management. For a smart home system an electric vehicle is also a device and can be integrated into the home energy management system (HEMS) for bi-directional charging applications (smart charging). Locally over-produced energy may be stored in the traction battery of the vehicle. Demand for mobility and for feedback of energy into the house is applicable, adjusted and controlled by the user.

Home Energy Management System [IAV]

347

Grid integration e-mobility – Developments and challenges A smart home system uses a communication system which is capable to integrate various and different devices. Most of the available smart home systems are based on proprietary communication protocols. Devices from other suppliers are not compatible and cannot be integrated into proprietary systems. Future proof smart home systems use communications protocols which are specified in international standards. Available HEMS standards for communication are SPINE, SEP 2.0 and Echonet Lite. They can be used in combination with the established communication standard ISO 15118 between electric vehicle and charging station. In the charging point of the home a gateway function is required to map information between HEMS and electric vehicle. All these standards contain already electric vehicles as a device and are mature enough that they can be used for implementation of smart home systems including smart charging functions. Electric vehicles have high potential to serve for mobility and for supporting local load management in homes.

7 Conclusion Acceptance of e-mobility by users is key for market success. E-mobility has to be easy to use and transparent. The users want to charge everywhere and expect security for their personal data. The main technical challenges for introduction of e-mobility are: ● Unique communication procedures and data security in the whole system ● Load management and proper integration into the grids ● Charging infrastructure with various charging power ● Grid stability ● Dynamic load management There are already first solutions available. To establish the e-mobility system providing all benefits to the users further effort in international standardization is required.

348

Grid integration e-mobility – Developments and challenges

Bibliography 1. NPE: Fortschrittsbericht 2018 – Markthochlaufphase; Berlin; 2018 2. IEC 61851-23: International Electrotechnical Commission; Electric Vehicle Conductive Charging System; Part 23: d.c. electric vehicle charging station; Frankfurt am Main; 2014 3. IEC 62196-3: International Electrotechnical Commission; Plug, Socket-Outlets, and vehicle Couplers – Conductive Charging Of Electric Vehicles; Part 3: Dimensional compatibility and interchangeability requirements for dedicated d.c. and combined a.c./d.c. pin and contact-tube vehicle couplers; 2014 4. DIN EN 50160 Merkmale der Spannung in öffentlichen Elektrizitätsversorgungsnetzen, Frankfurt: Beuth Verlag, 2011-02 5. DIN EN 61000-2-2 Elektromagnetische Verträglichkeit (EMV) Umgebungsbedingungen; Verträglichkeitspegel für niederfrequente leitungsgeführte Störgrößen und Signalübertragung in öffentlichen Niederspannungsnetzen, Frankfurt: Beuth Verlag, 2003-02. 6. IEC 61851-1 (2013): Electric vehicle conductive charging system – Part 1: General requirements 7. ISO 15118 (2013): Road vehicles – Vehicle to grid communication interface –Part 1: General Information and use case definition 8. ISO 15118 (2013): Road vehicles – Vehicle to grid communication interface –Part 2: Network and application protocol requirements 9. EEBus SPINE Technical Report – Introduction, Version 1.0.0, 29.4.1016, http://www.eebus.org/download-standard/ 10. EEBus SPINE Technical Report - Protocol Specification, Version 1.0.0, 29.4.1016, http://www.eebus.org/download-standard/ 11. EEBus SPINE Technical Report - Resource Specification, Version 1.0.0, 29.4.1016, http://www.eebus.org/download-standard/Bibliography Title 12. Open Charge Alliance (2014): Open Charge Point Protocol 2.0 - Interface description between Charge Point and Central System, URL: http://www.openchargealliance.org/sites/default/files/OCPP%202.0%20Release%20Candidate%202.pdf [Access 14.02.2019] 13. DIN SPEC 70121:2014-12: Electromobility – Digital communication between a d.c. EV charging station and an electric vehicle for control of d.c. charging in the Combined Charging System

349

Validation of range estimation for electric vehicles based on recorded real-world driving data Patrick Petersen, Jacob Langner, Stefan Otten, Eric Sax FZI Research Center for Information Technology, Karlsruhe, Germany Stefan Scheubner, Moritz Vaillant, Sebastian Fünfgeld Dr. Ing. h.c. F. Porsche AG, Weissach, Germany

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 M. Bargende et al. (Hrsg.), 19. Internationales Stuttgarter Symposium, Proceedings, https://doi.org/10.1007/978-3-658-25939-6_29

351

Validation of range estimation for electric vehicles based on recorded real-world driving data

1 Abstract Electrification of vehicles is a growing trend in the automotive industry. Battery electric vehicles offer the potential to reduce greenhouse gas emissions, but short maximum range and missing charging infrastructure limits user acceptance. Range anxiety is a great challenge for battery electric vehicle drivers, therefore accurate methods for range estimation are required to satisfy customer needs. Novel algorithms for range estimation include many information sources such as traffic, geographical and weather data. The increase of algorithmic complexity and the behavior of various non-deterministic influences lead to a high demand for intensive verification and validation methods for such predictive features. In this paper, we propose a methodology for the verification and validation of range estimation algorithms based on recorded real-world driving data. The steadily growing pool of recorded test drives from test campaigns enables statistically significant testing. A generalization step enables the usage of all recorded data regardless of the vehicles architecture. New software versions of range estimation algorithms can be tested and a higher feature maturity can be achieved without the need for cost- and time-intensive real-world test drives. This approach is demonstrated and evaluated in a small test campaign.

2 Introduction The current trend of electrification in the automotive industry promises a solution for the dependency on fossil fuels, reduction of greenhouse gas emissions and other environmental issues [1]. A current study has shown that roughly 90 % of vehicles on the road could be replaced by battery electric vehicles (BEVs) [2]. One of the main barriers to the adoption of BEVs is the perceived range limitation and the lack of widespread charging infrastructure [3]. This leads to the prominent barrier known as range anxiety [4]. Drivers tend to reserve large safety margins to prevent being stranded with a depleted battery, approximately 20-25 % of the battery capacity is reserved as such a safety margin [5]. Studies have shown that more experience with BEVs also increases the trust in the displayed range limits [6]. However, to overcome range anxiety and increase the acceptance of battery electric vehicles worldwide, more accurate range estimations algorithms are required. These algorithms try to achieve an accurate estimation of the remaining range by minimizing uncertainties during the estimation of energy consumption on a selected road. Different approaches exist to solve this challenge. They vary from adaptive algorithms [7], machine learning approaches [8], [9] to improved knowledge about the battery states [10]. Sautermeister et al. [11] demonstrated the following approach: To reduce range estimation uncertainty, the driving resistances of the vehicle are estimated during the trip and incorporated in the prediction. Another approach by Olivia et al. [12] is to combine particle filtering with Markov chains to

352

Validation of range estimation for electric vehicles based on recorded real-world driving data reduce estimation uncertainties of stochastic factors such as the driving style or the traffic situation. These learning algorithms are often verified and validated by test drives from test campaigns and country approvals. To enable statistically significant testing, thousands of kilometers must be recorded. These test drives are cost- and time-intensive and must be repeated for each software version of the range estimation algorithm. A more economical approach for the development and testing can be achieved by utilizing the continuously growing data-pool of already recorded prototype test drives and country approval campaigns, which are gathered by the Original Equipment Manufacturer (OEM). To execute a range estimation algorithm under development in a Software-in-the-loop (SiL) environment enables exhaustive testing on already recorded real-world data. Large-scale simulations and statistically significant assertions regarding the feature maturity in early development stages can be achieved. In this paper, we present a concept for using recorded real-world driving data for simulation-based testing and evaluation of range estimation algorithms regardless of the vehicles architecture. In Section 3 we give an overview of the state of the art of Automotive Systems Engineering (ASE) and verification and validation (V&V) methods for feature development and validation of range estimation algorithms. In Section 4 we present our approach and the core concepts for testing range estimation for BEVs based on recorded real-world driving data. Our prototypical implementation of this concept as well as a demonstration of a small test campaign evaluation is presented in Section 5. A conclusion and an outlook on future work are given in Section 6.

3 State of the art In systems engineering a common approach to reduce the system complexity is the decomposition of the system into subsystems, components and units. In the automotive domain an established method to guide the development of systems is the V-Model [13]. The Automotive SPICE®1 process reference model adapts the traditional V-Model to the specific demands and constraints of ASE [14]. As shown in Figure 1 the V-Model consists of two legs, each dedicated to different tasks. Best practices in the Automotive SPICE® V-Model are traceability and consistency. Traceability addresses the existence of references between each phase of the V-Model. Consistency describes if the developed system complies with the defined requirements in each phase.

1

Automotive Software Process Improvement and Capability Determination

353

Validation of range estimation for electric vehicles based on recorded real-world driving data

Figure 1: Automotive SPICE® process model as a development process

The right-hand side of the process model describes the V&V of the designs and implemented software. Every phase of the V&V follows different goals. During software unit verification the correct functionality and requirements compliance of each software unit is tested. This is done by applying unit tests. Integrations tests aim to verify and validate the integrated software regarding the interaction between the software components. Software qualification tests ensure the fulfillment of the requirements. Afterwards, the software is integrated in the electronic control unit (ECU). System integration and system qualification tests ensure the correct interaction with the mechanical elements. For each phase of the V&V activity different methods are needed for securing the functionality and requirements compliance. This varies from manual code reviews to applying simulations and real-world tests. During software unit verification, unit tests are used to verify the functionality of small software units. These tests can be created automatically or manually. The fulfillment of unit tests is required for the software integration phase. For integration tests on the component level, testers utilize simulation-based approaches such as SiL [15]. SiL simulations, which allow early dynamical software tests to be done without a real ECU, such as component analysis and code coverage analysis. By emulating a real system, software component integration into an ECU can be done with Hardware-in-the-loop (HiL) [16]. Additional test drives and test campaigns are used to further validate the integration and system quality of the developed feature. The increasing system complexity raises the challenge for exhaustive test coverage [17]. Boehm described a software program as a mapping from a space of inputs into a space of outputs [18]. Therefore, the complete space of inputs has to be covered to achieve

354

Validation of range estimation for electric vehicles based on recorded real-world driving data exhaustive test coverage. Automated test case generation fails due to the inherent complexity of these approaches. To cover the complete parameter space various scenarios and situations have to be considered [19]. Covering all combinations of the real world is not feasible with automated test cases or simulations. In the field of computer vision using recorded video data is a common method to solve this challenge [20]. In the automotive field, collected real world data is primarily used for the validation of a specific feature. The recorded data includes information, which can be further exploited to cover other features under development. In previous work we suggested the use of collected real-world driving data in the whole development process [21], [22]. The suggested data-driven development approach utilizes the steadily growing pool of recorded real-world driving data for the feature development executed in a SiL environment. In [23] we introduced the Reactive-Replay approach, that enables the execution of a closed-loop feature on system level utilizing recorded real-world driving data. This increases the feature maturity in early development stages without the need for new test drives. Carefully selected test cases reduce redundancy and simulation time without losing the previously gained test coverage [24]. Range estimation algorithms are primarily tested by real test drives [25], [26]. Thousands of kilometers are collected for statistically significant evaluation. For each new software version of a range estimation algorithm new test drives need to be conducted. Additionally, they lack scalability. Another method is to use simulations instead of real test drives [27], [28]. However non-deterministic factors from environmental, traffic and driver influences make a complete and realistic simulation environment difficult and not feasible. Much effort needs to be invested to create complex models for achieving a certain degree of realism [19]. Therefore, a hybrid of simulation and real-world data is preferred, which combines the benefits of both methods without their drawbacks.

4 Concept for simulation-based verification and validation of range estimation Our goal is to combine simulation-based testing of range estimation algorithms on a large pool of existing data from real-world test drives. This allows the usage of realistic data in virtual simulations for open- and closed-loop functional elements. By using realworld driving data, the driver behavior, realistic traffic and scenarios are automatically included without the need of expensive modeling and time-consuming simulation. Accurate estimation of the remaining range relies on such realistic data. The range estimation algorithm is used in a SiL environment which enables early testing of new software versions without the need to record new real-world driving data. Another benefit of this approach is that data collected from different vehicle architectures like internal combustion engine vehicles (ICEVs) can be utilized. The missing electric energy

355

Validation of range estimation for electric vehicles based on recorded real-world driving data consumption is then simulated by a verifiable energy consumption model. This enables statistically significant evaluation over thousands of kilometers.

Figure 2: Concept image for the verification and validation of range estimation algorithms for battery electric vehicles

Figure 2 shows the 5 steps of our concept for a simulation-based verification and validation of range estimation. It starts with the data conversion and generalization of different data formats and vehicle architectures (a). Followed by a data enrichment process (b) and the simulation execution of the range estimation algorithm (c). The simulation results are then evaluated (d) and visualized (e).

4.1 Data-pool and preconditioning Our data-pool consists of already recorded test drives from earlier test campaigns. The main advantage of using these records is that they already exist in very large quantities during development. These automotive full-logs include data from different vehicles, architectures and feature versions. A variety of drivers and routes are included as well. This allows early testing based on real-world driving data without conducting expensive test campaigns during integration and system qualification test. Also, geolocationdependent behavior in different countries and regions can be evaluated at the developer’s desktop. The data-pool consists of different vehicle architectures, the respective data logs might differ in data formats as well. Real-world driving data is recorded in formats such as MDF2, DAT3 or DCM4. To manage a large and varying data-pool a generalization procedure is needed. This step transforms the given data to a generalized data format which then can be accessed regardless of its previous differences in data 2

Measurement Data Format, Vector Informatik GmbH, https://vector.com/vi_mdf_tools_en.html Automotive Data and Time-Triggered Framework, Elektrobit, https://www.elektrobit.com/products/eb-assist/adtf/ 4 Data Conservation format, ETAS, https://www.etas.com/download-center-files/products_ASCET_Software_Products/TechNote_DCM_File_Formats.pdf 3

356

Validation of range estimation for electric vehicles based on recorded real-world driving data representation or formats. Due to the heterogeneous and old data, missing and defective signals shall be substituted if possible. Up to date most collected real-world driving data is primarily acquired from ICEVs. Therefore, it is necessary to enrich the data with electric energy consumption of the powertrain for testing range estimation algorithms based on electric vehicles.

4.2 Data enrichment Enrichment of the generalized real-world driving data with additional data sources can provide beneficial information for the evaluation. This can be used for designing performance indicators, deducing influencing factors or comparing recorded data with data from external sources. Algorithms such as [11] use online traffic data to estimate the remaining range. External data sources from providers e.g. HERE offer mapping and location data as well as additional data such as information about temperature or weather conditions which are normally not included in the recorded data. Reference data provide more information about situations which are not previously recorded. This increases the degree of realism. Comparing recorded data with data gained from external sources enables further development improvements. For simulating the electric consumptions for ICEVs we used the longitudinal powertrain model as described in [29]. Testing with generated data based on generated consumptions must be treated with caution. Evaluations deducted from this kind of data is at best as good as the built model itself. Therefore, a verified energy consumption model is needed for sufficient accuracy during the feature development and testing. By generalizing recorded real-world driving data and enriching with simulated energy consumption enlarges the usable databasis - ICEVs data can now be used as well. Thus, over hundred thousand of already recorded data can be utilized to evaluate the maturity of the developed range estimation algorithm.

4.3 Simulation of range estimation algorithms By providing enriched real-world driving data the open-loop range estimation algorithm can be executed. Also, different parameterization sets for the range estimation algorithm shall be provided. Using different parameterizations allows to test and analyze the impact of parameter changes and optimizations. For our concept the range estimation algorithm in [11] was used. This algorithm requires data of the driven route, traffic and energy consumption to estimate range. The algorithm shall be executed on single test drives and on combinations of different recorded test drives. This simulates a longer test drive on which the algorithm can be adjusted to the road, traffic as well as to the driver’s behavior. The continuous output of the range estimation algorithm as well as the real-world driving data with its simulated energy consumption from the energy

357

Validation of range estimation for electric vehicles based on recorded real-world driving data model can be utilized for further evaluation and visualization of feature specific metrics. The actual remaining range of the recorded test drive can be calculated from the simulated data from the used energy model. The estimated remaining range from the range estimation algorithm and the actual remaining range can then be compared and evaluated.

4.4 Evaluation There are two approaches to evaluate the simulation results of the data-pool: File-based evaluation and global evaluation. Metrics which use recorded real-world driving data and simulation results of the range estimation algorithm are calculated for the evaluation of the range estimation algorithm’s performance. For the file-based evaluation these metrics are computed locally. Global analyses can be performed by aggregating all calculated evaluation results to achieve a summarized overview of the feature performance. Providing context-based filtering and classification of the calculated metrics to evaluate segments by e.g. street type, slope or vehicle velocity. Drill-down functionality for selected cases is provided to enable exploratory analyses of specific test drives. The modular framework concept considers a flexible, on-demand metric selection and configuration of new metrics. To add new metrics for the evaluation of range estimation algorithms shall be considered as well.

4.5 Visualization of results For visualizing the results of the range estimation algorithm, a flexible set of visualization options is required. Therefore, selecting and customizing easy to understand visualizations must be provided. As well as adding new visualizations must be possible. The evaluation results for data sets shall be visualized globally or file-based. For the global visualization the local computed evaluation results shall be aggregated to provide plots with a summarization of the performance of the range estimation algorithm. Drilldown plots for single data sets are required to further analyze influencing factors.

5 Prototype application and demonstration To demonstrate the feasibility of our suggested concept we implemented a prototypical framework in MATLAB. From the existing data-pool, a set of 10 test drives were selected for the evaluation. Each test drive followed the same route with the same vehicle but was executed by a different driver at a different time. The visualizations in Figure 3 to 6 are built modularly to best suit the needs of the current user.

358

Validation of range estimation for electric vehicles based on recorded real-world driving data

Figure 3: Aggregated range estimation evaluation based on a calculated metric

Figure 4: Single range estimation evaluation based on a calculated metric

Figure 5: Test drive route with given coloring based on a calculated metric

Figure 6: Route elevation of a test drive

Displaying the evaluation results of the exemplary test campaign. Figure 3 shows the aggregated range estimation evaluation based on a calculated metric for all drives. This in each metric is gained by the ratio 𝑐 of the estimated remaining energy 𝐸 by the energy model. time step 𝑡 and the actual remaining energy 𝐸 𝑐

(1)

The colors indicate the performance of the range estimation algorithm. If the estimated remaining energy tends to overestimate 𝑐 1 or underestimate 𝑐 1 the actual remaining energy a dark color is used. Overestimation is the most critical case because being stranded by a depleted battery must be avoided at all cost. Underestimation of the remaining energy is not desired but not critical if it happens during a drive. For specific test drives the simulation results can be displayed in another plot as shown in Figure 4. Finding influencing factors to the developed algorithm’s behavior is necessary to display additional information. Figure 5 displays a map snippet showing the driven route. To ensure discoverability where the range estimation algorithm performed poorly the same coloring as in the other visualizations is used. Additional visualization of road

359

Validation of range estimation for electric vehicles based on recorded real-world driving data specific information such as the elevation are shown in Figure 6. This allows the user to further explore different influences on the performance of the range estimation algorithm. Other visualization options and metrics shall be provided to allow more statistical evaluations. Visualization of performance distribution via plots such as heat maps help to validate the maturity of the range estimation algorithm.

6 Conclusion We opened this contribution by introducing the current trend of electrification in the automobile industry to reduce the emission of greenhouse gas. This trend still struggles with the social acceptance of battery electric vehicles mainly caused by range anxiety. To overcome this barrier accurate range estimation algorithms are required. This raises the importance of comprehensive testing methods to further increase accurate range estimation and feature maturity. The current software development process in the automotive industry based on the Automotive SPICE® process model was analyzed and described. We identified the need for new verification and validation methods to tackle the increasing complexity of feature development for exhaustive test coverage. In this paper, we proposed concepts and methods for a hybrid approach consisting of simulation and recorded real-world driving data for the verification and validation of range estimation algorithms. The reutilization of recorded real-world driving data for simulation-based execution of range estimation algorithms allows time- and cost-efficient testing for all stages and activities of range estimation algorithm development. New software versions of range estimation algorithms can be immediately tested and evaluated. By converting and generalizing different recorded real-world driving data enables large scale statistically significant testing evaluations. The potential of the generalization process to enable usage of different data formats and vehicle architectures as well as proper data management concepts will be examined in following studies. Another benefit of this method is the high degree of realism while executing the algorithm in a SiL environment. The measured data includes the driver’s behavior and driven scenarios without the need of developing complex simulations and models. By applying feature specific metrics, it allows to evaluate the range estimation algorithm regarding its global performance throughout the test data as well as with the help of additional drill-down mechanics to further explore a specific dataset. We have shown that our concept of utilizing recorded real-world driving data can be executed in an open-loop and closed-loop context for feature development. This approach was and is frequently utilized and evaluated during development of range estimation algorithms.

360

Validation of range estimation for electric vehicles based on recorded real-world driving data Future work will focus on developing more methods and tools to analyze data sets. Global and exploratory tools are required for improved feature development of range estimation algorithms. Suitable concepts for data management must be considered to handle the steadily growing pool of heterogeneous recorded real-world driving data. Scalability concepts are required for large-scale simulations over thousands of kilometers for an exhaustive test coverage. Therefore, migrating this concept to a cluster or cloud infrastructure for more computational power is necessary. Testing the adjustment of driver’s behavior and management of consumers such as heating or air conditioning to further improve the remaining range need to be considered as well. Applying data science concepts such as machine learning can improve the testing process. Advantages of the concepts are for example automated data analyses to discover new knowledge, reducing test sets by detecting novelty or anomalies in data sets and automated parameter optimization approaches.

Bibliography 1. A. Mahmoudzadeh Andwari, A. Pesiridis, S. Rajoo, R. Martinez-Botas and V. Esfahanian, “A review of Battery Electric Vehicle technology and readiness levels”, Renewable and Sustainable Energy Reviews, vol. 78, no. May, pp. 414–430, 2017. [Online]. Available: http://dx.doi.org/10.1016/j.rser.2017.03.138 2. Z. Needell, J. McNerney, M. Chang and J. Trancik, “Potential for widespread electrification of personal vehicle travel in the United States”, Nature Energy, vol. 1, p. 16112, Sep 2016. 3. A. Adepetu and S. Keshav, “The relative importance of price and driving range on electric vehicle adoption: Los Angeles case study”, Transportation, vol. 44, no. 2, pp. 353–373, 2017. 4. J. Neubauer and E. Wood, “The impact of range anxiety and home, workplace, and public charging infrastructure on simulated battery electric vehicle lifetime utility”, Journal of Power Sources, vol. 257, pp. 12–20, 2014. [Online]. Available: http://dx.doi.org/10.1016/j.jpowsour.2014.01.075 5. T. Franke, “Nachhaltige Mobilität mit begrenzten Ressourcen: Erleben und Verhalten im Umgang mit der Reichweite von Elektrofahrzeugen”, 2013. 6. T. Franke, M. Günther, M. Trantow and J. F. Krems, “Does this range suit me? Range satisfaction of battery electric vehicle users” Applied Ergonomics, vol. 65, pp. 191–199, 2017. [Online]. Available: http://dx.doi.org/10.1016/j.apergo.2017.06.013 7. L. Rodgers, D. Frey and E. Wilhelm, “Estimating an Electric Vehicle’s “Distance to Empty” Using Both Past and Future Route Information”, 2013.

361

Validation of range estimation for electric vehicles based on recorded real-world driving data 8. J. Felipe, J. C. Amarillo, J. E. Naranjo, F. Serradilla and A. Diaz, “Energy Consumption Estimation in Electric Vehicles Considering Driving Style”, IEEE Conference on Intelligent Transportation Systems, Proceedings, ITSC, vol. 2015-Octob, pp. 101–106, 2015. 9. S. Grubwinkler, M. Kugler, and M. Lienkamp, “Vehicle independent prediction of the variation of propulsion energy consumption for EVs”, pp. 1–4, 2013. 10. M. A. Masrur, D. Sutanto, V. R. Tannahill, and K. M. Muttaqi, “Future vision for reduction of range anxiety by using an improved state of charge estimation algorithm for electric vehicle batteries implemented with low-cost microcontrollers” IET Electrical Systems in Transportation, vol. 5, no. 1, pp. 24–32, 2015. [Online]. Available: http://digital-library.theiet.org/content/journals/10.1049/iet-est.2014.0013 11. S. Sautermeister, F. Ott, M. Vaillant and F. Gauterin, “Reducing Range Estimation Uncertainty with a Hybrid Powertrain Model and Online Parameter Estimation”, 2017 IEEE 20th International Conference on Intelligent Transportation Systems, ITSC, pp. 910–915, 2017. 12. J. A. Oliva, C. Weihrauch and T. Bertram, “Model-based remaining driving range prediction in electric vehicles by using particle filtering and markov chains”, World Electric Vehicle Journal, vol. 6, no. 1, pp. 204–213, 2013. 13. E. Sax, Automatisiertes Testen Eingebetteter Systeme in der Automobilindustrie. München, Hanser, Carl, 2008. 14. “Automotive SPICE Process Assessment / Reference Model, 3rd ed., VDA QMC Working Group 13 and Automotive SIG, Berlin, Germany, 7 2015” [Online]. Available: http://www.automotivespice.com/ 15. H. Shokry and M. Hinchey, “Model-based verification of embedded software”, Computer, vol. 42, no. no. 4, pp. 53–59, 2009. 16. E. Bringmann and A. Krämer, “Model-Based Testing of Automotive Systems”, 2008 International Conference on Software Testing, Verification, and Validation, pp. 485–493, 2008. [Online]. Available: http://ieeexplore.ieee.org/document/4539577/ 17. J. Mazzega, F. Köster, K. Lemmer and T. Form, “Testing of Highly Automated Driving Functions” ATZ worldwide, vol. 118, no. 10, pp. 44–48, 2016. [Online]. Available: https://doi.org/10.1007/ s38311-016-0101-x 18. B. Boehm, “Software Engineering”, IEEE Transactions on Computers, vol. C-25, no. 12, pp. 1226–1241, 1976. [Online]. Available: http://ieeexplore.ieee.org/lpdocs/ epic03/wrapper.htm?arnumber=1674590

362

Validation of range estimation for electric vehicles based on recorded real-world driving data 19. T. Helmer, L. Wang, K. Kompass, and R. Kates, “Safety Performance Assessment of Assisted and Automated Driving by Virtual Experiments: Stochastic Microscopic Traffic Simulation as Knowledge Synthesis”, IEEE Conference on Intelligent Transportation Systems, Proceedings, ITSC, vol. 2015-Octob, pp. 2019–2023, 2015. 20. M. Cordts, M. Omran, S. Ramos, T. Rehfeld, M. Enzweiler, R.Benenson, U. Franke, S. Roth, and B. Schiele, “The Cityscapes Dataset for Semantic Urban Scene Understanding” 2016. [Online]. Available: http://arxiv.org/abs/1604.01685 21. J. Bach, J. Langner, S. Otten, M. Holzapfel and E. Sax, “Data-driven development, a complementing approach for automotive systems engineering”, 2017 IEEE International Symposium on Systems Engineering, ISSE 2017 - Proceedings, 2017. 22. J. Langner, J. Bach, S. Otten, E. Sax, C. Esselborn, M. Holzäpfel and M. Eckert, “Framework for using real driving data in automotive feature development and validation”, 2017. 23. J. Bach, K.-l. Bauer, M. Holzäpfel, M. Hillenbrand and E. Sax, “Control based driving assistant functions’ test using recorded in field data”, Proc. 7. Tagung Fahrerassistenzsysteme, 2015. 24. J. Bach, J. Langner, S. Otten, E. Sax and M. Holzäpfel, “Test scenario selection for system-level verification and validation of geolocation-dependent automotive control systems”, 2017 International Conference on Engineering, Technology and Innovation (ICE/ITMC), no. July, pp. 203–210, 2017. [Online]. Available: http://ieeexplore.ieee. org/document/8279890/ 25. C. C. Rolim, G. N. Gonc¸alves, T. L. Farias and O. Rodrigues, “Impacts of Electric Vehicle Adoption on Driver Behavior and Environmental Performance”, Procedia Social and Behavioral Sciences, vol. 54, pp. 706–715, 2012. [Online]. Available: http: //linkinghub.elsevier.com/retrieve/pii/S1877042812042504 26. V. R. Tannahill, K. M. Muttaqi and D. Sutanto, “Driver alerting system using range estimation of electric vehicles in real time under dynamically varying environmental conditions”, IET Electrical Systems in Transportation, vol. 6, no. 2, pp. 107–116, 2016. [Online]. Available: http://digital-library.theiet.org/content/journals/10. 1049/iet-est.2014.0067 27. Z. Yi and P. H. Bauer, “Adaptive Multiresolution Energy Consumption Prediction for Electric Vehicles”, IEEE Transactions on Vehicular Technology, vol. 66, no. 11, pp. 10 515–10 525, 2017. 28. A. Enthaler and F. Gauterin, “Method for reducing uncertainties of predictive range estimation algorithms in electric vehicles”, 2015 IEEE 82nd Vehicular Technology Conference, VTC Fall 2015 - Proceedings, 2016.

363

Validation of range estimation for electric vehicles based on recorded real-world driving data 29. S. Sautermeister, M. Falk, B. Baker, F. Gauterin and M. Vaillant, “Influence of Measurement and Prediction Uncertainties on Range Estimation for Electric Vehicles”, IEEE Transactions on Intelligent Transportation Systems, pp. 1–12, 2017. [Online]. Available: http: //ieeexplore.ieee.org/document/8101542/

364

Smart grids in mobile fleet operations M.Sc. Dusko Mitrovic, Dipl.-Ing. Manuel Klein, Dipl.-Ing. Leonardo Uriona, M.Sc. Marius Klein, M.Sc. Kristian Binder, Dipl.-Ing. Edward Eichstetter EKU Power Drives GmbH Prof. Dr. Michael Weyrich IAS Universität Stuttgart

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 M. Bargende et al. (Hrsg.), 19. Internationales Stuttgarter Symposium, Proceedings, https://doi.org/10.1007/978-3-658-25939-6_30

365

Smart grids in mobile fleet operations

Abstract In a mobile industrial fleet application different machinery is powered by each its own reciprocating internal combustion engine. EKU Power Drives (EKU) is focusing on increasing the energy efficiency of these fleet applications. Therefore, the combustion engines are enhanced to mild hybrid units and connected to a smart grid. This semi-automated fleet setup allows to dynamically share load between units or manage the usage of equipment to streamline operational procedures. Single combustion engines are entirely replaced by electric drives and powered by the excess power of the remaining hybrid engines and the grid. The increased load on the remaining engines results in specific fuel consumption advantages. This increases the efficiency of the entire fleet. Economically the achieved fuel savings are only a portion of the gained benefits. On the fleet perspective the necessary number of engines has been reduced and the individual load on the remaining has increased. Operators’ biggest cost savings result from less different engines to maintain and store spare parts. With the power need of the grid and the capability of the electrified auxiliary engine drives, concepts for electric motors and batteries are defined. A suitable net topology concept is developed, and the independent grid is controlled by a distributed cooperative closed loop control. This enables a configuration-less connection setup between different machinery of the fleet.

1 Introduction In off-grid, high horsepower off-highway and industrial applications (i.e. tug boats, mine haul trucks, excavators), Diesel-fueled engines and power drives remain an instinctive choice because there are currently little attractive alternatives without performance trade-offs. Compared to automotive systems, industrial engines have 3 major advantages that support the implementation of alternative drive train concepts. ● Despite low load profiles and high idle times, industrial drive trains accumulate a lot of operating hours and therefore have a very high fuel consumption. Investments in more efficient propulsion solutions can pay of very quick. ● In industrial applications the operators and owners of the heavy machinery are used to provide the refueling infrastructure by themselves. Operators do not need to wait for political decisions or other companies to provide any investment in public infrastructure. ● Driveways, work cycles and ambient conditions are at least semi predictable. This enables to tailor innovative power train solutions to specific applications for an optimum ratio of operational cost savings versus initial investment.

366

Smart grids in mobile fleet operations

1.1 Mobile fleet operations In industrial applications fleet operations are very common. In harbors multiple tug boats are pushing or pulling the same large container vessel. In mining a fleet of dump trucks is working in a team with large excavators for earth moving. In oil and gas several pump units join to pump water for well stimulation purposes. While in some applications direct load sharing with energy transfer between units is feasible in others the main goal is intelligent orchestration of machinery to distribute work between several units to achieve a common performance.

1.2 Fleet topology of initial application For well stimulation a fleet of 12 to 20 pump units is accompanied by a sand supply unit, a sand-water blending unit, an acid mixing unit and other power generating units, each one carrying its own diesel engine. Fig. 1.1 illustrates a typical pump unit, which consists of a 1.678 kW Diesel engine C (Caterpillar 3512 C) connected to a 7-gear transmission D (Caterpillar TH55-E70) that drives a quintuplex plunger pump E (FMC WQ2700).

Fig. 1.1: Setup of a pump unit [1]

1.3 Recurrent operation procedures The process of pressure pumping lasts approximately 2 hours per stage. Before each stage the well is prepared for pumping operation. Generally, this takes up to 1.5 hours. If the process or logistics are interrupted, the non-operating time can last up to a few days. During all this non-operating time the engines of all fleet members are on idle. Fuel consumption of the pump unit engine with all hydraulic auxiliaries (for fan and

367

Smart grids in mobile fleet operations power end lubrication pump) applied is up to 14 gal/hour. Of the 12 installed units with a total capacity of 20 MW propulsion power in average only 9-10 units are used to deliver the required pumping power. Fig. 1.2 shows the number of working units during a well stimulation process.

Fig. 1.2: number of working units during well stimulation process [1]

Pump units have a very high usage with up to 4.500 operating hours per year, but due to overall process economics 40% to 60 % are idle time, which economically is unused investment of fuel and material wear.

2 Electrification of Equipment EKU Power Drives (EKU) develops hybrid industrial power technology at the leading edge of clean energy, electric motor technology, industrial battery assembly and management, industrial internet of things (digitized, intelligent, interconnected industrial equipment), and compliance with increasingly stringent regulation (both environmental and industry specific).

2.1 ESC – Engine Standby Controller EKU has developed a stand-alone power module (Fig. 2.1) based on Lithium-Ion battery cells that replaces traditional lead-acid batteries of heavy-duty machinery. The

368

Smart grids in mobile fleet operations power module has been designed to deliver up to 4.000 A peak power to electrically start various engines in a power range of 500 – 3.000 kW and in different mobile industrial applications. Integral part of the power module is the Engine Standby Controller. The modular controller design features various multi-purpose inputs and outputs and communication interfaces to J1939 and Ethernet networks. The application software and the Battery Management System (BMS) is integrated into the control unit. A set of 24 programmable overload-protected high side switches enable enhanced monitoring and controlling of the power outputs.

Fig. 2.1: Power Module of Engine Standby Controller

EKU integrates the ESC system in the pump units shown in 1.2 replacing the hydraulic starting system that relies on an external power source (e.g. Tractor trailer engine). The lithium ion battery sources power for the application energy management. The embedded software in the ESC controllers takes over the powertrain availability responsibility. The operator/user is not required to monitor the powertrain and can focus on the application. A typical EKU user today manages up to 20 powertrains simultaneously, and each ESC manages the powertrain independently, assuring load operation at any time. In the pump unit example, when the operator sets the transmission into break, the ESC automatically stops the engine instead of idling. The powertrain transitions to the standby state and can restart at any time with the operator’s pump power request. During the standby phase, the user is concentrated on the application. The ESC system ensures immediate powertrain availability by maintaining the operation temperature and continuous lubrication on the powertrain. Furthermore, the ESC system monitors the battery energy, to assure enough electric power to restart operation. In case that the

369

Smart grids in mobile fleet operations powertrain is exiting the OEM specifications to operate at full load, or the available battery energy is running low, the ESC system automatically restarts the engine and restores the required temperature and battery charge quickly. When the ESC system reaches the standby required values, it automatically shuts down the engine and the powertrain transitions back again into standby state. The integrated automation software enables the pump unit to be fully available at all time.

Fig. 2.2: Load Factor of pump unit with ESC

With an ESC system installed, the amount of engine idle time can be reduced up to 98% and due to the reduced non-load operation, the engine load factor increases from 24% to 49%. Fig. 2.2 shows the load factor after 800 operating hours of a pump unit with ESC system installed. Additionally, the ESC includes the required hardware for creating IOT (Internet of Things) interfaces. Industrial application usually lacks on the required interfaces for IOT implementations.

2.2 Mild Hybrid with bi-directional power The concept for the next step of electrification is the enhancement of the pump unit to a mild hybrid. Therefore, all hydraulically driven auxiliaries are powered by electric motors. The necessary power is generated from the engine’s power take offs. Fig 2.3 shows the available excess power during the well stimulation process.

370

Smart grids in mobile fleet operations

Fig. 2.3: Residual engine power during well stimulation process [1]

The generated power is stored in an additional Lithium-Ion Battery with a nominal voltage of 466 VDC. The mean residual engine power is divided to power the radiator fan (Fig. 1.1 B), the lube pump of the power-end (Fig. 1.1 E) and external systems connected to the micro grid. The nominal on-board link voltage of the pump unit is 1.000 VDC.

Fig. 2.4: Schematic setup of the pump unit board grid [1]

The mild hybrid pump units of the array are connected on the link voltage level to form a local microgrid that enables power transfers between them. The high link voltage of 1.000 V helps to reduce currents and cable diameters between the grid members. In a mobile application it is crucial to take into account the weight and practicability of cable and connector handling as the system has to be reconfigured very often and in short time.

371

Smart grids in mobile fleet operations

Fig. 2.5: Schematic setup of the fleet micro grid [1]

3 Energy Management in distributed Systems The distributed system of this presentation is designed for the power management of mobile industrial fleet applications without creation of a single point of failure. The mobile systems operate a technical process and can be removed or added during operation. The main goal is to increase the efficiency of the entire industrial fleet without creating considerable weaknesses in its reliability. The distributed system allows to dynamically share load between units of the fleet. Furthermore, it allows to use surplus energy to supply an islanded microgrid and thus operate indirectly connected systems, such as tools and command trucks in the fleet. To be able to determine the requirements of the network, attention must be paid to the distributed control system. Various control system designs were examined regarding the required network traffic, network interconnection and needed computing power.

372

Smart grids in mobile fleet operations

Fig. 3.1: Mobile industrial fleet connected to an islanded microgrid [2]

The local power grid is supplied by pump units (Fig. 3.1) which are locally distributed in a ring of 12 to 20 pump units. The units are placed approx. 2.5 m apart from each other and directly connected by short link cables to build the local power grid. Operating this islanded microgrid without a single point of failure requires that the individual units control the global average voltage and dynamically share the load of the power grid without a centralized control system. The load of the microgrid must be shared dependent on the current position and the power capacity of the individual units, while there is no exact information about losses on the power transmission lines and the positions of the loads. As this is a mobile application, the network topology changes frequently.

Fig. 3.2: Load sharing in microgrid applications [2]

373

Smart grids in mobile fleet operations In the fleet topology of the initial application, all units are equipped with the power modules of the ESC-Engine Standby Controller and power electronics for the connection to a DC-Microgrid like shown in (Fig. 2.1) and (Fig2.4). The Power Module of the ESC-Engine Standby Controller is based on an embedded system and provides the necessary connectivity and computing power for the operation of the distributed system. The Engine Standby Controller communicates with the subsystems of the pump unit using the internal j1939 CAN-Network. The communication in the fleet is based on an industrial ethernet network. A cooperative control algorithm for multi-agent systems [7] was designed for the initial application. Compared to alternatives like model predictive controllers it allows a global transient response in a distributed control system while executing classic control algorithms on the single units. No optimization algorithm must be calculated, and processing power can be saved on the embedded system. The global transient response can be calculated with the resulting Laplacian matrix from the directed communication graph of the system. It represents the interconnection of the communication in the distributed system.

Fig. 3.3: Interconnection examples of the possible directed communication graphs

The cooperative control algorithm is a solution for the tracking synchronization problem of coupled dynamical systems, where all agents are required to act as one group towards one synchronization goal [7]. This control design ensures a stable control loop even when communication failures occur between the units. It is also possible to remove or add units during operation. However, the settling time of the global transient response will be affected by changes in the interconnection of the directed communication graph. This must be considered in the design of further control systems for low-inertia microgrids.

374

Smart grids in mobile fleet operations

Fig. 3.4: Cooperative control system on the units of the mobile industrial fleet

The control system on every unit consists of a primary and secondary control loop. The primary control loop operates the power electronics output based on the reference 𝑣 and the target voltage ∆𝑣 . The target voltage ∆𝑣 is calculated by the secondary control loop. It consists of a voltage and a current control loop. The target values 𝑣̅ and ∆𝐼 are calculated by observers based on the dynamic consensus algorithm of multi-agent systems. The observers need every unit to communicate their estimated global average 𝑣 ,𝑖 periodically. By processing the esvoltage and normalized current value Ψ timations of the other units, the observer updates its own estimations. As long as the network remains in the minimum spanning tree, all observers will converge to each other.

4 Data Distribution The design of a reliable, adaptive distributed system requires a communication middleware for dynamically scalable networks without a Single-Point-Of-Failure. The Data Distribution Service is an Open-Standard middleware Specification by the Object Management Group [6] which combines a high scalability with the performance needed for real-time communication systems. It is based on the Real-Time Publish/Subscribe Protocol – RTPS from the Real-Time Industrial Ethernet Suite IEC-PAS-62030 and combines other protocols for a data-centric design of the communication network [5]. A list of vendors is published by the object management group with implementations of the specification for variety of target hardware and software applications. Considering the target hardware some aspects of the DDS-Specification are not implemented or replaced by proprietary concepts in the different vendor-specific implementations, thus making them sometimes incompatible. There are, however, vendors that are working together for a better interoperability of their implementations. For the initial application and simulation of the discussed distributed systems just some functions of the Real-

375

Smart grids in mobile fleet operations Time Publish/Subscribe Protocol are necessarily needed for the evaluation of the concept. Fig. 4.1 shows the different entities of the Real-Time Publish/Subscribe Protocol and how they are associated to each other.

Fig. 4.1: Data Distribution Service communication entities of the specification [4]

RTPS describes the physical network node as a Domain with logical network Participants. A Participant consists of one or more Publisher or Subscriber processes. Each Publisher and Subscriber shares meta-data about each other via IP multicast packets. This is done through data Reader and Writer processes, following the RTPS network protocol. The meta-data includes Topic and QoS-Information (Quality of Service) of the Publisher and Subscriber. Topics are unique data structures which specify the name and data types of the shared information. Quality-of-Service policies are used to get control and visibility into the communication behavior, including data throughput, timing and resource utilization. The communication between Publisher and Subscriber entities is established automatically if the Topic and QoS information matches. Fig. 4.2 shows an exemplary logical network design based on the data distribution service specification.

376

Smart grids in mobile fleet operations

Fig. 4.2: Network design based on the data distribution service specification

Every Participant corresponds to one pump unit of the fleet. In the initial application each Participant is represented by one Publisher and Subscriber with the two Topics Observer and Alarm. The connection is established automatically by the DDS Standard when the shared Topic and QoS-Policy information matches. The Application software with the implemented control algorithm will be informed by callback functions if a new matching Publisher is found. The voltage and current observer of the application software will automatically start to calculate their estimations according to the additional data.

5 Data management For the overall goal of optimizing the efficiency of the mobile factory, data management is a key factor to success. While the ESC system is making its contribution in a local context on each machine and the mild hybrid allows real-time cooperation by sharing power between multiple machines in the field, the next consequent step towards smart usage of resources is the combination of live and previously recorded data in a global orchestration of the machinery. Under the project name ‘Sophia’, EKU is currently developing a data management solution for monitoring and strategy planning that works together closely with the ESC system and Mild Hybrid network, but also allows integration of third-party devices and applications. Sophia is a hybrid cloud system, combining local interconnection of edge-computing devices, embedded in the control system of each machine, with a global internet- or

377

Smart grids in mobile fleet operations inter-machine communication not only using internet connectivity, but also in isolated stand-alone networks on remote temporary factory or construction sites, while on the other hand managing data in a globally synchronized infrastructure. To accomplish a gapless monitoring of the machine’s states and activity, a ‘digital twin’ of its components is hosted on the embedded computer system. This allows to keep track on all events and system data relevant to the determination of system condition, wear level and key performance indicators, as well as providing information about the machine’s capabilities and current restrictions to other systems in the network. Based on this information, distributed cooperative control systems can make automated decisions to optimize the overall system efficiency and wearout. While the ‘digital twins’ of all machines primarily ‘live’ on the embedded computers, they are asynchronously sending their current state to the global cloud back-end and receive operational parameters that influence their processes and interaction to follow a global fleet strategy, such as optimization towards maximum productivity or more towards energy-efficiency or minimized wearout.

Fig. 5.1: Sophia Cloud Topology [3]

As the data management solution allows a deep insight into the fleet’s current condition, it will help maintenance mangers to plan repairs and overhauls, give feedback of the

378

Smart grids in mobile fleet operations overall fleet performance to the managing directors and allows machine learning algorithms to transfer system knowledge from one system to another, therefore increasing the benefits with the number of similar units managed by the system. The Sophia system will further support customers by implementing common open standards, such as the XML-based WITSML and open REST APIs to interface with approved industry-specific software. The open, loosely coupled system architecture allows fast integration of additional interfaces for today’s fast changing requirements on data processing.

Bibliography 1. Klein, M.: Development of a mild hybrid power train concept for grid remote industrial engines in fleet operation, Universität Stuttgart,2018 2. Mitrovic, D.: Entwicklung eines adaptiven verteilten Systems ohne Single-Point-ofFailure, Universität Stuttgart, 2018 3. Binder, K.: Entwicklung eines Datenbackends zur Verwaltung von Mess- und Konfigurationsdaten mobiler Industrieanlagen in Form eines Hybrid-Cloud-Systems, Universität Stuttgart, 2018 4. eProsima „http://www.eprosima.com,“ [Online]. Available: http://www.eprosima.com/ index.php/resources-all/rtps. [Accessed on 22 May 2018]. 5. Object Management Group (OMG), The Real-time Publish-Subscribe Wire Protocol DDS Interoperability Wire Protocol Specification, Version 2.2, OMG, 20014. 6. Object Management Group (OMG), Data Distribution Service (DDS), Version 1.4, OMG, 2015. 7. A. N. V. D. A. L. F. Bidram, Cooperative Synchronization in Distributed Microgrid Control, Arlington, TX: Springer, 2017.

379

Infrared-based determination of the type and condition of the road surface Lakshan Tharmakularajah, Jakob Döring, Karl-Ludwig Krieger Institute of Electrodynamics and Microelectronics (ITEM.ae) University of Bremen

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 M. Bargende et al. (Hrsg.), 19. Internationales Stuttgarter Symposium, Proceedings, https://doi.org/10.1007/978-3-658-25939-6_31

381

Infrared-based determination of the type and condition of the road surface

Abstract The knowledge of different road conditions is a significant factor in determining their impact on fully automated driving. As part of the research project "SeeRoad" funded by the German Federal Ministry for Economic Affairs and Energy (BMWi), a system with various technological approaches is being developed, in order to estimate road conditions. One of these approaches determines the condition of road surfaces as well as the distance between the vehicle or sensor and the road surface, based on infrared (IR). In order to distinguish between a wet and a dry surface, the water film height (WFH) is measured by infrared rays. We present a procedure for determining the type and condition of the road surface with the help of infrared radiation, which forms a basis for the multiple sensor system. This procedure is tested in a laboratory setup that is suitable for reproducible measurements. A new approach to detect road surface wetness by using a combination of infrared sensors, which work with different principles, is presented. The first principle is based on the intensity measurement of a reflected IR signal to determine the distance. Here the intensity depends on the material surface. At a constant distance, different intensities are measured for the absorbed and reflected signal, depending on the color and texture of a material. In the second principle, the transit time of the emitted and reflected beam is measured, which is independent of the intensity. Thus, the combination of both methods allows the distance and type of road surface to be calculated.

1 Introduction Common causes of personal injury in traffic are due to poor road conditions [1]. Between the years 2014 to 2017, slippery roads due to rain account for around 20 % of all accidents in Germany. Another 11-18 % are due to ice, and the trend is rising. Because of slippery roads, drivers regularly lose control of their vehicle. Accordingly, there is an increasing demand for technologies that reliably detect critical road conditions of this type. Moreover, the increase of automation in the automotive sector, environment detection is becoming increasingly important, including the knowledge of the road conditions. This is required for autonomous driving in weather-related driving situations. The environmental sensors, which are already installed in vehicles, are not sufficient, because they can only determine the road condition indirectly. For example, the rainlight sensor is only applicable on a wet road surface during rainfall. However, if the road is wet but it does not rain, it can not be derived on a wet road. There is a approach of indirect road condition monitoring with the use of capacitive sensors in the wheel arch liner [2]. To monitor the road surface directly visible image sensors, infrared image sensors, and millimeter-wave radar sensors are used [3] [4]. These sensors are expensive and some are not suitable for vehicles. Therefore, the motivation is to build a cost-

382

Infrared-based determination of the type and condition of the road surface effective solution with commercial and in-home developed sensors, to determine the road type and conditions.

2 Infrared sensor system Infrared sensors are commonly used for measuring distances as for example in robotics for obstacle avoidance. These low-priced sensors depend on the reflectance properties of the object surfaces but they have a non-linear characteristic. Therefore, the properties of the surface must be known. In this case, the knowledge about how the infrared light is scattered, reflected and absorbed by the surface is required to interpret the sensor output as a distance measure [5]. Generally, an IR sensor consists of a transmitter and receiver. The sensor uses a receiving photodiode to measure the intensity of the reflected light (emitted by the transmitter) to estimate the distance from an object [6]. Here, the intensity depends on the material surface. At a constant distance, different intensities are measured for the absorbed and reflected signal, depending on the color and texture of a material. To measure the distance to an object, regardless of the surface properties, time-of-flight (TOF) is used. In [7] the direct time-of-flight method is explained. This method uses a narrow pulsed laser, with a time-to-digital converter (TDC) measuring the difference in time between transmission and first photon reception. Thus, the combination of both methods allows the distance and type of road surface to be calculated. The first step is to determine the distance using the time-of-flight method and then determine the surface in combination with the intensity sensor.

2.1 Sensing principle of intensity measurements For the intensity measurements, emitted light is reflected on the surface and recorded by a photodiode [4]. The reflection and absorption depends on the surface of the medium. It can be seen from the beam path shown in Figure 1 that the emitted light beam 𝐼𝐼0 splits up into two unknown amounts upon the impact on the water film surface. On the one hand, 𝐼𝐼0 divides into the reflected beam 𝐼𝐼1 and on the other hand into the transmitted beam 𝐼𝐼2 . The transmitted beam 𝐼𝐼2 is absorbed by the water depending on the wavelength. At the surface of the ground 𝐼𝐼3 is reflected again. The reflected beam 𝐼𝐼3 is again absorbed by the water. During the transition from water to air, the beam 𝐼𝐼3 is refracted. The refracted beam 𝐼𝐼4 finally hits the sensitive area of the photodiode. Thus, the measured light intensity at the receiving photodiode is dependent on the reflected beams 𝐼𝐼1 and 𝐼𝐼4 . The reflection at the water surface depends on the wave-dependent refractive index 𝑛𝑛(𝜆𝜆). In addition to the rays shown, multiple beams appear due to multiple scattering and multiple reflections, which are not shown in the schematic diagram in Figure 1.

383

Infrared-based determination of the type and condition of the road surface Emitter

Receiver ‫ܫ‬଴

‫ܫ‬ଵ ‫ܫ‬ସ

Air

WFH ‫ܫ‬ଶ

Water

‫ܫ‬ଷ

Ground Figure 1: Sensor model and beam path, adapted from [4]

2.2 Sensing principle of time of flight measurements The time of flight measurement calculates the required time for a light pulse from the transmitter to the measurement target and back to the receiver. This method allows to measure the distance independent of target reflectance [8]. The emitted light pulse moves at a constant speed 𝑐𝑐 = 299,792,458 m/s in air. The returned distance 𝑠𝑠 can be calculated by (1)

𝑠𝑠 = 𝑐𝑐 ∙ ∆𝑡𝑡,

where the required time is represented by 𝑡𝑡. Since here the way s of the light is set back and forth twice, for the distance 𝑟𝑟 half of the way must be considered. The distance is determined by 𝑟𝑟 =

𝑠𝑠

2

=

𝑐𝑐 ∙ ∆𝑡𝑡 2

,

(2)

where ∆𝑡𝑡 is the time-of-flight.

3 Experimental study In this section, the setup of the measurement environment, the measured experiments and the results are presented. First, a test bench is set up, which reproducibly generates water film heights with an accuracy of 1 mm with the support of a strain gage balance. Subsequently, measurements on dry and wet surfaces are carried out. Finally, the results of the measurements are presented.

384

Infrared-based determination of the type and condition of the road surface

3.1 Measurement setup In order to carry out the measuring experiments reproducibly, a test bench is set up. The concept depicted in Figure 2 consists of four basic components: the pump and the nozzle, the surface, the linear guide and the strain gage balance. The pump sprays a fine mist vertically through the nozzle onto the surface of the substrate. By spraying, the formation of droplets can be counteracted to produce a homogeneous film of water on the surface. The linear guide allows the nozzle and the sensor to be aligned exactly above the surface by a vertical and horizontal movement. Water tank

Vertical movement

Linear guide Horizontal movement

Pump

Nozzle

Water

Surface Balance

Figure 2: Concept of the test bench

To ensure reproducibility for the respective water film height, the surface is placed on a strain gage balance. The strain gage balance measures the weight, while the pump applies the fine spray to the substrate. According to Equation 3, the given density 𝑝𝑝 of the water allows the weight 𝑚𝑚 to be calculated into the volume 𝑚𝑚

𝑉𝑉 = 𝑝𝑝 .

(3)

385

Infrared-based determination of the type and condition of the road surface By the calculated volume 𝑉𝑉 and a given surface of a tile 𝑎𝑎 ∙ 𝑏𝑏 the water film height 𝑉𝑉

𝑊𝑊𝑊𝑊𝑊𝑊 = 𝑎𝑎²

(4)

can be determined. In this case 𝑎𝑎 = 𝑏𝑏. Therefore, the area of the surface is 𝑎𝑎².

This measuring principle enables an indirect, non-contact measurement of the water film height. Contact-based measurement principles would change and falsify the water film height due to the surface tension and capillary forces.

Linear guide White tile

Black tile

Concrete

Asphalt

Sensors

Nozzle Surface

(a) Test bench

(b) Surfaces

Figure 3: Experimental setup

Therefore, the minimum water film height of this test bench is 1 mm. The test bench and the sensors are controlled with a specific graphical user interface based on Matlab. The linear guides can be approached exactly to a millimeter in vertical and horizontal direction. The two sensors are positioned at the same height to have the same distance to the surface. A border was attached to the edges, so that the water does not flow from the surface, as seen in Figure 3 (a). The four different surfaces are shown in Figure 3 (b).

386

Infrared-based determination of the type and condition of the road surface

3.2 Experiments For the intensity measurements, an in-home developed sensor with a photodiode and an LED with a wavelength of 920 nm is used. To measure the time-of-flight a commercial sensor VL6180X is used [8]. This sensor is specified for a range from 0 cm to above 10 cm. The emitted light beam has a wavelength of 820 nm. To differentiate the surfaces, they are measured with both sensors at different distances of 50 mm to 150 mm in steps of 10 mm. The measurements with a wet surface are made at a distance of 50 mm. The first wet measurement has a constant distance of 50 mm between sensor and the surface. The second measurement has a constant distance of 50 mm between sensor and the water film height. The water film height is measured from 1 mm to 5 mm in steps of 1 mm. In order to avoid disturbances and influences from extraneous light, the measuring tests are carried out in a shielded area.

3.3 Results Four different surfaces, white tile, black tile, concrete and asphalt, are investigated for the measurements. In the first measurement, the two sensors are positioned at a distance of 50 mm to 150 mm to the dry surfaces. Subsequently, the measurements are carried out at 50 mm with different water film heights of 1 mm to 5 mm.

3.3.1 Measurements with dry surfaces The sensors are positioned at a distance of 50 mm to 150 mm from the dry surface. In Figure 4 the line graph presents the measurements with the intensity sensor of four different surfaces. The white tile reflects the emitted light the most and the black tile the lowest. The voltage difference between the white and black tile is about 0.88 V. Because the concrete is lighter than the asphalt, the reflection of the infrared rays is stronger. The asphalt is lighter than the black tile because it contains some white and grey stones. Regardless of the background, the intensity decreases exponentially with the growing distance. From a distance of above 130 mm, there is barely a significant difference between the three surfaces concrete, asphalt and tile. The measurement results of the VL6180X sensor show that the sensor measures the distance, regardless of the surface of the substrate. In Figure 5 it can be seen that the measured data from 100 mm have larger deviations. This is due to the fact the VL6180X is designed to measure in a range of 0 mm to 100 mm [8]. With increasing distance, the distance value of the sensor also increases. Furthermore, the measured data show that the sensor has a linear behavior.

387

Infrared-based determination of the type and condition of the road surface 1.2 White tile

1

Concrete Asphalt Black tile

Voltage [V]

0.8 0.6 0.4 0.2 0 40

50

60

70

80

90

100

110

120

130

140

150

160

Distance [cm]

Figure 4: Intensity measurement on dry surfaces from 50 mm to 150 mm with the in-home developed sensor

The VL6180X sensor provides nearly identical distance values across the different surfaces. Due to the different absorption of the infrared rays and the distance value of the VL6180X, the surface can be determined. Therefore, the results show that the four dry surfaces can be differentiated by a combination of both sensors. 160 White tile

140

Concrete Asphalt

Distance [mm]

120

Black tile

100 80 60 40 40

50

60

70

80

90

100

110

120

130

140

150

160

Distance [mm]

Figure 5: Distance measurement on dry surfaces from 50 mm to 150 mm with the VL6180X

388

Infrared-based determination of the type and condition of the road surface From a distance of about 110 mm, the intensity values of the intensity sensor are very close to each other, so that from this height a distinction of the surfaces is difficult. The difference between asphalt and concrete at a distance of 110 mm is approximately 5 mV, which is within the standard deviation. Therefore, a distinction between the two surfaces over 110 mm is not possible.

3.3.2 Measurements with wet surfaces In these measurements, first the distance to the surface was kept constant at 50 mm and then the distance to the water film height was kept constant at 50 mm. This helps to investigate the influence of water on the different surfaces with the two measuring principles. In Figure 6 the solid lines indicate the constant distance between the sensor and the surfaces. The dashed lines indicate the constant distance between the sensor and the water film height. When the surface gets wet (from 0 mm to 1 mm water film height), the intensity decreases sharply. The reason is that some of the emitted infrared rays are absorbed by the water. Afterwards (1 mm to 5 mm water film height), the intensity decreases almost linearly. Here it can be seen that the intensity decreases with increasing water film height. Due to the constant distance of 50 mm to the water film height, the distance to the surface increases with the water film height. Due to the higher distance to the surface, the intensity decreases more. White tile surface

White tile WFH

Concrete surface

Concrete tile WFH

Asphalt surface

Asphalt WFH

Black tile surface

Black tile WFH

1.2

Voltage [V]

1 0.8 0.6 0.4 0.2 0

0

1

2

3

4

5

Water film height [mm]

Figure 6: Intensity measurement on wet surfaces with a water film height from 1 mm to 5 mm

The measurements with the VL6180X show that the water on the surfaces has little or no influence. The first line chart in Figure 7 shows that the sensor measures almost

389

Infrared-based determination of the type and condition of the road surface constant the 50 mm to the surface. The second line chart shows that the sensor is still measuring the distance from sensor to the surface and not to the water film height. Over 3 mm water film height the measured distance value decreases minimally. The reason could be that the light pulse is reflected at the water surface. For the detection of wetness on the surface this is a support. Normally with a smaller distance the intensity has to increase. However, the intensity decreases with increasing water film height. As the water film height increases, the accuracy of the sensor decreases. Nevertheless, this inaccuracy is negligible for determining the distance between the sensor and the surface. White tile

Concrete

Asphalt

Black tile

56 Distance [mm]

54 52 50 48 0

1

2

3

4

5

4

5

Water film height [mm]

56 Distance [mm]

54 52 50 48 0

1

2

3

Water film height [mm]

Figure 7: Measurement with the VL6180X on wet surfaces

Summary and Outlook Our intensity sensor emits infrared light with a wavelength of 920 nm and measures the intensity of the reflected light with the help of a photodiode. The time-of-flight sensor emits a light pulse with a wavelength of 820 nm and measures the time until the light is reflected back. The combination of the two sensor principles makes it possible to differentiate the surfaces in the dry state. Furthermore, it is possible to distinguish the surfaces between a dry and a wet state. The measurements have shown that the VL6180X measures the distance to the surface regardless of water on a surface. This is because the VL6180X works with a light pulse at a wavelength of 820 nm. The transmission of the infrared rays at a wavelength of 820 nm in water is high. The reflection

390

Infrared-based determination of the type and condition of the road surface of infrared light with a wavelength of 920 nm on a wet white tile is higher than the other three dry surfaces. In order to allow a distinction between the four surfaces and the two surface states, an intensity sensor with a higher absorption of water is needed. A distinction from the surfaces black tile, asphalt and concrete and their two states wet and dry is possible with the combination of the two presented sensor principles. For future use, a real-time system with the two presented sensors will be designed to determine the road condition. For this case, the distance between sensor system and road surface has to be optimized. Furthermore, vehicle tests and investigations with ice and snow on the road surface will be carried out.

Bibliography [1] Statistisches Bundesamt (Destatis), "Fachserie 8 Reihe 7 Verkehr Vekehrsunfälle," Wiesbaden, 2018. [2] J. Döring, L. Tharmakularajah, J. Happel and K.-L. Krieger, "A novel approach for road surface wetness detection with planar capacitive sensors," J. Sens. Sens. Syst. 8, pp. 57-66, https://doi.org/10.5194/jsss-8-57-2019, 21 January 2019. [3] R. Kurata, H. Watanabe, M. Tohno, T. Ishii and H. Oouchi, "Evaluation of the Detection Characteristics of Road Sensors under Poor-visibility Conditions," in IEEE Intelligent Vehicles Symposium , Parma, 2004. [4] F. Holzwarth, Entwicklung eines Verfahrens zur berührungslosen Messung der Wasserfilmdicke auf Fahrbahnen, Dissertation, Universität Stuttgart: IKFF, 1996. [5] P. Novotny and N. Ferrie, "Using infrared sensor and the Phong illumination model to measure distance," in International Conference on Robotics and Automation, Detroit, Michigan, 1999. [6] L. Korba, S. Elgazzar and T. Welch, "Active Infrared Sensors for Mobile Robots," in IEEE Transactions on Instrumentation and Measurement, 1994. [7] D. P. Baxter, "Proximity sensor having an array of single photonavalanche dodes and circuitry for switching off llumination source and assocated method, computer readable medium and firmware". GB/Edinburgh Patent US 8,610,043 B2, 17 12 2013. [8] STMicroelectronics, "VL6180X Proximity and ambient light sensing (ALS) module," Datasheet, Genf, 2016.

391

Essential predictive information for high fuel efficiency and local emission free driving with PHEVs M.Sc. Tobias Schürmann, Dr.-Ing. Daniel Görke, Dipl.-Ing. Stefan Schmiedler Daimler AG Dipl.-Ing. Tobias Gödecke, Prof. Dr.-Ing. Kai André Böhm Esslingen University of Applied Sciences Prof. Dr.-Ing. Michael Bargende University of Stuttgart

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 M. Bargende et al. (Hrsg.), 19. Internationales Stuttgarter Symposium, Proceedings, https://doi.org/10.1007/978-3-658-25939-6_32

393

Essential predictive information for high fuel efficiency and local emission free …

Abstract An intelligent selection of the operating modes can improve the fuel efficiency of plugin hybrid electric vehicles (PHEVs) and allow them to drive local emission free. In order to align these goals and hence to improve the mobility especially with air pollution problems in urban areas, predictive information about future driving situations is necessary. To achieve this target and furthermore to design and calibrate predictive control strategies accordingly, the sensitivity of predictive information on the fuel efficiency is analyzed in the presented simulation study. Traffic simulations are used which enable reproducible driving situations regarding traffic, traffic control and driving characteristics by their parameterizable settings. By calculating fuel optimal strategies with Dynamic Programming (DP) for a PHEV in P2 topology, the impact of predictive information about future driving situations on the fuel efficiency is evaluated. The results show which driving situations are suitable for charging and discharging and assess the efficiency of local emission free driving by comparing the fuel savings to the costs of the electric energy demand.

1 Introduction Determining efficient control strategies for hybrid electric vehicles (HEVs) is one of the most researched topics in the field of alternative powertrains, lately. The possibility to understand the powertrain control of HEVs mathematically as an optimization problem leads to various approaches, see [1, 2, 3] for further reading. Particularly implementations of DP are highly researched because of its ability to determine the global optimum and handle constrained, mixed-integer and non-linear optimal control problems. However, in order to calculate the global optimum all information, such as the exact driving cycle and boundary conditions, have to be known a priori. Since this is only possible in simulation studies, the global optimum cannot be determined with real-time capable applications and thus some of the fuel saving potential remains unexploited. Nonetheless, global optimal control strategies are useful to benchmark other strategies and to derive general correlations for designing control strategies [4]. To be able to cover various driving situations beyond a certain driving cycle, real-time capable operating strategies are determined on the basis of stochastic distributions of different driving cycles and realworld measurements [1, 5, 6]. Thereby the stochastic distributions are used to find the most probable next vehicle states and to calculate the strategy via an adapted dynamic programming implementation. With an appropriate prediction horizon the implementation reveals its disadvantageous. The prediction is no longer sufficiently accurate and the computational effort too high for being real-time capable. Even though it is the right approach to develop such strategies with real world measurements, it is not ensured that

394

Essential predictive information for high fuel efficiency and local emission free … the stochastic distributions used for the calibration represent the usage of the HEV afterwards. Thus, a large part of the efficiency improvements can be lost. Also strategies based on learning and self-adapting algorithms suffer from being dependent on historical data. For improving real-time capable strategies significantly, the accurate prediction of future driving situations is decisive. By predicting the essential information accurately, optimal strategies can be approximated much better and more robustly. As investigated by Cummings, Bradley and Asher [7] even a controller with a slightly erroneous prediction of the traffic respectively the traffic control leads to an improvement of the fuel efficiency compared to a baseline controller without prediction. However, an imprecise prediction of the route causes a noticeable increase in the fuel consumption. Fu, Ozguner, Tulpule and Marano [8] analyzed noise in the velocity prediction. Their results show that even noisy velocity predictions yield to close-to-optimal fuel economy and that representative driving cycles can be used instead of more accurate predictions. Still, they state that further analyzes are required for the confirmation of these findings. Also analyzed for a HEV, Romijn, Donkers, Kessels and Weiland [9] state that prediction horizons are necessary, which are long enough to meet a target state-of-charge (SOC) of the high-voltage battery. Based on the simplified assumption that the load of the drive cycle is known and can be predicted by the controller, they achieve higher fuel efficiencies with prediction. Since PHEVs can use a significant higher energy content of their high-voltage battery, small errors in the prediction can have a much greater effect on the fuel efficiency if the controller is not sufficiently robust. As analyzed by Zhang and Vahidi [10], the fuel economy of PHEVs can substantially be enhanced with only partial preview. Their online controller based on an implementation of the equivalent consumption minimization strategy (ECMS) performs close to the optimum with only the knowledge of the trip length. As their further analyses show, this is true for level roads and the knowledge of the altitude is additionally valuable. Based on these findings this paper provides an in-depth understanding of the influences of predictive information about future driving situations on fuel efficient control of PHEVs. The influences studied in here are twofold: Static influences are the trip length, the altitude and the speed limits of the route. The traffic, the traffic control and the driver characteristics belong to the dynamic ones. As already pointed out in the previous analysis about the electric energy demand for local emission free driving [11], especially the dynamic influences can only be investigated appropriately in a simulation study as it is done in the following part. The analyses reveal the necessary information for achieving a high fuel efficiency in combination with local emission free driving. The paper is organized as follows: Section 2 presents the proposed approach with an indepth understanding of the simulation environment. The sensitivity of the predictive information is analyzed in section 3. Section 4 summarizes the results and gives an outlook.

395

Essential predictive information for high fuel efficiency and local emission free …

2 Proposed Approach A virtual test environment is built up consisting of the traffic simulation SUMO [12] and a backward longitudinal powertrain simulation of a PHEV. In the traffic simulation the amount of other vehicles, the traffic control and the driving characteristics can be set up for specific routes. The routes are chosen on maps which can be loaded and converted from OpenStreetMap [13]. For the speed limits of 30 km/h, 50 km/h, 70 km/h, 100 km/h and 130 km/h own maps and routes are defined. Measurements are performed in these speed limits which are characterized by a moderately high target speed and medium accelerations and decelerations. The traffic simulation is enhanced by a new driver model which learns and represents measured driving behaviors, accurately. The driver model is trained by these measurements. In the traffic simulation the traffic is varied in certain steps from no traffic to traffic jam. Additionally, different lengths of the green phase of the traffic control are analyzed. Therefore the green phase index GPI is defined as follows: 𝐺𝑃𝐼 ≔ 𝑛 ∈ ℤ | 𝑛

1∧𝑛

9

(1)

Small values of the GPI stand for a high interference through red lights and hence lead to many decelerations and accelerations. Thus, it is possible to create parameterizable driving situations which can be described by their settings regarding the traffic, the traffic control and the driving characteristics. These settings represent possible predictive information about future driving situations. The traffic simulation outputs are velocity profiles for which the fuel optimal control strategies are calculated via DP as it is illustrated in figure 1.

Figure 1: The virtual test environment

The implementation of DP uses the backward longitudinal powertrain model of a Mercedes-Benz E-Class with a 155 kW 2.0 liter turbocharged 4-cylinder gasoline engine and a 90 kW electric motor in P2 topology. The battery net capacity is 9 kWh. The simulation model is based on static efficiency maps of the powertrain components and validated with measured data. By calculating fuel optimal control strategies the costs for engine starts are considered by an additional state variable.

396

Essential predictive information for high fuel efficiency and local emission free …

3 High fuel efficiency and local emission free driving Firstly, the focus is on driving situations which are efficient for charging the high-voltage battery via the combustion engine. Secondly, situations which are suitable for charge depleting are analyzed. Afterwards, the efficiency of local emission free driving is investigated.

3.1 Charge Increasing By choosing the operation modes of a HEV in an intelligent manner, the emissions in urban areas can be reduced. Further reductions are only possible by using stored electric energy. Therefore, it can be efficient to charge the battery via load point shifting in previous rural or highway driving situations. As a representative of such situations the charging in the speed limit 100 km/h is analyzed in the following. Therefore different charging gradients are considered. These multiplied with the trip length indicate by how much the target SOC as a boundary condition of the global optimal strategy is higher than the start SOC. “CH03” indicates a gradient of 0.3% SOC per kilometer. For comparability reason the fuel consumption of the optimal charging strategy is subtracted with the fuel consumption for the optimal charge sustaining strategy and afterwards normalized to 1% SOC per kilometer. The results are illustrated in figure 2.

Figure 2: Fuel expenses for different charging gradients for a speed limit of 100 km/h on level roads

397

Essential predictive information for high fuel efficiency and local emission free … In the left subplot the brightest graph shows the normalized fuel costs of the charging gradient “CH03” in case of no traffic over the length of the green phases. The other gradients considered are displayed darker. A GPI of 9 indicates a long green phase and a GPI of 1 a short green phase. Every subplot shows a different traffic situation with increasing traffic to the right. As can be seen, lower charging gradients generally have lower normalized fuel costs and thus are more efficient for charging. Depending on the driving situation up to 3% fuel can be saved by operating the vehicle with a charging gradient of 0.3 % compared to 0.9 %. The reason for this correlation becomes evident by analyzing the Willans lines of the engine and its slopes, see figure 3. Starting on a certain basic load, the expenses for load point shifting and their slope are plotted for different engine speeds nICE.

Figure 3: Willans lines of the engine and its slopes for a certain basic torque

As shown in figure 3, up from a certain torque the slope is relatively steep. As long as the control strategy is able to avoid these operation areas, efficient operating is possible. In cases which need more electric energy for efficient driving due to exemplarily traffic, these areas are used for load point shifting and thus the fuel costs increases significantly, see the right subplot of figure 1. In these considerations, it has to be taken into account that with lower charging gradients longer distances are needed for the same charged energy content. The reasons for the slight variations and the not precise proportionality of the normalized fuel costs in the first four subplots of figure 1 are twofold: Firstly, the traffic simulation is to a certain extent random and secondly, the calculation of the fuel optimal strategies considers engine start costs. Up to high traffic there are only small differences in the costs for charging the high voltage battery. In real time capable implementations these differences cannot be represented. On the one hand these are the results of global optimal control strategies which are not real-time capable and on the other hand noises

398

Essential predictive information for high fuel efficiency and local emission free … in the prediction are inevitable in reality. Only in cases of traffic jam the costs for charging the high voltage battery are noticeable higher as compared to other driving situations. For understanding the correlation regarding the traffic, the load points of two driving situations are examined. In case of no traffic and GPI 9 the velocity profile and the load points are illustrated in figure 4.

Figure 4: Global optimal control strategy in case of CH09, no traffic and GPI 9

As can be seen, there are only a few load points which are driven electrically. Most of the basic load points (bright gray) are shifted towards higher torque in highly efficient areas (black). Additionally, there are only some high torque requests while accelerating which are shifted towards lower torque and only few situations which are driven electrically (dark gray). Overall it is revealed that this driving situation is suitable for charging the high voltage battery efficiently. In comparison to this the global optimal charging strategy in traffic jam with a GPI of 3 is shown in figure 5. The engine map with the load points reveals that there are a lot more situations for efficient electric driving even though the load points have to be shifted towards higher, in some cases less efficient areas for achieving the desired target SOC. Additionally, there are a lot more

399

Essential predictive information for high fuel efficiency and local emission free … accelerations which need electric energy for reducing the torque of the combustion engine. In total, charging is not efficient in this driving situation. As shown previously, other driving situations without such heavy traffic are more suitable.

Figure 5: Global optimal control strategy in case of CH09, traffic jam and GPI 3

A significant impact on the basic load and hence the efficiency has the altitude. Therefore, constant slopes of 2 % and 4 % are analyzed in two different cases. In the first case a constant positive slope leads to an overall altitude difference, “w/ AD”. In the second case there is a turning point in the middle of the trip and afterwards negative constant slope. Thus, there is no overall altitude difference “w/o AD” in the end of the trip. The investigations are based on the moderately charging gradient of 0.6 % SOC per kilometer. The results are displayed in figure 6 and show that driving situations with constant slopes of 4 % are expensive for charging. The basic load points are in areas which are efficient to operate the combustion engine but which are not suited for further load point shifting to higher torque. Even a slope of 2 % without overall altitude difference leads in some cases to significant higher expenses. The reason is that half of the trip is downhill and thus the combustion engine is turned off for electric driving. The

400

Essential predictive information for high fuel efficiency and local emission free … electric motor is used for regeneration and to overcome the low driving resistances. Therefore, only half of the trip can be used for charging via load point shifting. The resulting increase of the charging gradient causes higher costs. Up to high traffic the increasingly frequent accelerations lead to the use of the combustion engine even though in downhill driving. While using the combustion engine in such accelerations even though it is going downhill charging becomes more efficient.

Figure 6: Fuel expenses for the charging gradient CH06 in the speed limit of 100 km/h with the influence of slopes

The shown correlations of the influence of traffic and the charging gradients on the efficiency are also apparent in other speed limits as it is illustrated in figure 7. Therefore, the normalized fuel costs of representative driving cycles of the speed limits 30, 50, 70, 100 and 130 km/h are displayed for the situations “no traffic” and “high traffic”. Even though these are only representatives, the clear correlations are evident. The efficiency is higher at low charging gradients and in driving situations without traffic in high speed limits on level roads. At rising charging gradients the advantage of certain, suitable driving situations decreases. Thus the assumption of the suitability of urban and highway driving situations for charging is confirmed bearing in mind the traffic situation.

401

Essential predictive information for high fuel efficiency and local emission free …

Figure 7: Fuel expenses for different charging gradient CH06 in different speed limits on level roads

The reason for these findings are given earlier by presenting the Willans lines of the engine in figure 2. At higher speeds the base load is higher mainly due to the increasing air resistance and thus the operation as well as the load point shifting are more expensive as can be seen at the slopes of the Willans lines.

3.2 Charge Depleting For analyzing which driving situations are suitable for using electric energy in hybrid mode, the fuel savings of charge depleting in comparison to charge sustaining are illustrated in figure 8. The calculation of the fuel savings in charge depleting is analogous to the fuel expenses for charging. Thereby, the charge depleting gradient multiplied with the trip length states the negative SOC difference between the start and the target SOC of the global optimal control strategy. The fuel consumption in charge sustaining is offset against the ones in charge depleting and then normalized to -1 % SOC per kilometer.

402

Essential predictive information for high fuel efficiency and local emission free … Each of the considered charge depleting gradients are illustrated in one subplot in which the normalized fuel savings of the different speed limits are shown. The analyzed driving situations are the same as in figure 7. As it can be seen, most of the fuel can be saved in the lower speed limits 30 km/h and 50 km/h. Additionally, driving situations with traffic are more suitable for charge depleting. In conclusion these findings confirm that the usage of the stored electric energy is particularly efficient if the driving situation is characterized by many load points in low torque and speed. This is on the one hand due to the low efficiency of the combustion engine in these operating points and on the other hand due to the low losses of the electric system in these areas.

Figure 8: Fuel expenses for different charging gradients in different speed limits on level roads

By comparing these results with those presented in figure 6 the operation in charge and charge depleting is in some combinations of driving situations more efficient than a charge sustaining strategy within certain SOC thresholds. In these cases the savings by using electric energy in one driving situations are greater than the costs for charging in the other one which is due to their load points. In this investigation there are still some accelerations operated in hybrid mode, even at charge depleting gradients of -0.9 % SOC per kilometer. Thus in the following chapter the operation in purely electric mode is investigated.

403

Essential predictive information for high fuel efficiency and local emission free …

3.3 Purely Electric Driving For reducing the air pollution in urban areas, PHEV offer the possibility to drive local emission free. In order to know whether this operating strategy is efficient, typical urban driving situations at a speed limit of 50 km/h are further investigated. Therefore the electric energy demand for purely electric driving is calculated while considering the maximum electric power of 90 kW. The influences of traffic, traffic control and driving characteristics on the electric energy demand are analyzed in a previous study [11]. In order to evaluate the efficiency of local emission free driving an equivalence factor is used for the conversion of the electric energy demand into fuel consumption. This equivalence factor indicates the expenses of electric energy in fuel mass for a specific global optimal control strategy with its load points, see [4] for further reading. The resulting normalized fuel savings for electric driving including the influence of the slope are presented in figure 9.

Figure 9: Fuel savings for local emission free driving in speed limit 50 km/h

Therefore, the same velocity profiles are simulated with the different constant gradients of 0, 2 and 4 % slope. In addition, driving situations are calculated with a turning point of the slope in the middle of the trip. After this point the slope value is set to be negative

404

Essential predictive information for high fuel efficiency and local emission free … in order that no overall altitude difference arises. The fuel savings are plotted for different GPI in four subplots which each represent another traffic situation. By converting the energy demand with the equivalence factor of the charge sustaining strategy the results show high efficiency for electric driving in high traffic. In addition driving situations in minor traffic and without red lights on level roads are suited. In contrast, many accelerations and decelerations without any limitations due to leading vehicles are not so suitable for local emission driving. On one hand the loads are often too high while accelerating, and while decelerating the driver brakes often too sharp for efficient regeneration. Mostly independent of the traffic and traffic control, slope leads to significant decreases of the efficiency. Even 2 % without an overall altitude difference has a higher influence than traffic or traffic control in extreme situations. The reason are the resulting shift towards higher torque. In direct comparison to the fuel expenses for charging in suitable rural and highway driving situations at a speed limit of 100 km/h, there are only a few combinations of driving situations in which charging for local emission free driving is efficient, see figure 10. The different fuel expenses for charging at a speed limit of 100 km/h are presented in black, the grey ones the converted fuel savings for electric driving at a speed limit of 50 km/h at different slopes with overall altitude differences.

Figure 10: Fuel savings for local emission free driving in speed limit 50 km/h compared to the fuel expenses for charging at a speed limit of 100 km/h

405

Essential predictive information for high fuel efficiency and local emission free … It has to bear in mind that the previous analyses investigate the fuel efficiency and therefore the CO2 emissions. By considering all emissions, figure 10 shows further driving situations which are suited for electric driving. Especially when considering common driving scenarios consisting of long highway driving followed by a short urban trip, low charging gradients would only increase the fuel consumption slightly. Nonetheless, as shown mainly the altitude of the urban trip should be considered while designing and calibrating such a predictive control strategy.

4 Conclusion In the presented simulation study the influences of predictive information about future driving situations on the fuel efficiency and local emission free driving are analyzed. For realizing these goals the speed limit, the altitude and the traffic are essential information as the presented study reveals. The fuel efficiency is highly sensitive to the altitude, even 2 % uphill slope decreases the efficiency significantly. In contrast the influence of the traffic control is only noticeable in the fuel savings at electric driving on level roads. However, in this case it is questionable if a decision should be made based on this information. For the design and calibration of predictive control strategies general conclusions can be drawn. It is efficient to use the electric energy mainly in lower, urban speed limits as well as in traffic jam situations. While considering solely the fuel efficiency local emission free driving is only in a few combinations of driving situations efficient. Exemplarily, a suitable combination consists of long highway drives which are followed by a short urban trip on level roads without many influences by the traffic control. Especially while driving local emission free the altitude has a significant influence on the energy demand and therefore its fuel saving potential. Nevertheless, bearing in mind the air pollution in urban areas, further situations become apparent which only result in a slight increase of the total fuel consumption. As long as the driving situations and the combination of fuel efficient and local emission free driving enables to avoid the operation of the combustion engine in increasingly more expensive load points, the goals can be achieved efficiently. These correlations also apply to other vehicle in p2 topology but then the specific costs and savings largely depend on the component’s design. The operation is characterized by the lowest possible charging gradient in driving situations which are limited by their highest speed as well as without high slopes. These predictive information should therefore be considered for local emission free driving while designing and calibrating predictive control strategies. Such implementations are subsequently planned for further investigations, especially regarding real-time applicability.

406

Essential predictive information for high fuel efficiency and local emission free …

Bibliography 1. Guzzella, L.; Sciarretta, A.: Vehicle Propulsion Systems, Introduction to Modeling and Optimization, Springer-Verlag Berlin Heidelberg, 2013 2. De Jager, B.; von Keulen, T.; Kessels, J.: Optimal Control of Hybrid Vehicles, SpringerVerlag London, 2013 3. Hofmann, P.: Hybridfahrzeuge, Ein alternatives Antriebssystem für die Zukunft, Springer-Verlag Wien, 2014 4. Görke, D.: Untersuchung zur kraftstoffoptimalen Betriebsweise von Parallelhybridfahrzeugen und darauf basierende Auslegung regelbasierter Betriebsstrategien, Dissertation Universität Stuttgart, Springer Vieweg, 2015 5. Lin, C.-C.; Peng, H.; Grizzle, J.W.: A Stochastic Control Strategy for Hybrid Electric Vehicles, American Control Conference, 2004 6. Vagg, C.: Optimal Control of Hybrid Electric Vehicles for Real-World Driving Patterns, PhD Thesis, University of Bath, Department of Mechanical Engineering, Great Britain, 2014 7. Cummings, T.; Bradley, T. H.; Ahser, Z. D.: The Effect of Trip Preview Prediction Signal Quality on Hybrid Vehicle Fuel Economy, IFAC, 2015 8. Fu, L.; Ozgunuer, U.; Tulpule, P.; Marano, V.: Real-time Energy Management and Sensitivity Study for Hybrid Electric Vehicles, American Control Conference, 2011 9. Romijn, T. C. J.; Donkers, M.C.F.; Kessels, J.T.B.A.; Weiland, S.: Receding Horizon Control for Distributed Energy Management of a Hybrid Heavy-Duty Vehicle with Auxiliaries, IFAC, 2015 10. Zhang, C.; Vahidi, A.: Route Preview in Energy Management of Plug-in Hybrid Vehicles, IEEE, 2012 11. Schuermann, T.; Goerke, D.; Schmiedler, S.; Strenkert, J.; Boehm, K.A.; Bargende, M.: Impact of predictive information on the energy consumption, Stuttgarter Symposium, 2017

407

Analogy considerations for the design of hybrid drive trains Michael Auerbach, Oliver Zirn Hochschule Esslingen

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 M. Bargende et al. (Hrsg.), 19. Internationales Stuttgarter Symposium, Proceedings, https://doi.org/10.1007/978-3-658-25939-6_33

409

Analogy considerations for the design of hybrid drive trains

1 Introduction Development engineers for hybrid drive chains need a suitable understanding of the application of the internal combustion engine (ICE) as well as the electric motor (EM) subsystem. The actual literature in applied textbooks consider EM’s from the perspective of the electrical machine developer with emphasis on the magnetic circuit, winding techniques and power electronic control [2]. In theoretical literature, EM’s are treated on a quite abstract and analytical level (flux linkage and field-oriented control) for the electrical engineer. This makes it difficult for developers coming from classic automotive engineering to accomplish a deeper understanding of electric drives. Based on the analogy of mean effective pressure in the combustion chamber and magnetic surface thrust in the air gap, however, a multi-physical similarity over the displacement and the rotor volume, the torque and the power can be shown along with losses between the subsystems. Based on the classic input values in the fuel tank (caloric value and fuel volume flow), the developer can thus develop a better understanding of the electrochemical energy storage and of the electrical quantities traction voltage and battery current. Mechanical / electro-technical equivalents also exist for the combustion engine-specific mechanical / flow-dynamic limitations, which are explained below.

2 Analogy of Systems

Figure 1: Schematic design of an electric and an internal combustion drivetrain

The analogy of the drivetrain systems is elucidated for an ICE drivetrain and an EM drivetrain in figure 1. All combinations of these subsystems resulting in a hybrid drivetrain can be designed using the considerations shown below.

410

Analogy considerations for the design of hybrid drive trains

2.1 Energy Storage The battery is both, electro-chemical energy storage and converter to electrical energy. Although the direct analogy of the complex battery and the simple fuel tank is only a crude consideration, it is suitable for systems design as a value for the quantity and one for the quality of the stored energy can describe both. The heating value (quality) and the fuel mass (quantity) in equation (1) determine the energy stored in the tank. Equation (2) gives the quantitative electric charge and the qualitative value open circuit voltage. 𝐸 𝐸

𝐻 ∗𝑚

(1)

𝑈 ∗𝑄

(2)

Analogy table 1 Quality

𝐻

Quantity 𝑚

[kJ/kg] Heating value [kg] Fuel mass

 

𝑈 𝑄

[V = J/As] [As]

Open circuit voltage Electric charge

Figure 2: Measured voltage as a function of SOC for LiFePO batteries depending on the state of age

411

Analogy considerations for the design of hybrid drive trains While the quantities of the energy storages are linearly dependent either on the amount of fuel or the state of charge (SOC), the analogy for the quality is more complex. The heating value depends on the fuel once filled up and afterwards considered as constant. A different performance is found for the battery voltage that depends on the SOC and varies slightly as shown in figure 2, each time charge is added or extracted, e.g. by propulsion, regenerative braking or during charging. Once materials and composition of the cell are chosen, other dependencies, as temperature, age, cyclization or current are considered as negligible here. So evidently it should be taken into account that the lower the battery charge is, the lower the quality of your electric energy or in other words the cell-voltage will be and this will affect the systems performance. Common values for the tank system are fuel masses of 30 kg-100 kg derived from the tank volume in a range of 40-120 l. Considering heating values for gasoline 11,1 kWh/kg and diesel 11,8 kWh/kg results in a total energy between 320 kWh and 1200 kWh. For battery systems, serial applications go up to a total energy of 100 kWh where voltages are chosen between 400 V and 800 V resulting in capacities of 250 Ah respectively 125 Ah. With actual energy densities for cells at 250 Wh/kg, this leads to 400 kg cell mass and in addition to this the weight of housing, cooling, armoring and battery management.

2.2 Drive Units

Figure 3a: Torque over speed for an ICE

Figure 3b: Torque over speed for an EM

For the analogy of the drive units, the electrical machine and the power electronic control are treated as one sub-system as the losses of the power electronics are negligible compared to the electrical machine and the presented approach is on system level design.

412

Analogy considerations for the design of hybrid drive trains Figure 3 shows examples for the torque-characteristic of an internal combustion engine and an electrical machine. The ICE torque characteristic curve at full load in Figure 3a is divided into 3 typical operating ranges I-III. Internal combustion engines are not operated at very low speeds: on one hand to perform a safe and stable combustion, on the other hand to avoid frequencies leading to unwanted vibrations in the drive train. Starting from a minimum speed, the so-called idle speed, a reliable and comfortable operation is possible. Beginning with this first operating point, the achievable torque increases over higher speeds and the usual compression in turbo engines leads to reach the maximum torque value. The first attainment of this maximum torque is referred to as the "low end torque" point (LET) and completes the first operating range. Starting from this LET point, the torque is not further increased to mechanically and thermally protect components, in diesel engines also for emission reasons, in some cases also for reasons of product diversification. The amount of fuel converted per cycle varies only slightly with respect to engine speed, depending on the engine efficiency. In a combustion process with an air ratio of λ = 1, this limit therefore also corresponds to the maximum achievable air flow rate per working cycle. This second operating range is called the torque-plateau and ends at the operating point with the highest exhaust gas temperature, at most with an air ratio of λ = 1. In diesel engines the rated power is reached at this edge point. In diesel engines at speeds above this edge point, the cylinder filling needs to be reduced in favor for a higher in-cylinder turbulence and thus higher burning speeds. In addition to this effect, the lower amount of converted energy also promotes a faster combustion process. In gasoline engines, the limitation of the exhaust gas temperature is due to component protection reasons and enforces earlier ignition time setting, unless knocking occurs or permissible peak pressure is exceeded. Although higher heat losses during the prolonged combustion lead to energy efficiency disadvantages and lower torques, exhaust gas temperatures can be limited. Independently of the combustion process, measures are taken to enable an operation even at higher rotational speeds, above the rated power edge point. In addition to the engine issues also the maximum transmissible power of the gearbox is taken into account, hence torque values drop below the plateau and decrease approximately inverse to the rotational speed.

413

Analogy considerations for the design of hybrid drive trains

Figure 4: Structure and equivalent circuit diagrams of FSM, PMSM and ASM [1]

For the description of the torque of the electric machine, the ranges I and II are considered together as the range up to the voltage limit enforced edge point of rated power (also called basic or constant maximum torque range). Here, the induced voltage 𝑈 is below the traction voltage 𝑈 (available after space phasor pulse width modulation, PWM), where the maximum torque producing current Iq =Imax is still reached according to Kirchoff’s law (mesh-rule), and thus the maximum torque can be displayed. 𝑈

414

𝑈

𝑈

𝐾 𝜔 ∗𝜔

√

∗ 𝑈 (= 50 bar) is the essential technology variable that defines the achievable torque and the maximum power via the displacement. In analogy for EM the magnetic surface thrust (together with the mechanically permissible maximum rotational speed) is thus the essential technology variable for the electrical drive developer in order to reliably estimate the required rotor volume and thus the size of the EM. Radial and axial flux machines (today the dominant types in series applications) reach about 𝜎 = 10kN/m2 for air-cooled and 𝜎 = 40kN/m2 for liquidcooled continuous operation. The short-term peak values are as high as 𝜎 = 50kN/m2 for air- and for water-cooled applications [1].

418

Analogy considerations for the design of hybrid drive trains

3 Summary Although there are now a large number of publications on hybrid propulsion systems in addition to standard automotive and electrical engineering textbooks, the analogy analysis presented here is not yet available in the current literature and has already proven to be very useful in several series of training courses by the authors. With the design tools presented here, an efficient drive design based on performance specifications and available installation space can be carried out - largely independent of the type of electric machine used and the applied battery / inverter topology.

Bibliography [1]

Zirn, O.: Elektrifizierung in der Fahrzeugtechnik - Grundlagen und Anwendungen, Hanser-Verlag, Leipzig, 2017, ISBN 978-3-446-45094-3

[2]

Fischer, R.: Elektrische Maschinen. Hanser Verlag, München Wien, 2011, ISBN 978-3-446-43813-2

Acknowledgements The research results presented in this contribution had been worked out by the endowed professorships "High Performance Powertrain" and "Electrified Commercial Vehicles", which since 2015 have been funded by the Stifterverband für die Deutsche Wissenschaft eV, Essen, as well as supported by Mercedes-AMG in Affalterbach and the Advanced Engineering Truck department of the Daimler AG in Untertürkheim.

419

Hybrid operating strategies in the trade-off between fuel consumption and emissions Sven Eberts, H.-J. Berner FKFS M. Bargende FKFS/IVK, Universität Stuttgart

This manuscript is not available according to publishing restriction. Thank you for your understanding.

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 M. Bargende et al. (Hrsg.), 19. Internationales Stuttgarter Symposium, Proceedings, https://doi.org/10.1007/978-3-658-25939-6_34

421

External water management: A predictive challenge Cameron Tropea, Johannes Feldmann, Daniel Rettenmaier, Patrick M. Seiler Institute of Fluid Mechanics and Aerodynamics, Technische Universität Darmstadt, Alarich-Weiß-Str. 10, 64287 Darmstadt Michael Ade Daimler AG, Aerodynamik PKW, 71059 Sindelfingen Daniel Demel BMW AG, Aerodynamik, 80788 München

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 M. Bargende et al. (Hrsg.), 19. Internationales Stuttgarter Symposium, Proceedings, https://doi.org/10.1007/978-3-658-25939-6_35

423

External water management: A predictive challenge

Abstract Predictive capabilities of film, rivulet and drop transport on vehicle surfaces are desirable, since experimental investigations are complex and only possible at late stages of vehicle design. However, simulations and/or experiments remain essential, either related to wing-mirror sight demands or, more recently, to insure proper operation of sensors and/or imaging devices. Major difficulties are presently apparent with such predictions, because physical models describing basic wetting phenomena encountered on typical vehicle surfaces and geometries are lacking or are not implemented in numerical codes. These include processes such as wettability influence on drop impact and splashing, shear-driven film-rivulet-drop transition, incipient motion of drops, drop interaction with grooves and material discontinuities, film/rivulet/drop stripping from edges, etc. Improvements to present models implemented in simulation codes can only be expected once a basic physical understanding of these processes is acquired. The present contribution summarizes the main challenges in developing models and outlines a series of generic experiments with the express purpose of meeting these challenges. The structure of such models at the micro through to the macro scale is described. Implementation of these models in existing codes is also discussed and the outlook for a prediction of external water management is drawn.

1 Introduction External water management on vehicles has always been an important safety issue, usually focused on maintaining good sight between the driver and the wing mirror under all driving conditions. Two circumstances have increased the importance and scope of predicting external water management in recent years. For one, demand has increased to make predictions available at an early design phase, since the practice of introducing major modifications after conducting wind tunnel measurements on later stage prototypes is no longer feasible or acceptable, although fine-tuning and concept confirmation will depend on wind tunnel tests for years to come. The second circumstance is that the safety issue has been extended to maintaining acceptable conditions of external water coverage for sensors and cameras associated with autonomous driving developments. It is therefore opportune to examine the present state of the art in predicting external water management and to identify major challenges and hindrances. The basic physical processes behind external water management issues are conveniently discussed in terms of three consecutive stages, as shown in Figure 1 and given by: ● Water impact and surface coverage ● Transport of wall-bound water (and accretion of particulate matter) ● Detachment of liquid fragments

424

External Water Management: A predictive Challenge whereby input boundary conditions and desired predictive output for each of these stages will be identified and discussed in the following sections. The accretion of particulate matter will not be addressed.

Figure 1: Consecutive stages in external water management defining basic physical processes.

The challenges for a two-phase flow simulation of such process are significant, since a numerical framework is needed that accounts for both the small scales of the interfacial physics and the large length and time scales of a full vehicle. Within this framework, a strong interaction between water and the complex airflow around the vehicle must be accounted for. Moreover, many of the fundamental phenomena related to wetting and de-wetting of surfaces are inadequately modelled and implemented in most existing CFD (Computational Fluid Dynamics) codes. Hence, any improvements to the state of the art involve issues related to both physical modelling and numerical implementation. These circumstances lead to the need for basic generic experiments in which the capabilities of models and numerical methods can be developed and validated for well-defined boundary conditions representing typical situations encountered on vehicles (Hagemeier et al. 2011). In the following sections, such experiments and the yielded numerical improvements, in particular in the codes Star-CCM+ and OpenFOAM will be discussed according to the stages defined in Figure 1.

2 Water Impact and Surface Coverage The input boundary conditions defining liquid impact and surface coverage include the aerodynamic flow field around the vehicle, surface properties such as wettability and morphology, and the local drop number, size and velocity distributions before impact. The desired predictive output is the drop impact probability (capture efficiency) and velocity (magnitude and direction), and the deposited (or re-emitted) liquid mass. Of interest is also the size and velocity of re-emitted secondary droplets. The outcome of this stage serves in several respects as the input boundary conditions to stage 2 Transport of Wall-bound Water and Accretion of Particulate Matter. Everyday experience of water running off a newly waxed car highlights the importance of surface wettability, normally characterized by the advancing and receding contact angles (a,r), the difference of the two known as the contact angle hysteresis. However, such information for typical vehicles is not widely known. Furthermore, these

425

External water management: A predictive challenge angles are not convenient to measure outside of the laboratory. What can be measured directly on vehicles is the equilibrium apparent contact angle (e), which lies between the advancing and receding angle. While this alone does not determine the adhesion forces of a drop on a surface, it is certainly indicative of the state of the surface.

Figure 2: a) Six locations at which contact angle measurements were performed on vehicle. b) Equilibrium contact angle measurements at the six measurement points (MP) over 240 days

To illuminate the long-term condition of a vehicle’s surface, a study was conducted at TU Darmstadt over a period of 240 days. The vehicle was used for short distance commuting and was not parked in a protected environment. Every seventeen days the equilibrium contact angle was measured a multitude of times at six locations on the vehicle (using a Krüss Mobile Surface Analyzer MSA), as shown in Figure 2a. The results are summarized in Figure 2b. Interestingly, even after 240 days, none of the measurement positions exhibited a steady state behavior and a decrease from the initial cleaned surface ranged between 20deg and 30deg. Nevertheless, comprehensive data concerning the contact angle hysteresis on vehicles is still lacking and should be a topic of future research. A second boundary condition concerns the size and velocity of impacting drops. Depending on the overall statement of the problem, this may entail computing the path lines of drops in the flow impinging onto the vehicle or in the case of wind tunnel tests, in the flow field between the spray bars injecting the drops into the tunnel and the vehicle. Also for this boundary condition little verification data exists and indeed, the condition becomes more complex when not only raindrops (wind-driven rain) are considered, but also drops from preceding vehicles or drops from the vehicle’s own tires (self-contamination). The difficulty arises when comparisons are drawn between simulations and experiments, since a comparison is only valid if the same distributions for size and velocity are used as input data. Hence, on-road tests or measurements in wind

426

External Water Management: A predictive Challenge tunnels of drop size and velocity immediately prior to impact are necessary. Furthermore, in wind tunnel tests, the consistency of spray nozzles from day to day must be monitored. None of these measurements is simple, since single realization measurement techniques of drop size and velocity are limited (phase Doppler, time-shift, shadowgraphy (Albrecht et al. 2013, Schäfer and Tropea 2014)) and difficult to implement in the field. Nevertheless, some literature exists about drop sizes from self-contamination (Gaylard et al. 2017, Schütz 2013) or from rain (Bouchet et al. 2014). Perhaps the weakest modelling link presently existing in this stage of simulation is an impact model of drops on a dry or wetted surface, which provides not only a splash limit, but also the size and velocity distribution of the secondary, splashed droplets and the deposition fraction of liquid on the surface. Models presently available in numerical codes are typically only valid within very narrow ranges of impact parameters, from which they were empirically derived. Moreover, very few existing models account for the surface wettability and no existing model has considered the hydrodynamics of drop impact with an imposed aerodynamic flow, as encountered in vehicle water management. Moreira et al. (2010) provide a comprehensive review of literature on this subject, albeit for a different application. A more recent treatment of theory and experiment of drop impact is available in Yarin et al. (2017). Systematic studies on this topic are of utmost importance and urgent. To meet this need one such experiment has been constructed and performed at TU Darmstadt in which all parameters could be well-defined and controlled. The target consisted of a hemisphere of different wettability placed in a vertical (downwards) wind tunnel with a spray nozzle in the upstream settling chamber. The drop size and velocity distributions were measured using a phase Figure 3: Hemispherical model for investigating drop impact, deposition and splashing Doppler at a plane immediately above the north pole and again at the equator. During the experiment, the volume flow rate through the nozzle was monitored and any liquid dripping from the target was collected. Prior to the measurements, the target was replaced by a sponge target of the same shape, which absorbed all impacting drops. This enabled the capture efficiency of the target to be computed and, after the experiment, the deposition rate. Furthermore, the flow field was computed numerically, so that the origin of all detected impacting drops and splashed secondary drops could be computed using a reverse Lagrangian tracking, i.e. the path lines were retroactively computed. This provides all necessary information for formulation of a complete drop impact model, which is presently in progress.

427

External water management: A predictive challenge

3 Transport of Wall-bounded Water Having impacted onto the vehicle surface, water is transported in the form of films, rivulets or drops, driven by gravity, aerodynamic shear and form drag forces or a combination of all, and resisted by viscosity and capillary forces, the latter being determined by surface tension and the advancing and receding contact angles. Moreover, the form of liquid transport will depend on the volume flow rate and small perturbations, which can lead to hydrodynamics instabilities. These instabilities can be responsible for transition from film to rivulet or rivulet to drops. Coalescence occurs when drops propagate with disparate velocities or when they encounter surface discontinuities such as grooves, gutters or material boundaries. Typical occurrences of liquid propagation on a vehicle are illustrated in Figure 4a. To understand the hydrodynamics of shear driven liquid on surfaces generic experiments are necessary and Figure 4b illustrates the transition of films to rivulets to drops in a generic wind tunnel experiment.

Figure 4: (a) Pattern of wet and dry elements on a car in an environmental wind tunnel and (b) in a comparable generic experiment (b). The three main liquid flow regimes are highlighted: films (green), rivulets (blue) and single droplets (orange).

Overarching all three stages of water management, but especially important for the transport of wall-bound liquid, are a number of fundamental modelling and numerical framework issues, which will first be outlined. Regarding the numerical framework, the main advancements concern iso-surface reconstruction, contact line models, contact line velocity, adaptive mesh refinement, dynamic load balancing and turbulence modelling, some of which are discussed below (Rettenmaier 2019). In computing wall-bound free surface flows using the Volume of Fluid (VOF) method, undue parasitic velocities arise if the local surface curvature, used in the Continuum Surface Force method (CSF) proposed by Brackbill et al. (1992) is not well estimated. Improvements introduced by Kunkelmann (2011) and Batzdorf (2015) yield a more accurate interface normal; however, issues still must be addressed when working with non-uniform unstructured grids (Rettenmaier 2019).

428

External Water Management: A predictive Challenge For moving contact lines, a contact line model is required and the empirical models by Hoffman (1975) related to Kistler (1993) are given in terms of an out-of-balance Young’s force, relating the contact angle to the Capillary number (Ca=Ucl/), where  is the dynamic viscosity and  the surface tension. The contact line velocity Ucl is however not the velocity in the grid cell next to the contact line, but is a wall adjacent velocity projected onto the interface then to the wall (Roisman et al. 2008). To properly account for the contact line hysteresis and thus for pinning, which is essential for vehicle water management, a mixed boundary condition (BC) is required at the wall. This Robin boundary condition invokes a Neumann BC for contact angles above the advancing or below the receding contact angle and a Dirichlet BC otherwise; hence, distinguishing between moving and pinned contact lines. Such a boundary condition has been proposed by Linder et al. (2015) and implemented and validated in OpenFOAM®. Although the VOF method cannot be applied for an entire vehicle, such simulations are necessary to develop more approximate numerical approaches and to invoke hybrid multiphase modelling strategies to address different areas of the vehicle. Nevertheless, a significant speedup of VOF methods can be obtained using adaptive mesh refinement and dynamic load balancing on parallel computing Figure 5: Speed up using 2 level refinement (quad) vs. platforms (Rettenmaier et al. 2019). equidistant/static Mesh. N=43x103 vs. N=560x103 For the validation case of capillary rise between plates, a speed-up of approximately one order of magnitude could be achieved, as illustrated in Figure 5. Model and numerical developments must be validated before application and for this, generic experiments or theoretical solutions are most suitable. One such exemplary flow is the study of drop incipient motion in a shear flow. Roisman et al. (2015) developed a model for incipient motion by equating the adhesion force arising from the difference between advancing and receding contact angles to the drag force, approximated using a

Figure 6: Force balance for drop incipient motion in a shear flow

429

External water management: A predictive challenge spherical cap cross-section of the wall-bound drop. Their theory agrees well with available experimental data of Hu et al. (2013) and Milne and Amirfazli (2009). Experiments have been carried out by Seiler et al. (2019) in a fully developed channel flow. They have found a scaling based on the force balance, taking into account the contact line hysteresis: 2 2 Ca  K (uattack  uincipient )3 2

(1),

where K and uincipient are determined only by the substrate wettability and roughness properties. The attack velocity is related to the velocity boundary layer at half the drop height (see Figure 6). The validity of this scaling is apparent from the data shown in Figure 7, at least for the range Ca2/3 30°. Investigations on the rear screen angle of the VW Polo were carried out by Janssen and Hucho [7]. Figure 4 illustrates the change of drag and lift with the rear screen slant angle. The results show a similar trend when compared to the results of the Ahmed body. The critical slant angle for the drag drop is however slightly larger than 30°, while for the lift drop it is smaller than 30°.

439

Aerodynamic study on the vehicle rear shape parameters with respect to ground simulation For the underbody region, Potthof has investigated the rear diffusor on the UNICAR [8]. The results presented in Figure 5 show that equivalent rear lift and drag can be achieved with a smaller diffuser angle and with longer diffuser length.

Figure 4 – Variation of the rear screen slant angle of the VW Polo I and corresponding drag coefficient [7]

Figure 5 – Lift and drag change with variation of the rear diffuser angle [8]

The studies presented above provide a valuable data base when studying the influence of different basic shape parameters on the aerodynamics of a vehicle. However, all the investigations were performed with a fixed ground. Thus, there is a need supplement these investigations with respect to ground simulation.

2.2 DrivAer Model The model used for all investigations in this work is a 25 % scale model of the DrivAer [9], as shown in Figure 6 (left). To reduce the simulation effort, the cooling air inlet grill was closed. Two rear end variants (notchback and estate back), as shown on the top right in Figure 6, are tested as the baseline. A detailed underbody geometry, shown on the bottom right in Figure 6, was chosen to simulate realistic underbody flow.

Figure 6 – DrivAer model (top left), notchback and estate back variants (top right), detailed underbody geometry (bottom right)

440

Aerodynamic study on the vehicle rear shape parameters with respect to ground simulation

3 Comparison between Moving Ground and Fixed Ground To study the difference between the results of the two ground simulation techniques, experimental and numerical methods are used. All investigations were done on the DrivAer model with notchback and estate back.

3.1 Experimental Setup Measurements were performed in the model scale wind tunnel of IVK University of Stuttgart (MWK). This Goettingen-type wind tunnel, operated by FKFS, has a nozzle area of 1.6 m² and is suitable for measuring models in 20 % and 25 % scale. The maximum wind velocity is 80 m/s. A state-of-the-art ground simulation system is installed in the test section. It consists of a 5-belt rolling road system. The center belt resembles the moving floor between the vehicles wheels. The wheels are rotated by the belts of 4 wheel-rotation units. To remove the boundary layer and generate a block profile of the flow, boundary layer suction and tangential blowing is used [10]. The use of all described subsystems is called moving ground. Measurements without the use of the described subsystems are called fixed ground. Figure 7 shows the DrivAer model with to investigated rears in the test section of the MWK.

Figure 7 – DrivAer model notchback (left) and estate back (right) in the test section of the model scale wind tunnel at IVK University of Stuttgart

3.2 Comparison of the Drag and the Lift Change The drag and the lift differences of the baseline configuration between the two ground simulation techniques are plotted in Figure 8. The results with fixed ground are subtracted from the results with moving ground. The light blue bars represent the results of the notchback; the dark blue bars represent those of the estate back.

441

Aerodynamic study on the vehicle rear shape parameters with respect to ground simulation The two rear end shapes show a different behavior of the drag change. The drag of the notchback is reduced by 0.001. The estate back shows a 0.005 higher drag value when compared to the results with fixed ground. These findings are similar to those reported by Wickern et al. [11], who presented results from a wide variety of vehicles. For the lift distribution, the diagram shows similar changes on the two rear end shapes. Smaller lift coefficients can be observed with moving ground. This is caused by the larger flow momentum and lower static pressure in the underbody area.

Figure 8 – Drag and lift difference (moving ground – fixed ground) of the DrivAer notchback and estate back

3.3 Numerical Setup Numerical simulations were performed with the commercial CFD code EXA PowerFLOW, which is widely used in the automotive industry. The simulation area, displayed in Figure 9 (left), is simplified as a box with a blockage ratio of 0.1%. The overall length is approximately 12 vehicle lengths from air-inlet to outlet. It extends 5 vehicle lengths upstream and 6 vehicle lengths downstream. In order to recreate the same moving ground conditions as in the MWK, the central moving belt and the 4 belts of the wheel-rotation units are modeled as moving wall. The tires are modeled as rotating wall and the rims of the wheels are modeled as sliding mesh. The rest of the floor is a frictionless wall, as shown in Figure 9 (right). For simulating the fixed ground, certain areas of the floor are modeled as standard wall with a specified roughness. The roughness parameters are adjusted to build a proper boundary layer thickness, which matches the boundary layer profile in the model scale wind tunnel [12, 13].

442

Aerodynamic study on the vehicle rear shape parameters with respect to ground simulation

Figure 9 – Simulation area (left) and the boundary conditions for the moving ground (right)

4 Parametric Study of Geometry Variations In the following sections, different rear end shape parameters of the DrivAer model are investigated with respect to the two ground conditions. As mentioned in section 1, the trunk lid slope angle, the trunk height, the rear screen slant and the rear diffuser angle differ between vehicle models. The findings from the literature, summarized in section 2.1, indicate that the variation of those rear end shape parameters changes the aerodynamic properties when the fixed ground is considered. Therefore, the four geometric parameters are investigated on the DrivAer model to show the influence of the moving ground on the achieved results. The investigations on the trunk lid slope angle, the trunk height and the rear screen slant are performed using numerical simulations. For the rear diffusor angle, variations with a smaller angle than the baseline were manufactured with a 3D-printer. These were tested in the MWK in addition to the numerical simulations. The variations shown in Figure 10 are: i. ii. iii. iv.

Variation of the trunk lid slope angle in the range from -10° to +10° on the notchback model. Variation of the trunk height in z direction in the range from -20 mm to +20 mm on the notchback model. Variation of the rear screen slant angle in the range from 22° to 50° on the estate model. Variation of the rear diffuser angle in the range from 0° to 18 ° on both notchback and estate back models.

443

Aerodynamic study on the vehicle rear shape parameters with respect to ground simulation

Figure 10 – Geometry variants of (i) trunk lid slope angle αH, (ii) trunk height Z, (iii) rear screen slant angle φ and (iv) rear diffuser angle αD

4.1 Trunk Lid Slope Angle To analyze the influence of the moving ground, the relative value of the aerodynamic coefficient is calculated. The result of the baseline DrivAer is defined as the reference, the results of the geometry variants are subtracted from this reference value. The results of the notchback model with different trunk lid slope angles and different ground conditions are plotted in Figure 11. The light blue dashed lines represent the results with fixed ground while the dark blue solid lines stand for the results with moving ground.

Figure 11 – Drag coefficient change ∆cD (left) and rear lift change ∆cLR (right) over the trunk lid slope angle αH, with moving ground and fixed ground

Different drag optimums can be observed for the two ground conditions. With fixed ground, the optimum slope angle is overestimated compared to the results with moving ground. The lift of the DrivAer is reduced with the trunk lid slope angle. The trend of both curves is similar. The ground condition shows only little influence on the trend of the lift.

444

Aerodynamic study on the vehicle rear shape parameters with respect to ground simulation

4.2 Trunk height The results for the different trunk heights are plotted in Figure 12. With moving ground, the lowest drag coefficient can be achieved with the baseline trunk height (0 mm). With fixed ground, the lowest drag can be achieved when the trunk height is increased by 10 mm – 20 mm. The drag increases again for a trunk height > 20 mm. This corresponds to a trunk height increase of 80 mm in full scale and correlates to the results obtained by Buchheim et al. [4]. Compared to the results with moving ground, the optimum trunk height is overestimated with a fixed ground. The rear lift decreases with increasing trunk height. Moving ground has also little influence on the tendency of the rear lift.

Figure 12 – Drag coefficient change ∆cD (left) and rear lift change ∆cLR (right) over the trunk height Z, with moving ground and fixed ground

4.3 Rear Screen Slant Angle The results of the rear screen slant angle are plotted in Figure 13. The trend of the drag and the rear lift change is similar to the results reported by Ahmed [6]. The drag increases with increasing slant angle and then abruptly decreases for angles > 30°. For slant angles > 35°, an almost constant drag value can be observed. The moving ground shows little influence on the trend of both curves. For the rear lift, the critical angle is reduced by about 1° with moving ground. For angles > 30°, the rear lift decreases. At a slant angle of 32° (see red bar in Figure 13), there is a difference of 0.120 between the two ground conditions.

445

Aerodynamic study on the vehicle rear shape parameters with respect to ground simulation

Figure 13 – Drag coefficient change ∆cD (left) and rear lift change ∆cLR (right) over the rear screen slant angle φ, with moving ground and fixed ground

In order to analyze the mechanism of the difference at this rear screen slant angle, the velocity distribution of the flow in the center plane is plotted in Figure 14.

Figure 14 – Comparison of the velocity distribution between fixed ground and moving ground in the center plane (top) and above the roof (bottom) of the DrivAer estate back with a rear screen slant angle of 32°

446

Aerodynamic study on the vehicle rear shape parameters with respect to ground simulation It can be observed that the separation area in the wake is longitudinally enlarged with moving ground. Compared to the fixed ground, the moving ground increases the flow rate under the vehicle. The underbody flow with moving ground obtains more momentum than with fixed ground. Therefore, upwash is formed behind the car. The separation area is enlarged behind the rear screen. The flow speed above the rear roof edge is lower with moving ground compared to the fixed ground. As a result, the static pressure on the roof surface is correspondingly affected. The pressure distribution, shown in Figure 15, with fixed ground shows lower pressure, which causes the higher rear lift.

Figure 15 – Comparison of the static pressure distribution on the surface of the DrivAer estate back with the rear screen slant of 32° between fixed ground and moving ground

4.4 Rear Diffuser Angle For the study on the rear diffuser angle, the variations from 0° to 6° were tested in the MWK. Due to the geometric limitations of the hardware model, larger diffuser angles could only be investigated with numerical simulations. For the MWK tests, some diffuser add-on parts were designed and manufactured with a Rapid Prototyping System. An undertray panel, shown in Figure 16, was designed to generate smooth flow under the diffuser. The diffuser angle can be varied with interchangeable wedge parts in a range from 0° to 6°. The maximum angle is limited by the original diffuser angle of the DrivAer.

447

Aerodynamic study on the vehicle rear shape parameters with respect to ground simulation

Figure 16 – CAD details of the diffuser add-on parts under the DrivAer

Figure 17 shows the drag change for different diffuser angles. Good agreement between the numerical results and the experimental results can be observed. Different optimum diffuser angles for the lowest drag are identified between the notchback and the estate back. The optimum angle for the notchback is larger than that of the estate back. Moreover, compared to the case with moving ground, the optimum angle is overestimated with fixed ground for both rear end shapes.

Figure 17 – Comparison of the drag value with different diffuser angles between fixed ground and moving ground for the DrivAer notchback (left) and the estate back (right)

The rear lift change for different diffuser angles is plotted in Figure 18. The optimum diffuser angle of the lowest rear lift also depends on the rear end shape. With the

448

Aerodynamic study on the vehicle rear shape parameters with respect to ground simulation notchback, a larger diffuser angle is required to reach the lowest rear lift compared to the estate back. With moving ground, the optimum angle for the notchback is reduced. However, for the rear lift of the estate back, moving ground shows only a small influence on the optimum angle.

Figure 18 – Comparison of the rear lift value with different diffuser angles between fixed ground and moving ground for the DrivAer notchback (left) and the estate back (right)

To analyse the reason of the reduced optimal diffuser angle caused by the moving ground, the differences in the wake structure of the notchback are compared. In Figure 19 the isosurface of cp,tot ≤ 0 is plotted to identify regions where losses in the flow occur.

Figure 19 – Comparison of the isosurface of cp,tot ≤ 0 in the wake of the DrivAer notchback between fixed ground and moving ground with the rear diffuser angle of 12°

449

Aerodynamic study on the vehicle rear shape parameters with respect to ground simulation It can be observed that the pressure loss area behind the stationary wheels (left) is larger than that of the rotating wheels in the moving ground case (right). This larger wake, obstructs the underbody flow at the diffuser. Therefore, a larger diffuser angle is needed to achieve the same effect. This corresponds to the findings of Wäschle [14].

5 Summary and Conclusion The results presented in this paper explain the mechanism that causes the difference in drag and lift between a fixed ground and moving ground for the DrivAer notchback and estate back. The moving ground causes the observed drag increase of 0.001 for the notchback, and the 0.005 lower drag for the estate back. The results also show that the lift for both vehicle shapes decreases in the same order. Furthermore, the systematic investigation of different rear shape parameters of the DrivAer model with the two ground conditions revealed that: – The optimum trunk lid slope angle and trunk height of the drag is overestimated with the fixed ground. – The optimum rear screen slant angle of the rear lift is overestimated with the fixed ground. – The optimum rear diffuser angle depends on the rear end shapes. To achieve the lowest drag and rear lift, the notchback requires a smaller diffuser angle than the estate back. The moving ground reduces the optimum diffuser angle. The tendencies of the drag and lift change is similar for the two ground conditions. However, the moving ground causes different optima of those geometric parameters compared to the fixed ground. The results presented in this study give a decisive complement to the historical investigations. Moving ground has to be considered if the optimum drag and lift values for different shape parameters is to be assessed.

Acknowledgement The authors would like to thank Dr. Felix Wittmeier and the staff of the MWK for the support during the experiments.

450

Aerodynamic study on the vehicle rear shape parameters with respect to ground simulation

Bibliography 1. J. Wiedemann, "Some Basic Investigations into the Principles of Ground Simulation Techniques in Automotive Wind Tunnels," SAE Technical Paper 890369, 1989. 2. E. Mercker and J. Wiedemann, "Comparison of Different Ground Simulation Techniques for Use in Automotive Wind Tunnels," SAE Technical Paper 900321, 1990. 3. J. Wiedemann, "The Influence of Ground Simulation and Wheel Rotation on Aerodynamic Drag Optimization Potential for Reducing Fuel Consumption," SAE Technical Paper 960672, 1996. 4. R. Buchheim, B. Leie and H. -J. Lueckoff, "Der neue Audi 100," in Ein Beispiel für konsequente aerodynamische Personenwagen-Entwicklung, ATZ, 1983, pp. 419-425. 5. L. Krüger and M. Lentzen, "Richtungsstabilität," in Aerodynamik des Automobils, Wiesbaden, Springer, 2008, p. 327. 6. S. R. Ahmed, "Influence of Base Slant on the Wake Structure and Drag of Road Vehicles," Transactions of the ASME, Journal of Fluids Engineering, 1984. 7. L. Janssen and W. -H. Hucho, "Aerodynamische Entwicklung von VW Golf und Scirocco," ATZ, 1975. 8. J. Potthoff, "The Aerodynamic Layout of UNICAR Research Vehicle," International Symposium on Vehicle Aerodynamics, Wolfsburg: Volkswagen AG., 1982. 9. A. Heft, T. Indinger and N. Adams, "Introduction of a New Realistic Generic Car Model for Aerodynamic Investigations," SAE Technical Paper 2012-01-0168, 2012. 10. F. Wittmeier, "The Recent Upgrade of the Model Scale Wind Tunnel of University of Stuttgart," SAE Technical Paper 2017-01-1527, 2017. 11. G. Wickern, A. Wagner and C. Zoerner, "Induced Drag of Ground Vehicles and Its Interaction with Ground Simulation," SAE Technical Paper 2005-01-0872, 2005. 12. O. Fischer, T. Kuthada, N. Widdecke and J. Wiedemann, "CFD Approach to Evaluate Wind-Tunnel and Model Setup Effects on Aerodynamic Drag and Lift for Detailed Vehicles," SAE Technical Paper 2010-01-0760, 2010. 13. T. Kuthada, D. Schröck, J. Potthoff and J. Wiedemann, "The Effect of Center Belt Roughness on Vehicle Aerodynamics," SAE International Journal Passenger Cars, pp. Mech. Syst. 2(1):841-848, 2009. 14. A. Wäschle, "Numerische und experimentelle Untersuchung des Einflusses von drehenden Rädern auf die Fahrzeugaerodynamik," Dissertation, University of Stuttgart, 2006.J. Wiedemann, "Some Basic Investigations into the Principles of Ground Simulation Techniques in Automotive Wind Tunnels," SAE Technical Paper 890369, 1989.

451

Development of an SUV reference model for aerodynamic research Max Tanneberger, Chenyi Zhang IVK, Universität Stuttgart Timo Kuthada, Felix Wittmeier, Jochen Wiedemann FKFS Juliane Nies Röchling Automotive SE & Co

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 M. Bargende et al. (Hrsg.), 19. Internationales Stuttgarter Symposium, Proceedings, https://doi.org/10.1007/978-3-658-25939-6_37

453

Development of an SUV reference model for aerodynamic research

Abstract With the introduction of WLTP, the effect of aerodynamics on the overall emissions is increased. Aerodynamic research can help to reduce the drag of a vehicle with new measures, resulting in a further reduction of the fuel consumption. Therefore, openaccess reference models are needed, that can be used for investigations and CAE method development. These vehicle models need to reproduce the flow details of the basic shape as well as the essential flow structures of the corresponding vehicle. In the global market, SUVs obtain the strongest growing market share. For this vehicle class no model is available, that features a high level of detail. Therefore, this work focuses on the development of the AeroSUV. To depict an SUV, this type of vehicle has to be characterized at first by an analyzation of the legal requirements of the EU and USA. This is added by mean dimensions of 17 models available in the market. The resulting requirements are used to create a first sketch of the model. To replicate a realistic shape, the AeroSUV is optimized using CFD. The resulting geometry is builtup in 25 % scale and investigated in the model scale wind tunnel (MWK) of the University of Stuttgart. The geometry of the AeroSUV is open-access and available on the ECARA website (http://www.ecara.org/).

1 Introduction With the introduction of WLTP, the portion of aerodynamic drag on the determined fuel consumption is increased. Additionally, it necessitates the consideration of each vehicle configuration. In average the procedure increases the determined consumption, hindering the accomplishment of the stricter CO2 regulations. Further aerodynamic research can help to achieve the legal limit values. Nowadays, SUVs are the strongest growing vehicle segment. The geometric characteristics of SUVs are e. g. a high ground clearance, large wheels and a large base body. This results in an increased cross-sectional area, combined with a comparable high drag coefficient. For investigations of this segment within research and development, no detailed model is available. To feature a high quality model that can be used by academia, OEMs and suppliers this work focuses on the development of the AeroSUV. To analyze the usage of generic models and to emphasize the necessity of the AeroSUV, the existing models are considered at first. Afterwards, the legal specifications as well as SUVs available on the market are analyzed, resulting in a first design. This is optimized in CFD to create a state-of-the-art model. The final geometry is built-up in 25 % scale and tested in the model scale wind tunnel (MWK) of University of Stuttgart. The results of these investigations are presented in the last section.

454

Development of an SUV reference model for aerodynamic research

2 Reference Models There are several generic models available for aerodynamic research. These models can be subdivided into simple bodies and basic car shapes [1], based on their level of abstraction. The most common simple body was introduced in 1984 by Ahmed [2] and is shown in Figure 1 on the left. The Ahmed body resembles a cuboid with rounded leading edges, that is placed on struts. By adapting the rear slant angle, different rear end geometries can be modeled. The flow around this reference model has been investigated multiple times in experiment and CFD, improving the understanding of the occurring flow structures.

Figure 1: Aerodynamic reference models. Starting from left, the Ahmed body, the Generic SUV corresponding Wood et. Al. and the DrivAer.

Also, simple bodies have been used for multiple investigations concerning CFD validation, transient flow mechanisms or interference effects. Of course, they cannot depict the flow details of vehicles available in the market due to the level of abstraction, but they capture the essential flow properties of bluff bodies in ground proximity. The second type of reference model is less abstracted. The basic car shapes depict different vehicle classes. For the segment of SUV, the first model was introduced in 2004 by Al-Garni et Al. [3]. It has been used e. g. by Krishnani and Zhou [4] for the investigation of drag reduction measures. This model provides an SUV like crosssectional area, but the slope of the front and rear bumper are not modeled and the wheels are fully integrated. This was improved in 2015 by Wood et. Al. [5] with the generic SUV model that is shown in the middle of Figure 1. Its basic shape is derived from multiple SUV models, released between 1970 and 2011. This model has been used by Forbes [6] for the comparison of different simulation methods or by Kabanovs [7] for the numerical investigation of spray impingement. Basic car shapes increase the complexity of the geometry, but neglect many details that are relevant for the distribution of the drag. For example, the shape of a real vehicles’ rear end often results in a combination of pressure and form induced flow separation and therefore affects the thickness of the separating boundary layer. Additionally, cooling air flow is often neglected, although this area is necessary for real cars and offers a high potential for drag reduction [8]. Also the wheels or at least their rotation is often

455

Development of an SUV reference model for aerodynamic research neglected. Investigations [9] reveal that the flow around rotating wheels accounts for up to 25 % of the drag. Another area that has a large influence on the drag is the underbody. The underbody is often flat at basic car shapes but the rough underfloor of production vehicles takes a large effect on vehicle drag [10]. Finally, the introduced models are often not open-access. The first fully detailed open-access model was introduced in 2012 by Heft et. Al. [11]. The so-called DrivAer model (right side of Figure 1) represents a mid-class car and possesses three rear end configurations (coupe, sedan, estate) as well as two underbody configurations (flat and detailed). The wheels are detailed and can be rotated. The model was extended by Wittmeier and Kuthada [12], adding a generic drivetrain and cooling air system, resulting in the so-called Open-Cooling DrivAer (OCDA) [13]. This model has been quickly adopted, resulting in multiple model built-ups, investigations of the occurring flow structures and drag reduction measures. Even though, this model features the desired level of detail, it cannot depict the aerodynamic characteristics of an SUV. The flow characteristics, resulting from e. g. the rotation of the larger wheels, the aspect ratio of the vehicle body or the increased ground clearance have a significant effect on the aerodynamic characteristics [14]. Hence, an SUV model with a high level of detail is desirable.

3 Development Process To characterize an SUV, the legal requirements of the EU and USA are analyzed and added by mean values of SUVs available in the market. Based on these restrictions, a first design is created and successively optimized. Afterwards, the final geometry of the AeroSUV is presented.

3.1 Requirements In the EU, there is no strict definition of an SUV, but for off-road vehicles. They are defined based on the M1G specification. Among other things, it defines the geometric characteristics as follows: The minimum ground clearance is 180 mm underneath the axles and 200 mm in the middle of the vehicle. The approach angle is set to at least 25° and the departure angle to a minimum of 20°. Finally, a break-over angle of not less than 20° is required. In the USA there is also no strict definition of this vehicle category. But the Code of Federal Regulations (CFR) §523.5 has to be fulfilled, to permit the desired registration as light duty truck. The geometric restrictions correspond to the M1G except the minimum approach angle of 28° and break-over angle of 14°. In both regulations one parameter can be neglected.

456

Development of an SUV reference model for aerodynamic research The regulations contain no information about the overall dimensions. Therefore, 17 midsized SUVs are analyzed, that have been released within the last 5 years. The mean overall length of these vehicles is 4630 mm, combined with a ground clearance of 210 mm, resulting in the desired dimensions of the first design. The mean aspect ratio (ratio of width to height of the base body) equals 1.3. Complementary, a cross-sectional area of 2.5 m² is aspired. For the drag coefficient, a range of 0.30-0.35 is chosen, to represent an aerodynamically sophisticated mid-class SUV. As there is only few information available concerning the lift distribution of production SUVs, the balance is preferred to be neutral. The AeroSUV shall be designed modular, by providing the same level of detail as a production car. Besides, common SUVs with estate rear end, the fastback or coupe SUV increases in popularity. Some car manufactures have also introduced notchback models. The DrivAer already provides these three rear end variations. Therefore, it was decided to design the AeroSUV model to be compatible with this well accepted reference model. Additionally, it shall adopt the parts of the under-hood compartment, drivetrain and the design of the wheels. Also, a modern look within the limitations of the described requirements is preferred. The resulting first design of the AeroSUV is optimized in the following.

3.2 Aerodynamic Optimization The aerodynamic optimization of the 25 % scale AeroSUV with estate back is performed in CFD with EXA PowerFLOW. This commercial solver is based on the Lattice-Boltzmann Method. The simulation volume is simplified as a box with a blockage ratio of 0.1 % and extends 9 vehicle lengths upstream and 13 vehicle lengths 7.8 ⋅ 10 fluid downstream of the model. The setup consists of approximately 𝑛 cells with a finest cell size of 0.6 mm. The floor is modeled using a 5-belt system, that is defined as moving wall and corresponds to the MWK. The residual floor is set as 50 m/s. This corresponds to frictionless wall. The upcoming flow speed is set to 𝑢 a Reynolds number of Re 2.3 ⋅ 10 , defined according to equation 1. Re

𝑢 ∙𝑙 𝜈

(1)

With 𝑙 being the wheelbase and 𝜈 the kinematic viscosity of air at ambient conditions. The wheel rotation is modeled by sliding mesh of the rims, combined with a rotating wall boundary condition on the tires. The tire contact area is modeled by a shift of tires by 1.25 mm into the floor. The pressure loss and the flow alignment of the radiator is modeled using a porous media with an infinite viscosity in cross-wise direction. The

457

Development of an SUV reference model for aerodynamic research configuration without cooling air flow is based on closed upper and lower grille, to simulate the typical measurement used in the wind tunnel to determine the cooling drag. The optimization focuses on the drag as well as the lift distribution. Due to the defined geometric requirements and the desired ability to adapt the DrivAer rear ends, the optimization is limited to basic geometric parameters. This comprises the slope and sweep angle of the front bumper, the inclination of the hood and the windscreen, the ground clearance as well as the shape of the rear diffusor. In the following, a short summary is presented. Previous investigations [15] revealed, that an increased sweep angle effectively reduces the area affected by the stagnation pressure and therefore the drag. The simulations of the AeroSUV confirm this correlation. Additionally, a slow transition between the lower leading edge of the front bumper and the underbody results in a reduced front lift. The geometric characteristics of this area were chosen based on the restriction of the minimum approach angle. The ground clearance is also optimized within the requirements. An increased ride height has a negative effect on the drag. Therefore, the pitch angle was modified by a contrary change of the front and rear ride height. The process results in a negative pitch angle of -1.5 %, in good agreement with the investigations of Zhang et. al. [16] for the DrivAer. Corresponding to Kuthada [8] a reduction of the leakage flow reduces the drag. As described, the geometry within the engine compartment was adopted from the DrivAer. This includes all parts of the drivetrain as well as the cooling module. At the AeroSUV, the redesigned cooling air duct reduces drag and increases the air flow through the radiator. The optimization of the lower rear leads to a slightly curved underbody, forming a diffusor. Even though the exhaust is integrated in the underbody, the layout was designed more symmetrically than at the DrivAer.

3.3 Geometry of the AeroSUV The final AeroSUV geometry is presented in Figure 2 with dimensions given in fullscale. It has a minimum ground clearance of 196 mm underneath the front axle and 212 mm in the middle, fulfilling all legal limit values. In this configuration, the approach, departure and the break-over angle are 25°, 24° and 19°, respectively. This satisfies the CFR except the approach angle and the M1G completely. As the ride height of the AeroSUV is adjustable in a range of -50 mm to +50 mm relative to the shown ride height, all legal requirements can be satisfied.

458

Development of an SUV reference model for aerodynamic research

Figure 2: Geometry of the AeroSUV in side view (left) and front view (right) including characteristic dimensions and the contour of the rear end (marked in red).

The wheelbase of the model equals 2786 mm, which is the same as the DrivAer model [16]. The front and rear wheel-track of the AeroSUV model are 1552 mm and the length is 4619 mm. The width without side mirrors is 1828 mm. The cross-sectional area amounts to 2.47 m², representing a mid-class SUV. The lower grille is partially blocked by a license plate. The three DrivAer rear ends can be mounted on the AeroSUV body. Figure 2 illustrates the AeroSUV with estate back and Figure 3 on the upper left the fastback and on the right the notchback configuration. The contour of the interchangeable rear end is continuous and marked in red for all rear ends. On the bottom of Figure 2, the bottom view of the model is shown. It can be seen, that most of the underbody is covered, except the wheel suspension, the exhaust pipes and the gear box. Corresponding to common vehicles, the cooling air outlets are placed in the wheelhouses and in the underbody cover (around the gear box). Their position and sizes are adopted from the DrivAer.

459

Development of an SUV reference model for aerodynamic research

Figure 3: Fastback and notchback rear end geometries with the contour of the rear end marked in red (top) and bottom view of the AeroSUV (bottom).

4 Aerodynamic Characteristics of the AeroSUV In the following section, the aerodynamic characteristics of the AeroSUV are presented. At first, the experimental results are shown. The investigated setup of the AeroSUV consists of the three presented rear end geometries. Afterwards, the properties of the AeroSUV are compared to the DrivAer based on CFD using the described simulation setup.

4.1 Experimental Investigation The AeroSUV was built-up in 25 % scale. The base body and the underbody are made of painted Ureol. To provide the necessary stiffness, the model consists of an aluminum base plate. The wheels are made of milled aluminum and can be rotated using the FKFS bearing system inside the wheel connector. All add-on parts are made using rapid prototyping techniques. The radiator is modeled using a radiator simulator, equivalent to the DrivAer [12].

460

Development of an SUV reference model for aerodynamic research The physical model is investigated in the MWK of University of Stuttgart, which is operated by FKFS. It is a Göttingen-type wind tunnel with a nozzle size of 1.65 m² and is suitable to measure models in 20 % and 25 % scale. Four struts with an adjustable length secure the model in the test section. A state-of-the-art ground simulation system is installed in the test section including a 5-belt rolling-road system and a boundary layer control system. The wind tunnel turntable allows the rotation of the model relative to the upcoming flow [17]. In the following, the experimental results are presented. This comprises the coefficients of the integral forces for drag and lift of the AeroSUV captured by the underfloor 2.3 ⋅ 10 and corresponds to the balance. The Reynolds number of all tests equals Re performed simulations during the development process. Both cooling air inlets are 0°, if not otherwise stated. opened and the yaw angle of the upcoming flow is 𝛾

𝑐

Estate back Fastback Notchback

Δ𝑐

0.314 0.286 0.286

𝑐

,

0.009 0.007 0.008

-0.009 0.019 0.023

𝑐

-0.014 0.089 0.086

Figure 4: Drag, cooling drag, front lift and rear lift coefficient of the 25 % scale AeroSUV, measured in the MWK at 0° yaw angle.

Figure 4 shows the derived drag, front and rear lift values for the 3 read ends, mounted on the AeroSUV body. The variable Δ𝑐 , represents the cooling drag and is defined according to equation 2. Δ𝑐

,

𝑐

,

𝑐

,

(2)

The estate back represents the common SUV rear end. At the AeroSUV this geometry 0.314 and an overall lift of 𝑐 0.023. For fastback and has a drag coefficient of 𝑐

461

Development of an SUV reference model for aerodynamic research notchback, the drag is 28 counts smaller than for the estate back. As expected, the slope of these two rear ends increases the rear lift. The low cooling drag for all rear ends is in the range of 1 point and emphasizes the good ducting in front of the radiator. On the left of Figure 5, the correlation between the drag and the yaw angle of the flow relative to the model is shown. As expected, the drag increases with increased magnitude of the yaw angle. In the range of -2.5 …+2.5° no drag increase can be 30° is observed. The drag increase for the maximum yaw angle of 𝛾 approximately 35 %. For positive and negative yaw angles the curve shows a similar behavior. On the right side of the same figure, the trend of the drag coefficient with a uniform variation of the ride height for front and rear is plotted. As expected, the drag rises with increasing ride height. The correlation is approximately linear.

Figure 5: Correlation of the drag coefficient to the yaw angle of the upcoming flow (left) and to the ride height (right) for the 25 % scale AeroSUV, determined in the MWK.

4.2 CFD based Comparison to the DrivAer Due to the level of detail of the AeroSUV, it does not belong to simple bodies or to classic basic car shapes. As the rear ends are adapted from the DrivAer it is obvious to compare the flow around those two reference models. Figure 6 shows the contours of AeroSUV and DrivAer in full scale in side view (left) and in front view (right). The most distinct variance can be observed on the upper contour. The body of the AeroSUV is 188 mm higher. This deviation starts at the hood and sustains until the upper edge of the rear end. Additionally, the slope angle of the AeroSUV front and rear bumper is larger. Also, the wheel diameter and the ground clearance are increased. Finally, the cross-sectional area is raised by 13.7 % and the base area by 11.7 %, relative to the DrivAer.

462

Development of an SUV reference model for aerodynamic research

Figure 6: Comparison of the contours of AeroSUV (black) and DrivAer (grey) in side view (left) and front view (right), both with estate rear end.

To investigate the aerodynamic differences between these two models, both are simulated in CFD using the described box setup. The overall drag of the AeroSUV is 5.0 % higher than for the DrivAer. Figure 7 shows the drag distribution by part for both models. As expected, overall the distribution to the displayed parts is quite similar. The portion of the base body to the overall drag is increased for the AeroSUV. Also the fraction of the front and rear wheels is raised from 17.8 % for the DrivAer to 21.3 % for the AeroSUV. This results from the larger wheels, the increased ground clearance and the therefore changed upcoming flow to the wheels.

Figure 7: Drag distribution by part for the AeroSUV and the DrivAer based on CFD.

The drag portion of the radiator is 8.5 % for the DrivAer and 6.1 % for the AeroSUV. Both models consist of the same porous media. Therefore, the difference can be explained by a combination of the blockage of the lower grille by the license plate at the AeroSUV and the lowered static pressure in the underbody region of the DrivAer,

463

Development of an SUV reference model for aerodynamic research resulting in a 6.0 % reduced cooling air mass flow. The values for the exhaust and the gear box match, but the engine shows a significantly lower drag portion at the AeroSUV. This can be explained by the changed flow filed inside the engine compartment. For information about the interaction between the flow field and the model, the local drag is considered [18]. This parameter is defined according to equation 3, with 𝐴 being the cross-sectional area and 𝑐 the total pressure. The formula is based on a differentiation of the momentum and continuity equation from the undisturbed approach flow to a plane downstream of the vehicle. It contains the total pressure loss as well as the losses due to the rotation of the flow field and shows the distribution of the drag contributors in the wake of a vehicle. The integration of this parameter approximately results in the overall drag of a vehicle. Δ𝑐

,

1 ⋅ 𝐴

1

𝑐

,

1

𝑢 𝑢

𝑢 𝑢

𝑢 𝑢

(3)

Figure 8 hows the local drag in an x-aligned slice, 125 mm downstream of the vehicles’ base. The contours of the reference models are marked in red. By comparing the lower area resulting from the underbody flow and the wheel, it can be observed, that the wake of the AeroSUV is more extended to the sides. Also, the local drag in the area of the underbody flow is decreased in comparison to the DrivAer. Both can be explained by the combination of the increased ground clearance and the increased influence of the larger wheels on the drag.

Figure 8: Local dag distribution in an x-aligned slice, 125 mm downstream of the vehicles' base for the AeroSUV (left) and DrivAer (right).

464

Development of an SUV reference model for aerodynamic research Overall, the wake of the AeroSUV is more symmetrical. Also, the wake of the side mirrors is more distinct and blurred compared to the wake for the DrivAer. The central region of reduced local drag indicates the base of the vehicles. The mean base pressure 0.1. In combination with the of both models is similar resulting in a value of 𝑐 , increased base area, this results in a higher influence of the pressure resistance on the drag. Finally, the distribution of the drag into pressure and friction drag is considered. Therefore, the static pressure and the shear stress on the surface are integrated in stream-wise direction. The drag resulting from the flow through the porous media cannot be clearly allocated. Therefore, this value is shown separately, in accordance with the drag by part. The resulting drag portions are displayed in Figure 9. The friction resistance for both models is similar in the range of 12 %. The pressure drag of AeroSUV is 8.5 % higher for the AeroSUV, resulting from a combination of the predefined geometric characteristics.

Figure 9: Drag distribution into pressure, friction and radiator for the AeroSUV and the DrivAer, based on CFD.

It has to be considered, that the drag coefficient as well as the cross-sectional area of the AeroSUV are increased. In summary, the combination of the raised base area, ground clearance and wheel diameter have a significant effect on the drag distribution and therefore indicate the necessity of a model like the AeroSUV.

465

Development of an SUV reference model for aerodynamic research

5 Summary In this paper, a new generic model, the so-called AeroSUV was developed and introduced. After an overview of existing generic models for aerodynamic research, an SUV was geometrically characterized. This contains the consideration of the legal requirements as well as SUVs available in the market. The resulting geometry was optimized in CFD and built-up in 25 % scale. The physical test model was investigated in the MWK. It features a realistic drag coefficient and a low overall lift, combined with an approximately neutral lift balance. The low cooling drag of the AeroSUV underlines the good cooling air management. Finally, a comparison of this model to the DrivAer based on CFD was performed. It was observed, that the deviating geometric details (wheels, ground clearance, base area, etc.) have a distinct influence on the drag. The final geometric data of the AeroSUV is open-access and can be downloaded on the ECARA website (http://www.ecara.org/).

Acknowledgement The Authors would like to thank Röchling Automotive and the Ministerium für Wirtschaft, Arbeit und Wohnungsbau Baden-Württemberg for the financial support. Also, we would like to thank Dr. Daniel Stoll for the technical support during the development process, the staff of the MWK for the support during the experiments as well as Domenik Schramm. Finally, we would like to thank ECARA for the allocation of the geometric data of the AeroSUV on their website.

Abbreviations CFD CFR ECARA KBA M1G MWK WLTP

466

Computational Fluid Dynamics Code of Federal Regulations European Car Aerodynamic Research Association Federal Motor Transport Authority of Germany Off-road Vehicle Category defined by European Community Model Scale Wind Tunnel Worldwide harmonized Light vehicles Test Procedure

Development of an SUV reference model for aerodynamic research

Bibliography [1] G. Le Good and K. Garry, "On the Use of Reference Models in Automotive Aerodynamics," SAE Technical Paper 2004-01-1308, 2004. [2] S. Ahmed, G. Ramm and G. Faltin, "Some Salient Features of the Time Averaged Ground Vehicle Wake," SAE Technical Report No. 840300, pp. 473-503, 1984. [3] A. Al-Garni, L. Bernal and B. Khalighi, "Experimental Investigation of the Flow Around a Generic SUV," SAE Technical Paper 2004-01-0228, 2004, https://doi.org/10.4271/2004-01-0228. [4] P. Krishnani and D. Zhou, "CFD Analysis of Drag Reduction for a Generic SUV," in ASME International Mechanical Engineering Congress and Exposition, Vol. 13, Lake Buena Vista, Florida, USA, 2009. [5] A. Wood, M. Passmore, D. Forbes, D. Wood and A. Gaylard, "Base Pressure and Flow-Field Measurements on a Generic SUV Model," SAE Int. J. Passeng. Cars - Mech. Syst. 8(1):233-241, 2015, https://doi.org/10.4271/2015-01-1546. [6] D. Forbes, G. Page, M. Passmore and A. Gaylard, "Computional study of wake structure and base pressure on a generic SUV model," in Institution of Mechanical Engineers - International Vehicle Aerodynamics Conference, Holywell Park, Loughborough, UK, 2014. [7] A. Kabanovs, A. Garmory, M. Passmore and A. Gaylard, "Computational simulations of unsteady flow field and spray impingement on a simplified automotive geometry," Journal of Wind Engineering and Industrial Aerodynamics Volume 171, pp. 178-195, 2017, https://doi.org/10.1016/ j.jweia.2017.09.015. [8] T. Kuthada, Die Optimierung von Pkw-Kühllftführungssystemen unter dem Einfluss moderner Bodensimulationstechniken, Dissertation, Universität Stuttgart, DE, 2006. [9] G. Wickern, K. Zwicker and M. Pfadenhauer, "Rotating Wheels - Their Impact on Wind Tunnel Test Techniques and on Vehicle Drag Results," SAE Techinical Paper 970133, 1997.

467

Development of an SUV reference model for aerodynamic research [10] A. Huminic and G. Huminic, "CFD Study Concerning the Influence of the Underbody Components on Total Drag for a SUV," SAE Technical Paper 200901-1157, 2009, https://doi.org/10.4271/2009-01-1157. [11] A. Heft, T. Indinger and N. Adams, "Introduction of a New Realistic Generic Car Model for Aerodynamic Investigations," SAE Technical Paper 2012-01-0168, 2012. [12] F. Wittmeier and T. Kuthada, "Open Grille DrivAer Model - First Results," SAE Technical Paper 2015-01-1553, 2015. [13] B. Huprtz, L. Krüger, K. Chalupa, N. Lewington, B. Luneman, P. Costa, T. Kuthada and C. Collin, "Introduction of a New Full-Scale Open Cooling Version of the DrivAer Generic Car Model," in Progress in Vehicle Aerodynamics and Thermal Management - 11th FKFS Conference, J. Wiedemann, Ed., Stuttgart, Springer, 2017, pp. 35-60. [14] J. Pitman and A. Gaylard, "An Experimental Investigation into the Flow Mechanisms Around an SUV in Open and Closed Cooling Air Conditions," in Progress in Vehicle Aerodynamics and Thermal Management, J. Wiedemann, Ed., Stuttgart, Springer, 2017, pp. 61-79. [15] C. Zhang, T. Kuthada, F. Wittmeier, N. Wiedecke and J. Wiedemann, "Parametrische Untersuchungen der Fahrzeugform zum Einfluss der Bodensimulation," Internal report of FKFS, Stuttgart, Germany, 2018. [16] C. Zhang, T. Kuthada, F. Wittmeier, N. Wiedecke and J. Wiedemann, "Aerodynamic Studies on the Underbody Parameters of DrivAer Model with Respect to Ground Simulation," in Advanced Autotive Aerodynamic Forum, Manchester, 2017. [17] F. Wittmeier, "The Recent Upgrade of the Model Scale Wind Tunnel of University of Stuttgart," SAE Technical Paper 2017-01-1527, 2017. [18] A. Cogotti, "A Straqtegy for Optimum Surveys of Pessenger-Car Flow Fields," SAE Technical Paper 890374, 1989, https://doi.org/10.4271/890374.

468

Lamborghini Aventador SVJ Aerodynamics: Route to breaking the super sport car’s record Ugo Riccio, A. Torluccio Automobili Lamborghini S.p.A.

This manuscript is not available according to publishing restriction. Thank you for your understanding.

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 M. Bargende et al. (Hrsg.), 19. Internationales Stuttgarter Symposium, Proceedings, https://doi.org/10.1007/978-3-658-25939-6_38

469

Attribute-based development of Advanced Driver Assistance Systems Prof. Bernhard Schick Head of ADAS/AD Research, University of Applied Sciences Kempten Dr.-Ing. Florian Fuhr Manager ADAS/HAD –Vehicle Dynamics and Application, Porsche AG Dr.-Ing. Manuel Hoefer Field Manager ADAS/HAD –Vehicle Dynamics and Application – Lateral Guidance, Porsche AG Prof. Dr. Peter E. Pfeffer CEO and Head of Vehicle Dynamics/Steering MdynamiX AG

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 M. Bargende et al. (Hrsg.), 19. Internationales Stuttgarter Symposium, Proceedings, https://doi.org/10.1007/978-3-658-25939-6_39

471

Attribute-based development of Advanced Driver Assistance Systems

1 Abstract Advanced Driver Assistance Systems and Automated Driving are a megatrend in the automotive industry. The following questions arise: Will vehicle manufacturers still be able to differentiate themselves "brand-specifically" in the future or will all vehicles be perceived the same when being driven? How can a brand DNA be implemented and how can the transfer of "fun to drive" to "fun to be driven" be achieved? In order to reach this, clear driving characteristic goals – in front of the customer – should be defined and the requirements for vehicle systems and components shall be derived from this. However, what are driving characteristics in the context of assisted and automated driving, Figure 1, and how can those specifically be achieved in the development? Porsche has addressed this question together with the University of Applied Sciences Kempten and MdynamiX. How can an attribute-based development look like and how can Porsche effectively design a brand-typical characteristic in this area?

Figure 1: Driving characteristics in the context of assisted and automated driving

2 Motivation Previous studies with regard to automated lateral control with over 120 test persons and current benchmark vehicles of the University of Applied Sciences Kempten and MdynamiX have shown the following: The functional characteristics and driving characteristics, which are currently achieved, still offer a great upward potential and customer acceptance is still relatively poor [1] [2] [3]. Additionally the challenge for vehicle manufacturers in the development of Advanced Driver Assistance Systems and

472

Attribute-based development of Advanced Driver Assistance Systems Highly Automated Driving (ADAS/HAD) lies in the difficulty of differentiating themselves (brand typical). The brand-specific characteristics and the brand position of the vehicle brands have hardly been taken into account in the ADAS/HAD development. For Porsche, it is very important to design both the product and the brand in the age of ADAS/HAD so that they can be experienced according to their brand. Customers shall experience special emotions, differentiable from other brands and products. Using the example of assisted lateral guidance, a generic procedure model should be developed to translate subjective customer experiences into subjective expert evaluations and finally into objective key performance indicators (KPIs) with defined driving maneuvers [4]. This should make it possible to define objective attribute targets for a Porsche typical characteristic and to validate them at any time in all phases of development from simulation up to road tests. The procedure model should then be transferable to the assisted longitudinal guidance and to driving functions of higher automation levels.

3 Evolution of Advanced Driver Assistance Systems at Porsche and Porsche typical character As a sports car manufacturer, Porsche only offered its customers few safety functions and a cruise control system until the first Panamera was launched in 2009. In the following years, Porsche pursued a late-follower strategy in the expansion of ADAS, focusing primarily on the Panamera and the SUV models. Since the introduction of the 2nd generation of the Panamera 2016, a trend reversal has been initiated. With preview longitudinal control functions such as Porsche InnoDrive, Porsche introduces for the first time an in-house developed ADAS function and optimizes existing functions by means of Porsche typical extensions, such as sportiness recognition [5], Figure 2.

Figure 2: Evolution of Advanced Driver Assistance Systems at Porsche

473

Attribute-based development of Advanced Driver Assistance Systems Porsche pursues the approach of offering driver assistance functions with their own DNA by complying brand-typical attributes such as reliability, sovereignty, performance, intelligence and trust. In order to address this claim for future assisted and highly automated driving functions, professional methods are required for an attributebased development.

4 Method A generic procedure model was developed using the example of assisted lateral guidance. Principles and approaches of vehicle dynamics and chassis development were applied. [6].

4.1 Evaluation Level Model In numerous expert workshops, benchmark tests and measurement campaigns, the relevant attributes for assisted lateral guidance were systematically developed. The defined subjective and objective characteristics were transferred to a so-called level model, Figure 3, and linked accordingly [4].

Figure 3: Evaluation Level Model

This model consists of the levels subjective customer evaluation, subjective expert evaluation, measurement signals and objective characteristic values (KPI - Key Performance Indicators) to be collected in defined driving maneuvers/driving scenarios. At the highest customer level, there are key criteria such as track guidance quality, vehicle reaction,

474

Attribute-based development of Advanced Driver Assistance Systems driver-vehicle interaction, availability, degree of relief, sense of safety and HMI (operation, display, monitoring and warning). At the expert level, the main criteria were broken down in 4-6 sub-criteria. In the next step, all relevant and measurable vehicle signals were worked out in expert workshops, in which the subjective expert criteria were expected to be clearly visible. Based on the expert knowledge, subjective and measurable vehicle signals then were linked. The experts rated the degree of visibility to be expected as high (9), moderate (3), low (1), none (0) or unknown (?). Therefrom, KPIs for the relevant signals were developed according to the individual expert criteria in analogy to characteristic values of vehicle dynamics (chapter 4.3).

4.2 Ground Truth measurement method New measurement and test methods had to be developed for the objective evaluation of driving characteristics in the ADAS/AD context. In the assessment of driving dynamics, it is a common knowledge that the driver - vehicle - environment control loop should consider the global vehicle evaluation. Therefore, driver input as well as road and traffic input, control intervention and the resulting vehicle reaction/movement should be evaluated in its 6 degrees of freedom. Derived from the automated lateral control, it is necessary to obtain a high level of knowledge of the road excitation (essentially road markings and surface geometry) and the driver input in order to be able to evaluate the resulting vehicle reaction accordingly. In the case of assisted longitudinal guidance, a high-level knowledge of the surrounding traffic is required. Like all sensors, environmental sensors such as camera, radar or lidar [7] are faulty and not available or sufficiently accurate in all situations. This can have a significant impact on the driving characteristics. For example, the camera may not be able to reproduce the curvature of the road accurately, which can cause difficulties for the lane-keeping controller. This repeatedly leads to uncertainties if the experienced driving characteristics are a result of the poor performance of sensors, trajectories, controllers, actuators or the poor response of the vehicle influenced by steering, axles, tires and chassis control systems. In order to investigate this cause and effect chain, a much more accurate reference measurement method should be used as “Ground Truth". The chosen approach was to integrate both a highly accurate measured vehicle position and movement into highly accurate digital "Ground Truth" maps, Figure 4. Atlatec has developed a method to generate digital "Ground Truth" 3D maps with high accuracy, which can essentially be used in simulation. It was used to measure various country roads and motorways around Weissach and Kempten. An ADMA pro from Genesys (IMU - Inertia Measurement Unit) with fiber optic gyroscope technology, Kalman filter, RTK-DGPS and SAPOS correction service was used for vehicle position and motion measurement [8], Figure 7.

475

Attribute-based development of Advanced Driver Assistance Systems The same "Ground Truth" 3D maps should be usable as digital twins in the simulation as well as for road testing, Figure 6. Great efforts have been made to achieve the objective of a relative accuracy between both absolute measurements (digital map and vehicle position) of less than +/-5 cm. Even in seemingly open terrain, satellite coverage by bridges, embankments or dips can be so poor that the IMU has to continue with pure coupled navigation and internal support. The longer the gaps in the GPS coverage, Figure 5, the greater the position errors. This depends on the drift quality of the IMU. By means of additional speed support methods and drift corrections or by forward/backward Kalman filters a sufficiently robust accuracy could be achieved [8].

Figure 4: „Ground Truth“ measurement method”

Figure 5: GPS coverage

Figure 6: "Ground Truth" 3D maps usable for road testing and simulation

476

Attribute-based development of Advanced Driver Assistance Systems In order to locate the vehicle precisely in the track during data processing, a route description format (CRO, Curved Regular Objects), based on OpenCRG was developed and the Atlatec measurements were transferred accordingly, Figure 6. In different layers, an orthogonal regular grid can be generated in any resolution. Object types can precisely be assigned and extended at any time. Layer 1 describes the 3D road surface, layer 2 the road marking, layer 3 the barriers and signs, layer 4 the buildings. The data format additionally was enriched with information such as curvature, course angle and attributes as a lock-up table. Regular grids allow the computation-efficient calculation of jump marks, e.g. to the currently measured vehicle position/direction or forecast. Thus, localization and motion calculation in the digital maps are also possible in real time [8]. In addition, a Kistler measuring steering wheel, Figure 7 right, was optimized and adapted to measure the real steering angle/speed and steering torque/gradient as a reference. It was important that the original steering wheel could be used to fully preserve haptics, control functions, hands-off detection and airbag function. The measurements can be characterized in terms of driver-vehicle interaction and allow an evaluation of the quality of the on-board sensors. The measurement concept was designed in such a way that the driving characteristics can be evaluated with all benchmark vehicles without bus connection, Figure 7. In order to be able to assign the availability and the system status of the LKAS to the driving status, the display in the instrument was converted into a measurement signal by means of a camera and image processing, Figure 7 right.

Figure 7: Vehicle measurement instrumentation

477

Attribute-based development of Advanced Driver Assistance Systems In addition, all bus signals, such as the object lists of the sensors, can be measured and precisely assigned. This allows the quality evaluation of the environmental sensors, trajectory, controller and the vehicle response to be examined throughout the entire cause and effect chain and the requirements of the individual components to be defined with regard to the overall vehicle characteristics.

4.3 Route and maneuver catalogue In order to be able to evaluate the driving characteristics in comprehensive driving situations, the previous standard procedures such as EuroNCAP [9] or ISO (International Organization for Standardization) are by far not sufficient. The variance in possible real road events such as road types, curvature, road markings, cross slopes, road entanglement and other road excitations is far too great. For this purpose, a comprehensive route catalogue was developed which corresponds to the intended use of the functions and represents the required excitation variance. In a route catalogue the routes, sections, area and events were subdivided and typified. The waypoints and GPS positions were precisely documented and all routes were generated as reference routes for the driving test and the simulation as digital "Ground Truth" maps.

Figure 8: Example of a driving maneuver definition

In addition, a comprehensive maneuver catalogue was created, in which each individual maneuver was precisely defined, Figure 8. In the so-called free ride, defined operating points in the sections, areas and events with different drivers and times of day were tested using driving instructions. Furthermore, specific driving maneuvers such as lane change test (with and without turn signals), transient test, feedback test, stationary cornering as drop and performance test, on-center handling test, step steer test were developed and described exactly in one document analogous to an ISO.

478

Attribute-based development of Advanced Driver Assistance Systems

4.4 Objective evaluation of driving characteristics using KPI's Using suitable algorithms, further signals could be calculated from the measurement data and then the KPI's could be generated automatically. For this purpose, e.g. reference signals of yaw rate and lateral acceleration, based on the "Ground Truth" curvature, were generated as target and the deviation from the actual measurement was evaluated. For free travel, statistical distributions or counting methods are used, such as the availability measurement, tracking precision measurement or jerk measurement, as well as finding specific states and events using an event finder. For example, the stationary states could be selected, from which the stationary lateral position above the lateral acceleration could be displayed. The following compressed chart, Figure 10, provides information, among other things, on how the vehicle is carried outwards (negative curve cutting gradient) or on how it cuts curves slightly (positive curve cutting gradient). In addition, the chart shows the center position when driving straight ahead (offset at ay=0) and the dispersion as a measure of precision. Furthermore, the lateral acceleration limits, steering torque limits, dropout limits, response, lock-in and lock-in times, steering torque gradients, steering hysteresis and drift speed are determined. In this scheme, over 80% of the subjective expert evaluations could be objectified, Figure 9.

Figure 9: Concept of objective criteria

Figure 10: Side offset for straight and curved driving

5 Driving characteristic evaluations in development 5.1 Benchmark studies and target definition Especially in the very innovative environment of ADAS/HAD it is necessary to observe the performance and solution approaches of competitor vehicles to learn from the good and to avoid the bad. Even in the age of ADAS/HAD, it is important for Porsche to design the product in a way that it can be experienced as typical of the brand and to

479

Attribute-based development of Advanced Driver Assistance Systems differentiate itself from others. This requires clearly recognizable driving characteristics that are associated with the Porsche brand and can be compared to the benchmark. Familiar brand attributes such as driving pleasure, performance, precision, driver feedback, transparency and reliability are to be addressed here as well, with a high degree of suitability for everyday use. Customers would expect a Porsche, for example, to follow a fluid driving line very precisely and always provide the driver with pleasant but not disturbing feedback on the driving condition. For this purpose, the desired brand attributes were linked to the criteria in the level model, Figure 3. This makes it possible to define objective targets for a typical Porsche characteristic and validate them at all times in all phases of development - from simulation to road tests. Figure 11 shows an example of the evaluation of track precision, as sub-criteria of the track guidance quality based on the calculated distance to centerline, of 3 benchmark vehicles, measured in the test free ride. The green area is defined with ±20 cm, the yellow area with ±50 cm and the red area is outside the yellow area. We would like to see more than 95% in the green corridor.

Figure 11: Lane precision distribution of 3 benchmark vehicles

Figure 12 shows the curve cutting coefficients of the 3 benchmark vehicles. The left vehicle is carried relatively and unnaturally to the outside of the curve.

Figure 12: Lane precision distribution of 3 benchmark vehicles

480

Attribute-based development of Advanced Driver Assistance Systems The middle vehicle shows a neutral curve-cutting coefficient and the right vehicle a positive curve-cutting coefficient – thus light curve cutting, which would be desirable for Porsche. The bandwidth also provides information about the lane precision.

5.2 Simulation-based development process In order to be able to validate the characteristic objectives in all phases of the development at the overall vehicle level, a modular simulation environment consisting of the environment simulation Vires VTD, Porsche driving dynamics model and a control system network including lane keeping control was set up. The co-simulation platform AVL Model.CONNECT represents the networking of the individual simulations/models and offers corresponding functions for their consistent use in MIL/SIL/HIL, Figure 15.

Figure 13: Implementation "Ground Truth" maps of the route

Figure 14: Implementation of maneuvers catalogue

A good steering model with effects in the on-center area is required in order to evaluate the track guidance quality, vehicle reaction and driver-vehicle interaction realistically. For this purpose, the Pfeffer steering model of MdynamiX was integrated into the Porsche driving dynamics model. The "Ground Truth" maps of the route and maneuvers catalogue were implemented analogously to the road tests. In order to obtain comparable results for the evaluation of development progress in all phases, the evaluation and evaluation algorithms were integrated into the Porsche postprocessing tool Veda Post. This can be used throughout, from the simulation to the road test, and always guarantees comparable results, Figure 15. Efficient calibrations for uniform driving characteristics across all model series and vehicle variants can thus be achieved.

481

Attribute-based development of Advanced Driver Assistance Systems

Figure 15: Simulation-based development process

6 Conclusion and outlook ADAS/HAD are becoming very important for the Porsche brand. Using the example of assisted lateral guidance, a procedure model was successfully established to show how a typical Porsche characteristic could be effectively achieved in an attribute-based development. The procedure model is currently being transferred to assisted longitudinal guidance and to driving functions with higher automation levels as well as to countryspecific calibrations, e.g. China. In the future, Porsche customers will be able to experience automated driving functions and the associated emotional driving pleasure, typical of the brand.

Bibliography 1. B. Schick, C. Seidler, S. Aydogdu und Y.-J. Kuo, Driving Experience vs. Mental Stress with Automated Lateral Control from the Customer's Point of View, München: ATZ, 2018. 2. C. Seidler und B. Schick, „Stress And Workload When Using The Lane Keeping Assistant - Driving Experience With Advanced Driver Assistance Systems,“ in 27th Aachen Colloquium Automobile and Engine Technology 2018, Aachen, 2018. 3. S. Aydogdu, B. Schick und M. Wolf, „Claim And Reality? Lane Keeping Assistant - The Conflict Between Expectation And Customer Experience“ in 27th Aachen Colloquium Automobile and Engine Technology 2018, Aachen, 2018.

482

Attribute-based development of Advanced Driver Assistance Systems 4. B. Schick, S. Resch, M. Yamamoto, I. Kushiro und N. Hagiwara, Optimization of steering behavior through systematic implementation of customer requirements in technical targets on the basis of quality function deployment, Yokohama/Japan: FISITA, 2006. 5. M. Höfer, Fahrerzustandsadaptive Assistenzfunktionen; Dissertation, Stuttgart: Fraunhofer Verlag, 2015. 6. P. Pfeffer und M. Harrer, Steering Hand Book, Springer Vieweg, 2011. 7. M. Maurer, J. Gerdes, B. Lenz und H. Winner, Autonomous Driving, Springer Vieweg, 2015. 8. D. Schneider, B. Huber, H. Lategahn und B. Schick, „Measuring method for function and quality of automated lateral control based on high-precision digital ”Ground Truth” maps,“ in VDI Tagung Fahrerassistenzsysteme und automatisches Fahren, Wolfsburg, 2018. 9. European New Car Assessment Programme (EuroNCAP), Test Protocol – Lane Support Systems Version 2.0.2, European new car assessment program (EuroNCAP), 2018.

483

ITC – Integrated traction control for sports car applications Dr. Lars König, Frieder Schindele, Andreas Zimmermann Bosch Engineering GmbH

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 M. Bargende et al. (Hrsg.), 19. Internationales Stuttgarter Symposium, Proceedings, https://doi.org/10.1007/978-3-658-25939-6_40

485

ITC – Integrated traction control for sports car applications

Abstract Due to the great success of Bosch Engineering’s nonlinear model based lateral dynamics controller Integrated Vehicle Dynamics Control (IVC), the concept has now been expanded to traction control systems. Based on the method of exact linearization, a feedforward algorithm is presented. This is done by taking the nonlinearities of tire behavior with combined lateral and longitudinal slip conditions into account. Asymptotic stability of the closed loop is established using a model based linear feedback controller, enhanced by gain scheduling and anti-windup algorithms as well as optimized initial condition values. Damping and robustness properties are investigated by applying frequency domain methods. In order to reduce communication delay, the engine ECU is integrated into the controller computations. By integrating the active differential into the traction control algorithm a multivariable controller is established. The performance of the proposed control algorithm is demonstrated by means of road tests that are carried out with a rear wheel driven sports car under high-µ conditions.

1 Introduction Traction control systems (TCS) were firstly introduced in series production cars in 1987 [1], primarily as a safety feature. In the meantime, TCS has become an essential element in order to accomplish excellent vehicle dynamics in high performance sports cars. When trying to achieve new lap records, an optimized traction control can make the difference, even for an experienced driver. Beside pure performance, drivability and driving pleasure are getting more and more into the focus of customers and manufacturers. For example, TCS driving pleasure functionalities are easing vehicle handling in drift situations [2] or enabling the driver to continuously adjust the system to his preferences [3]. However, the safety aspect is still one of the major requirements for any traction control system and has to be regarded within the design and calibration procedure. Another upcoming trend is that original equipment manufacturers (OEMs) are requesting a time and cost reduction in tuning of vehicle dynamics controllers. At the same time, the complexity in drivetrain architectures has increased through electrification and hybridization. In order to meet all these demands and to utilize new potentials provided by electrified drivetrain configurations, Bosch Engineering GmbH is introducing ITC (Integrated Traction Control) – an innovative, model based traction control algorithm. The theory behind ITC is presented in this paper which is organized as follows. Firstly, the plant model is derived and the architecture of the control loop as well as the design objectives are described. Then, the design of the controller with separate feedforward and feedback portions is formulated and the concept of integrated traction control is presented. Finally, the controller performance is investigated by means of road test results.

486

ITC – Integrated traction control for sports car applications

2 Controller design Similar to the concept presented in [4] by Bosch Engineering GmbH involving lateral vehicle dynamics control, here a traction slip controller is developed. A model based nonlinear controller design approach, based on the principle of exact linearization [5] is utilized, as illustrated in figure 2.1. -1 System

+

Target Value Generator

Feedback Controller

-

+

+ System

Figure 2.1: Control loop with feedforward control

The controller consists of three major elements: an inverted model of the controlled system (feedforward portion), a feedback controller and a generator of the target value. The feedforward and feedback elements are subject of this chapter, the target value generator is referred to in [11].

2.1 Derivation of plant model and feedforward control One major challenge in plant modelling for nonlinear controller design applications is finding a compromise between including all relevant effects on the one hand and limiting the model complexity to a level, which can be handeld in terms of model inversion and building time derivatives on the other. In figure 2.2, a “longitudinal quarter vehicle model” is presented. It mainly consists of one spinning wheel, which is connected to the body by a non-holonomic constraint. The differential equation of the quarter vehicle model with the moment of inertia of the drivetrain 𝐽 , the longitudinal tire force of the rear axle 𝐹 , the drivetrain torque 𝑀 , the angular acceleration 𝜔 and the wheel radius is denoted as 𝑟 𝜔

⋅ 𝑀

𝐹

⋅𝑟

.

(2.1)

487

ITC – Integrated traction control for sports car applications

Figure 2.2: Longitudinal quarter vehicle model

In order to consider the whole drivetrain’s moment of inertia, the principle of conservation of energy is applied ⋅𝐽 ⋅𝜔

2⋅ ⋅𝐽 ⋅𝜔

⋅𝐽

⋅𝜔

,

(2.2)

where 𝐽 describes the moment of inertia of all drivetrain elements rotating at wheelrepresents the moment of inertia of all drivetrain elements rotating at engine speed, 𝐽 represents the rotational velocity of the engine. By considering the speed and 𝜔 kinematic transmission ratio 𝑖 𝜔 /𝜔, (2.2) yield the drivetrain’s moment of inertia as, 𝐽

2⋅𝐽

𝐽

⋅𝑖 .

(2.3)

The friction force, 𝐹 , mainly depends on the longitudinal slip ⋅

𝜆

,

(2.4)

which is defined here according to [6], the sideslip angle 𝛼 and a set of tire parameters 𝒑 . The variable 𝑥 represents the longitudinal portion of the vehicle’s groundspeed at the mounting position of the wheel. By combining (2.1)-(2.4) with a tire model according to [10], the longitudinal quarter vehicle model’s equation of motion is obtained as 𝜔

𝑀 ,𝛼 ,𝜆 ,𝐽 ,𝑟

𝑓

,𝒑

.

(2.5)

The controller’s feedforward portion 𝑀

488

𝑓

𝛼 ,𝜆

,𝐽 ,𝑟

,𝒑

𝐽⋅𝜔

𝑥 ,𝑥 ,𝜆

,𝜆

(2.6)

ITC – Integrated traction control for sports car applications is derived by inverting (2.5), setting the wheel angular acceleration equal to its target value and splitting the actuating variable 𝑀 into two portions: a feedforward portion 𝑀 and a feedback portion 𝑀 . The desired angular velocity can be calculated as ⋅

𝜔

.

(2.7)

, which is referred to in chapter 3. Here It depends on the desired longitudinal slip 𝜆 the model inversion 𝑓 is not conducted exlusively with respect to the current system state and the inserted dynamics 𝜔 is not conducted exlusively relative to the desired system state, as proposed in [5] and performed in [4, 7]. In order to reduce model depend on a mixture of complexity and sensor noise influences 𝑓 as well as 𝜔 actual and desired state variables. Thus the dynamics of the feedforward controlled system becomes operation point dependent. Inserting the feedforward control law (2.6) into the system model (2.5) while considering major sources of model uncertainties yields the model 𝜔

𝜉

⋅𝜔

𝑀

𝜉

𝜉

⋅𝜔

𝜔

.

(2.8)

From a system dynamics point of view (2.8) equals a PT1-element with an additional input 𝜉 representing model uncertainties. The natural frequency 𝜉 mainly depends 𝑑𝐹 /𝑑𝜆. on the current longitudinal tire stiffness 𝑐

2.2 Derivation of feedback control Stabilizing feedback controllers with freely choosable error dynamics poles can be designed directly based upon (2.8). However, communication delay is a major factor in today’s traction control system layouts. In order to consider this effect without simplifications and to apply well known robustness measures the feedback control portion is designed in the frequency domain. The ITC control loop is represented by figure 2.3. The feedforward transfer function 𝐺 𝑠 can be derived directly from (2.8). represents measurement noise as well as high frequency The input signal 𝜉 drivetrain oscillations, which cannot be damped by feedback control due to the presence of communication delay 𝑇 . However, a Chebyshev low pass filter 𝐺 𝑠 [8] is inserted into the control loop to prevent the controller from being stimulated by 𝜉 . The plant transfer function 𝐺 𝑠 does not contain the nonlinear plant model (2.5), but only the remaining dynamics of the exact linearized system (2.8) and the communication delay. It is extended by a simplified drivetrain model, representing weakly damped eigenvalues within the frequency range below 10 Hz, which is relevant for traction control: 𝐺 𝑠

⋅

⋅

⋅ ⋅

⋅ ⋅

⋅𝑒

⋅

.

(2.9)

489

ITC – Integrated traction control for sports car applications 𝜉

𝑀

𝐺

𝐺

Δ𝜔

+

𝜔

𝑀

𝐺

+

+

𝑀

𝐺

+

+

𝜔

‐

𝜔

𝐺

𝜔

+

𝜉 +

Figure 2.3: ITC control loop

The feedback controller 𝐺

𝑠

⋅

⋅

⋅

⋅

⋅

⋅

(2.10)

is designed as a linear transfer function with an integrating portion, extended by an antiwindup function [9]. The control loop’s transfer function 𝜔 𝑠

𝐺

⋅𝜔

𝐺

⋅𝜉

𝐺

⋅𝜉

(2.11)

can be directly derived from figure 2.3 and is a combination of the reference transfer function 𝐺

⋅

𝑠

⋅

,

⋅

(2.12)

the disturbance transfer function 𝐺

𝑠

𝜉

⋅

(2.13)

⋅

and noise transfer function 𝐺

𝑠

⋅

𝜉𝑛𝑜𝑖𝑠𝑒

⋅ ⋅

⋅

.

(2.14)

As the reference action is mainly conducted by the feedforward portion, the feedback controller can be designed with major focus on dealing with model uncertainties and

490

ITC – Integrated traction control for sports car applications disturbances. A substantial aspect, which has to be regarded during the design proceof the plant dure, is the longitudinal tire stiffness dependent natural frequency 𝜉 model (2.9). Tire force characteristics derived from a nonlinear model including combined slip conditions according to [10] are illustrated in figure 2.4. When the sideslip angle is small and the operating point lies within the linear range of the longitudinal becomes large and (2.8) represents a stiff PT1-system. force characteristics, 𝜉 However, when the operating point lies within the saturation range of the longitudinal becomes small or even negative and (2.8) represents an force characteristics, 𝜉 integrator or even an unstable PT1-system. Furthermore, figure 2.4 illustrates that there can be a significant change in longitudinal stiffness for a constant longitudinal slip as well when the sideslip angle varies.

Figure 2.4: Longitudinal tire force dependent on traction slip 𝜆 and sideslip angle 𝛼

In order to obtain reasonable controller performance under any operating conditions the variation of 𝜉 is considered by a gain scheduling technique. Figure 2.5 represents the Bode Diagramm of the disturbance transfer function (2.13) when a PID controller 𝐺

𝑠

⋅

⋅ ⋅

(2.15)

is applied. It is featuring two typical oscilation issues, that have to be solved during the controller calibration procedure: weakly damped eigenvalues of the drivetrain in combination with the transportation delay are causing oscilations with a frequency of about 5 Hz. They can be damped by suitable calibration of the controller’s D-portion and are almost independet of 𝜉 .

491

ITC – Integrated traction control for sports car applications

Figure 2.5: Bode Diagramm of disturbance transfer function

Oscilations within a frequency range between 1 Hz and 2 Hz typically occur when the controller’s I-portion is too large with respect to the control plant’s current stiffness 𝜉 . On the other hand a too small I-portion massively worsens the disturbance attenuation properties. Therefore the I-portion has to be significantly scheduled dependent on the operating point, whereas the P- and D-portion can be kept almost may be inaccurate, robust constant. However, as the online estimation of 𝜉 performance has to be guaranteed for any potential operating point. For this purpose the evaluation of the Nyquist Diagram (compare [11]) is not sufficient as it represents only robust stability properties. Robust performance issues can be assessed by evaluating Bode Plots according to figure 2.5 under variation of 𝜉 for any operating point. A specialty of traction control systems is the extraordinary importance of initial conditions. The controller is typically never active for a long period of time but has to be activated and deactivated every time the driver demanded wheel torque exceeds the adhesion limits. As the initial condition response of a dynamic system has no representation in frequency domain it has to be dealt with by time domain analysis. Figure 2.6 illustrates ITC’s “control entry behavior” dependent on the initial conditions 𝒙𝟎 of a state space representation of (2.10). Typically the control deviation is positive when entering control, which means that the control starts before the actual wheel slip has reached its desired value. Due to the communication delay there is a short period of time (in this example 0.05s) where the control deviation is independent of the controllers initial conditions. After that setting 𝒙𝟎 𝟎 yields significantly higher overshoot and poorer damping capabilities compared to a setup with optimized initial conditions.

492

ITC – Integrated traction control for sports car applications

Figure 2.6: Initial condition response of closed loop

Road tests have shown, that the initial conditions should be chosen not only dependent on the initial control deviation but also dependent on the rear axle sideslip angle: the smaller the lateral dynamics stability margin of the rear axle, the less initial slip overshoot is acceptable.

3 Integrated approach One of the basic ideas behind ITC is the integration of all traction relevant actuators and ECUs into the control loop.

3.1 Active Differential Today, a typical setup regarding vehicles featuring active differentials is to implement independent controllers for both actuators – engine (TCS) and differential. However, as both controllers are focusing the same target (“traction optimization”) they are affecting each other by their control activities. This has to be taken into account during the calibration procedure – and is not only increasing the calibration effort but also limiting the achievable performance. In contrast, here an integrated approach is proposed: ITC is generating target slip values for both wheels of a driven axle within one single controller and is simultaneously utilizing both actuators (engine and active differential) to access the two control variables. In order to describe the basic concept of integrated differential and engine control a strictly simplified drivetrain model will be applied in the following. Extending (2.1) by a second wheel and an active differential connecting both wheels yields 𝜔

⋅

𝑀

𝐹

⋅𝑟

𝜔

⋅

𝑀

𝐹

⋅𝑟

,

(3.1a) ,

(3.1b)

493

ITC – Integrated traction control for sports car applications with 𝜔 denoting the rotational speed of the cornering inside wheel, 𝜔 denoting the denoting the rotational speed of the cornering outside wheel and 𝐹 , 𝐹 corresponding traction forces. The system of two algebraic equations (3.1) can now be solved for the unknown drive torque 𝑀 and differential locking torque 𝑀 𝑀

⋅

𝑀

𝜔

𝜔

𝜔 ⋅𝐽

𝜔 ⋅𝐽

𝐹 𝐹

𝐹 𝐹

⋅𝑟 ⋅𝑟

, ,

(3.2a) (3.2b)

dependent on the (target) values of traction forces and rotational wheel speed changes.

3.2 Engine ECU Integration In common TCS applications the traction controller is situated within the ESC ECU and transfers its torque demand to the engine ECU by CAN or Flexray communication. Often even routed by a gateway or so called “vehicle control unit” (VCU). Due to the high dynamics of the wheels’ rotationary motion, communication delay significantly affects the traction control performance (see [11]). In order to decrease communication delay ITC utilizes the engine ECU directly for its computations, like illustrated in figure 3.1.

Figure 3.1: Integrations of Engine ECU

The computation of the feedforward portion as well as the vehicle state dependent scheduling of the feedback gains is performed by the ESC ECU. Thus, the availability of sensor signals, failure detection algorithms and vehicle state and parameter observers, which are relevant for existing ESC functionalities as well, can be utilized. The wheel speed sensor signal is available for the engine ECU with some communication delay. Therefore the actual wheel speed has to be estimated within the engine ECU,

494

ITC – Integrated traction control for sports car applications based on the measured engine speed and the delayed wheel speed measurement information. By accessing the wheel speed estimation as well as the target value, the controller parameters and its initial conditions, which are provided by the ESC ECU, the feedback controller (see chapter 2.2) can be computed within the engine ECU. After that, the engine torque controller collects the feedback portion, the feedforward portion and additional driving state information, provided by the ESC ECU, and computes desired values for the engine’s actuators throttle position, ignition timing and turbo activation.

4 Test drive results Test drives are carried out under high-µ conditions with a rear wheel driven test vehicle (figure 4.1). The powertrain features an engine with 384 kW power and 670 Nm of torque. ITC is accessing the engine and the active differential simultaneously.

Figure 4.1: Bosch Engineering GmbH test vehicle featuring ITC

For evaluation of the driving results, an average percentage quotient for the feedforward portion is introduced as 𝐹𝐹%

| |

| ⋅ | |

| ⋅

⋅ 100% .

(4.1)

This quotient gives an indication about the ratio between feedforward and feedback portion.

495

ITC – Integrated traction control for sports car applications Measurement results of a corner exit acceleration maneuver are shown in figure 4.2. The average velocity is 100 kph whereas the lateral acceleration decreases from 10 m/s2 to 8 m/s2 and the longitudinal acceleration increases up to 4 m/s2. The end of the measurement does not represent the end of the corner exit but the end of traction control activity, as the requested wheel torque by ITC then exceeds the maximum available torque due to the engine’s physical limits. The corner inside (left side) target traction slip exceeds the one for the corner outside wheel. This is necessary due to kinematic restrictions of an active differential. The average control deviation is 0.2 % traction slip at the left wheel and 0.15 % traction slip at the right wheel. The feedforward quotient 𝐹𝐹% equals 95 %.

Figure 4.2: Measurement result of a corner exit acceleration maneuver

Measurement results of a drift maneuver with an average velocity of 90 kph and an average sideslip angle of 20° are illustrated in figure 4.3. The maneuver was performed by an unexperienced driver. Therefore, compared to test drives performed by an experienced driver (see [12]), the sideslip angle signal features significantly higher oscillations. This is due to overcompensating steering action. However, supported by ITC, obviously also an unexperienced driver is able to remain a drift situation for about 10 seconds with a 384 kW rear wheel drive sports car.

496

ITC – Integrated traction control for sports car applications

Figure 4.3: Measurement results of a drift maneuver

The target slip is not only adapted dependent on the sideslip angle (compare [11]), but also dependent on the sideslip angle gradient in order to enhance the vehicle’s yaw damping properties. The feedforward quotient 𝐹𝐹% during this drift maneuver equals 86 %.

5 Conclusion and Outlook A nonlinear model based traction control system – ITC – is presented in this article. The controller consists of a feedforward component, that is obtained by inverting the plant model and a feedback component that is designed to control a linear system with time-varying parameters. Nonlinear model inversion and the insertion of a desired dynamics are conducted dependent on a mixture of actual and desired state variables. Compared to a standard exact linearization setup, this approach reduces model complexity but yields a linear error dynamics with operation point dependent parameters. Performance and damping properties are addressed by gain scheduling and overcoming the effect of communication time delay by an integrated setup. Compared to other control tasks, initial conditions are having higher relevance for traction control systems and have to be chosen carefully. Due to the model based approach the amount of calibration parameters is significantly reduced compared to a standard traction control system.

497

ITC – Integrated traction control for sports car applications Road tests were carried out with a sports car on high friction surface with side slip angles up to 30°. Thereby the feedforward portion is ~90 % and the feedback portion about ~10 % of the controller output. This results in a performance that is appreciated by experienced drivers and also enables even non-professional drivers to control the vehicle at large sideslip angles. The ITC setup referred to in this article has now left its concept phase and entered series development stage within a supersports car project. Future research regarding ITC will focus on the extension of the concept to electric and hybrid drivetrains.

References [1]

Schöpf, H.-J.; Paul,J.: ASR Acceleration Skid Control - A Further Contribution Towards Increasing The Active Safety Of Daimler-Benz Vehicles, 22nd FISITA Congress, Dearborn MI USA, 1988

[2]

Peters, M.: Fahrbericht Ferrari 458 Speciale - Spezialist für schwierige Fälle, auto motor und sport 25/2013

[3]

Gebhardt, C.: Mercedes-AMG GT-R, sport auto 01/2017

[4]

König, L.; Walter, T.; Gutmayer, B.; Merlein, D.: Integrated Vehicle Dynamics Control – an optimized approach for linking multiple chassis actuators, 14th Stuttgart International Symposium for Automotive and Engine Technology, Stuttgart Germany, 2014

[5]

Isidori, A.: Nonlinear Control Systems. Berlin, Springer Verlag, 1995

[6]

Robert Bosch GmbH: Sicherheits- und Komfortsysteme, Vieweg Verlag 2004

[7]

König, L.: Ein virtueller Testfahrer für den querdynamischen Grenzbereich. Dissertation, Stuttgart University, 2009

[8]

Meyer, M.: Signalverarbeitung, Springer Verlag 2017

[9]

Adamy, J.: Nichtlineare Systeme und Regelungen, Springer Verlag 2018

[10] Pacejka, H.; Bakker, E.: The Magic Formula Tire Model. Proceedings of the 1st International Colloquium on Tire Models for Vehicle Dynamics Analysis, Amsterdam Netherlands, 1993 [11] Koenig, L.; Schindele, F.; Jyotishman, G.: ITC – model based feed forward traction control. Proceedings of chassis.tech 2018, München, Germany 2018 [12] Koenig, L.; Schindele, F.; Zimmermann, A.; Merlein, D.: Integrierte Traktionskontrole mit modellbasierter Vorsteuerung. ATZ - Automobiltechnische Zeitschrift, Ausgabe 01/2019, Seite 46-51

498

Computation time optimization of a model-based predictive roll stabilization by neuro-fuzzy systems Philipp Maximilian Sieberg, M.Sc.; Markus Schmid, B.Sc.; Sebastian Reicherts, M.Sc.; Prof. Dr.-Ing. Dr. h.c. Dieter Schramm University of Duisburg-Essen, Chair of Mechatronics, 47057 Duisburg, Germany

Computation time optimization of a modelbased predictive roll stabilization by …

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 M. Bargende et al. (Hrsg.), 19. Internationales Stuttgarter Symposium, Proceedings, https://doi.org/10.1007/978-3-658-25939-6_41

499

Computation time optimization of a model-based predictive roll stabilization by …

Abstract The present article discusses the possibility to reduce the computational effort of complex control algorithms by neuro-fuzzy systems. Thereby, great potentials can be released, especially in the automotive sector. A limiting factor for the design of control algorithms is the task of a real-time execution on cost-optimized control units [1]. The influence of this limitations can be reduced by neuro-fuzzy systems. This is shown exemplary for the model-based predictive control of the roll motion presented by Sieberg et al. [2]. The controller based on the adaptive neuro-fuzzy inference system is validated regarding the control quality and the computational effort. Thus it is compared to the origin model-based predictive control algorithm. The implementation and validation are based on a co-simulation of MATLAB/SIMULINK and IPG CarMaker.

1 Introduction Today’s modern vehicles are more than mere transportation systems. Therefore, vehicle manufacturers meet increasingly complex requirements. In order to persist in the highly competitive automotive sector, they must constantly face new challenges. Subparts of these lasting challenges are the enhancement of driving comfort and safety as well as the improvement of driving dynamics [3]. One possibility to influence the comfort, the safety and the dynamics is the control of the vehicles chassis. In addition to classical control approaches, such as PID controllers, more complex control algorithms are realized. Nowadays this includes for example the model-based predictive control, which Sieberg et al. present in [2] for active roll stabilization. The mentioned predictive control algorithm provides formidable control quality, but requires further improvement regarding the computational effort. Above all, the implementation of such control algorithms on cost-optimized control units in the vehicle itself poses major challenges to the vehicle manufacturers. This article presents a new approach to diminish the computational effort without compromising the control quality. Therefore, the conflict between the computational effort and the availability of computational capacity is avoided. This is shown for the model-based predictive control algorithm. Advances in machine learning are opening new options for training processes and thus for optimizing control algorithm. The described task is solved by using neuro-fuzzy systems. These systems represent the optimization of classical fuzzy systems through the enhanced structure of artificial neural networks. The article is structured as follows: Chapter 2 presents the fundamentals of model-based predictive control and neuro-fuzzy systems. The underlying simulation framework, the training process and the training results are described in chapter 3. Chapter 4 discusses the validation of the implemented neuro-fuzzy controller. Therefore, the control quality and the computational effort are compared with the ones of the original model-based

500

Computation time optimization of a model-based predictive roll stabilization by … predictive control algorithm. The validation refers to an unseen more general driving maneuver. Chapter 5 and chapter 6 conclude the article and deliver an outlook on future research in this area.

2 Fundamentals This chapter presents a short insight into the fundamentals of model-based predictive control algorithms. Furthermore, it focuses on the active roll stabilization of Sieberg et al. [2].The chapter finishes with the fundamentals of neuro-fuzzy systems, which result out of a combination of classic fuzzy logic and artificial neural networks.

2.1 Model-based Predictive Control The model-based predictive control presents the base as well as the starting point of this article. Generally two important attributes characterize such control algorithm. On the one hand they are based on a mathematical description of the system being controlled. On the other hand, this mathematical model can be used to predict the systems behavior. With this knowledge optimal control can be achieved. Model-based predictive control algorithms, which are used in the automotive sector, are presented in [4]. Besides the formidable control quality, the computational effort of model-based predictive control algorithm is very high. Hence, modifications respectively limitations have to be made, to facilitate a real-time implementation [5]. This article refers to a model-based predictive control of the roll motion. Regarding to DIN ISO 8855:2011 [6], the roll behavior describes the rotation of the vehicle body around its longitudinal axis 𝑥𝑥𝑉𝑉 . The roll behavior of a vehicle, in general, can be manipulated by different actuators. The simulated vehicle is equipped with two active stabilizers and four semi-active dampers. The mathematical model, which serves to predict the roll behavior of the vehicle, is described by equation (1). [7] 1 �(ℎ𝑠𝑠 − ℎ𝑤𝑤 )𝑚𝑚𝐴𝐴 𝑎𝑎𝑦𝑦 + (ℎ𝑠𝑠 − ℎ𝑤𝑤 )𝑚𝑚𝐴𝐴 𝑔𝑔𝑔𝑔(𝑘𝑘). . . 𝜃𝜃𝐴𝐴 2 2 𝑐𝑐𝐹𝐹,𝑣𝑣 + 𝑠𝑠𝐹𝐹,ℎ 𝑐𝑐𝐹𝐹,ℎ �𝜑𝜑(𝑘𝑘) −𝑠𝑠𝐷𝐷2 𝑢𝑢1 (𝑘𝑘)𝜑𝜑̇ (𝑘𝑘) − 𝑢𝑢2 (𝑘𝑘)] −2�𝑠𝑠𝐹𝐹,𝑣𝑣

𝜑𝜑̈ (𝑘𝑘) =

(1)

The roll motion is described by the roll angle 𝜑𝜑, the roll rate 𝜑𝜑̇ and the roll acceleration 𝜑𝜑̈ . Furthermore, two external inputs, the lateral acceleration 𝑎𝑎𝑦𝑦 and the acceleration due to gravity 𝑔𝑔, affect the roll behavior. As mentioned, the considered vehicle is equipped with active stabilizers and semi-active dampers, thus 𝑢𝑢1 represents the total attenuation factor and 𝑢𝑢2 constitutes the total counter roll torque. Moreover, 𝑢𝑢1 and 𝑢𝑢2 are the actuating variables in the control algorithm. Furthermore, the parameters of the vehicle setup, such as lever arms 𝑠𝑠𝑖𝑖,𝑗𝑗 and the distance between the center of gravity and the roll pole ℎ𝑆𝑆 − ℎ𝑊𝑊 , are taken into account. The prediction of the roll behavior regarding the

501

Computation time optimization of a model-based predictive roll stabilization by … input signals is achieved by the use of the semi-implicit Euler method [8]. Here, 𝑛𝑛𝑝𝑝 time steps are predicted. The predicted behavior of the vehicle is optimized considering a desired roll motion, which is expressed by 𝜑𝜑𝑟𝑟𝑟𝑟𝑟𝑟 . Therefore, the cost function 𝐽𝐽 is minimized. 𝑛𝑛𝑝𝑝

2

𝑛𝑛𝑝𝑝

𝐽𝐽 = 𝜆𝜆𝑅𝑅 � �(𝜑𝜑𝑟𝑟𝑟𝑟𝑟𝑟 (𝑘𝑘 + 𝑖𝑖) − 𝜑𝜑(𝑘𝑘 + 𝑖𝑖)� + 𝜆𝜆𝑢𝑢1 � 𝑢𝑢1 (𝑘𝑘 + 𝑖𝑖) . . . 𝑖𝑖=1 𝑛𝑛𝑝𝑝

+𝜆𝜆𝑢𝑢2 � 𝑢𝑢2 (𝑘𝑘 + 𝑖𝑖) 𝑖𝑖=1

𝑖𝑖=1

(2)

In addition to the control deviation, the actuating variables are part of the cost function. Thereby, extreme values of actuating variables are prevented. On top of that the actuating variables are restricted to a value facet considering the physical limitations. The actuating variables are expressed by polynomials in order to include their temporal aspects in the process of the optimization. Thus, discontinuities are eliminated (see [2]).

2.2 Neuro-Fuzzy Systems The main component of a neuro-fuzzy system is the fuzzy inference system. These fuzzy inference systems are based on fuzzy logic and enable the mapping of input and output quantities. Hence they can serve as a representation of a systems transfer behavior. Besides tasks in the control theory fuzzy-inference systems are suitable for data processing. [9, 10] A great challenge in the creation of fuzzy inference systems for control tasks is the definition of the membership functions, the rule base as well as the output functions. The membership functions convert the exact input variables into fuzzy degrees of membership. The rule base connects and processes the fuzzy input variables, so that the output function can merge the results of it through operations to be defined. The definition of fuzzy inference systems often requires expert knowledge [11]. Due to increasingly complexity of systems to be controlled, the manual definition is no longer effective. As an alternative, methods of machine learning are used to optimize the fuzzy inference systems. This results in neuro-fuzzy systems, which present a combination of fuzzy-inference systems and artificial neural networks. In Figure 1, an adaptive neurofuzzy inference system (ANFIS) is illustrated. The fuzzy inference system of the Sugeno-type is expressed as a feedforward neural network. [12]

502

Computation time optimization of a model-based predictive roll stabilization by …

Figure 1 Exemplary structure of a two-input ANFIS (own representation based on [10])

The ANFIS consists of five layers. Layer 1 calculates the degrees of membership of the given input variables. Layer 2 conducts the evaluation of the fuzzy-and operation. In Layer 3 the calculated firing strengths are normalized. The consequences of the rule base of the Sugeno fuzzy-inference system are interpreted in layer 4. Layer 5 presents the output layer of a feedforward neural network, which calculates the output of the ANFIS. During the training process only the parameters of the membership functions in Layer 1 and the parameters for the calculation of the rule base consequences in Layer 4 are adapted and optimized. The training process is carried out by supervised learning. [10, 12] Basically, adaptive neuro-fuzzy inference systems are used to model nonlinear functions and systems [12]. Furthermore ANFIS are utilized to classify data as well as to predict a systems behavior [13, 14]. In [15], an adaptive neuro-fuzzy inference system is applied to a control task. In this case the adaptive neuro-fuzzy inference system replicates a linear-quadratic Gaussian control algorithm. It is shown that the ANFIS is capable of replicating the origin control algorithm and substituting it in a closed loop system successfully.

3 Training In this chapter the software framework is introduced. Moreover, chapter 3 presents the training process, especially the used driving maneuvers. It finishes with the illustration of the training results in form of the neuro-fuzzy controller itself.

503

Computation time optimization of a model-based predictive roll stabilization by …

3.1 Simulation Framework The underlying simulation framework refers to the one used in [2]. Generally, the development of the adaptive neuro-fuzzy inference system can be divided into three different tasks. First of all, training data is constituted to represent the model-based predictive control algorithm. Subsequently, the training process is conducted. Finally the generated ANFIS is validated by the application of active roll stabilization itself. The training process of the adaptive neuro-fuzzy inference system is performed using MATLAB, the collection of training data and the validation of the implemented ANFIS is facilitated by a co-simulation of MATLAB/SIMULINK and IPG CarMaker. The underlying structure of this co-simulation is illustrated in Figure 2.

Figure 2 Simulation Framework

The simulation in the software IPG CarMaker considers the dynamic behavior of the vehicle as well as the interaction between the virtual driver, the environment and the vehicle itself. The actual task of the active roll stabilization is carried out in MATLAB/SIMULINK. Therefore, a state estimator, actuator models, the calculation of a desired roll behavior as well as the control algorithm are included. For the first task, namely the collection of training data, the control algorithm is based on the model-

504

Computation time optimization of a model-based predictive roll stabilization by … based predictive controller. For the validation, the model-based predictive control algorithm is exchanged with the controller based on the trained ANFIS. The training process of the adaptive neuro-fuzzy inference system is done with the help of the fuzzy-logic toolbox of MATLAB. In the first step of the training process, the initial structure of the fuzzy inference system is chosen. Every input variable is assigned to three membership functions, which are based on generalized bell functions. Four input variables, namely the lateral acceleration 𝑎𝑎𝑦𝑦 , the actual roll angle 𝜑𝜑, the actual roll rate 𝜑𝜑̇ and the desired roll angle 𝜑𝜑𝑟𝑟𝑟𝑟𝑟𝑟 as well as linear output functions result in 441 adaptable parameters, which are optimized during the training process.

3.2 Training Maneuvers To successfully replicate the model-based predictive control algorithm by the adaptive neuro-fuzzy inference system, reasonable chosen training data are needed. Due to the great influence of the lateral acceleration 𝑎𝑎𝑦𝑦 , especially during cornering, three different driving maneuvers concerning lateral excitation are utilized. The first two driving maneuvers represent the steady-state circular driving behavior defined by ISO 4138:2012 [16]. The first chosen method includes driving maneuvers with constant velocity and varying radii. To consider bilateral influences on the driving behavior, both left and right turns are simulated. From the velocities of 50 km/h and 70 km/h, the radii of 60 meters and 80 meters as well as both cornering directions result eight training maneuvers for the stationary case in terms of the velocity. The second driving maneuver refers to method two of ISO 4138:2012 [16]. These driving maneuvers focus on the curve acceleration behavior of the vehicle. The vehicle accelerates from standstill up to 60 km/h and 80 km/h in curves with radii of 70 meters and 100 meters, respectively. Moreover left turns as well as right turns are taken into account. This results in another eight training maneuvers for the unsteady case in terms of the velocity. The third driving maneuver presents the transient behavior between driving in straight line and cornering. For this purpose, the double lane change as defined in ISO 3888-1:2018 [17] is adopted. This driving maneuver is executed at the velocities of 50 km/h, 70 km/h, 90 km/h and 100 km/h. Therefore, including the four maneuvers describing the transient behavior, a total number of twenty maneuvers forms the database of the training process.

3.3 Training Results The results of the training process are shown exemplary with the input variables of roll angle 𝜑𝜑, desired roll angle 𝜑𝜑𝑟𝑟𝑟𝑟𝑟𝑟 , lateral acceleration 𝑎𝑎𝑦𝑦 and the output variable, the counter roll torque 𝑢𝑢2 . For this purpose, fuzzy surfaces are used. These surfaces illustrate the relationship between two input variables and one output variable, whereas the

505

Computation time optimization of a model-based predictive roll stabilization by … other two input variables are kept constant. In Figure 3, the counter roll torque 𝑢𝑢2 is plotted against the actual roll angle 𝜑𝜑 and the prevailing lateral acceleration 𝑎𝑎𝑦𝑦 .

Within this section, the actual roll rate 𝜑𝜑̇ and the desired roll angle 𝜑𝜑𝑟𝑟𝑟𝑟𝑟𝑟 are set to zero. In particular the wide range of the input variables are recognizable, especially in terms of the lateral acceleration 𝑎𝑎𝑦𝑦 . From this, it can be concluded that the resulting ANFIS can describe a wide range of the driving dynamics, despite the relatively low number of training maneuvers.

Figure 3 Fuzzy surface regarding roll angle 𝜑𝜑, lateral acceleration 𝑎𝑎𝑦𝑦 and counter roll torque 𝑢𝑢2

Figure 4 illustrates the interrelationship between the actual roll angle 𝜑𝜑, the desired roll angle 𝜑𝜑𝑟𝑟𝑟𝑟𝑟𝑟 and the counter roll torque 𝑢𝑢2 . For the fuzzy surface shown, the remaining input variables are set to zero. While Figure 3 presents the influence of a disturbance variable, the lateral acceleration 𝑎𝑎𝑦𝑦 , Figure 4 illustrates the impact of the control deviation. The control deviation is defined as the difference between the desired roll angle 𝜑𝜑𝑟𝑟𝑟𝑟𝑟𝑟 and the actual roll angle 𝜑𝜑.

Figure 4 Fuzzy surface regarding roll angle 𝜑𝜑, desired roll angle 𝜑𝜑𝑟𝑟𝑟𝑟𝑟𝑟 and counter roll torque 𝑢𝑢2

506

Computation time optimization of a model-based predictive roll stabilization by … To check the results of the training process, all input quantities are set to zero. From a physical point of view, this setup should not result in a counter roll torque 𝑢𝑢2 . The trained ANFIS generates a counter roll torque 𝑢𝑢2 = − 0.0145 Nm. Taking into account the actual magnitude of this output variable, it is almost equal to zero. Furthermore, the asymmetric structures of the fuzzy surfaces are based on the underlying training data. Due to the usage of only three different driving maneuvers, some threshold regions of the roll dynamics are more excited than others. With the help of more and in particular differing training maneuvers, the training database can be increased and so on these regions would be considered additionally.

4 Validation The trained controller is validated based on its control quality and the computational effort. In this study, the controller is compared to the original model-based predictive control algorithm. The driving maneuver, which is conducted, represents a part of the race course “Hockenheimring”. This test track satisfies the demand of an unseen maneuver as well as of a realistic route. In particular the altering velocities, the occurring accelerations as well as the whole routing differ from the training maneuvers. Thus, the vehicle dynamics are excited in unseen respectively untrained regions. The track layout is shown in Figure 5. To validate the control quality of the adaptive neuro-fuzzy inference system, the controller is integrated into the simulation framework shown in Figure 2. In the course of this it replaces the model-based predictive control algorithm.

Figure 5 Validation Track “Hockenheimring“

In the following, the roll behavior of the two different control algorithms is compared for the illustrated part of the “Hockenheimring”. For this purpose the resulting roll angles are plotted against time, which is illustrated in Figure 6. Besides that, Figure 6 also shows the resulting roll angles of a passive vehicle without an active roll stabilization

507

Computation time optimization of a model-based predictive roll stabilization by … as well as the desired roll behavior. Figure 6 clearly depicts that the desired roll angle 𝜑𝜑𝑟𝑟𝑟𝑟𝑟𝑟 reduces the roll behavior compared to the passive vehicle by approximately 75 %. It should also be noted that in addition to the model-based predictive control, the controller based on the trained ANFIS also follows the desired roll behavior. To compare both control algorithms the absolute control deviation for the given driving maneuver is calculated. This deviation is calculated using equation (3). 𝑡𝑡𝑒𝑒𝑒𝑒𝑒𝑒

𝑅𝑅𝐴𝐴𝐴𝐴𝐴𝐴 = ��𝜑𝜑𝑟𝑟𝑟𝑟𝑟𝑟 − 𝜑𝜑� 𝑡𝑡=0

(3)

The model-based predictive control algorithm shows an absolute control deviation of 𝑅𝑅𝐴𝐴𝐴𝐴𝐴𝐴,𝑀𝑀𝑀𝑀𝑀𝑀 = 73.5656 rad. Furthermore, the control algorithm based on the trained adaptive neuro-fuzzy inference system results in an absolute control deviation of 𝑅𝑅𝐴𝐴𝑏𝑏𝑠𝑠,𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴 = 84.7286 rad. The datasets are recorded with a step size of 1 millisecond. For one time step, this results in an average deviation of ∆𝜑𝜑𝑀𝑀𝑀𝑀𝑀𝑀 = 5.6007 ∙ 10−4 rad for the model-based predictive control algorithm and ∆𝜑𝜑𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴 = 6.4506 ∙ 10−4 rad for the control algorithm based on the ANFIS. Related to the averaged desired roll angle of this driving maneuver, the relative deviation between the two control algorithms is 2.3706 %. Besides the validation based on the control deviation, the difference between the two controlled roll angles can also be used. For the driving maneuver of the “Hockenheimring”, this deviation can be expressed as the root mean squared error, in which the value is 𝑅𝑅𝑅𝑅𝑆𝑆𝑀𝑀𝑀𝑀𝑀𝑀−𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴 = 8.4829 ∙ 10−4 rad.

Figure 6 Roll Behavior on the “Hockenheimring“

Since the adaptive neuro-fuzzy inference system is proven to be compatible to the model-based predictive control algorithm in control quality, the computational effort of

508

Computation time optimization of a model-based predictive roll stabilization by … both control algorithms is compared. To determine the computational effort, the simulation time for the given maneuver is measured. The calculations are executed with maximum speed. The model-based predictive control algorithm possesses a total simulation time of 𝑡𝑡𝑀𝑀𝑀𝑀𝑀𝑀 = 332.0903 seconds, whereas the controller based on the ANFIS requires a simulation time of 𝑡𝑡𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴 = 28.1698 seconds. The duration of the driving maneuver is defined by one lap, which is equal to 𝑇𝑇 = 131.4 seconds. Hence, the execution time factor 𝐸𝐸𝐸𝐸𝑊𝑊 is defined by equation (4). It represents the ratio between the duration of the maneuver 𝑇𝑇 and the duration of the simulation 𝑡𝑡. 𝐸𝐸𝐸𝐸𝑊𝑊 =

𝑇𝑇 𝑡𝑡

(4)

The execution time factor 𝐸𝐸𝐸𝐸𝑊𝑊 is additionally used to compare the computational effort of the control algorithm. The model-based predictive control possesses an execution time factor of 𝐸𝐸𝐸𝐸𝐹𝐹𝑀𝑀𝑀𝑀𝑀𝑀 = 0.3957. Meanwhile, the adaptive neuro-fuzzy inference system yields an execution time factor of 𝐸𝐸𝐸𝐸𝐹𝐹𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴 = 4.6646, which is almost twelve times faster than the origin model-based predictive control algorithm.

In addition to the examination of the execution time, the principle of operation can be compared further. Whereas the model-based predictive control algorithm solves a minimization problem and thus works iteratively, the adaptive neuro-fuzzy inference system constitutes the output variables directly. Regarding the ability of a real-time application [18], the controller based on the ANFIS can be used unmodified, meanwhile the maximum number of iterations of the model-based predictive control has to be limited.

5 Summary and Conclusion In this article a complex control algorithm with high computational effort is reproduced by the use of adaptive neuro-fuzzy inference systems. This complex algorithm is exemplary presented by the model-based predictive control algorithm of Sieberg et al. [2]. With the help of the ANFIS, the computational effort is significantly reduced without a degradation concerning the control quality. The implementation of the adaptive neurofuzzy inference system is carried out with the help of MATLAB and in parts with the help of a co-simulation of MATLAB/SIMULINK and IPG CarMaker. The process of the reproduction can be subdivided into three tasks. The first task presents the preparations of the training process: the generation of training data. Therefore, three significant driving maneuvers are chosen. The variation of parameters, like cruising speed, curve direction or curve radius results in twenty training maneuvers. In the second part, which is equal to the main part, the training process is carried out. The training is executed with the help of the fuzzy-logic toolbox of MATLAB. In the initial case three membership functions in form of generalized bell functions are assigned to every input variable. The chosen structure of a Sugeno-type fuzzy-inference system leads to 441 adaptable

509

Computation time optimization of a model-based predictive roll stabilization by … parameters, which are optimized during the training process. The last task presents the validation of the trained ANFIS. Therefore, the control quality and the computational effort is compared to the origin model-based predictive control. The validation is done with an unseen maneuver. More specifically a part of the “Hockenheimring” is used as the validation maneuver. Thus a realistic evaluation of the implemented ANFIS is possible. In the validation no significant deviation in terms of the control quality is determined. In general, the roll behavior of both control algorithm differs by the root mean squared error of 𝑅𝑅𝑅𝑅𝑆𝑆𝑀𝑀𝑀𝑀𝑀𝑀−𝐴𝐴𝑁𝑁𝐹𝐹𝐹𝐹𝐹𝐹 = 8.4829 ∙ 10−4 rad. However, the computational effort has been greatly reduced with the help of the ANFIS. In summary the simulation of the validation maneuver using the trained adaptive neuro-fuzzy inference system is almost twelve times faster than the simulation based on the model-based predictive control algorithm. Furthermore, the controller based on the ANFIS operates directly, meanwhile the optimization of the model-based predictive control algorithm works iteratively. Through the significant reduction in computational time using ANFIS, a real time implementation can be realized.

6 Outlook In future research the training process of the ANFIS will be optimized. As an example the optimization during the training process should be calculated by the GPU instead of the CPU. Furthermore, the robustness and the stability of the adaptive neuro-fuzzy inference system will be investigated. Finally more complex model-based predictive control algorithm can be implemented on a real-time system with the help of adaptive neuro-fuzzy inference systems. For example this could be used for an implementation in the vehicle itself.

Bibliography [1] T. Loderer, Echtzeitfähige Simulation steifer Modelle für Anwendungen in Fahrzeug-Steuergeräten, Heidelberg: Universität Heidelberg, 2017. [2] P. M. Sieberg, S. Reicherts and D. Schramm, "Nichtlineare modellbasierte prädiktive Regelung zur aktiven Wankstabilisierung von Personenkraftwagen," in IFToMM D-A-CH Konferenz, EPFL Lausanne, 2018. [3] B. Heißing and M. Ersoy, Chassis Handbook - Fundamentals, Driving Dynamics, Components, Mechatronics, Perspectives, Vieweg + Teubner, 2011.

510

Computation time optimization of a model-based predictive roll stabilization by … [4] L. del Re, F. Allgöwer, L. Glielmo, C. Guardiola and I. Kolmanovsky, Automotive Model Predictive Control - Models, Methods and Applications, SpringerVerlag, 2010. [5] E. F. Camacho and C. Bordons, Model Predictive Control, Springer-Verlag, 2013. [6] DIN ISO 8855:2011, Fahrzeugdynamik und Fahrverhalten – Begriffe, Deutsches Institut für Normung e.V., 2011. [7] D. Schramm, M. Hiller and R. Bardini, Modellbildung und Simulation der Dynamik von Kraftfahrzeugen, 2. Auflage ed., Springer, 2013. [8] A. Cromer, "Stable Solutions Using the Euler Approximation," American Association of Physics, pp. 455–459, 1981. [9] R. Kruse, C. Borgelt, C. Braune, F. Klawonn, C. Moewes and M. Steinbrecher, Computational Intelligence - Eine methodische Einführung in Künstliche Neuronale Netze, Evolutionäre Algorithmen, Fuzzy-Systeme und Bayes-Netze, Wiesbaden: Springer Vieweg, 2015. [10] J.-S. R. Jang, C.-T. Sun and E. Mizutani, Neuro-fuzzy and Soft Computing: A Computational Approach to Learning and Machine Intelligence, Michigan: Prentice Hall, 1997. [11] J. Jantzen, Foundations of Fuzzy Control: A Practical Approach, Chichester: John Wiley & Sons, Ltd, 2013. [12] J.-S. R. Jang, "ANFIS: adaptive-network-based fuzzy inference system," IEEE Transactions on Systems, Man, and Cybernetics, vol. 23, no. 3, pp. 665–685, 1993. [13] J. Zhang, Z. Y. He, S. Lin, Y. Zhang and Q. Qian, "An ANFIS-based fault classification approach in power distribution system," International Journal of Electrical Power & Energy Systems, vol. 49, pp. 243–252, 2013. [14] F.-J. Chang and Y.-T. Chang, "Adaptive neuro-fuzzy inference system for prediction of water level in reservoir," Advances in Water Resources, vol. 29, no. 1, pp. 1–10, 2006. [15] Z. Q. Gu and O. Oyadiji, "Application of MR damper in structural control using ANFIS method," Computer & Structures, vol. 86, no. 3-5, pp. 427-436, 2008.

511

Computation time optimization of a model-based predictive roll stabilization by … [16] ISO 4138:2012, Passenger Cars - Steady-state circular driving behaviour Open-loop test methods, International Organization for Standardization, 2012. [17] ISO 3888-1:2018, Passenger cars – Test track for a severe lane-change manoeuvre – Part 1: Double lane-change, International Organization for Standardization, 2011. [18] E.-R. Olderog and H. Dierks, Real-time systems: formal specification and automatic verification, Cambridge University Press, 2008.

512

Integrated approach for the virtual development of vehicles equipped with brake control systems Fabian Fontana, Jens Neubeck, Jochen Wiedemann IVK/FKFS, Stuttgart Ingo Scharfenbaum, Philippe Stegmann, Armin Ohletz, Uli Schaaf Audi AG, Ingolstadt

Integrated approach for the virtual development of vehicles equipped with brake control …

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 M. Bargende et al. (Hrsg.), 19. Internationales Stuttgarter Symposium, Proceedings, https://doi.org/10.1007/978-3-658-25939-6_42

513

Integrated approach for the virtual development of vehicles equipped with brake control …

1 Motivation 1.1 Introduction Since the introduction of the brake control system Electronic Stability Control (ESC) in 1995 an increasing number of vehicles was equipped with the system and it became statutory for new registered vehicles in 2014, e.g. in Europe, the United States and Australia [1] [2]. Representing a vehicle dynamic control system, the ESC assures predictable, stable and controllable vehicle behaviour, even at the limits of driving dynamics [3]. Several studies prove the effectiveness of ESC in reducing severe and fatal accidents [4] [5]. In consequence, the consideration of the brake control system is an important issue in every vehicle development. Simultaneously, today’s vehicle development is characterised by an increasing demand for a customer’s individualisation and a high competitive pressure. Moreover, the complexity and variety of the vehicles increase due to a higher number of interacting mechatronic components [6]. Examples are the degree of powertrain electrification or a rising number of chassis control systems that strongly interact with the brake control system. This leads to an extensive effort in application and testing of the different vehicle setups and conventional methods need to be extended by virtual methods and tools. Virtual approaches facilitate considering the vehicle variance, which results from different chassis, tyres, chassis control systems etc. In addition, they enable use of novel methods such as sensitivity analyses or optimising algorithms. A lack of reproducibility and expenditure of time are difficulties working with physical prototypes. Virtual methods enable obtaining reproducible results and efficient testing of several manoeuvres for different vehicle configurations. Moreover, the simulation is simultaneously available for multiple users, reduces costs compared to physical prototypes and is utilisable to homologate vehicle variants virtually in accordance with the law.

1.2 State of the Art Several publications deal with the application of virtual methods to develop vehicles with brake control systems. HAHN et al. utilise simulations as the basis for ESC certification according to legal requirements. For that purpose, they build up a hardware-inthe-loop environment [7]. MAO et al. picture the situation of increasing complexity and variety whereas the development time shortens. Thus, the authors show how to use simulations for the initial release without a physical prototype within a computer-aided design process. They forecast that virtual homologation can be supported by using the proposed tool chain [8]. LUTZ et al. describe methods used for homologation according to ECE regulation 13H (ECE R 13H). They show and validate a simulation environment

514

Integrated approach for the virtual development of vehicles equipped with brake control … and present the process of virtual development. For instance, the base application and robustness evaluations are done in a software-in-the-loop environment. In this context, objective characteristic values for evaluation of brake control systems are depicted. They mention the importance of considering interactions between brake control systems and other chassis control systems [9]. One important prerequisite for virtual development is objective evaluation criteria, especially if mathematic methods are utilised. In the topic area of brake control systems there are several criteria specified in literature. The most established ones are defined in the ECE R 13H [10]. This regulation defines the lateral offset, the maximum yaw rate and the yaw rate after particular periods during the sine with dwell manoeuvre. In detail, the manoeuvre is a sinusoidal steering input with a frequency of 0.7 Hz and a 500 ms dwell on the second peak. These values are crucial for homologation. FACH et al. propose the maximum lateral acceleration as an indicator for vehicle responsiveness and the integral under the measured curve of the slip angle for vehicle stability [11]. GERDES et al. suggest the evaluation of the slip angles at both the front and the rear axle and define criteria for responsiveness as well as stability [12]. A further established manoeuvre to assess brake control systems is the double lane change manoeuvre according to ISO 3888 [13]. However, the norm itself outlines that the manoeuvre has a lack of reproducibility and dependencies on the followed path or longitudinal dynamics that results in a considerable scatter in the data.

1.3 Objectives The increasing number of vehicle variants and the competitive pressure require new methods to develop vehicles with brake control systems. In addition to maintain conventional methods, virtual methods enable new potentials, such as analysing causeeffect relationships without impact of disturbances or even with specific disturbances. The focus of related publications are virtual homologation for vehicle variants according to legal requirements. Anyhow, there is a lack of consistency concerning the development based on vehicle characteristic targets. Moreover, interaction effects between brake control systems and further chassis control systems are not considered sufficiently. The objectives of this paper are the following: – Depiction of the required methods for the virtual development of vehicles equipped with brake control systems based on the V-model – Introduction and validation of a capable simulation environment to simulate vehicles with connected chassis control systems within the whole development process – Demonstration of the application of the introduced simulation environment

515

Integrated approach for the virtual development of vehicles equipped with brake control …

2 Virtual Development Process for Vehicles Equipped with Brake Control Systems 2.1 Required Methods for Virtual Development According to V-model The proposed process is based on the established V-model and is depicted in Figure 1 [14] [15]. The V-model is composed of the specification branch, the implementation and the verification branch. The virtual development in compliance with the V-model requires specific methods to meet the requirements of vehicles with brake control systems. The process begins with the definition of vehicle characteristic targets. These targets have to be defined in a testable and unambiguous way. This enables both a purposive development and a mathematical evaluation of the vehicle behaviour. As mentioned in Chapter 1.2, several publications deal with the objective evaluation of manoeuvres that trigger brake control interventions. The sine with dwell manoeuvre defined in the ECE R 13H is established because it is statutory for homologation. Besides, the regulation specifies characteristic values that describe the dynamic vehicle behaviour. Such characteristic values enable the utilisation of sensitivity analyses, optimisation algorithms and the like. Though objective characteristic values are published, a difficulty is the definition of the specific driving behaviour of the original equipment manufacturer (OEM). This means that the objective evaluation of further driving manoeuvres is required to transfer the engineer’s expertise to distinct evaluation criteria. Based on the vehicle characteristic targets, the specification of the function is derived. The function describes an implementation-independent formulation of influencing vehicle characteristics. Examples are reducing yaw rate or sideslip angle. The specification of the function is the basis for the specification of the system. Subsequently, the control of the yaw rate is used as an example to illustrate the approach. A yaw torque is necessary to affect the yaw rate of the vehicle. This can be generated by several systems, for instance, brake control system, electric power steering or rear-wheel steering. A knowledge database is requisite to assess which function meets the requirements best. If such a database is not available, the understanding is built up by means of sensitivity analyses that quantifies the influence of an active system. Besides, the knowledge about crucial vehicle variants is important for the purposive specification that considers the vehicle variance. In addition, it enables the decision which vehicle variants are built up as physical prototype. After selecting the brake control system to control the yaw behaviour, the actual design of the system is conducted. The system generally consists of mechanic, electric and software parts that are designed by interdisciplinary engineering teams based on the specification. That is, the brake

516

Integrated approach for the virtual development of vehicles equipped with brake control … control system is designed in a way that assures meeting of vehicle requirement targets. Using this general approach the system design is at most problem-oriented and leads to the best possible solution. A drawback of this proceeding is a lack of consideration of the established procedure, which is a collaboration of the original equipment manufacturer (OEM) and the supplier. That is, the vehicle itself and the brake control system are developed separately. Formal languages enable tracing of requirements during the whole process and the verification against the specification. Besides, they facilitate the clear connection between requirements and system design. An appropriate formal language is SysML [16]. Aims at stage  Required methods Vehicle characteristic targets  Objective evaluation of vehicle behaviour

Total vehicle level

Function specification System specification  Sensitivity Analysis  Knowledge database

Verification

Vehicle integration & calibration Function integration & testing  Sensitivity Analysis

Function level

System integration & testing

System level Component level Mechanic Software E/E

Implementation

Figure 1 – Methods within virtual development process according to the V-model.

After implementing the components, they are integrated and successively verified against the specification. This means, for instance, that the brake pressure is built up within the desired period of time or the function triggers when the driving dynamic conditions require it. The verification and testing process is extensive for the functional behaviour because of its complexity. Within the scope of verification and testing, the consideration of robustness is important too. By dint of simulations, the robustness of the whole vehicle variance is proved and quantified. The process according to the V-model closes with the characteristics of the full vehicle, which comprises the components. The characteristics are verified against the beforehand defined targets. The calibration of the brake control function is required to achieve the targets. The calibration process is challenging due to the high number of function parameters that define the control behaviour and interact extensively. Conventional methods meet this challenge predominantly by engineer’s expertise.

517

Integrated approach for the virtual development of vehicles equipped with brake control … However, the complexity and variety of today’s vehicles are acutely high why new methods allow both preserving expertise and better analysis of effects. Sensitivity analyses enable the identification of function parameters that have influence on the vehicle characteristic targets. The targets are defined objectively as characteristic values as mentioned beforehand. The identification of these parameters requires analysis methods that are able to deal with a large number of parameters. An established sensitivity analysis method to screen parameters is the elementary effects (EE) method. The decisive drawback of this method is that the number of simulations scales linearly with the number of parameters. The iterated fractional factorial design (IFFD) is a screening method that requires considerably less simulations to detect a few influential parameters within batches of hundreds or thousands [17]. The difference compared to the EE method or variance-based methods is the grouping of parameters. The parameters are allocated to groups randomly in every iteration and the parameters within one group are varied identically. A parameter that is frequently in an influential group is regarded as influential individually. This evaluation is conducted by a mathematical calculation. Based on such a screening, a variance-based sensitivity analysis provides a ranking of the remaining parameters according to their influence. Subsequently, the purposive calibration of the brake control system is conducted. The described analyses facilitate the conventional calibration process as well as virtual approaches.

2.2 Requirements for the Simulation Environment To enable the practical application of the proposed development process as described in Chapter 2.1 an appropriate simulation environment is necessary. The process comprises the whole development process starting with vehicle characteristic targets and closing with the complete vehicle considering the whole vehicle variance. From these framework conditions result the following requirements for the simulation environment. – Modular structure to enable, for instance, independent development of components and covering of the whole vehicle variety – Library of function, system and vehicle models to consider the whole virtual vehicle development process according to the V-model – Expandability by methods such as sensitivity analysis, system identification, calculation of characteristic values and comprehensive postprocessing – Support for structured validation process – Short computing times

518

Integrated approach for the virtual development of vehicles equipped with brake control …

2.3 Introduction of the Simulation Environment The simulation environment is composed of a modular structure to regard the framework conditions defined in the preceding chapters. This means that particular model elements are exchangeable according to the respective application. Figure 2 depicts both the structure of the model that corresponds to the one published by BRAUNHOLZ et al. [18] and the implemented methods. The modular basic structure consists of two main blocks: The active systems and the vehicle. The active systems consists of functions and systems and affect the basic vehicle through physical quantities, such as forces or angles. The active systems in turn react to vehicle dynamic quantities like yaw rate or velocity. The active systems are composed of functions and actuators, which are connected by a set value that the function transmits to the actuator.

Figure 2 – Modular structure of the simulation environment.

519

Integrated approach for the virtual development of vehicles equipped with brake control … The presented structure is strictly modular. The modules expandable and interchangeable. The components are chosen and parameterised by the user and the simulation model is built up automatically. Thus, the simulation environment meets the requirements on modularity, applicability within the whole virtual development process and covering of the entire vehicle variance.

2.4 Validation of the Simulation Environment In this section, an exemplary validation of the simulation environment is shown. In this context, the interactions between the brake control system and the active steering systems are considered in detail. The analysed vehicle is a luxury class sedan, which is equipped with a damper control, a rear-wheel steering and a superimposed steering at the front axle. The damper control operates in constant current mode. At first, the basic vehicle is validated, thus the brake control system is deactivated. Subsequently, the vehicle with brake control system is analysed. The outcomes base upon a nonlinear double-track model that interacts with several chassis control systems. The functions correspond with the original control unit code. Simplified models built up using elementary control loop elements predominantly model the actuators. Physical models are available if in-depth analysis of actuator behaviour is desired but increase simulation time significantly.

2.4.1 Basic Vehicle First, the validation of the steady-state driving behaviour is conducted by means of a steering wheel ramp. This manoeuvre is characterised by a slowly increasing steering wheel angle and driving with a velocity that is typical on a motorway. Figure 3 depicts the quantities steering wheel angle 𝛿 , the rear-axle sideslip angle 𝛽 and the roll angle with respect to the lateral acceleration 𝑎 in comparison of simulation (black line) and measurement (grey line).

Figure 3 – Steering wheel ramp manoeuvre to validate the steady-state vehicle behaviour.

520

Integrated approach for the virtual development of vehicles equipped with brake control … The steering wheel angle 𝛿 versus lateral acceleration 𝑎 characterises the selfsteering behaviour of the vehicle. The comparison shows that simulation and measure0.75𝑎 , . The curve has a noisy character in ment coincide, especially for 𝑎 the dynamic limits due to the tyre behaviour. The illustration of the rear-axle sideslip angle 𝛽 versus lateral acceleration 𝑎 characterises the vehicle stability. The curves of simulation and measurement coincide within the whole range. The last curve presents the comparison of simulated and measured roll angle 𝜑 with respect to the lateral acceleration 𝑎 . It should be noted that the compared curves deviate from each other especially in the area of medium acceleration up to 17 % and coincide for low and maximum accelerations. The seen deviation is not significant and often observed since the inclination of the vehicle dynamic test area is not expected to be zero. In addition to the steady-state validation of the vehicle behaviour, the dynamic characteristics are analysed. For this purpose, a sine-sweep steering wheel manoeuvre is evaluated. This means the vehicle is stimulated by a harmonic steering wheel input with constant amplitude and increasing instantaneous frequency. The initial frequency correlates with a slow steer manoeuvre and the final frequency with a highly dynamic one that considerably exceeds the yaw eigenfrequency. The steering wheel amplitude is determined in a steady-state driving manoeuvre with occurring lateral accelerations that allow linearisation of vehicle behaviour. Figure 4 shows the vehicle response to the described steering wheel input in the freΩ from steering wheel quency domain. First, the estimated transfer behaviour 𝜓⁄𝛿 angle 𝛿 to vehicle yaw rate 𝜓 in both magnitude and phase with respect to frequency Ω is depicted. This transfer behaviour represents the frequency–dependent selfsteering behaviour of the vehicle. It significantly determines the reaction of the vehicle to steering wheel inputs. The mathematical description as a transfer function is strictly speaking only valid for linear systems. Even though vehicle behaviour is characterised by several nonlinearities, the appraisal of the coherence proves the validity of the approach. The curves of simulation and measurement coincide in the whole frequency range. Consequently, the static gain, the amplitude peak and the eigenfrequency are simulated accurately. Furthermore, the phase response deviates at most 7 %. In addition, Figure 4 shows the transfer behaviour 𝑎 ⁄𝜓 Ω from yaw rate 𝜓 to lateral acceleration 𝑎 with respect to the excitation frequency Ω. This curve represents the relationship between front axle and rear axle dynamic. Regarding the amplitude curve, and the shape at high frequencies cothe static gain, the minimum near Ω 0.5 Ω incide. The eigenfrequency that is recognizable in the measurement is barely visible in the simulation and a deviation of approximately 5 % occurs. Considering the phase curve, it should be noted that simulation and measurement concur apart from one significant deviation in the minimum of the curve that represent 28 %.

521

Integrated approach for the virtual development of vehicles equipped with brake control …

Figure 4 – Sine sweep manoeuvre to validate the dynamic vehicle behaviour.

In Figure 4 the relationship between vehicle rear axle sideslip angle 𝛽 and lateral acceleration 𝑎 is shown. The sideslip angle is determined using a satellite-supported inertial platform that estimates the angle with the aid of a Kalman filter. Surveying the amplitude curve, a striking divergence between simulation and measurement is recognisable in the range of static gain. The reason is an offset drift in the sideslip angle signal that occurs during several measurements.

2.4.2 Vehicle with Brake Control Systems An established manoeuvre that triggers interventions of the brake control system is the double lane change. Figure 5 depicts the vehicle quantities with respect to time. The steering wheel angle 𝛿 is read in from measurement and is thus congruent for simulation and measurement. The comparison of simulated and measured lateral acceleration 𝑎 shows a good conformity. The curves coincide qualitatively and the deviation at the peaks are less than 10 %. The maximum acceleration is an indicator for the attainable lateral offset and thus relevant for critical evasion manoeuvres. Moreover, the figure shows the vehicle yaw rate 𝜓 versus time. The yaw rate is important for assessing vehicle behaviour according to ECE R 13H and thus for homologation. It is noticeable that the ascending branches of the maxima equal in simulation and measurement. However, the simulated curve does not reach the magnitude of the measurement and have a faster decline. This

522

Integrated approach for the virtual development of vehicles equipped with brake control … circumstance is explained by occurring deviations regarding the brake interventions of the brake control system, as will be discussed later. The sideslip angle curve progressions of the simulated and measured curve have some considerable differences. The positive amplitudes deviate by maximum 30 % from the measurement regarding the second peak. The negative amplitudes have deviations up to 47 % at the first peak. The sideslip angle estimated by a Kalman filter differs distinctly depending on the considered measurement and an offset exists that is compensated in the figure. The simulated and measured roll angle 𝜑 coincide except for a remarkable discrepancy near the second peak relating to both amplitude and dynamic. This tendency can be seen in every comparison, yet in different extents.

Figure 5 – Double lane change manoeuvre to validate the dynamic vehicle behaviour.

The behaviour of the vehicle is highly dependent on the chassis control systems. These are the damper control (constant current mode), the all-wheel steering and the brake control system. The simulation of the superimposed steering, the rear-wheel steering and the brake control system is analysed in detail. Figure 6 depicts the superposition with respect to time for measurement and simangle of the superimposed steering 𝛿 ulation in comparison. It is noticeable that the curves not entirely coincide and a deviations of 27 % on the second peak and 44 % on the last peak occur. This is due to utilising an analogous model of the superimposed steering function.

523

Integrated approach for the virtual development of vehicles equipped with brake control … Figure 6 also shows the angle of the rear-wheel steering 𝛿 with respect to time. Simulation and measurement have a considerable high coincidence. The behaviour of the rear-wheel steering is important since it influences the dynamic behaviour of the vehicle. Finally, the validation of the brake control system is presented in Figure 7. It shows exemplarily the brake pressures of the front axle on the left and right side 𝑝 , 𝑝 with respect to time in comparison between simulation and measurement. Wheel individual pressure sensors measure the values. The rear right sensor is faulty and the measurement is not utilisable for the analysis.

Figure 6 – Double lane change manoeuvre to validate the behaviour of steering systems.

It is important to note that the comparison between simulation and measurement shows a considerable deviation within a wide range. The pressure curves have three perceptible peaks. Regarding instant of time and gradient, the pressure build-up phases of the simulation coincide with the measurement. That is, the function triggers at the same time comparing simulation and measurement.

Figure 7 – Double lane change manoeuvre to validate the brake control system.

524

Integrated approach for the virtual development of vehicles equipped with brake control … The emerging peak heights and the subsequent pressure drops show several discrepancies. The built up pressure is maintained for a longer period in the measurement. This is caused by a different intervention of the brake control function. A possible reason is a slightly different vehicle behaviour and thus a different control intervention. Besides, the simulation processes ideal measurement data. This means that quantities such as the sideslip angle are available without any error or uncertainty. Consequently, the internal processed data differ between simulation and measurement even though the vehicle is modelled accurately. This leads to divergent functional behaviour and thus different brake interventions. Since suppliers provide the brake control system, it is a black box that internal signals cannot be analysed in detail.

3 Exemplary Application This section illustrates the application of the introduced simulation environment. First, the application of a sensitivity analysis is shown to identify influential function parameters in terms of vehicle dynamic behaviour. Subsequently, a parameter variation on the basis of the identification is pointed out. As explained in Chapter 2.1, the brake control system has several thousand parameters that determine its behaviour. By means of a screening method, the most influential parameters are identified. The applied method is the IFFD as described in Chapter 2.1, based on SALTELLI [17]. Figure 8 depicts the results of such a parameter screening in terms of the so-called main effect (ME) and quadratic effect (QE). The analysis refers determined by dint of a sine with dwell exemplarily to the maximum yaw rate 𝜓 manoeuvre. The method is equally applicable to other characteristic values.

Figure 8 – Results of an IFFD-screening applied to the function parameters.

525

Integrated approach for the virtual development of vehicles equipped with brake control … The abscissa shows the name of anonymised function parameters. The ordinate shows the normalised parameter impact on the characteristic value. The bars represent the main effect (black line) and the quadratic effect (grey line) and are an indicator of the influence a parameter has on the characteristic value, in this case the maximum yaw rate. The figure presents the ten parameters that have the highest main effect. The method provides values for the whole number of parameters though. The parameters are ranked according to the main effect the IFFD method provides. The ranking based on these effects is strictly speaking not valid but helpful to demonstrate the method. A valid parameter ranking is generated afterwards by means of a variance-based sensitivity analysis that is applied on a sub-group of the parameters that results from the IFFD method. The variance-based analysis identifies the parameter X1586 as most influential. A comparison with the results of the IFFD method shows that the parameter has the fifth largest main effect regarding the maximum yaw rate. In addition to the mathematical, abstract analysis, the simulation environment allows users to change parameters and analyse the resultant physical vehicle quantities. Such a parameter variation is shown below. Figure 9 depicts the yaw rate and the sideslip angle of the vehicle versus time. Two setups are compared. Setup 1 depicts the initial setup and Setup 2 represents an adjusted one. The changed parameter is the one identified above, thus X1586. It represents the maximum lateral acceleration, and consequently limits the maximum yaw rate.

Figure 9 – Influence of parameter variation on the vehicle behaviour.

526

Integrated approach for the virtual development of vehicles equipped with brake control … This means a decrease of the selected parameter reduces the yaw rate limit and thus leads to a smaller sideslip angle. This relationship shows Figure 9. The black curve represents the adjusted setup. The yaw rate 𝜓 reaches lower peak values without any distinct change in the qualitative behaviour. Similar findings can be seen regarding the sideslip angle 𝛽. The adjustment leads to at most 37 % deviation in terms of the sideslip angle amplitude of the fourth peak. The sideslip angle curves coincide qualitatively for both variants except for the shape of the fourth peak. In addition to the analysis on the vehicle level, the influence of the parameter adjustment is analysed on the function and system level in terms of, for example, brake pressure or internal functional signals. The difference between the vehicle setups is caused by the variation of only one function parameter. Higher deviations would be attainable if the variation range increases or further parameters are considered. It should be underlined that the function has more than thousand parameters and only one is adjusted. This makes clear that simulations reproducibly enable the unambiguous assessment of the influence a parameter has. Thus, simulation is ideally suitable for the examination of cause-effect relationships. In contrast, such an analysis is hardly possible in a physical vehicle.

4 Summary and Outlook The paper presents the required methods to develop vehicle with brake control systems with the aid of virtual methods according to the established V-model. Two essential methods are objective evaluation of manoeuvres that trigger brake control system interventions and sensitivity analysis. Sensitivity analysis facilitate the identification of crucial vehicle variants and influential function parameters. In this regard, the necessity of a modular simulation environment is presented in detail and the implementation of such an environment is shown. The environment meets the defined requirements in terms of modularity, several model accuracies, computing time and implementation of methods for various applications. Further, the validity of the environment is demonstrated. For that purpose, comparisons between simulations and measurement data are analysed for several steady-state and dynamic manoeuvres. This comparison shows that the simulation is capable for both the basic vehicle and the vehicle equipped with all chassis control systems. Whereas the active steering systems are simulated accurately, the simulation of brake control system interventions do not coincide with the measurement completely. Consequently, the sole virtual development up to the full vehicle level is not possible yet. Finally, the ability of the simulation environment to adjust function parameters and analyse their influence on the vehicle dynamic behaviour is shown. The application of the sensitivity analysis methods and the structured analysis of the interaction effects between brake control systems and further chassis control systems

527

Integrated approach for the virtual development of vehicles equipped with brake control … will be presented in a separate paper. The support of the development process by formal languages and the establishment of a knowledge database as proposed by SCHARFENBAUM is a further topic [15]. Besides, the objective evaluation of driving behaviour regarding vehicles with brake control systems is not conclusively researched yet, especially how to define the OEM’s specific driving behaviour.

Bibliography [1] E. K. Liebemann, K. Meder, J. Schuh and G. Nenninger, “Safety and Performance Enhancement: The Bosch Electronic Stability Control (ESP),” Convergence International Congress & Exposition On Transportation Electronics, 2004. [2] European Union, “Regulation No 661/2009,” 2009. [3] H. Winner, S. Hakuli, F. Lotz and C. Singer, Eds., Handbuch Fahrerassistenzsysteme, 3rd ed., Wiesbaden: Springer, 2015. [4] A. Lie, C. G. Tingvall, M. Krafft and A. Kullgren, “The Effectiveness of Electronic Stability Control (ESC) in Reducing Real Life Crashes and Injuries,” Traffic Injury Prevention, vol. 7, no. 1, 2006. [5] T. Pete, A. Chouinard and J.-F. Lécuyer, “A study of the effectiveness of Electronic Stability Control in Canada,” Accident Analysis and Prevention, vol. 43, no. 1, 2011. [6] A. Wagner, “Potentials of virtual chassis development,” 14. Internationales Stuttgarter Symposium, 2014. [7] K. M. Hahn, H. Holzmann, F. Weyer, M. Roemer, J. Webb and S. Boltshauser, “Simulation-based Certification of ESC Systems for Passenger Vehicles in Europe,” SAE International Journal of Passenger Cars – Electronic and Electrical Systems, vol. 5, no. 1, 2012. [8] Y. Mao, J. Wiessalla, J. Meier, W. Risse, G. Mathot and M. Blum, “CAE Supported ESC Development/Release Process,” Proceedings of the FISITA 2012 World Automotive Congress, vol. 7, 2012. [9] A. Lutz, B. Schick, H. Holzmann, M. Kochem, H. Meyer-Tuve, O. Lange, Y. Mao and G. Tosolin, “Simulation methods supporting homologation of Electronic Stability Control in vehicle variants,” Vehicle System Dynamics, vol. 55, no. 10, 2017. [10] European Union, “Regulation No 13H,” 2014.

528

Integrated approach for the virtual development of vehicles equipped with brake control … [11] M. Fach, J. Breuer, F. Baumann, M. Nuessle and T. Unselt, “Objective Assessment Methods for Wheel-Brake-Based Systems of Active Safety,” XXV. Internationales [My]-Symposium, 2005. [12] M. Gerdes and S. Dittrich, “Objektive Bewertung der Stabilität von Fahrzeugen mit ESP,” VDI-Berichte Nr. 1931, 2006. [13] International Organization for Standardization, “ISO 3888-1,” 1999. [14] Verein Deutscher Ingenieure, “VDI 2206,” Beuth, Düsseldorf, 2004. [15] I. Scharfenbaum, Funktionale Grundauslegung von Fahrwerkregelsystemen in der frühen Entwicklungsphase, Göttingen: Cuvillier, 2016. [16] R. Sell and M. Tamre, “Integration of V-model and SysML for advanced mechatronics system design,” Int. Workshop on Research & Education in Mechatronics, 2005. [17] A. Saltelli, T. H. Andres and T. Homma, “Sensitivity analysis of model output. Performance of the iterated fractional factorial design method,” Computational Statistics & Data Analysis, vol. 20, 1995. [18] C. Braunholz, W. Krantz, J. Wiedemann, I. Scharfenbaum, U. Schaaf and A. Ohletz, “Vehicle simulation environment enabling model-based systems engineering of chassis control systems,” 18. Internationales Stuttgarter Symposium, 2018.

529

New entry OEM – A global phenomenon David Ludwig Magna Steyr Engineering Graz, Austria

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 M. Bargende et al. (Hrsg.), 19. Internationales Stuttgarter Symposium, Proceedings, https://doi.org/10.1007/978-3-658-25939-6_43

531

New entry OEM – A global phenomenon

1 Motivation The number of so-called new entry OEM is growing constantly and is not only restricted to the Chinese marketplace (refer to figure 1). Besides, the business motivation for these companies to enter the automotive markets differs significantly. This presentation deals with the new entry OEM’s heritage, motivation and the technical challenges that have a significant impact on the automotive industry and engineering service providers like Magna Steyr. This presentation describes two different approaches – the “Uber”-approach and the “Tesla”-approach. Uber-approach companies try to use autonomous driving technology to distribute business models which basically consist of user data marketing and mobility as a service. These companies are not really interested in a respective vehicle differentiation but in the implementation of their AD devices. Dealing with vehicle development this leads to specific technical questions and challenges. Tesla-approach companies use the driving forces of the electromobility hype, here we have a local concentration not only, but especially in China. In their requirements and questions these new entries are more similar to traditional OEM, as they are less interested in fleet management etc. unlike Uber-OEMs. Nevertheless, also here we face distinctive questions and challenges that are different from traditional OEMs.

Figure 1: Agglomeration of new entry OEMs all over the world

532

New entry OEM – A global phenomenon

2 The “Uber”-Approach 2.1 Business Model and Requirements Companies with the “Uber”-approach focus more on fleets than on introducing new models to the market. As they do not have an automotive footprint yet, their business model is to sell mobility as a service and data marketing and do not follow conventional automotive rules. They are perceived of what is called the disruptive evolution in today’s automotive industry. “Uber” companies extend their focus to achieve cost efficiencies and to meet CO2 targets. This includes reducing driver costs (currently >50% of total costs), reducing operational costs and meeting legal demands. To achieve this, they undertake massive efforts in manpower and investments to develop their AD (autonomous driving) master intelligence unit (named “brain” in figure 2), which basically consists of sensors and a central processing unit to implement sensor fusion and algorithms for autonomous driving. As a consequence, they do not focus on the must-have activities in vehicle development and approach engineering service suppliers such as Magna Steyr, basically searching for quick solutions to set up their fleets, not driven by conventional industry standard KPI like costs per piece but by costs per mile. To fulfil the unusual requirements especially regarding timeframes (~1.5 years), the engineering service basically consists of finding a suitable (and accessible) donator platform with maximum carry-over content to integrate the master AD unit. This leads to specific technical issues which are described in the next chapter.

Figure 2: “Uber” companies need a system integrator

533

New entry OEM – A global phenomenon

2.2 Challenges of the “Uber” Approach The functionality of the AD unit is always kept a secret by the new entry OEM, so it is up to the engineering partner to perform black box system integration and validation that is able to fulfil functionality requirements of the complete vehicle. To achieve these highly challenging goals a complex validation chain has to established, starting with MiL simulation, HiL tests and AD simulation. As these are laboratory they are verified on real-life data on proving grounds and (public) roads. Taken alltogether, this data is used to do a performance analysis and finally parametrize the respective AD functions (refer to figure 3). Despite the technical challenges there are other non-development related challenges to overcome. First and most important, the legal situation – at least in most countries – is unclear and very often contradictory to the timing requirements. And finally, there is the search for a donator vehicle platform for which partnerships are required. Experience shows that they can be hard to establish. One major reason for this is that these partnerships cause product interdependencies which are very often unexpected and underestimated by the new entry OEM. Only to mention a few as exemplary examples: How to deal with technical platform or lifecycle adaptations as they can cause a chain reaction in both partner’s products? How to maintain the vehicles as they need the same maintenance infrastructure (diagnostic tester, flashing, cod

Figure 3: Development challenges – validation

534

New entry OEM – A global phenomenon

3 The “Tesla”-Approach 3.1 Business Model and Requirements Unlike the “Uber” companies the “Tesla” companies deal with electromobility vehicle products, i.e. Battery electric vehicles (BEV) with or without range extender (REX), Fuel cells (FC) or plug-in hybrid electric vehicles (PHEV). With these concepts they are by far closer to the big role model Tesla and other (traditional) OEMs including all advantages and drawbacks (refer to section 3.2). Very often these companies are privately owned, sometimes they occur to be syndicates of companies. One thing they all have in common which is to set up their brand, development and manufacturing they utilize government subsidies. It appears that these companies very often have some value-add – derived from the heritage of the company or its core product (e.g. entertainment, consumer electronics, connectivity, battery, etc.) – which they are keen to maintain also for their new vehicle product. Unlike the “Uber” companies the “Tesla” are much more driven by product development, whereas they try to differentiate in different areas such as styling, HMI, passive safety or driver assistance systems – as mentioned, this differentiation can be also influenced by the heritage of this company. For engineering suppliers such as Magna Steyr this means that – additionally to the system integration - they serve as system developers, as the value-add systems might not be automotive grade or other differentiation elements need to be developed (refer to figure 4).

Figure 4: “Tesla” companies need a system developer

535

New entry OEM – A global phenomenon

3.2 Challenges of the “Tesla” Approach Same as “Uber” companies the “Tesla” approach new entry OEMs are not experienced in the automotive field. But unlike the “Uber” companies they are in need of a complete vehicle and manufacturing turn-key solution. As their product will be introduced to the market as a brand they need to set up a development with simultaneous engineering (SE) work including all interfaces, i.e. logistics, purchasing, plant integration, quality assurance etc. This needs a lot of fast know-how, investment build-up and training already during development stage. Additionally, the before mentioned core components need automotive qualification which also means a lot of development and especially industrialization support. Furthermore, for the non-core components things can be even more difficult as new entry OEMs are not number one priority for Tier-1 suppliers and it can be very hard to find a reliable sourcing for every component in their BOM. Finally, “Tesla” companies have to consider their maintenance strategy already in early development stages – which is very often underestimated or even neglected. Creating maintenance strategies for a product lifecycle after SOP affects architecture developments considering diagnostics or variant management. All these challenges have to be tackled under the monitoring of the funding and/or subsidy stakeholders which push for fast results. Beyond the challenges described above, there are two additional challenges which are special about setting up business for the Chinese market. First, there is the limited amount of EV manufacturing licenses in China which forces these companies – if they do not own one – into collaborations or joint ventures if they do not want to be forced to produce outside China. Additionally, there is still a strong dependency on Chinese high voltage battery suppliers, despite the fact that Chinese government started to release the ban of foreign suppliers by neglecting them on the so-called “whitelist” in 2018. Nevertheless, it is still unclear how things will evolve in the next years.

4 Summary and Conclusion The new entry OEM is a global phenomenon that is not only restricted to the Chinese market. We encounter two mainstream approaches that differ significantly in their impact to vehicle engineering and beyond. Nevertheless, these two approaches face a lot of (different) challenges on their way to stand its ground in the automotive marketplace which needs a lot of involvement of independent engineering service suppliers. As a forecast, it is rather unlikely that all new entry OEMs – independent from their approach – will survive, those who can deal more clever with the different challenges will succeed.

536

Consistent application of systems engineering and simulation for cross-domain function integration Marcus Boumans, Dr. Martin Johannaber, Ulrich Schulmeister Robert Bosch GmbH

Consistent application of systems engineering and simulation for cross-domain …

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 M. Bargende et al. (Hrsg.), 19. Internationales Stuttgarter Symposium, Proceedings, https://doi.org/10.1007/978-3-658-25939-6_44

537

Consistent application of systems engineering and simulation for cross-domain …

1 Challenge & Motivation The automotive megatrends (electrification, automation, connectivity) and new mobility concepts lead to an increasing interaction of mechatronic systems with their surrounding systems and finally to an increased interaction with the entire vehicle. This leads to changed boundary conditions for OEMs and Tier1 suppliers.

1.1 Increase of cross-domain functions The creation of new, innovative customer functions are more and more based on the interaction of formerly separate vehicle domains. Examples are the use of the braking system for an advanced steering behavior or the use of the electric drive system for improved braking functions in electrified vehicles. The most obvious example of such cross-domain functions is highly automated driving. The driving task has to be completed by an effect chain that starts with the perception of the environment by a suitable sensor cluster, followed by a situation analysis and driving trajectory planning by advanced computer algorithms and finally the correct control of the steering-, powertrainand brake subsystems by deeply embedded control units and control algorithms. The development of those new functions requires a suitable cross-domain engineering.

1.2 Changed sourcing of car manufacturers and mobility providers For reasons of risk minimization and encapsulating complexity, some established OEMs are more and more requesting system solutions for e.g. an electrical axle or an automated driving kit from a Tier 1 supplier rather than sourcing single components. New players enter the automotive market with new business models. They expect suppliers to have functioning system solutions at a higher level of integration e.g. a rolling chassis with which they can provide their mobility services. In addition, there are generally higher demands for faster time to market.

1.3 Increased complexity Cross-domain functions together with increased requirements regarding safety and security lead to a highly increased technical complexity (more system elements, more interfaces). Furthermore, automotive development organizations are typically structured along to the classical vehicle domains. For the development of cross-domain functions, this vertical structuring leads to an additional organizational complexity (distributed knowledge, responsibilities and financing models).

538

Consistent application of systems engineering and simulation for cross-domain … For the Robert Bosch GmbH as an automotive supplier, it is essential to respond to these challenges to be able to offer competitive products and services and to be an attractive development partner on system level as well.

2 Approach The above-described challenges require a holistic vehicle view also for a Tier1 supplier to develop competitive products and services (components, subsystems, higher integrated systems). Hence, the Bosch Business Sector Mobility founded a systems engineering (SE) organization for the development of cross-domain vehicle functions subsequently called BBM-SE. The mission is to build up cross-domain knowledge and establish engineering competency on vehicle level. The systems engineering approach is starting with scoping, use cases and stakeholder analyses on vehicle level. BBM-SE identifies new features and derives system requirements, system architectures and finally component requirements for these. This front loading approach helps to achieve an optimal fit between the Bosch subsystems and components and minimizes integration effort. In order to accomplish that task there are two methodical key elements for efficient and effective cross-domain systems engineering (see Figure 1): 1. A MBSE (Model Based Systems Engineering)-framework as an enabler for the usage of a cross-domain model based systems engineering approach to handle the increased complexity, reduce the risk of improper/missing product features and minimize the integration effort. 2. A cross-domain vehicle simulator for a fast and robust assessment of solution variants and elicitation of requirements addressing time to market and development costs.

Figure 1: Overview BBM-SE

539

Consistent application of systems engineering and simulation for cross-domain … BBM-SE works in innovation projects where the methodical key elements are applied and if necessary refined. This step is an important prerequisite for a later roll out of the methods to series production projects. The following two subsections give a deeper insight into the methods. Section 3 discusses an example for the usage of the methods.

2.1 Cross-Domain Model Based Systems Engineering 2.1.1 Short Introduction to MBSE Systems engineering (SE) is a well-known, systematic, top down approach for the development of complex technical systems. It separates the problem analysis from the solution finding and starts with a clear description of the system under development and its context. Requirements are derived by focusing on stakeholder needs. In order to handle complexity, a systematic decomposition of the system into smaller subsystems and the definition of their interfaces is mandatory (s. Figure 2). Based on the solution-independent analysis of the system behavior system designs and their realizations are developed in the logical & technical viewpoints.

Figure 2: Key principle to handle complexity

In MBSE, the results of these engineering steps are stored in a computer-analyzable system model. This leads to a consistent and redundancy free system description and enables the reuse of artefacts in the different engineering tasks e.g. reuse artefacts from the functional viewpoint for quality measures like a FMEA. Impact analysis of changed requirements can be performed with a high confidence level compared to a manual, document based process. In addition to handling the complexity, MBSE reduces the risk of missing out requirements and therefore select a non-optimal solution and lowers the integration effort respectively the number of iterations [1] [2].

540

Consistent application of systems engineering and simulation for cross-domain …

2.1.2 MBSE-Framework For the introduction and application of the model-based systems engineering approach, BBM-SE develops and maintains a MBSE-framework that has to fulfill following toplevel requirements: 1. The method must be applicable for the development engineers (usability) 2. The framework has to support the description of architectures 3. The information in the different artefacts has to be stored in a consistent and redundancy free way 4. Non-formal (textual) as well as formal requirements engineering has to be supported 5. Traceability between requirements, design elements and tests has to be possible 6. Reuse of artefacts in different projects must be supported 7. The method has to enable the organization of technical and organizational complex projects by defining structure and organizing interfaces. In order to satisfy these requirements the MBSE Framework follows the principles described in Section 2.1.1. A MBSE approach based on the SysML modeling language has been chosen. SysML is used for the formalized documentation of requirements and the description of the different architectural viewpoints. For the textual, non-formal description of requirements, standard requirements engineering methods are used. A meta-model is required to ensure consistency between the data that is gathered during the engineering process. It describes which engineering artefacts have to be created and how they depend on each other. For MBSE several meta-models are available, e.g. SYSMOD, FASE or Harmony SE. Our MBSE-Framework uses a meta-model based on the Software Platform Embedded Systems 2020 (SPES). SPES was developed with focus on mechatronic systems and covers important aspects for the development of automated and connected systems like safety and open context problems and therefore suites very well to the demands of BBM-SE. Figure 3 shows an overview of the basic idea of SPES methodology.

541

Consistent application of systems engineering and simulation for cross-domain …

Figure 3: Overview SPES meta-model

The core concept uses granularity layers and different views and viewpoints to describe the system. For a detailed description of the SPES approach, see [3][4].

Figure 4: MBSE-Framework

As shown in Figure 5, the MBSE-Framework consists of two main pillars. 1. The MBSE-Platform contains the adaptation of the SPES meta-model to the BBM-SE specific use cases and demands. For a concrete application of MBSE, a tooling for modeling, requirements engineering and traceability is required. The platform uses tool environment that allows the linking between requirements and design elements. In addition, further methods and processes like an interface concept, safety analysis, variant management, and verification respectively validation methods are contained.

542

Consistent application of systems engineering and simulation for cross-domain … 2. The MBSE-Library collects artefacts for reuse in projects. It provides for example, templates, profiles and stereotypes that permit an easy application of MBSE. The building block element contains reusable requirement documentation and reusable architecture elements. The library also contains reference systems. Reference systems contain a complete set of models and requirements for a concrete system, e.g. the energy management for a battery electrical vehicle and can be used as starting point for new development tasks. The combination of the MBSE-Platform with the base methods and tooling together with the MBSE Library with reusable elements is a key enabler for the application of MBSE in cross-domain development projects.

2.2 Cross-Domain Simulation The increased connectivity and interactions between the software controlled systems also increases the scope and complexity and of cross-domain simulation environments used for development and testing. However, simulation is the only way of developing and testing connected software functions fast and early enough in the development process. Since the scope to be modeled is usually very high, one major target is the re-use of models and the integration of the actual control software in the models. For this we use a dedicated co-simulation middleware as integration platform as depicted in Figure 5.

Figure 5: Cross-domain co-simulation platform

543

Consistent application of systems engineering and simulation for cross-domain … It enables the integration of models from different authoring tools and compiled modules of the real controller software. It also enables the connection to state-of-the-art commercial simulation tools like CarMaker from IPG to easily realize a complete vehicle simulation including the environment of the vehicle. The middleware provides a coupling algorithm to reduce numerical errors induced by the co-simulation of several solvers coming with the different authoring tools. More details on numerics of co-simulations and algorithms to reduce numerical errors can be found in [6]. Furthermore, the co-simulation environment has to provide an easy handling for the simulation users even with the high complexity of a multi-tool cross-domain simulation. This is realized by the automation of tasks like script-based, automatic configuration and execution of the co-simulation. The use of preexisting models from different domains and authoring tools significantly speeds up simulation work, even if the models must be adapted for new or changed requirements. The models from the different domains need to be provided in a coordinated way and the support of the different domain and model experts is still required after model provisioning. The BBM-SE approach for this cross-domain and cross-sectoral collaboration is to set up a dedicated simulation team maintaining the cross-domain simulation, which we call xDomain Vehicle Simulator. Part of the xDomain Simulator (see Figure 6) is a cross domain model library, a simulation platform with technical solutions and documentations and integrated complete vehicle models called vPFD (virtual Platform Demonstrators). In addition to the models itself the technical solutions and documentations are crucial for the cross-domain simulation. Besides the co-simulation middleware provisioned in an easy to use way, it comprises a central repository suitable for collaboration in a team, i.e. a Git repository. A further element is the architecture development and specification based on SysML models, an information hub in the Bosch intranet and methods for providing control software in the simulation. An especially important and helpful element for the common development of simulation models in a team is continuous testing, as it is common in software development. This means every time a change is committed to the repository the simulation models are automatically checked out, simulated on a Jenkins server, a collection of tests is performed and test reports are automatically generated. This speeds up and simplifies the integration of changes in the main development branch and increases quality by the continuously execution of an ever increasing number of tests.

544

Consistent application of systems engineering and simulation for cross-domain …

Figure 6: Cross Domain Vehicle Simulator and organizational set-up

The library contents and the vPFDs features are driven forward by simulation use cases. For this systems engineering methods are used again for the elaboration of the simulation use cases, derivation of the model requirements and definition of the model architecture. Although the cross-domain simulation platform and libraries offer a very flexible and modular simulation solution, the challenge for the simulation still is to decide which simulation tool and environment is most suited to accomplish a task. The possibility of future reuse should always be considered in the decision.

3 Example Recuperative Braking The application of the described engineering method to a cross-domain function development is shown taking recuperative braking as an example. Recuperative braking is a state of the art function of modern electrified vehicles. The electrical propulsion system offers the possibility to store energy in the electrical storage system allows decelerating the vehicle not only by using the friction brake, but also by using the electrical machine as generator. The generated energy is stored for later propulsion and therefore energy is saved. As the propulsion system, the electrical storage system and brake system have to work together in a coordinated way recuperative braking is a typical example for a function that acts across different domains. After scoping and stakeholder requirements, analysis the next step in the MBSE process is to describe the functional viewpoint of the system. The top level of this viewpoint is documented in a Rhapsody SysML model as shown in Figure 7.

545

Consistent application of systems engineering and simulation for cross-domain …

Figure 7: Functional viewpoint of a recuperative braking system

The required functions are described in a solution independent way. The function decomposition comprises sensing the driver demand, provide vehicle systems state (e.g. the state of charge of the energy storage). Based on the information an energy distribution between electrical and mechanical brake energy is calculated and propagated to the respective energy conversion functions. Based on the functional analysis solution designs can be derived in the logical viewpoint, see Figure 8. Here different solution concepts have to be analyzed and assessed.

Figure 8: Logical viewpoint of the recuperative braking system

546

Consistent application of systems engineering and simulation for cross-domain … Figure 8 shows one possible solution where the functional elements are mapped to solution elements like a high voltage battery for storing the electrical energy and a ‘BrakeSystemActuator’ that operates a hydraulic system. The assessment of the different solution concepts is performed in a suitable simulation model. In our approach, the logical viewpoint in the model can directly be used to define a simulation model for this use case as shown in Figure 9.

Figure 9: SysML model for definition of the simulation setup

The simulation model of the complete vehicle can be used as a virtual prototype to do concept evaluations, integration tests and functional test of the software control in a closed loop setup. This way it also enables a first verification and validation of the function. The recuperative braking function shown here can freely distribute brake torques between the electric powertrain and the mechanical brake when the brake pedal is pushed. For this, a complex brake system is necessary, in this case consisting off a Bosch iBooster, ESP and a second actuator (to reduce mechanical brake torque when the brake pedal is pushed). Moreover, the control of this system has to interface with the control of the electric powertrain. Figure 10 shows exemplarily a simulation result, which can be used for testing the correct function of the recuperative braking control. The solid line shows the desired brake torque resulting from an interpretation of the brake pedal. At the beginning of this maneuver until roughly second 1.6, the electric machine is at high speeds in the field weakening area. Thus, the complete possible generative braking torque is applied simply by

547

Consistent application of systems engineering and simulation for cross-domain … the release of the acceleration pedal and the mechanical brake realizes additional braking torque. Between second 1.6 and 4.3 the possible torque from the electric machine rises with the decreasing speed and the brake torque from the mechanical brake is reduced. The limit of the recuperative braking torque of 1500 Nm results not from the maximum torque of the electric machine, but from a calibrated limit for a maximum deceleration by recuperation.

Figure 10: Simulation results from test of recuperative braking control

The simulation can not only be used for verification, but also to assess the energetic potential of different recuperative braking solutions in real life scenarios. This is possible with an extended simulation setup. Figure 11 shows how this is done for an assessment of the energetic benefit of the recuperation systems. For this the vehicle simulation is coupled to a traffic flow simulation. This enables the fast creation of stochastic traffic scenarios. The complete setup consists of many different models and tools integrated in the co-simulation middleware. It integrates the detailed models of the electric powertrain, the mechanical brake system, the powertrain and brake controls, vehicle dynamics in IPG Carmaker and the traffic simulation in SUMO from DLR. This cross-domain co-simulation enables to consider the interaction of the vehicle with the surrounding traffic. The density and the initial conditions of the traffic can be varied and enable a statistical study of effectiveness of the recuperation and the resulting benefit in energy consumption.

548

Consistent application of systems engineering and simulation for cross-domain …

Figure 11: Cross Domain Vehicle Simulation for system evaluation under real conditions

The recuperative braking example shows how MBSE and simulation can be used to specify and develop a complex software intense system across different domain and disciplines. Especially highly automated driving systems have many more interfaces and interactions, which make the application of these methods mandatory.

4 Outlook For a series development of cross-domain functions, further steps have to be taken. First, a common engineering process has to be defined that considers a holistic view on the vehicle and includes elements, which check how a new function fits into the overall system. In the next step, a system decomposition starting at vehicle level and a harmonized cross-domain interface definition for all involved development areas and vehicle domains have to be established. This is necessary to handle complexity by allocating requirements and responsibilities for smaller system elements to different development entities and by defining interfaces for their system of interest. With this approach, the different system elements can be integrated with minimal effort to a cross-domain system with the desired features. In order to combine development artefacts from different development departments and perform computer-based automated checks, rules for a common cross-domain MBSE process and method have to be defined. It is essential to find the optimum between a centralized governed standard and the flexibility in the adaption of the process to domain or customer specific demands. On the one hand, the standardization minimizes the integration effort, the risk for non-fitting products and enables cross-domain system analysis such as safety analysis. On the other hand, the effort for the common description has to be as small as possible and the development departments must be able to optimize their development processes in their specific manner. Finally, a seamless cross-domain engineering toolchain is required in which all the

549

Consistent application of systems engineering and simulation for cross-domain … described steps can be handled in a consistent and redundant free manner and information can be seamlessly exchanged between different project teams. For the simulation the target is to use the SysML model from MBSE seamlessly to generate the co-simulation models for the dynamic simulation of the system. For this, additional information and requirements have to be added to the model and the corresponding standards like SSP (System Structure and Parametrization, see [7]) have to be established. The next step in the area of cross-domain simulation is to use the simulator for virtual integration, verification and validation. For this, the functional models of the control software have to be replaced by close-to-series software and the simulation model has to be adapted to the parameters of the selected design. With this virtual testing environment, potential errors can quickly be identified and corrected before the actual vehicle integration. In the long term, the use of simulation is also aimed for system releases. This would increase development efficiency further. Moreover, the possibility of at least a partly virtual release is mandatory for some systems, e.g. for automated driving. However for simulation-based releases various questions such as the rating of model quality and suitability for releases must be clarified, see [5]. From our point of view MBSE and system simulation are crucial to meet the challenges in the automotive industry, which origin from the increasing complexity and the disruptions caused by electrification, connectivity and automation. Still it is a big challenge to establish them in the development process inside a big organization.

5 Bibliography 1. Embedded Market forecasters, Jerry Krasner, October 2015, How Product Development Organizations can Achieve Long- Term Cost Savings Using Model-Based Systems Engineering (MBSE) http://www.omgwiki.org/MBSE/lib/exe/fetch.php?media=mbse:how_product_development_organizations_can_achieve_long-term_savings_1_.pdf 2. TUM, Florian Deißenböck, Martin Fritzsche, Daniel Méndez Fernández, 12‘2010 Wirtschaftlichkeit der modellbasierten Entwicklung 3. Advanced Model-Based Engineering of Embedded Systems, Klaus Pohl, Manfred Broy, Heinrich Daembkes, Harald Hönninger, Springer 2016, ISBN 978-3-31948003-9 (eBook) 4. Systems Engineering II, Hermann Winner, Günther Prokop, Markus Maurer, Springer 2018, ISBN 978-3-319-61607-0 (eBook)

550

Consistent application of systems engineering and simulation for cross-domain … 5. Software-in-the-Loop simulation for early verification and validation of complex systems in the product life-cycle. Indrasen Raghupatruni, Andreas Thuy, Peter Baumann, Thomas Huber, Daniel Seiler-Thull, paper submitted for IEEE European Test Symposium 2019, Baden-Baden, Germany. 6. NEPCE – A Nearly Energy Preserving Coupling Element for Weak-coupled Problems and Co-simulation. Benedikt, M.; Watzenig, D.; Zehetner, J.; Hofer, A.. IV International Conference on Computational Methods for Coupled Problems in Science and Engineering, Coupled Problems 2013, Ibiza, Spain 7. MA - Project “System Structure and Parameterization” – Current Status and Plans, Jochen Köhler, Pierre R. Mai, FMI User Meeting 2017-05-15, Prague

551

Automotive megatrends and their impact on NVH Georg Eisele, Michael Kauth, Christoph Steffens FEV Europe GmbH Patrick Glusk FEV Consulting GmbH

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 M. Bargende et al. (Hrsg.), 19. Internationales Stuttgarter Symposium, Proceedings, https://doi.org/10.1007/978-3-658-25939-6_45

553

Automotive megatrends and their impact on NVH

Abstract The ongoing transformation of individual mobility will decisively change future vehicles. The acronym CASE, created by Daimler in 2016, sums up four megatrends in one abbreviation: Connected, Autonomous, Shared and Electric. This paper is meant to discuss the resulting consequences regarding NVH. The strong impact on vehicle NVH requirements is evaluated and suggestions for a suitable sound design are made. Sharing the same vehicle by many users will reduce the importance of creating an individual brand specific sound, instead active sound design would allow tuning the interior noise to the preferences of the current user. Electrified vehicles, which feature a very quiet drivetrain, create new challenges for the NVH development in several ways: unpleasant, high frequency noise from the drivetrain, less masking of disturbing noise such as road and wind noises, basic questions concerning brand sound with the completely new sound character of the electric drivetrain. NVH requirements for autonomously operating vehicles will differ significantly from conventional vehicles. The situation of the user is similar to that of a “train passenger”. Comfort oriented NVH features will become more important, acoustic load response is expected to lose relevance.

Introduction The automotive industry is facing substantial and disruptive changes. More than a few insiders consider the upcoming changes in the automotive world to be the greatest changes this branch of industry and life has ever gone through. Moreover, the change is not only about to happen in the future, it is already here. Driven by the increasing demand for the use of sustainable energy sources, the more and more complete digitalization of our world, the need for new mobility concepts to address progressing urbanization and population growth and, last but certainly not least, vast technological advances in the field of car2car, car2road and car2world connection and autonomous driving, the current automotive megatrends influence how the car of tomorrow will incorporate the many new developments and how it will differ from the cars we are currently used to drive. While it is yet unclear how exactly the car of tomorrow will look like, which features it will provide, and how the underlying technology will work in detail, it is obvious now that all this change will come. It is now that the members of the automotive industry and their development partners have to anticipate the course of events and use as much as possible the opportunities to shape the path ahead and to influence what the car of tomorrow will be.

554

Automotive megatrends and their impact on NVH Today’s automotive megatrends have been labelled CASE by Daimler [7], a term widely accepted in the automotive community. CASE stands for Connected, Autonomous, Shared and Electric. None of these megatrends is NVH. However, each of these megatrends will influence the NVH behavior and properties of tomorrow’s cars – some to a larger, some to a smaller extend. An overview of the global megatrends, drivers, new technologies and there effect on NVH is shown in Figure 1. Moreover, while for some developments the expected differences to today’s cars are rather obvious and the related NVH influence is already subject of detailed investigation and analysis (electrification), other developments will lead to changes in vehicle NVH that can hardly be precisely anticipated as of today.

Figure 1: Trends and drivers of the technological transition in automotive industry and their effect on NVH

How will the car of tomorrow look like, what will it be able to do, how will it sound? What will it be like inside of tomorrow’s car? To which extend can personal taste be addressed when cars are no longer personal property but have largely become shared

555

Automotive megatrends and their impact on NVH goods? What kind of environment will the automotive industry create inside of an autonomously driving car and which role does NVH play in this environment? How much and which information will be provided aurally, and how? To many of these questions, there is many answers or none, yet. However, the authors of this paper would like to offer some ideas of theirs and encourage the reader to develop some of his own.

Automotive Megatrends Since the fact that human-made carbon dioxide emission contributes to global warming has become scientifically widely accepted, strategies to reduce traffic-related CO2emissions have been a major factor in (conventional) powertrain and vehicle development. For some years, downsizing combustion engines to reduce fuel consumption was a most important field of development, driving and fueling a number of technologies’ research activities. Downsizing had and has significant influence on NVH aspects. Especially the drivetrain and the powertrain mount systems had to be adapted to address increased vibration excitation by higher combustion forces and in the case of cylinder number reduction, decreased excitation frequencies. Recently, electrification of the powertrain has become the strongest trend with regard to CO2-emmission reduction. Hybridization in varying degrees already is widespread in the automotive market. Sales numbers for hybrid vehicles increase, especially for plug-in hybrid vehicles. However, fully electric vehicles sales numbers are most strongly on the rise (Figure 2). Currently, the question whether there will be many fully electrical vehicles in the future has changed to the questions: How many, by when, and where? For different markets, very different distributions of electric vehicles are expected. China will likely solidify its lead in regard to the elective vehicle market share. A second trend that already is visible in many instances is the increasing degree of shared mobility. Ride sharing for example with colleagues living in close proximity between home and work has been well-known prior to the internet. In the late 90s and early 2000s, platforms evolved to offer and find ride-sharing with strangers. Recently, companies with strong commercial focus explore the field of shared mobility with enormous growth rates. The American transportation and delivery provider Uber reported net revenues of 6.5 billion $ in 2016 (with adjusted net losses of 2.8 billion $), up from 1.5 billion $ in 2015 and 495 million $ in 2014 [5]. Uber’s Chinese counterpart, DiDi Chuxing currently is valued more than 50 billion $ [8][5]. With regard to ongoing urbanization and population increase, the demand for public and shared mobility can be expected to increase at the cost of motorized individual traffic.

556

Automotive megatrends and their impact on NVH

3.217

38%

+65%

2.053 31%

23%

1.309 23%

10% 40%

431

2013

13%

759

5% 4% 4% 3%

31%

39%

2014

2015

6% 6%

27%

2016

China

Japan

France

Netherlands

USA

Norway

UK

Germany

12% 2017

Rest of the World

Figure 2: Global EV Fleet Development (‘000 units) [8]

Two more megatrends can be identified in the automotive world which are – to a large extend – linked to each other: connected vehicles and autonomous driving. Evolution and improvement of wireless communication standards, steadily increasing computational power and – in general – vast improvements in the field of artificial intelligence again changed the question from if to when and how. FEV expects the first fully autonomous cars in ~2027 with an expected market share of about 12 % in 2030. Fully autonomously driving vehicles can be expected to become a very different environment from conventional vehicles. With no need for a human driver, the way we use the time during travel might change very much, and so the way to design the vehicles will address this situation. Connecting the vehicles to each other, the road and the world not only is a prerequisite to enable autonomous driving, it will also provide much information of various kinds to the driver/passengers. This, too, will have an impact to the NVH properties demanded from the vehicle.

557

Automotive megatrends and their impact on NVH

Shared Different car brands and models distinguish themselves from others through various attributes. Besides rather objective properties like weight, power, fuel consumption, size or price, soft skills are even more important as selling points for many customers. Only a few people will buy a car that objectively fits to all demands if they dislike its looks. Exterior and interior design transport to a large degree the brand’s character. A car’s NVH behavior can be considered a comparable soft skill. The sound character inside and outside the vehicle transports quality perception and emotion. Different car brands often position themselves quite distinctively on a map of NVH characteristics. FEV uses evaluation criteria to describe a car’s NVH characteristic. Two very important parameters of these are applied to describe a car’s NVH: pleasantness and dynamic. Each parameter is determined by jury evaluations during vehicle benchmark projects. With these two parameters, a map can be spanned in which different cars are positioned. Figure 3 gives an impression on the spread regarding these ratings for serial cars from FEV’s NVH database.

Figure 3: Dynamic/pleasantness-ratings for benchmarked serial cars from different manufacturers and classes, potential targets for shared vehicles and estimated variation range for tuning through Active Sound Design

Generally, cars that are more dynamic tend to be observed less pleasant and vice versa. However, especially cars of luxury classes manage to combine dynamic and pleasant

558

Automotive megatrends and their impact on NVH characteristics – for a price. The customer chooses the car that – in his view – fits best to him: Technical specification, looks, price, and NVH. Manufacturers develop the vehicle’s NVH properties to meet their prospective customers’ expectations. If cars in future more and more are not owned and used by one customer but used by many participants of shared mobility, how should such a car be positioned for example on a map describing dynamic and pleasantness? A potential scenario is that future concepts for shared mobility are tuned to address some kind of uniform taste. It can be assumed that at least for NVH characteristics, such a uniform NVH characteristics layout might be designed to foremost accomplish a high rating of pleasantness while dynamic characteristics will be second or even fully discarded. A second scenario could be that there will be different cars with different characteristics available for the customer to choose from (compare Figure 4).

Figure 4: All purpose vehicle vs. specific purpose vehicle [9]

However, the span of variation of NVH characteristics would be significantly less than what we see in today’s automotive market. Figure 3 gives an example for such target values. Both scenarios mentioned above do not take into account the possibilities of individualization of each distinct car. Especially the more and more market-penetrating means of active systems, e.g. active sound design (ASD), offer numerous possibilities to tune the vehicle’s NVH character to current customer’s taste, see Figure 3. Current active sound devices influence vehicle NVH both for interior and exterior noise. One possible scenario is e.g. that the current vehicle user is identified by his chip-card, comprising also his personal NVH preferences. The systems in the vehicle, which allow

559

Automotive megatrends and their impact on NVH taking active influence on the interior noise, are then controlled in the way which is favored by the user. There are many possibilities of active control of interior noise: 1. Vehicle sub-systems with adjustable physical properties, e.g. engine mounts with variable stiffness/damping or exhaust system transfer behavior controlled by flaps 2. Mechanical systems, which create additional sound components: e.g. sound pipe from intake system to vehicle interior with flap control, mechanical shaker acting on vehicle boy 3. Additional sound emitted from the vehicle’s series audio system a.

to increase or reduce certain noise shares (engine orders) including feedback loop with additional microphones in the vehicle interior

b.

to increase certain noise shares based on the actual driving condition (engine load, engine speed, gear, …) in feed forward control (no feedback loop)

Especially with the latter, there is much potential to shape a car’s NVH behavior much to the individual driver’s taste. Variant 3b) currently is introduced in the market in some vehicles and a much greater spread is expected in the next years. Main reasons for this are on the one hand that even this simplified variant without feedback loop allows significant improvement of interior noise concerning pleasantness and especially regarding dynamic. On the other hand, the additional hardware costs are minimal, because most components are from the series HIFI-system and only an additional controller is needed.

Connected Increasingly connected vehicles will have access to much information of various kinds. Some of this information will be used solely by the car itself, for example for autonomous driving, and some of the information will be forwarded to driver and/or passengers. Such information will be presented visually, aurally or for example by movements of the seat or comparably. However, this largely is already the situation today. The increased connectivity thus is expected to influence tomorrow’s vehicles’ NVH rather indirectly by enabling other technologies like autonomous driving.

560

Automotive megatrends and their impact on NVH

Electrification The electrification of automobile propulsion is probably the most advanced of the here considered trends. Electrified vehicles, as hybrid version in combination with a combustion engine or as fully electric vehicle, have been successfully introduced in series production by many OEMs. Currently the most severe obstacles for a faster growth of the electric vehicle market share are battery price, electric driving range and the infrastructure for reloading the battery. Nevertheless, a significant growth of the market share of the different types of electric vehicles is expected in the next decades. Globally, expected are about 17% pure electric vehicle in 2030 – major market for battery electric vehicle is China with approx. 9 million units (Figure 5). For the year 2030, a market share of more than 70% is expected for vehicles, which allow pure electric drive: full hybrid, plug in hybrid and battery electric vehicle. The technology of the electrified vehicle itself is already advanced; nevertheless, many degrees of sophistication from NVH point of view are observed in the market.

Figure 5: Passenger car powertrain type forecast for 2030 in million units

Different subject areas arise when NVH of electrified vehicles is considered (Figure 6). Safety: Some countries already prescribed an additional vehicle warning noise for pedestrian safety. Here the challenge is not to lose the advantage of traffic noise reduction

561

Automotive megatrends and their impact on NVH with quieter electric vehicles by deliberately created annoying warning noise. In addition, the warning noise should not be perceived by the driver or passenger. Therefore a good isolation between the loudspeaker and interior has to be implemented. Annoying interior noise: Although electric vehicles are often considerably quieter than comparable vehicles with combustion engines, the interior noise is marked by highfrequent noise components, which typically are subjectively perceived as annoying and unpleasant. Tonal noise components are especially critical. Masking: Moreover, disturbing noise is no longer masked by combustion engine noise. That means, annoying noise from the drivetrain itself, as well as unpleasant noise shares from other vehicle systems come into the fore. Due to the missing noise from the ICE, road and wind noise will become more important. This frequency dependent background noise is used to define target lines for other noise shares from e-motor or transmission. Sound character / brand sound: There is a lot of experience in the automotive industry in designing the interior sound of vehicles with combustion engines. In addition, the customers have certain expectations concerning brand sound and the sound character regarding pleasantness and dynamic. Quiet electric vehicles offer new creative prospects, which have to be filled by the responsible engineers. More and more, synthetic sound is used to give the driver load feedback and thus create a dynamic sound impression.

Figure 6: NVH phenomena of electrified vehicles

562

Automotive megatrends and their impact on NVH The creation of a pleasant interior sound, which is the base for further elaboration of a more dynamic or brand sound character, requires effort on a wide range of vehicle subsystems: e-motor, generator, transmission, vehicle body, battery and cooling system, chassis and auxiliaries. Even though the vehicle is different, the general procedure can be adopted from conventional vehicles. Subsystem targets are derived from predecessor, competitor or more systematically from full vehicle targets by target cascading using a physical transfer path vehicle model [1], [2]. CAE methods like Finite Element Analysis or Multi Body Simulation need to be extended by modules considering electro-magnetic forces and are applied in the early development phases for subsystem design. When hardware is available, the simulations are validated and complimented by hardware testing and analysis. Mule and prototype vehicles are analyzed by transfer path analysis to identify critical NVH topics as soon as possible. Vehicle components and related NVH topics are e.g.: ● E-motor and generator: Whining noise created by magnetic forces ● Magnetic forces excite e-motor housing ● Fluctuating torque due to non-linear magnetic forces ● Magnetic forces due to inverter influence ● Unbalance howling ● Transmission: Optimization of gear meshing, T/M housing ● Vehicle body: Light weight design, damping of drivetrain noise, damping of exterior noise, wind and road noise ● Sound pack: Damping of high frequency noise ● Chassis and tires: Road noise, excitation by tire road interaction, energy transfer by suspension components ● Auxiliaries: Electrified auxiliary components like HVAC, braking system, steering system For the subsystem e-motor/generator the excitation mechanism, possible counter measures and test methods are given in Figure 7. Main excitation mechanisms are magnetic forces inducing rotational speed irregularity, torsional excitation of rotor shaft and axial forces (depending on angular orientation), high frequency magnetic forces from inverter switching frequency and unbalances of the rotor. Counter measures are based on optimizing angular orientation of magnets, increasing housing stiffness, increasing torsional stiffness of rotor, improvement of inverter switching frequency and PWM control and better balancing of rotor. The effect of the counter measures can be evaluated by measuring the corresponding physical property.

563

Automotive megatrends and their impact on NVH

Figure 7: Excitation mechanism, possible counter measures and test methods for vehicle subsystem e-motor/generator

If the above mentioned NVH relevant vehicle components are considered with sufficient depth during the vehicle development process, electric vehicles offer the possibility of very quiet and pleasant interior noise. But especially for sporty vehicles, the powerful and emotional sound may be missing. Even for small vehicles, which have the full torque direct from drive off due to the electric motor properties, the vehicle noise feedback does not fit to the feeling of strong acceleration. Active Sound Design (ASD) by playing virtual sounds via the series audio system provides the ability to have a significant influence on the interior sound atmosphere while driving and to express a brand-specific character. To demonstrate the possibilities of ASD for electric vehicles, FEV installed an ASD system to a series production Renault Zoe BEV and designed different sounds that are played back via the existing audio system. The sound fits to the actual driving situation based on parameters from the CAN-bus e.g. speed and load of the electric motor and vehicle speed. Figure 8 shows Campbell diagram of different interior noise variants created by changing the calibration parameters for the ASD system. The change of interior sound with motor load is essential for the driver experience of dynamic sound character. The here analyzed sawtooth driving condition includes several load changes, which are underlined by the load dependent change of ASD sound. One possible interior sound suitable for a series application is the so called “comfort sound”. Some mid-frequency orders are load

564

Automotive megatrends and their impact on NVH dependent enhanced, giving the driver response to his load request without being annoying. When the driver prefers to have an emotional drive with a combustion engine, he may choose one of the additional selectable sounds similar to V6 or V8 combustion engine sounds. Furthermore, futuristic sound options are additionally created to show the great bandwidth of ASD to create suitable interior sounds for different vehicle types and brands.

Figure 8: Interior noise variants of a FEV demonstrator vehicle created based on a series Renault Zoe. Campbell diagrams of a sawtooth driving condition with several load changes.

In the context of the other trends in automotive industry, the vehicle electrification offers interesting possibilities: ● Autonomous: Option to create more quiet and comfortable interior noise and less vibration as it is expected from a driver not concentrating on the traffic ● Shared: Quiet electric vehicles offer more options to create customized interior noise for the specific user by active sound design ● Connected: The sounds, which inform the driver about the surrounding, can be created more discreet due to less masking noise of the drivetrain

Autonomous Autonomously driving vehicles offer the chance to use the travel time for other things, which can be for occupational or recreational purpose. It is expected, that highly automated and connected vehicle concepts will enter the markets in the next 10-15 years.

565

Automotive megatrends and their impact on NVH Vehicles with different degrees of autonomous driving functionalities will be introduced step by step. Vehicles can be sorted into different groups 0-5 according to their degree of autonomous functionalities from no automation (SAE level 0) over partial automation up to full automation (SAE level 5) (Figure 9) [4].

Figure 9: Levels of automated driving [4]

The introduction of autonomous driving will not be a sudden change but will be phased in step by step. Figure 10 shows the expected growth of market share of automated vehicles in the EU [10]. FEV predicts that in 2035 about 28% of EU new vehicle sales are automated passenger vehicles according to SAE level 5 and about 34% are at SAE level 3&4. Level 3 and 4 automation sales will pick up strongly between 2025 and 2030.

Figure 10: Expected new vehicle sales in EU according to SAE automation level [10]

566

Automotive megatrends and their impact on NVH The transition from user driven to autonomously driving vehicles will have a strong effect on the user/buyer expectation and thus on the vehicle requirements. The user of a fully autonomous vehicle does not actively take part in the traffic events; he is out of the control loop. His situation is similar to a passenger in a train or a businessperson on the backseat of his chauffeur driven luxury class vehicle and very different from the driver of a conventional vehicle. Probably his main interest will be that the driving events do not disturb his activities like talking, reading, phoning, watching movies or working with the laptop. Maybe the only wanted disturbance is the information when the estimated arrival time is changing. Considering this, it can be expected that all comfort oriented vehicle features will become more important, whereas other features like acceleration performance or engine power will become less important. Interior noise and vibrations (N&V) are main influencing factors for a comfortable impression of the surrounding, thus all pleasantness criteria of interior N&V probably get a higher weight in the vehicle requirement book. The dynamic and sporty oriented noise features will be not needed because the user does not want a feedback about the driving situation and they are even probably experienced as annoying. A user surveillance confirmed this general evaluation. Together with experts for the different components and features of vehicle and powertrain, the relevance of different vehicle specifics of an automated vehicles in relation to a conventional vehicle is derived (Figure 11) [3]. For the rating of the vehicle specifics, two very different fields of vehicle application are distinguished: a specialized "City Pod" vehicle is purely designed for mobility within a city, a “Highway Pod” is designed for long distance travel. In such a future mobility concept, vehicle hubs can be assumed, where the passenger changes from city-pod to highway-pod and vice versa.

Figure 11: Change in relevance for vehicle specifics in dependence of use case [3]

567

Automotive megatrends and their impact on NVH According to this assessment, engine features like nominal power and torque will become less important, especially for the city-pod. Low-end torque and responsiveness will become less important especially for the highway-pod. Comfort-related NVH features like jerkiness of start, interior vibration, high frequency quality of interior noise and low- and high-speed boom are expected to be significantly more relevant for automated vehicles. Again the importance of the NVH criteria differs according to the vehicle use case: the behaviour at vehicle start clearly is more important for the city-pod, whereas high-speed behaviour is more relevant for the highway-pod.

Summary Fundamental changes for future vehicles are caused by several trends in the automotive industry, driven by social and political changes and new technology possibilities. The acronym CASE subsumes four megatrends: Connected, Autonomous, Shared and Electric. Although none of these megatrends is NVH, each of these will influence the NVH behavior and properties of tomorrow’s cars. The development of electrified propulsion systems is proceeding already on a high level. Disturbing noises are no longer masked by the combustion engine and new acoustic players like an e-motor get into the game. Annoying noise from the drivetrain itself, as well as unpleasant noise shares from other vehicle systems come into the fore. Due to the missing noise from the ICE, road and wind noise become more important. Today’s methods and tools allow to face a lot of these new challenges and develop completely new propulsion configurations. Target setting for these new systems and sub-systems is one of the challenges. Experience based approaches are not applicable since experience is missing. Methods have to be used, where systematically component targets are derived from full vehicle targets by target cascading using a physical transfer path vehicle model, considering the masking effect from road/tire and wind noise. Active sound design is an efficient possibility to overcome the typical lack of dynamic sound shares from electric vehicles. Sharing cars will lead to the fact that in future most of the cars are not owned and used by one customer but used by many participants of shared mobility. The question has to be answered how such a car should sound. Nevertheless, possibilities have to be taken into account for individualization of each distinct car. Especially the more and more market-penetrating means of active systems, e.g. active sound design (ASD), offer numerous possibilities to tune the vehicle’s NVH character to current customer’s taste. The transition from user driven to autonomously driving vehicles will have a strong effect on the user/buyer expectation and thus on the vehicle requirements. The user of a fully autonomous vehicle does not actively take part in the traffic events; he is out of

568

Automotive megatrends and their impact on NVH the control loop. Comfortable impression and therefore excellent interior noise and vibration will determine the requirement book of autonomous vehicle. Other features like acceleration performance or engine power will become less important. Active systems will play a decisive role in this context. Finally, it can be stated that the substantial and disruptive changes in the automotive industry have a big influence on NVH development. New acoustic players and boundary conditions have to be considered and new demands on vehicle NVH have to be addressed. Nevertheless, the physical background for NVH engineering will not change so that today’s tools and methods are a good base to face these new challenges.

Bibliography 1. Eisele, G., Angermaier, F., Genender, P., Wolff, K.: “Systematic concept for definition of NVH targets for vehicle subsystems and components”. Aachen Acoustics Colloquium, 2011 2. Kauth, M., Steffens, Ch., Eisele, G., van Engelen, A.; “NVH Targets for Passenger Car Transmissions”, Aachen Acoustics Colloquium, 2016 3. Will, P., Nase, A., Straschill, R., Hirsch, R.: “Connected and Automated Vehicles: New Degrees of Freedom to Improve the Powertrain”, 25th Aachen Colloquium Automobile and Engine Technology, 2016 4. Taxonomy and Definitions for Terms Related to On-Road Motor Vehicle Automated Driving Systems, Standard J3016_201401, SAE International, 2014 5. Uber Financial Numbers Communicated to Bloomberg. URL: https://www.bloomberg.com/news/articles/2016-12-20/uber-s-loss-exceeds-800-million-in-third-quarteron-1-7-billion-in-net-revenue 6. Bibi Chuxing Market Value: http://money.cnn.com/2017/04/27/technology/didivaluation-50-billion-uber/index.html 7. Zetsche, D., Mondial de l’Automobile 2016, Press Conference Mercedes-Benz Cars, 2016 8. Zentrum für Sonnenenergie- und Wasserstoff-Forschung Baden-Württemberg (ZSW), 02/2018 9. McKinsey & Company, Automotive revolution – perspective towards 2030, January 2016 10. FEV Consulting GmbH, Internal Analysis, 2017

569

Evaluation of the effects of trends on vehicle concepts based on a forecast of travel demand Peter Lukas Peters, Dr.-Ing. Rainer Demuth BMW Group – Total Vehicle Development Prof. Dr.-Ing. Dr. h.c. Dieter Schramm University of Duisburg-Essen, Chair of Mechatronics

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 M. Bargende et al. (Hrsg.), 19. Internationales Stuttgarter Symposium, Proceedings, https://doi.org/10.1007/978-3-658-25939-6_46

571

Evaluation of the effects of trends on vehicle concepts based on a forecast of travel demand

1 Introduction Today, vehicle concepts are developed on the basis of technical design premises derived from attributes such as e.g. the electrical range in the case of electrified vehicles. For the respective positioning of the concept, the customer requirements of the relevant target group determined by market research and the competitive comparison are decisive. Technological trends (e.g. fully automated driving1; FAD) and business model innovations (e.g. on demand mobility; ODM) have the potential to change the mobility behavior of users and thus the characteristics of vehicle concepts. Furthermore, due to these trends, the modal values of mobility, as for example the individual acceleration behavior or comfort demand, may become less important for selected customer use cases. This is especially the case for demand-oriented ODM services which will gain a new degree of flexibility through FAD vehicle concepts. Therefore, in the future, the conventional approach of determining design premises for total vehicle concepts could no longer be appropriate alone. The modeling of the effects of disruptive trends on the individual mobility behavior is currently not documented on the basis of vehicle sensor data logs, in contrast to mobility data from travel surveys. In the field of transport research, the effects of FAD and new ODM services on mobility behavior have been examined for several years. The introduction of FAD mode alternatives is expected to have a significant impact on mobility (Trommer et al. 2016). The overall approach presented in this paper focuses on deriving objective design premises for future vehicle concepts based on a forecast of travel demand and on the analysis of the associated effects on the technical realization, e.g. with regard to the design of vehicle architectures. For this purpose, findings from a future scenario of mobility based on National Household Travel Survey (NHTS) data are linked to the vehicle concept design under the prerequisite of FAD, ODM and vehicle electrification. As the variables obtained in the initial NHTS data set have not been intended for the derivation of total vehicle design premises, a transformation has to be carried out within the overall methodology. The application is exemplified by a vehicle portfolio, taking into account additional vehicle concepts of a scaled commercial FAD ODM fleet. Therefore the data basis has been evaluated as a frequency distribution with regard to the design premises under consideration. Linking the findings of transport research with the technical design of vehicle concepts represents a new methodology in the early phase of product development in the automotive industry. For the determined vehicle concepts, basic design parameters are estimated. This is followed by an analysis of the corresponding technical realization in terms of the dimensioning of the vehicle architecture design. The developed methodlogy of

1

VDA Level 5: Full Automation (VDA 2015).

572

Evaluation of the effects of trends on vehicle concepts based on a forecast of travel demand

evaluation is initially depicted on the basis of a current vehicle product portfolio. This paper focuses on the analysis of the results of the methodology applied with regard to the technical design of vehicle concepts. A more detailed description of the methodological approach will be part of further publications.

2 Methodology The methodology presented in this paper is divided into the analysis of the vehicle concept and architecture design in the automotive industry and into the derivation of objective total vehicle design premises based on a forecast of travel demand. For the overall approach, it is the aim to determine the characteristics of vehicle concepts and architectures to cover future mobility needs from an objective point of view.

2.1 Analysis of the vehicle concept design in the automotive industry In literature and in the automotive industry, terms dealing with vehicle concept and architecture design are not uniformly defined. In order to achieve a better understanding of the terms used in this paper, a definition has been set. According to Achleitner et al. (2016), a vehicle concept describes the design premises of a product idea that ensures its technical realization in principle. The vehicle concept includes the specification of the body style, the number of seats and the space required by the passengers, the size of the trunk, the main geometric dimensions as well as the concept of the drive train under consideration of the overall vehicle package. Derived from the general definition of a product architecture, which describes the functional and geometric technical solution concept of products (Ulrich 1995), a vehicle architecture in this paper is defined as the general technical solution concept for a number of interdependent vehicle concepts which differ to a certain extent. Based on this definition, a vehicle architecture describes the functional and geometric limits for the realization of vehicle concepts. The interfaces between the modules of a vehicle architecture are based on standardized principles and design guidelines, in order to use carry-over parts. Figure 1 is intended to further illustrate the link between the technical solution concept of a vehicle architecture, the corresponding design limits (boundaries) and technical concept leaps. If the limit of a design premise, e.g. the gross vehicle weight, is exceeded within the technical solution concept of a vehicle architecture, the respective vehicle concept can no longer be realized according to the requirements. A further increase of the design limits of the vehicle architecture in this case, however, would lead to a reduction of the characteristics (individual product optimum) for lighter vehicle concepts and would thus reduce the total benefit. The design optimum within the vehicle architecture is usually set on a reference product, which is to be prioritized

573

Evaluation of the effects of trends on vehicle concepts based on a forecast of travel demand

regarding the relevance in the product portfolio, e.g. in terms of the highest sales volume or contribution margin. The vehicle architectures can differ in terms of technical concept leaps which enable a shift in the design limits. The design limits of the vehicle architectures of an OEM (Original Equipment Manufacturer) are determined by the design premises of the product portfolio and the technical solution concept.

Figure 1: Schematic illustration concerning the distinction of vehicle architectures

The product portfolio and vehicle architectures of an OEM are linked in a mutual relationship. Thereby, the product portfolio represents all differentiable vehicle concepts and thus defines the design limits in the long term (Tesch 2010). In the automotive industry, a distinction is typically made between vehicle concepts within the product portfolio on the basis of the characteristics of the vehicle segment and the body style. Furthermore, the differentiation can be made with regard to the respective drive train concept (Schneider 2006). The decision on future product portfolio planning is based on assumptions about the development of sales markets and customer requirements (Kieckhäfer et al. 2012). As an example Figure 2 shows the product portfolio of the BMW Group2 in a matrix consisting of the category of the body style and the vehicle segment. The vehicles of the product portfolio are assigned to the respective technical solution concept using the color coding (see legend). According to Cornett (2002), the main aspect for the design and differentiation of a vehicle architecture in today's solution space is the configuration of the drive train and the choice of the corresponding major drive axle.

2

Communicated product portfolio for 2018 for the sales region Germany excluding BMW i, BMW M GmbH and Rolls-Royce vehicle models.

574

Evaluation of the effects of trends on vehicle concepts based on a forecast of travel demand

Figure 2: Classification of the product portfolio of BMW and Mini available in Germany (2018)

The following diagrams illustrate the relationship between the vehicle concept and architecture design using selected vehicle models from the product portfolio described above. As database, the published data of the series product portfolio of the open source ADAC car catalogue was used (ADAC 2018). The aim of the evaluations is to analyze the relationship between specific vehicle concept clusters and the corresponding vehicle architectures. Figure 3 initially shows the lower and upper limits of the vehicle height (H-100-B) in relation to the vehicle length (L103) as well as the respective assignment with regard to the drive train configuration. On the basis of the evaluations, the vehicle models with a transverse front engine configuration can be differentiated into two clusters, flat (Hatch) and high (SUV, MPV) vehicle models, depending on the body style. The vehicle models with a front longitudinal engine configuration – on the other hand – have a higher spread of the geometric design limits, as three car segments and five types of the body style are covered by the respective vehicle architecture. Three major clusters which differ significantly with regard to their geometric characteristics can be identified by the evaluation.

575

Evaluation of the effects of trends on vehicle concepts based on a forecast of travel demand

Figure 3: Analysis of length and height differences for different vehicle architectures

Figure 4 shows the limits of the engine output in relation to the gross vehicle weight. For each vehicle model, two configurations with the lowest and highest engine output have been considered3. This evaluation also shows that the vehicle models with a transverse front engine configuration are limited in the maximum engine output and have a significantly lower spread of the functional design limits compared to the vehicle models with a front longitudinal engine configuration. According to a similar analysis independent of the product portfolio of a specific OEM, the differentiation of the current vehicle architectures is generally based on technical concept leaps, as for example the drive train configuration. Based on the evaluated vehicle data – as already schematically mentioned in Figure 1 – it can be seen that with increasing differences between the design premises of the vehicle concepts, a further vehicle architecture with a different technical solution concept can offer advantages, compared to a single vehicle architecture which has to cover all vehicle models. This is especially the case when considering current and future trends such as vehicle electrification or new FAD ODM business models.

3

Due to the large number of possible equipment configurations of the considered vehicle models in the evaluation, the resulting values are to be understood as an example.

576

Evaluation of the effects of trends on vehicle concepts based on a forecast of travel demand

Figure 4: Analysis of the gross vehicle weight and output differences for different vehicle architectures

Therefore, OEMs will face the challenge of taking into account new drive train concepts and different customer- and application-specific requirements when developing future vehicle concepts and architectures. The evaluation methodology presented in this section is to be carried out again as part of the application example for a product portfolio, taking into account the trends mentioned.

2.2 Derivation of objective total vehicle requirements based on a forecast of future mobility demand The future scenario of mobility is simulated in a travel demand model using data from the NHTS "Mobilität in Deutschland (MiD)" (Follmer et al. 2010). By parameterizing the mobility behavior, the simulated mobility data are subsequently linked to design premises and evaluated using a cluster analysis. The aspatial travel demand model used in this paper is an adaptation of a previous model the by DLR Institute of Transport Research which was developed in the context of the study "Autonomous Driving − The Impact of Vehicle Automation on Mobility Behavior" (Trommer et al. 2016) sponsored by the Institute for Mobility Research (ifmo – a research facility of the BMW Group), Munich. For the analysis presented in this paper, the modeling approach has been

577

Evaluation of the effects of trends on vehicle concepts based on a forecast of travel demand

modified and simplified in order to comply with the overall methodology. Furthermore, findings from Kröger et al. (2016) and Kröger et al. (2018) are used for the simulation model. Hence, the approach of the aspatial travel demand model integrates microscopic (generation of simulation population, trip generation) and macroscopic (mode choice) elements. The result of the aspatial travel demand model is a micro-data set for a chosen future scenario of mobility for Germany in the year 2035, taking into account a new FAD ODM service as an additional mode alternative, in the following referred to as autonomous car sharing (ACS). In the scenario selected, the ACS service is only available for trips within cities with >100,000 inhabitants, as urban areas are to be considered the most likely area of use for this business model. That approach has been chosen in order to determine the specific trip characteristics of the considered mode alternatives. The ACS represents a new mode alternative which breaks the link between car usage and car ownership as well as the ownership of a driver’s license. Thus, a description different to conventional mode alternatives is necessary to enable a significant differentiation in modeling. For the distinction, appropriate premises are set in order to consider the effects of FAD vehicles in general and the ACS business model in the aspatial travel demand model. A more detailed explanation of travel demand modeling in general can be found in Hensher and Button (2007) as well as in Ortúzar and Willumsen (2011). Additional information about the model approach will be part of further publications, as the main focus of this paper is on the derivation of implications for future vehicle concepts and architectures based on the results of the simulated mobility behavior. The derivation of design premises based on the mobility behavior is conducted in four steps (see Figure 5): (1) parameterization of the individual mobility behavior based on the variables describing a trip, (2) linkage of the selected parameters as input variables for the relevant design premises, (3) conversion of the input variables into specific values per design premise per trip and (4) evaluation of the coverage of the need for mobility per design premise.

Figure 5: Derivation of design premises based on the mobility behavior

578

Evaluation of the effects of trends on vehicle concepts based on a forecast of travel demand

Based on the individually derived design premises, an additional approach has been developed to objectively determine the interdependent design premises of total vehicle concepts for different user groups or use cases, e.g. privately-owned cars or vehicle concepts as part of a commercially used ACS fleet. The methodology developed is divided into three steps: (1) The implementation of a cluster analysis for the identification of separated mobility clusters, (2) the determination of the relevance of the mobility clusters by identifying the share of trips and traffic volume of the mobility demand and (3) the transfer of the mobility clusters into specific vehicle concepts in the context of a scaled portfolio under consideration of technical concept leaps. Since the methodology depends on the chosen mode alternative, only selected variables have to be taken into account.

3 Results of the application example The following chapter is divided into the description and discussion of the determined design premises of the identified vehicle portfolio to cover future mobility needs and into the subsequent analysis of the technical realization. Finally, the validity of the methodology will be discussed.

3.1 Determination of objective design premises for a scaled future vehicle portfolio The introduction of the additional ACS has led to a significant shift in the trips taken and the traffic volume between the mode alternatives under consideration. Overall, the ACS achieves a modal share of the trips of 6.2% and of the traffic volume of 3.0% for Germany as a whole. The relative change is determined compared to the simulated baseline scenario. The comparatively low share of traffic volume compared to car trips can be explained due to the restriction in the availability of the ACS service to trips within cities >100,000 inhabitants. In particular the ACS substitutes shorter car trips and longer distances of non-motorized individual traffic. By only considering the distances traveled by households within cities >100,000 inhabitants (see Table 1), the ACS substitutes a significantly larger share of mobility from other mode alternatives with 25% of the trips and 31% of the traffic volume. In principle, a comparison of the identified market share of new mode alternatives with other studies is only possible to a limited extent, as the respective scenario parameters and the structure of the modeling usually differ. The identified shares of ACS trips and traffic volumes in this study should be understood as upper limits, as for example, no capacity restraint functions concerning road capacities and no ACS supply model are considered.

579

Evaluation of the effects of trends on vehicle concepts based on a forecast of travel demand

Table 1: Modal split for trips within cities >100,000 inhabitants

Walk Bike Car Public transport ACS

Modal split ACS scenario 2035 Trips Traffic 25.5% 9.0% 9.7% 6.3% 29.2% 37.9% 10.3% 15.4% 25.3% 31.4%

Relative change to the baseline scenario Trips Traffic -22.0% -28.3% -19.2% -27.9% -32.0% -36.7% -16.6% -18.6% -

The evaluation of the results of the travel demand model illustrates that a significant change in mobility behavior can occur by introducing new mode alternatives which differ in their specific characteristics from the existing ones. The results of the methodology for the data evaluation of design premises with regard to the coverage of the mobility needs (see section 2.2) are described in the following section. The analysis of the traveled distances shows, for example, that for car trips 99% of the distances can be covered without recharging if the vehicle has a range of at least 135 km. However – to enable 99.9% of all trips – a range of at least 475 km is required and over 700 km must be maintained for 99.99%. ACS vehicle concepts – on the other hand – can be designed with the aim of meeting specific requirements. In this context, the objective is not to design the vehicle concepts based on worst-case scenarios for the mobility needs but rather to ensure an efficient use of the entire vehicle fleet. For example, this could correspond to a coverage of 90% of mobility needs. For an ACS vehicle concept, a range per trip of 13 km would be sufficient to cover 90% of the trips. A comparable observation can be made with regard to the number of persons per trip. Due to a differentiation in the level of fulfillment of the design premises and the varying trip characteristics, there are different design recommendations for ACS vehicle concepts and privately-owned cars. For privately-owned cars, it is assumed that the entire mobility need of a user has to be covered over the year and thus – from an objective point of view – the resulting multi-purpose vehicle concept will still be oversized for most use cases. In contrast, a significant reduction in requirements can be assumed for ACS vehicle concepts, depending on the degree of fulfillment of the design premises for the use cases to be considered. The trips determined from the aspatial travel demand model represent the simulated travel demand for the future scenario in Germany and can be evaluated depending on the mode alternative. Based on the developed cluster analysis for this application case,

580

Evaluation of the effects of trends on vehicle concepts based on a forecast of travel demand

the relevant design premises derived from the trip characteristics can subsequently be classified into four different mobility clusters for the ACS and privately-owned cars respectively. The total vehicle design premises of the four clusters determined per considered mode alternative are shown in Table 2 for the ACS and in Table 3 for privatelyowned cars. The values given in the tables do not only list the maximum value of the design premises but also the average values. The objects per cluster have a high degree of homogeneity. For example, the ACS cluster (1) combines all trips with low requirements for the different design premises, e.g. the limitation of seating to two persons. The evaluation of the ACS clusters reveals that certain combinations of design premises do not have to be provided. Table 2: Characteristics and relevance of the mobility cluster for the ACS #

Share

Trips Traffic 1 87.7% 75.8% 2 4.3% 17.3% 3 6.0% 5.2% 4 2.0% 1.7% * Comfort requirement

Range per trip [km] ∅ Max. 5.0 18 21.3 35 5.0 16 5.1 17

Seating ∅ 1.2 1.5 3.7 1.4

Max. 2 4 7 4

Trunk size [l] ∅ Max. 40 261 36 180 81 269 442 600

Travel time (net) [min] CR* ∅ Max. 9.8 25 No 30.2 42 Yes 9.7 25 No 9.8 25 No

Additionally, the cluster analysis allows a specific combination of the minimum design premises which meet the requirements of the majority of the trips. It is possible to draw a conclusion about the relevance of the clusters through the respective share of trips and traffic volume. For example, the design premises in cluster (1) cover 88% of the ACS trips and 76% of the ACS traffic volume. From the perspective of a fleet operator, a reduction of the vehicle requirements in order to cover the basic mobility needs and the grading of the fleet portfolio, taking into account the respective frequency distributions, offers various advantages (Kampker et al. 2017). This is especially the case in the context of fully automated vehicle concepts. The dimensioning of all vehicles in the fleet for rare events cannot be seen as reasonable with regard to a demand-oriented design and the efficiency of the overall fleet. To sum it up, all vehicles in the fleet have to cover the requested mobility needs together (Pfriem 2016). Whereas the use of a privately-owned car can only be divided objectively into clearly differentiable mobility clusters to a limited extent. In general, all use cases of the specific user group have to be covered during the year. Nevertheless, the identified mobility clusters for the trips of privately-owned cars can be used as an objective starting point to scale the respective product portfolio.

581

Evaluation of the effects of trends on vehicle concepts based on a forecast of travel demand

Table 3: Characteristics and relevance of the mobility cluster for private cars Range per trip [km] Traffic ∅ Max. 69.6% 12.0 178 9.9% 16.8 316 1.7% 15.3 333 18.8% 235.7 700

Share

#

Trips 88.3% 8.8% 1.7% 1.2%

1 2 3 4

Seating ∅ 1.3 3.7 1.8 1.4

Max. 2 7 7 4

Trunk size [l] ∅ Max. 41 261 80 329 455 600 45 329

Travel time (net) [min] ∅ Max. 15.6 142 19.7 230 18.8 238 173.4 660

The scaled vehicle concepts of the overall product portfolio – taking into account specific ACS as well as privately-owned vehicle concepts – are primarily designed on the basis of the characteristics of the identified clusters for each mode alternative. In addition to that, the relevance of the clusters in terms of covering mobility needs and technical conceptual leaps in the overall vehicle context are taken into consideration. Table 4 shows the derived vehicle concepts and the corresponding defined design premises. Furthermore, the scaling of the portfolio for privately-owned cars is also based on today's concept segments, as described in section 2.1. Table 4: Selected design premises of the identified vehicle portfolio (ACS/Private) Comfort Seating standards

Vmax [km/h]

150

Trunk size [l] 100 (Cabin) 600

Label

Related cluster

Design

Range [km]

C2S

ACS 1

1 Box

Low

2

100

C5S

ACS 4

1 Box

Low

5

70 70

C7S

ACS 3

1 Box

Low

7

150

300

70

C2L

ACS 2

1 Box.

High

2

250

200

70

P2S

Private 1

1 Box.

Low

2

250

250

130

P5S

Private 2

2 Box

Medium

5

300 - 400

350

>130

P5S SUV

Private 2

2 Box

Medium

5

300 - 400

350

>130

P7S MPV

Private 2

2 Box

Medium

7

300 - 400

350

>130

P5T

Private 3

2 Box

Medium

5

350 - 450

600

>130

P5T SUV

Private 3

2 Box

Medium

5

350 - 450

600

>130

P4L

Private 4

3 Box

High

4

500 - 600

500

>130

P5L SUV Private 4 2 Box High 5 500 - 600 500 >130 C = ACS, P = Private, S = Standard, L = Premium/Comfort, T = Transport (Wagon)

582

Evaluation of the effects of trends on vehicle concepts based on a forecast of travel demand

In contrast to the ACS vehicle concepts, a differentiation is carried out according to the engine output (minimum/maximum) in order to face different individual user demands. It is also assumed that with a higher engine output, the electrical range also varies for each model. The design is based on the current state of science, e.g. for the energy density of the battery and not for the scenario year 2035.

3.2 Vehicle concepts to cover future mobility needs For the design premise of the defined vehicle concepts described in Table 2, the vehicle parameter vehicle length (L103), vehicle height (H100-B), gross vehicle weight as well as the electrical engine output (peak) are estimated. The technical design of the ACS vehicle concepts is based on the following premises: electrified powertrain (BEV), reduced initial acceleration for vehicles in commercial use following Pfriem (2016), reduced transverse dynamic requirements, adapted seating angles due to the elimination of the driving task (no driver’s workplace) as well as the use of new seat configurations in the interior, optimized for ACS use according to Baumgartner (2018), future length growth of the passengers as well as the utilization of the potential for lightweight construction due to the reduced basic requirements. In addition, the premise is set that the C2S vehicle concept is designed as a vehicle of the L7e-CP vehicle class (KBA 2014) in order to achieve a further reduction with regard to the requirements of the vehicle concept to be maintained. For the design of the vehicle concepts for private ownership, the electrified powertrain (BEV) as well as the assumed length growth of the passengers are also a prerequisite. Since privately-owned vehicles still have a driver's workplace in the presented study – even if fully automated – the changes in the interior design are considered to be less significant. The scaling of the engine output for the respective minimum and maximum configuration follows the current exemplary power-to-weight ratios of the vehicle segments “subcompact car” to “full size luxury car” and should thus represent the spread of a scaled engine portfolio. The design parameters determined for the vehicle portfolio are – analogous to the presented evaluation methodology in section 2.2 – inserted into the described diagrams (see Figure 6 and Figure 7). The analysis of the geometric design parameters in terms of the vehicle length and height shows that the ACS vehicle concepts with at least five seats have – with regard to the geometrical parameters – similarities to the vehicle concepts for private ownership, which can be assigned to the “MPV” or “SUV” body style categories. Furthermore, the vehicle concepts with two seats build a geometric vehicle cluster. On the other hand, there are no overlaps with the flat vehicle models from the vehicle concepts for private ownership. The ACS vehicles tend to have a shorter vehicle length with an increased vehicle height due to the changed seating angles and the new seat configurations in the interior.

583

Evaluation of the effects of trends on vehicle concepts based on a forecast of travel demand

Figure 6: Analysis of length and height differences to cover future mobility needs

Figure 7: Analysis of the gross vehicle weight and engine output differences to cover future mobility needs

584

Evaluation of the effects of trends on vehicle concepts based on a forecast of travel demand

The analysis of the gross vehicle weight and the engine output (peak) shows major differences between the ACS vehicle concepts and the vehicles for private ownership. These differences can be explained due to the lower requirements of driving dynamics, the reduced maximum speed and the potential weight savings, e.g. because of a reduced battery capacity. Only the P2S vehicle model has some overlap with the ACS vehicle concepts in this evaluation. To cut it short, it can be stated that in comparison to a future product portfolio without purpose ACS vehicles, the spreading of the analyzed technical design parameters to be covered will potentially increase. For this application example, there are substantial geometric and functional differences between ACS and vehicle concepts for private ownership. Therefore, a separate technical solution concept respecttively vehicle architecture design optimized for the differentiated vehicle categories seems to be reasonable.

3.3 Discussion The chosen modeling approach is limited in its level of detail. In general, when creating travel demand models and cluster analyses, the trade-off between the simplest possible modeling approach and thus the higher traceability of the results and a high degree of detail by modeling many explanatory variables always has to be considered. The underlying methodology for transferring mobility data to a scaled vehicle portfolio represents a strict objective approach. Hence, financial aspects regarding production costs or the ACS business model as well as emotional customer needs are generally not taken into account. Due to these restrictions, the design of the vehicle portfolio for privatelyowned vehicles can only be carried out to a certain extent.

4 Summary and outlook This paper describes a methodology to link findings from the forecast of future mobility behavior with the vehicle concept design in the early phase of development. The aim is to systematically identify future changes in mobility behavior through disruptive trends and to objectively analyze the associated effects on the technical realization of vehicle concepts. The methodology is applied using the example of a scaled vehicle portfolio, additionally taking into account vehicle concepts for the use in a FAD ODM fleet, referred to as autonomous car sharing (ACS) in this paper. The mobility behavior is simulated for the passenger transport in Germany for the scenario year 2035. Using an aspatial travel demand model, changes in the modal split and the usage characteristics of the considered mode alternatives are simulated. By parameterizing the individual mobility behavior, selected mobility parameters are linked with design premises for vehicle concepts. Based on a cluster analysis, the analyzed design premises are

585

Evaluation of the effects of trends on vehicle concepts based on a forecast of travel demand

subsequently grouped into separate clusters, differentiated between privately-owned vehicles and vehicle concepts for a commercial ACS fleet use. Based on the identified mobility clusters and under consideration of technical concept leaps, scaled vehicle concepts are determined for both purposes. For this application example, it turns out that there are considerable differences in the design premises and the related geometric and functional design parameters of privately-owned and ACS vehicle concepts. Due to the limited technical synergies between the analyzed categories, an optimized and separated technical solution concept respectively vehicle architecture design seems to be reasonable. The further development of the methodology intends to a conclusion regarding the spread of vehicle concepts for different future scenarios of mobility, for example taking into account ACS for long distance trips. Acknowledgments The authors would like to thank Mr. Lars Kröger (DLR – Institute of Transport Research) for his valuable and constructive contribution to the development and implementation of the travel demand model as part of the overall methodology.

5 Bibliography 1. Achleitner A, Antony P, Berger E et al (2016) Formen und neue Konzepte. In: Pischinger S, Seiffert U (eds) Vieweg Handbuch Kraftfahrzeugtechnik. Springer Fachmedien Wiesbaden, Wiesbaden, pp 131–251 2. ADAC (2018) ADAC Info - Autodatenbank Markenauswahl. https://www.adac.de/ infotestrat/autodatenbank/autokatalog/default.aspx. Accessed 1 October 2018 3. Baumgartner M (2018) Entwicklung von Lösungsansätzen für Innenraumkonzepte in vollautomatisierten zukünftigen Fahrzeugkonzepten. Master’s thesis. Technische Universität München 4. Cornett AP (2002) Plattformkonzepte in der Automobilentwicklung. Zugl.: Koblenz, Wiss. Hochsch. für Unternehmensführung, Diss., 2000. 1st edn. Dt. Univ.-Verl., Wiesbaden 5. Follmer R, Gruschwitz D, Jesske B, Quandt S (2010) Mobilität in Deutschland 2008: Ergebnisbericht: Struktur – Aufkommen – Emissionen – Trends. Bundesministeriums für Verkehr, Bau und Stadtentwicklung 6. Hensher DA, Button KJ (2007) Handbook of Transport Modelling. 2nd edn. Emerald Group Publishing Limited; Turpin Distribution Services Limited [distributor], Bingley, Biggleswade

586

Evaluation of the effects of trends on vehicle concepts based on a forecast of travel demand

7. Kampker A, Gerdes J, Schuh G (2017) Die StreetScooter-Mobilitätslösung. In: Kampker A, Gerdes J, Schuh G (eds) Think Big, Start Small: StreetScooter die e-mobile Erfolgsstory: Innovationsprozesse radikal effizienter. Springer Berlin Heidelberg, Berlin, Heidelberg, pp 201–210 8. KBA (2014) Korrigierte Endfassung 11. KBA-Bekanntmachung zur Fahrzeugsystematik (SV 1). Kraftfahrt-Bundesamt 9. Kröger L, Kuhnimhof T, Trommer S (2016) Modelling the Impact of Automated Driving. Private AV Scenarios for Germany and the US. Presented to the European Transport Conference. German Aerospace Center, Berlin 10. Kröger L, Kuhnimhof T, Trommer S (2018) Does context matter? A comparative study modelling autonomous vehicle impact on travel behaviour for Germany and the USA. Transportation Research Part A: Policy and Practice. doi: 10.1016/ j.tra.2018.03.033 11. Ortúzar JdD, Willumsen LG (2011) Modelling transport. 4th edn. Wiley, Chichester u.a. 12. Pfriem M (2016) Analyse der Realnutzung von Elektrofahrzeugen in kommerziellen Flotten zur Definition einer bedarfsgerechten Fahrzeugauslegung. Dissertation. Karlsruher Institut für Technologie (KIT) 13. Schneider A (2006) Die strategische Planung des Produktportfolios bei Automobilherstellern: Konzeption eines Instruments zur Bewertung des Cycle-Plans. 1st edn. Nomos, Baden-Baden 14. Tesch F (2010) Bewertung der Strukturvariabilität von Pkw-Karosseriederivaten. Dissertation. Technische Universität München 15. Trommer S, Kolarova V, Fraedrich E, Kröger L, Kickhöfer B, Kuhnimhof T, Lenz B, Phleps P (2016) Autonomous Driving: The Impact of Vehicle Automation on Mobility Behaviour. ifmo - Institute for Mobility Research 16. Ulrich K (1995) The role of product architecture in the manufacturing firm. Research Policy 24:419–440. doi: 10.1016/0048-7333(94)00775-3 17. VDA (2015) Automation: From Driver Assistance Systems to Automated Driving. Verband der Automobilindustrie e. V. (VDA)

587

Potentials of modular autonomous vehicles for variable scenarios of public transport Dennis Wedler, Prof. Dr.-Ing Thomas Vietor Institute for Engineering Design

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 M. Bargende et al. (Hrsg.), 19. Internationales Stuttgarter Symposium, Proceedings, https://doi.org/10.1007/978-3-658-25939-6_47

589

Potentials of modular autonomous vehicles for variable scenarios of public transport

1 Motivation On the way to new forms of mobility, a number of autonomous vehicles were built in recent years. In addition to tests with classic cars and trucks, which still have a driver's seat with all the necessary control elements, the vehicle type of a small passenger shuttle has also been developed. These passenger shuttles have already completed some operation time successfully in a controlled environment. Included are companies and university campuses, public areas such as train stations and airports as well as defined routes in city traffic (Table 01). The fact, that the designs of the passenger compartment doesn´t fit well with some of the tested use cases, can be a result of the test period. Nevertheless more variants of passenger compartments are necessary. Instead of different vehicles a modularisation of the vehicle can be the solution. Table 01: Overview of selected cities and application areas of autonomous vehicles

2 Procedures The starting point is a benchmark of existing and announced autonomous passenger shuttles with regard to their geometric information. The analysis is focused on own elaborations with the focus on the influence of geometry and movement patterns of a human body on the usable package space. The underlying human models are based on own research results and will be used as a reference later on. Based on the data analysis, relevant configurations of the passenger transport area are derived. This is used as a basis for the identification and design of modules based on defined usage scenarios.

590

Potentials of modular autonomous vehicles for variable scenarios of public transport

2.1 Data analysis Market-ready or announced autonomous vehicles can be divided into two groups: vehicles for private transport and vehicles for public transport. An overview of the vehicles included in this study can be found in Table 02. The classification (private transport and public transport) is based on the statements and advertisements of the manufacturers. In this case, public transport refers to a replacement system that can carry several groups of people. These vehicles can be used parallel to buses and trams or in exchange for them. Private transport vehicles are used to transport closed groups (family, company, circle of friends, etc.). On the one hand the data is based on provided data sheets of the respective manufacturers (homepage) and on the other hand the data is based on official press releases about these vehicles. At this point it should be noted that not all manufacturers provide the required data for their vehicles. To ensure that these vehicles could still be included in the analysis, the necessary dimensions were determined on the basis of existing sources (images and videos). Table 02: Overview of all vehicles included in the study

591

Potentials of modular autonomous vehicles for variable scenarios of public transport The evaluation of the basic dimensions shows that both areas (public & private) almost do not overlap. The vehicles of the private use are tending towards smaller dimensions. Furthermore regarding the length, it can be seen that the dimensions correlate with classic vehicles of private and group transport. The vehicles of the manufacturer Volkswagen are listed here as comparative data. An interesting result is that the classic vehicles represent the borderline between the two areas (Figure 01). The data of the manufacturer Volkswagen refer to the vehicle models (Up!, Polo, Golf, T5 short-wheelbase, T5 long-wheelbase). The project name "Swipi" stands for a concept discussed later in this paper.

Figure 01 – Comparison of length and width

Figure 02 – Comparison of height

If the height is also included (Figure 02), three general statements can be made. ● The majority of vehicles have a length of 3500 mm to 5200 mm, regardless of their actual function. ● Autonomous private transport vehicles are predominantly narrower than classic cars, but have a greater height. ● Autonomous public transport vehicles are wider than classic vehicles and high enough walking upright inside the vehicle. One exception is the MayMobility vehicle, which has the entrance and seat arrangement of a classic automobile. In the further course of the considerations, the focus will only be on public transport vehicles. Their special feature of being walkable inside, as well as the larger vehicle width, allows a higher variation of seating configurations. These boundary conditions allow a larger spectrum of usage scenarios to be covered, which in turn favour modularity in the structure. When analysing the existing data, two aspects are particularly noticeable. On the one hand, there are hardly any vehicles that are modularly adapted to different transport tasks (usage scenarios). On the other hand, it is not clear how the existing vehicles can

592

Potentials of modular autonomous vehicles for variable scenarios of public transport carry the specified number of passengers. This second aspect has an influence on the modularisation discussed later. Therefore it will be examined in more detail below.

2.2 Analysis of passenger capacity (number of seats) The following considerations are generally valid, but since the underlying data set is based on the latest series measurement in Germany, the findings must always be transferred in relation to the region of use [1][2]. Furthermore, an average physique was used. This represents a simplification for the first considerations. The influence of restrictions and body weight (circumference/dimensioning) were explicitly not considered. For a better understanding of the analysis, the definition of the required package space is described below. As shown in Figure 03, it is not assumed that there is a standard seat space, but rather the volume of space taken up by a person sitting on this standard seat space [3]. Even with the 5-percentile reference model of a woman, it can be seen that the area of the shoulders protrudes laterally beyond the seat. Furthermore, there is a considerable area in front of the seat which is stressed by legs and feet. This area plays a decisive role when the occupancy of persons in a vehicle is to be analysed. In particular, if only the seat surface is used as geometry, it can quickly lead to overcorner arrangements or to curve benches. However, the persons sitting on these seats always have an overlap of their foot areas in front of the seats. The area a person occupies in a seated posture is determined by the distance between his elbows in a relaxed seated posture and the distance between back and toes.

Figure 03 – Dimensions (volume box) of a 5th percentile woman and a 95 percentile man

The volume boxes, that are created in this way, have a base area of 800 x 650 [mm] for the 95th-percentile man and almost half this area of 650 x 440 [mm] for the

593

Potentials of modular autonomous vehicles for variable scenarios of public transport 5th-percentile woman. These volume boxes or their base areas were used to analyse the transport capacity of public transport vehicles. The manufacturer's seat configurations and the previously derived volume boxes were compared with the external dimensions of the vehicles. This step was necessary because no information regarding the interior was available. Nevertheless, initial results can be derived even from these simplified considerations. The result is shown in Figure 04. All current versions of the examined vehicles can easily accommodate persons of the 5th-percentil female. With the 95th-pecentile male all vehicles show problems. The vehicle models “Olli” and “EZ10” seem to have enough space, but at this point it has to be considered that the whole floor space (up to the outer skin panel of the vehicle) has to be used. Theoretically the “Apolong” also has enough space to accommodate people of the 95th percentile. Similar to the vehicles mentioned above the space is very limited. Although this vehicle was developed for the Asian market and its reference persons are smaller, however the transportability should be given according to first estimation [4].

Figure 04 – Use of package space depending on volume boxes (95th man and 5th woman)

These considerations correspond to a worst-case analysis and it has not been investigated yet which percentile-sex combinations for the vehicles allow an optimal utilization of the seats.

2.3 Results Almost all vehicles are not designed for large persons within the defined number of persons. With the smallest reference models, the transport capacity can be implemented. Only the "Olli", the "EZ10" (limited) and the "Apolong" can theoretically transport large persons to their full extent. Vehicles with benches allow greater variability,

594

Potentials of modular autonomous vehicles for variable scenarios of public transport especially when combining people of different sizes. Predefined seats (seat shell) limit the sensible use of the seating areas. If individual seats are used, the distance between them must be designed for the largest persons in the intended use scenario. Otherwise, benches are a more flexible solution. But even using benches, a sensible dimensioning must be found for the seat width that is oriented towards the taller person. If all potential users have great comfort during use, the acceptance for re-use will increase. This is a hypothesis that has emerged from an interim outcome of a survey. In this survey people who regularly use public transport were asked about their comfort. This comfort was particularly related to the sitting posture adopted in double seats while they were used alone or together with another person. The analysis shows that the targeted consideration of the user groups has a large influence on the actual use. On the one hand a good concept for local public transport can be developed, on the other hand it is possible to derive specialised vehicles based on the definition of passenger groups. This idea is the starting point to take a closer look at the modularisation of autonomous vehicles. Current concepts (e.g. Rynspeed or Toyota) or implemented vehicles (Bakulin Motors Group) are used to show the first geometrically meaningful separation (Figure 05). The first mentioned concepts use a separation between drive and life module (skateboard design with function structure). The Rynspeed company additionally allows its coupling process to be carried out completely autonomous. This means that the function structure can be switched off for further use (outside the transport function/activity) and the drive module can couple itself with another functional unit. On the other hand Bakulin Motors Group's approach is to provide a one end-coupleable drive unit for each axis. The advantage and disadvantage of these realizations are seen in Table 03.

Figure 05 – Separation levels of existing concepts

595

Potentials of modular autonomous vehicles for variable scenarios of public transport Table 03: Advantage/disadvantage of existing modules for public transport vehicles Rynspeed advantages Decoupling hat and drive module allows the organizational use of several hats by one drive module. (in chronological order) Automatic decoupling allows a usage process that can be controlled via an external system. Matryoshka and Toyota advantages Any vehicle length can be mapped as long as the interface on the axle modules is maintained. If battery system mounted on axis module: quick change/maintenance possible.

disadvantages High safety effort required to ensure that the coupling process always functions properly.

disadvantages If battery system mounted on hat module: Maintenance requires shutdown of entire hat.

3 Approach The own concept provides for a separation into two main modules (life module, basic module), whereby the basic module is divided into a further unit (drive module). The additional subdivision is based on production and maintenance aspects. The drive module itself is divided into four separate individual wheels. Through this the area for the battery system in the basic module is maximized. The basic module itself has a flat and even floor. This allows the life module above to have the largest possible usable interior space. The exact division of the modules is shown using the basic units of a road-bound vehicle (Figure 06). The following modules can be derived: Life module ● Environment recognition, control element, driving posture, driving position support structure Basic module ● Signal processing, energy storage, support structure Drive module ● Drive unit, powertrain, mechanics for changing the direction of movement, locomotion mechanics In addition, there is a control unit for the personnel (maintenance module) which can be connected instead of the hat module for maintenance purposes. The connection between

596

Potentials of modular autonomous vehicles for variable scenarios of public transport life modul and basic modul is made in the area of the wheel arches. A maintenance flap accessible from the interior allows the complete connection for energy, information and material flow.

Figure 06 – Modularisation based on the basic elements of a vehicle

Based on the selected separation planes, the following results are obtained. Advantages: ● Products from different manufacturers can be combined (standard interfaces). ● Existing supplier industry can become original equipment manufacturer. ● Drive module: Wheel-drive-steering ● Basic module with energy storage (battery housing and vehicle structure are one unit increased battery safety) ● life module ● Reduction of required vehicles (number)Resources ● A new usage scenario just requires a new hat. ● Cheaper than a complete vehicle. ● Can be implemented by many companies. ● Maintenance of hat and basic module can be implemented by many companies. ● Manual assembly of life module and basic module allows automatic maintenance during the process.

597

Potentials of modular autonomous vehicles for variable scenarios of public transport Disadvantages: ● Requirement analysis necessary (life module: time period, location) ● Decentralised depots (storage of life modules depending on the scenarios) ● Standardization: official regulations ● Manual assembly and testing is time-consuming and requires skilled personnel.

4 Application A validation of the approach of modularisation is carried out on the basis of three usage scenarios. These scenarios are based on conditions in Germany and therefore it is not possible to transfer them universally to other countries. However, the basic principle behind this approach is generally applicable. Scenario 1: general public passenger transport This is a reference scenario based on the existing vehicles that served as the basis. In order to achieve this, a transport capacity of at least 10 seats is targeted. In addition, sufficient areas should be provided for standing room and at least one wheelchair user [5][6]. This vehicle can be used on public routes but also in enclosed areas such as company premises. All persons should be able to enter and leave the vehicle independently. Therefore an automatic ramp has to be included in the basic module. Scenario 2: transport for nursing homes and care homes This is a special application that is deliberately aiming at transporting elderly people and people with restrictions in their musculoskeletal system. In addition, the assumption has been made that the vehicle can also be used by younger, healthy people who provide assistance to the elderly. Scenario 3: kindergarten transport This is a special scenario as the majority of the people transported are children aged 3-6. This scenario is based on the fact that in Germany kindergarten groups (2-3 adults & 20+ children) use public transport regularly to visit either the swimming pool or a museum. A vehicle with these requirements has to provide at least the space for the described number of persons as well as sufficient storage space (hand carts and backpacks). Children in wheelchairs must also be carried in the vehicle. The general technical dimensions are based on the vehicles examined. Only the width is 100mm longer than the reference. This value is based on §32 of the stVZO and the usable width of 2500 mm [7].

598

Potentials of modular autonomous vehicles for variable scenarios of public transport Basic dimensions: ● length: 5000 mm ● width: 2200 mm ● height: 2500 mm By combining the two geometric forms (base area and payload objects), different variants can be derived. In this case this was done manually (brute force), but for a comprehensive analysis with variable external dimensions this is also possible automatically [8]. Figure 07 shows a sensible solution for each usage scenario. All variants have in common that the seats are arranged on the wheel arches. This implementation also corresponds to the existing vehicles. In contrast to these, however, the alignment is transverse to the direction of travel, whereby a greater length of the interior can be used. The arrangement of the seats on the wheel arches and a standard seat height automatically results in the interior floor height. This is particularly important for the children in scenario 3.

Figure 07 – Overview of the concepts as 2-D top view

Since the kindergarten transporter has a special feature compared to the other two solutions, it will be examined in more detail below. Children are smaller in their dimensions (body length max. 1270 mm) than the adults escorting them [9][10]. However, they form the largest group of accompanying persons. In this way they define the basic dimensions of seating areas and floor height. Like the adults in the other concepts, the children sit on the wheel arches. Due to their smaller dimensions, it makes no sense to orientate on the seat height geometry of adults. For this reason, the

599

Potentials of modular autonomous vehicles for variable scenarios of public transport floor in this area was raised according to the seat height for young children [11]. This turns the vehicle from an adult transporter used by children into a real children transporter. The increase of the floor in the interior has the advantage of the child-friendly design and the disadvantage that the interior is reduced at a given maximum vehicle height. With a height of 2070 mm, the room is in principle accessible by persons of the 95th percentile man. Decisive at this point is the thickness of the roof which has not been detailed yet. Therefore it can be assumed that the use for large persons will be restricted. (Figure 08).

Figure 08 – Comparison of interior height between scenario 1 and scenario 3

In addition, the height difference must be made possible for the children by a step. This step can be positioned in two places. ● The first variant positions the elevation at the entrance to the seats. In the central section opposite the door there would be a sufficiently high ceiling so that all adults could use this area without any problems. ● The other variant envisages placing the step directly in the entrance area. Thus the restriction of the height would affect the entire interior area. A comparison of the advantages and disadvantages can be found in Table 04.

600

Potentials of modular autonomous vehicles for variable scenarios of public transport Table 04: Advantage/disadvantage of the different position of a step in the passenger compartment Entrance to the seats advantages The central area is accessible by every person up to the 95th-percentile male. An existing vehicle for adult transport (scenario 1) can be converted on one or both sides. Use of a ramp. (scenario 1 and scenario 2) Step in the entrance of the vehicle advantages Continuous floor in the interior. (Accessibility for children) The younger children (3 years) can be supported by adults to enter the vehicle.

disadvantages Inside, additional safety devices are required to prevent the children from falling down the stairs.

disadvantages Limitation of the height in the entire vehicle interior. Load of handcart / wheelchair must be lifted by two adults.

It is important to know that this evaluation has so far been carried out theoretically and that it will be subsequently tested by groups of test persons (kindergarten children and teacher). These tests (using an adapted modular seating buck with VR superimposition) have not been carried out yet at the time of publication.

5 Conclusions ● Ergonomic investigations and evaluations can be carried out in a simplified manner as long as the underlying database displays reality sufficiently. ● With simplified human geometries, vehicle concepts can be designed and evaluated at an early stage of development. ● Modular autonomous vehicles can increase passenger transport scenarios without changing the number of vehicles on the road. ● Defined interfaces and geometries enable market access for many companies. ● Special solutions in small quantities are possible for the life module.

601

Potentials of modular autonomous vehicles for variable scenarios of public transport

Bibliography 1. DIN 33402-2. Ergonomie – Körpermaße des Menschen – Teil 2: Werte: Beuth 2. Seidl, A., Trieb, R., Wirsching, H.-J.: SizeGERMANY – the new German Anthropometric Survey: Conceptual Design, Implementation and Results. In: IEA 17th World Congress on Ergonomics (2009) 3. Din 25105 Nahverkehrs-Schienenfahrzeuge; Sitze, Hauptmaße: Beuth 4. Anthropometric study among adults of different ethnicity in Malaysia. / Karmegam, K.; Sapuan, S. M.; Ismail, M. Y.; Ismail, N.; Shamsul Bahri, M. T.; Shuib, Solehuddin; Mohana, G. K.; Seetha, P.; TamilMoli, P.; Hanapi, M. J. In: International Journal of Physical Sciences, Vol. 6, No. 4, 18.02.2011, p. 777–788. 5. DIN EN 12183:2014 Muskelkraftbetriebene Rollstühle – Anforderungen und Prüfverfahren: Beuth 6. DIN 18040-2 Barrierefreies Bauen – Planungsgrundlagen – Teil 2: Wohnungen 7. StVZO - § 32 Abmessungen von Fahrzeugen und Fahrzeugkombinationen 8. Kuchenbuch, K.: Methodik zur Identifikation und zum Entwurf packageoptimierter Elektrofahrzeuge, AutoUni-Schriftenreihe; Band 25, Logos-Verlag , Wiesbaden, Berlin; 2012 9. Hesse, V., U. Jaeger, H. Vogel, K. Kromeyer, K. Zellner, I. Bernhardt, A. Hofmann und A., Deichl (1997(a)). "Wachstumsdaten deutscher Kinder von Geburt bis zu 18 Jahren.", Sozialpädiatrie. Kinder- und Jugendeilkunde 19 10. Kromeyer-Hauschild, K., Wabitsch, M., Kunze, D. et al. Monatsschr Kinderheilkd (2001) 149: 807. https://doi.org/10.1007/s001120170107 11. DIN EN 1729-1:2015. Stühle und Tische für Bildungseinrichtungen – Teil 1: Funktionsmaße: Beuth

602

Method for concept design and optimization of twist beam axles Xiangfan Fang, Kanlun Tan and Jens Olschewski Institute of Automotive Light weight Design, University of Siegen, Germany

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 M. Bargende et al. (Hrsg.), 19. Internationales Stuttgarter Symposium, Proceedings, https://doi.org/10.1007/978-3-658-25939-6_48

603

Method for concept design and optimization of twist beam axles

1 Introduction – Twist beam axle and their concept development The twist beam axle (TBA) is one of the most common type of rear axles. It was developed at first for the VW Polo [1, 2] based on the rigid axle. In comparison to a rigid axle whose wheels cannot move independently due to the direct connection between the two wheel carriers by a cross member, the relative movement of the two wheels in a twist beam is possible. There, the cross member is placed towards the vehicle front direction and connect the two trailing arms in a position between the wheel center and the body mount. During parallel wheel travel, the motion of a twist beam axle causes only slight changes in track, toe and camber that is similar to a rigid axle. However, during an opposite wheel travel the twist beam axle can provide a similar behavior like a multilink suspension [3]. It is thus called a semi-independent axle. The performance of TBA contains different axle stiffnesses, kinematic characteristics for both parallel and opposite wheel travel and durability. They are determined by design variables such as hard points, shape of the components in 3-dimensional package and their cross section size as well as the selected materials. As in Fig. 1 shows, all requirements are simultaneously influenced by all concept variables. On the one hand these complex interactions enable an enormous conception potential because the large number of the variables can provide a large variety of concept possibilities. On the other hand, the optimization of the product becomes very difficult. The local improvements of one requirement may have negative effects on the other requirements. That is the main challenge of the concept design of twist beam axles.

Figure 1: Complex interactions between concept variables and axle properties [5]

604

Method for concept design and optimization of twist beam axles In the conventional approach for the concept development phase, engineers start usually with several “empirical” or benchmarking based concepts. At first, these initial concepts are roughly designed with CAD software and then converted into FEM and MBS models. Their stiffness and kinematics are calculated by simulations. A relatively good concept that may have a good optimization potential can be identified. This selected concept will be further optimized via many CAD and CAE iterations until it meets all concept targets. There are two significant problems in this conventional process. First, many manpower and computing resources must be invested in the evaluations of all initial concepts. Moreover, the quality of the initial concepts is highly dependent on the personal experiences of the engineers. It is possible that even the selected concept could only have a limited potential and might not reach the targets after a long time of optimization. The conventional approach cannot meet the need of current development, which requires shorter time, fewer resources and better results. The development of efficient methods is therefore strongly desired and the target of this current work [4, 5] will be presented below.

2 Analytical calculation of axle properties Beside of the strength and durability, the axle stiffness and kinematics are the most important target properties of TBA and must be already met in the concept development phase. Since the strength and especially the durability are determined by the local design details and the axle kinematics can be described well by the Machinsky method [6] this capital concentrates on the axle stiffness of TBA.

2.1 Stiffness under lateral forces There are three major stiffnesses under lateral forces to be considered: a.) The lateral stiffness Cy which is the ratio between lateral force Fy at the wheel contact point with a short offset RASFS and the displacement of this contact point in the vehicle transverse direction ΔyRASFS. b.) Similarly, the toe stiffness Ctoe and c.) the camber stiffness Ccamber are the ratios between the same lateral force and the corresponding angle change of the toe and camber, Δδ and Δγ 𝐶

𝐹 ∆𝑦

,

𝐶

𝐹 , ∆𝛿

𝐶

𝐹 ∆𝛾

The calculation process of these three stiffnesses in this work will be described in the following four steps.

605

Method for concept design and optimization of twist beam axles

2.1.1 Step 1: Idealization of an axle as a beam model Because of the simple and similar structures of all twist beam axles, it can be simplified as a beam model with proper positioning of beams and suitable beam cross-sections. (Figure 2). Each of the three main components is idealized as five straight and homogeneous beam sections or beams in this work. With a suitable positioning of the beams, the three-dimensional paths of the components can be modelled accurately. To describe the diverse cross-sections of the axle components, different cross sections are defined for each beam. For example, the cross member of the example axle is a U-shaped profile with two reinforcement plates at both ends in its connection areas to the trailing arms. The beam in the middle is a U- or C-shaped profile and the two beams in the transitional areas have partially closed cross-sections. Therefore, the two beams or beam sections at both ends of the cross member can be idealized as rectangular cross sections. In this work, the axle is idealized as fifteen beams. Theoretically more beams can be introduced to improve the accuracy of the model. However, experience shows that the extra beams complicate the calculations significantly with only little improvements on accuracy.

Figure 2: Idealization of the example axle as a beam model

2.1.2 Step 2: Calculation of the loads on bearings and nodes In order to calculate the bearing loads analytically, the beam model is mounted as a statically determined system (Figure 3). To calculate the node loads the beam model can be separated at each node. Figure 4 shows an example for the nodes Ql, Qr and HRl. After the separation of the beams in many sub-sections, the forces and moments of each part must remain in equilibrium so that all loads on all nodes can be calculated analytically.

606

Method for concept design and optimization of twist beam axles

Figure 3: Reaction forces of the bearings and the names of different beam nodes and bearings

Figure 4: Loads on Ql, Qr and HRl

Figure 5: Separation of TWA in three parts and their bearing conditions

607

Method for concept design and optimization of twist beam axles

2.1.3 Step 3: Calculation of the deformations of each beam With the loads on different beam sections determined in step 1 and 2, the individual beam deformations can be calculated analytically. The deformations include tension, compression, bending and torsion. In order to determine the deformations, the TBA is divided into three parts (Fig. 5). The bearing conditions for the three parts for deformation calculation are now different from that in Fig. 3.

2.1.3.1 Deformation of the cross member The cross member bends because of the lateral force on wheels. Figure 4 already shows the loads at the two ends of the cross member. These loads can be divided into two bending directions X and Z of the vehicle coordinate system as shown in Figure 6.

Figure 6: Loads on the cross member

The bending moment on the cross member can be described by the following two equations. Its value decreases from left to right side in a trapezoid form. 𝑀 𝑠

𝑀

𝐹

∙𝑠

(1)

𝑀 𝑠

𝑀

𝐹

∙𝑠

(2)

In Figure 6 the nodes of the cross member in Figure 3 are renamed using alphabetic letters from A to F. The two fundamental bending equations (3) and (4) are applied to the five beam sections and their ten ends. In total, 20 equations are thus formed section wise. 𝑤, 𝑠

608

𝑑𝑠

𝐶

(3)

Method for concept design and optimization of twist beam axles 𝑤 𝑠

𝑑𝑠𝑑𝑠

∬

𝐶 𝑑𝑠

𝐶

(4)

Besides the 20 equations, two additional equations can be found directly from the bearings of the cross member, where the deflections of the two ends must be zero. 𝑤 𝑠

0

𝑤 𝑠

𝑠

0

(5)

0

(6)

In total, there are 22 independent equations for the description of bending. They contain exactly 22 unknowns, namely the six bending deflections 𝑤, the six bending angles 𝑤 , , of the nodes and the ten integral constants of the five beam sections of the cross member from AB1 to EF2. This equation system can be formed as a matrix and solved analytically.

2.1.3.2 Deformation of the trailing arms In order to calculate the deformations of the trailing arms, each trailing arm has been separated into five beam sections, as shown in Figure 7. All beam loads can be determined by the method shown in step 2. It must be noted that the fixed ends of the beams are different. The first two beams next to the guide bearing are fixed in the positive x-direction while the other three beams are fixed in the negative x-direction. These different fixings are important for the next step [4, 7].

Figure 7: Separation of the trailing arm in five beams

At the beginning of the calculations, the beam loads are transformed into the local coordinate system (KOS) that relates to each beam. For demonstration of the principles, the beam section QAl-Ql serves as an example. For this beam, a beam vector can be

609

Method for concept design and optimization of twist beam axles built with the coordinates of its two ends in the global coordinate system KOS-O. Figure 7: Separation of the trailing arm in five beams 𝑎⃗

𝑥 ,𝑦

𝑥

𝑦 ,𝑧

𝑧

(7)

The global coordinate system KOS-O can then be translated into the local coordinate system KOS-Ql using the following transformation matrix of homogeneous coordinates. 𝑻

1 0 0 𝑥

0 1 0 𝑦

0 0 1 𝑧

0 0 0 1

(8)

Subsequently, the new KOS at Ql is rotated around its y-axis with the angle β and around its z-axis with the angle γ. So the x-axis of the coordinate system KOS-Ql goes exactly along with the beam Ql-QAl. 𝛽

arcsin 𝑧

𝛾

sgn 𝑦

𝑧 𝑦

/|𝑎⃗|

(9) /| ⃗ |

∙ arccos

β

(10)

The rotation matrices of the two angles are thus 𝑹𝒚

𝑹𝒛

cos β 0 sin β 0 cos γ sin γ 0 0

0 1 0 0

sin β 0 cos β 0 sin γ 0 cos γ 0 0 1 0 0

0 0 0 1 0 0 0 1

(11)

(12)

All loads or vectors in the KOS-O can be transformed into the KOS-Ql by using the following equation. 𝑥, 𝑦, 𝑧, 1

_

𝑥, 𝑦, 𝑧, 1

_

∙ 𝑻 ∙ 𝑹𝒚 ∙ 𝑹𝒛

(13)

With such a transformation the loads on each beam can be disassembled into six regular components which are either along or perpendicular to the beam. These loads lead to simple tension, compression, bending and torsion of the beam sections. They can be calculated analytically with basic mechanical equations. Since most trailing arms are round profiles, no warping and warping torsion must be considered. In the next step, the deformation of section QAl-Ul and after that the section Ql-HRl can be calculated in the same way.

610

Method for concept design and optimization of twist beam axles

2.1.4 Step 4: Assembly of the axle deformation from the individual beam deformations In the last step, the individual beam deformations are assembled to determine the complete axle deformation. The assembly starts from the node Ql on the left trailing arm. It is assumed that this node is fixed in the space. Figure 8 shows the assembly of beams Ql-QAl and QAl-Ul with an exaggerated representation so that the process can be presented more clearly.

Figure 8: Assembly of the deformed beams Ql-QAl and QAl-Ul

In Figure 8 (a) the two beams are not deformed. In (b) the beam Ql-QAl deforms and the beam QAl-Ul remains undeformed. The position of QA1-U1 is changed due to a kinematic effect. In (c) the deformation of QAl-Ul occurs. During the analytical calculation, the displacement and rotation of the node QAl can be determined in KOS-Ql due to the deformation of the beam Ql-QAl. Then they are transformed into KOS-QAl and used as start conditions for the calculation of beam QAl-Ul. Subsequently, the position and the orientation changes of the node Ul are determined. In reality, Ql is not fixed and Ul is unable to displace. Thus, the deformed beams have to be moved to meet these real constraints as shown in (d).

Figure 9: Assembly of the individual beams of the left trailing arm

611

Method for concept design and optimization of twist beam axles By using the same method, the deformation of the other half of the trailing arm from Ql to HRl can also be determined analytically. In this virtual time point, the deformation of the cross member has not been considered yet. Because of a kinematic effect the whole axle shifts slightly to the right side as shown in Figure 9. The deformed cross member in 2.1.3.1 is now assembled into the beam model in Figure 9. This approach is a geometric conversion with the following boundary conditions. The 90 degree angles of the infinitely small areas between the cross member and the trailing arms (see Figure 10a) must remain constant before (dashed line) and after (bold solid line) the deformation because of the assumption of an infinitely stiff and infinitely small node between the trailing arm and the cross member.

Figure 10: Assembly of the cross member into the axle

Mathematically, the left trailing arm rotates around Ul, in x-direction with the angle u′ s and in z-direction with the angle w ′ s . These two angles are identical to the bending angles of the left end of the cross member. Similarly, the right trailing arm rotates around Ur with the angles u′ s and w ′ s which equal the bending angles of the right end of the cross member. The rotation of the guide bearings results in a further kinematic drift of the axle. The solid line in 1 shows the deformed axle while the dashed line represents the initial state of the axle. The deformation calculation of the axle is now completed. In concept design of axles, the wheels and wheel carriers are idealized as rigid bodies. Therefore, the wheel position depends only kinematically on the displacement and the rotation of the node HR which connects the wheel carriers with the trailing arms. A local coordinate system KOS-HR can be set up und fixed at the node HR.. With a coordinate translation from KOS-HR into the global coordinate system, the real changes of the wheel position can be calculated. By substituting the new wheel position into formulas (1, 2 and 3), the axle stiffness can finally be determined.

612

Method for concept design and optimization of twist beam axles

2.2 Vertical force and roll stiffness If a vertical force is applied at the wheel contact point of for example the left wheel, the axle will be twisted. The roll stiffness is the property of the axle for this load case. It is the radio of the applied force Fz to the resulting displacement of the wheel center R in . It describes the torsion deformation of the axle the vertical direction ΔzR. 𝐶 ∆ during cornering. With the approach described in 2.1, the roll stiffness can also be calculated analytically as well. Details can be found in [7].

3 Calculation time and validation of the method With this method, the complex interactions between 12 important axle properties and about 40 concept variables are systematically investigated, mathematically formulated and programmed as a user-friendly software called FastCon-TBA (fast conception of twist beam axles). Once concept variables are entered, the axle properties can be calculated in a very short time with high accuracy and without using CAD and CAE. To validate the correctness and accuracy of this analytical calculations, different axles have firstly been calculated with FEM programs. Beside of camber stiffness, the deviations are mostly much less than 10% [5,7]. The analytical calculation of TBA is thus accurate enough for concept development.

4 Automatic concept design and optimization 4.1 Optimization methods and their evaluation The most common practise in design optimization is the usage of CAE optimization. In this method, CAD und CAE programs are coupled with genetic algorithms. It may achieve a good quality of the end solution. However, the coupled CAD-CAE optimization has strong limitations in practical applications. Firstly, a long modeling time is needed to build a parametrical CAD-CAE-model with all concept variables. A Concept of twist beam axle can contain more than 40 important variables. They require certainly a long modeling time. Secondly, the CAE optimization chain is often problematic concerning time and license cost. A single CAE calculation is already time-consuming. An optimization with many variables and objects contains usually an exponential number of iterations and requires therefore extremely large computation time and amount of license. Beside of these general problems, current CAE optimization cannot be applied simultaneously to optimize the stiffness and kinematic performance.

613

Method for concept design and optimization of twist beam axles

4.1.1 Mathematical optimization and its simplification Based on the analytical calculations of TBA described above, its optimization is mathematically speaking a multi-dimensional non-linear optimization problem [8]. For this multi-dimensional non-linear optimization, each target performance value represents one dimension. For this current work, we determined 12 target performance values for a normal TBA with about 40 variables in component design. This means that the optimization is a 12 dimensional problem, which is obviously quite complex. In order to simplify the system, a so-called marking system is introduced. In this system, each performance value is weighted and the sum of all weighting factors must be 100% (Table 1). For example, as a requirement, the lateral stiffness should be higher than 1700 N/mm. If the actual lateral stiffness exceeds 1700 N/mm, the best mark of 10 can be achieved. In the case of the example axle (see Figure 2), the actual value of the lateral stiffness of the initial axle design does not reach the target value. Therefore, the mark in Table 1 is only 8.09. This mark is determined by the user-defined marking rule in equation 14 [5,7] with a penalty factor of 0.2. It means when the actual value is smaller than the target value, each 10% deviation from the target value leads to a mark deduction of 2. In the case of the example axle in Table 1, the total Mark is 7.37 according to equation 15 since at initial design many performance values are underperforming. Table 1: Setting of the marking system and the Optimization of an example axle with SQP

𝐼𝑓

𝑎𝑐𝑢𝑎𝑙_𝑣𝑎𝑙𝑢𝑒 𝑡𝑎𝑟𝑔𝑒𝑡_𝑣𝑎𝑙𝑢𝑒 𝑚𝑎𝑟𝑘 10 𝐸𝑙𝑠𝑒 𝑚𝑎𝑟𝑘 10 1 𝑎𝑐𝑡𝑢𝑎𝑙_𝑣𝑎𝑙𝑢𝑒⁄𝑡𝑎𝑟𝑔𝑒𝑡_𝑣𝑎𝑙𝑢𝑒 ∙ 𝑝𝑒𝑛𝑎𝑙𝑡𝑦_𝑓𝑎𝑐𝑡𝑜𝑟 𝑇𝑜𝑡𝑎𝑙 𝑎𝑥𝑙𝑒 𝑚𝑎𝑟𝑘 ∑ 𝑝𝑟𝑜𝑝𝑒𝑟𝑡𝑦 𝑚𝑎𝑟𝑘 ∙ 𝑤𝑒𝑖𝑔ℎ𝑡 𝑤𝑖𝑡ℎ 𝑖 𝑛𝑢𝑚𝑚𝑒𝑟 𝑜𝑓 𝑝𝑟𝑜𝑝𝑒𝑟𝑡𝑖𝑒𝑠

614

(14)

(15)

Method for concept design and optimization of twist beam axles Besides the reduction of the optimization dimension from 12 to 1, the marking system has the advantage, that different evaluation criteria of different OEMs for different vehicles can be adjusted in the optimization, since usually no perfect concept can be found which meet all the requirements perfectly due to the complex interactions of the parameters shown in Fig. 1. This marking system enables a quantitative comparison of different concepts with the best, but customer dependent concept. The smaller the mark is, the worse the concept is. With this marking system, the automatic concept design of twist beam axles becomes to a nonlinear constrained scalar optimization, which is a common problem in engineering practice. Many different algorithms have already been developed for this type of optimization. With an extensive investigation of various algorithms regarding their applicability, robustness and efficiency the following 5 algorithms are applied in this work: Sequential Quadratic Programming (SQP) [9], Interior Point, Active Set, Trust Region Reflective and Genetic Algorithm (GA) [10].

4.1.2 Parametric definition of package space Based on the analytical calculation we can now optimize the geometric design variables within a given package space by using a mathematical optimization method, which is very fast and unambiguous. For the design sensitivity analysis and optimization, only the package space, the target value of the axle performance concerning stiffness and kinematic, weight and materials as well as their possible thickness range are necessary. In the beam model in Figure 3, the max. and min. section diameters for the sections Ul-QAl, QAl-Ql, Ql-QEl, QEl-HFl and HFl-HRl of the left trailing arm and the corresponding right arm defines the package space of the trailing arm. The max. and min. width and depth define the space of the cross member which can also be section-wise. All dimensions meets the real reference vehicle restrictions. The position of Ql is also variable.

4.2 Example of the optimization of an existing axle To examine the potential, the described optimization method was used to develop a concept axle, which is then compared to an existing one.

4.2.1 Restricted Optimization At first, a restricted optimization using an existing axle at original unchanged thicknesses of all parts and original hard points within the package space was carried out. The aim was to see if the concept created by the new analytical calculation and optimi-

615

Method for concept design and optimization of twist beam axles zation results is different or similar to the geometric design. The optimization parameters were the spatial arrangement and the cross-section of both trailing arms (sectionwise variable) and cross member as well as the weld seam length of the spring /damper seat to both trailing arms and cross member. Furthermore, the angular orientation of the cross member was considered in the optimization.

Figure 11: Overlay plot – new concept axle and existing axle

In Fig.11, the optimization results (values of the diameter or width and depth of the different sections) of the new concept is translated to a simple CAD design. The comparison of the new concept (light color) and the original axle (dark color) shows several differences. At first, the cross-section of the new cross member is widened and its position is shifted backward in comparison to the original axle. The spatial arrangement of the trailing arm changes strongly. It goes more inwards. Due to the highest loads occurred in the rear contact zone between the cross section and the trailing arms (highlighted in Fig.11 by a circle), the cross-section of trailing arm is locally increased. In addition, the orientation of the cross member is now in angle 𝛼 which is different from most known existing axles using 0° or 90° angles. This allows the cross member to be placed closer to the force reaction point (wheel contact patch + 30 mm pneumatic trail ), while keeping the shear center fixed. In this way, the kinematic behavior during opposite wheel travel remains the same, while the bending moments in the cross member are reduced, which leads to the reduction of weight.

4.2.2 Free optimization For the second optimization, besides of the part geometry, also the part thicknesses are variable. Only the hard points of the axle remain unchanged. The optimization result shows in addition to a clear change of the geometric design also a strong thickness reduction which results in a significant weight reduction of 32 % (Table 2). This is possible because the durability issues i.e. the peak stress values are not considered in the current calculation. In the next step, also the stress will be conducted so the thickness calculation for the durability consideration.

616

Method for concept design and optimization of twist beam axles Table 2 Comparison – new concept axles and existing axle Property

Unit

Original TBA

Restricted Free Opt. Opt.

Toe stiffness

N/min

100 %

0.0%

0.0%

Camber stiffness

N/min

100 %

1.67%

0.0%

Roll stiffness

N/mm

100 %

-1.0%

-1.0%

Lateral stiffness

N/mm

100 %

-0.84%

-0.84%

Weight

Kg

100 %

-0.9 %

-31.9%

Summary and outlook With the presented analytical method and the corresponding software FastCon, it is possible to analytically evaluate a concept of TBA and further optimize its design within a few seconds without using CAD and CAE. This method can substantially reduce the development time and also improve the concept quality. In future developments, this method should be extended to 1.) better predict the influences of different spring damper seats; 2.) calculate the peak stress in components and its connection area for durability check. 3.) The elastic behavior of bushings should also be integrated so that the software is applicable not only for the component level but also for the axle level development. It may be coupled with the existing MBS software.

Reference 1. F. Gose and A. Fischer: "Handing performance requirements of automobiles – discussed in context of the VW golf", in International Automobil Engineering and Manufacturing Meeting, Toronto, Canada, 1974. 2. S. Gies and C. Brendes: "Der neue Golf - Modularität als Prinzip", ATZextra, vol. 16, no. 6, pp. 56-63, 2012. 3. B. Heißing, et al. (2013) Fahrwerkhandbuch Grundlagen, Fahrdynamik, Komponenten, Systeme, Mechatronik, Perspektiven, 4. ed., Vieweg+Teubner, Wiesbaden. 4. X. F. Fang and K. Tan: "Analytical modeling of twist beam axles", International Journal of Vehicle System Modelling and Testing, Vol.13, No.1, 2018, 1-25 5. K. Tan and X.F. Fang: “Automatic concept design and optimization of twist beam axles”, Automotive and Engine Technology, Vol. 3, 2018, 141-158

617

Method for concept design and optimization of twist beam axles 6. W. Matschinsky: Die Radführungen der Straßenfahrzeuge. Verlag TÜV Rheinland, 1987. 7. K. Tan, Effiziente Auslegungs- und Optimierungsmethode für den Konzeptentwurf von Verbundelenkerachse, PHD thesis University of Siegen, 2018 8. A. Schumacher, "Optimierung mechanischer Strukturen Grundlagen und industrielle Anwendungen", 2. ed. Berlin Heidelberg: Springer Vieweg, 2013, pp. XII, 306 S. 9. P. T. Boggs and J. T. Tolle, "Sequential Quadratic Programming", Acta Numerica, vol. 4., pp. 1-51, 1995. 10. M. Mitchell, "An Introduction to Genetic Algorithms". A Bradford Book The MIT Press, 1998.

618

Technologies for a modular vehicle concept used in passenger and goods transport Christian Ulrich, Horst E. Friedrich, Jürgen Weimer, Robert Hahn, Gerhard Kopp, Marco Münster German Aerospace Center, Institute of Vehicle Concepts, Stuttgart

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 M. Bargende et al. (Hrsg.), 19. Internationales Stuttgarter Symposium, Proceedings, https://doi.org/10.1007/978-3-658-25939-6_49

619

Technologies for a modular vehicle concept used in passenger and goods transport

Current situation and challenges At the beginning of the year 2019, vehicles with diesel internal combustion engines (Euro 4 or older) have been banned from entering the city of Stuttgart. This is the temporary climax of current efforts to address traffic-induced air pollution. More than 500 sanctions in several European cities try to ban private cars from cities. Most likely, a lot of today car-owners will switch to public transport in the future, thus increasing the issue with peak-times. During these phases (primarily commuting) the public transport is overstrained. During “off-peak”-phases equipment like busses operate very inefficient or stand idle in the operators’ depot. New vehicle concepts are required to support the operators in on-peak-phases or revolutionize the transportation system completely. However, not only the individual and public transport is affected by this transition, but also commercial traffic. This is in contrast to the steadily growing transport volume of goods. For instance, the market for parcel delivery grew by 96% from 2000 to 2017 [1]. New urban transportation systems to solve these issues are under development: e.g. new city hubs with truck swap bodies, more small electric vehicles, or freight bicycles for the last mile. There are several indicators showing that future vehicle concepts should combine automated urban delivery and people transportation; e.g. the European Commission is asking clearly for the new design of innovative shared, connected, cooperative and automated vehicle concepts for mobility of people and delivery of goods [2]. On a vehicle-level, some players in the automotive market have already presented scalable and modular vehicle concepts (Figure 1). These vehicle concepts as well as the corresponding business models to address urban traffic issues are discussed in [3].

Figure 1: Vehicle concepts with scalable and modular structures (source: own compilation based on [4]).

620

Technologies for a modular vehicle concept used in passenger and goods transport Figure 2 shows ideas for a better transport mode transition based on new on-the-road modular vehicle concept ideas. Our proposed vehicle concept, which is described in the following sections, follows the modular approach.

Figure 2: Vehicle concepts with different possibilities to separate driveboards and capsules (own visualization).

A novel on-the-road modular vehicle concept: MAUDE The German Aerospace Center, Institute of Vehicle Concepts proposes a disruptive vehicle concept (MAUDE1) for transporting people and goods which is based on dynamic on-the-road modularization. Since it is designed for various applications, it offers answers for many areas of today's urban traffic. Our system is derived from a demand analysis to enable a successful business development in the urban transportation sector. The concept consists of an autonomous and electrified driveboard and separated capsules for the transportation of goods and people (Figure 3). Key functions have been designed more usable compared to other modular concepts (Figure 1): for instance, standardizable multi-purpose capsules or a driveboard-based lifting system. Other modular vehicle concepts are discussed in [3].

Figure 3: MAUDE – driveboard and capsules for various applications (own visualization).

1

Modular, Autonomous, Updateable, Disruptive, Electric

621

Technologies for a modular vehicle concept used in passenger and goods transport The main USP of the MAUDE concept is the interface between the capsules and the driveboard in form of a U with an integrated lifting system, which reduces the costs for the capsules and avoids expensive infrastructure – this is a major distinction to existing concepts which include the lifting system in the capsules or use handling robots. However, this results in a sophisticated packaging, because the driveboard’s space is very limited. Our approach offers the possibility of depositing and exchanging capsules on the ground without a lifting unit integrated in the infrastructure or in the capsules. This allows for effective and fast loading of cargo capsules on ground-level, e.g. with a trolley, which is especially useful for urban operation. The rear opening of the driveboard’s structure offers the further advantage that capsules with different lengths can be transported with the MAUDE concept. For the capsules a variety of different designs is possible: capsules for the transportation of passengers, parcels, waste, construction tools, etc. [3]. The people capsules must have their own distributed energy storage for the supply of heating, air conditioning, electronics, lighting, etc. As the passenger pods mounted on the side are designed with overhang, pods also have advanced vehicle sensors for autonomous driving on board. Simple cargo capsules come without an overhang, but can be built very easily and inexpensively without further electronics, similar to an ISO-Container. The following sections will give an overview to the key topics of the MAUDE vehicle concept, form structural design, drivetrain, and possible automation concepts.

Structural design of the vehicle concept By separating the driveboard from the capsules, the necessary requirements plays an essential role to define and optimize different levels of modularization as well as to design the concept for loading and unloading (possible variations: laterally, from behind, from the front, from above) but also for getting in and out of persons (possible variations: from behind, from front, from the side). These further functionalities of the vehicle concept lead in particular to additional weight which has to be compensated in order to realize the necessary payload and axle loads (Figure 4). When comparing a conventional vehicle with a battery electric, autonomous vehicle, the additional weight e.g. for the energy storage and additional features for automatization reduces the payload. If the modularization is added, the payload drops again. This additional weight due to the modularization must be countered with intelligent and economical lightweight construction.

622

Technologies for a modular vehicle concept used in passenger and goods transport

Figure 4: Necessary lightweight construction to ensure the necessary payload, axle loads and resource conservation [5].

In order to analyze the necessary load paths, which are effective for different vehicle concepts and package specifications, structural optimization procedures were used. This is essential to derive mechanical interfaces and load carrier points of the driveboard and capsules in detail (Figure 5). Based on simulation results (e.g. structure optimization in relation to vehicle torsion and bending), the first concept designs for a selected driveboard were implemented and analyzed (Figure 6).

Figure 5: Load path analyzes for different requirements and package specifications [5].

623

Technologies for a modular vehicle concept used in passenger and goods transport

Figure 6: Structural analyzes and concept designs for different load assumptions (excerpt; own visualization).

Requirements and concepts for the drive system An urban, modular-on-the road vehicle concept is a unique solution and requires a closer look to the driveline design. The following figure based on an initial MAUDE study gives a first overview with (1) rear wheel with suspension, (2) front wheel drive and (3) battery:

Figure 7: Driveline components (own visualization).

There are some important restrictions which have to be considered. The performance of the drives of a driveboard could be low compared to usual conventional vehicles in this size. A “Sprinter size” MAUDE application has a drive with 2*45 kW and 1600 Nm at the wheel and allows a max speed of 70 km/h and a grad ability of 20% and 6% at 60km/h. Based on the planned use cases [3], these are fully sufficient values. The urban application defines the max speed of about 70km/h (with 8% grade) and max. 20% grade should be sufficient for urban regions (e.g. garage ramps). Usual values above of 40% and more are defined as not relevant for the application (no rutted lanes or no start at road curbside).

624

Technologies for a modular vehicle concept used in passenger and goods transport

Figure 8: Requirements for the powertrain (own visualization).

Battery and energy management The new modular design opens new opportunities for the energy management. As shown in the schematic Figure 9, the MAUDE concept uses capsules with own batteries. To increase the usability the stand by time for charging of the driveboard should be as low as possible. A usual approach is the use of high power chargers with the known disadvantages of installed power, reduced battery lifetime, or costs. We follow a new way and equip some of the capsules with an own battery systems.

Figure 9: Energy Supply concept (own visualization).

Use-case investigations show, that the number of capsules is much higher than the number of driveboards (1/3 … 1/10 depending on application) [3]. The capsule could be charged in the infrastructure in hubs or at customer site with low power and sufficient time, so charging infrastructure cost are low. During normal operation the driveboard can drive with capsules without batteries, so driving range is limited by driveboard battery capacity. During the day the overall operational control manages several transport tasks with capsules with battery. In this case the driveboard battery will be charged via a converter by the capsule battery. Typical capsules with batteries could be e.g. passenger capsules or refrigerator goods capsules. In these both examples there is an additional energy demand for the use case, heating and cooling for cabin climate or cooling of goods.

625

Technologies for a modular vehicle concept used in passenger and goods transport This new system and the different combinations of driveboards and capsules have some more design parameters which have to be defined. For the additional degree of freedom new design rules have to be defined. Based on vehicle size and usual vehicle design parameters known from existing EVs, the energy consumption will be appr. 30 kWh/100km (Table 1). With the assumption of an overall average speed of 25 km/h (a typical urban bus line during week day has an av. speed of about 15-20 km/h) a daily distance of 600 km at 24/7 operation is possible. For autonomous vehicles with this maximum utilization, battery size and charging during operation are very free to design. A quite suitable battery size for the driveboard will be an 8h shift operation with a capacity of 60kWh. The necessary energy of 180kWh for a whole day operation will cause to large battery sizes for the driveboard, so for a full 24h autonomous operation, the disposition of capsules during the day operation has to be managed in this way that during a shift a capsule with a battery has to be operated. Depending on the usecase and capsule type different capsule design configurations are possible. Table 1: Initial battery design estimates. DB: Specification, Performance Energy consumption

30

kWh/100km

Average speed

25

km/h

Average power

7,5

kW

DB: One battery load Battery Capacity Range Operating time

60 200 8

kWh km h

System: Daily energy, range Energy/day

180

kWh

Range/day

600

km

CAP: Battery v1 Battery capacity

60

kWh

DC/DC - Power

20

kW

No C * Operating time

3*3

h

CAP: Battery v2 Battery capacity

60

kWh

DC/DC - Power

60

kW

No C * Operating time

626

3*1

h

Technologies for a modular vehicle concept used in passenger and goods transport The simplified table shows the impact: e.g. for a passenger capsule which will be used in an on-peak public transport operation, the operating time will be presumed 3h and during this time the Driveboard-battery (DB-Battery) has to be charged with 60kWh in the related shift. So Capsule-Battery (CAP-Battery) will be 60kWh+x (x for 3h cabin operation, climatisation, etc.) and installed DCDC Inverter Power is only 20 kW for continuous and effective charging of the DB-Battery. Operational estimates show that for some longer PT on-peak shift passenger capsules could be sufficient but might be critical for night operation, so some of the goods capsules have to be equipped with a CAP-Battery. The table CAP-Battery V2 shows the simple difference. Goods capsules in general have a much lower operation time in the DB e.g. on the way from hub to customer. With the exemplarily assumption of 2-way delivery a 30 min driving the energy for the 3rd shift has to be transferred from CAP-Battery to DB-Battery in one hour resulting in a DCDC Inverter Power of 60 kW. As shown in the chapter above, the wheels will carry in coupled state and max. payload the biggest share of the load. So normally the rear wheels of a vehicle could be equipped with twin wheels. However, the concept in Figure 7 does not allow a large wheelset due to the limited width of the alongside stage. Considering a further specification point, the available space in the rear wheel section is very limited. An integration of an electrical drive is very challenging. The location of the heavy driveboard battery of the vehicle (Figure 7) could be disposed in the head of the driveboard. There are several reasons for this position. A packaging in the stage in general is possible but only with an ineffective and expensive subdivision in modules. The place in front has the big advantage of load on the front axle and, more important, reduces the load of the critical rear wheels and allows a bigger overhang at the back. These design discussion shows the complexity of the overall energy system design. It depends very strong on the use-cases and has to be designed individually for each city application. However counter-arguments are known. It is obvious, that the overall system capacity is higher than the quick charging solution, but the first business model calculations show a cost advantage for the capsule battery solution for highest utilization. Furthermore a positive impact on lifetime of the battery is expected. Experience with 24h operation of battery systems in autonomous vehicles does not exist worldwide. So relative continues operation without temperature spikes (during quick charging) is positive for the lifetime of each technical system.

627

Technologies for a modular vehicle concept used in passenger and goods transport

Automation of the vehicle concept: first considerations and deriving of research questions Urban, on-the-road modular vehicle concepts are basically designed for full autonomous operation. The driveboard itself is more a robot without a driver workplace or seat and the capsules are also not equipped with a steering wheel etc. According to SAE Standards the concept will be designed for SAE 5 level. Compared to today’s autonomous movers our concept likes to follow an unconventional way not only in the design but also in the field of automation. We try to find an optimum for the location of the “automation control” between vehicle and infrastructure. Nearly all of the autonomous vehicles which have been demonstrated up to now have vehicle control system for operation and safety on different layers mainly installed in the vehicle. This usual approach results in a huge number of vehicle sensors and very high computation power in the vehicle. Considering the evolutionary development during the last years, integrating more and more ADAS (Advanced Driver Assistance Systems), and existing safety standards, this is a reasonable way established carmakers have to go. However, some applications have been presented where the vehicle control is managed within the infrastructure. Bosch has already demonstrated its valet parking based on infrastructure supervision and only using simple vehicles sensors and actuators [6]. There are different drivers to follow within our automation concept. Based on geometrical reasons it is appropriate to mount a variety of sensor at the top or in upper corners of the vehicle to get the best picture of the surrounding situation. However, we have analyzed from a business case of view, that most necessary components should be placed in the driveboard and not in the capsules. The business model estimates have led to the conclusion that the capsules should be as cost-effective as possible due to the much higher number of capsules compared to the driveboards. Driveboard with sensors is close to 100% operation rate while capsules only 5-20% depending on usecase. So, only very cheap commodity sensors should be allowed. A further aspect for specifying automation control in the infrastructure is the unsymmetrical design. We use different capsule sizes: long capsules for passenger transport (side entrance) and goods transport are allowed to protrude the driveboard. As Figure 10 shows, the driveboard with the long capsule (1) has a surrounding sensor view (2) but a sensor shadow at the back (3).

628

Technologies for a modular vehicle concept used in passenger and goods transport

Figure 10: Driveboard and capsule with sensor view (own visualization).

In normal forward operation this is usually not critical but for maneuvers like reverse driving and unloading the automation needs sufficient sensors for control and safety. Thus, it could be a possibility to install hubs, where the vehicle has to park backwards as smart zones (Figure 11). The MAUDE vehicle (1) drives for loading and unloading in a smart zone (2) where e.g. several cameras (3) are used for control.

Figure 11 Traffic Sequence Chart – Unloading in smart zone (own visualization).

The general discussion of “Autonomous Control” installed in the vehicle and / or in the infrastructure is ongoing. The research question is: Which are the advantages and disadvantages for solutions with focus on vehicle or infrastructure support for urban automated driving. In our specific on-the-road modular approach it is very important to determine how many (if any) sensors need to be installed in the capsules. The best and cost optimized solution is still open and has to be investigated in further research.

629

Technologies for a modular vehicle concept used in passenger and goods transport

Summary This paper shows a new highly on-the-road modular vehicle concept. Based on demand analyses and inquires with different user groups, we have specified the vehicle and done a first conceptual analysis. We discussed generally the impact of modularity on the weight of the overall vehicle and outlined a lightweight design approach. A load path analysis shows the stress of components compared to other vehicle concepts. The urban driving performance results in a very moderate design of the powertrain. The unique approach with a split battery in driveboard and capsule allows a real 24/7 operation of the driveboard. Furthermore, we discussed different automation concepts. This specific on-the-road modular approach is unique and has been evolved by the German Aerospace Center, Institute of Vehicle Concepts. Further conceptual work will be continued together with additional research partners.

Acknowledgment We thank Mr. Marc Rohrer2, Dr. Stephan Schmid2, and Mr. Christian Wachter2 for their assistance that greatly improved the manuscript.

Bibliography 1. “KEP-Studie 2018 – Analyse des Marktes in Deutschland,” BIEK, Jul. 2018. 2. “Digitising and Transforming European Industry and Services: Automated Road Transport,” European Commission, Brussels, H2020-DT-ART-04, 2018. 3. H. E. Friedrich, C. Ulrich, and S. Schmid, “New vehicle concepts for future business models.,” in 19. Internationales Stuttgarter Symposium, 2019. 4. Vehicle manufacturers and companies (internet query), Jan-2019. 5. TMG Consultants and DLR Institut für Fahrzeugkonzepte, “Ökonomischer und ökologischer Nutzen des Konzept-Leichtbaus: Ungenutzte Potentiale heben,” Leichtbau BW GmbH, Landesagentur für Leichtbau Baden-Württemberg, 2019. 6. “Automated Valet Parking,” Bosch Mobility Solutions. [Online]. Available: https://www.bosch-mobility-solutions.de/de/highlights/automatisiertemobilit%C3%A4t/automated-valet-parking/. [Accessed: 15-Jan-2019].

2

German Aerospace Center, Institute of Vehicle Concepts, Stuttgart

630

Experimental investigation of Miller cycle combustion technology with water injection Nils Neumann, Normann Freisinger, Guido Vent Daimler AG Prof. Dr. Thomas Seeger University of Siegen

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 M. Bargende et al. (Hrsg.), 19. Internationales Stuttgarter Symposium, Proceedings, https://doi.org/10.1007/978-3-658-25939-6_50

631

Experimental investigation of Miller cycle combustion technology with water injection

Abstract The future RDE legislation as well as the associated fleet consumption targets require a high efficiency of the gasoline engine under real driving conditions. Throttled operation is increasingly losing in significance for RDE in downsized gasoline engines. Conversely, engine operating points are increasingly being shifted toward high brake mean effective pressures in the boosted range. The knock limit frequently results in late centers of combustion, which are not optimal in terms of efficiency. Consequently, a high knock resistance of the gasoline engine is crucial for low fuel consumption under real driving conditions. In consideration of this, the Miller combustion cycle with early intake closing offers advantages with regard to efficiency. The additional use of water injection leads to a reduction of the cylinder charge temperature, which further increases the knock resistance. The combination of these two technologies in the interests of improved efficiency represents an attractive approach for future powertrain configurations. This article evaluates port water injection in comparison with direct water injection with regard to the potential for efficiency improvements. The article also demonstrates the thermodynamic effect of water injection on combustion and analyzes the emissions behavior. The experimental investigations are conducted on a newly developed singlecylinder engine based on the downsizing concept and with a high geometrical compression ratio.

1 Introduction Downsizing concepts offer efficiency advantages regarding friction behavior, dethrottling at low part-load, heating-up after cold-start and weight. Therefore this concept established itself in the past in turbocharged gasoline engines. With regard to the future RDE legislation, however, the throttled low part-load range is increasingly losing in significance. Instead, there are frequently high brake mean effective pressures under real driving conditions in downsized engines. The knock limit requires the ignition timing to be retarded, which results in correspondingly late centers of combustion. Consequently, there is a clear loss of efficiency of the engine at these operating points.

632

Experimental investigation of Miller cycle combustion technology with water injection

Figure 1: Impact of technology configuration on efficiency at full-load

Combustion cycles with high knock resistance are key for the further CO2 reduction of future engine generations. In consideration of this, the further development of the Miller combustion cycle with early intake closing plays an increasingly important role. The early closing of the intake valve during the intake stroke leads to an intermediate expansion phase, which reduces the effective compression ratio. This results in a reduction of the final compression temperature, which increases the knock resistance. The reduction of the effective compression ratio by the Miller valve timing allows the use of a higher geometrical compression ratio. The thereby lengthened geometrical expansion phase improves the efficiency. The use of water injection technology provides a new additional potential to improve the knock resistance. This allows a further increase of the geometrical compression ratio. Figure 1 shows schematically the impact of different technology configurations on efficiency at full-load. The following experimental investigations evaluate the combination of the Miller combustion cycle with water injection technology as an approach for future powertrain configurations. Especially the focus is on the potential of port and direct water injection to improve the engine efficiency.

2 Fundamental principles of water injection The active principle of water injection for increasing the knock resistance and, therefore, improving the efficiency of the gasoline engine is based on two basic mechanisms: ● Removal of heat from the working gas by: o Evaporation of water o Heating of water and evaporation up to boiling point

633

Experimental investigation of Miller cycle combustion technology with water injection o Vaporization of water from boiling point o Superheating of water vapor beyond boiling point ● Increase of the specific heat capacity of the cylinder charge

[1]

At high part-load, a reduction of the final compression temperature by the removal of heat through vaporization provides the most important potential to increase the knock resistance. Water has a vaporization enthalpy of 2257 kJ/kg, while premium-grade gasoline has a vaporization enthalpy of 420 kJ/kg [2]. This means that the vaporization enthalpy of water is more than five times higher than that of gasoline. To remove a high quantity of heat from the working gas up to the instant of ignition, it is necessary to vaporize as much water as possible. The higher the actual temperature reduction up to the instant of ignition in comparison with the theoretically maximum temperature reduction from complete vaporization of the water, the higher is the water injection efficiency. Wall wetting, incomplete vaporization, only local cooling and unfavorable working gas stratification can have a disadvantageous effect on the water injection efficiency. Currently published studies show that the main heat extraction of the working gas through vaporization of the water does not take place until in the compression phase [1]. The conditions in the combustion chamber during the intake phase lead merely to heating and evaporation of the water. Generally, the dilution of the working gas by inert water vapor and the increase of the specific heat capacity lead to lower process temperatures. With regard to the use of water injection in a gasoline engine, it is possible to distinguish various concepts. The water injection concepts are characterized by the integration of the injectors into the particular component group. Consequently, water injection can take place in the plenum, in the intake port or directly in the combustion chamber. As far as direct water injection into the combustion chamber is concerned, in addition to the injection of just water, it is conceivable to inject a water/fuel emulsion. The feasible positioning of the water injectors in a gasoline engine is presented in Figure 2. [1]

Figure 2: Water injection concepts [1]

634

Experimental investigation of Miller cycle combustion technology with water injection Direct water injection requires higher injection pressures than water injection into the plenum or intake port. Direct water injection therefore has high technical demands on the water injectors and pumps. A feedback-controlled high-pressure water system requires not only a control logic, but also further components such as a high-pressure rail, control valves and pressure sensors. Consequently, the concepts differ with regard to installation space requirements and system costs. A comparative assessment of the requirements for the individual water injection concepts is presented in Table 1. Table 1: Comparison of water injection concepts requirement

Plenum

Port

Direct

Direct-Emulsion

injection pressure

•

•

•••

•••

system complexity

•

••

•••

•••

packaging

•

••

•••

••

•

•

•••

••

system costs

[•] medium, [• •] high, [• • •] very high In the case of plenum or port water injection, the injection timing is theoretically freely selectable over a range of 720° CA. However, injection with the intake valves closed favors the formation of a film of water on the walls of the intake manifold and on the intake valve head. This can increase the diameter of the droplets entering the combustion chamber, thereby lengthening the vaporization time. Generally, wall wetting leads to a heat extraction out of the relevant component instead of the working gas, which has a disadvantageous effect on the water injection efficiency [1]. In the case of direct water injection, the injection timing can be independent of the gasoline injection timing. This results in a high degree of freedom, which allows specific injection strategies, such as multiple water injection or additional water injection on the exhaust stroke. Regarding the direct emulsion injection, the water injection timing is always coupled to the gasoline injection. While plenum injection leads to a cooling of the intake air in the intake manifold, port water injection and, in particular, direct water injection are aimed at a high removal of heat in the combustion chamber.

635

Experimental investigation of Miller cycle combustion technology with water injection

3 Specifications of the experimental engine Table 2: Specifications of the single-cylinder experimental engine Displacement [cc] Compression ratio [:1]

375 12.8

Combustion mode

homogenous

Gasoline injector

multi-hole solenoid

GDI pressure GDI-SOI Water injection

350 bar 290°CA BTDCF PWI, DWI

PWI pressure

7 bar

DWI pressure

200 bar

Intake camshaft

narrow Miller cam

Valve timing

early intake closing

All experiments were conducted on a single-cylinder engine with external boosting. The specifications of the experimental engine are compiled in Table 2. Designed according to the downsizing concept, the single-cylinder engine has a displacement of 375 cc. The applied Miller combustion cycle with early intake closing is combined with a high compression ratio of ε =12.8. In relation to this, the narrow Miller intake cam is designed for operation across the entire engine map. Operation with early intake closing has a disadvantageous effect on charge motion and, therefore, on combustion stability, especially at high load. To guarantee a sufficiently high ride comfort, measures in relation to charge motion are necessary. Consequently, the intake port is equipped with a tumble flap, which closes off the lower half of the port. This improves the charge motion to such an extent that operation of the single-cylinder engine at high load with early intake closing is possible. Concerning the fuel injection concept the gasoline direct injection (GDI) injector is positioned in the center of the combustion chamber, while the spark plug is integrated at the side between the exhaust valves. Figure 3 presents the detailed installation positions. The injector for port water injection (PWI) is positioned above the intake port, while the injector for direct water injection (DWI) is at the side of the combustion chamber below the intake valves. As well the gasoline as the port and direct water injectors are special prototypes. The common integration of the two water injection concepts on the same engine allowed a direct experimental comparison.

636

Experimental investigation of Miller cycle combustion technology with water injection

Figure 3: Positioning of GDI, PWI and DWI injectors on experimental engine

In all experiments, the boost pressure, exhaust backpressure, intake temperatures, liquid temperatures and pressures were adapted to the conditions of a similar multi-cylinder engine. The experimental single-cylinder engine is equipped with high-pressure indication in the combustion chamber as well as with a lambda meter. Temperatures, pressures and flowrates are measured at relevant measuring points by the test rig system. A Horiba MEXA-ONE exhaust measuring system is used to analyze the gaseous emissions, while an AVL Particle Counter is used to measure the particulate emissions.

4 Experimental investigations The following section demonstrates the potential of port and direct water injection for improving the efficiency of the Miller combustion cycle with a high compression ratio. There is also an analysis of the emissions behavior. The investigations are conducted exclusively at boosted higher part-load in the lower engine speed range. The operating point of 2000 rpm and 16 bar IMEP, which was used in all the experiments, represents an operating range that, due to the knock sensitivity, offers a high potential for improving the efficiency of high-compression downsized gasoline engines under real driving conditions. Because early intake closing requires charge motion measures, the lower port half is always closed off by the tumble flap in this operating range. The singlecylinder engine is constantly operated with a stoichiometric air/fuel ratio of lambda=1.

637

Experimental investigation of Miller cycle combustion technology with water injection

4.1 Influence of PWI with Miller combustion cycle

Figure 4: PWI water volume flow variation

The influence of port water injection with the Miller combustion cycle is investigated on the basis of three different injection timing points: early injection 350° CA BTDCF soon after opening of the intake valves; injection 300° CA BTDCF at approximately maximum intake valve lift; and 270° CA ATDCF with the intake valves closed. Figure 4 shows that, without water injection into the intake port, there is a comparatively low knock resistance. This resulting in a relatively late center of combustion of 21.5° CA. The knock resistance rises as the injected water volume flow is increased. Injection with the intake valves open offers virtually no advantage in comparison with injection while the intake valves are closed. From a water volume flowrate of 1.1 l/h, equivalent to a water/fuel ratio of 50%, it is not possible to observe any notable further reduction of fuel consumption. Overall, port water injection at this operating point achieves a 2.4% reduction of fuel consumption. There is no significant improvement of combustion stability. The results of the experiments illustrate various conflicts of objectives, which represent a challenge with regard to a further possible improvement of the efficiency of port water injection with the Miller combustion cycle. Due to the short opening time of the intake valve, the water injectors must be designed to guarantee the injection of a sufficient quantity of water. At the same time, a minimum possible diameter of the water droplets is desirable in order to guarantee high vaporization rates in a short time. The shape of the Miller valve lift curve is limited by the maximum allowable

638

Experimental investigation of Miller cycle combustion technology with water injection cam surface pressure as well as by valve lift-off due to acceleration forces. Therefore, the valve lift is reduced especially in the case of narrow Miller intake cams. Small valve lifts favor the wetting of the intake valve head. The Miller high-tumble intake port as well as the positioning of the port injector must be designed to prevent wall wetting as much as possible. At high flowrates in the intake port, it is possible for accelerated water droplets to wet the valve seat area as well as the cylinder wall due to their inertia. In addition, downsizing concepts favor wall wetting because of the small combustion chamber.

4.2 Influence of DWI with Miller combustion cycle

Figure 5: DWI water volume flow variation

To investigate direct water injection with the Miller combustion cycle, the influence of two injection timing points is analyzed. As already considered in the case of port water injection, one injection timing point is at 300° CA BTDCF. This allows a direct comparison of potential for the two water injection concepts at the same injection timing point. The second analyzed injection timing point for direct water injection is at 80° CA BTDCF in the compression phase. In water/SOI variations conducted as part of preliminary investigations, this injection timing point showed the greatest influence on knock resistance. Figure 5 illustrates that, in comparison with port water injection, direct water injection fundamentally leads to a higher knock resistance with the Miller combustion cycle. This becomes especially clear in the case of an identical injection

639

Experimental investigation of Miller cycle combustion technology with water injection timing point for the two water injection concepts. At an injection timing point of 80° CA BTDCF, at a water flowrate of 1.1 l/h, equivalent to a water/fuel ratio of approx.51%, direct water injection results in a significant advancing of the center of combustion. Overall, direct water injection achieves a maximum reduction of fuel consumption of 4.7% at the operating point. With higher water flowrates, there is no further improvement of efficiency. The experimental results show that, with the Miller combustion cycle, direct water injection leads to a higher water injection efficiency in comparison with port water injection. The high injection pressure, along with the possibility to inject in the compression phase with the intake valves closed, results in fine atomization with small droplet diameters. Consequently, a high quantity of heat is removed from the working gas in a short time, resulting in a high increase in knock resistance. With regard to combustion stability, direct water injection provides an improvement in comparison with port water injection.

Figure 6: Multiple direct water injection

With regard to injection timing, direct water injection offers a high degree of freedom. Moreover, the high injection pressure leads to short injection durations. This advantage can be exploited to realize special injection strategies, such as multiple direct water injection, in addition to single injection. Figure 6 illustrates the influence of multiple direct water injection in comparison with single direct water injection and port water injection. With multiple direct water injection, the same total quantity of water is split

640

Experimental investigation of Miller cycle combustion technology with water injection between three consecutive injections. The first injection is at 200° CA BTDCF on the intake stroke, while the second injection at 140° CA BTDCF and the third injection at 80° CA BTDCF is in the compression phase. In comparison with port water injection and single direct water injection, this water injection strategy results in a significant increase of knock resistance and combustion stability. At a water volume flow of 1.4 l/h, the maximum consumption reduction is 7%. A further increase of the water volume flow does not lead to any further improvement of efficiency. As the water volume flow is still increased, it is possible to shift the center of combustion up to 8° CA. The vaporization rate is improved by short consecutive injections. In relation to this, an impact of the water distribution in the combustion chamber though multiple injections and reduced wall wetting can be assumed. Although the injection timing point 80° CA BTDCF shows the highest knock resistance in the case of single direct water injection, multiple direct water injection with further advanced injection timing provides a higher efficiency potential. Consequently, the previously mentioned effects with multiple direct water injection predominate for improving the water injection efficiency.

4.3 Influence of PWI and DWI on emissions behavior

Figure 7: Emissions behavior of water injection concepts

Figure 7 presents the untreated emissions behavior of the water injection concepts. For all three water injection variants, an increase of the water volume flow shows a steady reduction of the NOx concentration at the same rate. The inert water vapor leads to a

641

Experimental investigation of Miller cycle combustion technology with water injection lowering of the temperature of the working gas, which provides a reduction of the formation of thermal NOx. Water injection also leads to slightly reduced CO emissions. One feasible reason for this can be related to a reduction of the endothermic dissociation reactions, which form CO in correlation to the working gas temperature [2]. Especially in the case of multiple direct water injection, the HC and particulate emissions increase significantly with rising water volume flow. A possible interference with the formation of a homogeneous air/fuel mixture favors the creation of particulates. The low working gas temperature also causes quench-effects and hinders oxidation processes.

5 Summary and outlook The experimental investigation of water injection with the Miller combustion cycle shows that, at the current development status, direct water injection, in comparison with port water injection, offers a significantly higher efficiency potential. The possibility to inject water into the combustion chamber, even with the intake valves closed, has decisive advantages particularly if using short intake valve opening times. The improved combustion stability also speaks in favor of the combination of direct water injection with the Miller combustion cycle. Multiple direct water injection in particular leads to a highly increased knock resistance, despite a more critical behavior with regard to emissions. Table 3: Future development topics “Water injection with Miller combustion cycle” technology dimensioning variables

Relevance for water injection concept

water injection • • • • • •

droplet size pressure temperature targeting position injection strategy PWI + DWI

intake port • • • •

intake valve train

wall wetting • valve lift tumble • cam width flow rate • valve timing charge cycle work

PWI

PWI

A high water injection efficiency offers decisive advantages for a customer-focused application scenario. This applies especially where water injection is used frequently for improved efficiency in the engine map. Without in-vehicle water recovery systems a low water consumption represents the key factor for minimizing the weight, water tank installation space and refilling intervals. Table 3 shows various development topics as an approach for possible improvement of the water injection efficiency with the

642

Experimental investigation of Miller cycle combustion technology with water injection Miller combustion cycle. To further harness the potential, it is necessary for the water injection concept, intake port and valve train to be matched to each other and further developed. At the same time, it is necessary to minimize the conflicts of objectives between the individual design variables. Overall, the experimental investigations illustrate that water injection with the Miller combustion cycle represents an attractive approach for a highly efficient powertrain technology combination of the future.

Abbreviations ATDCF

After top dead center firing

BTDCF

Before top dead center firing

CAC

Charge air cooler

CO

Carbon monoxide

DWI

Direct water injection

GDI

Gasoline direct injection

HC

Hydrocarbons

IMEP

Indicated mean effective pressure

MFB50%

Mass fraction burned 50%

NOx

Nitrogen oxides

PWI

Port water injection

SOI

Start of injection

TC

Turbo charger

TDC Water/fuel-ratio

Top dead center Ratio water (kg/h) to fuel consumption (kg/h)

Bibliography [1]

Hunger, M., Böcking, T.: Potential of Direct Water Injection to Reduce Knocking and Increase the Efficiency of Gasoline Engines. Günther, M., Sens, M.: Knocking in Gasoline Engines. 5th International Conference, December 12–13, 2017, Berlin, Germany, Springer

[2]

Pischinger, R., Klell, M., Sams, T.: Thermodynamik der Verbrennungskraftmaschine. List, H.: Der Fahrzeugantrieb. 3. Auflage, 2009, Wien, New York, Springer

643

Reduction of cold start emissions with microwave heated catalytic converters Viola Papetti, Panayotis Dimopoulos Eggenschwiler, Daniel Schreiber Empa, Swiss Federal Laboratories for Materials Science and Technology Automotive Powertrain technologies, CH-8600 Dübendorf, Switzerland. [email protected], Tel +41 58765 4775

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 M. Bargende et al. (Hrsg.), 19. Internationales Stuttgarter Symposium, Proceedings, https://doi.org/10.1007/978-3-658-25939-6_51

645

Reduction of cold start emissions with microwave heated catalytic converters

1 Introduction Emissions of Euro-6 gasoline powertrains at normal operating temperatures are almost zero (Fig. 1). On the contrary, during cold starts, characterized by low temperatures for the catalytic converters, CO and unburned hydrocarbon (HC) emissions are 2-3 orders of magnitude higher than during cruise operating conditions [1]. The situation is even more severe with hybrid powertrains. Here the intermittent engine operation often leads to numerous cold starts. In parallel, the steadily increasing efficiency of modern engines has dramatically lowered exhaust gas temperatures, both in gasoline and diesel engines (first law of thermodynamics). All near-future engine concepts will face increasing difficulties conflicting with the fuel saving imperative in providing the necessary heat for the pollutant aftertreatment. Depending on the application, load conditions and aftertreatment technology, sometimes “artificial” supply of energy to the catalytic converter is required in order to meet emission limits. One potential solution was the utilization of electrically heated catalytic converters. Electrical Heated Catalysts (EHC) that uses mechanical energy from the engine to generate the electrical energy to heat up the exhaust gas upstream the catalytic converter is proposed in [2, 3, 4, 5].

Figure (1) – Passenger car (Euro- 6) emissions with a gasoline powertrain during the IUFC-15 cycle (measurements at Empa APTL).

In literature, microwave heating is often applied to chemical reactions: [6] discusses different designs for microwave field applied to Small-Scale Chemical Processing. In the exhaust aftertreatment field, microwaves have been used for initiating active regeneration of particle filters [7], as well as for sensing the urea level in an SCR catalytic converter [8, 9]. In this study, we describe a microwave heating system applied to catalytic converters for automotive applications and we show its relevant contribution in emissions reduction during cold start conditions. Experiments with synthetic exhaust gas and inflow gas temperature of 20°C have shown conversions of more than 60%. A vehicle size system mounted at the end of the tailpipe of a gasoline vehicle achieved a reduction of

646

Reduction of cold start emissions with microwave heated catalytic converters 15% of cumulative emissions in the first 600s after cold start. Further optimizations related to higher microwave power, substitution of conventional honeycomb (HC) with additive manufactured polyhedral catalytic converters and to improved layout have the potential to further increase cold start emission conversions.

2 System design Heating the catalytic converter with microwaves requires a microwave source, a wave guide, which is an extra volume responsible for the wave transmission from the source towards the cavity, and the cavity, where the catalytic converter is placed and the stationary microwaves develop. Each of these components has to be dimensioned and tested. A household (kitchen appliance) magnetron was chosen as a microwave source. The power of 1.3kW was deemed as sufficient. The low price of such mass produced components was an additional argument for the choice. Moreover, the materials of the entire catalytic converter, or at least parts of it, have to be good microwave absorbents.

2.1 Material selection Conventional cordierite (MgO-SiO2-Al2O3) is not absorbing microwaves. SiC proves to be the material of choice. SiC is already widely used in exhaust aftertreatment based on its good thermal and mechanical properties. In order to achieve the best dielectric properties, different slurries of SiC have been prepared and analyzed. The catalytic converter used in the present study comprises of a sequence of slices, each of identical cylindrical shape, either out of SiC or Al2O3. SiC slices have been taken out of a non-coated DPF, 200cpsi, and have been coated in house with Al2O3/Pt (6070grPt/ft3). Coated SiC slices are denoted by SiCc. For testing only the microwave absorption, sometimes, non-coated SiC slice have been used. In these cases they are denoted by SiC. The Al2O3 slices have been reused from another project. They had a foam structure and have been coated with Al2O3/Pt by an industrial coater. The notion used therefore is Al2O3c. Thus, the catalytic converters used in this study have rather an oxidative and reduction behavior.

2.2 Cavity design A sophisticated system composed by the microwave guide and the cavity for vehicle size applications has been designed based on simulations. Proper system dimensions must be chosen in order to obtain stationary waves in the cavity. The dimensions of the waveguide are mainly given by the wave length (=v/frequency) of the microwave. The cavity diameter was chosen in order to have typical passenger car catalytic

647

Reduction of cold start emissions with microwave heated catalytic converters converter dimensions. The length was a result of the simulations in order to enable the best stationary wave (Fig. 2). The microwave cavity is made out of steel or aluminum with bolted connections. The designs avoided any welded metal connections since the interaction of austenitic steel with microwaves is not so clear. The presence of the catalytic converters was neglected in the simulations in order to obtain an optimal design in an acceptable computational effort. For further optimization and development the detailed modelling of the interaction of the catalyst with the microwave is indispensable. Based on the simulations results, the system described in Fig. 2 has been realized (Fig. 3). Unfortunately, there is no way to position a thermocouple in the cavity. Thermocouples are immediately destroyed when the microwave is on. All measurements have been accompanied by microwave sensors in order to identify possible escape of the microwaves form the cavity. No leaks have been ever identified. Thus, no further actions have been considered as necessary in this field.

Figure (2) – Technical dimensions of the microwave guide and of the cavity.

Figure (3) – Microwave guide and cavity apparatus.

648

Reduction of cold start emissions with microwave heated catalytic converters

3 Experiments 3.1 Preliminary experiments with synthetic exhaust gas A synthetic exhaust gas of 295 liter/minute comprising of 800ppm CO mixed with air at 20°C as inlet temperature was tested with the apparatus described in section 2. Several concepts in terms of number, sequence and position of different slices out of SiC and Al2O3c have been tested in order to define the optimal one. It is worth noting that in all the experiments the synthetic exhaust gas had no water vapor. Microwave heating started at 50s and it was kept switched on for 200s. Fig. 4 reports the different configurations that have been tested: A. 1 SiC slice (uncoated acting as microwave absorbent) and 1 Al2O3c slice (oxidative catalytic coating as described in section 2.1), B. 2 SiC and 2 Al2O3c slices, C. 1 SiC and 2 Al2O3c slices, D. 1 SiC and 2 Al2O3c slices, in an optimized position with respect to c), based on the simulation results.

Figure (4) – Schematic of the tested concepts in the designed cavity, in orange the wave guide.

Fig. 5 represents the outlet temperature profile in time with the different configurations described in Fig. 4. It is evident that:

649

Reduction of cold start emissions with microwave heated catalytic converters ● The flow enters at 20°C. The microwave heating system is increasing the flow temperature immediately. ● The microwave heating increases the temperature at discrete spots in the catalytic converters, where the reactions start happening. The larger part of the gas passes through the unheated parts and remains at 20°C. Thus, the temperature measured in the catalytic converter exit corresponds to the mix of cold and heated gases. ● The highest temperature is reached with case A, corresponding to the configuration with lower thermal inertia (lower mass). As long as cases C and D, characterized by lower thermal inertia with respect to case B, have lower temperature with respect to case B, this means that, in case B, more reactions are happening. ● Among cases C and D, case D confirmed the optimization. ● Due to the higher inertial mass, case B reaches maximum temperature later in time with respect to the other cases.

Figure (5) – Catalytic converter Outlet Temperature with a synthetic gas flow of 295l/min and 20°C at inlet.

Figure (6) – Cumulative CO profile in ppm with a synthetic gas flow of 295l/min and 20°C at inlet.

650

Reduction of cold start emissions with microwave heated catalytic converters In Fig. 6 the corresponding cumulative outlet CO emissions are reported. Main conclusions are: ● The optimized configuration, case D, resulted in the highest emission reduction. Instantaneous CO oxidation reached 75%, while cumulative CO oxidation 42%, this at gas inflow temperature of 20°C. ● The higher conversion reached by case D in respect to case C is clearly due to the optimized positioning of the SiC slice; a better positioning in respect to the location of the next maxima of the stationary (micro-)wave. ● Case B with 2 SiC and 2 Al2O3 slices reaches higher conversion in respect to cases A and C although it has higher thermal inertia. The additional SiC slice in case B absorbs more energy from the microwaves. ● Case A reaches the highest temperature (Fig. 5) but is characterized by the lowest conversion (Fig. 6). This evidences that additional catalyst slices and additional microwave absorption capacity have the potential to further improve the system's behavior.

3.2 Extended measurements with synthetic exhaust gas Based on the first results, a vehicle size catalytic converter with a volume of approx. 2lt has been configured. Measurements with synthetic exhaust gas (with more complex composition) and at higher gas flows aimed to identify a configuration suitable for vehicle applications. The behavior in the preliminary measurements suggested testing of catalytically coated SiC slices, thus slices absorbing microwaves as well as acting as reactors, should the local temperature exceeds certain levels. The synthetic gas flow was set at 600 lt/min, comprising of 1000ppm CO, 1000ppm C3H6, 5%O2 and N2 for balance at 20°C as inlet temperature. It is worth noting that in all the experiments the synthetic exhaust gas had no water vapor. Similar to the measurements presented in the former section, the gas flow enters the cavity with ambient temperature (20°C). New configurations with additional slices have been compared in terms of HC and CO oxidation (Fig. 8). In all the configurations except for case e, the microwaves are switched on after 50s from the beginning of the flow. In case e a slightly different strategy was followed. The microwave was switched on 75s before the entrance of the gas flow in the catalytic converter, thus case e is characterized by 50s of preheating of the catalytic converter. In all configurations the microwave heating was operating for 150s. Fig. 7 shows a schematic of the tested assemblies:

651

Reduction of cold start emissions with microwave heated catalytic converters a.

Configuration number 1, with 1 SiC uncoated slice, 2 SiC coated slices and 1 Al2O3c slice and x=30mm,

b. Configuration number 1, with 1 SiC uncoated slice, 2 SiC coated slices and 1 Al2O3c slice and x=50mm, c.

Configuration number 2, with 1 SiC uncoated slice, 2 SiC coated slices and 2 Al2O3c slices and x=55mm,

d. Configuration number 2, with 1 SiC uncoated slice, 2 SiC coated slices and 2 Al2O3c slices and x=50mm, e.

Configuration number 2, with 1 SiC uncoated slice, 2 SiC coated slices and 2 Al2O3c slices and x=55mm. Thus identical to case c, but 75s of the 200s of preheating of the catalytic substrates is without flow through.

Figure (7) – Tested configurations for identifying the optimal catalyst for vehicle testing.

652

Reduction of cold start emissions with microwave heated catalytic converters (a)

(b)

Figure (8) – Temperature profile (a) and C3H6 emissions in ppm (b) with a synthetic gas flow of 600l/min and 20°C at the cavity entrance.

Fig. 8 presents the outlet temperature profile and the C3H6 outlet concentration respectively, over time, obtained with the different configurations. Without switching on the microwaves, no reactions and no conversion have been detected as well as no temperature increase over the cavity. Case e with the 75s of preheating of the slices with the microwaves resulted in the highest conversion and among the highest gas temperature increase over the cavity. C3H6 oxidation exceeds 50% and this at gas inlet temperature of 20°C. All cases (c,d,e) with the higher number of slices resulted in higher conversion. The preheating, not only increases conversion, but also contributes in reaching it earlier. The results confirm that the temperature and emissions are sensitive to the position of the catalytic converters.

3.3 Vehicle Experiments A system similar to the configuration 2 in Fig. 7 has been mounted at the end of the tailpipe of a gasoline vehicle. (Fig. 9). It is impossible to mount the microwave heated catalyst system in the conventional position of the serial catalytic converter without the collaboration of an OEM and access to the ECU. The position for microwave system is the most unfavorable, since, during cold start, the exhaust has to flow through and warm up the original vehicle catalytic converter (near the exhaust manifold) and the entire exhaust gas guiding system (pipes and mufflers), thus arriving with the lowest temperatures at the additional, microwave heated catalyst system. The exhaust gas temperatures at the tailpipe, and thus upstream the microwave heating system, remain below 100°C even 600s after engine start.

653

Reduction of cold start emissions with microwave heated catalytic converters

Figure (9) – Installment of the new catalytic converter at the end of a vehicle tailpipe on the chassi dynamometer.

The vehicle was tested on the chassis dynamometer where the first 15 minutes of a WLTC cycle have been followed. A series of cold starts (20°C) has been performed. After each cold start, no vehicle activity followed for 4 hours in order to allow all relevant parts of the vehicle to cool down to the ambient temperature of 20°C. The electrical power consumed by the magnetron for generating the microwaves has been kept at 1.3kW, a pre-heating time of 100s was selected while the microwave was switched on for 300s in total. The emissions before and after this "tailpipe" catalytic converter have been analyzed. Fig. 9 shows the vehicle installment as well as the catalytic converter positioning and the emissions analysis setup.

Figure (10) – Temperature difference over the catalytic converter during 5 cold starts (4 with microwave and one without microwave preheating).

654

Reduction of cold start emissions with microwave heated catalytic converters Fig. 10 shows the temperature difference over the catalytic converter during 5 different cold starts. The temperature difference over the catalytic converter is negative without the microwave heating (solid dark line). The exhaust gas loses heat in order to heat up the cold converter. In contrast, the temperature difference over the catalytic converter is positive in all cases as long as the microwave heating is switched on. The differences among those cycles are attributed to the differing catalyst activity. This in turn, can be attributed to the variation of the unburned hydrocarbon amounts coming from the engine in the different cold starts.

Figure (11) – Cumulative Total Hydrocarbon emission during a series of cold starts.

In Fig. 11, the solid dark line represents one measurement with no microwave heating. Microwave heating system allows a cumulative emission reduction in the initial 600s after cold start of around 10-15%. Based on these measurements and observations it is expected that a similar system will reduce cold start emissions for more than 30% if positioned near the exhaust manifold, a position typical for state of the art aftertreatment systems.

4 Conclusions A System for heating up the catalytic converter during cold starts has been developed and tested. Based on a conventional 1300W magnetron, the system comprises of a waveguide, a cavity and microwave absorbing components. As microwave absorbents, SiC honeycomb substrates have been used. Such substrates, catalytically coated, showed the highest pollutant conversion performance at low temperatures.

655

Reduction of cold start emissions with microwave heated catalytic converters In synthetic gas supplied at 20°C, pollutant conversions have exceeded 50% with the optimized configuration. At the tailpipe of a vehicle a similar system has achieved a reduction of 15% of the emissions in the first 600s of the WLTC cycle. The system's performance will increase if mounted in position nearer to exhaust manifold and by using a stronger magnetron. Further optimization of the configuration can also result to additional benefits.

Acknowledgements The authors gratefully acknowledge financial support from the Swiss National Science Foundation (SNF) for the project “Cases: Catalyst Substrates from Engineered Structures for state of the art powertrains: High pollutant conversion (including cold-start) with low precious metal requirements”, Project No 20PC21_161571/1, as well as from the Swiss federal Office for the Environment (FOEN) for the project “Development and manufacturing of an automotive catalyst concept for cold start low exhaust temperatures”, Project No UTF 522.14.15 IDM 2006.2423.485.

Bibliography 1. INFRAS (2014); Handbook Emission Factors from Road Transport (HBEFA) 3.2 2. M. Presti, L. Pace, L. Poggio, V. Rossi, “Cold Start Thermal Management with Electrically Heated Catalyst: A Way to Lower Fuel Consumption”, 2012 SAE International Congress and Exposition, Detroit, Michigan (USA) 3. A. Schatz, T. Knorr, D. Ellmer, "Elektrischer Heizkatalysator zur Optimierung der Emissionsn von Mildhybridsystemen", MTZ 2/2016 4. R. Brück, P. Hirth, B. Hu, C. Schorn, "Neue Katalysatorträger-Innovation für RDE- und SULEV-30-Grenzwerte", MTZ 1/2017 5. J. A. Avila, "Zukunftspotenziale heben", MTZ 2/2018 6. G.S.J. Sturm, “Microwave Field Applicator Design in Small-Scale Chemical Processing”, Phd thesis, Delft University of Technology, 2013 7. J. M. Strack, "Regeneration von Dieselpartikelfiltern durch Mikrowellen" PhD Thesis, Technische Universität Bergakademie Freiberg, 2008 8. D. Rauch, "Microwave-based Characterization of the Ammonia Loading of SCR Catalysts Materials" PhD Thesis Universität Bayreuth, 2016 9. D. Rauch, D. Kubinski, U. Simon, R. Moos: Detection of the ammonia loading of a Cu Chabazite SCR catalyst by a radio frequency-based method, Sensors and Actuators B: Chemical, 205, 88-93 (2014), doi:10.1016/j.snb.2014.08.019

656

Extended cylinder deactivation strategies to improve CO2 and pollutant emissions for light-duty Diesel engine applications Kai Deppenkemper, Markus Schönen, Bernd Lindemann FEV Europe GmbH Mauro Scassa FEV Italia s.r.l. Matthew Younkins, Robert Wang Tula Technology Inc.

Extended cylinder deactivation strategies to improve CO₂ and pollutant emissions …

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 M. Bargende et al. (Hrsg.), 19. Internationales Stuttgarter Symposium, Proceedings, https://doi.org/10.1007/978-3-658-25939-6_52

657

Extended cylinder deactivation strategies to improve CO₂ and pollutant emissions …

Abstract Further improvement of the trade-off between CO2- and pollutant emissions motivates the development of new diesel engine concepts. The deactivation of one or more cylinders of a light-duty diesel engine during low load operation can be a sophisticated method to improve fuel economy and reduce especially NOX emissions at the same time. This publication presents the potential of extended cylinder deactivation (CDA) strategies by means of steady-state, 1D simulations in combination with transient powertrain simulations. The transient cycle investigations have been conducted using the Worldwide Harmonized Light Vehicle Test Procedure (WLTP) and Real Driving Emissions (RDE) cycles. A representative mid-size SUV, powered by a state-of-the-art 2.0 l, 4-cylinder diesel engine, was considered for these investigations. The engine is equipped with both high and low pressure EGR and a closed-coupled DOC/SCRF system with an additional underfloor SCR. The potential of a selective CDA of separate cylinders by Dynamic Skip Fire (DSF®) technology, as part of an extended CDA strategy, has been analyzed in detail. The results show a CO2-benefit of approximately 2 % under a RDE cycle, while simultaneously reducing tailpipe NOX emissions by approximately 11 %. These improvements have been achieved by an increased engine efficiency due to a shift in the engine operation conditions by the extended CDA strategies. Implementing these CDA strategies also results in higher exhaust temperatures which further improves the conversion efficiency of the aftertreatment system. Furthermore, the investigations have been extended to show the potential of this strategy when applied to an 8.0 l medium duty diesel engine application.

1

Introduction: Motivation and approach

The cylinder deactivation on a diesel engine has showed potentials on the one hand side to further reduced pollutant emissions, while on the other hand to gain some fuel economy in parallel. This has been demonstrated by several investigations in the past. Nevertheless, a static deactivation of half of the cylinders is limited by their operation range. An additional dynamic deactivation of several cylinders delivers further degrees of freedom that could provide an extension of the cylinder deactivation operation range. The authors have used of different simulation tools such as 1D steady-state engine process model and transient mean value model to represent the possibilities of a dynamic cylinder deactivation on diesel engine applications.

658

Extended cylinder deactivation strategies to improve CO₂ and pollutant emissions …

2

Boundary conditions of the simulation models

The authors have used state-of-the-art diesel engine for passenger car (PC) and medium duty (MD) truck applications for the investigation program. Table 1 summarizes the powertrain and vehicle specifications. Table 1 – Powertrain and vehicle specification Parameter

Segment C

Compact SUV

MD truck

Vehicle mass corresponding to homologation cycle

1470 kg

1946 kg

14583 kg

Transmission

7-speed DCT

8-speed DCT

6-speed AT

Engine

Displacement:  2.0 L diesel engine Performance:  60 kW/L Boosting system:  Single-stage VNT EGR system:  Uncooled HP-EGR  Cooled LP-EGR

Displacement:  7.7 L diesel engine Performance:  30 kW/L Boosting system:  Single-stage WG EGR system:  Cooled HP-EGR

Closed-coupled DOC & SDPF Passive underfloor SCR (Aged system)

Closed-coupled DOC, DPF & SCR (Aged system)

Aftertreatment system

For the PC applications a 2.0 l 4-cylinder diesel engine with a single stage boosting system and a compression ratio (CR) of 15.5 has been considered. Further engine applications have been an advanced exhaust gas recirculation (EGR) system (uncooled high and cooled low pressure EGR path) and a 2000 bar fuel injection system (FIS). It has been decided to investigate two different vehicles, a C segment vehicle, as well as a compact SUV. Those have been equipped with a 7- and 8-speed dual clutch transmission (DCT). The exhaust aftertreatment system has installed a closed-coupled DOC, SDPF as well as a passive underfloor SCR. All EATS components have been used as

659

Extended cylinder deactivation strategies to improve CO₂ and pollutant emissions … aged system. The cycle investigations have considered the standard WLTC and a RDE operation. The MD truck has been powered by a 7.7 l 6-cylinder diesel engine. The air path has a standard wastegate turbine (WG) boosting system together with a cooled HP-EGR system installed. The combustion system has considered a 2400 bar FIS and a CR of 17.7. A state-of-the-art EATS based on closed-coupled DOC, DPF and SCR has been installed. For the MD truck application the WHTC has been considered.

2.1 1D engine process simulation model The commercial 1D engine process simulation software GT-SUITE has been used to investigate the thermodynamic reactions of the different exhaust gas heating strategies. The 1D engine model has considered the entire engine configuration, such as the boosting system, the air and exhaust path, the EGR path (high pressure and low pressure) and combustion chambers. The burn rate of fuel combustion has been implemented through profile arrays from several engine operation points of the entire engine operation range. Those have been generated by a standard 0D approach of cylinder pressure analysis of steady-state experimental engine measurements. The entire EGR control of the model has been modified from a mass flow control to an oxygen concentration control. The fuel injection pattern and rail pressure as well as boost pressure set points have been kept constant. This 1D model can operate in the entire map range, and allows simulation throughout the entire engine operation range. Standard PID-controllers have been used to control components like EGR valves or turbocharges in order to regulate EGR rates or boost pressure under steady-state investigations. Finally, a sub-model for engine-out emission predictions had been added to the engine model. This uses the physical correlation approach of in-cylinder O2-concentration to predict engine-out NOX and soot emissions [1, 2]. Thus, transient effects on emission production have been considered, which usually occur at dynamic engine operation. In addition, HC and CO emissions have been implemented by steady-state maps which dependent on engine speed and load. The approach describes the standard at FEV Europe GmbH and has been used in the past (see Deppenkemper [3]). To obtain an accurate results, the 1D model has been validated to surrogate data. The accuracy of boost pressure showed a deviation of maximum 1 %. The calibration level of the emission models were more challenging and provided a maximum deviation of 5 %.

2.2 Map calibration for considered heating strategies To investigate the exhaust heating potentials of the different exhaust heating strategies within the mean value powertrain model (MVPM), the baseline engine-out maps have

660

Extended cylinder deactivation strategies to improve CO₂ and pollutant emissions … to be adjusted, based on the results of the 1D model simulations. For this purpose, differential and factorized maps have been generated and added incorporated into the base engine maps. Together with the differential and factorized maps, a new engine calibration with a specified exhaust heating strategy has been considered. The adjustment process is shown in Figure 1.

Figure 1 – Generation of engine-out maps for the transient heating strategies investigations

2.3 Mean value powertrain model The FEV Complete Powertrain Simulation Platform, a pre-cursor of FEV’s advanced VCAP calibration platform was utilized in this study. The powertrain model [4] has integrated five main sub-models for boundary/ambient conditions, vehicle settings, transmission, engine and the aftertreatment system. The boundary/ambient condition sub-model described the different road conditions, emission test cycles and different driver behaviors. Inside the vehicle model the rolling resistance as well as road influence, aerodynamics and gravity were considered to model vehicle longitudinal dynamics. The main transmission and driveline components were modelled with ideal torsional systems, subjected to a distinct efficiency at different oil temperatures. Based on those sub-models, the main objective was to calculate the required inputs for the engine, mainly actual engine speed and load request. The engine model provided than on the specific operation point the corresponding engine out conditions, which were described by calibration maps at different coolant temperature. Further information on FEVs mean value powertrain model can be found in [5].

661

Extended cylinder deactivation strategies to improve CO₂ and pollutant emissions …

2.4 Selective cylinder deactivation by Dynamic Skip Firing Dynamic Skip Fire (DSF) is an advanced cylinder deactivation technology. A DSFequipped engine has the ability to selectively deactivate cylinders on a cylinder eventby-event basis in order to match the torque demand at optimum fuel efficiency while maintaining acceptable noise, vibration and harshness (NVH). To illustrate this concept, Figure 2 shows an example of DSF operation in a four cylinder engine. A varying torque request is shown in green, which results in cylinders being fired (red) or skipped (grey). The combined firing pulse train for all four cylinders is in blue. As torque demand increases, the density of firing cylinders also increases. When torque demand is zero or negative, no cylinders fire. This is termed DCCO, or deceleration cylinder cutoff. [6]

Figure 2 – Dynamic Skip Firing operation [6]

3

Evaluation of simulation results

The authors have substituted the evaluation process into two tasks. The first task has dealt with the steady state simulation investigations of the different heating strategies by means of 1D engine process models. Whereas the second task has focused on transient cycle investigation.

3.1 Analysis of steady state 1D engine process simulation results The 1D steady-state investigation have been obtained for partly loaded operation. Those investigations have been done under four different fire density (FD) levels, where 1 indicates full cylinder operation. A FD of 0.25 is equal to a single cylinder operation out of this 4-cylinder engine. The steps in between are defined as 0.75 and 0.5. The engine operation at a FD below 1 has led to an anomalous turbocharger operation due to the changed exhaust gas dynamics. Therefore, reduced boost pressure levels have been achieved and resulted in a limitation of the maximum engine load operation. Figure 3 shows a schematic of maximum engine operation loads that can be achieved at different FD levels.

662

Extended cylinder deactivation strategies to improve CO₂ and pollutant emissions …

Figure 3 – Schematic of Dynamic Skip Fire at firing densities equivalent to individual cylinder deactivation

Since the deactivation of one or more cylinders the load at the remaining fired cylinders have been increased to hold a constant engine power output. The increased inner load has provided than a higher exhaust temperature at a higher engine efficiency. Figure 4 summarizes the relative simulation results at a FD = 0.5 of engine efficiency improvement by BSFC and absolute exhaust temperature increase in the lower part load area. It can be seen, that FD of 0.5 has provided a fuel consumption benefit of 15 % in average in the shown operation area. At the same time an exhaust temperature increase of almost 130 K at 3 bar of BMEP has been achieved in comparison to a 4-cylinder operation. Additionally to the mentioned advantages other effects have occurred by a steady-state cylinder deactivation. On the one hand a reduction of the exhaust mass flow rate has obtained by deactivating cylinders. Hence, also a lower emission engine-out mass flow rate has been achieved. While this has delivered on the other some degrees of freedom to lower the steady-state EGR calibration to keep the same NOX engine-out mass flow rate compared to a 4-cylinder operation.

663

Extended cylinder deactivation strategies to improve CO₂ and pollutant emissions … 5 Delta BSFC / % (FD 0.5 - FD 1)/FD 1

BMEP / bar

4 3

-4

3

-6

-6

-1

-5 -10

2

-15 -20

1 0 5 Delta exhaust temperature / K (FD 0.5 - FD 1)

BMEP / bar

4 2

130

128

3

11

2500

3000

60 40

1 0 1000

100

80

2

129

120

1500

2000 Engine Speed / (1/min)

Figure 4 – Steady-state simulation results of FC potential and exhaust temperature increase with FD 0.5

3.2 Evaluation and assessment of transient MVPM simulation results To determine the impact of DSF on relevant cycles, the WLTC and RDE were simulated for the PC application, and the WHTC was simulated for the MD application. The results of those simulations are described here. Figure 5 shows transient results of C segment and compact SUV application over WLTC. It depicts the fire density, exhaust temperature upstream SDPF as well as the cumulated tailpipe (TP) NOX emission. The WLTC begins at an ambient temperature of 23 °C. A minimum coolant temperature limit of 60 °C is imposed to represent hardware constraints, and effectively eliminates DSF operation until 140 s. The exhaust temperature traces upstream SDPF have showed only slightly increase after cold start and warm-up phase, due to the thermal mass of the DOC. Afterwards, an exhaust temperature increase by around 20 K has

664

Extended cylinder deactivation strategies to improve CO₂ and pollutant emissions … been achieved under DSF operation at segment C vehicle. That increased exhaust temperature has improved the NOX conversion of SDPF and dropped the TP NOX emission down to 43 mg/km. It represents a reduction by 4.4 % compared to the 4-cylinder operation of segment C. Additionally, this improved results have been achieved with a benefit in CO2 emission by 1.5 %. Fire Density / 1

1.0 0.5 0.0 500

Temperature before SDPF / °C

400 300 200 4-cylinder Segment C 4-cylinder Compact SUV

100 0 1.5

Cumulated TP NO

X

DSF Segment C DSF Compact SUV

mass / g

1.0 0.5 0.0 150 100 50 0

Vehicle speed / km/h

0

200

400

600

800 1000 Time / s

1200

1400

1600

1800

Figure 5 – Simulation results showing firing density, SDPF inlet temperature and cumulated TP NOX emissions over WLTC cycle for C segment and compact SUV

The results of compact SUV have showed a lower NOX reduction potential by DSF operation. This heavier vehicle application has led to a higher engine operation with at an increase exhaust temperature level. Furthermore, the DSF operation has been reduced based on the higher load request. Thus, only a slightly exhaust temperature increase has entered the SDPF. Nevertheless, an improvement in CO2 emission by around 1 % has been obtained. Table 2 summarizes the simulation results of WLTC and RDE. The results under RDE have provided than additional improvements at the trade-off between NOX and CO2 emissions.

665

Extended cylinder deactivation strategies to improve CO₂ and pollutant emissions … Table 2 - Summary results of C segment and compact SUV application benefits of DSF in WLTC and RDE RDE

Compact SUV

Segment C

WLTC

Figure 6 shows the simulation results of the MD truck application under cold stared WHTC. It can be seen, that the activation of DSF has increased the exhaust temperature upstream SCR by 10-30 K in a wide range of the cycle. Thus, an improved NOX conversion has occurred and provided a tailpipe reduction by 15 % compared to base configuration. Also fuel consumption benefit has achieved of around 1.6 % due to the dynamic cylinder deactivation.

666

Extended cylinder deactivation strategies to improve CO₂ and pollutant emissions … Fire Density / 1

1.0 0.5 0.0 400

Temperature before SCR / °C

300 200 100

6-cylinder MD truck

0 30

DSF MD truck

Cumulated TP NO X mass / g

20 10 0 30 20 10 0

BMEP / bar

0

200

400

600

800 1000 Time / s

1200

1400

1600

1800

Figure 6 – Simulation results showing firing density, SCR inlet temperature and cumulated TP NOX emissions over cold stared WHTC for MD truck

Table 3 – Summary results of MD truck application benefits of DSF in weighted WHTC (cold and warm started)

MD Truck

WHTC

Table 3 shows the summary results of MD truck in weighted WHTC. The weighting factors consider a distribution of 14 % cold started WHTC and 86 % hot started WHTC. The investigations have showed a tailpipe BSNOX improvement of around 30 % in parallel to BSFC benefit of 1.6 %

667

Extended cylinder deactivation strategies to improve CO₂ and pollutant emissions …

4

Summary and conclusion

An analytical model was created in GT-Suite to simulate the potential benefits of Dynamic Skip Fire (DSF) technology for passenger car and medium duty truck application. The study of PC was conducted on segment C as well as compact SUV vehicle under WLTC and RDE, and the MD truck was analyzed under WHTC. The aim of this research was the comparison of conversion efficiency and thermal behavior of each aftertreatment components to the baseline results of current production vehicles. The simulation results predicted fuel economy benefits with CO2 reduction in the range of 1 % to 2 %, while realizing tailpipe NOX emissions reduction up to 11 % under RDE. The reduction of tailpipe NOX was achieved mainly by improved conversion efficiency in the SDPF system by better managing the exhaust temperatures with DSF technology.

5

Bibliography

1. Adolph, D., Busch, H., Pischinger, S., Kolbeck, A., et al., “Advanced Diesel Combustion – A Method Demonstrating Favorable Untreated Engine Emissions with Improved Consumption Characteristics”, MTZworldwide 69, 2008, No. 01, pp. 24-31. 2. Körfer, T., Schnorbus, T., Miccio, M., Schaub, J., “Emission-based EGR Strategies in Diesel Engines for RDE Requirements”, MTZworldwide 75, 2014, No. 9, pp. 10-17. 3. Deppenkemper, K., Özyalcin, C., Ehrly, M., Schoenen, M. et al., “1D Engine Simulation Approach for Optimizing Engine and Exhaust Aftertreatment Thermal Management for Passenger Car Diesel Engines by Means of Variable Valve Train (VVT) Applications”, SAE Technical Paper 2018-01-0163, 2018, doi:10.4271/2018-01-0163. 4. Blanco-Rodriguez, D., Vagnoni, G., Holderbaum, B., “EU6 C - Segment Diesel vehicle s, a challenging segment to meet RDE and WLTP requirements”, 8th IFAC Symposium Advances in Automotive Control, Sweden, 2016. 5. Blanco-Rodriguez, D., Vagnoni, G., Aktas, S., Schaub, J., “Model-based Tool for the Efficient Calibration of Modern Diesel Powertrains”, MTZworldwide 77, 2016, No. 10, pp. 54-58. 6. Chen, K., Wang, R., Younkins, M. Scassa, M., et. al., “Dynamic Skip Fire Applied to a Diesel Engine For Improved Fuel Consumption and Emissions”, 4th International Conference Diesel Powertrain 3.0, 2018, Coventry.

668

A new method for the objective assessment of ADAS based on multivariate time series classification Uwe Moser, Nick Harmening BMW Group Dieter Schramm University of Duisburg-Essen

A new method for the objective assessment of ADAS based on multivariate time …

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 M. Bargende et al. (Hrsg.), 19. Internationales Stuttgarter Symposium, Proceedings, https://doi.org/10.1007/978-3-658-25939-6_53

669

A new method for the objective assessment of ADAS based on multivariate time …

1 Introduction Advanced Driver Assistance Systems (ADAS) such as Adaptive Cruise Control (ACC) are appearing increasingly more often in modern premium cars and have great potential for increasing both comfort and safety. These gains are only appreciated by the customer, when the driving behavior of the system is perceived as positive, e.g., comfortable [1]. Therefore, the different calibrations of an ADAS must be assessed during the development process to ensure positive perceived comfort. However, the general practice is that the new calibrations of an ADAS are assessed subjectively by experts in test drives [2]. Although this system basically works, it has several disadvantages. The perception and consequently the assessment vary from person to person, are dependent on the drivers’ physical and mental conditions (e.g., tiredness and concentration) and also vary with the duration of the experiment (e.g., adaption) [3]. As a consequence, it is highly desirable to increase the objectivity in the assessment in order to assess driving comfort based on objectively measurable variables, which are determined in an experiment with a representative group of customers. To fulfill this vision, the subjective perception must be described by objectively measurable variables, which is nontrivial due to the complexity of the human perception process, the variety of driving scenarios and the complexity of the influences on the driver. [4] presents a method to determine the most influential variables in ADAS perceptions. Based on these variables, it is possible to build models that can predict the subjective perceptions of different ADAS calibrations in order to compare them in an objective way. The contribution of this paper is to present two Multivariate Dynamic Time Warpingbased classifiers for the objective assessment of ADAS, namely, a k-Nearest Neighbors classifier and a kernel density-based classifier. Both classifiers can handle the characteristics of time series, have low user involvement, deal with multivariate relations and lead to a transparent classification [4].

2 State of the art Objective assessment originates in driving dynamics and is defined as the description of the subjective perceptions using objectively measurable variables and mathematical rules [3]. In driving dynamics, the focus mostly is on lateral dynamics, such as the vehicle’s handling or the steering feel [5], assessing the current calibration or simulation. To predict the subjective perceptions based on objectively measurable variables, correlation (Bravais Pearson), (linear) regressions and multiple regressions are mainly used [3]. With the increasing distribution of ADAS in different car segments, the need for an objective assessment is increasing, but only a few models are currently available [6]. So far, multiple regressions have been the most frequently applied method in the objective assessment of ADAS. [7] uses linear regressions for the objective assessment of

670

A new method for the objective assessment of ADAS based on multivariate time … different ACC calibrations in order to determine the calibration with the highest comfort. [2] uses linear regressions to describe the relationship between the lateral movement and path curvature in order to generate an optimal comfort trajectory. [8] uses linear regressions for the objective assessment of different ACC scenarios in order to optimize the ACC function with a model-in-the-loop approach.

2.1 Problems and requirements During the objective assessment of driving dynamics, the focus is mainly on the interactions between the driver and the vehicle, and systems such as the lane-keeping assistant work in complex environments in which various variables influence the subjective perception and interact with other vehicles. This results in a massive number of possible situations (in comparison to standardized driving maneuvers in driving dynamics). As a consequence, the current procedures for the objective assessment of driving dynamics must be rethought for the assessment of ADASs. This is especially true in the context of highly automated driving (HAD) in which the complexity of scenarios is increasing even more. [1] uses a neural network for the objective assessment of a lane-keeping assistant, which is based on the lateral deviation, the average steering angle, the average lateral acceleration and curvature radius. Machine learning therefore seems to be a promising approach. However, the functioning of neural networks can be complex to understand and so they often appear as a black box. To fully understand the underlying process in the comfort perception of ADASs, a central requirement of the model is its interpretability. In summary, the problems with the state-of-the-art methods are highlighted in the following. Variance of the subjective perceptions (P1): The perception varies from person to person, is dependent on the drivers’ physical and mental conditions (e.g., tiredness and concentration) and varies with the duration of the experiment (e.g., adaption) [3]. High User Impact (P2): The impact of the subjective judgment of the expert that designs the characteristic values (or features) of variables is high. The process should be standardized in order to be applied quickly to different use-cases of objective assessments. Information Loss (P3): Since the time series must be reduced to a characteristic value in order to use a classical regression, the main information about the characteristics of the time series is lost. This is not negligible since the perceptions arise over time, and, therefore, the course of the time series has a high impact on the perception [9]. Problems occur with existing approaches and requirements can be formulated that are necessary for an effective process tor the objective assessment in ADASs. Multivariate relations (R1): The human perception of an ADAS relies on multiple variables. Therefore, the classifier must use multiple objectively measurable variables, too. Especially the interrelations between the variables must be considered. Transparency of the classification (R2): [10] states that the classification process must be transparent for the user in order to detect the possible failures or limits of the classifier. Expandability (R3): Objective assessment is an iterative process in which

671

A new method for the objective assessment of ADAS based on multivariate time … new conditions (e.g., new ADASs and new regulations) frequently occur. To adapt the existing classifier to the new conditions, it must be easily expandable (e.g., by adding new variables and changing weights). The discussed problems and requirements show that the objective assessment of ADASs must be rethought and a new method is necessary. In the following Chapter, P1-3 and R1-3 are used to propose a framework that has the potential to solve the problems and fulfill the requirements.

3 Methodology Since classical statistical methods are limited due to the problems described in Chapter 2, the research field of pattern recognition and machine learning seems to be promising. Therefore, the problem of objective assessments in ADASs can be formulated as a time 𝑿𝒒 , 𝑳 contains series classification problem according to [11]. The training set 𝑫𝒒 𝑿𝒊 |𝑖 a set of q multivariate time series (e.g. measurements in a driving test) 𝑿𝒒 1, … , 𝑞 with n variables, the length t and a set of class labels 𝑳 (e.g., comfort assessments) such that each time series 𝑿𝒊 carries a class label 𝐿 ∈ 𝐿. Therefore, in the time → 𝐿 learns from 𝑫𝒒 such that for any series classification problem a classifier 𝐶 ∶ ℝ time series 𝑿∗ of objectively measurable variables, 𝐶 predicts the class label (e.g., comfort) of 𝑿∗ using 𝐶 𝑿∗ . To solve this problem, a suitable classifier must be developed. According to P1, the main difference between most time series classification problems and the objective assessment problem is that there is no definite ground truth in subjective perceptions. Therefore, the subjective perception can be interpreted as noisy data.

3.1 Methods for time series classification [12] divides the methods for time series classifications into feature-based classification, model-based classification and sequence distance-based classification. In feature-based classification, the time series is transformed into a vector of features through feature selection. However, the feature selection process is nontrivial because the sequential nature of the time series must be represented [12]. As a result, it is highly dependent on the expert (P2) performing the feature selection and is therefore is not further considered in this work. Model-based approaches assume that sequences in a class are generated by an underlying model [12]. However, this model is based on assumptions that must be made, which also includes the parameter settings and is highly dependent on the expert (P2) designing the model. Therefore, this category of time series classification is discarded, too. Sequence distance-based classification methods rely on a distance function that measures the similarity between two sequences [12]. This similarity can be used for existing classifiers, such as the k-nearest neighbors classifier (k-NN) [11]. It is a parameter-free method that does not require feature selection and discretization using the raw data of the entire time series, therefore solving P2 and P3 [11].

672

A new method for the objective assessment of ADAS based on multivariate time …

3.2 Sequence distance-based classification: MDTW Sequence distance-based classification consists of three steps. In step 1, the distance between different time series is calculated using a distance function. This distance is used in step 2 to classify the new observations. Afterwards the prediction performance can be assessed in step 3. This process is displayed in Figure 1. In this process, the distance function determines the quality of the classification [12].

Figure 1: The process of sequence distancebased classification (compared to [12]).

Figure 2: The process of probability-based classification (e.g., KDE).

The Euclidean distance is a widely adopted distance function for simple time series classifications. However, it has two disadvantages. On the one hand, it requires two time series to have equal lengths and, on the other hand, it is sensitive to distortions in the time dimension [12]. Notably, these two aspects frequently occur in driving scenarios. Since there are many different ways to conduct a lane change, it is uncommon for two observations to have the same length. The same argument holds for distortions in the time dimensions, because the different phases of a lane change can be stretched or contracted when comparing two observations. Dynamic Time Warping (DTW) results in a distance function that avoids these two flaws because it can cope with time series with different lengths and equalize distortions in the time dimension. For a detailed description of DTW, the reader is referred to [13]. Although DTW originates from speech recognition, it has also already been successfully applied in the automotive domain. [14] uses DTW to cope with the different executing speeds of lane change maneuvers. In this paper, a k-NN approach based on DTW is used for lane change detection based on the steering angle. [15] uses Dynamic Time Warping for driving style recognition based on the lateral acceleration gained from the vehicle’s CAN bus. These two examples show that DTW can be used for comparing driving dynamic signals such as steering angle or lateral acceleration. Therefore, it seems to be a promising approach for the objective assessment of ADASs. However, the human perception of comfort is a highly complex phenomenon that relies on multiple objectively measurable variables [2].

673

A new method for the objective assessment of ADAS based on multivariate time … Consequently, a univariate DTW distance function can lead to insufficient results because it only processes the information of a single variable (R1). Therefore, a Multivariate Dynamic Time Warping (MDTW) approach is necessary for the objective assessment of ADASs. However, the extension of DTW to the multivariate case is not trivial. There are multiple approaches to implement it, each of which can lead to different distance calculations and, consequently, different predictions. For a demonstration of this problem on a database, the reader is referred to [16]. The independent MDTW calculates the DTW independently for every dimension. As a consequence, the interrelations between the variables get lost, and this approach is therefore not considered in this paper. The dependent MDTW calculates one warping over all variables and keeps the interrelations; therefore, it is used in this paper [15]. There are further approaches that cannot be classified as independent or dependent MDTW because these are integrated approaches, similar to in [17]. [4] shows a comparison of the promising MDTW approaches and concludes that the approach of [18] outperforms the other approaches. [18] treats a multivariate time series as a sequence of vectors. 𝛿 𝒙, 𝒚

∑

𝑐 |𝑥

𝑦|

(1)

Stating that in principle any p-norm (Equation 1) can be used to calculate the distance of two vectors, [18] uses the 1-norm, which is the sum of the absolute differences in all dimensions. [4] shows that for a lane change scenario the 2-norm leads to the best results, because the other norms distort the distance metric too much and lead to poor results. As a consequence, Equation 1 simplifies to the following: 𝛿 𝒙, 𝒚

𝑐 𝑥

𝑦

(2)

Figure 3 shows the functioning of the dependent MDTW with the 2-norm that is used as the distance function in this paper. It is easily extendable up to n variables, which meets requirement R3. Therefore, the difference between a univariate DTW approach and the dependent MDTW approach is that a multivariate distance metric 𝛿 𝒙, 𝒚 is used instead of a univariate distance metric. 𝑀𝐷𝑇𝑊 𝑿, 𝒀

𝐷𝑇𝑊 𝑿, 𝒀 , with 𝛿 𝒙, 𝒚

(3)

3.3 Classification with k-Nearest Neighbors With the MDTW transforming multivariate time series into a 1-dimensional distance space, this distance can be used for a k-NN classifier. The k-NN classifier takes the k points that are nearest to the test input 𝑿∗ in the training set into account, counts how many members of each class are in this set, and returns the empirical fraction as the estimate. Following [19], the k-NN classifier is formally defined as follows:

674

A new method for the objective assessment of ADAS based on multivariate time … 𝑝 𝐿∗

𝐿 |𝑿∗ , 𝑫𝒒 , 𝑘

∑

∈

𝑿∗ ,𝑫𝒒

𝕀 𝐿

𝐿

(4)

where 𝑁 𝑿∗ , 𝑫𝒒 are the indices of the k nearest points to 𝑿∗ in 𝑫𝒒 calculated by the MDTW distance function 𝑀𝐷𝑇𝑊 𝑿, 𝒀 (Equation 3), i is a certain class label (e.g., 5), and 𝕀 𝑒 is the indicator function. 𝕀 𝑒 is defined as follows: comfort 𝐿 𝕀 𝑒

1 if 𝑒 is true 0 if 𝑒 is false

(5)

Figure 3: The functioning of the dependent MDTW with the 2-norm.

A major advantage of the MDTW-based k-NN classifier is that the process of classification is transparent and easy to understand (R2). This is exemplified in Figure 4 where the black line represents the new observation 𝑿∗ and the other three lines are the closest observations from the training data set. This example clearly shows that DTW can compensate for distortions in the time dimension due to a phase shift of the black observation, compared to the other observations and the algorithm still recognizes the time series as similar. Based on these three neighbors, the classifier would predict comfort level 5 because it occurs most frequently. It is also transparent for the user when the algorithm does not find a similar observation using the apparent dissimilarity. Figure 5 shows the multivariate case with two variables. The weighting of the variables has a great impact on the classification and can be determined via clustering [4]. While considering multiple variables, the prediction of the classifier still remains interpretable. However, with an increasing amount of variables it is also increasingly difficult to find similar observations in the training data (the curse of high dimensionality [19]).

675

A new method for the objective assessment of ADAS based on multivariate time …

Figure 4: Exemplary data for a 3-NN classification based on univariate DTW.

Figure 5: Exemplary data for a 3-NN classification based on MDTW.

3.4 Classification with Kernel Density Estimation To calculate the confidence value of the prediction, the probabilities 𝑝 𝐿∗ 𝐿 |𝑿∗ can be estimated for every class 𝐿 in order to predict the class with the highest probability: 𝐶 𝑿∗

𝐿 |𝑿∗ , 𝑫𝒒

argmax 𝑝 𝐿∗

(6)

This process is displayed in Figure 2. Similar to the k-NN approach from Chapter 3.4, the distance is calculated with MDTW. Afterwards, the probability for every class 𝐿 is estimated with the KDE and the class with the highest probability is chosen to predict the comfort. Afterwards, the prediction performance is measured. The probability based on the KDE gives an estimation about the confidence of the classifier. To calculate 𝑝 𝐿∗ 𝐿 |𝑿∗ , 𝑫𝒒 , Bayes Theorem [20] can be used: 𝑝 𝐿∗

𝑿∗ 𝐿∗

𝐿 |𝑿∗ , 𝑫𝒒

𝐿 , 𝑫𝒒

|𝑫𝒒

where 𝑝 𝐿 , 𝑫𝒒 is given as the number of observations 𝑚 𝐿 divided by the total number of observations q. 𝑝 𝐿 |, 𝑫𝒒

(7)

𝑿∗ |𝑫𝒒

𝐿

𝐿 | 𝑞

𝐿

𝐿 | with Label

(8)

However, 𝑝 𝑿∗ |𝐿∗ 𝐿 , 𝑫𝒒 and 𝑝 𝑿∗ |𝑫𝒒 cannot be calculated directly because the probability distribution 𝑝 𝑿|𝑫𝒒 over 𝑿 is unknown. To solve this problem, the Kernel Density Estimation (KDE) can be used. It is a simple nonparametric density model, where no model fitting is required [19]. In contrast to the k-NN, the value of k does not have to be specified in the KDE. The KDE assumes that the observations are being drawn from some unknown probability density 𝑝 𝑥 in some D-dimensional space [20]. Using a Gaussian kernel [20], the KDE takes the following form:

676

A new method for the objective assessment of ADAS based on multivariate time … 𝑝 𝑥

1 𝑁

1

|𝒙

𝑒𝑥𝑝

2𝜋ℎ

𝒙 | 2ℎ

(9)

Using Equation 9 to estimate the probability distributions 𝑝 𝑿∗ |𝐿∗ 𝑝 𝑿∗ |𝑫𝒒 over q observations gives 1 𝑞

𝑝 𝑿∗ |𝑫𝒒

1

𝐿 , 𝑫𝒒 and

𝑀𝐷𝑇𝑊 𝑿∗ , 𝑿 2ℎ

𝑒𝑥𝑝

2𝜋ℎ

(10)

with MDTW according to Equation 3 and 𝛿 𝒙, 𝒚 according to Equation 2. In the same way, it can be determined that 𝐿 , 𝑫𝒒

𝑝 𝑿∗ |𝐿∗

1 𝑚

1

𝑒𝑥𝑝

2𝜋ℎ

𝑀𝐷𝑇𝑊 𝑿∗ , 𝑿 2ℎ

(11)

holds, where only the observations with label 𝐿 are regarded. Using Equations 10 and 11, it is possible to calculate 𝑝 𝐿∗ 𝐿 |𝑿∗ , 𝑫𝒒 (Equation 7 for every label 𝐿 ). Afterwards the classifier chooses the label with the highest probability (Equation 6).

4 Experimental Results In this Chapter, different variations of the proposed MDTW classifiers are tested against the state-of-the-art methods of multiple linear regressions and a neural network. Two experiments are used for this comparison, in which the comfort was assessed by the drivers on a scale of 1 to 7 (Table 1). In both experiments, the test vehicle was a BMW 7 series and the data were logged using a Vector CANape from the bus system of the car, using a Vector VX1132 for the Ethernet and a Vector VN7600 for the Flexray bus. A detailed description of the experiments follows in Chapters 4.1 and 4.2. Table 1: Comfort scale of the experiments 1

2

3

4

5

6

7

Almost crash

Strong irritation

Irritation

Strong dissatisfaction

Slight dissatisfaction

Satisfaction

Great satisfaction

To compare the performances of the different classifiers, a suitable metric must be selected. [12] proposes classification accuracy, which is widely used in the literature. However, with respect to subjective perception, there are always variations (compared to P1). Since there is no exact ground truth, it is very unlikely that a classifier predicts

677

A new method for the objective assessment of ADAS based on multivariate time … exactly the same class as the driver. Therefore, it is necessary to allow a small offset of +1 or -1 to detect if the classifier recognizes the tendencies of the human assessments. 𝑿𝒑 , 𝐿 with For the performance evaluation of the classifiers, a validation set 𝑫𝒑 𝑿𝒋 |𝑗 1, … , 𝑝 is used in which each time series 𝑿𝒋 carries a class label 𝐿 ∈ 𝐿. 𝑿𝒑 The adjusted prediction accuracy is the percentage of class labels generated by the classifier 𝐶 that match the class labels of the validation set 𝑫𝒑 and the labels that are +1 or -1 next to the real class labels: 𝑿𝒋

𝐶, 𝑫𝒑

𝐴𝑐𝑐

𝑿𝒋

𝒋

𝑿𝒋

𝒋

𝒋

(12)

In the objective assessment, the class labels generated by the classifier 𝐶 are the subjective assessments (e.g., comfort), which are predicted by the classifier based on the objectively measurable variables of an experiment, as in Chapter 3. Unfortunately, even the adjusted prediction accuracy has two major flaws. First, when dealing with highly imbalanced classes (a large proportion of high comfort levels and a small proportion of low comfort levels), a classifier that always predicts the majority class will have a high accuracy, even if there is no real information [21]. Second, small deviations from the real class are punished equally to high deviations. To tackle these shortcomings, the 𝐿 measure (loss function) with 𝐿

∑

𝐶 𝑿𝒋 |

|𝐿𝒋

(13)

calculates the average deviation of the correct class [22]. Therefore, a high deviation from the correct class is punished harder than a low deviation. To deal with the high imbalance of classes, macro averaging according to [23] is used. This results in the macro averaged 𝐿 measure with 𝐿

1 𝐿

1 𝑝

|𝐿

,

𝐶 𝑿

,

|

(14)

The following comparison comprises two datasets for every use case. The performance on the training set is used to assess the performance on the dataset of the original experiment. To assess the generalizability of the classifier to the variations of the use case, the performance on a test set must be evaluated as well. This is the data of a new experiment that was conducted after the original experiment in order to test how well the classifiers perform in changing environmental conditions. A high performing classifier has a high 𝐴𝑐𝑐 in combination with a low 𝐿 .

4.1 Lane keeping dataset The first experiments are conducted on a lane keeping dataset. In this study, the impacts of the frequency and amplitude of an oscillation of the lane keeping assistant on the

678

A new method for the objective assessment of ADAS based on multivariate time … comfort was examined. On a test site, different frequencies and amplitudes of the oscillation were presented to the subjects on a straight route. In this study, 394 observations were measured with 21 subjects. The observations had a length of approximately 120 seconds where the oscillation was stationary. However, in real traffic, only fragments of this oscillation occur. Therefore the observations were cut into smaller pieces (approximately 10 seconds), which resulted in a total amount of 2213 observations. Afterwards a small expert study was conducted to create a test set. The 3 experts drove and created 248 oscillation observations in real traffic. This aimed to demonstrate the transferability of the data from the test site to the real world. The lane keeping dataset contains only one important variable that describes the oscillation, namely, the lateral deviation (ld). Therefore, it is a univariate use case (univariate DTW). To find the optimal number of neighbors for the k-NN approach, k is varied from 1 to 5 neighbors. Most of the papers in objective assessment use multiple linear regressions (MLRs) [3]. Therefore, MLR and a simple neural network (NN) proposed in [1] are tested in this paper. Since the most important characteristics of an oscillation are the frequency (f) and amplitude (A), the max of these two is taken as the features for the MLR and the NN1. The accuracy (Equation 12) and the 𝐿 -measure (Equation 14) on the training and test sets are displayed in Table 2.2 Table 2: Accuracies of the lane keeping dataset Classifier DTW 1-NN DTW 2-NN DTW 3-NN DTW 4-NN DTW 5-NN DTW KDE MLR NN Random Cl.

Variables ld ld ld ld ld ld max f, max A max f, max A none

Training 𝑨𝒄𝒄𝒂𝒅 𝑳𝟏𝑴 84% 0.75 83% 0.84 85% 0.77 87% 0.74 87% 0.74 84% 0.75 69% 1.12 76% 0.91 46% 2.15

Test 𝑨𝒄𝒄𝒂𝒅 79% 78% 78% 79% 78% 72% 70% 61% 46%

𝑳𝟏𝑴 1.05 1.02 1.03 1.02 1.03 1.42 1.33 1.59 2.13

The 4-NN has the highest accuracy and the lowest 𝐿 on the training and the test sets. Therefore, it is clearly the approach with the highest performance on both datasets. The 1

2

The NN consists of an input layer with 2 neurons (features), one fully connected hidden layer with 50 neurons and one output layer with 7 neurons (comfort levels). The nearest neighbor has been ignored in the training dataset (leave one out cross validation [20]). Otherwise the classification accuracy of the 1-NN approach would be 100%.

679

A new method for the objective assessment of ADAS based on multivariate time … NN and the MLRs have lower accuracies and higher 𝐿 s than all the DTW-based classifiers on both datasets. In general, the difference between the training and the test dataset is rather small, because the accuracies decrease by approximately 6% and 𝐿 increases by approximately 0.30 with respect to the mean of all the approaches. This shows the high generalizability of the data from the test site to the real traffic. The DTW approaches can deal very well with the variance in the subjective assessments (P1) in this use-case. The reason for that is the large number of observations (2,213) in combination with an equal distribution of the comfort levels among these observations. This is because the experiment was conducted on a test track, where the different levels of the oscillation could be varied on purpose. As a result, there are enough observations for the different levels of oscillation and k can be set relatively high in order to compensate for the variations in the subjective assessments. The user impact (P2) in this use case is low because the oscillation can be described by the lateral deviation which can directly be inserted into the algorithm. This feature engineering is problematic for the NN and the MLR. The most prominent features of the lateral deviation are the frequency and oscillation. Since it is truly challenging to calculate these features for a nonperiodic signal, it leads to a poorer predictive performance. The performances of the DTW approaches show that they can indeed extract a higher information gain (P3) from the data than the current methods. As a summary, the lane keeping use case showed that DTWbased classification is easily designed, highly transparent and achieves superior predictive performance based on the high information gain by using the entire time series.

4.2 Lane change dataset The study on which the lane change dataset is based was conducted on a highway near Munich in real traffic where all lane changes were conducted manually. After the subjects conducted the lane change, the last 20 seconds were triggered by the study director who immediately asked the subjects to evaluate the comfort of the lane change following Table 1. In this study, there were 23 test objects that conducted 425 lane changes in total. Afterwards, a small expert study was conducted to create a test set. The 6 experts conducted 143 lane changes in total. The objective assessment of a lane change is much more complex than the objective assessment of an oscillation in Chapter 4.2. This is because the perceived comfort of a lane change is not dependent on only a single objective variable. Therefore, an MDTW-based k-NN classifier and an MDTW-based KDE classifier are promising methods to increase the prediction performance.3

3

This paper uses the variables (with their respective weights) longitudinal acceleration (𝑎 =0.3), lateral acceleration (𝑎 =0.1) and steering angle (stea=0.6), which were determined in [4].

680

A new method for the objective assessment of ADAS based on multivariate time … The features for the MLR and the NN4 that are used can be found in the literature, namely, max abs 𝑎 , max abs 𝑎 and max abs stea (steering angle) [1]. The accuracy (Equation 12) and 𝐿 (Equation 14) of the training and test sets are displayed in Table 3. Similar to Chapter 4.1, the nearest neighbor has been ignored in the training set. Table 3: Accuracies of the lane change dataset Classifier MDTW 1-NN MDTW 2-NN MDTW 3-NN MDTW 4-NN MDTW 5-NN DTW 2-NN DTW 2-NN DTW 2-NN MDTW KDE DTW KDE DTW KDE DTW KDE MLR NN MDTW 2-NN5 MDTW KDE5 Random Cl.

Variables 𝑎 , 𝑎 , stea 𝑎 , 𝑎 , stea 𝑎 , 𝑎 , stea 𝑎 , 𝑎 , stea 𝑎 , 𝑎 , stea 𝑎 𝑎 stea 𝑎 , 𝑎 , stea 𝑎 𝑎 stea max abs of 𝑎 ,𝑎 and stea max abs of 𝑎 ,𝑎 and stea 𝑎 , 𝑎 , stea 𝑎 , 𝑎 , stea none

Training 𝑨𝒄𝒄𝒂𝒅 𝑳𝟏𝑴 60% 2.0 60% 1.73 61% 1.82 67% 1.85 70% 1.88 66% 1.47 59% 2.19 51% 1.94 35% 1.65 76% 1.39 73% 1.82 70% 1.84 79% 1.29 80% 1.73 40% 2.3

Test 𝑨𝒄𝒄𝒂𝒅 57% 52% 54% 55% 58% 56% 51% 51% 35% 72% 68% 65% 82% 75% 76% 66% 38%

𝑳𝟏𝑴 2.22 1.95 1.98 2.20 2.20 2.11 1.67 1.91 2.30 2.12 1.49 2.14 1.46 1.59 1.17 1.44 2.4

The 2-NN classifier with 𝑎 has the lowest 𝐿 in the training set, while the 2-NN classifier with 𝑎 has the lowest 𝐿 in the test set. The 2-NN classifier with 𝑎 has the highest accuracy on both datasets, with respect to the univariate classifiers. In the comparison of the KDE-based classifiers, the KDE 𝑎 has the lowest 𝐿 on the test set. This shows the same effect as with the univariate k-NN classifiers. The longitudinal acceleration (𝑎 ) seems to be a better predictor in the training set, while the lateral acceleration (𝑎 ) seems to be a better predictor in the test set. The KDE 𝑎 has the best accuracy on the training and test sets. On the lane change dataset, the MLR is the best

4

5

The NN consists of an input layer with 3 neurons (features), one fully connected hidden layer with 50 neurons and one output layer with 7 neurons (comfort levels). Combination of the training and test dataset.

681

A new method for the objective assessment of ADAS based on multivariate time … performing approach, having the highest accuracy on the test set and the lowest 𝐿 in the training and test sets. The NN has the highest accuracy on the training set, and an 𝐿 equal to the MDTW 2-NN approach. On the test set 𝐿 is between the KDE 𝑎 and the k-NN approaches. This comparison shows that the multivariate approaches did not outperform the univariate approaches clearly. A reason for this is the lack of data regarding the different driving situations. The higher the dimensionality (number of variables), the bigger the space of possible situations. To find the 3 nearest neighbors that are as similar as possible to the new observation, the neighbors must be similar in every time series of every dimension. Here, the multivariate KDE classifier performs truly badly (even worse than random). A reason for that is the sparse data on which the probabilistic models are built. In the training set, there are only 2 observations that have been rated at a comfort level of 1. Therefore, the model has truly weak confidence and overestimates the low comfort levels. A lane change is a truly complex phenomenon with many possible variants. The results on the lane change dataset showed that especially for sparse data, a robust approach such as the MLR leads to the best predictions. However, it comes with a serious flaw. The model that was built in the MLR did not predict the comfort values 1, 2 and 7 because the plane is too flat (a limitation of linearity). Especially in these border cases, the MDTW approaches perform well. However, they lead to more heterogeneous predictions that are not as robust as the MLR prediction. In conclusion, the MLR is a robust approach for the detection of a first tendency and the MDTW approach seems to be more promising for predicting extreme situations. When the MDTW 2-NN database is extended with the test set (Table 3), the approach outperforms the MLR with respect to the 𝐿 of the test set. This shows the high improvement of the classifier with an increasing database. The same improvement effect can also be seen in the case of the MDTW KDE. Therefore, it is assumed that the performance of the MDTW classifiers will overtake the MLR with the increase of the data.

5 Conclusion and Outlook The experimental results in Chapter 4 showed the potential of a MDTW-based classification approach for the objective assessment of ADASs. When assessing the lane keeping assistant, the proposed methods clearly outperform well established methods, and they are easily designed and highly transparent. However, the lane change dataset showed that there is still room for improvement. Especially for sparse data, the MDTWbased approaches lack the robustness of the MLR. Since no hypotheses must be formulated before the experiment, the process of data generation is easier with the MDTW approaches. With regards to this increasing data generation, the MDTW clearly becomes a superior method for the objective assessment of ADASs. The MDTW approaches and the MLR, both have their strengths and weaknesses. While the MLR is the more robust approach, the MDTW approaches are able to recognize special or more

682

A new method for the objective assessment of ADAS based on multivariate time … extreme cases. As a consequence, it is recommended to use both approaches and even a combination would be promising. Since the test set is only based on a small validation study that is not representative enough, the main task for future work is to validate the methodology on a larger dataset with more drivers.

Bibliography [1]

P. Schöggl, M. Schrauf, J. Holzinger, and E. Bogner, “Objective assessment of comfort and safety of advanced driver assistance systems,” ATZ Worldwide, 2014.

[2]

M. Wegscheider, “Modellbasierte Komfortbewertung von Fahrerassistenzsystemen,” Dissertation, 2009.

[3]

M. Decker, Zur Beurteilung der Querdynamik von Personenkraftwagen, 2009.

[4]

U. Moser and D. Schramm, “The Use of Multivariate Dynamic Time Warping to Determine the Most Influential Variables in ADAS Comfort Perception,” 14th International Symposium on Advanced Vehicle Control; AVEC '18, 2018.

[5]

C. Fernandes and F. J. Okano, “Vehicle Dynamics Objective Metrics,” in SAE Brasil 2003 Congress and Exhibit, 2003.

[6]

M. Elbanhawi, M. Simic, and R. Jazar, “In the Passenger Seat: Investigating Ride Comfort Measures in Autonomous Cars,” IEEE Intell. Transport. Syst. Mag., vol. 7, no. 3, pp. 4–17, 2015.

[7]

T. Pawellek, L. Liesner, R. Henze, and F. Kücükay, “Objektivierungsverfahren für eine kundenoptimale ACC-Systemauslegung,” ATZ-Automobiltechnische Zeitschrift, vol. 118, no. 4, pp. 74–79, 2016.

[8]

F. Krauns, A. Sonka, R. Henze, and F. Kücükay, “Objektivierung kombinierter Längs- und Querführung,” in AAET - Automatisiertes und Vernetztes Fahren: Beiträge zum gleichnamigen 18. Braunschweiger Symposium vom 8. und 9. Februar 2017, Stadthalle, Braunschweig, Braunschweig: ITS automotive nord e.V, 2017.

[9]

H.-P. Krüger and A. Neukum, “Bewertung von Handlingeigenschaften – zur methodischen und inhaltlichen Kritik des korrelativen Forschungsansatzes,” pp. 245–262.

[10] M. Ribeiro, S. Singh, and C. Guestrin, ““Why Should I Trust You?”: Explaining the Predictions of Any Classifier,” Proceedings of the 2016 Conference of the North American Chapter of the Association for Computational Linguistics: Demonstrations, pp. 97–101, 2016.

683

A new method for the objective assessment of ADAS based on multivariate time … [11] Z. Xing, J. Pei, and P. S. Yu, “Early prediction on time series: A nearest neighbor approach,” IJCAI International Joint Conference on Artificial Intelligence, pp. 1297–1302, 2009. [12] Z. Xing, J. Pei, and E. Keogh, “A brief survey on sequence classification,” SIGKDD Explor. Newsl., vol. 12, no. 1, p. 40, 2010. [13] E. Keogh and C. A. Ratanamahatana, “Exact indexing of dynamic time warping,” Knowl Inf. Syst., vol. 7, no. 3, pp. 358–386, 2005. [14] Y. Zheng and J. H. L. Hansen, “Lane-Change Detection From Steering Signal Using Spectral Segmentation and Learning-Based Classification,” IEEE Trans. Intell. Veh., vol. 2, no. 1, pp. 14–24, 2017. [15] D. A. Johnson and M. M. Trivedi, “Driving style recognition using a smartphone as a sensor platform,” IEEE ITSC, vol. -, pp. 1609–1615, 2015. [16] M. Shokoohi-Yekta, J. Wang, and E. Keogh, “On the Non-Trivial Generalization of Dynamic Time Warping to the Multi-Dimensional Case,” in SDM, pp. 289–297. [17] Z. Bankó and J. Abonyi, “Correlation based dynamic time warping of multivariate time series,” Expert Syst. Appl., vol. 39, no. 17, pp. 12814–12823, 2012. [18] G. A. ten Holt, M.T.J. Reinders, and E. A. Hendriks, “Multi-dimensional dynamic time warping for gesture recognition,” ASCI, vol. 300, 2007. [19] K. P. Murphy, Machine learning: A probabilistic perspective. Massachusetts: The MIT Press, 2012. [20] C. M. Bishop, Pattern recognition and machine learning. New York NY: Springer, 2006. [21] Y. Tang, Y.-Q. Zhang, N. V. Chawla, and S. Krasser, “SVMs modeling for highly imbalanced classification,” (eng), IEEE transactions on systems, man, and cybernetics. Part B, Cybernetics : a publication of the IEEE Systems, Man, and Cybernetics Society, vol. 39, no. 1, pp. 281–288, 2009. [22] K. Janocha and W. M. Czarnecki, “On Loss Functions for Deep Neural Networks in Classification,” Feb. 2017. [Online] Available: http://arxiv.org/pdf/1702.05659v1. [23] A. Sun and E.-P. Lim, “Hierarchical Text Classification and Evaluation,” Proceedings of the 2001 IEEE International Conference on Data Mining, pp. 521–528, 2001.

684

Test methods of visual distraction by interaction with driver information- and assistance systems Thomas Stottan AUDIO MOBIL

Test methods of visual distraction by interaction with driver information- and …

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 M. Bargende et al. (Hrsg.), 19. Internationales Stuttgarter Symposium, Proceedings, https://doi.org/10.1007/978-3-658-25939-6_54

685

Test methods of visual distraction by interaction with driver information- and …

Abstract Basierend auf vorhandenen Richtlinien zur Testung des Ablenkungspotentials von Fahrerinformations- und assistenzsystemen (NHTSA, ISO/TS 15007-2) wurden mit verschiedenen Fahrzeugtypen insgesamt vier Tests auf einer Rundstrecke durchgeführt. Die visuelle Ablenkung wurde mittels eye-tracking-System bei der Durchführung vorgegebener Tasks im Fahrversuch mit Probanden ermittelt. Fragebögen zur User-Experience und Expertenevaluierungen ergänzten die Beurteilung der Serienfahrzeuge. Eine statistische Auswertung der Ergebnisse erlaubt eine reproduzierbare Einschätzung des Grades der Ablenkung in einer realen Umgebung. Diese Vorgehensweise hat sich als tauglicher Kompromiss zwischen einer Simulator-Studie und Testfahrten im öffentlichen Verkehr erwiesen. Die Durchführung dieser Tests wird beschrieben, Ergebnisse zum Ablenkungsgrad vorgestellt und die Überleitung dieses Prozederes in eine Ö-Norm, die der Feststellung der durch Fahrerinformations- und assistenzsysteme verursachten visuellen Ablenkung dienen soll, aufgezeigt.

Einleitung Die individuelle Mobilität wird sich in den kommenden Jahren vom Automobil hin zur intelligenten Vernetzung unterschiedlicher Verkehrsträger verändern. Das Auto entwickelt sich zu einem, in späterer Folge immer autarkeren Teilnehmer am Ökosystem „Digitale Mobilität“. In Betrachtung gezogen werden muss daher, welche unterschiedlichen Anwendungen es bei der Bereitstellung von Funktionen und Daten gibt. In weiterer Folge werden – abhängig von der ersten Fragestellung – die unterschiedlichen technischen Zugangsmöglichkeiten bewertet. Generell ist bei der Bereitstellung von Informationen und Funktionen die Relevanz zur eigentlichen Fahraufgabe zu evaluieren. Diese Relevanz wird sich in den kommenden Jahren aufgrund des Übergangs-Szenarios vom Selbstfahren zum hochautomatisierten Fahren maßgeblich ändern. Demnach gilt es zu berücksichtigen, welche Funktionen zu welchem Zeitpunkt und in welchem Kontext zur Verfügung gestellt werden können und dürfen. Schon heute ist die Hauptunfallursache nicht die überhöhte Geschwindigkeit sondern Unachtsamkeit und Ablenkung (Abb.1).

686

Test methods of visual distraction by interaction with driver information- and …

Abb. 1: KfV Statistik Austria 2017 – Download 20.05.2017 [13]

Es ist immer auf minimale sensorische und kognitive Ablenkung zu achten! Dabei kann es notwendig sein, die Funktionen und die Informationen „automotiv-tauglich“ bereitzustellen. Damit ist gemeint, dass je nach Informationsgehalt und Relevanz zur Fahraufgabe eine dynamisch, situative Bereit-stellung gemacht werden muss. Neben reinem Unterhaltungs-Content darf die Gruppe der mobilitätsbeeinflussenden Daten nicht außer Acht gelassen werden. Mobilitätsbeeinflussende Daten sind z.B.: Echtzeitinformationen über Verkehrslagen oder aufbereitete Informationen aus einem digitalen Baustellen-Management. Zudem kann eine Event-abhängige Verkehrsinformation den Fahrkomfort maßgeblich beeinflussen. Immer größere Bedeutung gewinnt dabei die Überwachung von Assistenzsystemen und gegebenenfalls temporäre Übergabe der Fahraufgabe samt der damit verbundenen Übergabe- und Übernahmemodalitäten. Daneben entstehen neue Anforderungen aus dem Bereich der alternativen Antriebe wie z.B.: Stromtankstellen oder drahtlose Power-Stations zur alternativen Ladung (Resonanz / Induktion) von Fahrzeugen. Der Unterhaltungs-Content kann wiederum nur situativ genutzt werden – entweder von Mitfahrern und Passagieren, beziehungsweise abhängig vom Automatisierungsgrad der momentanen Fahrsituation, dem Fahrer zur Verfügung gestellt werden. Aufgrund fehlender einfacher, einheitlicher und praxistauglicher Prüfmethoden wurden im Rahmen der Forschungseinrichtung Christian Doppler Labor für Contextual Interfaces an der Universität Salzburg gemeinsam mit dem Industriepartner AUDIO MOBIL in mehrjähriger empirischer Forschungsarbeit Grundlagen hierfür geschaffen.

687

Test methods of visual distraction by interaction with driver information- and … HMI-Tests (steht für „Human-Machine-Interaction“) zur Feststellung der visuellen Ablenkung als Basis für die Ö-Norm V 5090 stellen die erste Stufe der Fahrzeugvalidierung in Bezug auf die Bediensicherheit dar. Ablenkung – und hier vor allem die visuelle Ablenkung – von der Fahraufgabe ist mittlerweile wie bereits erwähnt zur Unfallursache Nummer 1 geworden. Die stete Zunahme von Infotainment- und Fahrerassistenzsystemen in den Fahrzeugen verleitet die Lenker zu übermäßiger Beschäftigung damit auch während der Fahrt. Das Ablenkungspotential von Telefonie im Auto wird schon seit den 80er Jahren wissenschaftlich untersucht, viele weitere Studien sind seither zu diesem Thema verfasst worden. Auch die NHTSA (National Highway Traffic Safety Administration) hat sich der starken Zunahme von durch Ablenkung verursachten Verkehrsunfällen angenommen und Richtlinien zur Evaluierung des Grades der Ablenkung erstellt. Die Tests dazu finden allerding im Simulator und mit einer sehr vereinfachten Fahraufgabe statt. Audio Mobil hat in den vergangenen Jahren, ausgehend von den Empfehlungen der NHTSA[1] und von ISO/TS 15007-2 [2], einige Testreihen mit unterschiedlichen Fahrzeugtypen, standardisierten Tasks und eye-tracking als primärer Testmethode auf einem Rundkurs durchgeführt. Weiter wurden in wissenschaftlichen Untersuchungen diverse Bereiche des Auto-Cockpits und der Interaktion der Insassen mit den Infotainment-Systemen beleuchtet. Die Ausgangssituation im Fahrzeug wurde als eine von „Drei Welten“, nämlich jener des Fahrers, des Beifahrers und der Passagiere auf der Rückbank, definiert [3]. Unterschiedliche Aspekte, wie die Untersuchung der Kollaboration von Fahrer und Beifahrer bei Navigationsaufgaben mittels ethnographischer Studien [4], oder die methodische Vorgehensweise für Tests von Interfaces im Auto [5], [6] wurden berücksichtigt. Ein anderer Schwerpunkt war die Fragestellung, wie weit das Lenkrad als zentrale Bedieneinheit, d.h. als Ersatz für alle anderen Bedienelemente im Cockpit inklusive der Displays, in Frage kommen könnte [7], [8], [9] – was sogar mit einem funktionsfähigen Prototypen getestet wurde. Um den Entwicklern von Mensch-Maschine-Schnittstellen ein hilfreiches Werkzeug an die Hand zu geben, wurden in Zusammenarbeit von Forschern und Praktikern in einem mehrstufigen Prozess sogenannte „Patterns“ generiert, die praktikable Lösungen für ablenkungs-relevante Design-Probleme im Automobilbereich vorstellen [10]. Eine andere Möglichkeit, die Qualität von Interfacedesigns zu überprüfen, stellt die von Stevens vorgestellt Checkliste für IVIS (In-Vehicle Information System) im Hinblick auf gutes, ergonomisches Design (was letztendlich auch wieder der Sicherheit dient) dar [11]. Die ÖNORM V 5090 wurde auf Grundlage der ISO/TS 15007‐2:2014 ausgearbeitet und konkretisiert die in dieser Technischen Spezifikation beschriebenen Methoden. Diese ÖNORM bezieht sich auf die Interaktion eines Fahrers mit einer Funktion des

688

Test methods of visual distraction by interaction with driver information- and … TICS (Transport Information and Control System) während der Fahrt auf einer definierten Prüfstrecke. Durch die voranschreitende Integration von neuen Funktionalitäten und Interaktionsmöglichkeiten mit Bedienelementen im Fahrzeug besteht zunehmend die Gefährdung, dass Fahrer abgelenkt werden. Dies ist darauf zurückzuführen, dass die allgemeine Leistungsfähigkeit des Menschen (kognitiv, perzeptuell und motorisch) begrenzt ist und eine immer komplexere Interaktion mit dem Fahrzeug die Sicherheit im Straßenverkehr gefährdet. Daher wurde ein einheitliches Prüfverfahren zur Feststellung Messung der visuellen Ablenkung eines Fahrers von der Fahraufgabe durch die Interaktion mit Fahrerinformations‐ und ‐assistenzsystemen (TICS) während der Fahrt ausgearbeitet. Das Prüfverfahren kann somit auch für Fahraufgaben mit einem assistierten und hochautomatisierten Fahrzeug angewendet werden.

Grundlagen Bei den Richtlinien der NHTSA zur Messung der visuellen Ablenkung von Fahrern während der Fahrt wird ein Simulator-Test mit 24 Probanden empfohlen, wobei die Fahraufgabe in der Nachfolge eines vorausfahrenden Autos mit gleichbleibender Geschwindigkeit besteht. Der Benchmark-Task besteht in der manuellen Einstellung eines Radiosenders. Die Blicke weg von der Straße (gemessen mittels eye-tracking) sollen prinzipiell 2 Sekunden nicht überschreiten und bei zumindest 21 von 24 Probanden sollen nicht mehr als 15 Prozent der Blicke länger als 2 Sekunden sein. All diese Vorgaben stellen ein ziemlich vereinfachtes und nicht sehr realistisches Szenario für die Testung der visuellen Ablenkung dar. Für die Messung der Blickabweichungen existieren mit den ISO-Standards 15007-1 und 15007-2 bereits Regularien. Bei der Okklusionsmethode, die einen noch theoretischen Ansatz darstellt, können keine Blicke die länger als 1,5 sek. dauern. Es ist also nicht möglich festzustellen, ob ein Fahrer – auch wenn er einen Task in den hier geforderten 1,5 (2 Sekunden bei der eye-tracking Methode) beenden kann – nicht doch länger von der Straße wegblickt. ISO 16673:2007 spezifiziert die Blickintervalle bei der Okklusions-Methode auf 1,5 sek. Verschluss offen gefolgt von 1,5 sek. Verschluss zu. Bei NHTSA sind 1,5 sek. offen und 1 sek. zu üblich. Die beschriebenen Testmethoden sind naturgemäß nicht sehr realitätsnahe. Ein Test von Bedienelementen und Eingabemodalitäten in Serienfahrzeugen im öffentlichen Verkehr mit relativ hoher Geschwindigkeit hingegen würde ein bedeutendes Sicherheitsrisiko – Gefahr durch zu große Ablenkung sowohl für die Fahrer als auch für die anderen Verkehrsteilnehmer - darstellen.

689

Test methods of visual distraction by interaction with driver information- and … Daher bietet sich als Kompromiss ein Test auf einer abgeschlossenen Teststrecke bei niedriger Geschwindigkeit mit Serienfahrzeugen oder auch Prototypen an. Die einfachste Variante einer Teststrecke ist ein Rundkurs. Dieser hat einen geringen Platzbedarf und die Fahrt im Kreis stellt eine doch nicht ganz anspruchslose Fahraufgabe dar. Ein eye-tracking System zur Messung der visuellen Ablenkung kann in jedes Fahrzeug mit relativ geringem Aufwand eingebaut werden. Derartige Tests – insgesamt vier – wurden mit verschiedensten Fahrzeugen von 2011 bis 2016 auf der hauseigenen Teststrecke von AUDIO MOBIL mit Unterstützung von Auto-Zeitschriften und der Universität Salzburg durchgeführt. Die daraus gewonnenen Erkenntnisse führten zum Vorschlag für eine Ö-Norm zur Messung der visuellen Ablenkung von TICS [12], welche nach passieren der Evaluationsschritte und Gremien als Ö-Norm 5090 im Frühjahr 2018 veröffentlicht wurde.

Abb. 2: Test-Setup ÖNORM V5090; AUDIO MOBIL 2018

690

Test methods of visual distraction by interaction with driver information- and …

Abb. 3: Test-Setup-Vergleich NHTSA vs. ÖNORM V5090; AUDIO MOBIL 2019

Prozedere Das prinzipielle Prozedere bei den HMI Tests ist im Folgenden kurz angeführt:

1.1 Nutzerstudie Für die Durchführung der HMI-Tests hat sich die Kombination aus Probandenversuchen und Expertenevaluierungen als sinnvoll erwiesen. 36 Testteilnehmerinnen und – teilnehmer, die durch Inserate rekrutiert und geschlechterparitätisch in drei Altersgruppen (20-35, 36-50, 51-65 Jahre) aufgeteilt wurden, fuhren - je 12 mit einem Fahrzeug – mit den drei Autos auf dem Rundkurs der Firma AUDIO MOBIL. Folgende Frage stand im Mittelpunkt der Untersuchung: Mit welchem System bzw. mit welcher Bedienart lenkt die Interaktion weniger von der Fahraufgabe ab und ist daher sicherer? Die Beantwortung dieser Frage erfolgt unter Berücksichtigung der folgenden Parameter: ● Aufgabendauer: Messung der Zeit zwischen Beginn und Beendigung der Aufgabe. Grundsätzlich ist anzustreben, dass Aufgaben im Fahrzeug in so kurzer Zeit wie möglich durchführbar sind. Nichtsdestotrotz spricht eine längere Aufgabendauer nicht zwingend für eine höhere Gesamtablenkung und kann daher nur in Zusammenhang mit weiteren Messungen gesehen werden, wie z.B. Eye-Tracking. ● Eye-Tracking: Messung der Blickrichtung um zu erfassen, wann ein Benutzer das interaktive System fokussiert, und wann die Straße. Details zu den Parametern sind unten angeführt.

691

Test methods of visual distraction by interaction with driver information- and … ● Lenkverhalten: Messung des Lenkwinkels um zu erfassen, wie sehr sich die Interaktion mit dem System auf die Lenkbewegung auswirkt. ● Wahrgenommene Anstrengung: Auch die Wahrnehmung der Benutzer spielt eine wichtige Rolle. Die wahrgenommene Belastung durch Aufgaben kann sich von der tatsächlichen Belastung unterscheiden. ● Usability (Gebrauchstauglichkeit) des Systems: Wie gut bedienbar schätzen die Benutzer das jeweilige System ein? Welche positiven/negativen Aspekte bei den Systemen fallen den Benutzern auf? Diese Frage wurde gestellt um die Wahrnehmung der Systeme auf Benutzerseite zu erheben. Zur Beantwortung dieser Frage wurden die Benutzer gebeten, die drei besten und die drei schlechtesten Aspekte der Systeme zu nennen. Die Antworten helfen so dabei, Gründe für Probleme bei der Aufgabendurchführung zu erheben. Daraus ergaben sich die folgenden Frage-Stellungen für die HMI Studie: Mit welchem System bzw. mit welcher Bedienart lenkt die Interaktion weniger von der Fahraufgabe ab und ist daher sicherer? Die Beantwortung dieser Frage erfolgt unter Berücksichtigung der folgenden Parameter: ● Aufgabendauer: Messung der Zeit zwischen Beginn und Beendigung der Aufgabe. Grundsätzlich ist anzustreben, dass Aufgaben im Fahrzeug in so kurzer Zeit wie möglich durchführbar sind. Nichtsdestotrotz spricht eine längere Aufgabendauer nicht zwingend für eine höhere Gesamt-ablenkung und kann daher nur in Zusammenhang mit weiteren Messungen gesehen werden, wie z.B. Eyetracking. ● Eyetracking: Messung der Blickrichtung um zu erfassen, wann ein Benutzer das interaktive System fokussiert, und wann die Straße. Details zu den Parametern sind unten angeführt. ● Lenkverhalten: Messung des Lenkwinkels um zu erfassen, wie sehr sich die Interaktion mit dem System auf die Lenkbewegung auswirkt. ● Anstrengung: Messung der Herz-Rate um die kognitive Anstrengung zu erfassen ● Wahrgenommene Anstrengung: Auch die Wahrnehmung der Benutzer spielt eine wichtige Rolle. Die wahrgenommene Belastung durch Aufgaben kann sich von der tatsächlichen Belastung unterscheiden. Welche positiven/negativen Aspekte bei den Systemen fallen den Benutzern auf? Diese Frage wurde gestellt um die Wahrnehmung der Systeme auf Benutzerseite zu erheben. Zur Beantwortung dieser Frage wurden die Benutzer gebeten die drei besten und die drei schlechtesten Aspekte der Systeme zu nennen. Die Antworten helfen so dabei, Gründe für Probleme bei der Aufgabendurchführung zu erheben.

692

Test methods of visual distraction by interaction with driver information- and … Welche negativen Aspekte sind bei den Systemen aus HCI Sicht auszumachen? Zur Beantwortung dieser Frage analysierten drei HCI Experten die Systeme. Ziel war es die größten Schwachstellen der Systeme aus Expertensicht zu identifizieren.

1.2 Aufgaben und Interaktionsmodalität Folgende Aufgaben wurden von jeder Person (im Stand zur Übung bzw. während der Fahrt) durchgeführt: Es wurde ein Setting „between subject design“ benutzt; das bedeutet, dass jede Person die Aufgaben nur mit einem Fahrzeug durchführt. Im Gegensatz dazu gäbe es auch noch das Setting „within subject design“, bei dem jede Person die Aufgaben mit jedem Fahrzeug durchführt. Aufgrund der verwendeten Mess- Hardware ist das „between design“-Setting für diese Studie praktikabler. Telefon: Eingetragenen Telefonbuch-Kontakt aus Liste wählen und anrufen ● manuell ● mittels Spracheingabe ● Navigation: Navigationsziel eingeben und Navigation starten ● manuell ● mittels Spracheingabe ● Internetaufgabe 1: Wetter ● Wetterinformationen für den aktuellen Ort abrufen und die angezeigte Temperatur dem Versuchsleiter mitteilen Internetaufgabe 2: News Die erste Nachricht eines bestimmten Nachrichtenkanals abrufen und sich vorlesen lassen bzw. (abhängig von den Möglichkeiten des Systems) die erste Nachricht eines bestimmten Nachrichten-kanals abrufen und per eMail verschicken. Alle Internetaufgaben wurden ausschließlich mit manueller Bedienung durchgeführt. Im Stand: Die erste Durchführung der Aufgaben wurde nicht in die Auswertung aufgenommen und diente nur zur Eingewöhnung in das System und dem Vermeiden von Effekten durch eine erstmalige Benutzung. Es gab hier keine maximale Zeitvorgabe und die Personen konnten so lange üben, bis sie sich mit der jeweiligen Aufgabe sicher fühlten. Während der Fahrt: Hier wurden dieselben Aufgaben durchgeführt, jedoch mit unterschiedlichem Inhalt, d.h. es wurde eine andere Adresse eingegeben, ein anderer

693

Test methods of visual distraction by interaction with driver information- and … Name angerufen, bzw. ein anderer Nachrichtenkanal gewählt. Hier wurde jeweils eine maximale Aufgabenzeit definiert. Danach wurde Aufgabe als „nicht erfolgreich“ gewertet und abgebrochen.

1.3 Expertenevaluierungen Welche positiven und negativen Aspekte sind bei den Systemen aus HCI Sicht auszumachen? Zur Beantwortung dieser Frage analysierten HCI Experten die Systeme. Ziel war es, die größten Schwachstellen und Pluspunkte der Systeme aus Expertensicht zu identifizieren.

Abb. 4: HMI Test 3, Testfahrzeuge und Teststrecke

Abb. 5: HMI Test 4, eye-tracking System eingebaut

Abb. 6: HMI Test 4, Expertenevaluierung

694

Test methods of visual distraction by interaction with driver information- and …

Ergebnisse Exemplarisch seien hier einige Ergebnisse – die bei allen vier HMI-Tests ähnlich gelagert waren – an Hand von HMI Test 3 angeführt: Einer der wichtigsten Aspekte bei der visuellen Ablenkung ist die Verweildauer des Blickes weg von der Fahrbahn. Diese ist von einigen Faktoren, wie der Position des Displays, der Vertrautheit mit dem System und vor allem der Komplexität der ausgeführten Tätigkeit, wie untenstehendes Beispiel zeigt, abhängig. Für die Fahrzeuge des HMI-Test 3, zeigt sich folgendes Bild: Betrachtet man über alle Aufgaben gerechnet die Gesamtdauer, in der der Blick während der Aufgabendurchführung nicht auf der Straße war, so ist die visuelle Ablenkung beim BMW am geringsten, gefolgt von Audi und Mercedes. Diese Unterschiede sind jedoch nicht statistisch signifikant (Abbildung 7).

Abb. 7: HMI Test 3, mittlere Verweildauer weg von der Fahrbahn

Generell zeigt sich, dass die visuelle Ablenkung bei der Telefonier-Aufgabe deutlich geringer ist, als bei der Navigationsaufgabe (p)59=


The electrical energy used by the EM in this case is equivalent to an amount of fuel consumed by the engine in order to restore the energy level of the battery. Therefore, the total amount of fuel spent in this event corresponds to the fuel used by the ICE when providing power to the wheel and the fuel equivalent “electric” amount indirectly consumed by the e-motor.

E% ' D E%  C E%  $"

(7)

The BSFC in hybrid boost is then dependent on the total fuel flow from the equation above and the driver power demand:   D

!E7?6:).B ,

D

!E7?6:)/+- !E7?6:)6:64>400 Nm), an improvement in SCR efficiency can be achieved due to the higher temperature level. It could be shown that EIVC and the higher temperature level resulting from the reduction in mass flow can contribute to an advantage in nitrogen oxide conversion. Particularly in the lower or partial load range, it leads to an efficiency improvement of up to 66 percent-points. Due to the disadvantages in the warm up behaviour, this is only interesting for warm engines. Therefore, the very early Miller valve timing can be a proven method to reach or maintain the SCR conversion at low load cycles. This can also be seen in Figure 5, where the results shown above are summarized based on linear extrapolation over the entire map range. 2400

ηSCR EIVC70

ηSCR RefIVC

MEngine / Nm

100 90

1800

80 60

1200

40 20

600

0 0 600

1000 1400 nEngine / rpm

1800

2200 600

1000 1400 nEngine / rpm

1800

2200

Figure 5: Improvement potential in the SCR system by miller valve timing based on linear extrapolation

3.3 Alternative Combustion Processes In this paper the influence of inlet valve closing timing on a combination of the HCLI and HPLI process is presented. Whereas in the HCLI process the injection takes place before the TDC and the start of combustion is delayed by high EGR rates after the TDC,

549

Potential of air path variabilities for heavy duty Diesel engines in the HPLI process the fuel is injected after top dead center. Therefore the combustion takes place during expansion. The HCLI process thus produces high pressure gradients, which significantly limits the load range. The HPLI process, on the other hand, achieves higher load ranges. However, fuel consumption increases significantly due to the late MFB50. A target conflict between load expansion and acceptable fuel consumption occurs. [7] Both early and late inlet valve closing are used within the investigations. The tests are carried out in the main speed range at 1275 rpm and are started at a load of 6 bar mean indicated pressure and then extended. In order to evaluate and compare the mixture formation resulting from the boundary conditions, a homogenization time is defined analogous to [14]. The detailed results are shown in [7]. In principle, it could be observed during the tests that an increase of the ignition delay and therefore of the premix proportion is a result of later MFB50 and later or earlier inlet valve closing. This results in a longer homogenization time and the soot emissions decrease. Thus, the soot NOX trade-off could be significantly improved compared to the conventional process. Soot emissions of less than 0.003 g/kWhi can be achieved with NOX emissions of 1 g/kWhi. The resulting pressure gradients of approximately 12 bar/°CA can be evaluated critically. It can be seen that with increased EGR rates the fuel conversion is slower and therefore lower pressure gradients appear. However, soot, CO and HC emissions increase as a result of incomplete combustion. With a moderate increase in fuel consumption of approximately 6 % and NOx emissions of 0.5 g/kWhi, soot emissions of less than 0.012 g/kWhi can still be achieved. Figure 6 shows the results of the 6 bar IMEP operation point compared to the conventional process. In addition, the results of a 0.3 g/kWhi NOX point are shown here. In principle, it is therefore possible to fulfil the current Euro 6 NOX limit with this process without the help of an EAT system. However, fuel consumption increases by about 15.5 g/kWhi. In addition, soot emissions are also increased to 0.08 g/kWhi. A loss analysis shows that the reason for increased fuel consumption is mainly due to the unfavourable MFB50 and the real gas composition. Finally, due to the reduction of the pressure gradient, a load increase up to 8 bar IMEP could be carried out by early and late inlet valve closing. However, a further load increase was not investigated due to the pressure gradients of up to 14 bar/°CA. Here also low NOX emissions of 0.5 g/kWh and PM emissions of 0.04 g/kWh were obtained.

550

Potential of air path variabilities for heavy duty Diesel engines 0.020

Conventional, CA50=11, 1PI & 1MI

0.018 0.016

Miller PCCI, CA50=16, 1MI, EIVC 70, 1g/kWh sp. NOx

sp. PM / g/kWh

0.014

Miller PCCI, CA50=16, 1MI, EIVC 70, 0.5g/kWh sp. NOx

0.012 0.010 0.008 0.006 0.004 0.002 2

4

6 8 sp. NOx / g/kWh

10

225

490

220

450

215

410

TExh. / °C

bi / g/kWh

0

210 205 200 195

14

370 330 290 250

CI PCCI PCCI PCCI 4g 1g 0.5g 0.3g

CI PCCI PCCI PCCI 4g 1g 0.5g 0.3g

0.7

14

0.6

12

0.5

10

dp/dφ / bar/°CA

sp. HC / g/kWh

12

0.4 0.3 0.2 0.1 CI PCCI PCCI PCCI 4g 1g 0.5g 0.3g

8 6 4 2 CI PCCI PCCI PCCI 4g 1g 0.5g 0.3g

Figure 6: Comparison of the Miller PCCI process with conventional diesel combustion process

551

Potential of air path variabilities for heavy duty Diesel engines

4 Summary and Conclusion In this paper the potentials of Miller valve timing at full load and the temperature increase for exhaust gas management at partial load were investigated. In addition, the influence of variable intake valve timing on partially homogeneous combustion was investigated. On the basis of the results shown here, it can be established that variable valve timing can be an effective method to reduce fuel consumption and emissions. By using the Miller process in the upper load range and closing the WG of the highpressure turbine, a significant reduction in fuel consumption at 1400 rpm could be achieved with a standard turbo charging group. However, the improvement decreased up to 1600 rpm and a fuel consumption disadvantage at rated speed was observed. In order to further increase the potential for improvement, the charging group was subsequently adjusted and Miller valve timing was applied. Therefore, a further fuel consumption reduction in the relevant operating range of the commercial vehicle engine in long-distance traffic without any notable disadvantages at nominal speed was achieved. A validated EAT model was used for the exhaust gas management investigations. For the potential analysis, in addition to the reference valve timing, early intake valve closing and EEVO were also investigated and compared with a conventional method (intake air throttling). It is shown that EEVO is an efficient method for a faster heat up und therefore earlier AdBlue injection. However, EIVC can also contribute to an improved cool down behaviour, particle regeneration and SCR efficiency. Finally, the potential of valve timing for alternative combustion processes was investigated. A combination of the HCLI and HPLI processes was used. The load range with conventional diesel hardware has been extended to 8 bar indicated mean effective pressure due to the extended ignition delays caused by Miller valve timing. The results show emissions close to the Euro VI standard, which can be achieved with only moderate fuel consumption increases of approx. 6 %. Due to the good controllability of the process and the low soot and NOX emissions, this may represent a potential method to fulfil the nitrogen oxide limits during the warm up of the EAT system in view of future legislations.

Acknowledgement This paper is the scientific result of a research project undertaken by the FVV (The Research Association for Combustion Engines eV) and performed by Marius Betz and Dávid Kovács at the Institute of Internal Combustion Engines at the Technical University Braunschweig under the direction of Prof. Dr.-Ing. Peter Eilts. Special thanks are due to the AiF (German Federation of Industrial Research Associations eV) for funding the project. The project was supported by an expert group led by Dr.-Ing. Christian

552

Potential of air path variabilities for heavy duty Diesel engines Weiskirch (TRATON Group). The authors gratefully thank the support received from the chairman and from all members of the project user committee. The research project was carried out in the framework of the industrial collective research programme (IGF No. 18762 N/1). It was supported by the Federal Ministry for Economic Affairs and Energy (BMWi) through the AiF (German Federation of Industrial Research Associations eV) based on a decision taken by the German Bundestag.

Bibliography 1.
DieselNet [Online]. Available: https://www.dieselnet.com/standards/. [Zugriff am 31.07.18]. 2.
Europäische Kommission 2018. [Online]. Available: https://ec.europa.eu/clima/ policies/transport/vehicles/heavy_de. [Zugriff am 31.07.2018] 3.
DieselNet [Online]. Available: https://www.dieselnet.com/standards/us/fe_hd.php. [Zugriff am 31.07.2018]. 4.
S. Gehrke, D. Kovács und P. Eilts, Investigation of VVA-Based Exhaust Management Strategies by Means of a HD Single Cylinder Research Engine and Rapid Prototyping Systems, SAE Technical Paper 2013-01-0587, 2013. 5.
D. Kovács und P. Eilts, „Potentials of the Miller Cycle on HD Diesel Engines Regarding Performance Increase and Reduction of Emissions,“ SAE International 2015-24-2440, 06 09 2015. 6.
S. Gehrke, D. Kovács und P. Eilts, „Abschlussbericht über das Vorhaben Nr. 1065 (IGF-Nr. 17080N) – Potential von Luftpfadvariabilität am NFZ-Motor zur Leistungssteigerung bei Optimierung des Schadstoff- und des CO2-Ausstoßes,“ Forschungsvereinigung Verbrennungskraftmaschinen e.V. Heft 1060-2015, Frankfurt am Main, 2015. 7.
M. Betz, D. Kovács und P. Eilts, „Abschlussbericht über das Vorhaben Nr. 1199 (IGF-Nr. 18762 N/1) - Potenzial von Luftpfadvariabilitäten am Nfz-Motor zur Leistungssteigerung bei Optimierung des Schadstoff- und des CO2-Ausstoßes,“ Forschungsvereinigung Verbrennungskraftmaschinen (FVV) e.V., Frankfurt am Main, 2018. 8.
S. Gehrke, „Beitrag zum gaspfadseitigen Abgasmanagement an Nutzfahrzeugmotoren,“ in Berichte aus dem ivb – Dissertation TU Braunschweig, Aachen, Shaker Verlag GmbH, 978-3-8440-3669-5, 2015.

553

Potential of air path variabilities for heavy duty Diesel engines 9.
B. M. Lump, E. Trapel, M. Münzenmay, M. McMackin, E. Bouillon und K. Zimmermann, „Desktop Simulation and Calibration of Diesel Engine ECU Software using Software-In-The-Loop Methodology,“ in GT-SUITE European Conference, Frankfurt am Main, 2013 10.
P. Glonegger, C. Weiskirch, M. Erath, P. Pribyl und R. Cadle, „MAN D26 Two Stage development, heavy duty commercial engines optimized with respect on low fuel consumption and emissions,“ in Aufladetechnische Konferenz Dresden, Dresden, 2016. 11.
E. Beineke, Rechnerische Untersuchung der ein- und zweistufigen Abgasturboaufladung mittelschnellaufender Viertaktdieselmotoren, Braunschweig: Dissertation Technische Universität Braunschweig, 1975. [11] 12.
M. Rößler, „Investigation of the effect mechanisms of the formation of NO2 in Diesel engines,“ in WKM-Symposium, Braunschweig, 2017. 13.
H. Theißl, T. Kraxner, H. Seitz und P. Kislinger, „Miller-Steuerzeiten für zukünftige Nutzfahrzeug-Dieselmotoren,“ MTZ11/2015, p. 18 14.
F. Doosje, F. Willems und R. V. D. M. Baert, „Experimental Study into a Hybrid PCCI/CI Concept for Next Heavy Duty Diesel Engines,“ SAE paper 2012-01-1114, 2012.

554

Automated & cloud-based load profilegeneration and evaluation of lithium ion batteries Dipl. -Phys. Alexander Kohs (Author) CTC cartech company GmbH, Hanns-Klemm-Straße 40, 71034 Böblingen, [email protected], +49 703120405854 Dr. -Ing Thomas Freudenmann (Co-Author 1) EDI GmbH – Engineering Data Intelligence, Wöschbacher Str. 73, 76327 Pfinztal, [email protected], +49 72115663022 Fabian Back (Co-Author 2), CTC cartech company GmbH Dr.-Ing. Mohannad El-Haji (Co-Author 3), EDI GmbH Tobias Schilling (Co-Author 4), CTC cartech company GmbH Praveen Kumar Kuppusamy (Co-Author 5), CTC cartech company GmbH

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 M. Bargende et al. (Hrsg.), 19. Internationales Stuttgarter Symposium, Proceedings, https://doi.org/10.1007/978-3-658-25939-6_107

555

Automated & cloud-based load profile-generation and evaluation of lithium ion …

Abstract Today lithium ion batteries (LIB) are the state-of-the-art technology for traction batteries in electric vehicles due to their high energy capacity and output as well as power density. However, the determination of the aging and performance of the LIB remains a challenge, since these essential parameters are highly dependent on the vehicle, user behaviour and environmental conditions. Therefore, traction batteries have the necessity to be constantly monitored by the vehicle electronics. In addition, new methods are needed for the efficient evaluation and diagnosis of the LIB, which are used during development, production and subsequent operation. In this paper, a fully automated, cloud-based process is introduced, which includes the measurement of the LIB during operation, the generation of load profiles at cell level and transfer to battery test benches. To determine the load profile, the combination of the driving profile of the vehicle (for example GPS data) with the load acting on the installed LIB cells takes place. With the help of a central cloud-based IT platform, data loggers, battery test stands and Key Performance Indicators (KPIs) for the aging process of LIB are networked to automatically integrate the generated load profiles into laboratory test series, thus enabling efficient and customer - oriented development. Thus, all battery types in the lab can be loaded with the critical load profile of the specific vehicle and user behaviour, the best technology for the intended use can be determined and vehicles with optimized range, charging time and lifetime can be developed.

1 Introduction As the world move towards sustainable transportation, demand from various customers and manufacturing technology has made automobile manufacturers to introduce electric vehicles (EVs), such as hybrid EVs, battery EVs and plug-in hybrid EVs. To power the vehicle, one common unit used in all EVs is a large electrochemical battery. Properties such as high power, higher cycle life, high energy density make Lithiumion (Li-ion) chemistries, currently the best candidate for EV applications

1.1 Background Performance and ageing of lithium ion batteries (LIB) heavily depend on external ageing factors (EAF) like discharge and charge current, state of charge (SOC), temperature or depth of discharge (DOD). False treatment of the battery may lead to a thermal

556

Automated & cloud-based load profile-generation and evaluation of lithium ion … runaway resulting in an exothermic reaction. Therefore, it is necessary to monitor the battery during its entire life span. The battery life cycle is subdivided into four parts: ●
Cell development and production ●
Integration into applications ●
Operation of the battery ●
Second life and recycling sector

Figure 1: LionTelligence – Battery Life Cycle Management [1]

LionTelligence developed by CTC cartech company in cooperation with EDI [4], operated on the customizable EDI hive cloud platform which can run on different cloud providers, e.g. Azure, AWS, GCE (Figure 2), functions as the central, intelligent platform. It provides algorithms and battery models that determine not only the state of health (SOH) but also the residual useful life and the state of value of the battery for various applications. By continuously gathering data from all Lion-Products, these algorithms improve constantly and provides smarter services for every company. Hence, the economic benefit of the battery is optimized [1].

557

Automated & cloud-based load profile-generation and evaluation of lithium ion …

Figure 2: HMI of smart service ”Load profile generation”

1.2 Aim LionTelligence provides different functionalities, which are called smart services. This papers aim is to give a documentation of one smart service, focusing on automated load profile generation on EVs, evaluation of lithium ion batteries in the laboratory and reporting the health of the batteries. The general approach is: 1.
Generation of vehicle drives cycles: Data logging, data visualization and data transfer to the laboratory. 2.
Analysis of the measured data: Simpler load profiles are generated which are used to test batteries in the lab. 3.
Generation of test plan: Using of design of experiments (DOE), the amount of required test runs is significantly decreased. 4.
Modelling of important battery parameters: A battery model is generated by using linear regression to determine and predict State of Health and remaining lifetime.

1.3 Outline Section 2 provides an Overview of Automated load profile generation and brief steps involved in it. Section 3 demonstrates the data acquisition process such as data logging, visualization and analysis. Section 4 involves entire processes in the battery laboratory such as load profile analysis, generation of test plan, battery modeling and

558

Automated & cloud-based load profile-generation and evaluation of lithium ion … updating the health of lithium batteries. Section 5 gives a brief summary of the results produced and Section 6 provides the possible future work involved in this paper.

2 Overview Smart Service

Figure 3: Automated load profile generation overview

The various steps involved in automated load profile generation is described in Figure 3. ●
Data acquisition ●
The data logger records all data during the test drives ●
The recorded signals are visualized CANalyzer ●
The required signals can be now stored in the cloud database ●
The User Interface (UI) provides the visualization of the load profiles ●
Depending on the complexity of the load profile, the micro services now calculates a new corresponding load profile and classifies it as worst/easy module ●
Battery laboratory ●
At the CTC’s laboratory, Lithium ion cells are aged with various ageing factors (temperature, C-rates, SOC, DOD) ●
The output response of the cells is compared with the statistical battery models and ageing parameters are derived

559

Automated & cloud-based load profile-generation and evaluation of lithium ion … ●
These parameters are used in the load profile to evaluate the condition of Lithium ion batteries Thus, user, battery and vehicle data is semantically captured from all Lion devices at a central point of information, the LionTelligence HUB, powered by the EDI hive cloud platform (Figure 3). It may be used to improve battery models and to generate battery specific decisions and user instructions.

3 Data Acquisition Data acquisition (DAQ) is the method of measuring physical or electrical signals such as voltage, current or temperature with a system. This section consists of EV’s drive cycle generation, data logging and data visualization.

3.1 Drive cycle Initially the Smart ED is tested with various defined driving cycles. They include Standard cycles (50 km) for Autobahn, urban and rural roads. Three drivers have to drive these cycles two times (aggressive / moderate driving styles), so that an average can be made over all driving cycles. The platform allows you to visualise these routes and stores them in the cloud platform [Figure 4]. The cloud platform can provide detailed information such as test drive type, battery type, EV type, driving style, driving style etc.

Figure 4: Drive cycles

560

Automated & cloud-based load profile-generation and evaluation of lithium ion …

3.2 Data Logging Telemotive AG data logger is being used to record all data sent during the test drives using the HV BMS and Hybrid CAN. With the help of a universal adapter, both CANs are connected to the data logger. As soon as the data logger is supplied with 12V DC, it boots up and starts to record data according to its current configuration.

Figure 5: Data logging

The data logger is also able to record the GPS coordinates of the test drive, if set to do so in the configuration. Incorporating map data, the ascending/descending slope of every part of the test drive can be considered. All relevant data is stored in proprietary Telemotive AG format but can be converted into common industry standards, like CANoe ASCII Format *.asc data. CANalyzer (Figure 6) is used to translate the recorded CAN signals using the *.dbc database into known channels and values.

561

Automated & cloud-based load profile-generation and evaluation of lithium ion …

Figure 6: Vector CANalyzer

3.3 Data Visualization After data logging process, the EV battery parameters are stored in the cloud platform. The user can visualize these data such as voltage, SOC, temperature (Figure 7) and get information about the performance, health of the battery pack.

Figure 7: EV battery parameters

562

Automated & cloud-based load profile-generation and evaluation of lithium ion …

3.4 Data transfer The visualized data such as temperature, cell voltage, current, velocity, SOC is now transferred to the laboratory. Once the load profile of the EV is generated, accelerated ageing is performed in the laboratory with similar profile. This ensures a rapid ageing of cells without the presence of actual vehicle. Finally, this simulation of this load profile helps to determine the battery behaviour and through which battery model is derived. The process involved in the laboratory is explained in upcoming section.

4 Battery Laboratory (LionTrace) Lithium-ion battery dynamics is complex to understand and factors such as power fading, capacity decrease do not originate from a single cause, but from various processes and their interactions. This section explains about the load profile analysis, statistical plan and battery modeling.

4.1 Load profile The analysis of resulting load profiles become important premises for determination of vehicle status and health. The could be divided into realistic load profile (EV) and a simpler load profile.

4.1.1 Simple Load profile The state of health (SOH) is a figure of merit that presents the battery condition, compared to its ideal conditions. It is the ratio of current cell capacity to cell’s initial capacity. Battery ageing comes under two categories ●
Calendar aging ●
Cyclic ageing Calendar aging is strongly linked to electrolyte reduction and oxidation and the growth of surface films on the active materials. It strongly depends on time, SoC, and temperature. Cycle aging comprises also the mechanisms of structural and mechanical changes in the battery components. It is substantially more complex than calendar aging and depends also on a variety of additional parameters, such as charging and discharging currents, cycle depth, and charge throughput [3]. Hence the main Lithium ion battery’s SOH ageing factors are as follows ●
Depth of Discharge (DOD) ●
Maximum (SOC)

563

Automated & cloud-based load profile-generation and evaluation of lithium ion … ●
Charge rate (C+) ●
Discharge rate (C-) ●
Temperature (T) At the CTC lab, aging of batteries is done at various temperatures, SOCs, C-rates, DODs. Thus, the characterization and analysis of the battery performance is mandatory to gain knowledge about the dependence of electrochemical processes to the external ageing factors (EAF). Figure 8 shows a simple load profile.

Figure 8: Simple Load Profile

4.1.2 Realistic Load profile The vehicle data obtained from data logger is visualised and further analysed in the laboratory. The resulting data from one such vehicle is shown in Figure 9. The battery voltage is decreasing as the EV is driven. Once this realistic load profile is scaled down to a simpler one, modelling the battery’s behaviour is easy and predicting the SOH, remaining useful life can be carried out in a systematic approach.

Figure 9: Vehicle Load profile

564

Automated & cloud-based load profile-generation and evaluation of lithium ion …

4.2 Test plan A process for the derivation of a battery ageing model (BAM) has been developed. Table 1: Ageing table

Temp Group [°C]

Charge rate [C]

Energy Discharge SOC DOD Number throughput / rate [%] [%] of cells cycle [C] [Ah]

1

-20

4

10

100

95

5

4.75

2

-20

4

10

10

5

5

0.25

3

-20

4

5.05

55

5

4

0.25

4

-20

4

10

100

5

4

0.25

The goal is to obtain the following improvements when compared to conventional ways of developing BAMs: Shorter time period for ageing cycles at the laboratory, these BAMs can cover the entire parameter space. This process is applicable to other battery models and cell chemistries. Also it is an independent physical battery model which means, that different physical battery models can be used Since the SOH of a battery depends on various ageing factors, a statistical plan is developed and the cells are aged accordingly. An example of such plan is shown in Table 1. The cells capacity is also calculated periodically and cell’s SOC is adjusted accordingly. Design of Experiment (DoE) is used to generate a test matrix, since this method enables the investigation of many parameters with moderate testing effort.

4.3 Battery model Modeling the battery behavior is complex as they show non-linear effects during discharge. In general, Battery impedance is a nonlinear multivariable function of temperature, aging, current and SOC. In the development of model-based battery management system, battery model plays a vital role in electric vehicles (EVs) simulation and accurate estimation of batteries states. The SOC of a battery is analogous to a fuel

565

Automated & cloud-based load profile-generation and evaluation of lithium ion … gauge of a conventional vehicle, thus precise calculation is required in order to utilize the battery’s full capacity [2]. At CTC’s lab, all cells are aged according to their load profile. As shown in Figure 10, the cell’s initial SOH is 100%. As the cell ages, SOH starts decreases over Energythroughput. Using the slope value, statistical model can be derived which gives the residual health and remaining useful life of the battery.

Figure 10: SOH Modeling

5 Summary EV’s salability mainly depends upon the battery’s performance as it is the core energy source. Overview Smart Service (Section 2) explained the overall steps involved in this paper, which is powered by the EDI hive customizable cloud platform. Data acquisition (Section 3) gave an overview of drive cycle generation, how data logging is performed at the vehicle side, conversion of these data to desired format, visualization and data transfer through Smart services. Battery laboratory (Section 4) provided a look into various load profiles analysis at the laboratory, various statistical test plans and modeling of lithium ion battery’s SOH. Thus, an entire work process is demonstrated in this paper to generate load profile automatically and to evaluate the performance of lithium ion batteries.

566

Automated & cloud-based load profile-generation and evaluation of lithium ion …

6
Outlook With the number of EVs rising, small test devices, especially for workshops, which characterize the battery by using the data provided by the on-board diagnosis system or the battery management system will be of high demand. This upcoming workshop diagnosis tester ‘LionAid’ provides complex test routines with easy ‘traffic lights’ interfaces, to ensure ideal maintenance of the LIB. Additionally, when the first automotive batteries have to be replaced, 2nd Life applications like energy storages for used automotive batteries will be a big market. Hence a health observing modular energy storage device called ‘LionHome’ is also being developed. All project results are incorporated into this cloud-based platform, which is accessible via internet browser to the user from any device with online connection. It provides the combination of electrochemical battery models, deep battery knowledge along with big data algorithms and intelligent cloud software. Every Lion Product has access to ‘LionTelligence’, which empowers all devices to provide smart services and solutions for all entities along the battery life cycle. The overall goal is to provide an economical, ecological and efficient value to Battery Life Cycle Management.

Bibliography 1.
Kohs, Alexander, Ing Hans-Christian Reuss, Ing Ferdinand, and Heiner Schmeck Panik. " LionTelligence Intelligent Battery Life Cycle Management." 2.
P. Kuppusamy, “Analysis of lithium ion cell data by pulse fitting measurement and model validation to improve cell-specific SOH quality without high measurement accuracy,“ Ulm, Germany, 2017 3.
Keil, Peter. "Aging of Lithium-Ion Batteries in Electric Vehicles." (2017). 4.
Freudenmann, Thomas; El-Haji, Mohannad: “Applied AI for Industry 4.0 - New Assets for Manufacturing”, In: Data Productivity in Automotive Manufacturing, Berlin, 2018

567

Requirements for battery enclosures – Design considerations and practical examples Jobst Kerspe TEB Dr. Kerspe; Mauer Michael Fischer König Metall–GVI; Karlsbad-Ittersbach

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 M. Bargende et al. (Hrsg.), 19. Internationales Stuttgarter Symposium, Proceedings, https://doi.org/10.1007/978-3-658-25939-6_108

569

Requirements for battery enclosures – Design considerations and practical examples

Summary Requirements for battery housings in e-vehicles are extensive: regulatory requirements; functional requirements; consideration of the installation conditions, transformation of forces and torques into the vehicle structure as well as wishes and demands of the end customer for trouble-free operation under a wide variety of climatic conditions. Space and weights are scarce resources in electric vehicles; this means lightweight construction and multifunctionality are stringent requirements for all functional units. The multifunctional battery housing - the B: HOUSE® in GVI® technology - offers new and highly efficient solutions. This concept allows effective passive and active thermal management, vibration and crash-proof housing and fixation of the battery cells / modules, fire protection in all directions, EMC safety, environmental protection, lightweight construction – which means resource saving and weight-optimization in one functional unit.

1 Introduction When designing e-mobiles - and thus the batteries or battery cases - there are some basic requirements that have to be taken into account, both from the technology as well as from various regulations. For example, it is now well known that the usable electrical energy stored in the battery is limited (compared to the energy content of fossil fuels). It is also foreseeable in the short and medium term, that this will not change fundamentally. On the other hand, it should be noted that in case of damage, a multiple of chemical reaction potential can be released. Furthermore, it should be noted that the running during charging and discharging of the battery, chemical reactions are more or less dependent on the temperature level - it is not without reason to speak about the "feel-good temperature" of the battery. What does this mean for the design of the battery or battery case? a) The enclosure must provide the necessary protection from external mechanical influences (crash, vibration-proof fixation, safe removal of high battery weight forces and transmission of forces and moments in the vehicle structure) b) Active and passive thermal management to maintain the "battery feel-good temperature" under all possible operating conditions.

570

Requirements for battery enclosures – Design considerations and practical examples c) Protection against external fires as well as active environmental protection – which means encapsulation for every case of an accident (in some cases you will also find the trivialized description "Thermal Event"). d) EMC resistance e) Space and resource-saving lightweight construction How to transform these different requirements into a fundamental new concept, will be explained now.

2 The B: HOUSE® concept The basic technology of the B: HOUSE® concept as well as the basic technology for this is the GVI® technology, which should be described first.

2.1 The GVI® technology: structure and properties 2.1.1 Types of Vacuum-Insulations and their Possibilities in Design GVI® stands for Assisted Vacuum Insulation - we are talking about a highly efficient thermal insulation. In contrast to the well-known "Thermos® vessels"(Fig. 1), however, the insulation gap / vacuum space is firmly "stuffed" with a filling material, which supports the vacuum-tight envelope against the external atmospheric pressure and thus gives a tremendous rigidity and mechanical strength to the insulating element - similar to a honeycomb sandwich element. While the vacuum insulations, known from Thermos® vessels, cannot be pressure-loaded and therefore can only be produced in rotationally symmetric shapes. In contrast GVI® structures can be produced in any desired shape, and the walls, forming the vacuum space, can also be made very thin, which is a basic for heavy-duty lightweight structures.

Figure 1: Dewar vessel (left) and thermos flask® (right; [Alfi])

571

Requirements for battery enclosures – Design considerations and practical examples The fine, microporous structure of the filler is the main reason for a further physical effect: while the insulation in Dewar or Thermos® vessels must be operated in the high vacuum range, the supported vacuum insulation already works in the rough vacuum range - this results in cost advantages and longer lifetimes. The basic structure of a GVI® housing is shown in Fig. 2; but also flat, planar elements can be produced, from which complete rooms can be assembled (Fig. 3), which may be interesting for stationary batteries. A structure according to Fig. 2 is recommended in the B:HOUSE® - this can minimize thermal bridge effects (increased heat flow via metallic elements - in this case the connecting membrane on the open side).

Figure 2: GVI®-Enclosure

572

Requirements for battery enclosures – Design considerations and practical examples


 
  
 



Figure 3: Two-dimensional GVI® element (left) and spatial structure composed of different elements

2.1.2 Physical properties of supported vacuum insulation (GVI®) a) Thermal Features Naturally, the outstanding insulating properties of the evacuated insulation should be mentioned here first. Figure 4 shows the specific thermal conductivities of different insulation methods. Compared with conventional (non-evacuated) thermal insulation, the GVI® has an insulation effect, that is up to 30 times better. As a result, evacuated insulation requires a significantly thinner insulation. The possible application temperatures for GVI® systems are very wide: in the standard version, these insulations can be used from absolute zero to about 500 ° C - with minor modifications, an application even up to > 800 ° C is possible. A further outstanding feature is, that the insulation effect of the GVI® can be worsened by up to 100 times, by filling up with very small quantities of good heat-conducting

573

Requirements for battery enclosures – Design considerations and practical examples gases. This is referred to as adjustable or switchable thermal insulation (RWD); this effect is exemplified in Figure 6. With appropriate design, this effect is done without pumps, valves or other mechanical actuators and can be cycled as often as desired, so the insulation is in the "optimal" state again at the end of a switching cycle. It is easy to imagine that for larger GVI®-unit’s considerable amounts of energy can be dissipated via the housing surface – provided, that a sufficient temperature difference between the housing internal temperature and the ambient temperature is given! Due to the fact, that the insulation layer is completely encapsulated, it never will occur - even at fluctuating temperatures - that moisture will penetrate the insulating layer! b) Mechanical Features When GVI®-systems are properly designed, evacuation results in rigging of the thinwalled shell material with the filler - the GVI®-structure behaves like a highly rigid sandwich structure. The supporting effect makes it possible to produce large-area, flat walled elements with thin shell walls or to support large, heavy units (for example battery blocks) in the insulating layer against the outer vacuum jacket without additional support elements (which would effect thermal bridges!). The weight loads of a battery are transformed completely without additional support elements over the insulating layer on the outer supports! A GVI®-element can withstand forces (preferably pressure), shearing strain and moments, induced into a GVI® housing - but it is important to ensure that this should be via a large area.

574

Requirements for battery enclosures – Design considerations and practical examples

Figure 4: Comparison of different Insulation-Types

Figure 5: Specific thermal conductivity as a function of the residual gas pressure

575

Requirements for battery enclosures – Design considerations and practical examples c) Vacuum Features The vacuum characteristics of GVI®-systems act in the range of 0.1 mbar according to Figure 5. Due to the fact, that this vacuum must be maintained over the required life time of a housing, there are some minimum requirements for the selection of the materials used, their pretreatment and the production technology.

Figure 6: Switchable Insulation

576

Requirements for battery enclosures – Design considerations and practical examples Not least for these reasons, the shells of the GVI® structures at König Metall are manufactured almost exclusively from austenitic stainless steel (for example 1.4301) and completely subjected to a helium leak test.

2.2 The B: HOUSE® Concept 2.2.1 Multifunctional Structure The basic idea of the B: HOUSE® is a complete integration of the required functions and properties into the housing. In a GVI®-housing, numerous functional units can be integrated into this double-walled construction. Not all of these additional functions must be installed in one housing - the respective structure depends on the requirements of the application. Depending on the required thermal properties, the total wall thickness is between 5 mm and 15 mm. One possible design variant of a fully enclosed battery is shown in Figure 7.

Chiller – between the walls / or as part of the inner wall

B:HOUSE® in GVI®-Technology with switchable Insulation

Heat storage by PCM – placed inside the GVI®

Fig. 7: Principle sketch of the B:HOUSE® Concept Figure 7: Principle sketch of the B:HOUSE® Concept

577

Requirements for battery enclosures – Design considerations and practical examples

2.2.2 Integrated functions With the concept of a multifunctional battery housing - B: HOUSE® - a wide variety of basic requirements can be fulfilled within one single unit.

Tight and stress resistant GVI®-

Large-area heat conduction via the com-

structure with switchable insulation

plete housing-surface – when switched

Battery-block

Inner wall of GVI® designed as heat exchanger

Internal heat-storage by PCM

for active cooling

Figure 8: Possible features of the Thermo management in the B:House®-Concepts

●
Passive thermal management: With the GVI® technology, both, a highly efficient thermal insulation and a regulation of the insulation effect for large-scale heat dissipation is possible. Thus, influences of the ambient climate can be cushioned without additional energy expenditure - therefore we speak of a passive thermal management ●
Additional assemblies for active thermal management - e.g. cooler or PCM heat storage - can be installed directly in the double-walled housing. For this purpose, for example, one or more of the inner walls may be designed as cooling surfaces, which are linked to a refrigeration unit outside the B: HOUSE®.

578

Requirements for battery enclosures – Design considerations and practical examples ●
In addition, such a vacuum insulation has excellent mechanical properties, so that mechanical loads on the cell can be optimally intercepted (crash safety - ECE R100 / 2, vibration-proof fixation of the cells). ●
With appropriate design, the housing can be installed as part of the vehicle structure. ●
The housings are EMC-proof (closed Faraday cage) without additional construction measures. ●
The housings protect the cells from external fire (ECE R100 / 2) and, with appropriate design, protect the environment even in the case of a "thermal runaway" of individual (or more) cells > delayed spread of such fires to the entire vehicle; no leakage of cell chemistry ●
Since all the features mentioned are combined in one housing, resource-saving lightweight construction concepts are possible.

3 Examples 3.1 Different designs with the respective advantages and disadvantages Tube model (Fig. 9) with stopper: Tubular or tunnel-shaped housing in which the battery pack is inserted via rail systems; the open end is insulated and sealed with a plug (conventional insulation or GVI® structure).

Figure 9: High-Temperature Battery (NaS) with GVI®-Housing

579

Requirements for battery enclosures – Design considerations and practical examples Advantages: relatively small thermal bridge zone; battery is relatively easy to open for maintenance; good and easy fixation of the cells / modules possible. “Babushka” (Fig.10) The battery pack is enclosed by 2 half-housings on all sides; with appropriate design of the connection zone, the thermal bridge effects can also be minimized here; the battery pack is also fixed here via rail systems etc. in the housing. Advantages: almost complete enclosure with GVI® housing with all safety advantages; low thermal bridge effects; very good force and torque transmission in the vehicle structure; good accessibility of the battery in case of revision.

Figure 10: “Babuschka”-Concept: 2 Envelope-Half shells

Trough with lid (Fig. 11) The battery modules can be installed from above in the trough and fixed with appropriate fasteners. Advantages: easy handling in battery construction Disadvantages: strong thermal bridge effects; depending on the size lower rigidity than other types

580

Requirements for battery enclosures – Design considerations and practical examples

Figure 11: Trough

Complete enclosure The battery pack is enclosed on all sides by the GVI® structure. Advantage: maximum safety gain (crash; encapsulation) Disadvantages: complex handling during maintenance; complex electrical feedthroughs

Figure 12: Integral housing of the battery with GVI®

581

Requirements for battery enclosures – Design considerations and practical examples

3.2 Example of a large truck housing 3.2.1 Calculation for passive air conditioning The battery housing (B: HOUSE®) has an approx. 15 mm thick GVI® insulation - open on one side. The insulating effect is sufficient to keep the battery pack at operating temperature for more than 12 hours without additional heating! (ambient temperature minus 20 °C; starting temperature of the battery 25 °C; temperature after 12 h 20 °C). Conversely, of course, even at high ambient temperatures (> 40 ° C) the battery does not unnecessarily absorb heat from the environment - even in this case without additional active cooling with the corresponding energy requirements. In the event that due to exothermic battery reactions, an active cooling is necessary, the GVI® ensures that the introduced cooling power benefits only the battery and not unnecessarily discharged to the environment!




Figure 13: Electric Truck with a Battery mounted to side of the Structure (FRAMO)

3.2.2 Weight ratios / carrying behavior In the largest version, the housing has approximately the following internal dimensions: approx. 700 x 600 x 1800 mm. The total wall thickness is about 18 mm - of which twice

582

Requirements for battery enclosures – Design considerations and practical examples sheet thickness 0.8 mm. This housing weighs about 75 kg empty; the battery weighs about 1000 kg in this case. The entire battery is fixed to the side rails of the truck via three cantilever arms and straps. Conclusion: The total weight forces of the battery (in the amount of 1000 kg) are completely supported by the lightweight housing and introduced shockproof in the longitudinal structure of the vehicle.

4. The battery case as part of the vehicle structure – presented in the concept of the ECTT 4.1 The concept of the ECTT (E-City-Thermo-Truck) The battery housing (s) in GVI® technology form a sturdy platform, which sits directly on the vehicle's longitudinal frame rails and at the same time serves as a supporting structure for the refrigerated box and forms a firmly connected unit with it. In this way, the space between the vehicle floor and the vehicle longitudinal members, which is necessary for the travel of the wheels, is optimally utilized, without additional components (for example crosswise arranged carrier structures), which otherwise bring only additional weight.

Figure 14: Side view of the ECTT with one unit - consisting of a refrigerated box and the B: HOUSE® units

583

Requirements for battery enclosures – Design considerations and practical examples At the same time, the battery housing formed in GVI® technology represents a highly efficient thermal insulation of the bottom of the case and thus prevents the heat absorption from the heat radiation of the road surface, which is not insignificant in the summer.

Bibliography 1.
Kerspe, J. et al, Vakuumtechnik in der industriellen Praxis, 3. Auflage, expert-Verlag, ISBN 3-8169-2196-5 2.
Smoluchowski, M., Über Wärmeleitung pulverförmiger Körper, Bericht II. Intern. Kältekongress, Wien; (1910) 3.
Froehlich, D., Earth-Laid Transport Pipes for -162°C to +400°C, Pipeline Technology Conference 2011, Hannover 4.
Neumann, H., Thermal Performance of Different Glass Microspheres in Comparison to Perlite between 77 and 300K 5.
Kerspe, J., Vielfältige Eigenschaftsprofile von gestützten Vakuum-Isolierungen – Einsatzmöglichkeiten für das aktive und passive Thermomanagement in Fahrzeugen; 3. VDI-Fachkonferenz „Thermomanagement für elektromotorisch angetriebene PKW“ 5./6.11.2014 6.
Kerspe, J., Passives und aktives Thermomanagement mithilfe von multifunktionalen Batteriegehäusen für die E-Mobilität; 6. VDI-Fachkonferenz „Thermomanagement für elektromotorisch angetriebene PKW“ 14./15.11.2017 7.
Kerspe, J., „Multifunctional Battery-Housings“ and Their Helpful Application for Thermal and Mechanical Tasks; 8. aabc, Feb. März 2018, Mainz 8.
N.N., Elektroautos im arktischen Winter; emove Magazin 01-2014 9.
N.N., Integrierter Brandschutz für Lithium-Ionen-Batterien; VDI Nachrichten 1.8.2014 10.
Stenner, P., VisioM – Leichtfahrzeugkonzept für urbane Elektromobilität; ecartec 2014

584

Prediction of the lifetime of urban electric bus traction batteries in the context of the overall system design Martin Ufert Dresden University of Technology, Chair of Vehicle Mechatronics

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 M. Bargende et al. (Hrsg.), 19. Internationales Stuttgarter Symposium, Proceedings, https://doi.org/10.1007/978-3-658-25939-6_109

585

Prediction of the lifetime of urban electric bus traction batteries in the context of …

1 Introduction Driven by the ongoing discussion about clean air in German cities, many municipal transport companies are pushing ahead with the conversion of their bus fleets from diesel buses to electric buses. In many cities, this is an elementary step in the package of measures to reduce emissions and comply with legal limits. In recent years, various pilot projects have been launched and completed in various calls for proposals (e.g. Berlin, Hamburg, Dresden) [1]. In Dresden, for example, a 12 m electric bus with a conductive high power charging system was extensively tested as part of the "Electric Bus Line 79" project [2]. The aim of all of these projects was to test the different systems available on the market with regard to readiness for use and to gain initial experience in handling electric buses and their use in regular passenger service. An overview of the pilot projects in Germany is given in [1]. In many of these projects mainly vehicles were put into operation, which are then used on specially selected routes – often with low daily mileage in order to avoid the range problem. Only in a few projects was the number of vehicles procured sufficient to operate an entire line with electric buses. In the coming years, the next step towards switching bus fleets to electric mobility is to be taken. Many public transport companies are planning to procure a larger number of vehicles with which entire lines can then be operated completely with electric buses. In order to maximise public awareness, there is often a desire to choose so-called volume lines. These are lines that have a high passenger volume and often a high daily mileage and require a corresponding range. The vehicles currently available on the market have very different ranges. The manufacturer's specifications range from approx. 150 km for a 12 m vehicle [3] to over 300 km for an 18 m vehicle [4]. However, depending on the choice of line, this range may be too short to ensure safe operation under all circumstances. In such cases, charging the energy storage device during operation is required in order to be able to perform the daily driving performance. This charging can be carried out in different ways. The main degrees of freedom are the location, duration and power of the charging. Switching from diesel buses to electric buses does not only mean simply replacing the vehicles, but is to be understood as a way of designing the system. Today, this design is often based on empirical values. For example, specific mean values (in kWh/km) are used to estimate the required energy content of the traction battery for a certain distance. For charging, known combinations of charging location and charging power are then used, which are often known from the pilot projects. However, in many cases such a system design has a rather heuristic character and often only a few design scenarios are

586

Prediction of the lifetime of urban electric bus traction batteries in the context of … compared with each other. This procedure is shown exemplarily in [5] and [6]. An optimal design, for example with regard to the required investment and operating costs, cannot be determined thereby. This paper presents a methodology which automatically calculates and evaluates a multitude of technically possible configurations for a given route under given boundary conditions. The core of the methodology is a detailed, multi-stage battery model, since the traction battery is currently the most expensive single component of an electric vehicle [7] and its consideration therefore has a particularly high priority in system design. Probably the most important differences between electric buses and diesel buses in daily operation are the limited range and the much more complex charging process (charging energy storage vs. refuelling in conventional diesel buses). Analogous to the refuelling process, charging the vehicles in the depot during the night break is the state of the art. The required charging infrastructure confronts transport companies with great challenges, especially if a large number of vehicles are to be charged simultaneously in the depot. However, the charging process in the depot and the required hardware will not be examined in detail in this paper. The focus of the methodology presented here is on the choice of a suitable energy storage device as well as a possibly necessary charging infrastructure along the route being considered. The following questions are to be answered: ●
What is the power requirement of a vehicle to cover the considered distance? ●
What is the required energy content of the traction battery? ●
How many charging points are required along the line and where are they positioned? ●
What charging power must be installed at the charging points? ●
Is the stop time planned according to the timetable at the charging points sufficient? ●
How does the intended operation affect the lifetime of the traction battery?

2 Framework for the Design of Electric Bus Lines The developed framework was especially designed to answer these questions. The aim is to carry out a system simulation with subsequent parameter variation within the framework. Thus, a large number of technical configurations can be calculated for a specific system and be evaluated in a post-process. The following variation parameters are defined for this purpose:

587

Prediction of the lifetime of urban electric bus traction batteries in the context of … ●
Energy content of the traction battery, ●
Number and position of charging points in the network, ●
Power to be installed at the charging points. The route to be investigated is first examined with regard to the possible charging point locations. All locations that could be used as potential charging points are marked. Criteria for this can be, for example, the amount of space required, the available network connection or a pause in operation. A table with all possible combinations is then created. For an example route with two potential locations, this results in 4 combinations: Table 1: Combination of charging points (CP) ID

Depot-Charging

Charging Point 1

Charging Point 2

(0/0) (1/0) (0/1) (1/1)

1 1 1 1

0 1 0 1

0 0 1 1

It is assumed that depot charging always takes place so that it is not part of the parameter variation. Only the combination possibilities of charging point 1 and charging point 2 are varied at this point. The variation parameters of charging power and battery energy content are varied within (user-defined) limits in discrete increments. All possible combinations of variation parameters are now calculated within the framework. The subsequent evaluation of the results is initially carried out with regard to technical feasibility. Often not all theoretically possible combinations are technically feasible. In the second step, the technically feasible configurations are economically evaluated and compared with each other. In the following chapters, the individual models of the system simulation are presented.

3 System Modelling The system simulation consists of two main modules: Module 1 determines the power requirement of the vehicle during operation on a given route. The structure and functionality of this module are explained in [8] and will not be discussed in detail here. In module 2 the dimensioning of the energy storage as well as the required charging infrastructure is carried out. The core of this module is a detailed, three-stage battery

588

Prediction of the lifetime of urban electric bus traction batteries in the context of … model. The system simulation is completed by an electrical model of the charging infrastructure as well as a cost model for the evaluation of the determined configurations. The individual models are explained below.

3.1 Battery Model The battery model consists of 3 sub-models: Electric, thermal and aging model.

Figure 1: Three-stage battery model

The sub-models are called serially. The power curve P(t), which is calculated in module 1 for the considered vehicle serves as input to the battery model. In the electrical model, the battery current resulting from the power requirement and the State of Charge (SOC) are calculated. Both values are transferred to the thermal model. Now the battery temperature is determined. Then the battery current, the SOC and the battery temperature are transferred to the aging model, in which the aging of the battery resulting from the specific load is calculated. Both the results of the thermal model and the aging model are fed back into the electrical model. On the one hand, the battery temperature determines the maximum permissible current, so that current limitation (and thus power limitation) may be necessary. On the other hand, due to battery aging, the parameterization of the electrical model must be adapted before a new calculation of the power curve can be performed.

589

Prediction of the lifetime of urban electric bus traction batteries in the context of …

3.1.1 Electrical Model An equivalent circuit diagram model is used to model the electrical behaviour of the battery. Different depths of modelling are known from the literature [9]. Common to all is the approach consisting of an open-circuit voltage source and an ohmic internal resistance. Then, depending on the desired accuracy, one or more RC elements are connected. As the number of RC elements increases, the accuracy of the results generally increases, but the computational and parameterization effort increases as well. As a compromise, an equivalent circuit diagram containing one RC element was chosen.

Figure 2: Equivalent circuit diagram model of a single battery cell

The equivalent circuit diagram is first parameterized for a single cell. Subsequently, an equivalent circuit diagram for the entire battery storage is drawn up using the known connections to the series and parallel connection. The following information is essential for the parameterization of the cell model: •
Open Circuit Voltage (OCV), •
Ohmic internal resistance Ri, •
Capacity C and resistance value R of the RC element. All mentioned components are not constant quantities and have non-linear progressions, so that corresponding characteristic diagrams have to be defined in the simulation. OCV The open-circuit voltage depends on the SOC of the individual cell. In addition, a capacity is assigned, since a battery cell can only absorb and release a limited amount of charge. The capacity of the cell depends on the temperature. Ohmic internal resistance The ohmic internal resistance depends on both temperature and SOC.

590

Prediction of the lifetime of urban electric bus traction batteries in the context of … RC element Both components of the RC element are temperature and SOC dependent. Some of the required parameters can be taken from cell data sheets. Especially the values of the RC element are rarely given, so that these values have to be measured in the laboratory. A procedure for this is explained in [10]. In addition to the parameterization of the individual equivalent circuit components, limit values for battery operation can be stored in the electrical model. In real operation, these limits are usually specified in the battery management system (BMS) and are intended to ensure safe and reliable operation of the battery. A simple and effective way of such a BMS parameterization are current characteristic diagrams depending on the battery temperature and the SOC (Figure 3).

Figure 3: Permissible battery current, left: discharge, right: charge

3.1.2 Thermal Model In the thermal model, the temperature of the entire battery storage device is determined. From the literature different modelling depths for single cells are known. In [11] the principal representation of 0D, 1D, 2D and 3D models is shown. The relationship between parameterization effort, model complexity, result accuracy and calculation time is also shown. For this methodology, a 0D model is chosen due to the low parameterization effort and the short computation times. Analogous to the electrical model, a model at cell level is first created here as well. However, the interconnection to the full storage is not trivial. Often individual cells are grouped into modules, which in turn are connected in series and parallel to the full storage device. In order to determine the battery temperature, the consideration of such a module is essential. In the 0D model, a heat flow is injected in the middle of a rectangular body (Figure 4, left). Part of the heat is then released across all 6 body surfaces according to the thermal

591

Prediction of the lifetime of urban electric bus traction batteries in the context of … cell properties. If several cells are connected together to form a module, the cell does not release the corresponding proportion of heat from the contact surface into the environment, but transfers it to adjacent cells.

Figure 4: Thermal model, left: single cell model, right: module model

Thus, the interconnection of 0D cell models results in a 1D module model (Figure 4, right), since the inner cells have a higher temperature due to the heat transfer from the outer cells and thus a temperature curve in the x-direction of the module results. The model used is only valid for rectangular bodies. Accordingly, only prismatic cells and pouch cells can be simulated. For round cells a differentiated analysis is required.

3.1.3 Aging Model The aging model was developed on the basis of the institute's own measurements. A detailed description of the model structure as well as the parameterisation and the required workflow for the determination of a concrete aging can be found in [12]. As state of the art, four essential influencing factors on the aging process of Li-ion batteries are known. ●
Battery temperature, ●
SOC swing or ΔSOC, ●
Battery current (C rate), ●
SOC. All influencing factors mentioned lead to the aging of the battery with different quantities. This has the following effects: On the one hand the usable capacity (reduced range) of the battery decreases, on the other hand the internal resistance increases (lower performance, e.g. regarding acceleration, max. speed or charging time). By definition, a

592

Prediction of the lifetime of urban electric bus traction batteries in the context of … Li-ion battery reaches its End of Life (EoL) if at least one of the following two criteria is met. ●
Doubling of the internal cell resistance (to 200%), ●
Reduction of usable capacity to 80% of the nominal capacity. In [12] it was shown by aging measurements that both the decrease of the capacity and the increase of the internal resistance can be approximated by linear functions. On the basis of these measurements, damage factors S were determined, which describe the respective share of an influencing factor on the total aging. This results in the following model equations for the determination of aging. 

             



(1)



             



(2)



Initial capacity



Charge throughput



Resulting (reduced) capacity



Specific damage factor

The denominations apply analogously to the equation of internal resistance.

3.2 Model of Charging Points A simple approach is used to model the charging points. They provide a constant power, which is composed of the charge voltage and the charge current. A constant efficiency can optionally be used to take conversion and transmission losses into account. A retroactive effect on the feeding grid is not considered. The selected constant charging power represents the maximum value. Depending on the operating point of the battery, a derating can be activated. In this case, the available charging current at the charging point is higher than the permissible battery charging current must be limited accordingly.

3.3 Cost Model A cost model is to be used to evaluate the technical configurations determined. The model does not aim at providing a classical TCO calculation, as it is done in [13]. Rather, at this point only the individual technical configurations are to be compared with each other. Therefore, a simple cost model is used here, which only contains the main

593

Prediction of the lifetime of urban electric bus traction batteries in the context of … cost components of an electric bus system, in which the considered configurations differ. Therefore, the term system costs is used here. The individual components of the model are shown in Figure 5. The model is parameterized exemplarily for demonstration purposes (Chapter 4). Conventional diesel buses are usually in service for 12 - 15 years. A similar period is also assumed for electric buses. It is expected that the traction battery will reach the EoL criteria at least once during this period and will have to be replaced. The number of battery replacements thus has a significant influence on the lifecycle costs of an electric bus, which is why the cost share for this is correspondingly two-stage (Figure 5).

Figure 5: Cost model incl. example parameters

4 System Simulation with Parameter Variation In this chapter the functionality of the system simulation with parameter variation will be demonstrated by means of an example. The following scenario is given: Table 2: Operational data and parameter boundaries Type of vehicle Type of battery Length of line Number of rounds Daily mileage Cycle time Vehicle life cycle max. ΔSOC Ambient temperature Battery voltage Max. charging time CP 1 Max. charging time CP 2

594

18 m articulated bus Lithium-Iron-Phosphate 38 km 6 6 · 38 km = 228 km 20 min 12 years 0.9 - 0.1 15 °C 420 V - 710 V 15 min 10 min

Prediction of the lifetime of urban electric bus traction batteries in the context of … The power requirement of the selected vehicle to cover the considered distance is determined by means of the approach described in [8] and is available as a function of time P(t). The following limits are defined for the variation parameters defined in Chapter 2 (Table 3). Table 3: Boundaries of variation parameter Number of charging points Position of charging points Charging power Energy content of battery

0–2 At turning points of the line 50 kW – 400 kW 68 kWh - 377 kWh

In the system simulation, all possible combinations of the variation parameters within the defined boundaries are calculated. In this example, the step size of the charging power is 10 kW. The step size of the energy content of the battery is 13 kWh and results from the fact that the calculation is based on real cells available on the market. In a first step, all technically possible configurations within the defined boundaries are determined. At this point, the evaluation criteria is the daily SOC curve, at which the lower limit SOCmin must not be violated.

Figure 6: Min. energy content of the battery as a function of the charging point combination and the charging power

Figure 6 shows the minimum possible energy content of the battery, which can be achieved with the respective combination of number of charging points and charging power. It can be seen that no feasible configurations exist for variants of the charging

595

Prediction of the lifetime of urban electric bus traction batteries in the context of … point combination (0/0). Therefore, the considered route cannot be operated as a depot charger. The maximum energy content of 377 kWh in connection with the defined SOC boundaries is not sufficient to cover the daily energy demand of the vehicle. Due to the lower available charging time (10 min), the charging point combination (0/1) has fewer feasible configurations than the combination (1/0) with 15 min charging time. The largest number of feasible configurations results consequently from the charging point combination (1/1).

Figure 7: System costs of all technical feasible configurations

The next step is to evaluate all calculated variants using the cost model (Figure 5). The expected life time of the traction battery is calculated and on this basis the required battery replacement during the defined vehicle life of 12 years is calculated. In order to avoid a distortion of the results due to integer rounding, decimal numbers for the necessary number of battery changes are permitted. Figure 7 shows the resulting system costs for all calculated configurations. It can be seen that the lowest system costs are for load point combination (1/1). The absolute minimum is not directly recognizable due to the very low variance of the results. The optimum cost can be determined using a minimum search function: ●
Charging point combination (1/1) ●
Energy content of the battery: 122 kWh ●
Charging power: 320 kW

596

Prediction of the lifetime of urban electric bus traction batteries in the context of … In order to classify the found optimum, this is to be compared with the minimum technically feasible configuration with the same charging capacity and CP. The minimum possible energy content of the battery is 81 kWh (Figure 6). In Table 4, both configurations are compared with each other. The results show that choosing a 50% larger battery results in a 70% increase of battery life. According to the cost model (Figure 5), the cost of a single battery set also increases by 50%. However, the system costs over the vehicle life cycle can be reduced by 17% due to the extended lifetime. Table 4: Predicted Lifetime of the traction battery at EoL-criteria min. feasible energy content of the battery acc. to Figure 6 81 kWh 100 % 1072 days of operation (2,9 years) 100 % 2761 TEUR 100%

energy content of the battery at cost minimum acc. to Figure 7 122 kWh 150 % 1822 days of operation (5,0 years) 170 % 2300 TEUR 83%

5 Conclusion In this paper a methodology was presented which enables the dimensioning of energy storage and charging infrastructure of electric bus lines. Therefore, a framework was developed which varies the parameters battery energy content, charging power as well as location and number of charging points within definable limits. Within this framework, individual lines can be analysed by means of system simulation. The core of the system simulation is a three-stage battery model consisting of electrical, thermal and aging model. The functionality of the methodology was demonstrated by an example. In this example, all technically feasible configurations within the set variation limits were first determined. A cost model was used to evaluate these configurations. It could be shown that an economically optimal combination does not necessarily correspond to the combination with minimum technical effort. This is mainly due to the aging of the energy storage device. In this example, a battery with a larger energy content leads to a longer service life of the storage device and thus to a reduction in system costs over the vehicle life cycle.

597

Prediction of the lifetime of urban electric bus traction batteries in the context of …

Literature 1.
UITP (Hrsg.): ZeEUS eBus Report #2: An updated overview of electric buses in Europe, 2017 2.
ALEXANDER BUNZEL; PETERSOHN, Ronny (Mitarb.); BÄKER, Bernard (Mitarb.): Dresden's bus route 79 turned into full electric - Opportunity and challenge. In: HOFF, Carsten; SIRCH, Ottmar (Hrsg.): Elektrik, Elektronik in Hybrid- und Elektrofahrzeugen und elektrisches Energiemanagement VII. Renningen: expert Verlag, 2016 (Fachbuch / Haus der Technik, Band 142). 3.
CORA WERWITZKE: Mercedes-Benz rückt Serien-eCitaro ins Rampenlicht. URL https://www.electrive.net/2018/07/10/mercedes-benz-rueckt-den-ecitaro-insrampenlicht/ – Überprüfungsdatum 2019-01-17 4.
SILEO GMBH: Technische Daten: Sileo S18 – Überprüfungsdatum 2019-01-17 5.
MARTIN UFERT: Systemauslegung batterie-elektrischer Stadtbuslinien unter Berücksichtigung des Alterungsverhaltens der Hochvolt-Traktionsenergiespeicher (6. Internationale E-Bus Konferenz). Solingen, 21.11.2018 6.
MARTIN UFERT: Dimensionierung von Energiespeicher und Ladeinfrastruktur am Beispiel von Elektrobussen (Symposium „Elektrische Fahrzeugantriebe und -ausrüstungen“). Dresden, 30.11.2018 7.
KAMPKER, Achim (Hrsg.); VALLÉE, Dirk (Hrsg.); SCHNETTLER, Armin (Hrsg.): Elektromobilität: Grundlagen einer Zukunftstechnologie. 2. Auflage. Berlin: Springer Vieweg, 2018 8.
BUNZEL, Alexander; BÄKER, Bernard: Energy consumption of electric city buses: Determination as a part of a technological and economic evaluation of bus lines with regards to their electrifiability (ESARS-ITEC). Nottingham, 07.11.09.11.2018 9.
EINHORN, Markus; CONTE, Valerio; KRAL, Christian; FLEIG, Jürgen: Comparison of electrical battery models using a numerically optimized parameterization method. In: IEEE Vehicle Power and Propulsion Conference (VPPC), 2011: 6 - 9 Sept. 2011, Chicago, IL, USA. Piscataway, NJ: IEEE, 2011, S. 1–7 10.
LUTZ MORAWIETZ, STEFFEN KUTTER, RAINER FALSETT, BERNARD BÄKER: Thermoelektrische Modellierung eines Lithium-Ionen-Energiespeichers fur den Fahrzeugeinsatz. In: Innovative Fahrzeugantriebe 2008: Tagung Dresden, 6. und 7. November 2008. Düsseldorf: VDI-Verl., 2008 (VDI-Berichte, 2030, CDROM), S. 299–318

598

Prediction of the lifetime of urban electric bus traction batteries in the context of … 11.
FLECKENSTEIN, Matthias: Modellbasiertes Thermomanagement für Li-IonenZellen in elektrischen Fahrzeuganwendungen. München, Verlag Dr. Hut. Zugl.: Technische Universität Dresden, Diss., 2013. 2013 12.
MARTIN UFERT; Andy Batzdorf (Mitarb.); Lutz Morawietz (Mitarb.): Prädiktion der Lebensdauer von Traktionsbatteriesystemen für reale Nutzungsszenarien. In: VDI Wissensforum GmbH (Hrsg.): Innovative Antriebe 2018: Der Ausblick auf die Fahrzeugantriebe für die kommenden Dekaden. Düsseldorf: VDI Verlag GmbH, 2018 (VDI Berichte, 2334), S. 33–48 13.
KUNITH, Alexander W.: Elektrifizierung des urbanen öffentlichen Busverkehrs: Technologiebewertung für den kosteneffizienten Betrieb emissionsfreier Bussysteme. Berlin, Technische Universität. Dissertation. URL http://dx.doi.org/ 10.1007/978-3-658-19347-8

599

The impact of pass-by noise legislation on the design of exhaust systems Jan Krüger, Peter Wink, Maike Werner Eberspächer Exhaust Technology GmbH & Co. KG, Esslingen, Germany

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 M. Bargende et al. (Hrsg.), 19. Internationales Stuttgarter Symposium, Proceedings, https://doi.org/10.1007/978-3-658-25939-6_110

601

The impact of pass-by noise legislation on the design of exhaust systems

1 Abstract In all major markets, the legislation on emissions (CO2, NOx, and PM) has been changed drastically and limit values have been lowered over the last years and will be even lower in the near future. Moreover, the new pass-by noise rules require more testing e.g. for additional sound emissions provisions (ASEP) and limit values are more stringent than ever according to R51.03, particularly with the phase coming into force in 2024. Besides the tire noise, the powertrain and here in particular the exhaust tailpipe noise is nowadays the main contributor to pass-by noise. This will have an enormous impact on the design of future passenger cars, their powertrains, and finally their exhaust systems with an inherent cost penalty as e.g. muffler systems need to be larger or more complex to emit even lower noise. The technology for Active Noise Cancellation has been investigated in the automotive industry for many years. The basic principles are well understood and active engine mounts as well as in-cabin active noise cancellation have been in series production for a few years. For exhaust muffler systems, several studies at prototype level have demonstrated technical feasibility and resulted in substantial benefits for acoustical performance and sound design. Moreover, advantages in package space, weight, and backpressure have been demonstrated for different engine set-ups. In this publication, active exhaust systems will be compared to conventional cold-end systems. Their main advantages in acoustics, weight, and package space under different scenarios will be analyzed.

2 Introduction The recent update of the well-known pass-by measurement standard ISO 362 [1] and the introduction of further tests relating to additional sound emission provisions (ASEP) have significantly increased the acoustic requirements for the development of vehicles [2]. Furthermore, new vehicle categories and sub-categories have been introduced and stringent noise levels have come into force in Europe and many other countries around the world. So, for vehicles with up to 120 kW/t, there is a new limit value of 72 dB(A) applicable as of 2016 in the first stage (phase 1), which will be reduced to 70 dB(A) as of 2020 (phase 2), and even reduced to 68 dB(A) as of 2024 (phase 3). Note that a reduction of 4 dB(A) corresponds to a reduction to around 60 % in terms of emitted output – a standard met by only few vehicles today. Moreover, another challenge arises from the tough CO2 targets, which have been agreed upon by the authorities worldwide. Since more stringent noise and CO2 limits are imposed in parallel, this will cause an enormous boost on innovative engine technologies, which will allow massive downsizing with increased power density of > 80

602

The impact of pass-by noise legislation on the design of exhaust systems kW per L displacement for diesel engines and > 100 kW per L displacement for gasoline engines [3, 4, 5, 6]. In today’s automotive business, OEM’s develop the vehicles in platforms to reduce development cost and utilize synergies between similar vehicles. Typical platforms have in common most relevant vehicle and powertrain properties, e.g. longitudinal or transversal mounting of the internal combustion engines (ICE) or limousine or sports utility chassis. Each platform is developed in a limited time frame and in close cooperation between the OEM and exhaust supplier according to the specification of requirements which form together with the commercial aspects the basis of the nomination. While in smaller vehicles (A- and B-segment) engines with 3 and 4 cylinders are common, in larger vehicles (C- and D-segment) typically more engine power is requested. Thus, in the premium platforms today the vehicles are mostly propelled by gasoline and diesel engines with 4 and 6-cylinders.

3 Exhaust Volume Requirements Depending on the OEM requirements the exhaust system layout is determined mainly by the number and total volume of the mufflers. Generally, the muffler volume requirements rise directly with engine power which is traditionally derived from engine displacement. However, due to the charging technology, the power density can vary enormously and thus the relation of muffler volume to the engine power is more appropriate than to the displacement. Furthermore, the backpressure as well as acoustic targets play an important role in the definition of necessary muffler volume. As an example, Figure 1 illustrates the muffler volumes of current passenger vehicles with gasoline engines with different numbers of cylinders which were introduced into the market in the last 5 years. Here, the large distribution reflects different application conditions such as backpressure and acoustic targets. However, a clear trend shows that higher engine power drives larger muffler volumes. Furthermore, it can be deduced that with higher engine power active valves are more common and 70 L form a sort of limit even for large passenger vehicles.

603

The impact of pass-by noise legislation on the design of exhaust systems

Figure 1: Statistical Survey of Current Production Exhaust Systems of Gasoline Engines and their Overall Muffler Volume versus Engine Peak Power

Active valves are often used since they significantly reduce the need for large mufflers by about 20%. They are closed for low mass flow rates and in particular at low engine speeds with inherent low frequency excitation. With only one of two tailpipes open, the noise can be reduced more effectively. At higher mass flow rates, conversely, increasing flow noise requires the opening of the second tailpipe. The active valve technology was introduced some decades ago and the actuator of former models was mainly driven by air pressure. However, in recent years, the design was more and more improved to reduce leakage and production cost. Further innovation was introduced driven by the change to an electric actuator which offers more flexibility including partial opening operating schemes (Figure 2).

604

The impact of pass-by noise legislation on the design of exhaust systems

Figure 2: Eberspächer Active Valve with Electric Actuator

4 General Active System Information The ANC technology is well known for many years for its ability to efficiently reduce low frequency noise. It has been thoroughly investigated particular in the automotive industry with increasing low engine order challenges and applications for the interior cabin of a vehicle [7] or active engine mounts [8] have been introduced in automotive series production in several brands and vehicles. For exhaust systems of ICE it has been found, that the main advantages of the ANC technology can compensate for extra costs which are associated with the additional components such as actuator and electronics [9,10,11]. The potential advantages of active mufflers are as follows: –
high efficiency in reducing dominant engine orders, leading to smaller mufflers, –
simplified development procedures and reduced development costs and time, –
uniform muffler construction leading to more carry-over-parts and avoiding costly tooling, –
limitation of acoustic muffler variants due to slightly different packaging, sound and tailpipe trim variants, –
low backpressure possible resulting in higher engine performance, –
software control of engine orders allowing an adaptive sound design and easy sound customization, –
reduced number of additional mechanical parts e.g. heat shields, exhaust and body side hangers, isolators, –
improved driving dynamics and fuel economy due to lowered exhaust system mass.

605

The impact of pass-by noise legislation on the design of exhaust systems In recent years, the active components (actuators, electronic hardware and software) were consistently improved and their durability professionally validated [12]. For passenger vehicle exhaust systems, the rear muffler is replaced by an actuator which receives its input from an ActiveSilence Control Unit (ASCU). A muffler in front or center position reduces flow noise and engine orders (EO) to a level, which the actuator can cancel (Figure 3).

Figure 3: Typical Set-up of an ActiveSilence® – Exhaust System

5 Passive muffler system of a 150 kW engine compared to an ANC-system In an initial study, a premium vehicle with a 4-cylinder gasoline engine and a peak power of ca. 150 kW was chosen. In this study, the exhaust aftertreatment remained unchanged as well as the hanger positions and ground clearance of the whole exhaust system. The production cold end has a rather complex muffler set-up with an overall volume of ca. 50 L and a total cold end weight of ca. 27 kg to achieve this vehicle’s package constraints, backpressure requirements, and premium sound quality targets at the same time. On the contrary, the active exhaust line for this application, depicted in Figure 4, contains only one muffler and one actuator with a total volume of ca. 23 L. Note, that in this case the reduction of muffler volume is more than 40%, thereby clearly exceeding, what would have been possible to achieve with an active valve

606

The impact of pass-by noise legislation on the design of exhaust systems technology. The most relevant exhaust system data of both cold ends are compared in Table 1. Table 1: Exhaust Data of the Cold-Ends of a 2.0 L 4-Cylinder Gasoline TurboCharged Engine with ca. 150 kW Cold end description Muffler volume (including Actuator) Backpressure Overall weight SPL 2EO @ 1500 rpm (Representing Booming) SPL OVL @ 2500 rpm (Representing Pass-by noise) SPL OVL @ 6000 rpm (Representing Flow noise)

Production system (dual exhaust line)

Active system (single exhaust line)

50 L

23 L

160 mbar 27 kg

160 mbar 19 kg

58 dB(A)

57 dB(A)

77 dB(A)

76 dB(A)

92 dB(A)

91 dB(A)

The acoustic measurements on a roller-dynamometer test bench under full load acceleration were carried out with the microphone located at a distance of 1.0 m centered behind the car and at a height of 1.5 m to allow a fair comparison of the noise radiation from dual tailpipes, with that of the active single exhaust. In Figure 4, the acoustic results are shown for the overall sound pressure level (bold line) and the 2nd engine order (thin line) as it is the firing frequency and thus the most important engine order. At low engine speeds the A-weighted overall SPL of the ANC system is up to 3 dB lower than with the production system due to good cancellation of all relevant EO. Moreover, at high engine speeds the overall SPL is dominated mainly by flow noise and a benefit of 2-3 dB is visible for the ANC system due to the careful CFD-design, an increased pipe diameter, and consequently lower Mach numbers in the pipe. For the 2nd EO the ANC system is superior mainly at low engine speeds up to about 1.500 rpm. Above this engine speed the 2nd EO SPL of the active system is similar with the production system but less relevant for the overall SPL.

607

The impact of pass-by noise legislation on the design of exhaust systems

Figure 4: Measured Overall and 2nd EO SPL at the Tailpipe of the 4-Cylinder Gasoline Engine at full Load (bold lines – OVL; thin lines – 2nd EO)

Another important vehicle and engine condition for an evaluation of the acoustic performance is the idle condition. Consequently, measurements were carried out at ca. 800 rpm with the production system as well as the ANC system after the engine and the exhaust system were shortly warmed up and with air-conditioning activated to increase engine load. It is evident in the measurement results shown in Figure 5, that the overall SPL is > 4 dB(A) lower for the ANC system, but EO SPL differ considerably. At this engine speed, all EO from 2nd to 20th were actively cancelled simultaneously. Even though, the frequency of the 2nd EO is very low, the ANC system performs very well. However, active cancellation of 6th and 20th EO should be analyzed in more detail to allow further optimization in future projects.

608

The impact of pass-by noise legislation on the design of exhaust systems

Figure 5: Measured SPL at ca. 800 rpm in Idle Condition at the Tailpipe of the 4-cylinder Gasoline Engine (grey columns – production system; black columns – ANC system)

6 Active valve with muffler system of a 200 kW engine compared to an ANC-system In a second prototype case study, another premium vehicle with a 4-cylinder gasoline engine and a peak power of ca. 200 kW was investigated. Again, the exhaust aftertreatment production part was used and also the hanger positions of the whole exhaust system were kept as in the production system. The production cold end has a muffler set-up with an overall volume of ca. 43 L and a total cold end weight of ca. 24 kg. It is a single exhaust line with two tailpipes out of the rear muffler. One of the tailpipes is connected to an active valve. However, the layout of the active exhaust for this application contains only one large center muffler and one actuator in the rear with a total acoustic volume of just ca. 29 L. Backpressure and weight are fairly similar to each other. The most relevant exhaust system data of the cold ends are compared in Table 2.

609

The impact of pass-by noise legislation on the design of exhaust systems Table 2: Exhaust Data of the Cold-Ends of a 2.0 L 4-Cylinder Gasoline TurboCharged Engine with ca. 200 kW Cold end description

Production system (dual exhaust line) 43 L + valve 80 mbar 24 kg

Muffler volume Backpressure Overall weight SPL 2EO @ 1500 rpm 57 dB(A) (Representing Booming) SPL OVL @ 2500 rpm 89 dB(A) (Representing Pass-by noise) SPL OVL @ 6000 rpm 97 dB(A) (Representing Flow noise)

Active system (single exhaust line) 29 L incl. actuator 90 mbar 22 kg 60 dB(A) 79 dB(A) 100 dB(A)

Figure 6 shows acoustic results measured on the roller dynamometer at full load and conditions similar to pass-by noise measurements. The conventional exhaust with active valve shows an increase of the overall SPL which reaches its maximum at about 2700 rpm just before the active valve is opened. After the valve opening, the flow noise is decreased but the 2nd EO increases immediately. In contrast, the overall SPL of the active system rises more steadily over engine speed. However, the attenuation of the 2nd EO with the active system is more limited and peaks occur in the SPL curve e.g. at 2000 and 4000 rpm since the single actuator works at its electro-dynamic limits. In contrast, a good noise reduction was achieved below 1400 rpm and about 3000 rpm.

610

The impact of pass-by noise legislation on the design of exhaust systems

Figure 6: Measured Overall and 2nd EO SPL at the Tailpipe of the 4-Cylinder Gasoline Engine at full Load (bold lines – OVL; thin lines – 2nd EO)

Additionally, measurements in idle condition at ca. 740 rpm were carried out with the production system as well as the ANC system after the engine and the exhaust system were shortly warmed up and with air-conditioning activated. In the measurement results depicted in Figure 7, the overall SPL is on a similar level for both systems. For the EO SPL there is an advantage for the active system visible up to the 6th EO. Note that, the benefits of the ANC technology at low frequencies will also support during cold start of the engine and the following phase with catalytic heating which is usually rather prominent in the tailpipe noise.

611

The impact of pass-by noise legislation on the design of exhaust systems

Figure 7: Measured SPL at ca. 740 rpm in Idle Condition at the Tailpipe of the 4-cylinder Gasoline Engine (grey columns – production system; black columns – ANC system)

To investigate the market readiness of the ActiveSilence-Technology in more detail, the durability of the described active exhaust system was verified in a thermomechanic endurance test on an engine test bench. In this test, the engine was run under similar conditions as in a typical full vehicle endurance test on a test track as specified by an OEM with a high percentage of full load conditions and in various engine speeds. After more than 400 cycles and 360 h the exhaust line was still fully operating and the actuator as well as sensor was functioning well according to their specifications. Finally, the vehicle was tested with both exhaust lines for pass-by noise according to R51.03 by a certified German service provider on an ISO test track. The measured pass-by noise levels are shown in Table 3. In both cases, these values yield a finally reported value of Lurban of 71 dB(A), which means that a limit value of 72 dB(A) corresponding to phase 1 of the European limit values for M1 vehicles with a power to mass ratio < 120 kW/t was reached safely. A more detailed investigation of the frequency spectrum of the active system’s pass-by noise revealed a dominant peak at about 350-400 Hz which could later be attributed to the 10th and 12th EO of the exhaust tailpipe noise. Since these EO were not actively cancelled during the test, a further optimization of the software parameter set could result in a substantial reduction of these engine orders (e.g. by 5 dB). With such an improvement, a reduction of passby noise of > 0.5 dB is estimated, which could result in a Lurban of just 70 dB(A). This

612

The impact of pass-by noise legislation on the design of exhaust systems would mean, that even phase 2 (coming into force in 2020) of the European limit values of such a M1 vehicle can be reached. Nevertheless, to fulfil the phase 3 target of just 68 dB(A) it is estimated, that a more combined effort on optimization of several noise sources such as engine, exhaust, and tire noise is necessary. Table 3: Pass-by Noise Level according to R51.03 of a passenger car with a 200 kW Gasoline Engine Pass-by Noise Level

Production system 70.7 dB(A)

Active system 70.9 dB(A)

7 Summary/Conclusions Internal combustion engines will remain the major source of propulsion power in passenger cars for the next decade. Thus, innovative exhaust systems are needed to fulfil the growing demands for lower emission, lower noise, light weight solutions as well as more sound quality. While the active valve technology in exhaust systems as well as active sound technology has already proven their maturity and customer satisfaction in the field, considerable progress has been made in the development of the more challenging ANC technology. Accordingly, the benefits of this technology have been increased and thereby found to compensate more and more the efforts which are associated with the additional components (e.g. loudspeaker, microphone, and electronics). The current status of the ANC technology and the gathered detailed application experience allow relatively easy adaptation to different customer requirements and package situations. Thereby, it is possible to reach the challenging pass-by noise targets of future legislation and decrease at the same time the necessary package space, number of metal parts, tooling cost, validation loops, and development time.

Bibliography 1.
ISO 362-1:2015, „Measurement of noise emitted by accelerating road vehicles -Engineering method - Part 1: M and N categories”. 2.
UN/ECE Regulation R51.03, “Uniform Provisions Concerning the Approval of Motor Vehicles having at least Four Wheels with regard to their Noise Emissions”. 3.
Meier, C., Bikker, S., “CAE-based NVH development of the new 4- and 6Cylinder Diesel engines”, Aachen Acoustic Colloquium 2017, p. 19-25. 4.
Durst, B. et al, „BMW Water Injection: Initial Experience and Future Potentials”, Wiener Motoren-Symposium 2017.

613

The impact of pass-by noise legislation on the design of exhaust systems 5.
Basso, G. et al, „The General Motors 1.6 Liter Midsize Diesel Engine for the North American Market”, Wiener Motoren-Symposium 2017. 6.
Reitz, D., “P2 Hybrid solutions for 48 V”, Hyundai-Kia International Powertrain Conference 2017. 7.
Schirmacher, R.; Kunkel, R.; Burghardt, M., “Active Noise Control for the 4.0 TFSI with Cylinder on Demand Technology in Audi's S-Series” SAE Technical Paper doi: 10.4271/2012-01-1533. 8.
Loheide, St.; et al, “Active Engine Mount,” Aachen Acoustic Colloquium 2012, 167-176. 9.
Garabedian, C.; Zintel, G., “Active Noise Control: Dream or Reality for Passenger Cars,” SAE Technical Paper 2001-01-0003, 2001, doi:10.4271/2001-01-0003. 10.
Bönnen, D.; Bamdad-Soufi, D.; Steinkilberg, H.; Abram, K., “Possibilities and Constraints for Lightweight in Exhaust Systems,” SAE Technical Paper 2014-012058, 2014, doi:10.4271/2014-01-2058. 11.
Anslinger, L., Driot, N., Robin, X., Yeap, K.Z., “Model-based development of an integrated ANC System,” ATZ extra 9/2016, p. 16-20. 12.
Krüger, J.; Koch, V., Buganza, F., „The effects of future noise limit values on the design of exhaust systems,” ATZ – Automobiltechnische Zeitschrift 09 (2016), Volume 118, p. 48-51.

614

Simulative research on the tire torsional vibration and its vehicle relevant influencing factors Dipl.-Ing. Wenrui Han, Yi Guo, Prof. Dr.-Ing. Günther Prokop Lehrstuhl für Kraftfahrzeugtechnik, Technische Universität Dresden Dr. Thomas Roscher AUDI AG

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 M. Bargende et al. (Hrsg.), 19. Internationales Stuttgarter Symposium, Proceedings, https://doi.org/10.1007/978-3-658-25939-6_111

615

Simulative research on the tire torsional vibration and its vehicle relevant influencing …

1 Introduction Tire pressure as an important vehicle parameter has a great impact on the vehicle safety, ride comfort, driving dynamics and fuel economy. Every registered vehicle in the USA needs to be equipped with a tire pressure monitoring system from 2007. This legal requirement has been published in the EU since 2012. The tire pressure monitoring system (TPMS) can be divided into direct and indirect systems. The indirect tire pressure monitoring system (iTPMS) is a cost friendly solution for the tire pressure monitoring. In this system the wheel speed signal has been used for detection of tire pressure loss. In iTPMS 2 algorithms have been used. One algorithm is the comparison of the rolling radius of 4 wheels, the other one is the analysis of the natural frequency of wheel speed signal. For 1-3 wheels pressure loss scenario (puncture) the both algorithms can be used. For the 4 wheels pressure loss scenario (diffusion) just the second algorithm can be used, because in this case there is no significant difference of the rolling radius between four wheels. At pressure loss the natural frequency of the wheel speed signal in a frequency band of 30-60 Hz will decrease, as figure 1 shows. This pressure-dependent natural frequency can be used to detect the four wheels pressure loss.

Figure 1: Frequency spectrum of wheel speed at 2 pressures

Tire pressure is one of the most important influencing factors for the pressure-dependent eigenmode. However, this eigenmode can also be influenced by other factors such as suspension components. The iTPMS is based on the wheel speed signal, which can represent the tire torsional vibration. In order to get a better understanding of the influence of suspension on the tire torsional vibration and realize an early predication of the iTPMS performance in

616

Simulative research on the tire torsional vibration and its vehicle relevant influencing … the concept phase. The influencing factors can be investigated via simulation. The models, which have been used in this investigation, will be shown in the next sections.

2 Research methods Although the tire torsional vibration is the important information for the iTPMS, this information is not widely used in the analysis of vehicle dynamics. Therefore, related research is less common in the literature. For the purpose of getting a better understanding of iTPMS and its influencing factors, systematic research methods of the tire torsional vibration need to be developed.

2.1 State of the art A suitable tire model is an essential prerequisite for the investigation of tire torsional vibration. A rigid ring model is a classical tire model, which has been developed by ZEGELAAR [1]. This model can be used to describe the iTPMS-related vibrations of a tire. In the [2] VOS used the rigid ring model to investigate the change of the tire torsional vibration due to the interaction between tire and suspension. However, the parameterization and validation of this model were not precisely described. In the [3] SUENDER proved the correlation between the tire rotational vibration and longitudinal vibration by using a simplified rigid ring tire model and various tests. A test procedure on a tire test rig [4] was developed in [5] in order to estimate the tire performance with iTPMS. The research of tire torsional vibration by using a rigid ring tire model can be done in depth. In addition, the influences of suspension and full vehicle need to be proved in the simulation and real-world tests. These investigations will be executed in this research.

2.2 Methods of investigation The aim of the research on the tire torsional vibration is to optimize the iTPMS. The tire vibration behavior can be influenced due to the interaction between suspension and tire. The influences of the suspension and the vehicle need to be considered in this investigation. For this reason, the research should be divided into three levels, see figure 2: component level, subsystem level and full vehicle level. As mentioned above, the rigid ring tire model is a suitable model for the investigation of the tire torsional vibration. An axle model and a full vehicle model can be created based on the rigid ring tire model. With these simplified models which have a limited number of parameters, some analyses such as a sensitivity analysis or a modal analysis

617

Simulative research on the tire torsional vibration and its vehicle relevant influencing … can be easily made. However, for a dynamic analysis is a more complex model needed. A more complex model can better describe some dynamic properties like the dynamic stiffness and nonlinearity of rubber components. In this case the complex multibody model is suitable. The vehicle test and rig test are definitely necessary for the whole research. On the one hand, some influences of suspension components can be directly tested and proved with the help of tests. On the other hand, they can be used to validate the models. A combination of the simulation and the tests is meaningful for the investigation of iTPMS.

Figure 2: Overview of the tools for investigation

The focus of this work is on the simplified models. The simplified models of component level and subsystem level will be introduced and analyzed in the next sections.

3 Introduction of simplified model A rigid ring tire model is the basic of the simplified model. As mentioned in the literature review, the rigid ring tire model can describe the iTPMS-related tire mode. In this section the rigid ring tire model and an axle model will be shown. Due to the complexity of the vehicle model, the full vehicle model will not be described in this article.

3.1 Rigid ring tire model The rigid ring tire model comes from ZEGELAAR [1]. This model is used to simulate the tire vibration on a tire test rig. Therefore, some simplifications of the tire model have been made in order to reduce the complexity of this model. The difference between the tire model in this work and the rigid ring tire model from [1] is that the rim just has

618

Simulative research on the tire torsional vibration and its vehicle relevant influencing … one degree of freedom of rotation, because the wheel is fixed on the rig during a test. This model has one more degree of freedom in vertical direction compared to the simplified rigid ring model, which was used in [6].

3.1.1 Description of the model The structure of the rigid ring tire model with a contact model is illustrated in figure 3. The differential equations can be seen below the figure.

Figure 3: Rigid ring tire model

  $ 

  %#  ! #  &

  % !  & " 

(1)

  $ 

  # 

   !    !      " 

(2)

  $

  #

% 

(3)

 &  

     " 

  $  !   %#  ! #  & !   % !  &

     " 

(4)

Contact model '

 

  (  #



    " ! ! #    

(5)

 "   

(6)

 " #  !   # 

(7)

The rigid ring model consists of following parameters: mass of tire belt ( ), moment of inertia of rim (
 ) and of tire belt (
 ), tire longitudinal stiffness and damping ( ,  ), tire vertical stiffness and damping ( ,  ), tire torsional stiffness and damping ( ,

619

Simulative research on the tire torsional vibration and its vehicle relevant influencing …  ), wheel rotational velocity (). It should be noted that the tire stiffness and damping in longitudinal and vertical directions are considered to be equal. The contact model has these parameters: longitudinal tread stiffness and damping (
,  ), slip stiffness (
 ), longitudinal slip ( ) and relaxation length ( ).  is the longitudinal velocity and  is the longitudinal slip velocity.   is the dynamic rolling radius. The rigid ring tire model has 4 degrees of freedom: rotation of rim (  ) and belt ( ), translation of belt in longitudinal and vertical directions ( ,  ). The slip in the contact model can be described with the degree of freedom of the rigid ring model.

3.1.2 Analysis with tire model This physical tire model is a useful aid for the investigation of the correlation between the tire vibration behavior and the physical parameters. 2 Examples of using the tire model will be shown below. The parameters for this analysis is from [3]. They do not need to correspond to a certain tire for these qualitative analyses. The influence of each parameter on the tire vibration behavior can be calculated with the help of this model. For example, figure 4 shows that, an increasing moment of inertia of the rim leads to a decrease of the natural frequency of the rotational mode. This phenomenon is similar to what happens when a tire pressure loss occurs. The amplitude becomes smaller, which is contrary to the phenomenon of pressure loss. The different influences on the tire vibration behavior can be clarified by using this model.

Figure 4: Frequency spectrum of wheel speed with different JF

After solving the differential equations, the eigenmodes of the rigid ring model can be calculated. This system has 4 eigenmodes, the first mode is the in-phase rotational mode.

620

Simulative research on the tire torsional vibration and its vehicle relevant influencing … The influences of all parameters on this mode can be analyzed by a sensitivity analysis. Figure 5 presents the result of a sensitivity analysis by changing of each parameter by +/- 30%. The main influencing factors can be derived from this result. Some conclusions need to be extracted in consideration with the change of these parameters under actual conditions. For example, the change of the rolling radius by 30% is too high in reality.

Figure 5: Sensitivity analysis of rigid ring tire model

3.2 Axle model based on rigid ring tire model In this section, the front and rear axle models based on the rigid ring tire model will be introduced. The left side of the figure 6 shows the front axle model. This model has six degrees of freedom: the rotation of axle and belt, the translation of axle and belt in longitudinal and vertical directions. This model has two more degrees of freedom compared to the axle mode, which was used in [7]. The following parameters are additionally defined compared to the rigid ring model to describe the front axle model: the longitudinal and vertical stiffness and damping ( ,  ,  ,  ) of the axle. The moment of inertia and mass of rim (
 ,  ) are replaced by the moment of inertia and mass of axle (
, ).

621

Simulative research on the tire torsional vibration and its vehicle relevant influencing …

Figure 6: Front axle (left) and rear axle (right) based on rigid ring tire model

Below are the equations of the free vibration system of the front axle model:   $  !   %# ! #  & !   % !  & !    !     % !  & " 

(8)

  $ !   %# ! # & !   % !  &

    % ! & " 

(9)

     " 

(10)     % !  & "  (11)

  $  !   %#  ! #  & !   % !  &   $

%

 &  #

% 

 &   !   #  !   

  $

%

 &  #

% 

 &   !   # !    !     % ! & " 

  $ 

  %#  ! #  &

  % !  & " 

(12) (13)

The equation of the contact model has been shown in the rigid ring model. The right side of the figure 6 shows the rear axle model. Due to a former research result in [8], the elastic connection between the rear subframe and the vehicle body, which can influence the tire torsional vibration, needs to be considered in order to get a better simulation results with the simplified model. The subframe of rear axle is connected with the vehicle body via two rubber bushings and two hydro bushings. One rubber bushing and one hydro bushing are modelled in the ½ rear axle model. The rubber bushing is regarded as a Kelvin–Voigt model. A model from ANGRICK [9] is applied for the hydro bushing. This rear axle model has three extra degrees of freedom compared with the front axle model: the translation of subframe in longitudinal and vertical directions ( ,  ) and the translation of fluid in the hydro bushing in longitudinal direction ( ). Other than the above defined parameters for the front axle, the following parameters are additionally defined for the rear axle model: subframe mass ( ), rubber bushing longitudinal and vertical stiffness and damping (
 ,  ,
 ,  ), longitudinal and vertical basis stiffness and damping of hydro bushing (
 ,  ,
 ,  ),

622

Simulative research on the tire torsional vibration and its vehicle relevant influencing … equivalent longitudinal stiffness and damping of fluid in hydro bushing ( ,  ), and equivalent mass of fluid in hydro bushing ( ). Due to the complexity of the rear axle model the equations and the analysis of the rear axle model will not be presented in this paper.

4 Parameterization of the axle model To obtain valid results from the simulation, the parameterization for the axle model needs to be carried out. This section will focus on the parameterization for the front axle model. The front axle model is required to be parameterized: the moment of inertia and the mass (
, ) of axle, the axle longitudinal and vertical stiffness and damping ( ,  ,  ,  ). The axle mass indicates the vibrating mass of the half axle including the rim, half shaft, links, etc., which could be directly weighed. The moment of inertia of axle refers to the rotational parts in front axle, which could be measured with a torsional pendulum. The longitudinal and vertical stiffness and damping of the front axle could be measured on a road-load-simulator at the Technical University of Dresden, which was introduced in [10]. The axle will be connected to the test rig with the mounting bolts at the damper and the subframe bushings. The hydraulic actuators are jointed with the rim of front axle. The actuators can move in longitudinal and vertical directions. The wheel center displacement and force can be measured and used to calculate the longitudinal and vertical stiffness and damping. Figure 7 shows an example for the calculation of vertical stiffness in the front axle.

Figure 7: Wheel center displacement and force from measurement

623

Simulative research on the tire torsional vibration and its vehicle relevant influencing … Table 1 shows the parameters of a test axle, which has been measured on the road-loadsimulator. Table 1. Parameters of the front axle model (tire model parameters not inclusive) Symbol     


Description

Value

Unit

Axle longitudinal stiffness Axle vertical stiffness Axle longitudinal damping Axle vertical damping Axle vibrating mass Axle rotational inertia

755.6 42.0 3000 2500 34 1.6

N/mm N/mm Ns/m Ns/m kg kg.m2

5 Analysis with axle model The eigenmode of the front axle model can be analyzed with the parameters of axle from measurements above and the parameters of rigid ring model from [3]. The first eigenmode of the axle model has a natural frequency of 10.67 Hz. It can be seen from the phase information that the movement of axle and tire in the vertical direction  and  participate mainly in this mode. According to the magnitude information of the eigenvector, the magnitude of the vertical movement of the axle is slightly higher (about 10%) than the vertical movement of the belt (see figure 8). The motions of  and  are also in phase.

Figure 8: Vibration analysis of the front axle model - 1. natural frequency

624

Simulative research on the tire torsional vibration and its vehicle relevant influencing … The second eigenmode at 15.92 Hz represents the in-phase vibration of the four degrees of freedom
 ,  ,
 and  (see figure 9). This means that the vibrations of longitudinal movement and rotation of the tire are in phase with the longitudinal movement and rotation of the axle. The magnitude information presents a primary role of the rotational degrees of freedom in this mode. Additionally, both the rotations and the longitudinal movements have the comparable magnitudes at the rim and belt.

Figure 9: Vibration analysis of the front axle model - 2. natural frequency

The third eigenmode corresponds to the in-phase rotational mode, which has a natural frequency of 41.54 Hz. According to the phase information, adding the axle stiffness to the tire model does not affect the characteristic of the tire vibration. The degrees of freedom ,
 and  continue to remain in phase. The strong coupling between the tire longitudinal translation and the wheel rotation is also maintained. It can be seen from the magnitude information, that the magnitude of the translational movement of the axle is greater than that of the belt. The magnitude of axle rotation is also significantly higher compared to the tire belt (see figure 10).

Figure 10: Vibration analysis of the front axle model - 3. natural frequency

625

Simulative research on the tire torsional vibration and its vehicle relevant influencing … The fourth eigenmode of the axle model has a natural frequency of 84.05 Hz. This eigenmode represents also a vertical vibration mode according to the magnitude information (see figure 11). The vertical degrees of freedom of axle
 and tire  are antiphase with each other in this mode. From the magnitude it can be seen that the vertical movement of tire belt  is decisive in this mode.

Figure 11: Vibration analysis of the front axle model - 4. natural frequency

The analysis of the fifth and sixth natural vibrations in figure 12 and figure 13 shows a high correlation to the anti-phase rotational modes (the third and the fourth eigenmodes of rigid ring tire model). The longitudinal movement of axle  shows an anti-phase relationship with the longitudinal tire movement  . The magnitude of  in the two eigenmodes are imperceptible.

Figure 12: Vibration analysis of the front axle model - 5. natural frequency

626

Simulative research on the tire torsional vibration and its vehicle relevant influencing …

Figure 13: Vibration analysis of the front axle model - 6. natural frequency

It could be concluded from the eigenmode analysis, that the four original vibration modes of tire model can still be found in the front axle mode. The phase relationship between ,
 and  stay unchanged in the rigid ring model and axle model. The natural frequency of the in-phase rotation shifts significantly to a higher frequency from 34.41 Hz (the first natural frequency of rigid ring tire model) to 41.54 Hz. A coupling between the vertical and longitudinal movements is negligible. Figure 14 shows a cleat test result (longitudinal force) of a front axle on a suspension test rig at 20 kph in frequency domain. Two vibration frequencies could be readily observed at about 15 Hz and 37 Hz, which present an accordance with the second and third eigenmodes of axle model. The second eigenmode of 15 Hz which can not be found in the eigenmodes of rigid ring model appears after adding the axle components. It can be proved that the 15 Hz vibration is from axle. The vibration frequency at about 90 Hz in the spectrum is also identifiable and corresponds to the fifth natural frequency. The comparison between the results of simulation and measurement demonstrates that the front axle model can indicate the eigenmodes of a real axle.

Figure 14: Frequency spectrum of longitudinal force of a front axle on a test rig

627

Simulative research on the tire torsional vibration and its vehicle relevant influencing … Figure 15 presents the sensitivity analysis for the in-phase rotational mode with a parameter variation of
30%. Axle mass , tire longitudinal stiffness  , dynamic rolling radius   and moment of inertia of axle
 present the greatest influences on the natural frequency. The mass and moment of inertia of front axle are normally invariable. The influence of axle longitudinal stiffness on the natural frequency is comparatively not significant. A conclusion can therefore be reached that the axle parameters have a certain influence on the natural frequency of the in-phase rotational mode.

Figure 15: Sensitivity analysis of front axle model

6 Conclusion The tire torsional vibration as an important tire property for the iTPMS has been investigated with the simplified models in this work. It has been proved that a rigid ring tire model can be used to present the iTPMS-related tire vibration behavior and to analyse the correlation between tire parameters and tire vibration. For the research of the influence of suspension on the tire torsional vibration the rigid ring model has been extended to an axle model. After the parameterization of the axle model a modal analysis of the axle model has been executed. The eigenmodes from simulation correspond to the natural frequencies which were found from frequency spectrum. With the help of the axle model, the physical meaning of the vibrations which can be found from the frequency spectrum can be clarified. The sensitivity analysis of the axle model shows that the axle parameters such as the axle mass, moment of inertia and stiffness have a certain influence on the iTPMS-related natural frequency and adding the axle components leads to a shift of the iTPMS-related natural frequency.

628

Simulative research on the tire torsional vibration and its vehicle relevant influencing …

Bibliography 1.
ZEGELAAR, P.: The dynamic response of tires to brake torque variations and road unevenesses, PhD Thesis, Delft University of Technology, 1998 2.
VOS, S.: Eine indirekte Methode zur Identifikation fahrsicherheits- und komfortrelevanter Reifenkenngrößen, Reihe 12, Nr. 551, Düsseldorf, VDI Verlag, 2003 3.
SÜNDER, R.: Entwicklung eines optimierten Verfahrens zur Reifenabsicherung für indirekte Reifendruckkontrollsystem, Dissertation, Technische Universität Dresden, 2017 4.
SARKISOV, P., KUBENZ, J., VAN PUTTEN, S., PROKOP, G.: Steifer Reifenprüfstand für höhere Messgenauigkeit, Automobiltechnische Zeitschrift (ATZ) 04/2015 S. 46-51 5.
ROSCHER, T., SÜNDER, R.: Tire Classification Project – A Rig Procedure for Evaluating Tire Compatibility with Indirect TPMS, 8th Annual Conference Intelligent Tire Technology, Darmstadt, 2012. 6.
PROKOP, G., SÜNDER, R., ROSCHER, T.: Evaluation of Tyre Properties regarding iTPMS in Measurement and Simulation, 4th International Tyre Colloquium, Guildford, U.K., 2015 7.
SÜNDER, R., PROKOP, G., ROSCHER, T.: Comparative Analysis of Tire Evaluation Methods for an Indirect Tire Pressure Monitoring System (iTPMS)" – SAE Int. J. Passeng. Cars - Mech. Syst., 2015 8.
HAN, W.: Untersuchung des Einflusses der Elastizität von Fahrwerkskomponenten auf das Reifenschwingverhalten mittels MKS, WKM Symposium, Darmstadt, 2018 9.
ANGRICK, C.: Subsystemmethodik für die Auslegung des niederfrequenten Schwingungskomforts von PKW, Dissertation, Technische Universität Dresden, 2017 10.
BINDAUF, A., ANGRICK, C., PROKOP, G.: Fahrwerkscharakterisierung an einem hochdynamischen Achsprüfstand, Automobiltechnische Zeitschrift (ATZ), 12/2014, S. 76-81.

629