Automotive Acoustics Conference 2017: 4. Internationale ATZ-Fachtagung Fahrzeugakustik [1. Aufl.] 978-3-658-20250-7;978-3-658-20251-4

Table of contents :
Front Matter ....Pages I-XIII
Challenges of achieving better NVH performance for Chinese domestic brands (Perry P. Gu, Fei Xiong, Hailan Zhao, Xie Kai)....Pages 1-20
Low- and high-frequency NVH CAE – test methods for development of a lightweight sedan design (Yuksel Gur, Jian Pan, David A. Wagner)....Pages 21-41
Vibration reduction in automotive applications based on the damping effect of granular material (Sebastian Koch, Fabian Duvigneau, Sascha Duczek, Elmar Woschke)....Pages 43-57
State-of-the-art digital road noise cancellation by Harman (Nikos Zafeiropoulos, Jürgen Zollner, Vasudev Kandade Rajan)....Pages 59-74
Acura NSX – Hybrid Acoustics, a new carpet concept for meeting stringent performance and weight targets (Jared Cox, Steve Eich, Andrea Martin, Rolf Balte)....Pages 75-89
Powertrain mounting systems for electric vehicles (David Roth, T. Ehrt, C. Bultel, H. Kardoes)....Pages 91-106
Noise radiated by electric motors – simulation process and overview of the optimization approaches (Jean-Baptiste Dupont, Henri Saucy)....Pages 107-121
New fuel-saving technologies and NVH refinement of powertrains (Léon Gavric)....Pages 123-139
Diesel engine control based on structure-borne noise – optimization and adaptation of parameters (Sebastian Schneider, Jan Hendrik Carstens, Jürgen Nobis, Hermann Rottengruber, Clemens Gühmann, Enrico Neumann et al.)....Pages 141-170
Combustion mechanical noise breakdown – turbocharger noise identification on a V8 engine (Karl Janssens, Fabio Bianciardi, Konstantinos Gryllias, Simone Delvecchio, Claudio Manna)....Pages 171-187
Simulation of exterior noise propagation for the acoustic load estimation of airborne model (M. Danti, G. Bartolozzi, M. Meneguzzo, C. Campagna)....Pages 189-205
On the role of simulation in accounting for the design complexity of engine encapsulation (R. D’Amico, R. Stelzer, J. Grebert, P. Chandler, G. Fossaert)....Pages 207-222
Pass-by noise synthesis from frequency domain exterior acoustic simulations (Alexis Talbot, Gregory Lielens)....Pages 223-236
An exploration study of automotive sound package performance in the mid-frequency range (Nicolas Schaefer, Bart Bergen, Tomas Keppens, Wim Desmet)....Pages 237-248
The noise reduction potential of lightweight acoustic metamaterials – a numerical and experimental study (Peter Schrader, Fabian Duvigneau, Hermann Rottengruber, Ulrich Gabbert)....Pages 249-272
Vibration damping behavior of flexible polyurethane foams under low and high strain regimes (Mark Brennan, Martino Dossi, Yueqi Wang, Maarten Moesen, Jan Vandenbroeck)....Pages 273-291
Predicting pass-by noise levels for trucks based on component test bench measurements – by using virtual assembly techniques (Patrick Corbeels, P. Van de Ponseele, M. Choukri, R. Sinnig, W. Kerres)....Pages 293-311
NVH development strategies for suspensions – challenges and chances by autonomous driving (Andreas Schilp, Hartmut Bathelt)....Pages 313-340
A framework for the sensitivity analysis of transfer paths combining contribution analysis and response modification analysis (Dejan Arsić, Matthias Pohl)....Pages 341-351
Tagungsbericht (Matthias Heerwagen)....Pages 353-356

Citation preview

Proceedings

Wolfgang Siebenpfeiffer Hrsg.

Automotive Acoustics Conference 2017 4. Internationale ATZ-Fachtagung Fahrzeugakustik

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

Wolfgang Siebenpfeiffer (Hrsg.)

Automotive Acoustics Conference 2017 4. Internationale ATZ-Fachtagung Fahrzeugakustik

Hrsg. Wolfgang Siebenpfeiffer Stuttgart, Deutschland

ISSN 2198-7432 ISSN 2198-7440 (electronic) Proceedings ISBN 978-3-658-20250-7 ISBN 978-3-658-20251-4 (eBook) https://doi.org/10.1007/978-3-658-20251-4 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

Editorial

Tightening regulations on fuel consumption, CO2 emissions, and pass-by noise are pressing toward more environmentally-friendly and energy-efficient vehicles. Accordingly, the environmental impact generated by vehicles and their NVH performance is in the spotlight more than ever. Furthermore, megatrends like powertrain electrification, autonomous driving, and connected vehicles are becoming increasingly important and are posing engineering challenges that will have to be addressed in order to ensure the necessary NVH comfort in the future. With the Automotive Acoustics Conference, Autoneum and ATZlive provide an essential global forum for debating these trends and technological challenges that are moving the industry. Also, the twoday conference provides opportunities for personal interaction and networking within an international community of experts. Starting with keynote lectures made by leading industry executives, the 4th Automotive Acoustics Conference will cover its customary broad spectrum of themes, giving participants a comprehensive, bi-annual snapshot of the state of the art in acoustics and NVH. In addition to 20 presentations, the 2017 program also features two parallel workshops aiming at the latest trends in the automotive industry: NVH Development of xEV vehicles, Trends and Opportunities in High Performance Composites for Automotive Applications. The 2017 edition will take place

V

VI

Editorial

at the renowned Swiss thinktank Gottlieb Duttweiler Institute (GDI) in Zurich/ Ruschlikon and remains under the patronage of Professor Dr. Paolo Ermanni of ETH. We look forward to meeting you in Zurich in July 2017! On behalf of the Scientific Advisory Board Dr. Davide Caprioli Head of Product Acoustic and Thermal Performance Autoneum Management Wolfgang Siebenpfeiffer Editor-in-Charge ATZ | MTZ | ATZelektronik

Inhaltsverzeichnis

Challenges of achieving better NVH performance for Chinese domestic brands Perry P. Gu, Fei Xiong, Hailan Zhao und Xie Kai Low- and high-frequency NVH CAE – test methods for development of a lightweight sedan design Dr. Yuksel Gur, Jian Pan und David A. Wagner Vibration reduction in automotive applications based on the damping effect of granular material Sebastian Koch, Fabian Duvigneau, Sascha Duczek und Elmar Woschke State-of-the-art digital road noise cancellation by Harman Dr. Nikos Zafeiropoulos, Jürgen Zollner und Dr. Vasudev Kandade Rajan Acura NSX – Hybrid Acoustics, a new carpet concept for meeting stringent performance and weight targets Jared Cox, Steve Eich, Andrea Martin und Rolf Balte Powertrain mounting systems for electric vehicles David Roth, Dr. T. Ehrt, C. Bultel und H. Kardoes Noise radiated by electric motors – simulation process and overview of the optimization approaches Dr. Jean-Baptiste Dupont und Henri Saucy New fuel-saving technologies and NVH refinement of powertrains Dr. Léon Gavric

VII

VIII

Inhaltsverzeichnis

Diesel engine control based on structure-borne noise – optimization and adaptation of parameters Sebastian Schneider, Dr. Jan Hendrik Carstens, Jürgen Nobis, Prof. Dr. Hermann Rottengruber, Prof. Dr. Clemens Gühmann, Enrico Neumann und Michael Joerres Combustion mechanical noise breakdown – turbocharger noise identification on a V8 engine Karl Janssens, Fabio Bianciardi, Konstantinos Gryllias, Simone Delvecchio und Claudio Manna Simulation of exterior noise propagation for the acoustic load estimation of airborne model Marco Danti, Dr. G. Bartolozzi, M. Meneguzzo und Dr. C. Campagna On the role of simulation in accounting for the design complexity of engine encapsulation Dr. Roberto D‘Amico, R. Stelzer, J. Grebert, P. Chandler und G. Fossaert Pass-by noise synthesis from frequency domain exterior acoustic simulations Alexis Talbot und Gregory Lielens An exploration study of automotive sound package performance in the mid-frequency range Nicolas Schaefer, Bart Bergen, Tomas Keppens und Prof. Wim Desmet The noise reduction potential of lightweight acoustic metamaterials – a numerical and experimental study Peter Schrader, Fabian Duvigneau, Prof. Dr. Hermann Rottengruber und Prof. Dr. Dr. Ulrich Gabbert Vibration damping behavior of flexible polyurethane foams under low and high strain regimes Mark Brennan, Martino Dossi, Yueqi Wang, Maarten Moesen und Jan Vandenbroeck Predicting pass-by noise levels for trucks based on component test bench measurements – by using virtual assembly techniques Patrick Corbeels, Dr. P. Van de Ponseele, M. Choukri, R. Sinnig und Dr. W. Kerres NVH development strategies for suspensions – challenges and chances by autonomous driving Andreas Schilp und Prof. Dr. Hartmut Bathelt

Inhaltsverzeichnis

IX

A framework for the sensitivity analysis of transfer paths combining contribution analysis and response modification analysis Dr. Dejan Arsić und Matthias Pohl Tagungsbericht Mathias Heerwagen

Autorenverzeichnis

Dr. Dejan Arsić Müller-BBM VibroAkustik Systeme GmbH, Planegg, Deutschland Rolf Balte UGN, Inc., Novi, USA Dr. G. Bartolozzi Centro Ricerche Fiat S.C.p.A., Orbassano, Italien Prof. Dr. Hartmut Bathelt AZL Technology Center GmbH, Lenting, Deutschland Bart Bergen Toyota Motor Europe, Zaventem, Belgien Fabio Bianciardi Siemens Industry Software NV, Ferrara, Belgien Mark Brennan Huntsman Polyurethanes, Everberg, Belgien C. Bultel Vibracoustic GmbH & Co. KG, Weinheim, Deutschland Dr. C. Campagna Fiat Chrysler Automobiles Group, Turin, Italien Dr. Jan Hendrik Carstens TU Berlin, Berlin, Deutschland P. Chandler Jaguar Land Rover Limited, Warwick, England M. Choukri Siemens Industry Software NV, Leuven, Belgien Patrick Corbeels Siemens Industry Software NV, Leuven, Belgien Jared Cox Honda R&D Americas, Inc., Raymond, USA Dr. Roberto D‘Amico Autoneum Management AG, Winterthur, Schweiz Marco Danti Centro Ricerche Fiat S.C.p.A., Orbassano, Italien Simone Delvecchio Siemens Industry Software NV, Ferrara, Belgien Prof. Wim Desmet KU Leuven, Leuven, Belgien XI

XII

Autorenverzeichnis

Martino Dossi Huntsman Polyurethanes, Everberg, Belgien Sascha Duczek Otto von Guericke Universität, Magdeburg, Deutschland Dr. Jean-Baptiste Dupont VibraTec, Ecully, Frankreich Fabian Duvigneau Otto von Guericke Universität, Magdeburg, Deutschland Dr. T. Ehrt Vibracoustic GmbH & Co. KG, Weinheim, Deutschland Steve Eich Honda R&D Americas, Inc., Raymond, USA G. Fossaert Jaguar Land Rover Limited, Warwick, England Prof. Dr. Dr. Ulrich Gabbert Otto von Guericke Universität, Magdeburg, Deutschland Dr. Léon Gavric PSA Groupe, La Garenne-Colombes, Frankreich J. Grebert Autoneum Management AG, Winterthur, Schweiz Konstantinos Gryllias KU Leuven, Leuven, Belgien Perry P. Gu Geely Automobile Research Institute, Zhejiang, China Prof. Dr. Clemens Gühmann TU Berlin, Berlin, Deutschland Dr. Yuksel Gur Ford Motor Company,Dearborn, USA Matthias Heerwagen Springer Fachmedien Wiesbaden GmbH, Wiesbaden, Deutschland Karl Janssens Siemens Industry Software NV, Ferrara, Belgien Michael Joerres Ford-Werke GmbH, Köln, Deutschland Xie Kai Geely Automobile Research Institute, Zhejiang, China H. Kardoes Vibracoustic GmbH & Co. KG, Weinheim, Deutschland Tomas Keppens Toyota Motor Europe, Zaventem, Belgien Dr. W. Kerres Daimler AG, Stuttgart, Deutschland Sebastian Koch Otto von Guericke Universität, Magdeburg, Deutschland Gregory Lielens MSC Software Belgium SA, Mont-Saint-Guibert, Belgien Claudio Manna Ferrari S.p.A, Modena, Italien Andrea Martin Honda R&D Americas, Inc., Raymond, USA M. Meneguzzo Centro Ricerche Fiat S.C.p.A., Orbassano, Italien Maarten Moesen Huntsman Polyurethanes, Everberg, Belgien

Autorenverzeichnis

XIII

Enrico Neumann IAV GmbH, Berlin, Deutschland Jürgen Nobis IAV GmbH, Berlin, Deutschland Jian Pan Autoneum North America, Inc., Novi, USA Matthias Pohl Müller-BBM VibroAkustik Systeme GmbH, Planegg, Deutschland Dr. P. Van de Ponseele Siemens Industry Software NV, Leuven, Belgien Dr. Vasudev Kandade Rajan Harman Becker Automotive Systems GmbH, Straubing, Deutschland David Roth Vibracoustic GmbH & Co. KG, Weinheim, Deutschland Prof. Dr. Hermann Rottengruber Otto von Guericke Universität, Magdeburg, Deutschland Henri Saucy VibraTec, Ecully, Frankreich Nicolas Schaefer Toyota Motor Europe, Zaventem, Belgien Andreas Schilp AZL Technology Center GmbH, Lenting, Deutschland Sebastian Schneider Otto von Guericke Universität, Magdeburg, Deutschland Peter Schrader Otto von Guericke Universität, Magdeburg, Deutschland R. Sinnig Daimler AG, Stuttgart, Deutschland R. Stelzer Autoneum Management AG, Winterthur, Schweiz Alexis Talbot MSC Software Belgium SA, Mont-Saint-Guibert, Belgien Jan Vandenbroeck Huntsman Polyurethanes, Everberg, Belgien Bart Verrecas Siemens Industry Software NV, Ferrara, Belgien David A.Wagner Ford Motor Company, Dearborn, USA Yueqi Wang Huntsman Polyurethanes, Everberg, Belgien Elmar Woschke Otto von Guericke Universität, Magdeburg, Deutschland Fei Xiong Geely Automobile Research Institute, Zhejiang, China Dr. Nikos Zafeiropoulos Harman Becker Automotive Systems GmbH, Straubing, Deutschland Hailan Zhao Geely Automobile Research Institute, Zhejiang, China Jürgen Zollner Harman Becker Automotive Systems GmbH, Straubing, Deutschland

Challenges of achieving better NVH performance for Chinese domestic brands Perry P. Gu, Fei Xiong, Hailan Zhao, and Xie Kai Geely Automobile Research Institute China

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 W. Siebenpfeiffer (Hrsg.), Automotive Acoustics Conference 2017, Proceedings, https://doi.org/10.1007/978-3-658-20251-4_1

1

Challenges of achieving better NVH performance for Chinese domestic brands

ABSTRACT The NVH performance of a vehicle largely influence customer’s perception and satisfaction of vehicle perceived quality. Vehicle manufacturers have spent significant amount of resources to optimize vehicle NVH to meet the increasing demand of consumers. Recently, the industry has become more aggressive in its lightweighting actions, and the rate of introducing new technology to improve customer’s experience is accelerating. The trend of lightweighting, vehicle electrification, V2X connectivity, intelligence, and autonomous driving as well as eco-design for customer health and environment protection has placed several challenges for vehicle NVH development. First, vehicle weight reduction is a well-known strategy for improving fuel consumption and reducing CO2 emission in vehicles. However, reducing vehicle weight has inevitable consequences for the strength and stiffness of the systems and structures affected. If not implemented carefully, lightweighting strategies can lead to NVH issues that are potentially difficult to remedy because these issues could arise from fundamental design of the vehicle body structure and complex system interactions. Secondly, the combination of lightweighting, electrification, increased demand for advanced driver-assistance system (ADAS), connectivity contents, and faster product cadences has affected overall vehicle cost amortization. This cost amortization limits NVH strategies for noise source design, structural solidity and sound package content. Thirdly, for Chinese car manufacturers, pursuing better NVH performance has become a very important competitive focus in China market to establish brand images and compete with well-known global brands. This paper presents NVH development strategies and techniques to combat these challenges.

2

Challenges of achieving better NVH performance for Chinese domestic brands

INTRODUCTION The NVH performance of a vehicle largely influence customer’s perception and satisfaction of the vehicle perceived quality [1-2]. Vehicle manufacturers have spent significant amount of resources to optimize vehicle NVH to meet the increasing demand of consumers. Chinese domestic auto OEMs, in the past ten years, have recognized the importance of NVH, affecting customer’s perception and satisfaction. These OEMs have gradually invested more resources in vehicle NVH development to improve their vehicle quality through training personnel, establishing NVH facility, developing NVH testing and CAE capability. Chinese cars were criticized in the past for their lack of finesse in term of quality。 Figure 1 shows the JD Power PP100 data from 2000 to 2016. As shown in the figure, the gap between global brands and Chinese domestic brands from 101 at 2015 is reduced to 32 in 2016. In the last five years, Chinese domestic brands have narrow the gap significantly. The NVH benchmarking data of imported and JV cars has shown that the NVH performance of major Chinese domestic brands such as Great Wall, ChangAn, Geely, and GuangQi,have reached or exceeded the levels of non-domestic brands for certain vehicle segments.

Figure 1. PP100 data of JD Power for comparison of Global brands and Chinese domestic brands from year of 2000 to 2016。

3

Challenges of achieving better NVH performance for Chinese domestic brands

In the new phase of the Chinese car industry entering, Geely as one of quick growing automakers is going to break out of its home market. Geely‘s cars now not only have a common, and quite pleasing design, but also better perceive quality. Figure 2 shows the vehicle sale volumes of Geely Automotive Company over the past ten years. The company targets between one million and 1.2 million units in 2017.

Figure 2. The vehicle sale units of Geely Automotive Company over the last ten year.

Recently, the industry has become more aggressive in its lightweighting actions, and the rate of introducing new technology to improve customer’s experience is accelerating. Vehicle weight reduction is a well-known strategy for improving fuel consumption and reducing C02 emission in vehicles. By reducing the mass of the vehicle, the inertial forces that the engine has to overcome are less, and the power required to move the vehicle is thus lowered. The trend of lightweighting, vehicle electrification, V2X connectivity, intelligence, and autonomous driving as well as eco-design for customer health and environment protection has placed several challenges for vehicle NVH development. First, vehicle weight reduction is a well-known strategy for improving fuel consumption and reducing C02 emission in vehicles. However, improving fuel consumption and cutting down CO2 emissions by reducing vehicle weight has inevitable consequences for the strength and stiffness of the systems and structures affected. In general, achieving low body mass and good NVH performance simultaneously is perceived to be a contradictory objective. If not implemented carefully,

4

Challenges of achieving better NVH performance for Chinese domestic brands

lightweighting strategies can lead to NVH issues that are potentially difficult to remedy because these issues could arise from fundamental design of the vehicle body structure and complex system interactions of different systems and sub-systems. Secondly, the combination of lightweighting, electrification, increased demand for advanced driver-assistance system (ADAS), connectivity contents, and faster product cadences has affected overall vehicle cost amortization. This cost amortization limits NVH strategies for noise source design, structural solidity and sound package content. Thirdly, for Chinese manufacturers, considering the fact that Chinese customers prefer more the quiet sound than the dynamic sound which is the preference of customers in European and North America markets[2], pursuing better NVH performance has become a very important competitive focus in China market to compete with wellknown established global brands.

LIGHTWEIGHT ACTIONS It has demonstrated that vehicle weight reduction has the potential to reduce fuel consumption. On average,across all available vehicle models, every 100 kg weight reduction will achieve a reduction of 0.6-0.8 L/100km in fuel consumption. For electrical vehicles, every 100 kg weight reduction will achieve a save of 0.5 KWH/100 km battery electricity for a vehicle of 1600 kg curb weight. Figure 3 shows the trend of vehicle weight reduction for European vehicles and Geely brands in China market. Over the past years, the European vehicles show a gradual reduction trend. However, the trend of Geely vehicle weight seems “creeping up”. There are several reasons attributing to the trend of weight increasing, one of the reasons is mainly due to “feature creeping”. The increasing number of new features that have been introduced into vehicles that improve utility such as comfort and safety, which is also add weight. Examples include power folding seats, heated seats, navigation systems, additional speakers, and safety features like side airbags, etc. This “Mass creep “phenomena, the weight of added safety, emissions and feature content, that can offset the weight saved during base vehicle development.

5

Challenges of achieving better NVH performance for Chinese domestic brands

Figure 3. The trend of equivalent body volume density of Europe vehicles and Geely vehicles.

There are several ways to reduce the weight of new vehicles. Weight reduction can be achieved by a combination of: 1) optimizing structure design; 2) substituting lightweight materials; and 3) optimizing the manufacturing processes by introducing new lightweight materials. Cost is an important consideration for lightweighting a vehicle, because we are interested in detailing the benefits associated with vehicle weight reduction at an acceptable cost of implementation. For weight reduction using lightweight materials, automakers have been reluctant to adopt new materials and manufacturing process, in part because of the established infrastructure, capital equipment, and knowledge base to promote use of conventional materials, and also because of the cost of substituting these alternative lightweight materials. Cost estimates of using lightweight automotive materials could be varying from parts to parts widely per kilogram of weight savings. However, it is practically acceptable for a new developing program that if the cost increase is in the range of 30 to 40 RMB per kilogram of weight savings. From the vehicle development point of view, it is recommended that if a vehicle is planned to achieve 10% lighter than previous model year vehicle or competitions, it is better to plan the mass reduction of 15%. Figure 4 shows a schematic diagram of vehicle curb mass variation during a development process as an example. As shown in the figure, at the begin of a vehicle development program, the vehicle weight target was established based on its Product Attribute Leadership Strategy (PALS), the vehi-

6

Challenges of achieving better NVH performance for Chinese domestic brands

cle weight typically is well controlled during early design and analysis phase before the first prototype coming. Once entering the prototype development phase, the vehicle weight will “creep” as attribute DV procedures progress, since more mass could be added due to the following reasons: 1) insufficient weight historical data to predicting actual weights for certain systems/subsystems such as powertrain, structural glue; 2) adjusting of attribute performance strategy such as safety and NVH performance; and 3) improper design found in the DV process. Therefore, it has been suggested, for practical application purpose, to have a different weight control roadmap during the entire development process as shown in the figure, in order to reach the established vehicle target at the Job1.

Figure 4. A typical example of vehicle weight variation during a vehicle development process.

LIGHTWEIGHT STRATEGY FOR NVH PERFORMANCE As the OEMs have to continue to improve the vehicles they are developing for the coming model years. Customers want the better performance which are associated with stiffer body which could be at a price of adding mass to achieve structural reinforcement. In general, achieving low body mass and good NVH performance simultaneously is perceived to be a contradictory objective. If not implemented carefully,

7

Challenges of achieving better NVH performance for Chinese domestic brands

lightweighting strategies can lead to NVH issues that are potentially difficult to remedy because these issues could arise from fundamental design of the vehicle body structure and complex system interactions of different systems and sub-systems Torsional stiffness is one of critical parameters related to NVH performance. It affects the vehicle solidity related performance such as squeak and rattle. Torsional stiffness has major influence on vehicle dynamics performance such as ride and handling. Figure 5 shows a collection of multiple vehicle torsional stiffness vs vehicle curb mass. For overall of the data, it is arguable if there is a proportional relationship. However, if we plot the torsional stiffness vs vehicle BIW mass, as shown in Figure 6, it is shown a proportional relationship between the torsional stiffness and BIW mass. It makes sense that the vehicle stiffness is primarily dominated by its BIW structural design. Although we tend to draw the conclusion from this set data that the heavier the vehicles are, the better torsional stiffness are, it shouldn’t ignore the factor that for a given BIW mass the torsional stiffness can be improved by 15% to 20% with proper structural design as the individual data shown in the plot. Therefore, the better structure design, combined with light weight material usage and associated new manufacturing processes, is a very critical steps to achieve lightweight. Relationship between curb mass & stiffness 35000

Stiffness

30000

25000

20000

15000

10000 1000

1200

1400

1600

1800

2000

Curb Mass

Figure 5. Torsional stiffness of a collected vehicles vs curb mass.

8

2200

2400

2600

Challenges of achieving better NVH performance for Chinese domestic brands

Figure 6. Torsional stiffness vs BIW mass.

Figure 7. The modal frequency comparison of a rear floor panel with different thicknesses and floor pattern designs.

9

Challenges of achieving better NVH performance for Chinese domestic brands

On a component level, the amount of weight savings resulting from better structural design depends on the application and design intent. For example, for a rear floor panel designed for modal responses to against certain vehicle excitations, , using better panel local reinforcement design, as shown in Figure 7, the thickness reduction from 0.7 mm down to 0.65 mm can still maintain or achieve better modal frequency as targeted. Sound package is a critical part of controlling vehicle interior airborne noise. Although the weight of a complete set of vehicle sound package, varying from vehicle to vehicle, is roughly 2% to 3% of the overall vehicle curb weight, as shown in Figure 8, the addition of a wide range of new features such as electronic equipment for safety and entertainment could lead to a demand of weight reduction for NVH sound package. One of approaches is to use lightweight dissipative constructions for floor trim and dash insulator to replace conventional barrier-based systems. Figure 9 show the transmission loss comparison of conventional barrier-based approach vs lightweight dissipative treatments. The material parameters are listed in Table 1. Using the dissipative construction saves about 40% of the weight.

Figure 8. The ratio of sound package weight vs vehicle curb weight data for multiple vehicles.

10

Challenges of achieving better NVH performance for Chinese domestic brands

Figure 9. The Comparison of Transmission Loss for Barrier-based dash vs. Dual-layer lightweight dash.

Table 1. Parameters of Barrier-based dash vs Dual-layer Lightweight dash Baseline Design

Lightweight Design

material

2mm EVA + 600 mg/m2 blended cotton and synthetic fibers

1000 g/m2 Blended Cotton+1400 g/m2 Blended cotton)

Weight

5.3 kg

3.0 kg

Cost

baseline

no change

Performance

Meet target

Meet Target

The comparison of vehicle level NVH performance for 3rd gear wide open throat (WOT) are shown in Figure 9 and 10, separately, for interior noise level in dB (A) and Articulation Index. The demonstration of this example using the dissipative system concluded that the interior acoustic dash, tuned properly, can results in a mass reduction of nearly 40% without degrading the NVH performance expected by customers.

11

Challenges of achieving better NVH performance for Chinese domestic brands

Figure 9. The example of vehicle interior noise comparison for different type of dashes shown in Figure 8 and Table 1. Vehicle operational condition: 3rd gear, Wide Open Throat (WOT).

12

Challenges of achieving better NVH performance for Chinese domestic brands

Figure 10. The example of vehicle interior Articulation Index comparison for different type of dashes shown in Figure 8 and Table 1. Vehicle operational condition: 3rd gear, Wide Open Throat (WOT).

Coatings technologies that offer collateral lightweighting benefits as well as friction reduction and NVH attenuation are receiving greater attention by suppliers. A proven example is the liquid acrylic sound-deadening costing (LASD) produced by Henkel that is replacing relatively heavy bitumen acoustic mats in critical areas. In one of our study, we have achieved mass reduction up to 30% for Henkel incumbent low density LASD for different vehicle applications as shown in Figure 11. To verify the performance of LASD, we conducted vehicle level testing for different operational conditions to compare these two materials and have found no detectable difference on vehicle levels between these two materials. Figure 12 shows the comparison of vehicle rear floor panel equivalent stiffness measurements for these two different materials. In the overall trend, LASD provides similar performance as bitumen acoustic mats.

13

Challenges of achieving better NVH performance for Chinese domestic brands

Figure 11. The weight comparison of bitumen pad vs. LASD for different vehicle applications

Figure 12 shows the rear panel driving point equivalent stiffness comparison for bitumen pad and LASD.

14

Challenges of achieving better NVH performance for Chinese domestic brands

OPPORTUNITIES FOR CHINESE DOMESTIC OEMs For latest Geely vehicle design, the usage of high strength steel has exceed 60%. The vehicle body lightweight coefficient is 3.05, in a leading position among domestic brands. However, looking at the historical data of Geely vehicles as shown in Figure 3, the weight of Geely vehicles is slightly “creeping up” due to “feature creeping” to improve the safety and quality. This trend will demand the OEM to take more aggressive effort to achieve more weight reduction than global brands in coming years in order to achieve better fuel economy to compete with other global brands. Furthermore, if we separate the Geely’s data in Figure 6 from other brands and plot on the same figure as shown in Figure 13, comparing the fitted data trends between these two groups, it is noticed that Geely has significant opportunities to improve its BIW design to achieve the similar or even better weight efficiency than other global brands. One of the critical parameters to assessing the effectiveness of lightweighting technology application is the ratio of BIW vs curb mass. For a given performance parameter, the smaller the ratio is, the better BIW design is. Based on the collected historical data, it has found that the ratio is typically about 19-21%. Figure 14a and 14b show the comparison of the Geely vehicles and European vehicles for sedans and SUVs, separately. The performance parameter is torsional stiffness. The ratio of BIW over curb mass for Geely brand is about 2% higher than the global brands, corresponding about 20-40 kg curb mass heavier than global brands. These two sets of data indicate that although Geely brand has improved the torsional stiffness significantly over the last few years, there is still a room for further improvement through structure design and optimization.

15

Challenges of achieving better NVH performance for Chinese domestic brands

Figure 13. The comparison of trends of Geely vehicles vs. Global brands.

Figure 14a. The torsional stiffness vs. the ratio of BIW mass and Curb Mass. The European vehicle data vs. Geely’s data for sedans. The typical ratio of BIW mass and curb mass is 0.19-0.21.

16

Challenges of achieving better NVH performance for Chinese domestic brands

Figure 14b. The torsional stiffness vs. the ratio of BIW mass and Curb Mass. The European vehicle data vs. Geely’s data for SUVs. The typical ratio of BIW mass and curb mass is 0.19-0.21.

Improving NVH performance can improve product quality, promote product brand image, and increase the profitability of the product. However, the quality improvement, in general, could be associated with cost increasing. Due to the lack of brand image, the domestic brands are sold at much low prices in order to gain more market shares. In Figure 15a, 15b,and 15c, the comparison of MSRP (Manufacturer’s Suggested Retail Price) for Compact sedan,Compact SUV and Mid-size Sedan, are shown respectively, for Chinese domestic brands and global brands. As shown in the figures, the MSRP of Chinese domestic brands are at the lower end of the MSRP of global brands. In terms of NVH performance, most of those domestic brands are at the same level as the global brands. For example, the NVH performance of Geely NL-3 vehicle was developed to target VW Tiguan. The actual product is as good as Tiguan, but at almost a half of Tiguan’s price. Therefore, on one hand, at the current stage, the profit per vehicle for domestic brands is low, comparing with global brands; On the other hand, once the MSRP of domestic brands can get close to or at the same level as global brands, their profit margin will be improved significantly. The domestic OEMs will have more money to invest in new “features”, to improve function and attribute performance to increase their competitiveness, and to get more market share.

17

Challenges of achieving better NVH performance for Chinese domestic brands

Figure 15a. The MSRP comparison of Chinese domestic brands and global brands for compact sedans

Figure 15b. The MSRP comparison of Chinese domestic brands and global brands for compact SUVs.

18

Challenges of achieving better NVH performance for Chinese domestic brands

Figure 15c. The MSRP comparison of Chinese domestic brands and global brands for Mid-size sedans.

As shown in Figure 1, on average, the gap between Chinese Domestic brands and global bands is gradually getting narrow. However, there are still significant gaps between domestic brands and well-established brands such as Audi, Benz, BMW and Lexus in terms of IQS NVH performance and high mileage NVH performance. The degradation of NVH performance due to high mileages is a critical factor for customers’ long term loyalty and the decision making of new vehicle purchase. It becomes imperative for Chinese domestic brands to improve their high mileage performance over time. Improving high mileage NVH performance will be another opportunity for Chinese domestic brands in coming years.

Summary From manufacturing good vehicles for everyone 20 years ago, then manufacturing most safety, environmental friendly and best fuel economy vehicles for last ten year, until creating boutique vehicles for everyone as new brand mission, like many other Chinese domestic OEMs, Geely has come a very long way in a very short time since it flooded the Chinese market 15 years ago with entry level vehicles. The progress of Geely has made along with other Chinese domestic brands is magnificent. So far, Geely has demonstrated that it is way ahead of the Chinese auto industry, and probably ahead of many carmakers in the world. If any Chinese Car company will take on the world, it is this one [6].

19

Challenges of achieving better NVH performance for Chinese domestic brands

The Chinese brands are on the trend of ascendancy. Chinese domestic OEMs have made tremendous progress in the past 10 year. However, there is still a gap between them and global leading brands such as Audi, Benz, BMW and Lexus, etc. Benchmarking and learning from leading global OEMs, the domestic OEMs need to invest more, in additional to the product development, in advanced technology development and fundamental theoretic study, in order to catch up with the leading global OEMs. In terms of lightweighting vehicles for new programs, what will be the construction of post-2025 passenger vehicles consist of [3]? If the “lightweighting” appears certain to remain a product-development mantra, what should NVH engineers do to achieve better or maintain the NVH performance? The trend of lightweighting, vehicle electrification, V2X connectivity, intelligence, and autonomous driving as well as eco-design for customer health and environment protection has placed many challenges for future vehicle NVH development.

References [1] Eric Frank, B. Engels, B. Naimipour and G. Rinaldi, “Consumer’s Demand for Better Sound and Vibration Quality”, Sound and Vibration, 2015. [2] Kyoung-Jin Chang, Ki Woong Jeong, and Dong Chul Park, “A Study on the Strategy and Implementing Technology for the Development of Luxurious Driving Sound”, SAE Technical Paper, 2014-01-0035. [3] Lindsay Brooke, Ryan Gehm and Bill Visnic, “ Lightweighting: What’s next?”,Auto motive Engineering, August, 2016. [4] Michael J Dunne, “Shanghai Paints A Misleading Picture of China’s Car Market“ [5] David Caprioli, Philippe Godano, Delphine Guigner, Marco Seppi, “Increased Func tionality in Lightweight Fibre-based NVH Packages”, Automotive Acoustics, September, 2016, ATZ, pp44-49. [6] Bertel Schmitt, “If Any Chinese Car Company will take on the world, it’s this one”, Forbes, April 24, 2017.

Authors Perry P. Gu, Fei Xiong, Hailan Zhao, and Xie Kai Geely Automobile Research Institute No. 918 BinHai 4th Road, HangZhou Bay New District Ningbo, Zhejiang Province China, 315336

20

Low- and high-frequency NVH CAE – test methods for development of a lightweight sedan design Yuksel Gur*, Jian Pan**, and David A. Wagner*

* Ford Motor Company, Research and Advanced Engineering Research Innovation Center ** Autoneum North America, Inc.

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 W. Siebenpfeiffer (Hrsg.), Automotive Acoustics Conference 2017, Proceedings, https://doi.org/10.1007/978-3-658-20251-4_2

1

Low- and high-frequency NVH CAE – test methods for development of a lightweight …

1 Introduction The Multi Material Lightweight Vehicle (“MMLV”) is a collaborative effort between Ford Motor Company, Magna International and US Department of Energy to design, develop and deliver a 23.3% weight saving over the baseline vehicle, “Vehicle A”, using lightweight materials and manufacturing technologies. MMLV vehicle design was developed by collaborative efforts of several Ford Product Development organizations, Magna International Inc., and suppliers [1]. MMLV prototypes have been built for durability, safety and NVH testing, and CAE correlation. In this paper, MMLV NVH lightweight sound package development, MMLV NVH CAE and Test results, and overall NVH performance of MMLV are presented. The limitations and use of CAE in vehicle design process is discussed. Low- and Mid-frequency NVH CAE vehicle interior acoustic transfer function results due to body attachment point excitations with corresponding test data is presented and then lightweight vehicle sound package development progress and sound package designs are presented with noise reduction test data to highlight the NVH performance of lightweight sound packages. Full vehicle Statistical Energy Analysis (SEA) model development for high frequency NVH simulations is also presented to evaluate the vehicle NVH performance of sound package designs. MMLV vehicle road noise, wind noise, and pass-by noise test results are shown to highlight the MMLV vehicle’s NVH performance.

Figure 1. Multi Material Lightweight Vehicle (MMLV)

2

Low- and high-frequency NVH CAE – test methods for development of a lightweight …

2 Vehicle Trimmed Body LP6 CAE and Test The LP6 test consists of measurement of trim body vehicle interior acoustic, drive point mobility, and tactile transfer functions for subframe, suspension, powertrain, exhaust attachment point excitations. The vehicle LP6 CAE analysis is very important for evaluation of vehicle’s structure borne noise performance in the design phase. The LP6 CAE and testing are used to obtain the dynamic characteristics of local attachment structures and vehicle acoustic and tactile sensitivities. These sensitivities and attachment stiffness are important part of the System Design Specifications (SDS) for body structures. The LP6 CAE analysis was used to drive the design of MMLV to reduce the structure borne noise. The LP6 trimmed body testing was conducted to verify CAE predictions and improve the CAE modelling prediction capability. Vehicle acoustic transfer functions are the vehicle interior acoustic response due to unit force excitation at each of the body attachment points. Each acoustic transfer function is calculated separately for each body attachment excitation. These transfer functions are very important for vehicle acoustic performance for the structure borne noise performance. There are targets for these acoustic transfer functions at the trimmed body level. Drive point mobility is defined as the frequency response function of velocity at attachment points where unit force is applied. Drive point mobility is one of the parameters determining the power input into the body. The higher the drive point mobility, the higher is the power input into the structure through that attachment point.

2.1 Vehicle Trimmed Body FEA Modelling for LP6 CAE Analysis The LP6 CAE analysis is used to predict the basic dynamic characteristics of local attachment structures and body acoustic and tactile sensitivities. The LP6 CAE analysis is used for analyzing the NVH performance of different design proposals and fixing potential NVH deficiencies during the vehicle development. The vehicle acoustic transfer functions are very important to evaluate the overall NVH performance of a vehicle design due to the road input. To calculate the vehicle acoustic transfer functions, the vehicle trimmed body FEA model needs to be coupled with vehicle interior acoustic cavity model. The FEA model which consists of trimmed body and vehicle interior acoustic models is used to predict the vehicle acoustic sensitivity (p/F) transfer functions.

3

Low- and high-frequency NVH CAE – test methods for development of a lightweight …

A coupled structural-acoustic analysis can be performed by creating the FEA model for the vehicle interior air cavity and combining it with the FEA model of the vehicle trimmed body model. The trimmed body model and the seat models are required as input to generate the vehicle interior acoustic model. AKUSMOD [2] is an auto meshing software used to model the vehicle interior cavity mesh. Shell elements enclosing the vehicle interior cavity as well as the seat surface elements are selected to define the vehicle interior cavity and holes such as door inner panel holes are covered by additional shell elements to generate the closed domain for vehicle cavity mesh generation. Figure 2a shows the trimmed body FEA model and Figure 2b shows the interior cavity FEA model of MMLV. AKUSMOD is used to automesh the vehicle interior acoustic volume up to the surface elements of the trimmed body and the seats. The coupling constraints, multi-point constraints (MPC) between seat volume mesh and interior acoustic mesh is generated by AKUSMOD. AKUSMOD is also used to generate the fluid-structural coupling between trimmed body and interior acoustic mesh.

Figure 2. (a) Trimmed body FEA model of MMLV vehicle; (b) Vehicle interior cavity mesh for MMLV vehicle

2.2 Vehicle Trimmed Body Vehicle Acoustic Transfer Function Test/CAE Correlation MMLV trimmed body acoustic transfer functions are obtained by conducting the LP6 tests and LP6 CAE analyses. For each body mount excitations, vehicle acoustic transfer functions between excitation points and driver's outboard ear (DOE) are obtained to determine the vehicle acoustic sensitivities for body mount excitations both experimentally and numerically up to a frequency of 400 Hz. Since “Vehicle A” was the baseline model for the MMLV vehicle development, “Vehicle A” trimmed body acoustic transfer functions are also obtained to establish the baseline response level for the MMLV vehicle. Figure 3 shows the MMLV trimmed body acoustic transfer function CAE vs. TEST comparison in the frequency range from 10 Hz to 400 Hz.

4

Low- and high-frequency NVH CAE – test methods for development of a lightweight …

Definitions of the body attachment labels shown along the vertical axis of the plots are described below: PEN_RB X:

Engine body mount, right body side, excitation along x-direction.

FSH_LB Y:

Front Shock mount, left body side, excitation along y-direction.

ROLRNEW Z: Engine Roll Restrictor, excitation along z-direction. LFLCABU1 X: Front LCA (Lower Control Arm) body mount, left body side, excitation along x-direction. RFLCABU1 Y: Front LCA (Lower Control Arm) body mount, right body side, excitation along y-direction. RSFF_LB X:

Rear Subframe Front, left body side, excitation along x-direction.

RSH_LB Y:

Rear Shock mount, left body side, excitation along y-direction.

RSFR_LB Z:

Rear Subframe Rear, left body side, excitation along z-direction.

Figure 4 shows the MMLV vehicle’s 80th percentile and 98th percentile correlation results between CAE and Test for body acoustic transfer functions. The correlations shown in this figure indicate that the percentile responses are better for the correlations and vehicle NVH target settings. The LP6 acoustic sensitivity CAE analysis to test correlation indicates that the average 80th and 98th percentile CAE to Test data difference are 2.0 dB and 2.8 dB, respectively.

Figure 3. MMLV CAE vs. TEST comparison for acoustic transfer functions for body attachment excitations.

5

Low- and high-frequency NVH CAE – test methods for development of a lightweight …

Figure 4. (a) MMLV 80th percentile ; (b) MMLV 98th percentile CAE vs. TEST comparison for acoustic sensitivities.

Table 1 shows the percentile body mount acoustic sensitivity differences between “Vehicle A” and MMLV vehicle. These test results indicate that average MMLV acoustic sensitivities are 1.6 dB higher for the 80th percentile and 1.5 dB higher for the 98th percentile compared to the “Vehicle A” indicating that a possible MMLV vehicle’s low and mid frequency road NVH degradation of 1.5 dB without taking into account of tire design effects.

6

Low- and high-frequency NVH CAE – test methods for development of a lightweight …

Table 1. Acoustic Sensitivity (“Vehicle A” – MMLV) TEST differences for both 80th and 98th percentile

FSH_LB X FSH_LB Y FSH_LB Z FSH_RB X FSH_RB Y FSH_RB Z PEN_RB X PEN_RB Y PEN_RB Z PEN_LB X PEN_LB Y PEN_LB Z RSFF_LB X RSFF_LB Y RSFF_LB Z RSFF_RB X RSFF_RB Y RSFF_RB Z RSFR_LB X RSFR_LB Y RSFR_LB Z RSFR_RB X RSFR_RB Y RSFR_RB Z RSH_LB X RSH_LB Y RSH_LB Z RSH_RB X RSH_RB Y RSH_RB Z AVERAGE

("Vehicle A" - MMLV) TEST Difference - ("Vehicle A" - MMLV) TEST Difference Acoustic Sensitivity 80th percentile Acoustic Sensitivity 98th percentile 0.0 -0.6 1.6 1.8 -0.6 0.7 2.0 -3.1 -2.8 0.4 -0.5 2.0 0.0 -0.2 -2.6 -8.6 -1.2 -2.3 -3.8 -5.1 -5.9 -0.9 -0.6 0.0 -4.2 -4.6 -5.7 -3.9 0.0 1.7 -6.2 -3.6 -0.7 -4.0 -0.4 0.8 -1.5 -1.1 -1.3 -2.5 -1.0 -1.6 -0.8 0.7 -2.9 -0.9 -5.0 -2.2 0.0 -0.8 -5.0 -4.1 -0.7 -2.3 1.3 0.8 0.6 0.2 -0.9 -1.1 -1.6

-1.5

3 Lightweight Sound Package Development The MMLV vehicle utilizes extensively advanced lightweight materials such as highstrength steel, magnesium alloys and aluminum alloys, etc. in the body structures. These lightweight body panels increase noise transmission into the passenger compartment. In order to improve acoustical performance of lightweight vehicles, a simple approach is to add additional mass to sound package components to compensate

7

Low- and high-frequency NVH CAE – test methods for development of a lightweight …

for the reduction of mass in the sheet metal. This approach will increase the weight of the sound package of the vehicle, and therefore limits the potential of total weight saving of the vehicle. A second approach is to increase packaging space of acoustic components, which will increase sound transmission loss as well as sound absorption of acoustic components. However, the second approach is often constrained by clearance requirements or interior dimension requirements and may not be feasible practically. A new approach to reduce interior noise in lightweight vehicles is required in order to maximize the weight saving potential of lightweight vehicles without significantly increasing packaging space [5].This new approach is derived based on analysis of twosubsystem Statistical Energy Analysis (SEA) model [5]. In the context of vehicle noise control, this new approach can be summarized as the following: 1. Increasing damping loss factor in the source and/or receiving subsystems, for example, inside the engine compartment and inside the passenger compartment, will reduce engine noise and tire noise. This is a simple concept widely known to the noise control community. Practically, there are two ways to increase absorption: (a). increasing absorption coefficient of an existing component by using more absorptive materials, or by utilizing multilayer absorbers to increase low to mid frequency absorption. (b). increasing the surface area of absorbing components by increasing the size of existing absorption components, or adding absorptive function to non-absorptive surfaces. 2. Increasing damping loss factor in the source and/or receiving subsystems, e.g., inside engine compartment and inside passenger compartment, is more effective than reducing coupling loss factor between source and receiving subsystems, e.g., between engine compartment and passenger compartment, i.e., doubling in damping loss factor results in larger reduction in interior noise than doubling in coupling loss factor between noise source and receiver. This principle is less known in the noise control community. Practically, increasing damping loss factors (i.e. increasing absorption) requires less weight than reducing coupling loss factor (i.e. increasing STL). This principle is the foundation for lightweight sound package design and development of MMLV vehicle [3][4][5]. A number of light weight acoustical materials have been developed for applications in the engine bay, vehicle underbody and under engine environment, as well as vehicle interior. These materials provide high sound absorption at broad frequency range from 400 Hz to 10,000 Hz, while delivering “adequate” sound transmission loss. “Adequacy” here is determined by application, and by the available weight of the associated sound package component. Theta-CellTM (TC) is semi-rigid lightweight polyurethane foam with high sound absorption performance at a broad frequency range. It helps reduce the weight of engine

8

Low- and high-frequency NVH CAE – test methods for development of a lightweight …

bay absorbers in a limited packaging space. It is designed for applications in high temperature environment such as engine bay. Theta-FiberCellTM (TFC) is a dual layer acoustical panel made of lightweight structural fiber and semi-rigid polyurethane foam that has high thermal resistance and mechanical properties suitable for engine compartment applications. RUSTM is a lightweight underbody technology made of 100% uniquely engineered PET (Polyethylene terephthalate) that has strong mechanical properties and can survive the harsh environment of vehicle underbody, and provide good sound absorption. PET is the most common thermoplastic polymer resin of the polyester family. Ultra-LightTM ECO+ floor insulator is a lightweight carpet consisting of a dense fibrous top layer and a loft fibrous decoupler layer, with perforated foil in between the top and the loft layers. This lightweight insulator is designed to provide high absorption and balanced insertion loss for vehicle interior applications. This material is lighter than traditional spring-mass insulators, and it is intended to replace traditional spring-mass system at a lighter weight, while providing equivalent or better acoustic performance in vehicle interior applications. Sound absorption and insertion loss characteristics of all MMLV lightweight acoustical materials are presented in the SAE paper [5] in details. Both engine bay material and underbody material are designed to have high mechanical properties (flexural strength, tensile strength, etc.) and acoustical performance at the lowest surface weight. In addition, the engine bay material must also withstand the high temperature typically seen in the engine compartment. Figure 5a and 5b show the absorption coefficient and insertion loss of a typical dual layer TFC panel at 15 mm.

Figure 5. (a) Insertion Loss of Lightweight Dual Layer Engine Bay Material; (b) Sound Absorption Coefficient of Ultra-Light Weight Underbody Material [5]

9

Low- and high-frequency NVH CAE – test methods for development of a lightweight …

3.1 MMLV Sound Package Designs Based on the new lightweight acoustic development approaches discussed in previous section, it is desirable to significantly increase sound absorption in the engine compartment, wheel wells and the exterior underbody to reduce engine noise and tire noise. An engine encapsulation was developed using the engine bay material discussed in the previous section to significantly reduce radiated noise from the engine. Two weight neutral sound package designs, defined as sound package one and two, SP#1 and SP#2, were developed using “Vehicle A” as a surrogate vehicle (“Vehicle A”) and SP#2 sound package was implemented on MMLV[3]. SP#1 includes the engine encapsulation and engine top cover as well as the ultra-light carpet prototypes. The ultra-light carpet is lighter than baseline carpet and it has higher absorption characteristics [4].Weight reduction achieved by the ultra-light carpet is used to offset the weight addition due to the engine encapsulation prototype shown in Figure 6. SP#2 lightweight sound package design includes new ultra-light sound package components besides SP#1 components. These new ultra-light sound package components shown in Figure 7 are the ultra-light underbody acoustic shields; ultra-light engine under shield, fiber inserts in front wheel house liner, and dual layer RIMIC tunnel heat shield. This sound package SP#2 also does not increase the total sound package weight of “Vehicle A”. It actually reduces the total sound package by 0.8 kg (see Table 2). Table 2 shows the weight summary of baseline and new light weight sound packages (SP#1 and SP#2). Figures 6 and 7 show the SP#2 sound package prototypes installed on MMLV vehicle.

10

Low- and high-frequency NVH CAE – test methods for development of a lightweight …

Figure 6. Theta Fiber Cell Engine Encapsulation and Engine Cover prototypes on MMLV vehicle [3]

Figure 7. Other lightweight sound package materials in MMLV vehicle [3]

11

Low- and high-frequency NVH CAE – test methods for development of a lightweight …

Table 2. Weight summary of the baseline and new light weight sound packages (SP#1 and SP#2) on “Vehicle A” [3] Baseline Sound

New Sound

New Sound

Package

Package (SP#1)

Package (SP#2)

SP Parts Carpet Engine Top Cover

Baseline Part

SP#1 Part

SP#2 Part

Weight [kg]

Weight [kg]

Weight [kg]

10.35

8.56

8.56

0.83

0.56

0.56

NONE

2.04

2.04

0.91

0.91 (Baseline)

0.95

1.17

1.17 (Baseline)

1.00

1.18

1.18 (Baseline)

1.00

1.49

1.49 (Baseline)

1.17

1.55

1.55 (Baseline)

1.12

Tunnel Heat Shield

0.45

0.45 (Baseline)

0.76

Total Weight

17.93

17.91

17.16

Engine Encapsulation Engine Under Body Panel Driver Side(DS) Under Body Panel Passenger Side(PS) Under Body Panel Front DS Wheel Well Liner Front PS Wheel Well Liner

3.2 MMLV NVH Performance with Lightweight Sound Packages Noise reduction is defined as the difference between the average sound pressure level at the source side SPL_inp and the average sound pressure level at the receiver side SPL_rec. NR = (SPL_inp) – (SPL_rec) [dB] In a vehicle, it represents the acoustic attenuation provided by vehicle body and panels to exterior noise sources (powertrain, driveline, tires, road, and wind). Vehicle Engine Noise Reduction (ENR) tests [3] are conducted using the Reciprocal Point Source Method (RPSM). The RPSM method is setup by placing a High Fre-

12

Low- and high-frequency NVH CAE – test methods for development of a lightweight …

quency Sound Source (HFSS) at the desired interior positions and placing microphones in the vehicle engine compartment. For each source excitations, sound pressure levels of 18 surface microphones placed closely to the engine surfaces are measured. The Vehicle Engine Noise Reduction is obtained by

ENR  SPLref  SPLengine _ surfaces Where

SPLref

[dB]

is the reference SPL of the HFSS and

SPLengine_ surfaces

is the

power-averaged SPL of engine surface microphone responses. MMLV vehicle with baseline (“Vehicle A”) sound package was tested for Engine Noise Reduction (ENR) and Tire Patch Noise Reduction (TPNR) performance. After completing these tests, sound package of the MMLV vehicle was replaced with the SP#2 sound package and tests were repeated to determine the NVH performance improvements due to the ultra-light sound package of SP#2 [3][4]. Test data indicates that MMLV with SP#2 has higher ENR performance above 550 Hz and lower ENR in the damping controlled region below 550 Hz compared to “Vehicle A”. Test data also shows that SP#2 sound package increases the engine noise reduction performance of MMLV significantly in the entire frequency range. Figure 8a shows the MMLV vehicle’s engine noise reduction (ENR) test results for the baseline sound package and SP#2 sound package contents. Test data in Figure 8a indicates that SP#2 sound package prototypes increase the engine noise reduction performance of MMLV vehicle by 3.3 dB.

Figure 8. (a) ENR performance of MMLV (b) TPNR performance of MMLV with baseline and ultra-light sound package designs [3]

Figure 8b shows the MMLV vehicle’s tire patch noise reduction (TPNR) test results as a function of frequency in the front seat location for the baseline and SP#2 sound

13

Low- and high-frequency NVH CAE – test methods for development of a lightweight …

package contents. Figure 8b indicates that SP#2 sound package prototypes increase the tire patch noise reduction (TPNR) performance of MMLV vehicle by 1.2 dB. In summary, ultra-light sound package parts together improved the MMLV vehicle’s engine noise reduction (ENR) by 3.3 dB and front Tire Patch Noise Reduction (TPNR) by 1.2 dB without increasing the baseline sound package weight. MMLV with ultra-light sound package SP#2 has higher NVH performance compared to “Vehicle A” (0.8 dB higher for ENR, 0.7 dB higher for TPNR).

4 Vehicle Engine Noise and Tire Patch Noise Reduction Test/CAE Correlation Statistical Energy Analysis (SEA) models of “Vehicle A” and MMLV vehicle have been developed and full vehicle engine noise reduction SEA calculations have been conducted to evaluate the noise reduction performance of these two vehicles. Statistical Energy Analysis (SEA) is an energy based analysis technique used for acoustic and vibration predictions at higher frequencies for vehicle NVH development. Details of these SEA simulations and CAE-TEST correlation results are discussed in details in [4]. This correlated SEA model of “Vehicle A” is used to develop and evaluate the weight neutral sound package designs to determine the optimal vehicle sound package content. MMLV as-built model were also developed to evaluate the noise reduction performance of MMLV. Figure 9a shows the MMLV ENR Test to CAE correlation results as a function of frequency from 200 Hz to 10 kHz. Comparing the spectral shape of MMLV SEA data with corresponding test data indicates that SEA spectral data indicates a higher level of correlation except in the low to mid frequency region below 400 Hz and in the frequency range of 4 kHz and 5 kHz. This SEA model is used to verify the NVH performance of MMLV and can also be used for the sound package development of future lightweight vehicles. Figure 9b shows the MMLV TPNR Test to CAE correlation results as a function of frequency from 200 Hz to 10 kHz. Comparing the spectral shape of MMLV SEA data with corresponding test data indicates that SEA spectral data under predicts the response below 1.3 kHz and above 5.5 kHz. SEA predicts the response well in the frequency range between 1.3 kHz and 5.5 kHz. The correlation is very weak above 5.5 kHz which is the coincidence frequency range for MMLV. Possible reasons for this discrepancy in the correlation are some leakage problem occurred during the tests and also some modelling problems associated with the lightweight body panels including underbody acoustic panels.

14

Low- and high-frequency NVH CAE – test methods for development of a lightweight …

Figure 9. (a) MMLV Test-to-SEA correlation for Engine Noise Reduction; (b) MMLV Test-toSEA correlation for Tire Patch Noise Reduction

5 Vehicle Transparency Tests Vehicle transparency noise reduction level [5] is defined as the difference between the average sound pressure level at the source side SPL_inp and the average sound pressure level at the receiver side SPL_rec. NRL= (SPL_inp) – (SPL_rec) [dB] The MMLV vehicle transparency tests were also conducted to measure vehicle level sound transmissibility in the reverberant room test facility to evaluate the NVH performance of lightweight glazing designs. Vehicle transparency noise reduction is defined as the difference between the average sound pressure level at the vehicle exterior and the average sound pressure level at the vehicle interior. Figure 10 shows the MMLV and “Vehicle A”s’ transparency noise reductions as a function of frequency. Decrease in the Noise Reduction Level (NRL) spectrum for “Vehicle A” around 3.2 kHz is related to the coincidence frequency region of “Vehicle A” and this coincidence frequency region is shifted to 8 kHz for MMLV. This coincidence frequency shift to higher frequency range is related to light weight glazing designs and aluminum body panels used in MMLV. Vehicle body transparency response of MMLV is worse in the frequency ranges of 50- 200 Hz, 400-1250 Hz, and above 8000 Hz; and better in the wide frequency range from 1800 to 8000 Hz. The average NRL for MMLV is 0.8 dB better than “Vehicle A”. Figure 11 shows the NRL difference between MMLV and “Vehicle A” as a function of frequency. In summary, MMLV is 0.8 dB better compared to “Vehicle A” for the pass-by noise performance.

15

Low- and high-frequency NVH CAE – test methods for development of a lightweight …

Figure 10. MMLV and “Vehicle A” body transparency (NRL) test results

Figure 11. MMLV and “Vehicle A” body transparency performance differences as a function of frequency.

16

Low- and high-frequency NVH CAE – test methods for development of a lightweight …

6

MMLV Road NVH and Wind Tunnel Tests

Vehicle Road NVH tests were conducted to evaluate the road NVH performance of MMLV. Road NVH tests were conducted on the Glen Eagles (Rough Asphalt), Concrete and Smooth road surfaces. During these tests, vehicle interior noise levels were measured. Table 3 shows the Sound Pressure Level (SPL) test data for both Driver Outboard Ear (DOE) and Rear Passenger Outboard Ear (POE2) locations for different road surfaces. These road NVH test results are compared with the available “Vehicle A” road NVH test data in Table 3. Since MMLV smooth road NVH data spectral results indicate door seal leakage issues for this road surface, this road surface comparison is not included in Table 3. Table 3. MMLV and “Vehicle A” Road NVH performance difference

MMLV vehicle road NVH performance is 1.0 dB worse than “Vehicle A”’s performance on concrete road surface at 80 kph, and 1.7 dB worse on the Glen Eagles road surface at 80 kph. Wind noise performance evaluations of MMLV and “Vehicle A” are conducted in the aero-acoustic wind tunnel. The tests are repeated with and without taping the vehicle in order to determine the presence of any leakage. The wind tunnel tests are conducted on the MMLV for four speed conditions 

(100 kph, 130 kph, 160 kph, and 180 kph) and five different yaw conditions ( 0 ,

10 ,  20 ). Four “Head Acoustics Aachen Heads” are placed inside the vehicle to record the vehicle interior noise data. These Aachen Heads are placed on the front seat driver, front seat passenger, rear seat driver side, and rear seat passenger side locations. Table 4 shows the vehicle interior noise levels differences between “Vehicle A” and MMLV for all the wind tunnel test conditions and all the head microphone locations for the quietest (fully taped) condition.

17

Low- and high-frequency NVH CAE – test methods for development of a lightweight …

Descriptions of the head microphone locations in Table 4 are: DOE: DIE: POE: PIE: 2DOE: 2PIE:

Front Seat Driver Outboard Ear Front Seat Driver Inboard Ear Front Seat Passenger Outboard Ear Front Seat Passenger Inboard Ear 2nd Row Seat – Driver Side Outboard Ear 2nd Row Seat – Passenger Side Inboard Ear

The average interior noise level difference between “Vehicle A” and MMLV for all different head microphones are shown in Table 4. The quietest condition test data shown in Table 4 indicates that speed averaged MMLV vehicle’s interior noise level is 0.3 Sones higher for 100 kph and 0.9 Sones higher for 130 kph compared to the “Vehicle A”. This wind noise performance difference becomes larger for higher vehicle speeds (1.5 Sones for 160 kph and 2.0 Sones for 180 kph). Table 4. Interior noise levels differences between “Vehicle A” and MMLV for the Quietest Condition Speed (kph) 100 100 100 100 130 130 130 130 160 160 160 160 160 180 180 180 180 180

Yaw (Deg) 0 10 20 -10 10 20 -10 -20 0 10 20 -10 -20 0 10 20 -10 -20

DOE -1.1 -0.8 0.5 -1.6 -1.0 0.8 -2.3 -4.2 -2.0 -1.2 0.4 -3.7 -6.8 -2.8 -2.2 0.1 -4.1 -7.2

DELTA Sones ("Vehicle A" - MMLV) QUIETEST Condition (Taped) DIE PIE POE 2DOE 2DIE 2PIE -0.2 -0.2 -0.3 0.2 0.2 -0.1 -0.1 -0.2 -0.1 -0.6 -0.3 -0.3 0.8 0.2 0.3 0.1 0.5 1.1 -0.9 -0.8 -1.1 -1.0 -0.9 -1.1 0.1 -0.3 -0.3 -0.9 -0.2 0.0 1.2 0.1 -0.5 0.7 1.4 2.0 -1.6 -1.3 -2.3 -1.8 -1.8 -1.8 -2.1 -1.6 -1.5 -1.7 -1.3 -1.8 -0.8 -0.6 -1.0 -1.1 -0.7 -0.4 0.2 -0.2 0.0 -1.1 0.1 0.5 0.8 -0.5 -1.3 -1.1 1.4 1.9 -2.7 -2.7 -3.3 -3.2 -2.1 -2.2 -3.6 -3.4 -3.3 -2.7 -0.9 -1.2 -0.8 -0.7 -1.1 -2.0 -0.5 -0.4 -0.5 -1.2 -1.2 -3.2 0.0 0.2 0.5 -0.9 -2.7 -4.8 -0.2 -0.2 -2.7 -2.5 -3.1 -3.4 -1.5 -2.0 -3.7 -4.1 -3.9 -2.2 -0.8 -1.1

2POE 0.0 -0.7 0.2 -1.1 -0.9 0.5 -3.2 -2.3 -1.8 -1.0 -0.1 -4.2 -2.5 -2.4 -1.6 -1.9 -4.2 -2.8

Average of all head Microphones

-0.2 -0.4 0.5 -1.1 -0.4 0.8 -2.0 -2.1 -1.1 -0.3 0.2 -3.0 -3.1 -1.3 -1.2 -1.3 -2.9 -3.2

Average for each speed condition

-0.3

-0.9

-1.5

-2.0

7 Conclusions In this study, NVH CAE-TEST correlation results for MMLV are presented to highlight the limitation and use of finite element analysis (FEA) and statistical energy analysis (SEA) [3] in vehicle design. Lightweight MMLV vehicle’s NVH performance compared to “Vehicle A” is also presented. Ultra-light sound package design development and its impact on the NVH performance of MMLV vehicle are presented with experimental and CAE data. The results of this study could be used to determine the further improvement areas for both low and high frequency NVH modeling.

18

Low- and high-frequency NVH CAE – test methods for development of a lightweight …

Here are major conclusions which can be drawn from this research work: ● MMLV wind noise performance is 0.9 dB worse than “Vehicle A” at 130 kph. MMLV is 0.8 dB better compared to “Vehicle A” for the pass-by noise performance. ● MMLV has better noise reduction performance compared to “Vehicle A” (0.8 dB higher for ENR, 0.7 dB higher for TPNR). ● A number of industry-leading light weight acoustical technologies were applied to improve MMLV baseline vehicle’s acoustical performance. These technologies are: ● ● ● ●

Ultra-LightTM ECO+ floor carpet RUSTM lightweight underbody technology RIMICTM perforated acoustical heat shield Theta Fiber CellTM structural-acoustical material for engine compartment applications

Sound package parts build with these materials together improved the MMLV vehicle’s engine noise reduction (ENR) by 3.3 dB and front Tire Patch Noise Reduction (TPNR) by 1.2 dB without increasing the baseline sound package weight. ● A new approach to sound package design in lightweight vehicles was developed to reduce vehicle interior noise without addition of weight. ● MMLV acoustic sensitivities are 1.6 dB higher for the 80th percentile and 1.5 dB higher for the 98th percentile compared to the “Vehicle A” indicating MMLV vehicle’s low and mid frequency road NVH degradation without taking into account of tire design effects. ● MMLV road noise performance is 1.4 dB worse than “Vehicle A” for a vehicle speed of 80 kph. This result is in agreement with the LP6 acoustic sensitivity test data. ● The MMLV LP6 acoustic sensitivity CAE analysis to test correlation indicates that the average 80th and 98th percentile CAE to Test differences are 2.0 dB and 2.8 dB, respectively. ● LP6 CAE to Test correlation results indicate that the percentile (80th and 98th) responses for the acoustic and tactile sensitivities are better for the correlations and vehicle NVH target settings. ● MMLV as-built model were also developed to evaluate the noise reduction performance of MMLV. This SEA model is used to verify the NVH performance of MMLV and can also be used for the sound package development of future lightweight vehicles. ● SEA is an effective sound package optimization tool to evaluate the vehicle’s NVH performance provided that a correlated baseline vehicle level SEA model is available. Developing a correlated SEA model often includes challenges requiring significant SEA modelling efforts and robust and extensive experimental data.

19

Low- and high-frequency NVH CAE – test methods for development of a lightweight …

● Demonstrated that vehicle interior acoustic performance improvement can be achieved for a lightweight vehicle design with the use of new lightweight sound package materials without adding sound package weight

Acknowledgements The authors would like to thank the following individuals for their support, providing test data, and assistance in this project: Jeff Wallace, Gary Strumolo, Matt Zaluzec, Dan McKillip, Jeff Pumphrey, John Huber, Allen Li, Mike Azzouz, Jenyuan Her, Jared Shroyer, Wei Liu, Heesuk Kang, Hangxing Sha, Terry Young, Dan Gorsing, from Ford Motor Company, and Brandon Wichmann, Alex Rojas, and Tim Mason from Autoneum North America, Inc. This material is based upon work supported by the Department of Energy National Energy Technology Laboratory under Award Number No. DE-EE0005574. This report was prepared as an account of work sponsored by an agency of the United States Government. Neither the United States Government nor any agency thereof, nor any of their employees, makes any warranty, express or implied, or assumes any legal liability or responsibility for the accuracy, completeness, or usefulness of any information, apparatus, product, or process disclosed, or represents that its use would not infringe privately owned rights. Reference herein to any specific commercial product, process, or service by trade name, trademark, manufacturer, or otherwise does not necessarily constitute or imply its endorsement, recommendation, or favoring by the United States Government or any agency thereof. The views and opinions of authors expressed herein do not necessarily state or reflect those of the United States Government or any agency thereof. Such support does not constitute an endorsement by the Department of Energy of the work or the views expressed herein.

20

Low- and high-frequency NVH CAE – test methods for development of a lightweight …

References 1. Skszek, T., Zaluzec, M., Conklin, J., and Wagner, D. (2015): “ MMLV: Project Overview”, 2015 SAE World Congress, SAE International, SAE paper # 2015-010407, doi:10.4271/2015-01-0407. 2. MSC.AKUSMOD, User's manual, Version 2002, MSC Mechanical Solutions. 3. Gur, Y., Pan, J., Huber, J., Wallace, J. (2015): “MMLV (Multi Material Lightweight Vehicle) NVH Sound Package Development and Full Vehicle Testing”, 2015 SAE World Congress, SAE International, SAE paper # 2015-01-1615, doi:10.4271/2015-01-1615. 4. Gur, Y., Pan, J. and Wagner, D. (2015):” Sound Package Development for Lightweight Vehicle Design using Statistical Energy Analysis (SEA)”, SAE International, SAE paper # 2015-01-2302, doi: 10.4271/2015-01-2302. 5. Pan, J. and Gur, Y., (2015):” Sound Package Design for Lightweight Vehicles”, SAE International, SAE paper # 2015-01-2343, doi: 10.4271/2015-01-2343. 6. Gur, Y., Wykoff, R., Nietering, K., and Wagner, D.(2012): “NVH Performance of Lightweight Glazing Materials In Vehicle Design”, Paper # IMECE2012-89439, Proceedings of the ASME 2012 Int. Mech. Eng. Conference, Nov. 9-15, 2012, Houston, TX.

Authors Yuksel Gur*, Jian Pan**, and David A. Wagner* * Ford Motor Company, Research and Advanced Engineering Research Innovation Center, 2101 Village Road Dearborn, MI 48121 USA ** Autoneum North America, Inc. 29293 Haggerty Road Novi MI 48377 USA

21

Vibration reduction in automotive applications based on the damping effect of granular material Sebastian Koch, Scientific associate Fabian Duvigneau, Scientific associate Sascha Duczek, Postdoctoral fellow Elmar Woschke, Ass.-Professor

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 W. Siebenpfeiffer (Hrsg.), Automotive Acoustics Conference 2017, Proceedings, https://doi.org/10.1007/978-3-658-20251-4_3

1

Vibration reduction in automotive applications based on the damping effect …

Introduction The improvement in sound quality is an important aspect in the development of automobiles, as it influences the customers comfort and therefore the purchase decision. Furthermore, the legislature has passed several bills that aim at tightening the sound radiation threshold within the next decade. In order to improve the sound emission properties of lightweight designs, extensive research efforts have been devoted to develop sophisticated concepts that achieve an excellent noise reduction at a low mass. A thermo-acoustical encapsulation is a very promising passive approach to significantly reduce the noise radiation. In [1] an encapsulation made of foamed material applied to an engine is investigated. Often encapsulations are designed as a spring-mass-system and therefore they consist of a bimaterial system. Another passive concept is studied by Hering [2], where a structural intensity analyis computes the best position to add a small mass in order to reduce the resulting vibrations. In addition, a sound reduction can be achieved by modifying the geometry, e.g. by applying stiffening elements on the surface of the structure [3]. It is important to note that not only the sound level but also the human perception of the sound radiation is an important factor that needs to be taken into account. Therefore, psychoacoustic investigations need to be carried as was done in Duvigneau et al. [4]. Here, a numerical approach, which can be used to assess the quality of the sound radiation of an engine with respect to its psychoacoustic characteristics, has been developed. Therefore, a holistic numerical simulation is executed to predict the sound that is in a second step rated by test persons in a hearing test. In this way, the psychoacoustic design can be improved at an early stage of development process. This study presents a passive lightweight concept using granular materials, with the aim to achieve both a lower sound emission level and a reduced mass compared to the original configuration. This concept was introduced in a previous paper by the authors [5], where the excellent damping properties of granular materials was demonstrated. In their study they focused on automotive applications and used an oil pan to show the performance of novel approach. Since the granular material needs to store in the bottom plate of the oil pan a new design was developed. As the main result of the conducted study a lightweight concept was proposed, which offers both a reduced mass and a lower amplitude of the sound pressure. Moreover, the influence of the position of the granular material on the vibrations was analyzed. The position of the granular filling cannot be controlled, if only one large cavity and a partial filling are used. For this reason, the effect of the distribution of partial fillings was investigated in detail by Koch et al. [6] in follow-up article. In this publication, a combination of honeycomb structures with granular materials was proposed. Sandwich panels with a honeycomb core layer offer excellent stiffness to weight ratios while granular materials provide the required damping properties. The honeycomb structure allows a defined positioning of the granules, which is necessary to investigate the influence of the filling distribution on the damping properties of the novel concept. Furthermore, different types of sand as granular material were tested. The presented concept was lighter compared to the original oil pan and achieves a significant reduction in vibration amplitudes. The paper at hand continues the previous work of Duvigneau et al. [5] and Koch et al. [6]. Therefore, three new and important aspects are examined in this

2 2

Vibration reduction in automotive applications based on the damping effect …

study. First, different filling materials are examined and compared with the most effective sand type of the previous studies. Second, it is investigated whether selected modes can be suppressed by an adapted positioning of the granular material. Finally, it is examined how the vibration behavior differs in a horizontal and a vertical suspension.

Experimental setup The vibration behavior of the oil pan bottom and the honeycomb sandwich structure is measured with the help of a laser scanning vibrometer (model PSV-400 from Polytech). The vibrometer uses the Doppler Effect, which allows a contactless measurement of the surface velocity. The measuring object is suspended with cords. Consequently, free-free boundary conditions are realized. The excitation is realized with an electro-dynamic shaker (Mini Shaker Type 4810, Brühl & Kjær). The initiated forces are measured with a force sensor. In order to be able to compare the results, the excitation point and the number of measuring points are kept constant. The experimental setup is shown in Fig. 1. Both, the frequency response and the root means square (RMS) value of the vibration are used to show the influence of the granular material. The RMS amplitude of one mesuring point is determined by the amplitude xi at the frequency i and the number of used frequencies n by

(c)

(b)

(a)



� = � ∑ � .

(1)



The design of the original oil pan, which is shown in Fig. 2, has been adapted. While the upper part remains unchanged, the lower part is replaced by a new construction with one or more cavities, which can be filled with different granular materials. Both the original oil pan and the new constructed oil pan bottom are made of aluminum. If not stated otherwise, we deploy sand as the granular filling material, which has several beneficial properties such that it is environmentally friendly, cheap, recyclable and easy to handle. 95 % of the sand particle have a grain size between 0.1 mm and 0.48 mm. The average grain size is 0.3 mm and the maximum grain size is 2.6 mm.

Fig. 1: Experimental setup with (a) Laser scanning vibrometer, (b) Oil pan, (c) Shaker.

Fig. 2: Oil pan with original bottom.

3

3

Vibration reduction in automotive applications based on the damping effect …

Vibrational behavior of a one cavity oil pan

Amplitude [dB]

This section gives an overview about the first steps of the basic concept that was introduced by Duvigneau et al. [5], wherein a lightweightdamping concept is presented, which uses sand as a granular material in one large cavity. Fig. 3 shows the developed oil pan bottom configuration that is in the following referred to as prototype I. The individual components are connected via bolts. The new bottom weighs 1994 g and has a capacity of 760 g with respect to the chosen type of sand. The original bottom is significantly Fig. 3: Design of the oil pan bottom prolighter than the prototype I as it only weighs totype I, with one large cavity for granu1095 g. The frequency responses of the original lar material. oil pan, the empty and fully filled prototype I are shown in Fig. 4. It is easy to observe that the filled version of prototype I has significant lower amplitudes than the original bottom and the empty prototype bottom. However, the presented concept is 2.5 times heavier than the original configuration and thus no longer attractive for real applications. Therefore, the thickness of the oil pan bottom was significantly reduced from 4 to 1 mm and consequently a drastic reduction in mass to 553 g was realized. In this lightweight design also the height of the spacers has been reduced from 11 to 5 mm and consequently the capacity is also decreased to 347 g of sand. Thus, the fully filled lightweight concept is 195 g lighter than the original version. The frequency response of the original bottom as well as the filled and unfilled lightweight concept are given in Fig. 5. The lightweight concept shows lower vibration amplitudes than both the original bottom and the empty lightweight bottom. Table 1 summarizes the maximum and the averaged vibration amplitudes of the presented configurations. The filled lightweight concept is able to reduce the maximum amplitude by 26 dB and the average 150 140 130 120 110 100 90 original oil pan bottom 80 empty bottom prototype I 70 sand-filled bottom prototype I 60 0 500 1000 1500 2000 Frequency [Hz] Fig. 4: Frequency response of the original oil pan bottom, the empty and filled prototype I.

4 4

Vibration reduction in automotive applications based on the damping effect …

Amplitude [dB]

amplitude by 8 dB, at a 17 % loss in weight. In a last step of the study by Duvigneau et al. [5] it was shown that the sand position influences the vibration behavior, when partial fillings are used. 150 140 130 120 110 100 90 80 70 60

original oil pan bottom empty lightweight bottom sand-filled lightweight bottom 0

500

1000 Frequency [Hz]

1500

2000

Fig. 5: Frequency response of the original oil pan bottom, the empty and the filled lightweight bottom.

Table 1: Maximum and average amplitudes of the different bottom configuration and difference to the original configuration [5].

Configuration

Maximum

Original Empty prototype I Filled prototype I Empty lightweight Filled lightweight

148.24 dB 134.84 dB 109.55 dB 147.24 dB 121.78 dB

Difference to max. original -13.40 dB -38.69 dB -1,00 dB -26.46 dB

Average 113.38 dB 105.52 dB 92.83 dB 115.52 dB 105.1 dB

Difference to average orig. -7.86 dB -20.55 dB +2.14 dB -8.28 dB

Vibration reduction of an electric engine Another example, where the vibration behavior is reduced using granular materials stored in several large cavities, is presented by Duvigneau et al. in [7]. In this study the acoustic behavior of an electric engine is investigated. In this context, different filling materials such as sand, balsa wood and two different foams (OC Form 500 and OC Form 1000) are compared due to their damping properties. The cavities, which are placed in the front and back side of the engine housing, are fully filled, whereby the positioning has no influence. Fig. 6 shows the sand filled part of the housing of the electric engine before it was closed by an additional metal sheets and assembled with the other parts of the housing. The

5

5

Vibration reduction in automotive applications based on the damping effect …

corresponding frequency response of the closed part for selected fillings is presented in Fig. 7. It can be seen that both foam and sand allow a significant reduction of the vibration amplitude. Indeed, the mass of the inserted foam was significantly lower. For this reason, in section “Experimental investigation of alternative materials” additional filling materials are investigated. Furthermore, the modes are shown, which are only slightly influenced by the filling. It can also be observed that the filling causes a frequency shift. Fig. 6: Open part of the housing of an electric engine fully filled with granular material in large cavities [7].

Fig. 7: Frequency responses of the housing of the electric engine without filling, with Balsa wood filling, with sand filling and two different foam fillings. .

Combination of a honeycomb structure and granular material In [5] a significant influence of the position of the granular material on the vibration behavior of the oil pan bottom was demonstrated. In order to prevent an undesired distribution of the granular material, many small cavities are preferable, which allow an expedient filling. Therefore, a honeycomb structure is used, which offers also a high stiffness to weight ratio. The design consists of two plates. In contrast to the upper one, which seals the oil pan, the lower plate has a honeycomb structure attached and is shown in Fig. 8 with a partial sand filling in the region near the boundary of the plate. With the help of

6

6

Vibration reduction in automotive applications based on the damping effect …

Fig. 8: Honeycomb bottom with sand filling at the border area.

Fig. 9: Vibration behaviour of the (a) Empty honeycomb bottom; (b) Honeycomb oil pan bottom with a uniformly distributed sand filling (310 g).

this bottom design the influence of different distributions of the granular material on the resulting vibration behavior was investigated by Koch et al. [6]. The mass of the empty bottom is 730 g while it has a capacity of 1075 g of the chosen sand. One important goal is that the novel oil pan including the granular filling is lighter than the original one. Therefore, only a partial filling with 310 g of sand is allowed. Fig. 9(a) shows the vibration behavior of the empty honeycomb bottom and (b) the vibration behavior of the honeycomb bottom uniformly distributed filled of 310 g sand. It is clearly visible, that this mass is sufficient to achieve a significant vibration reduction. Different sand distributions (310 g) and the resulting RMS values are shown in Fig. 10. It is obvious that the positioning has a significant influence on the resulting vibration behavior. It was determined that the sand has to be placed at the position of the largest vibration amplitudes, due to the fact that the dissipative effect is a result of friction and impacts between the granular particles [6]. An improved filling strategy was determined by numerical (FEM) and experimental (laser scanning vibrometer) investigations. This methodology is illustrated in Fig. 11. Fig. 12 shows the frequency responses of the original, the empty honeycomb and the honeycomb oil pan bottom with the best experimentally determined filling. One easily observes that the filled honeycomb bottom shows an improved vibration behavior compared to the original bottom, although the honeycomb bottom has a lower mass, even including the 310 g sand filling. Furthermore, two additional types of sands with different grain size distributions have been examined. Compared to the initially chosen sand only negligible improvements could be achieved.

Fig. 10: Different sand distributions and the resulting vibration behaviour.

7

7

Vibration reduction in automotive applications based on the damping effect …

Fig. 11: Determination of the best granulate position using a laser scanning vibrometer. 110

Amplitude [dB]

100 90 80 70 original oil pan bottom empty honeycomb bottom experimentally optimized filling of the honeycomb bottom

60 50 0

250

500

750 1000 1250 1500 Frequency [Hz] Fig. 12: Frequency response of the original oil pan bottom the empty honeycomb bottom and the honeycomb bottom with the best filling of 310 g sand.

Experimental investigation of alternative materials In this sections the question will be answered, whether the damping effect of other materials like soft granules is better than the effect of the previously investigated sand. Besides, the influences of the shape and the grain size of the materials are studied. Therefore, the honeycombs are filled with different types of materials. Thereby, soft granulates and soft substances like gel, different kinds of glass and corundum are tested and compared against the previously used sand. Contrary to the previous sections a simple plate is investigated, but the honeycomb dimensions are the same as in the previous. Rectangular plates (300mm x 600mm) were used in this study. In contrast to the previous experiments, the plate is excited by a shaker with a coupling rod instead of an impulse hammer, in order to stimulate a higher frequency range using white noise. Fig. 13 shows the experimental setup of this investigation.

8 8

Vibration reduction in automotive applications based on the damping effect …

Filling materials We investigated 8 different materials such as (a) sand, (b) granular rubber, (c) two different grain sizes of granular glass, (d) corundum, (e) polystyrene, (f) glass balls, (g) silicone and (h) gelatin based gel. Fig. 14 depicts the investigated materials (a)-(f). Material (a) is the sand exhibiting the best performance in previous studies [6]. This sand can now be used as a reference to assess the properties of the other filling materials. Moreover, silicone (g) and a gelatin based gel (h) are used as soft nongranular material. As an example for a soft granular material granular rubber (b) is used. The influence of the particle shape can be roughly estimated by comparing the granular glass with the glass balls. The glass balls have a diameter of 3 mm and a low inner friction coefficient. Consequently, a low friction loss is expected. Two different grain sizes (0.2-0.4 mm and 0.4 – 0.6 mm) of granular glass (c), are investigated, in order to determine this influence. To determine the

(c)

(d)

(e) (b) (a)

Fig. 13: Experimental setup: (a) Laser-Scanning-Vibrometer; (b) Amplifier; (c) Shaker; (d) Honeycomb plate; (e) Control and postprocessing unit.

(a)

(b)

(c)

(d)

(e)

(f)

Fig. 14: Different investigated granular materials: (a) Sand; (b) Granular rubber; (c) Granular glass; (d) Corundum; (e) Polystyrene; (f) Glass balls.

9

9

Vibration reduction in automotive applications based on the damping effect …

influence of the inner friction, corundum (d) with a high friction coefficient is examined. Fig. 14(e) shows polystyrene, which is a very light granular material.

Results

Amplitude [dB]

In Fig. 15 the frequency response for a sand and corundum filling is compared. As long as no other specification are given, 620 g of the specific filling material are used, which is equal to the mass of the honeycomb plate. In addition, the frequency response of the empty honeycomb plate is shown in gray. Up to 165 Hz a negligible reduction of the vibration amplitudes is observed while a significant frequency shift towards lower frequencies is seen. The amplitudes of the mode at 166 Hz are even slightly elevated. Above 166 Hz a significant vibration reduction is realized. The aforementioned frequency shift can be attributed to the mass effect of the filling. The vibration behavior of sand and corundum is similar over the entire frequency range. The frequency response of glass particles of different grain sizes and shapes are shown in Fig. 16. It is easy to recognize that the frequency responses are quite similar. Apparently it is not of importance whether spheres or granular particles are used, also the grain size seems to have hardly any influence. The comparison of sand and glass balls is depicted in Fig. 17. Again, there is hardly any difference between the different granulates. The frequency response of the honeycomb plate filled with granular rubber, which is used as an example of soft granules is shown in Fig. 18 and compared to sand. Up to 50 Hz the vibration behavior is quite similar. At higher frequencies, a significant vibration reduction is seen using the granular rubber. Obviously, the soft granules cause higher dissipative effects than the stiffer ones. This leads to the question how materials such as silicone affects the vibration behavior, since it is soft but not granular. Fig. 19 compares silicone and granular rubber. While 620 g of granular rubber completely fill the honeycomb structure, 620 g of silicone are filling only a few cells. The lower frequency level of the plate with granular filling is clearly visible. A complete filling of a gelatin-based gel compared to granular rubber is shown in Fig. 20. The mass of this gel is 1300 g, thus twice as much as the rubber. Obviously, a larger frequency shift occurs caused by the additional mass. The vibration amplitudes are in the whole frequency domain higher than for the granular filled plate. As a further filling, polystyrene is tested and compared with granular rubber. Due to the low 145 140 135 130 125 120 115 110

Empty Plate Sand Corundum

0

50

100

150

200

250 300 Frequency [Hz]

350

400

450

500

Fig. 15: Frequency response of the honeycomb plate without filling, a filling of sand and corundum.

10

10

Vibration reduction in automotive applications based on the damping effect …

Amplitude [dB] Amplitude [dB] Amplitude [dB] Amplitude [dB] Amplitude [dB] Amplitude [dB]

145 145 140 145 140 145 135 140 135 145 140 130 135 130 140 135 125 130 125 135 130 120 125 120 130 125 115 120 115 125 120 115 110 110 120 115 110 0 0

Empty Plate EmptyGlass Plateballs Empty Plate Glass balls Granular glass I Empty Plate Glass balls Granular glass I glass II Granular Empty Plate Glass balls Granularglass glassIII Granular Glass ballsglass Granular glass III Granular Granular Granularglass glassI II Granular glass II

Amplitude Amplitude[dB] [dB] [dB] Amplitude Amplitude [dB][dB] Amplitude Amplitude [dB]

Amplitude Amplitude [dB] [dB] Amplitude [dB] Amplitude [dB] Amplitude [dB][dB] Amplitude

Amplitude Amplitude [dB] [dB] Amplitude [dB] Amplitude [dB] Amplitude [dB][dB] Amplitude

50 100 150 200 250 300300 350 350 400 400 450 450500 500 50 100 150 200 250 115 [Hz] Frequency 110 0 50 100 150 200 Frequency 250 [Hz] 300 350 400 450 500 110 [Hz] 300 0 Frequency 50 response 100 of the 150honeycomb 200 Frequency 250without 350 400glass 450 balls Fig. response plate filling, a filling of balls and 500 Fig. 16:16:Frequency the honeycomb plate without filling, a filling of glass and Frequency [Hz] 0 Frequency 50 100 sizes. 150honeycomb 200 250 without 300filling,350 Fig. 16: response of the plate a filling400 of glass450 balls and500 granular glassin intwo two grain granular glass grain sizes. Frequency [Hz] Fig. 16: Frequency response of the honeycomb plate without filling, a filling of glass balls and granular glass in two grain sizes. 145 145 Fig. 16: Frequency response of the honeycomb plate without filling, a filling of glassEmpty balls and Plate Plate granular Empty 145 140 glass in two grain sizes. 140 granular glass in two grain sizes. Empty Plate Sand 145 140 Sand 135 Empty Plate 145 135 Sand Glass balls 140 Empty Plateballs 135 130 Glass Sand 140 Glass balls 130 135 Sand 130 125 Glass balls 135 125 130 125 Glass balls 120 130 120 125 120 115 125 120 115 115 110 120 115 110 50 100 150 200 250 300 350 400 450 500 110 0 115 110 00 50 100 150 200Frequency 250[Hz] 300300 350350 400 400 450 450 500 500 50 100 150 200 250 11017:0 Frequency Fig. plate without filling, a350 filling of sand and Frequency [Hz] [Hz] 50 response 100 of the 150honeycomb 200 Frequency 250 300 400 450glass 500 50 response 100 of the 150honeycomb 200 Frequency 250without 350 400 450 glass500 [Hz]300filling, balls. Fig. 17:0Frequency Frequency plate a filling of sand and Fig. 17: response of the honeycomb plate without filling, a filling of sand and glass Frequency [Hz] filling, a filling of sand and glass Fig. 17: Frequency response of the honeycomb plate without 145 balls. balls. Empty Plate Fig. 17: Frequency response of the honeycomb plate without filling, a filling of sand and glass balls. 140 145 145 Sand balls. Empty PlatePlate 145 Empty 135 140 140 Empty Plate Granular rubber 145 Sand 140 130 Sand 135 Empty 135 Sand Plate 140 Granular rubber 135 125 Granular rubber 130 Sand Granular rubber 130 135 130 120 125 Granular rubber 125 130 115 125 120 125 120 110 120 115 120 115 50 100 150 200 250 300 350 400 450 500 115 110 0 Frequency [Hz] 115 110 110 0 50 100 150 200 250 300 350 400 450 500 Fig. response honeycomb plate without filling, a filling sand and granular 110 18: [Hz] 300 00 Frequency 50 100 of the 150 200Frequency 250 300 350 350 of 50 100 150 200 250 400 400 450 450 500 500 rubber. 0 50 100 150 200 250 300 350 400 450 500 Frequency [Hz]filling, a filling of sand and granular Frequency [Hz] Fig. 18: Frequency response of the honeycomb plate without 145 Frequency [Hz] Fig. 18: Frequency response response of plate without filling, a filling of Empty sand and granular Fig. 18: Frequency ofthe thehoneycomb honeycomb plate without filling, a filling of sand granular rubber. Plateand 140 Fig. 18: and granular 145Frequency response of the honeycomb plate without filling, a filling of sand rubber. rubber. Granular rubber Empty Plate 135 rubber. 145 140 145 Silicone Empty Plate Granular rubber Empty Plate 145 130 140 135 Empty Plate 140 Granular rubber 140 Silicone Granular rubber 125 135 130 Granular rubber 135 Silicone 135 120 Silicone 130 125 Silicone 130 130 115 125 120 125 125 110 120 115 120 0 50 100 150 200 250 300 350 400 450 500 120 115 110 Frequency [Hz] 115 115 0 50 100 150 200 250 300 350 400 450 110 Fig. 19: Frequency response of the honeycomb plate without filling, a filling of granular rubber 500 110 0 50 100 150 200 Frequency 250 [Hz] 300 350 400 450 500 and110 silicone. 50 response 100of the 150 250[Hz]300 300 350 350 of 400 450rubber 450500 500 Frequency 50 100 150 200200 250 400granular Fig. 19: 0Frequency honeycomb plate without filling, a filling [Hz][Hz] Frequency and silicone. Fig. 19: Frequency response of the honeycombFrequency plate without filling, a filling of granular rubber

11 Fig. Frequency plate without filling, a filling of granular rubber Fig. Frequencyresponse responseofofthe thehoneycomb honeycomb plate without filling, a filling of granular rubber and19: silicone. and silicone. and silicone. 11

11 11 11

11

Amplitude [dB]

Vibration reduction in automotive applications based on the damping effect … 145 140 135 130 125 120 115 110

Empty Plate Granular rubber Gelatin based gel

0

50

100

150

200

250 300 Frequency [Hz]

350

400

450

500

Amplitude [dB]

Fig. 20: Frequency response of the honeycomb plate without filling, a filling of 620 g granular rubber and 1300 g of a gelatine base gel. 145 140 135 130 125 120 115 110

Empty Plate 620 g granular rubber 110 g polystyrene 110 g granular rubber

0

50

100

150

200

250 300 Frequency [Hz]

350

400

450

500

Fig. 21: Frequency response of the honeycomb plate without filling, a filling of 110 g polystyrene and 110 g granular rubber.

density of polystyrene, only 110 g can be filled into the honeycomb structure. Fig. 21 compares the vibration behavior of polystyrene and granular rubber for a filling of 110 g. Even the small mass of 110 g reduces the peak amplitudes in the range of 125-250 Hz each by 2 dB. Which of the two materials has the better damping effect depends on the frequency under consideration, but the vibration behavior is largely comparable. In summary, it can be stated that the granular rubber and polystyrene have the best damping effect of all investigated materials. However, only a small amount of polystyrene (110 g) can be filled in the structure due to the limited volume of the honeycomb plate. A full filling with granular rubber (620 g) would cause significant lower vibration amplitudes (see Fig. 21).

Attenuation of selected modes In this section it is examined, whether it is possible to suppress selected modes by an adapted positioning of the granular material in the honeycomb structure. Therefore, the same plate and experimental setup as depicted in Fig. 13 are used. Fig. 22 shows the frequency response of the empty honeycomb plate and selected eigenmodes (a)-(d). In order to show whether it is possible to suppress certain modes, the best investigated material, granular rubber, is used. In addition, the frequency response of a partial filling of 234 g (e) is presented. The filling material is distributed at the positions with large amplitudes

12

12

Vibration reduction in automotive applications based on the damping effect …

Fig. 22: Frequency response of the empty honeycomb structure and with a filling of 234 g granular rubber (e). (a) – (d) show some eigenmodes, whereby (c) is the mode, which should be damped with the filling.

in the empty plate at 206 Hz (c). It is clearly visible that this mode is strongly damped. The mode at 106 Hz (a), which has low amplitudes at the positions of the filling, is nearly unchanged. The filling has a large influence on the modes at 166 Hz (b) and 233 Hz (d), but not as strong as the mode at 206 Hz (c). Consequently, it is possible to increase the damping for selected modes in a certain frequency. However, the filling also has an influence on other modes. The strength of the influence depends on whether the greatest amplitudes take place in the same regions.

Comparison of the honeycomb plate under horizontal and vertical suspension Finally, in this section the question is answered, how does the vibration reduction behavior in a horizontal and vertical suspension differ. Therefore, the plate and experimental setup depicted in Fig. 13 is rotated by 90°. Fig. 23 shows the frequency response of the empty and glass ball filled honeycomb structure horizontally and vertically mounted. One clearly sees that the amplitudes of the horizontal and vertical empty plate are quite similar although a frequency shift is visible. This results from the changed boundary conditions. While the amplitudes of the empty configurations are largely the same, they are significantly influenced in certain frequency ranges in the filled vertically mounted variant. This was unexpected, as the granular material is now not perfectly in contact with the measured

13

13

Vibration reduction in automotive applications based on the damping effect …

surface. Instead the main direction of motion is orthogonal to the gravitational force. Consequently, the particles can move more easily compared to the case of the horizontally mounted plate and therefore they can dissipate more energy by impacts and friction between each other. 145 Empty plate horizontal Empty plate vertical Glass balls horizontal Glass balls vertical

Amplitude [dB]

140 135 130 125 120 115 110 0

50

100

150

200 250 300 Frequency [Hz]

350

400

450

500

Fig. 23: Frequency response of the honeycomb plate without filling and filled with glass balls.

Summary The previous studies have shown that it is possible to reduce the vibration behavior without additional mass, using granular materials. Therefore, two concepts were presented, which use one or many cavities for the filling with granular materials. This investigation has shown that soft particles such as granular rubber have a larger damping effect than stiffer ones. It has also been shown that neither size nor shape have a significant influence on the vibration behavior, using solid granules. Additionally, the granular rubber results in an increased vibration reduction than a gel, which is soft but not granular. It could be confirmed that it is possible to reduce the amplitudes of chosen modes stronger than the remaining ones. However, the filling also has an influence on the other modes, depending on the similarity of the modes. In a last step it was shown that the presented lightweight concept can be used almost independent of the mounting (vertically or horizontally). Hence, the damping effect is only slightly effected by orientation.

Acknowledgements The presented work is part of the joint project COMO III (COmpetence in MObility), which is financially supported by the European Union through the European Funds for Regional Development (EFRE) as well as the German State of Saxony-Anhalt (ZS/2016/04/78118). This support is gratefully acknowledged. The authors would also like to thank the colleagues from the chair of Energy Conversion Systems for Mobile Applications (EMA) of the Institute of Mobile Systems of the Otto von Guericke University Magdeburg for the idea to use polysterene.

14

14

Vibration reduction in automotive applications based on the damping effect …

Literature [1] F. Duvigneau, T. Luft, J. Hots, J. L. Verhey, H. Rottengruber, U. Gabbert, “Thermoacoustic performance of full engine encapsulations - A numerical, experimental and psychoacoustic study”, Applied Acoustics, Volume 102, 15 January 2016, Pages 79-87. DOI: 10.1016/j.apacoust.2015.09.012 [2] T. Hering, Strukturintensitätsanalyse als werkzeug der maschinenakustik, Ph.D. thesis, Offenbach am Main (2012). [3] Schrader, P.; Duvigneau, F.; Luft, T.; Gabbert, U.; Rottengruber, H.: Development, Simulation and Experimental Investigation of a Function-Integrated and Foam Damped Oil Pan for a Two Cylinder Diesel Engine, 44th International Congress and Exposition on Noise Control Engineering – Inter-Noise 2015, San Francisco, 2015 [4] F. Duvigneau, S. Liefold, M. Höchstetter, J. L. Verhey, U. Gabbert, Analysis of simulated engine sounds using a psychoacoustic model, Journal of Sound and Vibration 366 (2016) 544–555. [5] F. Duvigneau, S. Koch, E. Woschke, U. Gabbert, An effective vibration reduction concept for automotive applications based on granular-filled cavities, Journal of Vibration and Control (2016) [6] S. Koch, F. Duvigneau, R. Orszulik, U. Gabbert, E. Woschke, Partial filling of a honeycomb structure by granular materials for vibration and noise reduction, Journal of Sound and Vibration. Volume 393, 2017, pp. 30-40. [7] F. Duvigneau, U. Gabbert, Numerische und experimentelle Schwingungsanalyse eines Radnabenmotors zur Entwicklung akustischer Maßnahmen, in: DAGA 2017 - 43. Jahrestagung für Akustik, 2017.

15 15

State-of-the-art digital road noise cancellation by Harman Dr. Nikos Zafeiropoulos, Jürgen Zollner, Dr. Vasudev Kandade Rajan

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 W. Siebenpfeiffer (Hrsg.), Automotive Acoustics Conference 2017, Proceedings, https://doi.org/10.1007/978-3-658-20251-4_4

1

State-of-the-art digital road noise cancellation by Harman

1 Introduction to digital RNC Structure-borne road noise dominates the cabin of modern vehicles, especially in the case of electric vehicles (EVs). Several RNC prototype systems based on analogue accelerometers are available in the market. So far the placement of these sensors has been based either on random optimisation methods or on a rule of thumb that at least 3-6 reference signals per wheel are necessary. An improved method for selecting sensor locations and optimising their performance is demonstrated in this paper. The method takes into account suspension and axle locations that are important for NVH/CAE prognosis, TPA and modal analysis of structure-borne road noise. The locations found from the method are exactly or at least similar to the NVH simulations and measurement methods. The sensor analysis method is developed on a multiple coherence basis, in order to create a coherence map between the reference acceleration inputs and the cancellation locations. In this way, the most coherent road noise dynamics are recognised for each structure-borne road noise resonance. In this analysis it can also be observed that below 80 Hz the sensor must be arranged according to modal analysis of front and rear suspension and definitely be symmetrical in order to control road drum. As for road rumble (80-150 Hz), the problem becomes more complicated since several front and rear axle contributions are important for different road and driving conditions. In this frequency range, the sensor locations and directions must be carefully selected as several dynamics contribute to the noise. For example, if the main source of road rumble is located at the rear axle, the front axle will still have some residual contributions on particular surfaces. This means that the controller must also be able to observe secondary structural contributions. At the frequency range of tyre cavity frequency range (170 – 250 Hz), the sensor selection can be obvious as the tyres are the sources, but in many cases several suspension parts can provide highly coherent inputs. This is caused by the fact that they are structure-borne propagation paths of the tyre’s resonant frequencies. These paths might also be known from TPA analysis and in this case, the phase reference input is crucial for controller stability issues. Fixed coefficient DSP filtering techniques will increase the noise level of the tyre cavity resonance in the cabin due to instabilities. These instabilities occur because a fixed filter system cannot track the amplitude and phase changes that appear dynamically in the tyre cavity. Above 250 Hz for mid and high frequency road noise, more localised vibrations occur and the sensor selection can become asymmetrical as these mid-high frequency road noise resonances do not appear in all passenger seats. All the previous locations are validated with a novel multiple coherence analysis. Lastly, predictions with the use of the adaptive algorithm that is implemented in the DSP processor are carried to validate the control strategy in the vehicle. As a second step, the audio amplifier that contains the adaptive algorithms is integrated with the vehicle electronics and audio system in order to start the RNC tuning on the road. The DSP

2

State-of-the-art digital road noise cancellation by Harman

amplifier uses digital automotive accelerometers that are mounted at the locations that are found from the previous coherence path analysis and their output is fed into a modern DSP audio amplifier platform. A novel adaptive algorithm processes the digital acceleration signals with high convergence and reaction time for various speed and surface ranges, in order to maintain high audible effects for the passengers. Several modern vehicle platforms are integrated with the presented digital RNC system with four ANC microphones at the headliners and the standard audio loudspeaker setup in order to prepare the technology with the existing audio layout of the vehicle and integrate with other active sound technologies.

1.1 Structure-borne road noise and multiple coherence Structure-borne road noise problems require multiple reference adaptive controllers that can observe and compensate for changes in structural vehicle inputs and in terms of the noise responses in the cabin. Therefore, before applying the RNC technology to a vehicle, it is crucial to understand the changes in the vibration inputs and how they correlate with the road noise perceived by the passengers in the cabin. Typically, TPA is the most appropriate tool to fully analyze the vehicle in terms of structural road noise sources and their propagation paths [1]. Transfer path analysis of structureborne road noise can identify the road forces fc at suspensions and other axle parts where road vibration can be generated or propagated. The following diagram presents the source-path-receiver model that is used for TPA as well as RNC coherence analysis since the same locations on the vehicle are used for the analysis and selection of sensor placement for RNC application.

Figure 1 – Source path receiver model for TPA analysis that is also assumed for RNC sensor analysis and placement. Source path receiver model (a): sources at the axle (b): receiver responses, (c): connections between the axle and the body structure.

3

State-of-the-art digital road noise cancellation by Harman

The forces at various points on the axle are denoted as fa and the forces at various connection points are presented as fc. The vibration paths between the tyre and the connections are shown as transfer mobilities Yac and the point mobilities at the connections are noted as Ycc. The vibro-acoustic paths between the connections and the interior sound pressure Hcb and the primary paths of structure-borne noise are noted as Hba. The error between the estimated road noise responses from TPA and the measured road noise responses is denoted as e. The structure-borne road noise inside the vehicle is a result of the excitation road forces that act on the vehicle multiplied by the primary vibro-acoustic paths, in the frequency domain as =

.

With use of source, path and receiver sub-structuring theory, the vehicle structure can be similarly decomposed as: 1. sources, where road forces are generated 2. paths, connection points between vehicle parts 3. receiver, the cabin. In terms of RNC and multiple coherence analysis, the sensors are also mounted at candidate locations as in TPA that can be either sources or paths and then later be selected for the application based on digital sensor placement constraints, cable harness and manufacturing. If the TPA part is substituted from figure 1 and replaced by a feed-forward active controller, then the similarity of TPA and RNC can be immediately highlighted.

Figure 2 – System diagram for Transfer Path Analysis and RNC. fa: road forces, Hba: vibroacoustic paths, T: Structural transfer paths from the source to the reference sensor locations, W: control _lter matrix, C: electro-acoustic paths from the loudspeakers to the microphones at the headrests, x: vibration responses as reference inputs to the controller, y: control signals that drive the loudspeakers in the cabin, pb, p^b: are the measured and synthesized road noise spectra and e: the residual error signals at the cancellation point in the cabin.

The system matrix T contains all the vibration paths that arrive at the sensor locations, which are also coupled depending on their location on the structure. This coupling

4

State-of-the-art digital road noise cancellation by Harman

causes cross-coupling between the directions and thus the acceleration signals at each location are partially correlated. This signal mixing can degrade the adaptation performance of adaptive algorithms, and several algorithms with decorrelation techniques have been proposed in the past. However, these algorithms can significantly increase the computational complexity of the RNC system and thus a novel adaptive algorithm is introduced in the following section. In figure 3, the input signal analysis for the sensor is presented that is multiple coherence-based in order to select the best location for achieving the NVH that is set by the OEM. From this coherence map between the sensors and the microphone, the most coherent locations can be identified that can support the algorithm to observe fast changes in road noise inputs.

Figure 3 – Multiple coherence map between the candidate paths and the 4 RNC microphones.

1.2 Novel RNC system The novel RNC control that is presented here is partially introduced in [2] and [3]. The algorithm features are explained in this paper and some performance examples are given with different modern vehicle platforms.

2 Novel RNC algorithm A FX-LMS algorithm was selected due to its advantages in convergence and speed [5] and its proven ability to handle ANC problems in various applications. In contrast to other feed-forward ANC systems, here the reference signal acquisition uses multiple strategy placed digital accelerometer sensors. These sensors measure the structure-

5

State-of-the-art digital road noise cancellation by Harman

borne road noise transmitted through the suspension and radiated via the vehicle body through the cabin to the passengers’ ears. This transmission path is known as the primary path. Between the error signals, which are recorded via error microphones located close to the passengers’ heads, and the vehicle loudspeakers, which are used as actuators, the secondary path is defined and well considered within the FX-LMS algorithm as shown in Figure 4.

Figure 4 – Block diagram for the multichannel fully digital RNC-System utilizing FX-LMS.

The challenge for the algorithm convergence is to generate a global road noise cancellation or at least local cancellation around the error microphone positions. It has to be considered that both the accelerometer channels as well as the error microphone channels are prone to disturbances such as other undesired cabin noise and audio from the entertainment system. In order to achieve a reasonable performance at all passenger seat positions, several considerations have been taken into account. Since the algorithm uses time domain, sample-by-sample - low latency, convolution for the reference signals xr within the forward path to calculate the output signals yl, the system is able to meet the causality criteria for stochastic signal cancelation. But spectral domain calculation techniques are used to generate spectral domain reference and error signals Xr and Em to reduce the calculation effort required to perform the update and to populate the w-filter matrix wr,l. Therefore, utilizing such multichannel MIMO FX-

6

State-of-the-art digital road noise cancellation by Harman

LMS algorithms is a challenge in design, implementation, testing, tuning and maintenance in a real-time embedded processing platform, since real-time applications implemented in automotive graded processors are limited in costs, which mostly implies limitations in memory space and calculation power. Additionally, the reference signals provided by the digital accelerometer sensors need special attention. Sophisticated signal conversion and pre-processing was implemented to maximize the signal quality without adding unwanted delay. Besides that, advanced algorithm enhancements, including normalization and other modifications, were made to improve the strength and stability of the algorithm without compromising on performance.

3 RNC system realization An RNC system in the real world is the amalgamation of know-how from different disciplines. These disciplines include signal processing and algorithmic development, NVH intelligence/information, MEMS and sensors development, loudspeaker technology, and hardware platform and software development. The algorithm and NVH aspects were explained in the previous sections. In this section, some of the other components of the RNC system are presented.

3.1 Accelerometer sensors The input signals for active feed-forward adaptive control play a significant role in the system’s performance. Below is a comparison between analogue automotive accelerometer (black line) and two different digital automotive sensors. The red line is the acceleration spectrum of a sensor with a high sensitivity and low acceleration g-range and the green line is the spectrum of high g-range digital sensor.

7

State-of-the-art digital road noise cancellation by Harman

Figure 5 – Accelerometer frequency response comparison. Black line: NVH measurement accelerometer spectrum as reference. Red line: Low g-range automotive sensor. Green line: High g-range.

The analogue sensor is used as a reference system as it can detect vehicle vibrations up to 8 kHz. The low g-range digital sensor can measure roughly up to 500 Hz, whereas the sensitivity of the high g-sensor above 200 Hz starts to roll-off. Another difference between the two digital sensors is the SNR difference as the low g-range has around 20 dB more SNR in this medium speed road noise measurement. This means that the sensor might have some impact on the RNC system performance as more gain in the amplifier side will be needed to boost the acceleration signals. However, this boosting must be carefully performed, in order to avoid enhancing the noise floor of the digital sensor, creating audible side effects for the passengers. The main benefit of the digital automotive sensors is that they can be integrated into the vehicle system and use modern standard digital protocols such as PSI-5, A2B etc.

3.2 Prototype platform In general, the amplifier platform is designed to offer a wide range of flexibility like a prototype device but incorporating OEM product features as well. The expected platform requirements have thus been evaluated.

8

State-of-the-art digital road noise cancellation by Harman

3.2.1 Platform requirements The elaborated platform is more than a development kit. It is designed to be a versatile tool for the NVH department and its functional owners to demonstrate digital RNC based on a higher level of maturity. Hence, the platform offers enough processing power and interface options to meet three major customer expectations. ● Meet the NVH targets ● Guarantee production readiness ● Support maintainability and customization The platform has to be an automotive graded prototype for the state-of-the-art digital RNC technology. So the amplifier is not only required to meet the NVH road noise targets and performance expectations. The algorithm also needs to be stable in all driving situations and withstand disturbances. Continuous fast adaptation without user interaction is mandatory to meet the above criteria, in addition to reliability and durability with regular usage. From a systems-engineering point of view, the tuning effort should stay reasonable and rapidly lead to a perceivable result. The platform is required to serve as a functional A-Sample; all embedded components have to be automotive graded and the component sourcing has to be aligned with a potential ECU development timeline and vehicle SOP. Besides that, the platform needs to enable seamless integration of the technology into the vehicle network, e.g. CAN network or the HU audio interface. The installed processing power needs to meet the standard ECU processing headroom of 30%, in order to allow individual customization and modifications. On top of that, the platform has to be embedded into an optional engineering service package, including a best practice digital RNC vehicle survey and reliable performance prediction, along experienced digital RNC integration consulting suggesting a cost effective solution.

3.2.2 Platform features Within all recent vehicle integrations, the presented platform shown in Figure 4 meets all above-mentioned requirements. The algorithm is executed on a modern DSP processor. In terms of I/O features, on the output side the amplifier includes twelve speaker output channels – providing up to 600W together. On the input side, the amplifier can handle up to 12 individual digital automotive acceleration sensors and up to 24 microphone inputs with a switchable bias support voltage. To interface a standard HU analogue output, four audio inputs are allocated. In addition, to access the vehicle telemetry and other comfort information, a low- and high-speed CAN interface was considered in the design. Besides that, eight auxiliary inputs and up to four robust battery voltage GPIOs are available on-board.

9

State-of-the-art digital road noise cancellation by Harman

Figure 6 – HARMAN digital RNC audio platform (top cover removed) using a modern DSP processor and digital automotive sensor inputs for acceleration reference channels.

Last but not least, a proprietary USB tuning interface was built in to enable online tuning of the digital RNC algorithm configuration as well as the audio processing elements or other useful on-board features. Once the platform hardware was available, the challenge of utilizing the FxLMS algorithm was to design, implement and test the software. As a last step, the RNC system is tuned and maintained such MIMO system on a real-time embedded platform.

3.2.3 Software architecture It is clear that the system must be implemented in a real-time framework. This basic requirement combined with memory and computational limitations poses considerable challenges in the software implementation of the RNC system. The causality condi-

10

State-of-the-art digital road noise cancellation by Harman

tion determines the latency that the RNC system can afford. This drives the processor and implementation to support sample-by-sample processing or at least small block size. The effort of software development is proportional to the underlying hardware and latency conditions. Furthermore, as stated, the software must also support parallel RNC and audio processing. This means that a multi-rate system supporting the standard 48 kHz audio sampling rate and the RNC sampling rate simultaneously is required. Finally, the software architecture must clearly identify blocks for potential optimization based on the signal flow shown in figures 3 and 5. Examples of such blocks include FFT/IFFT implementations, adaptive filter computations, filters in the audio processing block, etc.

3.2.4 Audio integration A major benefit of a software framework being able to handle FX-LMS algorithm and standard audio processing simultaneously is the integration of branded audio with RNC technology. Now, the platform prototype can be used as a standard audio amplifier ECU with RNC as an add-on feature.

Figure 7 – Signal flow illustrating audio processing in parallel to RNC.

An architecture example is depicted in Figure 7, but other architecture layouts could be considered due to the software framework’s flexibility. Several successful RNC demonstrations have shown the system readiness. The individual performance results are highlighted in the next chapter.

11

State-of-the-art digital road noise cancellation by Harman

4 RNC performance results Designing a complex system such as RNC requires a controlled test process. The system must also be validated across different vehicle types for performance. The results of the presented RNC system are divided into two parts: a. Bench performance under ideal conditions. b. Validation and performance in real road conditions.

4.1 Bench performance For the purpose of testing the performance under ideal conditions, a so-called Hardware-in-the-Loop (HiL) was implemented. This setup provides a lot of flexibility during the development of such systems. The idea of a HiL setup itself is slowly catching on across various automobile industries. A very simple setup consists of a workstation, a soundcard, and the device undergoing testing to simulate various driving conditions. Figure 8 shows the performance of the RNC system in a bench test scenario. The signal to be cancelled was chosen to be a shaped pink noise signal.

Figure 8 – RNC algorithm Hardware in the Loop performance of the Harman RNC system. Yellow line: Low-pass filter white noise for RNC off. Green line: RNC off.

It can clearly be seen from the plots that a cancellation of about 40dB is achieved. The broadband random noise is deliberately used as latency issues can be addressed, which helps optimise the DSP software and hardware. The harmonics that are seen in the graph as not audible and aliasing and other digital effects are at the noise floor. The DSP audio platform contains high quality ADCs and DACs and a floating point DSP so that these

12

State-of-the-art digital road noise cancellation by Harman

digital effects are kept at a very low level and do not interfere with the car audio. Therefore, RNC HIL tests are crucial not only for addressing latency issues but also algorithm signal quality that could potentially affect the system’s audio quality.

4.2 Vehicle performance The presented digital RNC with the prototype audio amplifier was integrated into several types, in order to achieve high NVH targets and audible effects. The operating range of RNC was defined according to the vehicle road noise problem and audible broadband reductions were achieved. The measurements that are shown here are NVH assessments with microphones in the headrests, far from the headliner microphones where cancellation levels are at their highest. Reductions up to 12 dB(A) were achieved in the main road noise resonances. The spectrogram analysis also reveals how consistent the performance is when driving over the same surface and that the road noise energy is also reduced to very low noise levels. The following measurement data present three cases ● a luxury sport vehicle driven at 50 km/h over rough asphalt in a city ● a luxury limousine driven over cobblestones at 30 km/h ● a hatchback on a highway at 80 km/h. In Figure 10, the spectrogram is presented up to 2 kHz in order to investigate any artefacts from the RNC. The system generates no artefacts due to the high audio quality floating point DSP and electronics that are used for achieving maximum SNR in the system.

Figure 10 – Spectrogram of the road noise for a luxury sport vehicle when it is driven at 50 km/h at a rough asphalt in a city.

13

State-of-the-art digital road noise cancellation by Harman

The second vehicle was a luxury limousine with great refinement, but with audible road noise issues at particular speeds and surfaces. The RNC systems achieves more than 10 dB(A) maximum reduction.

Figure 12 – Road noise spectrum of a luxury sport vehicle when it is driven at 50 km/h over rough asphalt in a city with high road noise cancellation above 10 dB(A) RNC.

The third case study was a high volume hatchback with dominant structure-borne road noise issues. The RNC system was able to cancel all of them and some test results from a test drive at a typical German Autobahn are presented here, where the vehicle still suffers from road noise and the NVH improvements from RNC were impressive.

14

State-of-the-art digital road noise cancellation by Harman

Figure 13 – Road noise spectrum of a hatchback vehicle when it is driven at 80 km/h on the German Autobahn.

5 Conclusion The latest state of the art multichannel ARNC amplifier platform development is summarized in this paper. A novel algorithm was implemented in a modern DSP audio amplifier and a new method for selecting sensor location in combination with the algorithms was presented. An RNC HIL test was described that is necessary for qualifying the system performance in terms of latency and audio signal quality before integrating the system into vehicle electronics. It was shown that the technology is ready to be industrialized, since the new platform addresses all OEM ECU development concerns and meets the NVH road noise cancelation targets. The platform can be fully integrated into vehicle architecture and then be tested 365 days a year. Since the adaptation is permanently active, the performance stays highly perceivable for targeted road noise scenarios. Daily life cases were also tested and RNC performance at different modern vehicles was presented after hours of testing different road conditions.

Acknowledgements The authors would like to dedicate this paper to the memory of Johannes Hammendinger, team member of the RNC core Harman, who actively contributed to the development of the RNC system. Special thanks to Gerhard Pfaffinger, Franz Lorenz, Dr. Markus Christoph, Dr. Gerald Bauer, Paul Zwickowski, Andreas Fritsch and Justin Cao for their contributions to the RNC technology and vehicle developments.

15

State-of-the-art digital road noise cancellation by Harman

References [1]

Elliott, A., A. Moorhouse, T. Huntley, and S. Tate (2013). In-situ source path contribution analysis of structure borne road noise. Journal of Sound and Vibration 332 (24), 6276-6295.

[2]

Zafeiropoulos N., Ballatore M., Moorhouse A., Mackay A. (2015). “Active control of structure-borne road noise based on the separation of front and rear structural road noise related dynamics”, Noise and Vibration Conference and Excibition, SAE International, Grand Rapids, Michigan, USA, 22–25 June, (2015).

[3]

Zafeiropoulos N., Christoph M., Zollner J., Kandade Rajan V. (2016). “Acive control of structure-borne road noise: A controller strategy based on the most significant inputs of the vehicle”, Proceedings of the Inter-Noise and Noise-Con Congress and Conference, Hamburg, Germany, pp. 551—557.

16

Acura NSX – Hybrid Acoustics, a new carpet concept for meeting stringent performance and weight targets Jared Cox, Steve Eich, Andrea Martin (Honda R&D Americas, Inc.) Rolf Balte (UGN, Inc.)

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 W. Siebenpfeiffer (Hrsg.), Automotive Acoustics Conference 2017, Proceedings, https://doi.org/10.1007/978-3-658-20251-4_5

1

Acura NSX – Hybrid Acoustics, a new carpet concept for meeting stringent …

Abstract The NVH development of the all new Acura NSX was a significant challenge. The innovative sports hybrid powertrain allows for a wide range of driving modes each with specific NVH requirements. From highly emotional engine sound to almost silent cruising in pure electric mode. To achieve these high NVH targets without a significant weight penalty required an innovative approach to the NVH treatment package. One key enabler was the carpet system which is based on the Hybrid-Acoustics Technology. Sufficient sound transmission loss performance was combined with a maximum of sound absorption in a lightweight system to meet the targets. This paper shows the theory behind “Hybrid-Acoustics” as well as comparisons of test and simulation results to classical mass based insulations and purely dissipative systems.

1 Introduction The ultimate goal for the new NSX is to be a leader in today’s cutting edge supercar market, but still pay tribute to the heritage of the original NSX, which was considered revolutionary and helped to build the Acura brand nearly 30 years ago. The primary focus for the sound concept was to maintain the heritage sound of the original naturally aspirated V6. The challenge was to achieve this with the new hybrid electric twin turbo V6 engine. A secondary goal was to also celebrate the EV driving capability of this next generation supercar. This was achieved by building on a foundation of a quiet interior, using sealing, NVH treatments, and road and engine isolation strategies...

1.1 Current Market Super Sports Car Landscape Automotive technology, construction and performance have changed significantly since the original NSX was launched, and the next NSX needs to satisfy the expectations of the customers in the current super sports car market. Traditional super sports cars are often defined by extremely sporty NVH characteristics, including loud combustion noise, powertrain mechanical noises, lack of road isolation, and strong vibrations. These vehicles tend to be raw, and unrefined, making for a very authentic sports sound experience, however they are also very tiring, tending to wear on a driver and making anything more than a short trip uncomfortable. This segment positioning is shown in Figure 1, along with the typical luxury sports sedan vehicle and Acura product placement.

2

Acura NSX – Hybrid Acoustics, a new carpet concept for meeting stringent …

Figure 1: 1990’s Sports Car Market Landscape

Figure 2: 2015’s Sports Car Market Landscape

Over the past 25 years, the introduction of electronic controlled powertrain and sound devices permitted manufacturers to better balance the general driving NVH refinement, and sports car sound. These technologies permit the driver to select their desired driving setting, within some performance window, as shown in Figure 2. Market analysis shows the customer expectation in this segment has been raised significantly since the release of the original NSX. Most buyers expect a vehicle that is reasonably luxurious and refined, suitable for longer drives, but that can also deliver a super sports car, visceral and raw engine sound experience.

2 Target Setting With this understanding of the current sports car market landscape and customer expectation, an engine sound concept was defined to deliver a New Sports eXperience (NSX). Every driving situation encountered by the customer was considered, from acceleration, to cruising, gear changes, etc. and then the desired sound performance in each of these driving situations and Integrated Dynamics System (IDS) modes is defined, as shown in Figure 3. IDS modes are key to this concept by creating a fully adjustable driving experience, where each mode provides the driver a different vehicle character including a tailored acoustic performance impression. Each IDS mode has a unique intended use, summarized in Figure 4.

3

Acura NSX – Hybrid Acoustics, a new carpet concept for meeting stringent …

Figure 3: Vehicle Sound Concept

Figure 4: IDS Mode Concept and Intended

Quiet Mode: The Quiet mode concept is to enable the driver to be un-detected by the outside world, minimizing noise and creating a comfortable, quiet cabin environment. Sport Mode: The Sport mode concept is to enable the driver to experience a New Sports eXperience (NSX), integrating hybrid vehicle driving with that of a super sports car with exceptional sound response to driver input, while maintaining everyday drivability. Sport+ Mode: Sport+ mode concept is focused towards winding or mountain road driving, eliminating IC engine stop and giving the driver increased on-road audible feedback and performance. Track Mode: Track mode concept is strictly for circuit use only, to achieve fast, repeatable lap times and give the driver the maximum acoustic feedback even audible while wearing a helmet.

3 Approach 3.1 Insulation Material Package Traditional super sports cars are more analogous to race cars when it comes to noise and vibration cabin quietness. Of course weight is the enemy of acceleration and handling performance, so it’s easy to choose lightweight or low coverage insulators in the interest of preserving maximum performance. The challenge with the NSX development is to achieve the concept of Quiet mode driving, while minimizing added weight from the material package.

4

Acura NSX – Hybrid Acoustics, a new carpet concept for meeting stringent …

This was addressed by specifying the package based on the noise source hierarchy. Figure 5 shows a representation of the input intensity of each noise source. Low Contribution

High Contribution

Figure 5: Air-Bourne Source Noise Contribution

As shown in Figure 5, the primary noise sources are behind the driver. Focus is then concentrated on treatment in this area, centralizing the weight in the rear bulkhead panel insulator, rear panel carpet, rear partition glass, and many other smaller parts to create high transmission loss from the engine room to the cabin. For the noise sources in the front of the vehicle, light weight absorptive materials are applied to the front bulkhead panel and wheel arch regions, where the heavier weight, barrier materials are not required.

3.2 Floor Carpet Initial Construction To maintain acceptable levels of road noise, the initial concept for the areas of the floor pan was to use a barrier based construction consisting of a 1.2 – 1.5 kg/m^2 heavy layer with foam pads. The idea behind this was to have sufficient insertion loss. NVH simulations undertaken by UGN, Inc. however indicated, that a new technology called “Hybrid Acoustics” could be used instead to even further lower the weight and improve the material NVH characteristics.

3.3 Hybrid Acoustics Concept The target of Hybrid Acoustics (HA) is to maintain the insulation performance of traditional barrier carpets below 1.5 kg/m^2 for the transmission loss layer (it is very difficult and expensive to make lightweight barriers with conventional EPDM heavy lay-

5

Acura NSX – Hybrid Acoustics, a new carpet concept for meeting stringent …

er) but at the same time generate additional absorption performance by using fiber technology. One of the main technical challenges is to avoid the usual insulation performance decrease when using fiber material on surfaces. By increasing the hard layer stiffness, it is possible to reduce the hard layer resonance. The most important acoustic effect is linked to a high dynamic compressional stiffness. By doing so, the static flexural (bending) stiffness is also increased which has the added benefit of good shape retention of the carpet The graph in Figure 6 illustrates the reduced insertion loss performance when simply converting a mass based barrier to a felt/film construction due to hard layer resonance.

Figure 6: Reduced Insertion Loss when using fiber based hard layers

The second dip at around 1600 Hz typically eliminates a fiber / film based hard layer from being used as a barrier, because of the degradation in insertion Loss. The correlation of hard layer resonance, its stiffness and the impact on the insertion loss was first investigated by Seppi et. Al. and the results were presented at the InterNoise 2012.1

1 Marco Seppi, Claudio Bertolini, Claudio Castagnetti (Autoneum Management AG) – “Analytical prediction of resonance effects for lightweight acoustic packages with increased insulation”, InterNoise 2012, New York City, USA

6

Acura NSX – Hybrid Acoustics, a new carpet concept for meeting stringent …

As stated in said paper, by adjusting the dynamic compressional stiffness of the hard layer, a potential alternative construction to the intended mass backed carpet in the NSX could be found (Figure 7). It has to be pointed out that the compressional stiffness is “physically responsible” for the resonance and not the bending stiffness.

Figure 7: Increasing Dynamic Stiffness in a Hybrid Acoustic Carpet Construction

3.4 Floor Carpet Simulation A Transfer Matrix Method (TMM) based simulation tool2 was used in order to efficiently investigate the proposed mass based systems and also various alternative carpet constructions using a variety of modelling parameters before making physical samples. Table 1 describes 6 of the constructions investigated used in the simulation including their surface weight. The total thickness of each construction was kept at 25mm. Carpet #1 is the originally proposed solution, while carpets #2 and #3 use the Hybrid Acoustics solution. Carpet #4 maintains the total mass weight of 1.3 kg/m^2 by splitting the massback layer in a felt and mass layer and carpet #5 increases the combined

2 P. Bertolo – “Visual SISAB: a visual tool for a quick and accurate assessment of the acoustic properties of car parts”, proceedings of Rieter Automotive Conference 2007

7

Acura NSX – Hybrid Acoustics, a new carpet concept for meeting stringent …

felt/mass layer to 2.6 kg/m^2. Last but not least, carpet #6 is a lightweight purely dissipative system with excellent absorption but poor IL – it was chosen to showcase an entirely different approach. Table 1: Carpet Constructions simulated # 1 2 3 4 5 6

Description Tufted carpet + EVA 1.3 kg/m^2 + PU 20mm Tufted carpet + DSL3 1300 gsm 3 mm + film 80 gsm + PU 17mm Tufted carpet + DSL3 900 gsm 2.5 mm + film 80 gsm + PU 18mm Tufted carpet + Felt 600 gsm 4 mm + EVA 0.7 kg/m^2 + PU 16mm Tufted carpet + Felt 600 gsm 4 mm + EVA 2.0 kg/m^2 + Felt 800 gsm Tufted carpet + AFR Felt 700 gsm 4 mm + Lofted Felt 700 gsm

g/m^2 2,840 2,840 2,480 2,680 4,040 1,900

The results of the simulation are presented in Figure 8 (Insertion Loss – IL) as well as in Figure 9 (Diffuse Field Absorption Coefficient).

Figure 8: Simulation Results – IL

3 DSL = Dynamic Stiffness Layer of a Hybrid Acoustic System

8

Acura NSX – Hybrid Acoustics, a new carpet concept for meeting stringent …

Figure 9: Simulation Results – Absorption

It is interesting to see, that the Hybrid Acoustic (HA) with a DSL 1300gsm construction (carpet #2) has a better IL than the same weight mass barrier (carpet #1) at frequencies below 3150 Hz. Only above this frequency has the mass system better insulation performance. On the other hand, in terms of absorption, the HA 1300 has far better performance over the entire frequency range than the mass based system of the same weight. Because of that fact, the assumption is, that the in-vehicle performance of the “Hybrid Acoustic” solution will be better. However, it is not enough to simply use any type of felt in this carpet system, but it is rather critical to have a DSL (Dynamic Stiffness Layer) felt. The compressional stiffness of this DSL layer must be tuned so that the IL does not drop significantly in the frequency range between 1250 and 4000 Hz. For example a drop of 8.1dB was observed with construction #4, and to some extent also in construction #5. It is possible to compensate for the drop in insertion loss at these critical frequencies without using the Hybrid Acoustic technology. However, the weight of the barrier mass has to increase to at least 2.0 kg/m^2 which makes that whole construction (#5) over 40% heavier than the original 1.3 kg/m^2 barrier construction (#1).

9

Acura NSX – Hybrid Acoustics, a new carpet concept for meeting stringent …

3.5 Floor Carpet Flat Samples In order to validate the findings from the simulation presented in the previous chapter, it was decided to construct and test flat samples of all of the simulated constructions with the exception of carpet # 4. Table 2 shows these constructions. Table 2: Carpet Constructions tested # 1 2 3 5 6

Description Tufted carpet + EVA 1.3 kg/m^2 + PU 20mm Tufted carpet + DSL4 1300 gsm 3 mm + film 80 gsm + PU 17mm Tufted carpet + DSL 900 gsm 2.5 mm + film 80 gsm + PU 18mm Tufted carpet + Felt 600 gsm 4 mm + EVA 2.0 kg/m^2 + Felt 600 gsm Tufted carpet + AFR Felt 700 gsm 4 mm + Lofted Felt 800 gsm

g/m^2 2,840 2,840 2,480 4,040 1,900

The results of the flat sample testing can be seen in Figure 10 (IL) and Figure 11 (Absorption).

Figure 10: Measured Results – IL

4 DSL = Dynamic Stiffness Layer of a Hybrid Acoustic System

10

Acura NSX – Hybrid Acoustics, a new carpet concept for meeting stringent …

Figure 11: Measured Results – Absorption

These results are in line with the simulation results shown previously with the exception of construction #5 performing worse than predicted in terms of IL. The difference is likely in the fact that no 800gsm felt was available at time of sample manufacture as a decoupler but rather a 600gsm material was used.

3.6 Floor Carpet – Conclusions With the information provided by the floor carpet simulation as well as the flat sample testing, it was decided to pursue the HA 900 (Carpet #3) solution, since it reduced the carpet part weight by about 1.0 kg without sacrificing the NVH insertion loss performance and improving the high frequency performance. The LHS portion of the carpet weighs 3.6 kg and the RHS portion 3.3 kg on the LHD vehicle. Early in the development it became clear, that one major challenge comes from the aggressive design of the part with its deep draw of more than 300mm and draft angles smaller than 5 degrees (Figure 12). With stretch factors greater than 50% in some area, initial manufacturing trials at the plant showed, that this is not achievable with mass barrier constructions, even by increasing the heavy layer weight significantly. The Hybrid Acoustic technology was the only way to follow the aggressive draws and

11

Acura NSX – Hybrid Acoustics, a new carpet concept for meeting stringent …

draft angles of the carpet not only in the tunnel area. The radii were reduced from initially 20 to 10 mm in the tunnel area which resulted in a very crisp design execution.

Figure 12: Carpet Draw / Draft Angles / Radii

The Hybrid Acoustics solution at 900gsm molds perfectly without issues. Last but not least, because of the high compressional stiffness of the HA 900 construction, which also increases the flexural stiffness, the carpet retains its shape without portions of it folding over, as it would be likely with a 1.3 to 1.5 kg/m^2 barrier construction. Wear and abrasion values are also in line with the specifications.

12

Acura NSX – Hybrid Acoustics, a new carpet concept for meeting stringent …

4 Summary The result of this strategic isolation concept using Hybrid Acoustics is dramatic and benefitted the vehicle concept in several ways: 1) Cabin Quietness: In Quiet and Sport modes the isolated cabin creates a smooth transition between IC engine running to full EV driving and is often imperceptible by the driver. EV driving is almost silent in the cabin and highway cruising noise level is more like a sports sedan than the typical super sports car. Figure 13 illustrates the interior road noise measured at 100kph on concrete road surface while the Cabin quietness competitiveness is shown in Figure 14, where NSX demonstrates best in class road noise performance.

Figure 13: Interior Noise – NSX vs. Competitor

13

Acura NSX – Hybrid Acoustics, a new carpet concept for meeting stringent …

Figure 14: Insulation Material Package – Road Noise Performance

2) Low Noise Floor: With a low noise floor, the undesirable noises are suppressed. This provides a clean palette for the addition of emotional combustion and power plant related sports sounds. Engine mechanical noise isolation during acceleration is shown in Figure 14, where NSX realizes best in class performance.

Figure 15: Engine Noise During Part Throttle Acceleration

14

Acura NSX – Hybrid Acoustics, a new carpet concept for meeting stringent …

Figure 15 shows how the NSX bridges the gap between sedan and super-car, accomplished through multiple personalities, available for the driver to access with the flick of a knob. This brings a truly New Sports eXperience, a no compromise super sports car, one that can be comfortably driven every day, short or long distances, spirited canyon road driving, or even on the racetrack. Application of the hybrid acoustic carpet was an integral part of the overall NVH strategy to achieve best in class performance at the lowest possible weight.

Figure 16: Final Sound Achievement Competitiveness

15

Powertrain mounting systems for electric vehicles David Roth, System Simulation Specialist, Dr. T. Ehrt, C. Bultel, H. Kardoes Vibracoustic GmbH & Co. KG, Germany

Zürich / Rüschlikon, Switzerland, 11-12 July 2017

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 W. Siebenpfeiffer (Hrsg.), Automotive Acoustics Conference 2017, Proceedings, https://doi.org/10.1007/978-3-658-20251-4_6

1

Powertrain mounting systems for electric vehicles

Introduction Initial situation Mounting systems for combustion engines are well established for many decades and are constantly being further developed and optimized. Even in the case of increasing demands, such as through downsizing, downspeeding, or lightweight construction on the one hand and increasing comfort demands on the other hand, the fundamental NVH (noise vibration harshness) issues for combustion engines are generally resolved through a wide range of relevant development tools and products. For the mounting of electric powertrains, the systems used today are typically based on concepts for combustion engines and adapted exclusively for the new application. However, electric powertrains differ significantly from conventional drives in relation to NVH, such as in their weight, mass inertia, torque, and excitation forces and moments. Consequently, it stands to reason that mounting systems optimized for use with combustion engines do not necessarily represent the best solution for electric powertrains.

NVH requirements for electric powertrains In relation to NVH, the most striking difference between electric motors and combustion engines is that electric motors do not have an idle state or engine start and stop. This has a significant effect on the boundary conditions, as the lower-frequency excitations in idle state do not need to be isolated and therefore the eigenmodes of the mounting system do not need to be kept beneath the low idle state excitation frequencies. The load cases resulting from starting and stopping the engine also no longer occur. Based on these modified boundary conditions, the mounts can then be given a stiffer design, leading to lower relative displacements between powertrain and body, which in turn has a positive effect on the space requirements of the powertrain (packaging) and durability of the mounts, as well as on driving dynamics. A further difference of relevance for vibration control technology is that electric motors are typically lighter than combustion engines and tend to have lower moments of mass inertia. This makes the vibrations of the powertrain caused by road excitation less relevant, both in terms of driving comfort and regarding the durability of the mounts. Any overlapping of powertrain bounce and wheel bounce frequencies should still be prevented in the design of the mounting system to avoid any powertrain movements noticeable to the occupants of the vehicle (motor shake). However, the need to reduce these powertrain movements with additional damping by means of hydro mounts is trending down.

2

Powertrain mounting systems for electric vehicles

Furthermore, electric motors produce excitations in a considerably higher frequency range than combustion engines. In the case of a four-cylinder engine, second and fourth-order excitations dominate, corresponding to a frequency of 200-400 Hz at an assumed maximum rotational speed of 6,000 revolutions per minute. With electric motors, the strongest excitations typically occur up to the high double-digit order range. In conjunction with the often higher rotational speeds, this leads to excitations in the clearly audible frequency range of up to 10,000 Hz and beyond. The driver perceives this as a “tonal” noise, similar to that emanating from streetcars, which differs significantly from the typical humming noise of a combustion engine. Part of these excitations is also transferred to the interior of the vehicle through the powertrain mounts as structure-borne noise. This presents the challenge that the powertrain mounts need to exhibit a low dynamic stiffness at a large and very high frequency range in order to achieve optimal isolation of these vibrations. As well as generating high torque, electric motors also generate particularly rapid torque changes, due to the immediately available engine torque during acceleration on the one hand and recuperation on the other hand when the driver removes his foot from the gas pedal. Depending on the design, the recuperation torque can well reach up to 80 percent of the drive torque in the case of electric vehicles. Support of the quasi-static drive torque and load cycle behavior (tip-in/let-off) then become key requirements in relation to vibration control technology. The aim in particular is to quickly limit powertrain movements and enter into the progressive snubbing range of the mounts as little as possible. All of these factors raise the question of whether soft, progressively designed mount concepts, such as those typically used for combustion engines and which also often use hydraulic mounts, are a suitable approach for electric powertrains.

Current mounting systems for electric powertrains Adapted combustion engine systems If we consider the electric vehicles currently on the market, it is evident that there are two different types of mounting systems. There are vehicles which were originally developed for an internal combustion engine (ICE) and subsequently adapted for electric motors, as well as those for which both combustion engines and electric motors are available. The same platform is used for both drive versions in the case of these vehicles. This means that the entire vehicle structure is more or less identical for both versions. Since the electric powertrains are usually smaller than the combustion engines, subframes are used between the side rails in some cases to bridge the distance between the fixing points on the body and the more compact electric powertrain. In

3

Powertrain mounting systems for electric vehicles

other cases, the mounts are positioned closer to the electric powertrain. However, in general, most adapted systems involve a traditional pendulum system with two engine mounts which primarily bear the static load as well as a torque roll restrictor. These types of systems are generally only suitable for the front axle, as there is usually not enough space at the rear axle. In some cases, hydro mounts are used to improve motor shake behavior or a second mounting level (double isolation) is employed to increase noise isolation in higher frequency ranges.

Dedicated electric motor systems The second established way to mount the electric powertrain involves dedicated systems. These mounting systems are used in vehicles and platforms exclusively developed for electric drive concepts from the very beginning. For mounting, a three or four-point system with relatively stiff bushings is typically used, positioned close to the electric motor or in a frame – for example in the subframe if it is installed at the rear axle. The frame itself is often mounted elastically, too, thus enhancing high frequency isolation. Hydro bushings are also occasionally used in these mounting concepts. Since this type of installation can be carried out in a particularly compact fashion, similar to the case of a typical rear axle differential, it can be used both at the front axle and at the rear axle.

Examples of application An example of an adapted combustion engine system is the Volkswagen e-Golf VII (2014), where the electric powertrain is just one of several different drive concepts. The system implemented here is a pendulum system where the mounts that bear the static load (right hand side mount, RHM, and left hand side mount, LHM) are, in contrast to the combustion engine variant, not positioned further out on the side rails, but rather further inward in a cross member between electric motor and power electronics (see Figure 1). The cross member is directly attached to the side rails, so no additional isolation of vibrations can be carried out. In the same way, the elastic mounting is implemented for the torque roll restrictor (lower torque rod, LTR). In this case, too, the cross member is directly connected to the body.

4

Powertrain mounting systems for electric vehicles

Figure 1: Example of an adapted combustion engine mounting system

An example of a dedicated mounting system is the Tesla Model S (2013), which was designed as a pure electric vehicle from the outset. We can see two bushings centrally at the rear axle, one in front of and one behind the motor (front mount and rear mount). Their soft, axial direction is oriented transversely to the vehicle, so they are primarily used to control the motor torque and the remaining loads in fore-aft and vertical direction. A third bushing to the left beside the transmission (left mount), which is installed with its axial direction in driving direction, absorbs loads in lateral and vertical direction in particular. The electric powertrain is installed in the elastically mounted subframe and is thus double-isolated (see Figure 2).

5

Powertrain mounting systems for electric vehicles

Figure 2: Example of a dedicated electric motor mounting system as currently on the market

Market overview If we consider the electric vehicles currently on the market and arrange them based on the different mounting concepts outlined above, it is evident that the large majority of current electric vehicles is still using adapted combustion engine systems for powertrain mounting. Pendulum mounting systems are used in most cases, with a simple three-point mounting also used in certain cases. Hydro mounts are also occasionally used to ensure that the drive unit does not shake excessively in the case of road surface excitations. Only a small number of vehicles such as the Tesla Model S and Model X, the Hyundai Ioniq Electric and the BMW i8 currently feature dedicated electric motor mounting systems based on a three or four-point solution (in the case of the BMW i8). However, it is evident that the future trend is clearly moving in the direction of dedicated systems.

Simulation of powertrain vibrations (rigid body modes) Methodology Below, simulations of rigid body modes of the powertrain will be used to discuss how the previously presented mounting system versions affect the system behavior as a whole. To calculate its rigid body modes, the powertrain mounting system is initially set to the desired static equilibrium state in simulation. The eigenmodes, i.e. the eigen-

6

Powertrain mounting systems for electric vehicles

frequencies and eigenvectors, of the system linearized in this state are then determined. While the focus for combustion engines is often on idle state, i.e. on the state without torque, but with low excitation frequencies, the state with maximum torque is particularly important for electric drives, as the largest excitations are expected here. The state “without torque” is still important for electric motors, however, as it also applies for all drive torques that are small enough for the mounts to still operate in the linear range of their spring characteristics, which corresponds to driving at constant speed and with low loads as often occurs in practice. However, the requirements decrease due to the lower excitations and missing idle state. The rigid body modes for both states are therefore determined and optimized when designing mounting systems for electric powertrains. The results can then be displayed in the form of bar charts (see Figure 3). Each bar represents an eigenmode (rigid body mode) of the mounting system. The bar color displays the type of movement here, i.e. in which of the six directions – three translational and three rotational – the engine moves or rotates. For example, the gray bar represents the rotation around the vehicle transverse axis, or the powertrain “pitch”. It is particularly important for laterally installed powertrains, as their drive torque is applied in this direction. The position of the bar along the horizontal axis corresponds to the respective eigenfrequency of each mode.

Typical challenges faced by current mounting systems In the case of combustion engines, the eigenmodes of the mounting system should be sufficiently beneath the dominant order of excitation in idle state to ensure effective vibration isolation. Consequently, the powertrain rigid body modes in modern systems are typically within a frequency band of 5-15 Hz and up to a maximum of 20 Hz (see Figure 3, top). In contrast to this, the eigenmodes spread out higher across a wider frequency range in the case of an electric powertrain with an adapted combustion engine mounting system due to the lower mass inertia of the powertrain (see Figure 3, center-left). The pitch mode in particular moves to higher frequencies, as the engine mounting needs to be sufficiently stiff in this direction in order to handle the high motor torque. If there is no torque load, or just a low loading, this is less significant from a vibration control perspective than in the case of combustion engines, as the electric powertrain then does not produce any excitations, or just low excitations. The specific modal layout in this operating condition is therefore less important for electric motors than for combustion engines. Looking at high torque ranges is rather recommended, as this is where the largest excitations are expected (see Figure 3, center-right). And this is where a significant drawback of typical, adapted combustion engine mounting systems becomes apparent. Due to the high motor torque and the relatively low basic mount spring rates, the mounts enter into the progressive snubbing range, particularly in the direction controlling the rotation around the vehicle transverse axis

7

Powertrain mounting systems for electric vehicles

(pitch mode), i.e. the spring force now rises quicker than the spring deflection. The mounts therefore become “harder” under high torque conditions, and are not able to isolate noise and vibration as effectively.

Figure 3: Illustration of simulated powertrain rigid body modes for various mounting systems in the form of bar charts

In the case of dedicated electric motor mounting systems, this “harder” system behavior and the subsequent poorer isolation effect in the case of high motor torque is also observed, however it is less severe since the bushings have a higher basic spring rate from the outset and therefore don’t engage the snubbers as much for the same level of load (see Figure 3, bottom). But there is another issue here: the bounce mode, whereby the powertrain vibrates in a vertical direction (dark red bar), is often located at the same frequency range at which the wheel hop and tramp modes are typically found (orange area marked in all bar diagrams). The overlapping of these modes can result in perceivable motor shake despite the lighter weight of the electric powertrain.

8

Powertrain mounting systems for electric vehicles

Based on these simulation results, Vibracoustic recommends two alternative mounting systems for electric powertrains, which both go one step further than the systems currently on the market.

Proposed new mounting concepts for electric powertrains Vibracoustic Soft Proposal The “Vibracoustic Soft Proposal” can be regarded as a further development of the adapted combustion engine mounting systems. It is a pendulum mounting system with a somewhat smaller spacing between the mounts in the vertical and transverse vehicle direction as well as relatively high spring rates in the longitudinal vehicle direction. The mounts are designed large enough in size to handle the maximum motor torque without significant stiffness increase. This mounting concept thus prevents the previously described drawbacks for isolation in the case of high torques (see Figure 4). The progression commences in the directions of force which handle the drive torque at an early stage, but the stiffnesses only rise very slowly subsequently. There is also no significant hardening of the mounts in the maximum torque range – the mounts roughly retain their linear spring rate and can continue to isolate the vibrations effectively. In the vertical direction, however, the mounts are soft enough to keep the powertrain bounce mode beneath the wheel hop and tramp modes. Hydro mounts can be used if necessary to prevent noticeable motor shake. With regard to the durability issues in relation to road excitations, the longer spring deflections as a result of the larger mounts do not pose a significant problem, as the typically lower weight of the electric motor produces lower dynamic loads than in the case of a combustion engine. To exploit the full potential of the mounting system in relation to noise isolation, the dynamic hardening of the mounts should be as low as possible and double isolation should ideally be provided. This can either by implemented by means of a second mounting level on the subframe or via Vibracoustic mounts with integrated double isolation.

9

Powertrain mounting systems for electric vehicles

Figure 4: Sample illustration of operating points and related rigid body modes for the “Vibracoustic Soft Proposal” mounting system compared to an adapted combustion engine mounting system

However, this mounting system does involve some limitations: in order for the three mounts used to be large enough, sufficient installation space must be available, while

10

Powertrain mounting systems for electric vehicles

the components are also heavier. This mounting concept is therefore primarily suitable for the front axle and should be used in particular if the overall vehicle concept emphasizes extremely high isolation and related comfort levels. Regarding the eigenmodes, the Vibracoustic Soft Proposal shows clear benefits compared to the adapted combustion engine mounting system in the case of high torque. The eigenfrequencies of the highest modes are significantly lower in the simulations, with the system exhibiting relatively constant, torque-independent isolation behavior throughout the entire torque range.

Vibracoustic Stiff Proposal The “Vibracoustic Stiff Proposal” can be regarded as a further development of the dedicated e-motor mounting systems currently on the market. In contrast to the first proposal, it is based on stiff, conventional bush-type mounts, yet still with minimal dynamic hardening, which handle drive torque in their vertical direction within a three or four-point mounting system. The bushings should be sufficiently stiff to fulfill two demands. First, the powertrain bounce mode should be above 20 Hz and thus well beyond the typical wheel hop and tramp frequency range. This will prevent motor shake issues without the need to use hydro mounts. Second, the spring characteristics of the mounts are practically linear throughout the entire torque range, so that even maximum torque can be handled without any progression. They consequently do not harden statically and, in the case of high torque, where the excitation is at its strongest, act more “softly” than the mounts of the dedicated electric motor mounting concepts described above (see Figure 5). The higher basic spring rate gives them a lower isolation effect in the case of lower torque. As carried out previously, however, the low torque load case is of less importance for electric vehicles due to the smaller excitations. In relation to the eigenmodes, it is apparent here that the highest eigenfrequencies in the state without torque are somewhat higher than in the case of previous dedicated electric motor mounting systems. But the system behavior hardly changes when torque is applied. In maximum torque condition the highest eigenfrequencies are below those of a typical dedicated system which can currently be found on the market, whereby the isolation is improved in comparison to those. As described, the powertrain bounce mode is above the wheel hop and tramp modes in the simulation to avoid motor shake issues. As small, conventional bushings are used, there are also benefits with regard to weight and packaging. The very small relative displacements within the stiff design improve vibration behavior in the case of the rapid torque changes (tipin/let-off) that typically occur with electric motors. Furthermore, they increase the durability of the already very robust bushings, which is primarily dependent on the number of torque changes in this case.

11

Powertrain mounting systems for electric vehicles

Figure 5: Sample illustration of operating points and related rigid body modes for the “Vibracoustic Stiff Proposal” mounting system compared to a dedicated electric motor mounting system as currently on the market

Finally, the reduced movements of the powertrain relative to the body also lead to lower mechanical stress in the power supply cables.

12

Powertrain mounting systems for electric vehicles

Based on its low installation space requirements, the Vibracoustic Stiff Proposal is suitable both for use on the front axle of electric vehicles and for the rear axle, where there is typically far less space available for mounting the powertrain. This concept can also be improved by a second isolation level which offers a significant benefit with regard to noise isolation and which should therefore be planned if possible.

Summary Despite the level of sophistication of mounting concepts for combustion engines, they are only suitable for electric powertrains to a limited extent, as those have their own specific requirements. The biggest differences include the lack of an idle state and engine start/stop load cases, the lighter weight, and the smaller mass inertia of the electric powertrain, as well as the considerably higher excitation frequencies of up to more than 10,000 Hz. Other factors include the high torque and rapid changes in torque when the driver accelerates or takes his foot off the gas pedal, with a considerably higher deceleration than is the case with combustion engines due to the recuperation. Consequently, for electric vehicles idle and engine start/stop load cases are of no importance, while driving with high torque and rapid load cycles take on greater importance. This should be taken into account accordingly when designing the mounting system. The mounting concepts established on the market to this point involve either adapted combustion engine systems or dedicated electric motor systems. The adapted systems can typically only be used on the front axle. The relatively “soft” engine mounts operate far within the progressive snubber range in the case of high torque, thus become much “harder”, and can then no longer isolate noise and vibration as effectively. This phenomenon is also noted in the case of dedicated systems, although it is less severe due to the considerably higher basic spring rate of the mounts. They can also be used both on the front axle and the rear axle. However, the powertrain bounce mode is often within the wheel hop frequency range, which can cause motor shake issues. Based on these evaluations carried out through simulations, Vibracoustic recommends two approaches for mounting electric powertrains. As an expansion of the adapted combustion engine systems, the “Vibracoustic Soft Proposal” is based on a traditional pendulum mounting system, where the mounts are large enough and have adequate stiffness in the longitudinal direction to manage without any significant progression even in the case of high motor torque. In the vertical direction in particular, however, the mounts are soft enough to keep the powertrain bounce mode beneath the wheel hop and tramp frequencies. Hydro mounts can be used if necessary to prevent noticeable motor shake. This concept is primarily suitable for the front ax-

13

Powertrain mounting systems for electric vehicles

le and only if there is sufficient installation space available. It should ideally be used if particular emphasis is placed on ensuring optimal isolation during the overall vehicle design.

Figure 6: Evaluation of described mounting concepts with regard to key NVH criteria

In all other cases, Vibracoustic recommends the Vibracoustic Stiff Proposal for mounting electric powertrains as a continuation of the current dedicated electric motor mounting systems. It recommends the use of conventional bushings in a three or fourpoint mounting system, with spring rates high enough to fulfill two conditions. First, the engine mounts have practically linear spring characteristics throughout the entire torque range. The strengths of this concept become apparent in the case of high torque, when the engine produces the largest excitations. The mounts handle even maximum torque without any significant progression and exhibit consistently effective isolation behavior. Second, the powertrain bounce frequency is moved above the range of the wheel hop and tramp modes. Due to the relatively low weight of the electric drive, motor shake issues can therefore also be prevented without the use of hydro mounts. Furthermore, the very low relative movements in the system improve vibration behavior in the case of rapid load cycles (tip-in/let-off), which are also a determining factor for the durability of the mounts in the case of electric vehicles. Finally, there are also clear benefits in terms of weight and packaging. The concept can also be used on the rear axle.

14

Powertrain mounting systems for electric vehicles

Outlook The next step should be to verify the results of the simulation using prototypes on an actual electric vehicle. In parallel, the calculation and simulation tools, measurement technology, and test benches need to be further developed for investigations in the kilohertz range, as these areas are particularly important for electric drives.

Figure 7: Sample illustration of a Vibracoustic QUIETTYPE® mount characteristic (left, green) compared to conventional mounts (left, red) and specific implementation examples (right)

Vibracoustic can call on its expertise from other areas of application here, where the problematic frequency ranges are similar in scale to electric motors, such as in the isolation of transmission whine. Based on this, the company has already introduced some initial specific product innovations in this field, such as the new QUIETTYPE® mounts. They use the specific design of a continuum-mechanical effect carried out via simulation to achieve a significantly reduced dynamic stiffness for the actual component across a broad and configurable frequency range, and thus better isolate problematic frequencies. More or less independently from the static stiffness, this enables the tuning of mounts and bushings in a high frequency domain, where electric motor specific issues can occur.

15

Powertrain mounting systems for electric vehicles

References Pflüger, Martin; Brandl, Franz; Bernhard, Ulrich; Feitzelmayer, Karl: Fahrzeugakustik. Der Fahrzeugantrieb. Edited by Helmut List. Vienna: Springer-Verlag, 2010. Trelleborg Vibracoustic (Hg.): Schwingungstechnik im Automobil. Grundlagen, Werkstoffe, Konstruktion, Berechnung und Anwendungen. Würzburg: Vogel Business Media, 2015.

16

Noise radiated by electric motors – simulation process and overview of the optimization approaches Dr. Jean-Baptiste DUPONT, Henri SAUCY

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 W. Siebenpfeiffer (Hrsg.), Automotive Acoustics Conference 2017, Proceedings, https://doi.org/10.1007/978-3-658-20251-4_7

1

Noise radiated by electric motors – simulation process and overview …

1 Introduction The automotive industry has entered a phase of change due to environmental considerations. Hybrid and electric vehicles have emerged. The vibroacoustic behaviour (knowledge of the source mechanisms and expertise regarding the integration of this source of noise and vibration inside the vehicle structure) of combustion engines is now well controlled. This knowledge cannot be directly transferred to these new vehicles, either hybrid or electric. Therefore, it is necessary to build a new expertise and new tools in order to understand and optimize electric motors and their integration. This paper takes part in this process: the goal is to predict the vibroacoustic characteristics of a motor during the design phase. Noise and vibrations generated by an electric motor can be divided into three main contributions related to three distinct sources [1]: mechanical, aerodynamic and electromagnetic. Mechanical noise and vibrations are related to the assembly. The friction and banging noise may be due to bearings, gears, possibly to the friction of the brushes on the rotor. These sources are highly dependent to the rotational speed. In order to cool the motor, a fan is often mounted at the shaft end. Its goal is to gather the air up and force it to flow through the air gap. Just as any fan, it radiates noise. It is commonly accepted [2] that the acoustic power radiated by an axial fan is proportional to the power 5 of the rotational speed. Finally, note that the salient rotor poles may be the cause of an aerodynamic noise contribution. The levels of aerodynamic and mechanical noise are the highest at high speed. Electromagnetic noise and vibrations are due to electromagnetic excitations that occur at the air gap of the motor. Many parameters influence the content of the excitation [3] such as the motor topology (number of slots and poles), the pole shape, the slot opening or the motor drive (currents shape). The amplitude of these excitations is not necessarily proportional to the rotational speed. Thus, the electromagnetic noise can be predominant at low speed. Figure 1 illustrates the noise generation process in electric rotating machines.

Figure 1 – Noise generation process

22

Noise radiated by electric motors – simulation process and overview …

A methodology has been developed in order to estimate the noise radiated by the machine due to electromagnetic phenomena. This requires the estimation of the forces applied to the stator, to calculate the dynamic response of the stator, and then deduce the radiated noise. This is a multiphysical problem for which it is necessary to combine elements of electromagnetic finite element calculations, dynamic finite element calculations and acoustic finite element calculations. This type of calculation has been presented and implemented by several authors. Neves et al. [4,5] applied it to a switched reluctance motor (SRM) by using simplifying assumptions. They showed the relevance of the approach. Furlan et al. [6] used this methodology to calculate the noise radiated by a DC electric motor. Schlensok et al. [7] applied it to an induction machine with a squirrel-cage rotor. Rainer et al. [8] computed the dynamic response of a skewed induction machine and studied the accuracy in the frequency domain. Pellerey et al. [3,9] also applied this methodology to a wound-rotor synchronous motor. The implementation of the simulation method has already been detailed [10]. This article focuses on the optimization opportunities given by this simulation method to reduce the noise radiated by electrical machines. Optimization actions can be implemented at all steps of the noise generation process: Ɣ

supply strategy,

Ɣ

electromagnetic design,

Ɣ

mechanical design,

Ɣ

integration.

Optimization cases are detailed in order to illustrate the possibilities offered by this method.

2 Simulation methodology The basic principle of the calculation is to perform a weakly-coupled electromagneticdynamic calculation. Figure 2 gives a scheme of the 3-step multiphysical calculation procedure. The electric motor is modelled using a finite element electromagnetic software program in order to calculate the electromagnetic excitations applied to the stator. As stated in [10], Maxwell pressure is taken as the main phenomenon responsible for noise and vibrations of the rotating electrical machines. Both radial and tangential contributions are estimated and converted into space-frequency excitation matrices. The influence of every relevant parameter is contained in this simulation: Ɣ

number of poles

Ɣ

number of stator slots (if any),

3 3

Noise radiated by electric motors – simulation process and overview …

Figure 2 - Basic principle of the calculation procedure. Ɣ

number of rotor slots (if any),

Ɣ

magnetic core geometry,

Ɣ

supply strategy (current shape),

Ɣ

saturation,

Ɣ

eccentricities (if any).

These parameters affect the excitation content in the time (and frequency) domain as well as its space distribution. These excitation matrices are projected onto the structural mesh of the e-machine with the aid of a dedicated mapping tool and a dynamic calculation can be performed using a finite element method. The calculation of dynamic responses is often based on the modal frequency response. In a first step, the modal basis of the structure is extracted. Then, the response of each mode under realistic electromagnetic excitation is calculated and the operating deflection shape is obtained by sum these contributions. In some rare cases, the direct approach may be preferred, but it often leads to longer computational times. As this kind of procedure is often included in an acoustic scope, the last step of the procedure is about the estimation of the noise radiated by the machine. The radiated sound power is obtained from the vibration velocity on the outer skin of the machine structure. In the field of the numerical methods, acoustic finite element method (FEM) or boundary element method (BEM) can be used. But these methods are time consuming and much faster analytical method are used if possible. Since the estimation of the noise ra-

4 4

Noise radiated by electric motors – simulation process and overview …

diated diatedby bya avibrating vibratingstructure structureisisa avery verytypical typicalproblem, problem,this thisstep stepisisnot notdetailed detailedininthis this paper. paper.

3.3. Analysis Analysisof ofthe thedynamic dynamicresponse response Highest Highestnoise noiseand andvibration vibrationlevels levelsare areobserved observedwhen whenresonances resonancesoccur. occur.AAresonance resonance happens happenswhen whenthere thereisisboth botha aspace spaceand andfrequency frequencycoincidence coincidencebetween betweenananexcitation excitation contribution contributionand anda astructure. structure.InInorder ordertotoanalyse analyseand andunderstand understandthe thedynamic dynamicbehaviour behaviour ofofthe themachine machinestructure structureunder underrealistic realisticelectromagnetic electromagneticexcitation, excitation,both boththe themodal modalbasis basis and andthe theelectromagnetic electromagneticexcitation excitationmust mustbebeconsidered. considered. InInterms termsofofNVH NVHbehaviour, behaviour,the themost mostimport importstructure structuremodes modestotoconsider considerare arethe thesosocalled calledcylinder cylindermodes modescharacterized characterizedby bytwo twointegers integers(m,n) (m,n)giving: giving: ƔƔ

m: m:the thenumber numberofofmaxima maximaover overthe thecircumference. circumference.ItItisiscalled calledcircumferential circumferentialspatial spatial order. order.

ƔƔ

n:n:the thenumber numberofofnodal nodalcircles circles(sections (sectionswith witha anull nulldisplacement) displacement)ininthe thelength. length.

Figure Figure33illustrates illustratesthe therelated relatedmode modeshapes. shapes. m=0 m=0

m=2 m=2

m=3 m=3

… …

n=0 n=0

… …

n=1 n=1

… …

… …

… …

… … Figure33––First Firstcylinder cylindermodes modes Figure

… …

… …

Thestructure structureisismore moreeasily easilyexcited excitedon onitsitslow lowspatial spatialorder ordermodes. modes.IfIfthe theexcitation excitationisis The considereduniform uniformthrough throughthe thestator statorlength, length,the themodes modeswith witha alongitudinal longitudinalbending bending considered ((2,1),(3,1), (3,1),etc.) etc.)are arenot notexcited excitedand andthe themost mostproblematic problematicmodes modesare arethus thus(m,0) (m,0)modes. modes. ((2,1), Thesemodes modesare arecharacterized characterizedby bytheir theirspatial spatialorder orderand andtheir theirfrequency. frequency. These

55 5

Noise radiated by electric motors – simulation process and overview …

In order to predict the resonance, the goal is to analyse the excitations in the same way. To compare the excitation and the cylinder modes, the electromagnetic excitation is decomposed, for each engine speed, into elementary rotating forces characterized by their frequency f and their spatial order m, leading to a spatial order-frequency excitation matrix that conveys the content of the electromagnetic excitation and providing, even before the dynamic calculations, a powerful tool to estimate the critical speeds and frequencies. Figure 4 shows an example of spatial order -frequency analysis for the electromagnetic excitation at a given speed: the x-axis gives the spatial order, the y-axis is the frequency (equivalent to the motor harmonic) and the colour gives the magnitude of the excitation.

Figure 4 – Example of spatial order-frequency analysis of the electromagnetic excitation It can be seen observed that the electromagnetic excitation is actually made by a small number of contributions: the stator is excited for a few spatial orders and a few frequencies that depend on the main characteristics of the machine (number of poles, number of slots, eccentricities…). These elementary rotating forces are the cause of the dynamic response of the stator (resonance or forced response) and its acoustic radiation. This analysis allows the prediction and the understanding of the operating deflection shapes.

66

Noise radiated by electric motors – simulation process and overview …

4. Optimization cases 4.1 Presentation The noise generation process in electric rotating machines is illustrated in figure 1. An optimization approach can be (must be) implemented at each of the main steps: Ɣ

optimization of the supply strategy,

Ɣ

optimization of the electromagnetic design (number of poles, number of poles, detail design…),

Ɣ

optimization of the structure design,

Ɣ

optimization of the machine integration to reduce the propagation of the noise and vibrations towards the outside.

The simulation method presented in section 2 follows the process of noise generation in electric rotating machines. It can be used to implement an optimization approach which aims at choosing at each step the best design parameters for the reduction of the noise radiated by the machine. Three optimization examples are presented in the following sections.

4.2 Optimization of the supply strategy 4.2.1 Presentation This case is about a railway traction motor. It is a 4 poles, 48 slots, PWM-fed induction machine. The optimization is only about the PWM frequency, it is not possible to modify the machine itself. The initial PWM frequency is 1000Hz, but the noise level is considered too high and an optimal value is to be determined.

4.2.2 PWM excitation contributions Pulse-width modulation (PWM) is a modulation technique used to control the power supplied to electrical devices, especially to motors. The average value of voltage fed to the load is controlled by turning the switch between supply and load on and off at a fast rate, the so called PWM frequency. The longer the switch is on compared to the off periods, the higher the total power supplied to the machine. Compared to a perfect sinus voltage, PWM creates high frequency voltage contributions that can be observed in the currents, in the flux density, in the electromagnetic excita-

7 7

Noise radiated by electric motors – simulation process and overview …

tion (Maxwell pressure) and in the noise. The frequencies of the high frequency contributions are rather well known [11]: ݂ ൌ ܽ ή ݂௉ௐெ േ ܾ ή ݂௦

(1)

Where ݂௉ௐெ is the PWM switching frequency, ݂௦ is the frequency of the fundamental sinus voltage to obtain and (ܽǡ ܾሻ two integers. The space distribution of the excitation contribution related to high frequency PWM voltage harmonics is to be determined with the aid of the method proposed in section 3. The magnitude of these excitation contributions depends on many parameters such as the machine topology and the PWM algorithm.

4.2.3 Resonances and optimization The main principle of this type of optimization is to avoid resonances is the speed range of interest of the machine. On the one hand, an accurate knowledge of the PWM excitation contributions is necessary (see 4.2.2). On the other hand, the modal description on the structure is required. The crossing of this information makes it possible to understand the response of the structure to the PWM excitation contributions. Figure 5 shows a Campbell diagram (frequency-speed diagram) that includes both the PWM excitation and the modal characteristics. Coloured boxes correspond to the main cylinder modes of the structure. For each mode, the frequency width of the box depends on the damping and the uncertainty in the natural frequency. Coloured lines correspond to the PWM excitation contributions. They are characterized by their spatial order and the magnitude of the excitation.

Figure 5 – Campbell diagram for ݂௉ௐெ =1000Hz

88

Noise radiated by electric motors – simulation process and overview …

A resonance happens when both a space and a frequency coincidence occur between an excitation contribution and a structure mode, in other words, when a coloured line crosses a box with the same spatial order. In the initial case (figure 5), both (4,0) mode and (0,0) mode are excited during the run-up. These resonances lead to high noise levels and should be avoided by modifying the PWM frequency. Figure 6 shows the Campbell diagram for the optimal PWM frequency (1230Hz). It can be observed that no resonance occurs over the run-up. Computations and tests show that the noise level at 1000rpm is reduced by 9dB without modifying the machine.

Figure 6 – Campbell diagram for ݂௉ௐெ =1230Hz

4.3 Optimization of the electromagnetic design

This example is about an alternator from the automotive industry. It is a 12-pole and 36slot claw pole machine. Figure 7 shows the considered alternator. The target concerned a specification about the radiated sound power. The initial design, based on the experience of the manufacturer’s engineers, exceeded the maximum level allowed. The objective was to understand the radiation mechanisms and to propose reduction solutions. The complete simulation procedure described in section 2 has been applied. In this case, due to the specific shape of the rotor, it was necessary to use a 3D electromagnetic simulation to correctly determine the excitations applied to the structure of the machine. In a second step, the finite element model of the machine structure was built. A model updating procedure has been performed in order to ensure the model reliability.

9 9

Noise radiated by electric motors – simulation process and overview …

Figure 7 – Considered alternator

Figure 8 – Campbell diagram of the sound power radiated by the alternator Finally, after the projection of the electromagnetic excitations, the sound power radiated under realistic excitations has been computed. As illustrated by the Campbell diagram given in figure 8, two stator modes and two harmonics have been identified in the noise generation process: H36 and H72. Unfortunately, only the rotor can be changed. It is not possible to modify the design of the stator. Thus, the modal behaviour of the radiating structure remains identical and the modifications concern only the electromagnetic excitations applied to the stator. Several geometrical parameters of the rotor claws have been considered (width, length, thickness, chamfer…). The most interesting point was to be able to evaluate both the changes in performances of the e-machine and the associated noise reduction without building any prototype.

10

Noise radiated by electric motors – simulation process and overview …

Figures 9 and 10 show the reduction of sound that could be obtained by modifying the rotor design (and thus the electromagnetic excitation) without changing the overall mechanical design. It can be seen that the maximum peaks are at the same frequencies, only the amplitudes have changed (the natural frequencies of the structure are not modified). As usual, these design improvements have drawbacks. In the present case, we see that optimization 1 leads to a very significant reduction on H36 (around 10 dB on the main peak), However, there is a slight worsening for H72. Whereas optimization 2 leads to a lower reduction but it has an effect on both Harmonics. This optimization approach resulted in the production of prototypes. The noise measurements confirmed the results of the simulations: the optimized designs led to 10dB reduction in the noise radiated by the alternator in the speed range of interest.

Figure 9 – Sound power radiated by the alternator (order tracking H36)

Figure 10 – Sound power radiated by the alternator (order tracking H36)

4.4 Optimization of the mechanical design The studied machine is a starter that generated a whistling noise. This noise is usually hardly heard by the driver. However, with stop-start systems this noise appears more frequently, thus it can be identified with numerous engine starts. The simulation-based approach consisted in determining the origin of the noise and proposing modifications to reduce it. The speed range of interest has been identified: 4000rpm-8000rpm. Figure 11 is a Campbell diagram showing the sound pressure level against time and frequency during an ICE start. A first noise frequency band is observed around 1200 Hz (red circle), this noise is due to the mechanical excitation of the first vibration mode by the cranking mechanism. The second noise emission zone is located around 2200 Hz (yellow circle). This is typically an electric contribution at 19 times the rotation speed of the electric motor, 19 being the number of slots.

11 11

Noise radiated by electric motors – simulation process and overview …

Pa

dB

100.00

s

Time

1.00

AutoPow er MIC1 WF 349 [0-1.74 s]

2896.79

0.00 0.00

Hz MIC1 (CH1)

85.00 5000.00

Figure 11 – Starter noise during an engine start The simulation method previously presented has been applied. Then, the first step was to calculate the electromagnetic excitations that apply to the structure of the starter. This has been done with the aid of a finite element electromagnetic solver. The model is not presented in this paper. As detail in section 3, the content of the electromagnetic excitations has been analysed, leading to the graph given in figure 12. As expected, the excitation is mainly borne by the 19th and 38th engine order, for each engine order the force can be decomposed into spatial orders. Engine order 19 has its highest components on the 15th and 19th spatial orders (red circle). However, the structure will be sensitive to low spatial orders, this is why for the same engine order (the 19th), the structure will mainly be excited by the 1st and 3rd spatial order (yellow circle). A finite element model of the starter has built. A model updating procedure has been performed in order to validate the model reliability. As it is rather well known, this step is not detailed in this paper. The modal basis of the structure has been extracted for representative boundary conditions. Among the structure modes, the most important ones are the so-called “cylinder modes” presented in paragraph 3. In particular: Ɣ

(2,0) modes at 850Hz and 950Hz

Ɣ

(3,0) modes at 2200Hz and 2400Hz

Ɣ

(4,0) modes at 3300Hz

12

Noise radiated by electric motors – simulation process and overview …

Figure 12 – Decomposition of the electromagnetic excitation into spatial order and engine order It is clear that the high level of radiated noise related to the engine order H19 around 2200Hz is due to the resonance of a structure mode. Indeed, a 3-lobe mode is located at 2200Hz. Since the electromagnetic excitation contains a contribution of spatial order 3 for the engine order H19, when the rotation speed is 6950rpm, a space and a frequency coincidence occur between an excitation contribution and a structure mode. That explains the high level of noise and vibration. At this stage of the design process, it is was no longer possible to modify the parameters of the electrical machine by itself (number of poles, number of rotor slots, slot shape, etc.). The main reduction perspectives thus concerned the structure design. The 3-lobe mode has been identified as the cause of the high noise level. Structural modifications have therefore been proposed to shift its frequency towards the high frequencies so that the resonance occurs outside the speed range of interest (4000rpm8000rpm). Figure 13 shows the radiated sound power (order tracking H19) for the initial structure and the modified structure. The resonance of the 3-lobe mode with the engine order H19 still can be observed, but thanks to the stiffening of the structure, it occurs at 8800rpm,

13 13

Noise radiated by electric motors – simulation process and overview …

that is, outside the speed range of interest. In the targeted speed range, the noise related to the engine order level H19 decreased by 12dB and the starter meets the specifications in terms of noise.

Figure 12 – Radiated sound power (order tracking H19). Initial and optimal designs

5. Conclusions In order to meet the new needs of the automotive industry, a methodology for the calculation of the dynamic response of an electric motor stator has been presented in this paper. This method can also lead to the calculation of the acoustic power radiated by the motor. It is a multiphysical method that accounts for the supply strategy (input voltages or currents), for the electromagnetic design of the active electromagnetic parts, for the structure design of the machine and for the integration of the electrical machine. This simulation approach is a powerful tool for the optimization in order to reduce the noise and the vibrations generated by the machine. At each step of noise generation process, the optimal parameters can be determined, as illustrated by 3 examples in this paper. It must be noted that the earlier this process is undertaken in the design phase, the more optimization parameters can be found, the more the optimization approach can lead to a significant reduction in noise and vibration levels.

14 14

Noise radiated by electric motors – simulation process and overview …

6. References 1. Gieras J. F., Wang C. and Lai J. C., “Noise of polyphase electric motors”, CRC Press, ISBN-13: 978-0824723811, 2005. 2. Sharland I. J., “Sources of noise in axial flow fans”, J. Sound & Vib., 1(3):302-322, 1964. 3. Pellerey P., Lanfranchi V. and Friedrich G. “Vibratory simulation tool for an electromagnetically excited non skewed electrical motor, case of the Wound Rotor Synchronous Machine”, ELECTRIMACS2011, Cergy-Pontoise, France, June 2011. 4. Neves C. G. C., Carlson R., Sadowski N., Bastos J. P. A., Soeiro N. S. and Gerges S. N. Y, “Calculation of electromagnetic-mechanic-acoustic behavior of a switched reluctance motor”, IEEE transactions on magnetics, 36(4):1364:1367, 2000. 5. Neves C. G. C., Carlson R., Sadowski N., Bastos J. P. A. and Soeiro N. S., “Forced vibrations calculation in a switched reluctance motor taking into account the viscous damping”, in Conf. Rec. IEEE-IEMDC-International Electric Machines and Drives Conference, Seattle, USA, 1999, pp. 110:112. 6. Furlan M., Cernigoj A., Boltezar M., “A coupled electromagnetic-mechanicalacoustic model of a DC electric motor”, COMPEL: The International Journal for Computation and Mathematics in Electrical and Electronic Engineering, Vol. 22 Iss: 4, pp.1155 – 1165, 2003. 7. Schlensok C., van Riesen D., Küest T. and Henneberger G., “Acoustic simulation of an induction machine with squirrel-cage rotor”, COMPEL: The International Journal for Computation and Mathematics in Electrical and Electronic Engineering, Vol. 25 Iss: 2, pp.475 – 486, 2006. 8. Rainer S., Bíró O., Weilharter B. and Stermcki A., “Weak Coupling Between Electromagnetic and Structural Models for Electrical Machines”, IEEE transactions on magnetic, 46(8):2807:2810, 2010. 9. Dupont J. and Bouvet P., “Multiphysics Modelling to Simulate the Noise of an Automotive Electric Motor”, SAE Technical Paper 2012-01-1520, 2012, doi:10.4271/2012-01-1520. 10. Dupont J., Bouvet P. and Humbert L., “Vibroacoustic simulation of an electric motor: methodology and focus on the structural FEM representativity”, XXth International Conference on Electrical Machines (ICEM'2012), Marseille (France), September 2-5 2012 11. Lo W. C. et al., “Acoustic noise radiated by PWM-controlled induction machine drives”, IEEE Transactions on Industrial Electronics, 47(4):880:889, 2000.

15 15

New fuel-saving technologies and NVH refinement of powertrains Leon GAVRIC, Vice President, Senior Expert NVH, PSA Groupe

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 W. Siebenpfeiffer (Hrsg.), Automotive Acoustics Conference 2017, Proceedings, https://doi.org/10.1007/978-3-658-20251-4_8

1

New fuel-saving technologies and NVH refinement of powertrains

1 Context of actual car-market The ‘ecological conscience’ of consumer, the CO2 reduction imposed by legislation and the low consumption tax policy adopted by majority of countries, motivate car manufacturers to develop new generation of low CO2 emission cars. The reduction of rolling friction (low resistance pneumatics), of air resistance and of overall mass is the design imperative for low consumption vehicles. Still, the most important reduction of fuel consumption and CO2 emission is to be achieved by new generation of powertrain. This implies different new technologies which can represent a marked impact on NVHbehaviour of engines. In the past, NVH was considered as a secondary performance. However, the rise of high quality products made car manufacturers change their point of view. NVH has become a selective performance, allowing the positioning a car in the upper commercial product range, which generates the best profits. Therefore, in actual highly competitive automotive market, the acoustic comfort of vehicle is a major issue, particularly when power train noise and vibration are concerned.

2 Powertrain of tomorrow 2.1 Internal combustion engine (ICE) and CO2 requirements The most of car market analysts are firmly convinced that internal combustion engine will power 90 percent of 2020-2030 vehicles. New and more energy efficient powertrain concepts driven by increased focus and demands on the reduction of fuel consumption and CO2 are to be developed by the automotive industry. New generation powertrains, should be efficient, have to respect the CO2 strategy of carmaker by minimizing fuel consumption, have to meet high performances concerning the pollutant emission, while still reaching the drivability and the NVH targets.

2.2 Powertrain, engine and transmission trends Since, the internal combustion engine should be dominating propulsion system for at least next 10 to 20 years, the reduction of C02 emission is one of the most important issues and a long-term target for car manufacturers – [2]. The new technologies, widely applied to diesel engines (ex. high pressure fuel injection systems, high performance turbochargers, EGR systems and high performance air intake systems with heat exchangers etc – proposed in 90-ties) were already led to small engines generating high power while having relatively low fuel consumption. Nowadays the similar technologies are also applied to gasoline engines, resulting in smaller engines, lower fuel consumption and CO2 emission and sometimes in smaller number of cylinders.The fuel reduction target for new generation of internal combustion engines is

2

New fuel-saving technologies and NVH refinement of powertrains

extremely high. The engine development strategy concerning NVH can be resumed as: Fuel-economy needs shall drive engine noise. Nevertheless, at the end of development the high NVH refinement of vehicles, powered new generation of engines, is to be reached. In order to fulfil all these requirements, some technology breakthroughs will be introduced: cylinder deactivation, variable compression, efficient thermodynamic cycles (Miller or Atkinson), new combustion technologies (ex. stratified), variable valve lift and timing, advanced turbocharging (ex. twin turbocharging, electric driven charger, etc.). The transmissions have to adapt to such technology allowing extreme engine downsizing by: high ratio spreads, short 1st gears, small gear steps, power-shift capability. Some complementary technological solutions will be added to the IC based powertrain: hybridization (electric or other), waste energy recover, stop & start devices, …

3 New powertrain technologies and NVH impacts 3.1 Downsizing and down-speeding of IC engines The basic idea concerning the downsizing consists of putting more air and fuel in a small engine, in order to generate more power. Since the losses are generally proportional to the engine size of engine, they are kept constant. As consequence, the downsized engine, is more efficient and consumes less fuel. Comparing to the standard engines, the downsized one is also lighter and generates higher torque. The downspeeding strategy, searches to decrease the fuel consumption by pushing the engine to run at low speeds. The trend toward downsized and dawn-speeded, turbocharged powertrains has led to a number of new issues that noise, vibration and harshness engineers must overcome. It seems that it is done successfully, as today, downsizing and down-speeding are reputed to further extend the vehicles’ fuel efficiency whilst avoiding to compromise safety, comfort, and fun to drive. However, these technologies may represent a marked impact on NVH-behaviour powertrains, due to: ● Lighter powertrain (less mass) ● more flexible engine mounts needed ● ‘powertrain on mounts’ eigen-ferquencies narrow the engine speed range ● lower vibration attenuation by engine mounts ● High torque ● higher pre-loads on the mounts (more rigid, less decoupled) ● higher torque irregularity (vibration excitation)

3

New fuel-saving technologies and NVH refinement of powertrains

● Lower engine speed ● low frequency excitation (may narrow low frequency resonances) ● larger intake manifold volumes needed (intake noise) ● larger exhaust muffler volumes needed (exhaust noise) Furthermore, stringent emission legislation enforces engine calibration schemes such as catalyst pre-heating and particulate filter regeneration which evoke negative acoustic side effects. Hence, the design of the exhaust system requires more effort to achieve customer acceptance, e.g. larger muffler volumes or increased backpressure.

3.2 Reducing number of cylinders The down-sizing can be applied by decreasing the volume of each cylinder while keeping the global engine design preserved. The most emblematic case is switching the 4 cylinder 2.0 litre diesel engines by equivalent 1.6 or 1.5 litre engines of equivalent power and torque, which is accomplished by quasi-totality of European carmakers during the last decade. Still, the technology limitations impose lower limit of cylinder volume, and ultimate ‘down-sizing’ of both diesel and gasoline engines, is to be reached, by reducing number of cylinders. The trend towards powertrain downsizing will lead to considerable market share gains for three-cylinder engines in the smaller vehicle segment up to medium-sized cars.

Figure 1 – Typical crank train of 3-cylinder engine with balancing shaft and torque irregularity as a function of engine speed: 3-cylinder vs 4-cylinder engines [3]

The NVH behaviour of 3-cylinder engine is different with respect to: inertial forces and balancing, torque irregularity, noise perception, etc. The excitation of 3-cylinder engine is ¾ times lower in frequency than of equivalent 4-cylinder engine. This may amplify the response of some low modes of vehicle structure/chassis, which are prac-

4

New fuel-saving technologies and NVH refinement of powertrains

tically not excited by 4-cylinder engine. The specific NVH behaviour of 3-cylinder engines can be summarized as follows: ● Inertial forces and balancing (if no balancing shaft) ● ● ● ●

low frequency vibration of powertrain at EO1 booming noise in medium and high engine speeds stop & start vibrations due to unbalanced crankshaft at EO1 specific “yaw-pitch” vibration pattern

● Speed and torque irregularities ● ● ● ●

low frequency vibrations of powertrain EO1.5 gearbox rattle (only load depended gas forces) booming noise at low engine speeds stop & start vibrations due to roll vibration at EO1.5

● Airborne noise level ● low frequency content of engine base noise (combustion/cylinder number) ● accessory and accessory drive noise (due to higher torque irregularity) ● Engine roughness and sound quality ● specific ratio of the gas and mass forces (different from 4-cylinder engine one) ● dynamic excitation of crankshaft (120°-cylinder lag) and balancing shaft ● Air intake and exhaust noise ● low frequency excitation content of (combustions/cylinder number) ● larger intake manifold volumes needed (intake noise) ● larger exhaust muffler needed (exhaust noise) ● others … Three-cylinder engines do not generally lead to amplification of interior noise level in the vehicle. In fact, from mid-rpm levels and up, the dB(A) level is lower than with comparable four-cylinder engines. Nevertheless, for heavier vehicles, smalldisplacement 3- cylinder engines may not be well suited and would have to run more frequently at higher rpm-s and consequently make more noise. The decision on whether or not to use a balance shaft with 3-cylinder engine should be evaluated as a function of NVH refinement targets.

5

New fuel-saving technologies and NVH refinement of powertrains

3.3 Cylinder deactivation and variable compression Cylinder deactivation, when engine working in partial loads, is that the fuel for several cylinders is cut off by some mechanism and the corresponding valves are closed whiles the rest of the cylinders can run efficiently. When vehicle accelerates fast or climbs, all cylinders will start work to enhance engine power output. Such engine behaves as “variable number of cylinder” engine, adapting the global working cylinder volume “on demand”. Cylinder deactivation reduces pumping and mechanical losses and improves fuel economy of engine. The fuel will be saved as much as 20% by employing the engine with cylinder deactivation. Under cylinder deactivation, the firing interval angle of remainder cylinder and the unevenness of running will increase. This causes modifications in torsional vibrations of the engine crank train. The resulting vibration depend on the global concept of the engine (flat, in-line, V-engines) on the number of cylinders (6, 4, 3 cylinders) and on the type of drive line (4-wheel drive, front-wheels drive, rear-wheel drive). The number of cylinders and engine design define the global ratio of gas and mass forces and their frequency (engine order) distribution. For 4-cylinder engines, the principal mass and gas forces are confined at the 2nd engine order (EO2). The deactivation of 2 cylinders (often 2 middle ones) does not change the EO2 contribution of mass forces. Since the combustion occurs less often, the corresponding gas forces create a supplementary excitation at 1st engine order (EO1). The similar can be observed when intake and exhaust noise is considered. The standard 4-cylinder engine generates the most of the exhaust/intake noise at the 2nd engine order (EO2). If 2 cylinders are deactivated and corresponding valves are closed, the intake and exhaust gas pressure peeks will be less frequent, crating the excitation at 1st engine order – EO1. For 3-cylinder engines, the gas forces are only excitation forces. They correspond to 1.5 engine order (EO1.5). The similar can be observed for the intake and exhaust noise excitation. The cylinder deactivation, when 3 cylinder engines are considered, is a tricky task. In order to preserve constant firing interval, the cylinders have to be deactivated one-by-one following the firing scheme. A specific mechanism for ‘rolling cylinder deactivation’ is to be used. It allows one cylinder to be active during one 4 stroke cycle and deactivated during the following. On the same principle, the mechanism than deactivate the next cylinder, while activating the previous one. The final result is that, in average, only one and the half cylinder is working during two rotation of the crankshaft. The resulting gas forces due to the combustion correspond to the 0.75 engine order (EO 0.75). The resulting excitation frequency is very low. The same is valid for the intake and exhaust noise. Here bellow is a ‘not-exhaustive’ list of potential NVH impacts due to deactivation of cylinders:

6

New fuel-saving technologies and NVH refinement of powertrains

● Speed and torque irregularity ● generation of supplementary excitation of the crankshaft ● 4-cylinder engine (if 2 cylinders deactivated – EO1 excitation) ● 3-cylinder engine (turning deactivation – EO 0.75 excitation) ● low DMF mode needed if ‘flexible driveline’ ● Engine vibrations (of roll type) ● lower attenuation due to low frequency excitation (EO1 or EO 0.75) ● resonances of ‘powertrain on mounts’ narrow the drivable speed range ● more flexible mounts needed ● Air intake and exhaust noise ● ● ● ●

low frequency excitation when cylinders deactivated (EO1 or EO 0.75) larger intake manifold volumes needed (intake noise) larger exhaust muffler needed (exhaust noise) turbocharger noises ● difficulties for smooth functioning (EO1 or EO 0.75 gas pulses)

Variable compression ratio (VCR) technology increases global engine efficiency under varying load. Petrol engines have a limit on the maximum pressure during the compression stroke, after which the fuel/air mixture detonates rather than burns. To achieve higher power outputs at the same speed, more fuel must be burned and therefore more air is needed. This would result in detonation of the fuel/air mixture unless the compression ratio was decreased. This can be done to greater or lesser extent resulting in noticeable increase of engine power. Thus, a small efficient family car engine can turn into a highly tuned. For automotive use, this needs to be done dynamically in response to the load and driving demands. The VCR system has a positive influence on the fuel consumption and also optimizes emissions for diesel engines. In addition, VCR allows free use of different fuels besides petrol e.g. LPG or ethanol. The VCR systems attain the reduction of CO2 emission and fuel consumption by: ● Optimizing the compression ratio and maximal cylinder pressure ● Using of efficient Miller type cycle (higher expansion work during power stroke) ● Fine adjustment of admitted cylinder volume Variable compression engines have existed for decades, but only in laboratories, for the purposes of studying combustion processes. Earlier variable compression engines have been highly desirable but technically too complex and difficult to control. However, new cost-efficient innovative design solutions are proposed, which may change the future of VCR engines. Some of the most promising VCR engine have variable engine

7

New fuel-saving technologies and NVH refinement of powertrains

kinematics, like “Gomecsys” 4-cylinder engine. The modified kinematics introduces the supplementary low frequency excitation, which impacts the NVH behaviours: ● Speed and torque irregularity: generation of supplementary low frequency excitation EO1, consequently very low DMF mode needed ● Engine vibrations: supplementary low frequency excitation EO1 (roll type), consequently very low stiff mounts needed

3.4 Stop and start devices and hybridization The stop and start systems provide an important contribution in fuel economy with respect to CO2-reduction and they are mandatory for full hybrid vehicles. The consumption advantages are considerable: in the NEDC, the stop time totals 240 s, which is 20 % of the entire cycle. The event of an engine start/stop becomes a major issue of vehicle NVH refinement. Since idle time is becoming shorter, it tends to substitute idle comfort in importance from the NVH point of view. In this context, automatic engine start/stop systems are more critical from the NVH point of view compared to key start/stop due to the customer’s expectations. The vibration should be imperceptible while the engine stops, starts or re-starts. The potential NVH impacts are: ● Vibrations due to stop/start system ● vibrations during engine stop ● vibrations during engine re-start ● Noise when functioning ● ● ● ●

“clonk” – impact type noise due to the high vibration amplitudes gear noise of drives (whining) belt noise of starter drive (flapping and squeal) electric whining of starter

A stop-start system is a lead-in technology toward the hybridization. The fuels saving and CO2 rejection objectives have prompted most vehicle manufacturers to embrace propulsion technologies with varying degrees and types of hybridization. Nowadays, many different hybrid vehicle systems are either on the market, or under development, even up to all-electric vehicles. Each hybrid vehicle configuration brings unique NVH challenges that result from a variety of sources. The noise from electric machines such as motors and generators manifests in the form of whine noise, i.e., tonal noise (typically in the 400 Hz – 2000 Hz range). The tonal nature of the whine noise from the electric machines can be annoying to the customer. Missing the covering sound of the combustion engine, in several operating conditions, the noise generated by electric machines is getting more and more relevant for the acoustics of the passenger cars. The absence of masking combustion noise leads to a completely different subjective

8

New fuel-saving technologies and NVH refinement of powertrains

perception. This impression is even intensified by noise sources with no linear correlation to speed or demanded load. The transitions between the different operating modes are generally critical for hybrid vehicles. For instance, during the transition from a purely electric mode to a mixed mode with the combustion engine turned on, the starting process should not produce unpleasant noise and vibration phenomena. In both operating ranges, a capable powertrain control is needed, to avoid sudden torque changes and torque fluctuations

3.5 New automatized transmissions For the years, the automatic transmissions (AT) have been reputed to generate high fuel consumption due to low efficiency of the hydraulic torque converter. In order to overcome this negative point, the automated manual transmission is introduced in the late 90’s (a manual transmission automated by the use of clutch and transmission actuators). Its cost price was lower and fuel consumption in the NEDC cycle outperformed even the manual transmission variants due to optimized shifting points. Lately we see a reduction of this transmission type and this is caused by severe drivability issues. Another transmission that entered the market was the DSG (or DCT) transmission developed by Volkswagen. This transmission combined the fuel saving performance of the manual transmission with the drivability of a premium automatic transmission. During the last decade the efficiency of automatic transmissions is tremendously improved. The torque converter design is upgraded and new efficient lockup strategies are developed in order to minimize the dissipation of energy in the drive line (containing transmission). The new generation of automatized transmissions is designed to enable the engine to run at lower speed (down-speeding), which aids fuel economy, too. The overall reduction in engine speed improves NVH comfort and therefore the pleasant sense of well-being on board. It can also cut down external noise, thus reducing the strain on the environment. The wide ratio spread delivers outstanding low-end performance while small gear-ratio steps contribute to smooth transitions. Shortened shift and reaction times ensure optimum spontaneity combined with outstanding ease of shifting.The potential NVH drawback of the automatic transmission is excessive use of engine at low speed and high torque conditions. Such fuel saving strategy may impact NVH behaviour due to: ● High torque irregularity causing torsional vibrations of driveline ● booming noise at low engine speeds and high loads ● gearbox rattle due to high torsional excitation ● Vibration of powertrain (roll) ● roll engine vibration low frequency during ‘vehicle start’ and ‘idling’ ● high torque pre-charges which make mounts more stiff

9

New fuel-saving technologies and NVH refinement of powertrains

In the earlier automatic transmissions, the attenuation of torque irregularities in drive line was provided by excessive use of un-locked torque converter. The vehicle vibrations were reduced, but the fuel consumption was excessive. Nowadays, the efficient torsional dampers are used to attenuate vibrations, allowing lock-up rapidly and limiting the energy dissipation in hydraulic converter. Moreover, the engine load/speed duty point can be optimized with respect to the consumption and/or NVH criteria. By adding an electrical motor to the automatized transmission, the mild and/or full hybrid solutions can be realized without significant modification of the internal combustion engine.

4 NVH refinement levers for modern powertrain 4.1 Low frequency excitation The excitation frequency of engines depends on its design. For 4 cylinder engines the principal excitation corresponds to engine order two – EO2. For such engine the dominant vibration excitation corresponds to twice of engine speed expression in frequency (rotation per second). Principal vibration excitation for balanced 3-cylinder engine is EO1.5. The excitation frequency of EO1.5 is only ¾ of engine speed expressed in rotation per second. This results in lower frequency vibrations, which are more difficult to attenuate. If a 3-cylinder engine is not equipped with a balancing shaft, the principal excitation frequency is due to inertial forcing and it corresponds to the first engine order EO1, which means that is equal to the rotational speed of engine. Such a low frequency excitation is extremely difficult to attenuate since it narrows the resonances of engine suspension system and the drive line. If then, cylinder de-activation is applied to the 3-cylinder engine the lowest excitation frequency corresponds to EO0.75. Such a low frequency excitation may excite the existing resonances of suspension system, driveline shafts and vehicle structure. The NVH control of EO 0.75 is a highly tough task. One of the simplest attenuator of different type of fluctuations/oscillations is single degree of freedom system. The physical realization of such an attenuator corresponds: ● for mechanical vibration, to the mass-spring system ● for pressure pulsations (noise), to the Helmholtz resonator

10

New fuel-saving technologies and NVH refinement of powertrains

Figure 2 – Low frequency excitation and attenuation of single DOF system

The mechanical oscillator driven by a harmonic force F responds by a harmonic motion of the excitation frequency. The response amplitude is proportional to the excitation force and to the oscillator amplification factor. The amplification factor involves, the oscillator mass and the relative position of the excitation frequency to the resonant frequency. When the excitation frequency is narrows the resonance, the excitation is not well attenuated. Moreover, it can be event amplified.

4.2 Engine balancing Unbalance forces, corresponding to the single piston mechanism, combine when multi-cylinder engines are considered. The resulting unbalance force depend on the number of cylinders. At the slide above, 4, 3 and 2 cylinder in-line engines and their EO1 inertial excitations corresponding to ‘heave’ and ‘pitch’ vibrations are presented. Due to the symmetry of 4-cylinder engine (two pistons go up, while two go down), there is no any EO1 excitation (only EO2 vertical excitation). When 3 and 2-cylinder engine are concerned, the principal inertial forcing is EO1 ‘pitch’ (reciprocating masses). Balancing of 2 and 3-cylinder engine consists of annihilating the EO1 ‘pitch excitation’. It should be noted that the optimal balancing of each single piston leads to the optimal balancing of whole crank train. However, the mass optimum can only be reached by further optimization of distribution of counterweight masses, while keeping the same overall effect on the engine.

4.3 Engine suspension system The excitation frequency of engines depends on its design. For 4 cylinder engines the principal excitation corresponds to engine order two – EO2. For such engine the dominant vibration excitation corresponds to twice of engine speed expression in frequency (rotation per second). Principal vibration excitation for balanced 3-cylinder engine is EO1.5. The excitation frequency of EO1.5 is only ¾ of engine speed expressed in

11

New fuel-saving technologies and NVH refinement of powertrains

rotation per second. This results in lower frequency vibrations, which are more difficult to attenuate. The powertrain mounted to the vehicle is excited by multiple sources: ● inertial excitation due to vehicle movement (accelerations, curvilinear movement) ● road excitation (road profile roughness and route irregularity) ● powertrain excitation (different engine orders and sudden torque changes) In the domain of low frequencies, a powertrain on elastic mounts can be modelled as rigid body fixed to the grounded springs. Six resonant frequencies, corresponding to the modes of such system, can be observed. The idea is, that each mode is excited by only one: torque irregularity (roll), moving masses and road (have, yaw, …). When one force excites multiple modes, one says that the excited modes are ‘coupled’ with respect to the given force. As rule, ‘coupling’ of modes should be avoided (ie mode shapes should match the excitation forces – 6 DOF-s). The resonance frequencies of suspension modes should be low enough to ensure the appropriate attenuation.

Figure 3 – Static loads of a powertrain suspension and related non-linearity

The powertrain suspension should support all ‘static’ and ‘dynamic’ loads. The powertrain static ‘load’ corresponds to: ● weight of powertrain ● powertrain mean torque In some specific condition the ‘static load’ is increased by: ● intensive breaking ● important acceleration (high torque demand of driver) ● high road excitation (sudden slope change) which is then addressed as ‘quasi-static load’. The dynamic (vibration excitation) load is superimposed to the ‘slow changing’ loads. The principal deformation of elastic el-

12

New fuel-saving technologies and NVH refinement of powertrains

ement comes from ‘static’ and ‘quasi-static’ load, resulting in highly non-linear stiffness-displacement characteristics. The dynamic stiffness, corresponds to the derivative with respect to displacement / deformation evaluated at the static operating point. For the high static loads, the saturation of elastic elements leads to high values of dynamic stiffness and consequently to a low capacity of attenuation of vibration. For the full load conditions, the mean engine torque can be expressed as a function of engine speed. Since it is the principal ‘static’ load, the static deformation of suspension and resulting dynamic stiffness can be expressed as a function of the engine speed, as well as the eigen-frequency values and engine orders (excitation). The resonant behaviour, for full load acceleration, with high vibration levels, corresponds then to the duty points where the resonance curves are crossed by the engine orders excitation curves (particularly when the excitation type/pattern matches to the mode shape)

4.4 Torque irregularity filtering With the introduction of the dual mass flywheel (DMF), during the nineteen eighties, the driveline is almost completely isolated from oscillations in the crankshaft speed due to the combustion process. The DMF replaces the conventional single-mass flywheel (SMF) previously fitted as standard to all engines and is installed between the crankshaft and the clutch. Due to its exceptional ability to reduce torsional vibration of driveline the DMF allows the vehicle and engine to be driven at lower speeds, thereby reducing consumption and CO2 emissions. Nowadays, the position of the DMF as a standard powertrain element is well established. The efficiency of DMF in reducing vibrations is high, if the corresponding mode of the driveline (called DMF mode) is low comparing to the excitation (from idle to max speed range). What makes things more difficult with the 3-cylinder engine is that the DMF resonance narrows the drivable “desired” speed range due to the shrinking order of excitation. For the same speed, the engine excitation is relocated for factor ¾ to the lower frequencies, if 3 and 4 cylinder engines are compared (EO1.5 vs EO2). Consequently, the attenuation capacity of the DMF is deteriorated. The stiffness of the axle shafts has a considerable influence on the position of the DMF resonance. While “flexible” drive trains have a beneficial effect on DMF designs, “stiff” drive trains are particularly suited for conventional systems. These are generally less excited. This is particularly true for 4cylinder engine because of the annulation of torque irregularities when engine runs at higher speeds. Therefore, for 3-cylinder engines, the optimal damping system must be selected depending on the drive train stiffness. Both DMF and in the clutch disc long travel dampers can be used. With 2-cylinder engines, conventional solutions are used exclusively. With 2-cylinder engines, a DMF can be ruled out due to the very low frequency excitation (EO1). Here, the DMF resonance would always be within the driving range. Recently, the new concepts of torsional vibration dampers, based on centrifugal pendulum, are proposed for mid-range category of cars. Such element behaves

13

New fuel-saving technologies and NVH refinement of powertrains

as a “tuned damper”, reducing vibrations in the vicinity of its resonance frequency. Its inertial component is due to the centrifugal force and therefore increases with the rotational speed. Consequently, it can be tuned to the engine speed order instead to the constant frequency. The centrifugal pendulum system can be added to the single mass flywheel (SMF) or to the DMF. When SMF driveline is considered, the centrifugal pendulum is fixed between the clutch springs and the gearbox, somewhere on the friction plate. If it is applied to the DMF, then it is fixed to the secondary flywheel. The DMF with the centrifugal pendulum absorber corresponds to the range with the highest comfort demands. The pendulum absorber is generally tuned to attenuate the principal engine order, i.e. EO1.5 for 3 cylinder engines. Unfortunately, the engine does not generate the excitation only at principal engine order. For example, a 3-cylinder engine generates relatively important torque irregularity at EO3. The pendulum absorber tuned to EO1.5 is completely ineffective when EO3 excitation is considered. In order to overcome this weakness, the multiple pendulum absorbers are mounted on a single clutch system, each one tuned to one EO excitation. ● The principal NVH drawback of such system consists in the fact that vibration attenuation is limited due to complexity of dynamic behaviour (some new resonances may appear). ● The supplementary NVH risk consists in NVH behaviour during stop end start of the engine. In such conditions, the pendulum mechanism is potential noise source (‘clonk’ noise).

4.5 Engine encapsulation Nowadays the noise reduction of premium diesel engines becomes an important issue. Engine- and transmission-mounted encapsulation brings a radical improvement in the insulation of powertrain noise, without being intrusive to other engine parameters like: combustion efficiency, pollutant emission and torque/power. Insulation at source has important benefits not only in terms of NVH but also of improved fuel efficiency via engine heat conservation. An aligned layout of acoustical and thermal components enables an overall weight reduction of the engine insulation elements. The encapsulation parts are molded foam covers with a fiber top layer. These have been placed around the belts, the sump and the cylinder head improve NVH refinement and real life fuel efficiency. Due to legal restrictions the heat conservation measures have no or almost no effect in NEDC. Therefore, the EU initiated Eco-Innovations, which adds a customer benefit of fuel reduction in real life. Engine encapsulation can be one of those Eco-Innovations – [7].

14

New fuel-saving technologies and NVH refinement of powertrains

4.6 Other noises: intake-exhaust, accessory Downsizing, down-speeding and reduction of number of cylinders result in sever pulsating flow excitation. Inside of the manifolds and pipes, the flow pulsation is converted to pressure pulsation. The pressure pulsation is then propagating within the waveguide (manifold). From the acoustic point of view, air intake system (manifold) is an acoustic filter tuned to attenuate audible noise. An acoustic filter consists of an acoustic element or set of elements inserted between a source of acoustic signal and the receiver, like atmosphere or open space. The acoustic elements are: surge volumes, accumulators, acoustic filters etc. If appropriately applied, they reduce the noise generation within the air intake manifold. The dimensions of acoustic elements, ensuring an efficient sound attenuation depend on excitation frequency. Generally, larger acoustic devices are needed for low frequency excitation. The speed irregularity of modern downsized and down-speeded engines is very high. Consequently, the accessory belt and timing belt are put under high tension, in order to support engine speed irregularity. The high tension, results in increased sensitivity of such drives to the whining noise, emerging from the well accepted engine masking noise.

5 Concluding remarks 5.1 Summary of NVH drawbacks The potential NVH drawbacks are due to low frequency excitation: ● Speed/torque irregularity ● Inertial forcing – balancing shafts may be mandatory ● Exhaust and intake low frequency noise Main sound quality issues are: ● Accessory and accessory drive noise (high torque irregularity) ● Engine roughness (dynamics of crankshaft) of 3 and 2 cylinder engines Principal design measures are: ● ● ● ●

Low stiffness mounts (high decoupling capacity) Low torsional stiffness driveline filtering elements Larger intake volumes (manifolds and air filters) Larger exhaust volumes needed (silencer, muffler)

15

New fuel-saving technologies and NVH refinement of powertrains

5.2 NVH design challenges and Murphy’s Law The NVH design of machinery is tricky engineering domain. The acoustic and vibration phenomena involve small energies. For the standard powertrain, the overall energy involved in acoustic radiation is generally lower than 100 mW. The mechanical energy produced by the same powertrain is generally more than 100 kW. Consequently, by controlling the principal engine process, the NVH engineer is in charge to control the residual physical phenomena caused by a minor fraction of the main power (1 over 1 000 000). The precision and robustness of any numerical prediction, in such context, is obviously not very high. So the design decisions based on numerical predictions are to be carefully examined. This seems to be rather common substance for experimented NVH engineers. For that reason, Professor Rajendra Singh from Ohio State University, has formulated a special version of Murphy’s Laws concerning the relations between Noise Expectation/Prediction and Product Design: ● First Law: If there are two or more ways to design a product, both or all will lead to more noise. Alternately, if there is a way to design a quieter product, it will never be implemented due to cost, weight or political considerations ● Second Law: If a designer can find a way to degrade noise and vibration, he or she will be successful 99% of the time. Alternately, if there a 50-50 chance of increasing noise, it will go up 100% of the time. ● Third Law: An effort to enhance the energy efficiency will also increase noise and vibration levels – but in a non-proportional manner. ● Fourth Law: Any solutions to reduce noise and vibration that are offered by a manager will always fail – though after much testing. In the meantime, the manager will have been promoted. ● Fifth Law: Strengths of noise and vibration sources can only be reduced by adding money and weight. Does your experience confirm these laws?

16

New fuel-saving technologies and NVH refinement of powertrains

References [1]

T Hoffmann, H Richter: “New vehicle and engine concepts and their impact on vibroacoustics” – Schlegel &Partner 2009

[2]

M van Besouw, S Huijbers: “Future of Automotive Powertrains: Trends and Developments in engine and transmission ” – ACM 2015

[3]

L Gavric, S Courtois: “Specific NVH behavior of 3 cylinder engines” – Acoustic Automotive Confer ATZ 2011, Zurich 7-8 Jully 2011

[4]

A Ihleman, N Nitz: “Cylinder Deactivation – A technology with the future or niche application” – Schaeffler Colloquium 2014

[5]

K Govindswamy, T Wellmann and G Eisele: “Aspects of NVH Integration in Hybrid Vehicles” – SAE 2009

[6]

J Kroll, A Kooy, R Seebacher: “Torsional vibration damping for future engines” – Schaeffler Symposium 2010

[7]

J Gallinat: “Engine encapsulation for new generation engines of BMW” – Acoustic Autom Conf ATZ 2015, Zurich 23-24 Jully 2015

17

Diesel engine control based on structure-borne noise – optimization and adaptation of parameters M. Sc. Sebastian Schneider, OvGU Magdeburg Dr.-Ing. Jan Hendrik Carstens, TU Berlin Dipl.-Ing. Jürgen Nobis, IAV GmbH Prof. Dr. Hermann Rottengruber, OvGU Magdeburg Prof. Dr. Clemens Gühmann, TU Berlin Dipl.-Ing. Enrico Neumann, IAV GmbH Dipl.-Ing. Michael Joerres, Ford-Werke GmbH

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 W. Siebenpfeiffer (Hrsg.), Automotive Acoustics Conference 2017, Proceedings, https://doi.org/10.1007/978-3-658-20251-4_9

1

Diesel engine control based on structure-borne noise – optimization and adaptation …

1 Introduction Today’s passenger car diesel engines are distinguished by low exhaust emissions and low fuel consumption. In the lower load and engine speed range, however, dominant combustion noises are produced by the higher combustion delay, particularly when the engine is started and warming up. Although multiple pilot injections are used nowadays to reduce irritating combustion noise of this type, they clash with higherpriority exhaust emissions. Cylinder-pressure-based engine-management systems in particular are used to reduce fuel consumption and, as a result of this, CO2 emissions in future diesel engines. To achieve this without cylinder pressure sensor, the pressure variables needed for control purposes can also be determined on the basis of structure-borne noise signals. The advantage here lies in the fact that structure-borne noise signals also contain information on combustion noise, thereby permitting evaluation of the irritation caused by combustion noise. The idea behind the “Diese Engine Control Based on StructureBorne Noise” research project is to use structure-borne noise signals to reduce emissions from cylinder-pressure-based engine control while at the same time evaluating combustion noises without using cylinder pressure sensors.

1.1 Approach and Content of the Paper Cylinder-selective engine control and active manipulation of combustion for minimizing acoustic emissions demand detailed analysis of the relationship between combustion excitation and development of structure-borne noise. The facilities and test benches needed for experimentation are described in Section 2. To determine the specific injection or combustion-related influences of solenoid valve injectors in the structure-borne noise signal as accurately as possible, it is beneficial to examine these influences separately (Section 2.4). For cylinder-selective engine management not requiring combustion chamber pressure sensors but capable of influencing the diesel noise rating, virtual sensors need to be used for cylinder pressure and airborne noise. These virtual sensors are presented in Sections 3 and 4, and provide the basis in Section 5 for presenting implementation of cylinder-selective combustion control with which the diesel noise rating can be optimized by varying the engine’s injection parameters.

2 Test Bench Setup and Measurements In order to carry out the studies, a production four-cylinder diesel engine was set up on the acoustic test bench at the Institute for Mobile Systems (IMS) at the Chair for Energy Conversion Systems for Mobile Application at the University of Magdeburg.

2

Diesel engine control based on structure-borne noise – optimization and adaptation …

The test bench chamber is lined with sound-absorbing walls, with the centrally positioned test object being spaced at a sufficient distance from the wall surfaces to perform reliable sound pressure measurements in the near and far field. Furthermore, an e-motor is used both as a starter as well as a brake for generating engine load. This is connected to the engine’s crankshaft by a one-and-a-half meter vibration-damping shaft. For acoustic reasons, the e-motor is accommodated in the adjacent room which, in turn, explains the length of the test bench shaft. The test engine is mounted on elastic bearings both at the front end as well as on the pressure side, the latter being provided in the form of a torque support.

2.1 Test Bench Configuration Unlike the research project [1], this project’s objective is to examine and implement a cylinder-selective diesel engine control system based on structure-borne noise under near-production conditions and over an extended engine speed and load range. For this reason, the injectors are controlled and actuated by means of the engine’s production control unit. The setpoint settings for the injection parameters as well as the control unit data are implemented by CAN bus using the ATI VISION data acquisition and calibration software from ATI Technologies Inc. In this application, only calibration parameters are overwritten, avoiding any integration or modification of the software structure.

Figure 1: Schematic diagram of test bench configuration

The algorithms required for processing engine speed, current, structure-borne noise and microphone signal data are then provided by IAV’s MATLAB®/Simulink®programmable IAV-MPEC (Modular Prototyping Engine Controller) rapid prototyping system. The MPEC system is based on an industrial PC, IAV’s FI (Flexible In-

3

Diesel engine control based on structure-borne noise – optimization and adaptation …

jection and Ignition for Rapid Engineering) injection control unit with two TRA cards (Thermodynamic Realtime Analysis). The FI system permits work-cyclesynchronous evaluation and processing of the structure-borne noise and microphone signals as well as the engine speed signals in real-time. Following this, the characteristic values computed from the signals are then sent to the industrial PC by CAN bus. These data are available to the control loops developed for cylinder-selective diesel engine control based on structure-borne noise. The control setpoint values are then transferred to the control unit by CAN bus using the ATI VISION software. Here, the control unit’s calibration parameters are merely overwritten. A schematic diagram of the engine test bench with its various system components is shown in Figure 2.

2.2 Testing Various engine measurements were conducted at steady-state working points for the analyses and for modeling the virtual sensors. Following coordination with the working group, the experimental designs were created using the “Design of Experiments” method to minimize the number of measurements needed. This involved examining and measuring various engine maps in the engine speed and load range of relevance to the diesel noise rating and also at salient map points in relation to the psychoacoustic variables of loudness and modulation (1250 rpm/25 Nm, 1750 rpm/100 Nm) relevant to the diesel noise rating. Additional parameter variations (modifying the injection parameters) were carried out at many of the working points at which the engine was operated.

3 Development of a Virtual Cylinder Pressure Sensor Information on the cylinder pressure profile is imperative for developing and calibrating a cylinder-pressure-based engine control system. Unlike pressure sensors, the characteristic variables of indicated mean effective pressure , maximum cylinder pressure and center of heat release are to be determined by means of a virtual cylinder pressure sensor. Based on the methods developed and preliminary studies carried out in FVV’s “Noise-Controlled Diesel Engine I & II” research project Es ist eine ungültige Quelle angegeben., these are adapted, extended and validated for the current test engine.

3.1 Preliminary Studies on COHR in the Time and Frequency Range In a combustion engine, the cylinder pressure profile plays a key part in exciting noise emissions. Acceleration sensor positioning has fundamental effects on the information contained in the structure-borne noise signals which can be induced by combustion in the

4

Diesel engine control based on structure-borne noise – optimization and adaptation …

cylinder. Depending on engine structure, amplitude is attenuated and a phase lag induced between pressure profile and structure-borne noise. In principle, it is possible to reconstruct the engine’s transmission behavior from the information provided by the cylinder pressure excitation spectrum and noise emission. In application practice, however, noise emissions are influenced by mechanical noise interference, resulting in incorrect determination of the engine’s noise transmission structure [4]. It does, however, make sense to examine the cylinder pressure excitation spectrum generated by combustion as a means of checking the plausibility of the structure-borne noise spectrum measured. For this reason, the combustion engine’s cylinder pressure excitation spectrum is determined first.

3.1.1 Excitation Spectrum of the Cylinder Pressure Profile Cylinder pressure development in a direct-injection diesel engine plays the key part in exciting the noise signals in the engine block. Depending on the cylinder pressure profile, different frequency ranges are excited in this context. In accordance with [6, 4, 7], the characteristic pressure variables can be matched up to specific frequency ranghas an effect up to a frees in the cylinder pressure spectrum. Peak pressure quency of approx. 10 ⋅ , maximum cylinder pressure speed max ( / ) up to a

frequency of approx. 40 ⋅ and maximum cylinder pressure acceleration max ( / ) beyond this (Figure 2). Above approx. 4 to 7 kHz, resonances dominate the cylinder pressure spectrum. These correlations are illustrated in Figure 2. Depending on combustion, the various frequency ranges may shift [7]. The cylinder pressure excitation spectrum (power density) is produced from the pressure spectrum P ( f ) ( )=2⋅

( )=2⋅

∗(

)

with the reference sound pressure

⋅

( )

,

> 0,

(1)

being defined as 20 μPa.

Figure 2: Qualitative cylinder spectrum with characteristic pressure variables [6]

5

Diesel engine control based on structure-borne noise – optimization and adaptation …

Figure 3 and Figure 4 show the excitation spectra for the working points 1250 rpm, 25 Nm and 1750 rpm, 100 Nm from a map measurement. The spectrum between 50 – 700 Hz falls at an attenuation of 40 dB/decade. This range is dominated by the maximum cylinder pressure speed. In the frequency range between 700 Hz to 2000 Hz, the excitation spectrum decreases by as much as 60 dB/decade as a result of maximum cylinder pressure acceleration. Above approx. 2 to 3 kHz, the spectrum in part rises again, induced by resonances in the cylinder pressure signal.

Figure 3: Cylinder pressure excitation spectrum for cylinder pressure signals − at working point 1250 , 25 Nm from map measurement

Figure 4: Cylinder pressure excitation spectrum for cylinder pressure signals − at working point 1750 , 100 Nm from map measurement

These analyses make it possible to confirm the qualitative correlation between cylinder pressure characteristic and excitation spectrum from [6, 4, 7]. Excitation signals up to 700 Hz are mainly triggered by maximum cylinder pressure speed max ( / ) at the two working points examined. Maximum cylinder pressure acceleration max ( / ) characterizes the frequency range of approx. 700–2000 Hz in the excitation spectrum. Higher frequency components from 3 kHz cannot be matched up to the combustion characteristic with any certainty. It is likely that these frequencies are dominated by cylinder pressure resonances.

3.1.2 Studies on Combustion Excitation in the Acceleration Signal by Means of Wigner-Ville Distribution Coherence analysis [3] for the project engine has revealed a linear relationship between cylinder pressure signal and structure-borne noise signal in the frequency range up to 3 kHz. Determining the frequency bands excited in the structure-borne noise

6

Diesel engine control based on structure-borne noise – optimization and adaptation …

signal by combustion involves a time-frequency analysis. For this purpose, SmoothedPseudo-Wigner-Ville distribution (SPWV) is used for producing an anglesynchronous graph showing the relationship between structure-borne-noise and cylinder pressure signal. This is followed by an analysis of the time-based correlations between cause and effect or excitation by cylinder pressure and signal inputs in the structure-borne noise. The aim of this analysis is to extract a position information characteristic from structure-borne noise that correlates with the center of heat release. The following looks at individual principles. Only the structure-borne noise signal from sensor is examined in closer detail. Sensors − are included in overall evaluation. A series of measurements was carried out at the 1250 rpm, 25 Nm working point to examine the principles involved. To this end, main injection was increased from -9 °CA to 0 °CA in fixed increments. For these measurements, the load was in each case adapted, with rail pressure and pilot injection being kept constant. Table 1 shows the characteristic variables for the measurement series. Table 1 Characteristic variables for measurement series at 1250 crement variation Operating point

, 25 Nm with in-

1

2

3

4

5

6

7

M [Nm]

24.7

24.9

24.7

25.1

25.0

25.0

25.1

[rpm ]

1250

1250

1250

1250

1250

1250

1250

2.7

2.7

2.7

2.7

2.7

2.7

2.6

60.2

57.6

55.2

52.2

48.9

46.2

43.7

3.4

4.6

6.0

7.8

9.7

11.4

14.0

4.3

3.6

3.0

2.4

2.0

1.7

1.7

-1.7

0.5

2.2

4.4

6.7

8.7

11.6

1.2

1.2

1.2

1.0

0.9

0.9

1.0

[bar] [bar] {

} [°CA]

max ( °CA]

/

) [bar/

max max ( / [bar/°CA ] max

[°CA] )

-3.5

-1.2

0.3

2.5

4.7

6.8

9.6

[°CA]

-16.9

-16.9

-16.9

-16.9

-16.9

-16.9

-16.9

[°CA]

-14.1

-14.1

-14.1

-14.1

-14.1

-14.1

-14.1

[°CA]

-11.2

-9.2

-7.3

-5.1

-2.8

-1.1

0.9

[°CA]

-6.5

-4.1

-2.4

-0.1

2.1

3.9

6.0

0.1

2.5

4.5

7.1

9.1

11.2

13.0

[°CA]

[°CA]

7

Diesel engine control based on structure-borne noise – optimization and adaptation …

[

]

365

365

365

365

365

365

365

By way of example, Figure 5 shows the graph-based evaluations of structure-borne and the cylinder pressure signal for one operating point used for the noise sensor measurements. The top part shows structure-borne noise signal from the sensor with the bandwidth up to 44.1 kHz. As there is coherence between structure-borne noise signal and cylinder pressure up to a frequency of approx. 3 kHz, structure-borne noise signal is additionally bandpass-filtered. Filtered structure-borne noise signal has cutoff frequencies of 200 Hz and 2.5 kHz. The lower cutoff frequency is selected on the basis of the fact that low-frequency mechanical vibrations are contained in the structure-borne noise signal. The maximum amplitude of bandpass-filtered signal is also plotted in the diagram. The middle section shows the SPWV of the structureborne noise signal, cylinder pressure speed / and cylinder pressure acceleration / . The lower section of the diagrams shows the characteristics of cylinder pressure , / as well as cylinder pressure speed / , cylinder pressure acceleration as a function of time. Here too, the results are discussed by way injector current of example on the basis of one working point and one operating point. A more detailed description can be found in [3]. Figure 5 initially provides conclusive confirmation of the excitation spectrum from the cylinder pressure profile (cf. Section 3.1.1). The cylinder pressure speed significantly dominates the lower frequency range up to 1 kHz, with clear evidence that cylinder pressure speed, resulting from the compression phase and expansion phase, generates a frequency excitation below approx. 500 Hz over the entire angle range.

8

Diesel engine control based on structure-borne noise – optimization and adaptation …

Figure 5: Time-frequency analysis of structure-borne noise signal of sensor in comparison / and cylinder pressure to cylinder pressure of the 1st cylinder, cylinder pressure speed / . Measurement series 1250 , 25 Nm, 1st operating point. acceleration

9

Diesel engine control based on structure-borne noise – optimization and adaptation …

In particular, cylinder pressure acceleration excites frequency ranges between 1 kHz and 2,5 kHz. Unlike the spectrum of cylinder pressure speed, cylinder pressure acceleration causes frequency excitation in a narrower angle interval. The spectrum of cylinder pressure acceleration is excited in the angle range from -5 to 0 °CA. In contrast to this, the significant spectrum of cylinder pressure speed lies in the angle range of 10 to 10 °CA. In comparison to the structure-borne noise signal, it can be seen that the spectrum correlates with the frequency excitations of cylinder pressure speed and cylinder pressure acceleration over time. Further evaluations showed that shifting the injection time of main injection ( ) shifts the time lag of frequency excitations in all signals. It must be noted in particular that a maximum of filtered structure-borne noise signal is clearly developed and also correlates over time with frequency excitation of the cylinder pressure speed and cylinder pressure acceleration signals. Angle position {max( )} was extracted in [1] as a characteristic of the center of heat release. Results of the preliminary studies suggest that this characteristic can also be used as position information for combustion. It was established in the FVV research project [1] that angular position {max( )} correlates with the angular position of maximum cylinder pressure speed (max { / }). The time-frequency analysis in Figure 5, however, shows in particular that cylinder pressure acceleration / induces frequency excitation between 1 kHz and 2.5 kHz.

10

Diesel engine control based on structure-borne noise – optimization and adaptation …

Figure 6: Comparison between the signal for cylinder pressure , cylinder pressure speed / of 1st cylinder, as well as of structure-borne / and cylinder pressure acceleration noise signal of sensor , while varying (measurement series 1250 , 25 Nm)

11

Diesel engine control based on structure-borne noise – optimization and adaptation …

To check plausibility in terms of which cylinder pressure excitation signals are capable of significantly influencing structure-borne noise, the pressure signal, pressure speed, pressure acceleration as well as the bandpass-filtered structure-borne noise signal for all operating points of the measurement series at 1250 rpm, 25 Nm and 1750 rpm, 100 Nm are presented in Figure 9. It can be seen from these diagrams that the angular position of the maximum amplitude of bandpass-filtered structure-borne noise signal lags behind the angular position of the maximum amplitude of pressure acceleration and runs ahead of pressure speed.

Figure 7: Correlation diagram of the characteristic from the structure-borne noise signals and position information from maximum cylinder pressure acceleration for the “DoE load/engine speed” measurement series

12

Diesel engine control based on structure-borne noise – optimization and adaptation …

Taking causality into account, it follows that the structure-borne noise signal is primarily excited by cylinder pressure acceleration / . This gives rise to the hypothesis that angular position {max( )} = represents a characteristic for _ ( )}. the center of heat release at maximum pressure acceleration { / Based on a series of complex measurements, it was possible to verify [3] that characteristic {max( )} from structure-borne noise provides reliable information on the center of heat release even after varying pilot injection, main injection, engine load, rail pressure and EGR. For a load range of 25 Nm to 250 Nm and an engine-speed range of 850 rpm to 2250 rpm, the characteristics from structure-borne noise sensors to were evaluated in relation to the corresponding cylinder pressure to , with the structure-borne noise signal being used for the nearest cylinder. Four signal pairs (structure-borne noise signal from – cylinder pressure etc.) are produced. Figure 10 clearly shows that the structure-borne noise characteristics exhibit a linear relationship to the position of maximum cylinder pressure acceleration.

3.1.3 Modeling The main center of heat release , maximum pressure and indicated mean effective pressure ̂ are key variables for control. They are modeled in relation to position information {max( ), start of injection ( , ), injection period (Δ , Δ ), rail pressure ( ), EGR as well engine speed ( ). EGR is influenced by the target position for the EGR valve. As deposits can cause the end position to vary from the EGR valve target position, the EGR rate is not considered any further for modeling. This produces the following model approach for the variables to be estimated: = ̂ ̂

= =

( (

,

_

,

_

(

with the notation tion.

, , ,

_ _

,Δ ,Δ ,

,Δ

,Δ ,Δ ,Δ

, , , , , ,

), ). )

(2) (3) (4)

= {max( )} being introduced for the sake of simplifica-

A gradual regression is in each case applied for determining these functions. Linear and quadratic model terms as well as interaction terms are used for this purpose. In this application, the significance of the parameters is evaluated by means of a hypothesis test [8]. If a parameter is not significant, it is not included in the regression function. As a criterion, a type 1 error probability of 10 % was applied in all models. To estimate the parameters for the regression models, the measurement data from a map measurement test series with ECU control (51 operating points) were combined with those from a “DoE load/engine speed range” test series (53 operating points). Twelve of the measurement data items were used here for offline verification. The

13

Diesel engine control based on structure-borne noise – optimization and adaptation …

center of heat release varied in the range from approx. 0 to 20 °CA. The following regression model was produced for estimating the center of heat release on the basis of the control unit variables and structure-borne noise characteristic : _ = + ⋅Δ

Δ +

⋅

+ ⋅ + ⋅ + ⋅ + ⋅ Δt + ⋅ _ ⋅ + ⋅ + ⋅ + ⋅ + ⋅ Δt + _ + ⋅ + ⋅ Δt ⋅ + ⋅p ⋅ Δt + ⋅ ⋅ ⋅ ⋅ + ⋅ ⋅ + ⋅ ⋅ + _ _ ⋅ + ⋅ ⋅Δ + ⋅ ⋅ _

(5)

Similar models for maximum pressure and indicated mean effective pressure can be found in [3]. Section 3.1 only examined the relationship between the structure-borne noise characteristics and the adjacent cylinder. ( in relation to cylinder pressure , as well as between and , and , and ). The structure-borne noise signals, however, also deliver information on combustion in more remote cylinders. It was possible to obtain COHR information from the sensor signals ( − ) for all cylinders. Combustion excitation of all cylinders is significant enough to detect COHR information by means of a single structure-borne noise sensor. This _ provides the capability of generating separate models of the main center of heat release for each cylinder using just one structure-borne noise sensor. The absolute error between measured and estimated main center of heat release is well below 2 °CA. The adjusted coefficient of determination is used as the criterion for evaluating the models. The relative RMS (root mean square) error in percent is used to determine the accuracy of the estimated value in relation to the measured value: % (y,

)=

1



− ∙ 100. max( ) − rpm( )

(6)

Regardless of the position of structure-borne noise sensors – , the adjusted coefficient of determination reaches values well above 0.98 for all estimated main centers of heat release for all four cylinders. This high level of model quality has a positive effect on the RMS error. The maximum RMS error is 0.630 °CA, equating to a relative error of 2.841 %. The smallest RMS errors were reached for the structurebornenoise sensors positioned closest to the cylinders. One reason for this relationship can be found in the distance covered by combustion noise and the measured structureborne noise signals because engine structure attenuates signal amplitudes more intensely the further the distance between the point of origin and sensor. Consequently, extracting characteristics from the structure-borne noise signal (cf. Section 3.1.2) comes with greater uncertainty at the main center of heat release.

14

Diesel engine control based on structure-borne noise – optimization and adaptation …

For indicated mean effective pressure, the center of heat release was varied in the range from approx. 2.5 to 20 bar in an engine speed range of 850 to 2500 rpm. At most operating points the absolute error between measured and estimated mean effective pressure was below 2 bar. The models for indicated mean effective pressure have a coefficient of determination of between 0.979 and 0.996 and RMS errors below 5 %. Variations in the range of approx. 30 to 170 bar in an engine speed range of 850 to 2500 rpm were evaluated for modeling maximum cylinder pressure. The models for indicated mean effective pressure show a coefficient of determination of between 0.973 and 0.983. The maximum relative error is below 7.5 %.

4 Development of a Virtual Noise Sensor The diesel noise rating computation method from FVV’s “Objectivizing Subjective Assessment” project [9] is used for evaluating diesel knock. Computing the diesel noise rating is based on airborne noise signals recorded using microphones. A model now needs to be created on the basis of the structure-borne noise signal that estimates the diesel noise rating for airborne noise (virtual noise sensor). The diesel noise rating is based on determining loudness and modulation. The basic principle for this this is described in [2] for example.

4.1 Diesel Noise Rating A measure of combustion noise is cylinder pressure [10, 11]. However, it lacks information on the subjective perception of irritation from combustion noise. [9] introduces a characteristic value for the diesel noise rating which permits subjective assessment of the difference between combustion noise and perceived noise on the basis of objective airborne noise signals. In particular, the focus here is placed on assessing diesel knock. The evaluation criteria are based on assessing the weighted loudness as well as on modulation of airborne noise. The diesel noise rating from diesel knock is defined in the steady-state part-load range as =

+

⋅

+ ⋅

,

(7)

where = 10.2; = 2.42 and = 0.1 are experiment-based regression parameters. The diesel noise rating has a range of 1 (unacceptable) to 10 (undetectable). [3] provides a detailed description of the principles applied. It examines the way in which the diesel noise rating is governed by the start of pilot injection, injected fuel quantity and start of main injection. The pilot injection angular position was varied at two operating points. This showed that the influence on the diesel noise rating de-

15

Diesel engine control based on structure-borne noise – optimization and adaptation …

pends on the engine’s operating point. The influence on the diesel noise rating at the 1250 rpm, 25 Nm working point is marginal. Shifting the pilot injection angular position from advanced to retarded at the 1750 rpm, 100 Nm working point, however, resulted in an improvement by rating grade of approx. 0.75. In a second step, the pilot injection quantity was varied at two operating points. Increasing the pilot injection quantity to as much as 1 mg/stroke led to a direct reduction in combustion noise irritation. The diesel noise rating rose by up to 1.5 grades. Above 1 mg/stroke and depending on working point, it was not possible to achieve any significant improvement in the diesel noise rating. In particular, diesel noise rating stagnated at the 1750 rpm, 100 Nm working point and fell slightly despite increasing the pilot injection fuel quantity. The last step involved examining the extent to which shifting the main injection angle affects the diesel noise rating. Shifting the angular position from advanced to retarded led to the diesel noise rating increasing by 1.5 grades, whereby, as a critical observation, retarding the main injection event too far can result in delayed combustion. Cylinder pressure development was significantly influenced in the same way as the position of main injection. Retarding the main injection led to a direct fall in cylinder pressure speed and maximum cylinder pressure. One drawback of influencing the diesel noise rating by shifting the main injection angular position, however, lies in the fact that cylinder pressure development, and therefore combustion, is influenced to a far greater extent than by the pilot injection parameters. At the same time, it can be assumed that exhaust emissions vary more widely than by manipulating the pilot injection parameters.

4.2 Determining the Diesel Noise Rating from Structure-Borne Noise Signals Given the need for airborne-noise signals, however, determining the diesel noise rating is limited to acoustic test benches because when used in the vehicle, interfering noises, such as driving noises, superimpose the signals. The research project [1] showed that structure-borne noise signals can be used for estimating airborne-noise signal loudness and modulation. The aim is to use the method presented in [1] for the current test engine and adapt it for an extended engine speed and load range. Estimating the diesel noise rating provides a way of evaluating combustion noise irritation online and integrating it into the cylinder-pressure-based engine control system.

16

Diesel engine control based on structure-borne noise – optimization and adaptation …

Figure 8: Methods for determining diesel noise rating

A high level of correlation between airborne and structure-borne noise signals, or loudness, modulation and diesel noise rating, was verified in [3]. The next step aims to show the diesel noise rating value range from airborne sound using structure-borne noise signals. This uses the idea of creating a regression model that estimates characteristic airborne sound values on the basis of structure-borne noise signals. This model is referred to as a virtual noise sensor. The approach reflects the concept described in Section 3 for virtual pressure sensors. The diesel noise rating can be determined indirectly by estimated and modulated from structure-borne noise. Regression models are defined for these two variables. The diesel noise rating can then be calculated from =

(

= =

, (

+

⋅

Control unit data ( .

)

(8) )

,

+ ⋅

(9) (10)

) can be used for determining regression models

and

As in Section 3.2, a gradual regression analysis is used for determining regression functions , . In this application, linear and quadratic model terms are included in regression analysis. This model approach incorporates the information provided by the control unit variables of start of injection ( , ), injection time (Δ , Δ ), rail pressure ( ) and engine speed ( ). To determine the regression function parameters, 86 measurements were conducted in the engine speed range from 1000 rpm to 2500 rpm and load range from 25 Nm to 250 Nm while varying the injection parameters. 21 measurements were selected at random from this dataset to ver-

17

Diesel engine control based on structure-borne noise – optimization and adaptation …

ify the regression modes. In particular, it can be seen from the detailed analyses in [3] that loudness reaches a coefficient of determination of over 0.93 no matter where the microphone and structure-borne noise sensors are positioned. Model quality is shown in Figure 12. Generally speaking, modulation from microphone signals recorded on the cold engine cannot be modelled quite as well. The higher stiffness of the engine block results in a far poorer correlation of modulation between the microphone’s airborne noise and structure-borne noise on the cold side. Consequently, the regression model selected for modulating the microphone signal on the cold side cannot be reproduced with sufficient accuracy. However, the models for modulating the hot and front end reach a coefficient of determination of over 0.92 for all structure-borne noise sensors. The low discrepancies between estimated and measured diesel noise rating are achieved with the structure-borne noise sensors at the main bearing. They lie between 5 % and 7 %, equating to an RMS error of approx. 0.2 of a rating grade. The greatest discrepancy of approx. 9 % results from using structure-borne noise information from the sensor positions on the hot side which equates to a discrepancy of approx. of 0.32 of a rating grade. To estimate the irritation caused by combustion noise, the models using the signals from the structure-borne noise sensor at the main bearing in particular produce the lowest discrepancies. For use at an industrial level, however, the practicability of retrofitting and maintenance at this sensor position is limited. As structure-borne noise sensors to are used for the virtual pressure sensor, using these sensors provides a way of reducing structure-borne noise sensors needed. With regard to estimating the diesel noise rating, the lowest discrepancies are produced for sensor – for this reason, the diesel noise rating is determined below using this sensor.

5 Cylinder-Selective, Noise-Controlled Diesel Engine Sections 3 and 4 looked at the development of virtual cylinder pressure and noise sensors. These virtual sensors provide the capability of estimating the combustion variables per working cycle as well as of evaluating the irritation of combustion noise from the test engine. On the basis of these virtual sensors, the aim is to implement a cylinder-selective combustion control system (Sections 5.1.1 and 5.1.2) as well as optimize the diesel noise rating by varying the engine’s injection parameters (Sections 5.2 and 5.3). In this application, control is to be implemented under near-production conditions. Under near-production conditions means controlling the injection parameters using the production control unit and leaving the basic software architecture unchanged. The entire control structure is shown in Figure 10.

18

Diesel engine control based on structure-borne noise – optimization and adaptation …

Figure 9: Diagrams showing correlation between virtual noise sensor by means of and diesel . Top: correlation between estimated noise rating computed by means of microphone loudness and computed loudness of the diesel noise rating. Center: correlation between estimated modulation and computed modulation of the diesel noise rating. Bottom: correlation between estimated and computed diesel noise rating.

19

Diesel engine control based on structure-borne noise – optimization and adaptation …

5.1 Cylinder-Selective Combustion Control The models developed for estimating cylinder-selective combustion variables , were presented in Section 3. These estimated variables are to be used to ̂ and ̂ realize a cylinder-selective combustion control system. Here, the control strategy involves giving all cylinders equal status in relation to indicated mean effective pressure and controlling the main center of heat release. Main injection fuel quantity and main injection angle .serve as the control variables. For this reason, it is necessary to examine whether the quality of the estimated values from the virtual pressure sensor is sufficient to use these for control purposes.

Figure 10: Diagram illustrating the principle behind the control structure

Below, the individual control loops are presented and integrated into the FI2RE Commander. The chosen control parameters are communicated to ATI-Vision by a CAN bus and the control variables overwritten in the control unit. This avoids any intervention in or change to the control unit’s software structure and is intended to make it easier to adapt the control structures for near-production application.

20

Diesel engine control based on structure-borne noise – optimization and adaptation …

5.1.1 Controlling Indicated Mean Effective Pressure Indicated mean effective pressure is a direct measure of engine work. In an initial approach, therefore, a system is to be provided for controlling indicated mean effective pressure. Here, indicated mean effective pressure is to be recorded on a cylinderspecific basis and corrected for each cylinder by means of the main injection fuel quantity. This is necessary because the injected fuel quantity can vary depending on the condition of the solenoid injectors. This must be expected in particular if aging effects are at play or the injectors are damaged. The principle behind the control structure based on the structure borne-noise sensor is shown in Figure 58. Proceeding from the regression models of sensors , … , , indicated mean effective pressure ̂ ( ,…, ) is ascertained for each cylinder. In this context, these estimation algorithms are referred to as virtual sensors. ) is resident in a map as a reference value. The desired mean effective pressure ( Furthermore, the main injection fuel quantity values calibrated by the control unit ( , ) are also read from this map. As a result, the controller must only compensate for the target discrepancies between desired reference value and estimated value of indicated mean effective pressure. For this purpose, a variable main injection fuel quantity (Δ ) is added to the control unit’s main injection fuel quantity. The to) produced in this way is then communicated to tal main injection fuel quantity ( the control unit as the setpoint value which subsequently adapts the main injection time of the respective solenoid injectors. Here, four separate PI controllers, each individually parametrized, serve as controllers. To verify the control structure, a step measurement of + 1 bar is carried out around the 1250 rpm, 25 Nm operating point ( = 2.8 ) (Figure 11). The measurements were conducted as follows: To begin with, the engine is taken to the operating point by the control unit. The cylinder-selective control structure is then activated. After a brief dwell time, the desired reference value ( ) is increased by 1 bar. Measurement is concluded when the controlled variable has reached a settled state. The diagrams in each case show the measured indicated mean effective pressure ( ), the estimated indicated mean effective pressure ( ̂ ) as well as the reference variable ( ) for each cylinder. It can be seen from the step responses that, for all control loops, transient behavior is concluded after approx. 50 working cycles, with the desired final value being reached for the reference variable. The error between measured indicated cylinder mean effective pressure and estimated cylinder mean effective pressure shows a brief, dynamic absolute error of below 0.35 bar in response to a change in reference variable. This behavior can be explained by the latency between control unit and the FI system.

21

Diesel engine control based on structure-borne noise – optimization and adaptation …

This error is minimized once the control loops are in the final steady state because variation in the control unit data decreases and run-time delay has no effect on the estimated values.

5.1.2 Controlling the Center of Heat Release Particularly when it comes to minimizing exhaust emissions it is essential to maintain the main center of heat release. Any deviation from the desired or calibrated center of heat release can influence emission values significantly [12]. In particular, deviations of this type can be caused by aging effects in the injection system, e.g. resulting in insufficient or too much fuel being injected. To avoid such actuator deviations, the injection parameters need to be adapted in the form of a control structure. At commercial level, the center of heat release can be determined by cylinder pressure sensors. However, the virtual pressure sensors developed in Section 3 provide the alternative for ascertaining the center of heat release on the basis of structure-borne noise sensors. The regression models generated provide the basis for estimating the respective main centers of heat release ( ,…, ) for each cylinder in synchrony with working cycle. Adapting the center of heat release involves controlling the injection angle. The desired center of heat release ( ) is fed into a map as a reference value. The angular position of the start of main injection is used for influencing the center of heat release. The main injection angle ( ) calibrated in the control unit is also used in the map for pilot control. The system controlling the center of heat release adapts this angle (Δ ) to compensate for the error between reference value ( ) and ac( ,…, ). This means that the injection angle of each injector can tual value always be varied for each cylinder. However, as the control unit used permits no direct software-based access to each individual main injection angular position, it is only possible to set one angular position for all injectors. This means it is not possible to vary the angular position on a cylinder-selective basis on the test engine. So, although all main centers of heat release were determined for the cylinders, only the main injection angle of cylinder 1 was used for control purposes. This means that adaptive angular position Δ was applied to all of the other cylinders.

22

Diesel engine control based on structure-borne noise – optimization and adaptation …

Figure 11: Step response of control to positive setpoint, plotted as a function of the working cycles (WC). The diagrams show the reference value, the measured and estimated indicated mean effective pressure in relation to the individual cylinders and structure-borne noise sensors

Figure 12: Step response of COHR control to positive setpoint, plotted as a function of the work cycles (WC). The diagrams show the reference value, the measured and estimated in relation to the individual cylinders and structure-borne noise sensors

A PI controller was configured for control purposes. To verify the control structure, a step measurement of + 1 °CA is carried out around the 1250 rpm, 25 Nm operating point ( ( ) = 7.8 ° ). Figure 12 shows the measured ( ) and estimated cen) as well as the reference variable ( ) for each cylinder. ter of heat release ( It can be seen that the steady-state final values are reached after approx. 40 working cycles.

5.2 Noise Control The specially developed virtual noise sensor from Section 4 can be used for evaluating combustion noise online. One question is how to use this diesel noise rating information for controlling the engine in an effort to reduce noise emissions from combustion. In the research project [1] as well as in [2], the center of heat release was shifted on the basis of a setpoint diesel noise rating. To compensate for any change in indicated mean effective pressure, the main injection fuel quality was also adapted. This concept is based on prior knowledge that the setpoint diesel noise rating can also be achieved by shifting the center of heat release. The drawback of this concept is that reaching this setpoint value must be available as information or the controller’s manipulated variable may be limited or combustion delayed. An alternative concept in this research project is not to influence the diesel noise rating by means of the center of heat release but to maximize the diesel noise rating by varying the pilot injection parameters.

23

Diesel engine control based on structure-borne noise – optimization and adaptation …

The center of heat release and indicated mean effective pressure are to be controlled here using the structure presented in Section 5.1. The desired reference values are adopted by the calibrated control unit. As a result, influence on the combustion variables can be compensated by varying pilot injection. The noise control structure is shown in Figure 10. Noise behavior, or the diesel noise rating of the combustion engine, is estimated by means of the virtual noise sensor. Varying the pilot injection quantity (Δ ) and angular position (Δ ) aims to maximize the diesel noise rating. The control law underlying these two controlled variables is provided by an optimization algorithm. Control variables Δ and Δ are added to reference values and which are resident in a map on , , the control unit. This produces the control variables for the setpoint injected fuel quantity and angle position of pilot injection which are communicated to the control unit. To develop a suitable optimization algorithm, the influence of the pilot injection on the diesel noise rating as well as on the exhaust emission levels is initially examined at the 1250 rpm, 25 Nm and 1750 rpm, 100 Nm example operating points. A heuristic optimization algorithm can then be developed to reduce the irritation from combustion noises.

5.2.1 Influence of Pilot Injection Parameters on Diesel Noise Rating To develop an optimization algorithm to maximize the diesel noise rating, it is necessary to ascertain the relationship between diesel noise rating and pilot injection parameters. To do this, the diesel engine is run at the 1250 rpm, 25 Nm and 1750 rpm, 100 Nm operating points while varying the injected fuel quantity as well as angular position of injection. In this test, indicated mean effective pressure and center of heat release, which are produced from the calibrated control unit data, are kept constant by the control loops from Section 5.1. The measurements provided the basis for generating a contour diagram and linear interpolation between individual measurement points. The results are shown in Figure 13 and Figure 14 for the 1250 rpm engine speed. In both tests, it was possible to extend the diesel noise rating by up to 2 rating grades by varying the pilot injection parameters. In particular, the maximum diesel noise rating levels were attained by increasing the pilot injection quantity. Although increasing the injected fuel quantity improves the diesel noise rating in general, it does not do so in every case. If the pilot injection quantity exceeds approx. 1.4 mg/stroke at the pilot injection angle of = −13 ° at the 1250 rpm, 25 Nm operating point, the diesel noise rating decreases. Although the diesel noise rating tends to improve on increasing the pilot injection quantity, local maximum diesel noise rating values are produced in relation to the pilot injection angle. Depending on the pilot injection quantity, the angular position of pilot injection has a positive influence on the diesel noise rating on varying the position from advanced to retarded. This is shown in particular at the 1750 rpm, 100 Nm operating point for a

24

Diesel engine control based on structure-borne noise – optimization and adaptation …

pilot injection quantity of 1.2 mg/stroke. Moving the angular position from -32 °CA to -18 °CA makes it possible to increase the diesel noise rating by approx. 2 rating grades, in relation to the minimum diesel noise rating level. It is furthermore interesting to note that the highest diesel noise ratings only occur in, what is generally speaking, a narrow angular position range. At approx. -22 to 18 °CA, the maximum levels are attained at the 1250 rpm, 25 Nm operating point. In contrast to this, the range at the 1750 rpm, 100 Nm operating point shifts to approx. 20 to -18 °CA. From examining the underlying principles, it can be concluded that the pilot injection parameters have a significant influence on noise emissions or irritation caused by combustion noises. Generally speaking, the pilot injection quantity provides a key to significantly minimizing combustion noise irritation or improving the diesel noise rating. This, however, makes it necessary to take account of the injection angle. At an adverse injection angle, too much preinjected fuel can reduce the diesel noise rating. The parameter combination of angular position and pilot injection quantity at which the maximum diesel noise rating is attained is governed by the engine speed and load range. These relationships, however, cannot be generalized.

Figure 13: Diesel noise rating (DN) at operating point , on varying pilot injection quantity ( ) and ) angular position (

Figure 14: Diesel noise rating (DN) at operating point , on varying pilot injection quantity ( ) and ) angular position (

5.3 Optimization Algorithm The aim is to optimize the diesel noise rating irrespectively of the operating point selected. Studies carried out so far have shown that combining angular position and pilot injection quantity can optimize the diesel noise rating across the board. Depending on the engine’s operating point, however, insufficient information is available for analyzing the value pair that provides the maximum diesel noise rating. On top of this, information on injector wear or aging effects would also be needed. This makes it nec-

25

Diesel engine control based on structure-borne noise – optimization and adaptation …

essary to develop an algorithm that defines optimum pilot injection values independently of exact information on operating point or injection behavior. The optimization idea is based on using a search algorithm that has the task of detecting the maximum diesel noise rating by gradually varying the injection parameters. This is the reason for selecting the method of steepest descent which has a relatively simple algorithm, making it easy to implement in a control architecture in synchrony with the working cycle. This heuristic algorithm can also be used for nonlinear optimization challenges. The gradient descent method increases the current injection parameter by a fixed increment. If the diesel noise rating increases, the injection parameter continues to increase until the gradient of the diesel noise rating between old and new injection parameter is zero. When the gradient turns negative, the last injection parameter set is adopted at the point at which the gradient produced a positive value. Figure 15 shows the characteristic curve for iteratively optimizing the diesel noise rating. Here, the algorithm was run cyclically once every second as a way of illustrating the algorithm’s functionality from the data measured. This also makes sure that the system has reached a steady state before recording the signals or controlling combustion. The potential for optimizing cycle time was not exploited to the full. Following initialization, the algorithm shifts the pilot injection angle from retarded to advanced. This increases the diesel noise rating from approx. 5.2 to 5.8. The diesel noise rating falls again from a pilot injection angle of -19 °CA. The pilot injection fuel quantity is then increased, resulting in a nearly proportionate rise in the diesel noise rating. From 1.5 mg/stroke, the diesel noise rating remains at approx. 6.9. Any maximization of the diesel noise rating can no longer be identified by increasing the pilot injection, and the maximum permissible total injected fuel quantity is also reached. This is the point at which the algorithm is stopped.

26

Diesel engine control based on structure-borne noise – optimization and adaptation …

Figure 15: Development of the diesel noise rating and pilot injection parameter over time on running the optimization algorithm

A comparison of control unit data calibrated at the 1250 rpm, 25 Nm operating point and the optimized pilot injection parameters is shown in Table 2. The emission levels are also indicated. This shows that although Δ and Δ increase by up to 10%, the other emission levels fall significantly. It must be noted that the relative values refer to the emissions levels produced when the engine is only operated with the control unit.

27

Diesel engine control based on structure-borne noise – optimization and adaptation …

Table 2: Comparison of calibrated control unit data and optimized injection parameters Calibration values ECU [rpm]

1250

Optimized diesel noise rating 1250

[Nm]

25

25

Δ

[%]

Δ

[%]

DN

6,4

6,9

]

-17

-19

+ [mg/stroke]

7.3

7.3

[°

Calibrati on values ECU Δ

[%]

0

Optimized diesel noise rating -41.1

0

-11.5

Δ Δ Δ

[%] .

[%] [%]

0

9.1

0

3.3

0

-36.2

0

-6.3

6 Summary and Outlook For cylinder-selective, noise-controlled engine management, a system has been developed to control indicated mean effective pressure as well as the center of heat release. For this purpose, regression models are successfully used for estimating the controlled variables on the basis of structure-borne noise signals or their characteristics and control unit data. Step responses from the control loops verified that the models provide a level of quality sufficient to reflect the actual physical values, making them suitable for control purposes. This provides the basis for adapting combustion by varying the pilot injection parameters at predefined operating points. The regression models were then additionally used for computing the diesel noise rating from the structure-borne noise signals. The aim was to vary and, ultimately, optimize the diesel noise rating using the pilot injection parameters. By examining the underlying principles, it was established that the pilot injection fuel quantity in particular plays a sensitive part in affecting the diesel noise rating. Furthermore, the maximum diesel noise ratings fall within a relatively narrow pilot injection angle range that is governed by the engine’s operating point. It was not possible to examine any generalization of the prevailing relationships on account of the limited data quantity. To reach the objective of optimizing the diesel noise rating, a heuristic algorithm was used which is based on a gradient method. The algorithm initially varies the pilot injection angular position until the point at which the maximum diesel noise rating is reached. The pilot injection fuel quantity is then increased. At the 1250 rpm, 25 Nm operating point examined, it was possible to achieve an improvement by half a diesel noise-rating grade in comparison to the diesel noise rating produced with the calibrat-

28

Diesel engine control based on structure-borne noise – optimization and adaptation …

ed control unit values. The nitrogen oxide and hydrocarbon emission levels and specific fuel consumption were also reduced at this optimized working point. As a result of the limited test data available, it was not possible to ascertain any generalization as to whether the diesel noise rating can be maximized independently of the operating point. However, the emissions analyzed at the 1750 rpm, 100 Nm operating point do reveal comparable tendencies. The control concepts implemented show a fundamental potential for reducing acoustic irritation from combustion noises. Further studies will need to reveal whether the model quality of virtual cylinder pressure sensors can be optimized by integrating further measured variables, such as engine speed, which is also used in various research projects for directly estimating cylinder pressure. Engine speed signals could also be used for obtaining information on modulation in airborne noise and matching it directly to the specific cylinder. Further analyzing the use of pilot injection parameterization to minimize combustion noises also provides research potential. The study results have shown that the maximum diesel noise rating lies in a local pilot injection parameter range. Regardless of engine speed and load ranges, any potential generalization could be used for integrating predictive control into the engine management system. This would lead to a clear reduction of the adaptation time needed to optimize the diesel noise rating. Pilot injection could also be corrected dynamically in the dynamic engine speed and load range. Injection information fundamentally provides a potential that could be used in particular for diagnosing the injection system. It would be conceivable to use it for detecting production tolerances or aging effects. This permits selective compensation and adaptation of system control data. This would involve conducting extensive analyses with example injectors.

7 References [1] M. Decker, S. Lucas, K. Hintz and J. Nobis, "Geräuschgeregelter Dieselmotor I & II," FVV Final Report, Final Report, 2013. [2] M. Decker, Körperschallbasiertes Motormanagement for PKW-Dieselmotoren, Aachen : Shaker, 2014. [3] J.-H. Carstens, C. Gühmann, S. Schneider, H. Rottengruber, J. Nobis and E. Neumann, "Körperschallbasierte Dieselmotorenregelung – Optimierung und Adaption der Parameter – Final Report", FVV, 2017.

29

Diesel engine control based on structure-borne noise – optimization and adaptation …

[4] K. Finger, Untersuchungen zur Kraftanregeung durch die Verbrennung beim direkteinspritzenden Common-Rail Dieselmotor unter Berücksichtigung des körperschallübertragungsverhaltens, Darmstadt: Darmstadt University of Technology, 2001. [5] D. Powel. and J. E. Manning, "Engine Monitoring Using Vibration Signals", in International Off-Highway and Powerplant Congress and Exposition, USA, 1986. [6] V. Göhringer, Beitrag zur experimentellen Bestimmung des Strukturübertragungsmaßes von Dieselmotoren, Renningen: Published by Expert Verlag, 2008. [7] D. Föller, Untersuchung der Aanregung von Körperschall in Maschinen und der Möglichkeiten für eine primäre Lärmbekämpfung, Darmstadt: Faculty of Mechanical Engineering at Darmstadt University of Technology, 1972. [8] N. R. a. H. S. Draper, Applied Regression Analysis, Hoboken, NJ: WileyInterscience, 1998. [9] J. Hoppersmanns, "Objektivierung subjektiver Beurteilungen", Project No. 841, RWTH Aachen University of Applied Sciences – FVV, 2006. [10] N. W. Alt, J. Nehl, S. Heuer and M. W. Schlitzer, "Prediction of Combustion Process Induced Vehicle Interior Noise", in SAE 2003 Noise and Vibration Conference and Exhibition, Warrendale, PA, 2003. [11] V. Göhringer, U. Philipp and M. Bargende, "Verbrennungsschallanalyse bei Pkw-Dieselmotoren", in Dieselmotorentechnik 2000, published by ExpertVerlag, 2000. [12] F. Sarıkoç, Untersuchungen zur Reduzierung der Stickoxidemissionen bei modernen Brennverfahren für Motoren mit Benzin-Dirketeinspritzung, Berlin: Published by Logos Verlag Berlin GmbH, 2009.

30

Combustion mechanical noise breakdown – turbocharger noise identification on a V8 engine Karl Janssens, Fabio Bianciardi Siemens Industry Software NV Konstantinos Gryllias KU Leuven Simone Delvecchio Siemens Industry Software NV Claudio Manna Ferrari S.p.a.

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 W. Siebenpfeiffer (Hrsg.), Automotive Acoustics Conference 2017, Proceedings, https://doi.org/10.1007/978-3-658-20251-4_10

1

Combustion mechanical noise breakdown – turbocharger noise identification …

Introduction The NVH character of an engine is composed of a mixture of various sources. Some of these have been categorized as combustion related noise and others as mechanical noise. The introduction of modern engine technologies, such as turbocharger, direct injection, etc., often results in additional engine vibrations and radiated noise. Therefore being able to extract specific engine noises is of prime interest in the engine development phase and as well as for engine diagnostic purposes. Several methodologies for combustion mechanical breakdown have been developed in the past, namely the multiple regression analysis, the classical Wiener filter and cyclostationary Wiener filter. In the present paper the techniques are applied to microphone recordings measured at one meter distance from a Ferrari V8 engine running on a test bench. Strengths and weaknesses of the techniques are evaluated. Moreover the results obtained by the first two aforementioned techniques are compared. Finally a source separation method is combined with microphone array measurements, allowing a localization and quantification of the combustion and mechanical sources of noise on the V8 engine. This is illustrated with an application example for a specific turbocharger noise.

Combustion/mechanical breakdown setup Operational tests on the Ferrari V8 (CaliforniaT) engine have been performed in a Ferrari engine test facility. The engine was mounted on the test-bench. Transmission bell and the transmission shaft were included in the setup as in the in-vehicle configuration. Figure 1 presents the rear view of the V8 engine in the engine test facility. Five microphones were positioned around the engine at 1m of distance from each of the engine surfaces (front, right, left, rear, top). One microphone was located below the engine as well, at a distance of 0.2 m from the engine bottom surface due to spatial constraint. Each cylinder was instrumented with a piezoresistive pressure sensor measuring the cylinder pressure evolution. Unfortunately during the tests the sensor in the cylinder number 2 got damaged and the signal from this sensor has not been used as reference in this analysis. An incremental encoder with 60ppr was instrumented on the crankshaft measuring the engine speed. Additionally a torque sensor was instrumented on the transmission shaft. The engine was run at different steady speed and torque levels. Additionally different engine run-up tests sweeping from 1000 rpm to 7500 rpm were performed for different constant torque levels, as well as torque sweeps from 0 Nm to 600 Nm were performed at different constant speeds. The total duration of the operational tests was 90 seconds. All the signals were synchronously acquired at 102.4 kHz of sampling frequency with the LMS Scadas.Lab acquisition system.

22

Combustion mechanical noise breakdown – turbocharger noise identification …

Figure 1 – Engine test setup. Rear view of the engine is shown.

Figure 2 shows the averaged Overall Sound Pressure Level (OASPL) measured around the engine at three torque levels (100Nm, 300Nm, 600Nm). The averaged OASPL has been obtained by averaging 4 microphones around the engine: front, rear, right and left. The top and bottom microphones have not been included in the analysis as strong flow noise was present in the former microphone during the tests while the bottom microphone was too close to the engine bottom surface. As can be noticed from figure 2, the measured OASPL shows a global behaviour of the radiated noise in function of speed and torque, but doesn’t provide information of possible engine improvement. A breakdown analysis would help the engineers to steer the improvements toward a much faster and engineered solution.

Pa dB(A)

10 dB

F F F

1000

2000

3000

original:S 100Nm original:S 300Nm original:S 600Nm

4000 rpm

5000

6000

7500

Figure 2 – Averaged OASPL measured around the engine for different torque levels (red line = 100Nm, green line = 300Nm, blue line = 600Nm).

3

Combustion mechanical noise breakdown – turbocharger noise identification …

For confidentiality reasons the real OASPL values have been omitted from the figures shown in this paper. An arrow located on the y-axis of each figure is indicating the relative values in decibels (dB). In addition, for sake of clarity, all the plots presenting OASPL values are shown with the same dB scale throughout the entire paper.

Combustion/mechanical breakdown techniques Combustion Mechanical Breakdown commonly refers to techniques attempting to provide an analysis of possible engine noise improvements. When dealing with engine acoustic optimization, NVH target requirements definition, and diagnosis of engine anomalies, specific noise separation techniques can be applied to help engineers to identify and quantify noise radiating sources with the objective of improving the engine acoustic radiation. The noise separation techniques, presented in this paper, attempt to decompose the overall sound pressure level of the radiated engine noise in two components, namely the noise related to the combustion, and the remaining noise (mechanical and accessories noise). These techniques can be used as pre-processing step to further enhance the results of possible further analysis which can be performed on the decomposed noise.

Multiple regression analysis One of the methodologies, used to decompose the overall radiated noise, has been proposed by Hirano, Kondo and others, which developed a multiple regression technique for combustion/mechanical breakdown [1]. The radiated engine sound power is represented by the equation (1). The engine noise is assumed to consist of three components: the load independent mechanical noise (SPm), the combustion noise proportional to the cylinder pressure (H∙CP) and the load dependent mechanical noise (G∙L). CP and L are the explanatory variables representing respectively the cylinder pressure power and the squared engine torque, as presented in equation (2). The SPm, H, and G are the partial regression coefficients, while the vector e represents the measurement noise. The index n represents different torque operating conditions. 𝑆𝑆𝑆𝑆 = 𝑆𝑆𝑆𝑆𝑚𝑚 + 𝐻𝐻 𝐻 𝐶𝐶𝐶𝐶 + 𝐺𝐺 𝐺 𝐺𝐺

𝑆𝑆𝑆𝑆1 𝑆𝑆𝑆𝑆2 [ ]= ⋮ 𝑆𝑆𝑆𝑆𝑛𝑛

1 𝐶𝐶𝐶𝐶1 1 𝐶𝐶𝐶𝐶2 [ ⋮ ⋮ 1 𝐶𝐶𝐶𝐶𝑛𝑛

𝑒𝑒1 𝐿𝐿1 𝑆𝑆𝑆𝑆𝑚𝑚 𝑒𝑒2 𝐿𝐿2 ] ∙ [ 𝐻𝐻 ] + [ ⋮ ] ⋮ 𝐺𝐺 𝑒𝑒𝑛𝑛 𝐿𝐿𝑛𝑛

Equation (2) can be rewritten in a more compact form as in equation (3):

4

4

(1)

(2)

Combustion mechanical noise breakdown – turbocharger noise identification …

(3)

𝑌𝑌 𝑌 𝑌𝑌𝑌 𝑌 𝑌𝑌 𝑌 𝑌𝑌

By means of a multiple regression analysis, the three noise components of the vector A can be estimated, as shown in the equation (4). The analysis is performed in third octave bands. 𝐴𝐴𝐴 𝐴𝐴𝐴 𝑇𝑇 𝑋𝑋𝑋−1 𝑋𝑋 𝑇𝑇 𝑌𝑌

(4)

The results of the multiple regression analysis are presented in figure 3. The noise is decomposed in combustion and mechanical noise. The mechanical noise is the sum of the load independent and load dependent noise. The colorplots in the figures represent the distribution of the overall sound pressure level in dB in function of engine speed (xaxis) and torque (y-axis). OASPL A-weighted - measured noise

600

120 110

20 dB

Torque [Nm]

500

100

400

90

300

80

200

70

100 1000

OASPL A-weighted - combustion noise

Torque [Nm]

4000 5000 speed [rpm]

600

120

20 dB

500 400

110

500

100

400

90

300

80

200

70

100 1000

3000

60 2000

3000

4000 5000 speed [rpm]

6000

7000

50

Torque [Nm]

600

60 2000

6000

7000

50

OASPL A-weighted - mechanical niose

120 110

20 dB

100 90

300

80

200

70

100 1000

60 2000

3000

4000 5000 speed [rpm]

6000

7000

50

Figure 3 – Combustion mechanical breakdown result using the multiple regression technique. Top figure: measured noise. Bottom figure left: combustion noise. Bottom figure right: mechanical noise

The method has proven to be effective. However, the main limitation of the method resides in the fact that the results consist of third octave bands not permitting any further post-processing study such as listening, subjective analysis, time-frequency analysis or computation of sound quality metrics.

5

5

Combustion mechanical noise breakdown – turbocharger noise identification …

Classical Wiener filter An alternative approach for decomposing the engine radiated noise is based on the coherence method proposed by Wiener [2]. The aim of the Wiener filter is to extract, from a signal measured in operating conditions y(t), the contribution x(t) associated to a specific source represented by a reference signal r(t), which is strongly coherent to x(t) and uncorrelated with the other noise sources. The scope is to estimate the best set of linear filters ℎ(𝑡𝑡) which, once applied to 𝑟𝑟(𝑡𝑡), will give the best estimation 𝑥𝑥𝑒𝑒𝑒𝑒𝑒𝑒 (𝑡𝑡) of 𝑥𝑥 (𝑡𝑡), minimizing the quadratic error as presented in equation (5). ℎ𝑘𝑘𝑘𝑘 (𝑡𝑡) = 𝐴𝐴𝐴𝐴𝐴𝐴𝐴 𝐴𝐴𝐴ℎ𝑘𝑘𝑘𝑘 (𝑡𝑡) 𝐸𝐸{|𝑦𝑦𝑘𝑘 (𝑡𝑡) − 𝑥𝑥𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒 (𝑡𝑡)|2 } , ∀𝑡𝑡

(5)

In the application of this paper, the method attempts to decompose the engine radiated noise in: combustion related noise and residual noise. y(t) represents the acoustic pressure measured by a microphone k, ri(t) are the cylinder pressure signals and bres(t) represents the residual noise non correlated to the combustion. Figure 4 shows a schematic representation of the classical Wiener filter concept. The combustion related noise is defined as the noise that is coherent with the reference in-cylinder pressure signals. While the residual noise is obtained by subtracting the combustion related noise from the measured radiated noise.

Figure 4 – Schematic representation of the classical Wiener filter

Residual noise is expected to contain all types of noise that are not coherent with incylinder pressure. Residual noise is sometimes referred to as mechanical noise; however this terminology might not be the most proper one, since also the turbocharger noise will be categorized as mechanical noise. Turbocharger noise is more correctly defined as the noise generated by aerodynamic phenomena. Therefore the noise not coherent with in-cylinder pressure will be named residual noise in the rest of the paragraph.

66

Combustion mechanical noise breakdown – turbocharger noise identification …

The method requires one or more reference signals related to the combustion process. The method consists of estimating the transmissibility functions between the in-cylinder pressure references and the acoustic responses measured in operating conditions. The estimation is performed in frequency domain per frequency line. Equation (6) presents the transmissibility function computed for a multi inputs – multi outputs system (MIMO). The Sii is the input autopower spectrum matrix, and Ski is the crosspower spectrum vector between the i=1,…, nc inputs (in-cylinder pressure signals/references) and the k outputs (microphones). The total contribution is finally obtained using the equation (7), where ri represents the reference sensors vector and nc is the total numbers of cylinders. 𝐻𝐻𝑘𝑘𝑘𝑘 (𝜔𝜔) = 𝑆𝑆𝑘𝑘𝑘𝑘 (𝜔𝜔) ∙ [𝑆𝑆𝑖𝑖𝑖𝑖 (𝜔𝜔)]−1

𝑥𝑥𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒 (𝜔𝜔) =  ∑𝑛𝑛𝑛𝑛 𝑖𝑖=1 𝐻𝐻𝑘𝑘𝑘𝑘 (𝜔𝜔 ) ∙ 𝑟𝑟𝑖𝑖 (𝜔𝜔) 

(6) (7)

As can be noticed from equation (7), the total combustion related noise is the sum of each in-cylinder pressure signal contribution. The advantage of using the Wiener filter method, compared to the multiple regression analysis, resides in the fact that this approach can be performed also in time domain. In order to obtain the required time domain results, FIR filters are constructed from the estimated frequency domain transmissibility model and applied to the measured incylinder signals by means of the fast FFT convolution technique. A proper filter preprocessing is required, avoiding to generate artifacts in the synthesized sound.

Application to stationary operating conditions One of the assumptions of the Wiener filter method consists of considering the reference signals fully decorrelated between each other. This ensures a full rank autopower spectrum matrix leading to a correct transmissibility estimation. This assumption is fundamental when estimating the partial contribution of each in-cylinder pressure to the measured radiated noise [3]. When applying the classical Wiener filter at constant speed the cylinders pressure are highly correlated. Figure 5 (left figure) presents the principal components analysis of the autopower spectral matrix for a constant speed test at 3000rpm. At the combustion excitation frequency lines only one singular value is dominant, implying that all the reference in-cylinder pressure signals are correlated. The effect derived from this aspect can be mitigated by considering the total combustion related noise contribution xest, instead of the partial contribution of each in-cylinder pressure reference. The total contribution would be always correct even if the partial contribution of each in-cylinder pressure might not be realistic due to the rank deficient autopower matrix [3]. This is true as

7

Combustion mechanical noise breakdown – turbocharger noise identification …

soon as enough references are considered for each uncorrelated sources and no sources are neglected. Principal component PRCM:0001:S/PRCM:0001:S Principal component PRCM:0002:S/PRCM:0002:S Principal component PRCM:0003:S/PRCM:0003:S Principal component PRCM:0004:S/PRCM:0004:S Principal component PRCM:0005:S/PRCM:0005:S Principal component PRCM:0006:S/PRCM:0006:S Principal component PRCM:0007:S/PRCM:0007:S

/2

50.00

-50.00 0.00

Hz

2000.00

F F F

Principal component PRCM:0001:S/PRCM:0001:S Principal component PRCM:0002:S/PRCM:0002:S Principal component PRCM:0003:S/PRCM:0003:S

F F F F

Principal component PRCM:0004:S/PRCM:0004:S Principal component PRCM:0005:S/PRCM:0005:S Principal component PRCM:0006:S/PRCM:0006:S Principal component PRCM:0007:S/PRCM:0007:S

dB

F F F F F F F

dB

/2

50.00

-50.00 0.00

Hz

2000.00

Figure 5 – Principal component analysis for a constant speed test (left figure), and for a speed sweep test (right figure)

Another condition for a correct estimation of the transmissibility functions consists of assuming the combustion related noise and residual noise fully decorrelated. The noise not coherent with the reference in-cylinder pressure signals is eliminated in the crossspectral term by averaging subsequent data blocks. However the classical Wiener filter applied to an engine running in stationary condition might suffer from this aspect. As reported by Lee et al. [4], some mechanical noise sources can be partly correlated with in-cylinder pressure, therefore some mechanical noises might be incorrectly separated as combustion related contribution.

Application to non-stationary operating conditions In order to overcome the risk of leaking of mechanical noise in the combustion related noise contribution, Lee et al. [4] proposed a randomization of the start of injection timing for each combustion event and cycle-by-cycle. The randomization process has been considered, in a stationary scenario, as a way to decorrelate, on one side the reference in-cylinder pressures, and on the other to better decorrelate combustion related and residual noise. The randomization process has been considered very effective in a stationary scenario, leading to similar results as for a run-up test. Essentially, by varying the engine speed, the engine control parameters are continuously adjusted and updated in order to achieve specific targets for combustion process efficiency. Therefore, in order to overcome the limitations presented by the stationary conditions, the classical Wiener filter has been applied to non-stationary operating conditions, more specifically to runup tests [5]. Estimating a transmissibility function from the entire run-up requires as a main condition the invertibility of the autopower spectral matrix. Averaging over the entire run-up

88

Combustion mechanical noise breakdown – turbocharger noise identification …

helps to decorrelate the in-cylinder pressure signals making the autopower matrix full of rank. The principal component analysis, obtained from a run-up test, shows that all the principal components have similar importance in each frequency line, as presented in figure 5 (right figure). For this application case all the available in-cylinder pressure signals have been considered as references for the transmissibility estimation. The averaging is performed over the entire run-up with a fixed frequency resolution of 4 Hz. The assumption of time invariant filter applied to non-stationary conditions may suffer of small inaccuracies. The best fit for the results is globally obtained at the high speed range, while at very low speed this approach may lead to small errors due to the estimated transmissibility function. Despite this aspect, the filtering technique applied to non-stationary operating conditions has been considered an effective approach to separate combustion related and residual noise. This solution is also in-line with the most common testing maneuver performed at OEMs consisting of speed run-up tests for different constant throttle positions or torque levels. Figure 6 shows the application of the Wiener filtering technique to a measured microphone during a run-up test at full load. 7500

10 dB

7000 6500 6000 5500

4500

dB(A) Pa

rpm

5000

4000 3500 3000 2500 2000 1500 1000 500 1000 1500 2000 2500 3000 3500 4000 Hz Frequency

7500

6500

6000

6000

5500

5500

5000

5000

4500

4500

4000

rpm

6500

dB(A) Pa

rpm

7500 7000

10 dB

7000

5000

10 dB

dB(A) Pa

0

4000 3500

3500

3000

3000

2500

2500

2000

2000

1500 1000

1500 1000 0.00

Hz Frequency

5000.00

0.00

Hz Frequency

5000.00

Figure 6 – Example of the application of the Wiener filtering technique to a measured microphone for a speed run-up test at the maximum torque level. Top figure: measured noise. Bottom figure left: combustion related noise. Bottom figure right: residual noise

9 9

Combustion mechanical noise breakdown – turbocharger noise identification …

Figure 7 shows the averaged overall sound pressure level, obtained by averaging the microphones around the engine, decomposed in the combustion related noise and the residual noise. The colurplots show the evolution of the overall sound pressure level in function of speed (x axis) and torque (y axis). Those maps can be used to identify possible noises occurring at specific speed and torque levels. OASPL A-weighted - measured noise

600

120 110

500

20 dB

Torque [Nm]

100 400

90

300

80 70

200

600

60 2000

OASPL A-weighted - combustion related noise

3000

110 100

400

90 80

300

6000

20 dB

7000

50

OASPL A-weighted - residual niose

600

120

500

Torque [Nm]

4000 5000 speed [rpm]

120

20 dB

110

500

100

Torque [Nm]

100

400

90

300

80

70 200 100

60 2000

3000

4000 5000 speed [rpm]

6000

7000

50

70 200 100

60 2000

3000

4000 5000 speed [rpm]

6000

7000

50

Figure 7 – Combustion mechanical breakdown results using the classical Wiener filter technique. Top figure: measured noise. Bottom figure left: combustion related noise. Bottom figure right: residual noise.

A more detailed insight into the decomposed results can be obtained by investigating the combustion and the residual noise evolution for different torque levels over the entire speed sweep, as presented in figure 8. The comparison of the breakdown analysis for three load cases (100 Nm, 300 Nm, 600 Nm) shows that the load is mainly affecting the residual noise in the speed range between 2000 rpm and 3000 rpm. While, in the same speed range, the combustion related noise assumes comparable values especially between 300 Nm and 600 Nm.

10 10

Combustion mechanical noise breakdown – turbocharger noise identification … 10 dB

Pa dB(A)

Pa dB(A)

10 dB

F F F 1000

2000

3000

F F F

combustion_related:S 100Nm combustion_related:S 300Nm combustion_related:S 600Nm 4000 rpm

5000

6000

7500

1000

2000

3000

4000 rpm

residual:S 100Nm residual:S 300Nm residual:S 600Nm

5000

6000

7500

Figure 8 - OASPL comparison of the radiated combustion related noise (left figure) and residual noise (right figure) for three different torque levels (red line = 100Nm, green line = 300Nm, blue line = 600Nm).

The possibility of using the Wiener filters in time domain represents a great advantage compared to the multi-regressive method for which only third-octave bands results are obtained. Combustion related and residual time-domain results can be used for multiple post-processing studies such as: listening, subjective analysis, time-frequency analysis, engine diagnosis, computation of sound metrics and sound source localization. The latter will be presented in the following section in this paper as an effective postprocessing application case to enhance source localization of a specific residual engine noise occurring during run-ups at full load.

Cyclic (synchronous) Wiener filter The dynamic of internal combustion engines presents periodic statistical patterns, in other words it is cyclostationary within the engine cycle. For this reason alternative and more advanced approaches based on the assumption of cyclo-stationarity have been proposed by Pruvoust et al. and Leclere et al. [6,7,8]. In order to process the measured signals by means of cyclic statistics that are locked onto the engine kinematics, it is convenient to use a constant angular sampling instead of constant time sampling. The synchronous (cyclostationary) Wiener filter exploits the property of cyclostationarity of the measured signals over the combustion cycle.

11 11

Combustion mechanical noise breakdown – turbocharger noise identification …

In this work all signals resampled using a tachometer which measures the speed of the crankshaft. The angular resampling practically enforces the signals to be cyclostationary. The signals corresponding to each cycle are considered as a realization of the same stochastic process [10]. As a result the statistical features can be assessed in a general and efficient manner using the “synchronous averaging” method [11]. The transmissibility functions are estimated similarly to the classical Wiener filter but the averaging is performed cycle by cycle based on a synchronous average in the angular domain. As already noted the raw pressure and vibroacoustic signals are strongly correlated leading to estimation problems. For this reason, in order to overcome the problem and lower the bias on the estimated transmissibility functions, the transmissibility functions are estimated from the residual/random part of the input and output signals obtained by removing the deterministic components using synchronous averaging [7,8,10,12]. Figure 9 shows an example of the raw and the residual signals of both the input/pressure and the output/sound.

Figure 9 – Example of a measured signal (in black), the deterministic signal after synchronous averaging (in red) and the residual signal after synchronous averaging removal (in blue) for a combustion cycle. On the left figure the application on a cylinder pressure signal and on the right figure the application on a microphone signal are shown.

The Wiener filter in the angle domain can be estimated as in equation (8): ̂𝑘𝑘𝑘𝑘 (𝜃𝜃 ) = 𝑆𝑆̂𝑘𝑘𝑘𝑘 (𝜃𝜃 ) ∙ [𝑆𝑆̂𝑖𝑖𝑖𝑖 (𝜃𝜃 )]−1 𝐻𝐻

̂ 𝑥𝑥𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒 (𝜃𝜃 ) =  ∑𝑛𝑛𝑛𝑛 𝑖𝑖=1 𝐻𝐻𝑘𝑘𝑘𝑘 (𝜃𝜃 ) ∙ 𝑟𝑟𝑖𝑖 (𝜃𝜃 ) 

(8) (9)

where 𝑆𝑆̂𝑘𝑘𝑘𝑘 (𝜃𝜃 ), 𝑆𝑆̂𝑖𝑖𝑖𝑖 (𝜃𝜃 ) are respectively the cross spectral and the autospectral power densities. A MIMO Wiener filter version is used in order to estimate the contribution of each cylinder/source/input to each measured noise/output. As pre-processing step, the

12

Combustion mechanical noise breakdown – turbocharger noise identification …

signals have been resampled in the angular domain and the residual part, after synchronous averaging, has been considered for the estimation of the filter. The estimated combustion related noise is finally obtained using equation (9). The analysis performed in this paper revealed how estimating transmissibility from the residual signals after synchronous averaging can be very challenging when applied to microphones recording signals at 1m of distance from the engine for a 8 cylinders engine. The coherence between the residual microphone noise and the residual cylinder pressure is extremely low as presented in figure 10. Therefore estimating a transmissibility function in a MIMO system is not recommended, leading to a risk of an underestimated transmissibility function.

Figure 10 – Partial coherence between one cylinder pressure residual signal and the left microphone residual signal after synchronous averaging removal

In a number of cases presented in the literature [8,9,10], the authors tried to estimate the contribution of each engine cylinder to the global engine noise. Applied mainly in 4 cylinder engines, the MISO problem has been transformed in a SISO gated cylinder analysis by using a Tukey window around the Top Dead Center (TDC). However, in this application case this approach has not been considered applicable due to the high overlap among cylinders. Nonetheless the cyclo-stationary technique will be further explored to potentially identify other noise phenomena occurring within the engine cycle.

13

Combustion mechanical noise breakdown – turbocharger noise identification …

Comparison of the combustion mechanical breakdown approaches Due to the different combustion noise definition, the results obtained by multiple regression analysis and the classical Wiener filter are presenting differences especially in the low frequencies. In this section a comparison of the combustion mechanical separation OASPL results computed for two different frequency ranges are presented. Figure 11 shows the combustion mechanical separation OASPL results obtained by considering the frequency range from 100Hz to 20kHz. Results are shown for different speeds at 300Nm of torque. As in the previous section, the overall sound pressure levels are computed by averaging the selected microphones around the engine. As can be noticed from figure 11, a clear difference in noise contributions is present between the two methods investigated in this work.

115 110

CMB at 300Nm

CMB at 300Nm

120

120

measured combustion related residual

10 dB

110

100 105

80

90

95 90

70 85

80

100 95 90 measured combustion related residual

60 80

75

75 70 1000

measured combustion related residual

10 dB

110

70 85

measured combustion mechanical

60 80 50 1000

CMB at 300Nm

100 105

100 dB-A

dB-A

dB-A

90

CMB at 300Nm

120 115

110

dB-A

120

2000

3000

2000

4000 5000 3000speed [rpm] 4000

6000

5000 speed [rpm]

7000

6000

50 1000

8000

7000

8000

70 1000

2000

3000

2000

4000

5000

6000

3000speed [rpm] 4000 5000 speed [rpm]

7000

6000

8000

7000

Figure 11 – OASPL comparison after combustion mechanical breakdown. On the left figure the results obtained from multiple regression analysis are shown. On the right figure the results obtained from classical Wiener filter applied to non-stationary operating condition are shown. The OASPL are computed by considering the frequency range 100Hz-20kHz.

When computing the OASPL by considering a higher frequency range, the combustion mechanical separation results obtained by multiple regression analysis and the classical Wiener filter are presenting a more similar behavior. Figure 12 shows the combustion mechanical separation OASPL results computed in the frequency range spanning from 1600Hz to 20kHz. Results are shown for different speeds at 300Nm of torque. By considering this frequency range, both combustion mechanical breakdown approaches are in agreement identifying the mechanical noise as the largest contributors to the noise. This analysis shows that the low frequencies represent a challenge for the noise separation approaches.

14 14

8000

Combustion mechanical noise breakdown – turbocharger noise identification …

115 110

CMB at 300Nm

CMB at 300Nm

120

120

measured combustion related residual

10 dB

110

100 105

80

90

95 90

95

80 70

measured combustion mechanical

60 80

75 70 1000

100

90

70 85

50 1000

measured combustion related residual

10 dB

110

100 105

100 dB-A

dB-A

dB-A

90

CMB at 300Nm

CMB at 300Nm

120 115

110

dB-A

120

2000

3000

2000

4000

5000

6000

3000speed [rpm] 4000 5000 speed [rpm]

7000

6000

85

measured combustion related residual

60 80

75

50 1000

8000

7000

8000

70 1000

2000

3000

2000

4000 5000 speed [rpm]

3000

6000

4000 5000 speed [rpm]

7000

6000

8000

7000

Figure 12 – OASPL comparison after combustion mechanical breakdown. On the left figure the results obtained from multiple regression analysis are shown. On the right figure the results obtained from classical Wiener filter applied to non-stationary operating condition are shown. The OASPL values are computed by considering the frequency range 1600Hz-20kHz.

Sound source localization: identification of a turbocharger noise Acoustic imaging is a technique to localize noise sources on the surface of an object emitting noise. The source localization can be used to spatially localize areas where the noise is emitted from in order to take specific counter measures. Sound source localization techniques combined with noise separation methods allow to enhance the sources spatial localization depending on their different contribution. In this application case the classical Wiener filter has been applied as a pre-processing technique to the microphones recording signals measured from a microphones array decomposing the total noise in combustion related residual noise. The array has been located in the front of the engine and the microphones signals have been synchronously acquired together with the in-cylinder pressure data. The analysis of the results obtained from the source localization of the residual noise contribution, in the speed range 2000rpm – 3000rpm for 600Nm of torque, revealed that the noise sources are localized on the two sides of the engine. These are the engine areas where the two turbochargers are located. As expected the turbocharger flow noise has been categorized as residual noise by applying the classical Wiener filter as preprocessing technique to the array microphones recordings. This can be also highlighted in the time-frequency maps previously presented in figure 6. Figure 13 shows the sound source localization results of the residual noise contribution in the speed range 2000rpm-3000rpm.

15 15

8000

Combustion mechanical noise breakdown – turbocharger noise identification …

8 dB

Figure 13 – On the left an image of the engine under test is shown; and on the right the sound source localization result of the residual noise for the speed range 2000-3000rpm is shown. Residual noise is obtained by applying the classical Wiener filter to the microphones array recorded signals.

Conclusions The paper presents the application of different combustion mechanical separation techniques on a Ferrari V8 engine. The investigated techniques revealed their strengths and weaknesses. The multiple regression analysis is a well-known technique for combustion mechanical breakdown; however the main limitation resides in the fact that the method results consist of third octave bands not permitting any further post-processing study. Another technique investigated in the paper is the coherence method, also known as Wiener filter. The method attempts to separate the combustion related noise, coherent with the reference in-cylinder pressure signals, from the global radiated noise. The residual noise is calculated by subtracting the combustion related noise from the global radiated noise. The classical Wiener filter has been applied to run-up tests, with the advantage of better decorrelating the combustion related and the residual noise sources. The possibility of using the classical Wiener filter in time domain represents a great advantage compared to the multiple regression analysis for which only third-octave bands results are obtained. Combustion related and residual time domain results can be used for several post-processing studies such as: listening, subjective analysis, timefrequency analysis, engine diagnosis, computation of sound metrics and sound source localization. The classical Wiener filter combined with the source localization revealed to be a powerful approach to detect and quantify a specific engine noise source. The technique used to separate the combustion related noise from the residual noise revealed to be efficient

16

16

Combustion mechanical noise breakdown – turbocharger noise identification …

in separating engine orders from broadband noise, identifying the flow noise effect of the turbocharger occurring at a specific engine speed range for high engine load. The classical Wiener filter applied to the recorded signals of a microphones array was able to identify and categorize the specific turbocharger flow noise as residual noise, thus not correlated with the in-cylinder pressure signals.

References 1.

I. Hirano, M. Kondo, Y. Uraki, Y. Asahara, “Using multiple regression analysis to estimate the contributions of engine-radiated noise components”, JSAE 9931233 (1999) 2. N. Wiener, “Extrapolation, interpolation and smoothing of stationary time series”, (The technology press of the Massachusetts Institute of Technology, New York: John Wiley & Sons, Inc., 1950). 3. F. Bianciardi, L. Britte, K. Janssens, “Critical assessment of OPA: effect of coherent path inputs and SVD truncation”, Proceedings ICSV20, July 2013 (2013). 4. M. Lee, J. S. Bolton, S. Suh, “Estimation of the combustion-related noise transfer matrix of a multi-cylinder diesel engine”. Measurement Science and Technology 0233/20/1/15106 (2009) 5. In-Soo Suh, “Combustion on radiated noise and mount vibration from a V8 gasoline engine”, SAE 2003-01-1730. 6. L. Pruvoust, Q. Leclere, E. Parizet, “Improvement of the spectrofilter – separation of coherent sources overlapping in time and frequency domains”, 19th international congress on acoustics, Madrid. 2-7 September 2007 (2007). 7. L. Pruvost, Q. Lecler, E. Parizet, “Diesel engine combustion and mechanical noise separation using an improved spectrofilter”, Mechanical Systems and Signal Processing, Vol. 23, (2009), pp. 2072-2087. 8. Q. Leclere, C. Sandier, O. Sauvage, “Source separation in diesel ingines using Wiener filtering: physical interpretations”, 8th Symposium Automotive NVH comfort - 22 & 23 October 2014 9. B. Lafon, J. Antoni, M. Sidahmed, L. Polac, “The concept of cyclic sound intensity and its application to acoustical imaging”, Journal of Sound and Vibration, Vol 330, (2011), pp. 2017-2121. 10. M. El. Badaoui, J. Daniere, F. Guillet, C. Serviere, “Separation of combustion noise and piston-slap in diesel engine – Part I : Separation of combustion noise and piston-slap in diesel engine by cyclic Wiener filtering”, Mechanical Systems and Signal Processing, Vol. 19, (2005), pp. 1209-1217. 11. B. Peeters, B. Cornelis, K. Janssens, H. Van der Auweraer, “Removing disturbing harmonics in operational modal analysis”, Proceedings of 2nd International Operational Modal Analysis Conference (IOMAC 2007), 30 April - 2 May 2007. 12. J. Antoni, J. Daniere, F. Guillet, “Effective vibration analysis of IC engines using cyclostationarity, Part I – A methodology for condition monitoring”, journal of Sound and Vibration, Vol. 257(5), (2002), pp. 815-837.

17 17

Simulation of exterior noise propagation for the acoustic load estimation of airborne model Eng. M. Danti Dr. G. Bartolozzi Mr. M. Meneguzzo Vehicle concept & integration / Performance Integration & Validation / Vehicle integration CRF – NVH, Centro Ricerche Fiat

Dr. C. Campagna Vehicle concept & integration / Performance Integration & Validation/ Noise Vibration & Harshness PCC

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 W. Siebenpfeiffer (Hrsg.), Automotive Acoustics Conference 2017, Proceedings, https://doi.org/10.1007/978-3-658-20251-4_11

1

Simulation of exterior noise propagation for the acoustic load estimation of airborne …

Abstract The time to market in the automotive industry is constantly decreasing pushing the carmaker companies to increase the efforts in numerical simulations and to decrease the number of prototypes. In the NVH field, the airborne models are widely used for many purposes: from the estimation of the Transmission loss of different parts to the assessment of interior noise due to very complex exterior loads, like aeroacoustic sources. In most cases, the estimation of the exterior load for SEA model is quite complicated and takes a long time to be completed. The aim of this paper is to present a methodology which is able to forecast the acoustic loads on the exterior surface by exploiting the numerical potentiality of the Fast Multipole Boundary Element Method (FMBEM). In addition, the paper shows a tool that is able to close the gap between the two different methods (FMBEM and SEA), by collecting and processing the data deriving from the FMBEM to be applied to the different panels, which are defined in the SEA method. This is done by a proper averaging – both spatially and in the frequency range – of the pressure for the different panels. The approach is therefore validated by means of a numerical experimental comparison at two different levels: the first one is on the wetted surface, where FMBEM and experimental measurements are compared, while the second one is at the interior cavity level, where the noise level predicted by experimental acoustic loads is compared with the level derived by numerically estimated sources

Introduction The reduction of the time required for the design of quiet cars is crucial in order to meet the customers’ expectations in terms of quality and perceived quality. Among all the characteristics, noise plays an important role as the continuous exposure to high level of sound during a trip by car is usually linked to a decrease of the attention threshold for the driver. The noise can be transferred into the passengers’ compartment by structure borne path or airborne path: the first is predominant at low frequencies while the second one is really important at high frequencies. The airborne noise propagation deals with frequencies ranging from 300-400 Hz up to the maximum audible frequency and therefore is still prohibitive for numerical deterministic simulation. In the late 60’s the statistical energy analysis method has been developed by Lyon [1] and applied in the late 80’s by many carmakers as a powerful tool even if it treats averaged quantities. The level of noise inside the passenger compartment is therefore assessed as an overall third-octave-band average when an acoustic load is applied to the vehicle. The loads can be measured experimentally if the car is available or when the new car model is an evolution of a previous one, but early phase modifications require anyway virtual evaluation. In recent years many new numerical

2

Simulation of exterior noise propagation for the acoustic load estimation of airborne …

methods have been developed. Danti et al. [2] and Miccoli at al. [3] have reached very similar conclusions dealing with different structures (in the first case a portion of a vehicle and in the second case a prototypal engine bay) stating that the new deterministic FE methods, e.g. the automated perfectly matched layer, are suitable for the exterior propagation up to around 3000-4000 Hz. Given the need for a proper numerical approach for the estimation of the exterior acoustic loads, different numerical techniques have been considered. The choice of FMBEM (see reference [4]) is almost straightforward, because up to now the new FE methodologies are really effective only for frequencies below 3000-4000 Hz, while the range addressed by this work should cover at least 8000 Hz third octave band. From this point of view, the work by Blanchet [5] et al. has inspired this work showing a suitable industrial process that has been extended in frequency and joined to a seamless data automation tool to exchange information between FMBEM and SEA models. The paper is organized as follows. The first section covers the choice of the proper numerical methodology, comparing results for an A-segment car, while the application case will then be on a C-segment car, both for the experimental campaign and for the numerical validation, in the following section.

Comparison of current methodologies for the numerical prediction of acoustic exterior field In recent times, many software tools have enhanced their capability in the simulation of acoustic exterior field. Some of them are still at research level like the Wave based method that have made huge progress since its presentation but it is still not completely practical for vehicle application. Preliminarily, the considered methods in this work were the following: – – – –

Ray tracing (RT) Finite element with Infinite element boundary constraint (IFEM) Finite element with automated perfectly matched layer (APML) Boundary elements with the fast multipole approach (FMBEM)

The applicability test was performed on a simple A-segment car. Ray tracing has proved to be the fastest as the technique consists of the evaluation of the reflection of different rays emitted by a source: the bigger the number of rays that hit the receiver the larger the noise in that point. In Figure 1 the triangular domain of the car and of the hemi anechoic chamber is shown. This method has a big drawback as it is not suited for the diffraction and the acoustic exterior field around a passenger vehicle is mainly due to the diffraction across the edges and reflection on very com-

3

Simulation of exterior noise propagation for the acoustic load estimation of airborne …

plicated geometry. The test performed on the A-segment car has suggested to skip this method for the application.

Figure 1: Ray tracing domain

The finite element model with infinite element boundary constraint on the outer surface of the exterior fluid domain is a robust method that exploits higher order integration scheme on the outer domain in order to damp the wavelength travelling in the normal direction with respect to the surface. It has the advantage that the boundaries of the fluid domain are intrinsically reflective and no further constraint has to be defined. On the other side the open field has to be specified with the IFEM elements.

4

Simulation of exterior noise propagation for the acoustic load estimation of airborne …

Figure 2 IFEM domain of A – segment car

An example of the IFEM model is reported in figure 2, where it can be noted the shape of the outer boundary: it must be a convex volume with a shape similar to an ellipsoid. This method is numerically accurate and fast but it has a drawback for the shape of the exterior domain and for the frequency extension as it is well suited for frequencies up to 4000-5000 Hz (from this point of view it could be a viable solution for the pass-by simulation) but it is not recommended for frequencies up to 8-10 kHz. The finite element method with automated perfectly matched layer is similar to the previous method with two extra advantages: the shape of the exterior fluid domain is arbitrary, but still convex, and the numerical algorithm is faster than the IFEM method. Nevertheless, the frequency limit – exclusively for the required time to solve each frequency – is still around 4000-5000 Hz for a complete vehicle. The last numerical method that was considered is the Fast Multipole BEM. It is a recent iterative solution that decreases the complexity of the standard BEM considering that in the near field the sources are evaluated taking into account only the nearest portions of the model-sources (the first phase of the FMBEM is in fact a geometrical decomposition of the complete volume into “cells” organized by an octree algorithm) while the far field considers all the subdomains. From a numerical point of view is not as efficient as the APML but it is the only one that can reach the desired target value of frequency of 8-10 kHz. In figure 3 a FMBEM model is shown. For this reason the chosen method has been the FMBEM.

5

Simulation of exterior noise propagation for the acoustic load estimation of airborne …

Figure 3 Fast multipole BEM domain. The green plane is the reflection plane

Experimental campaign The validation of the methodology has required quite a huge experimental data collection as the C-segment test vehicle was measured twice: – A first campaign has focused the attention on the engine bay as it is the most complex subdomain from a geometrical point of view. In this case six sources were placed around the engine bay (windshield, front hubs and front grille) and 30 microphones inside the engine bay were placed and measured. – A second extended campaign was then performed on the complete vehicle with the aim of having enough data to feed a SEA model and to compare one to one to the numerical model. For that reason the number of FRFs has exceeded 10000 units and has required a strong postprocessing, as the sources for the acoustic loading are 32 organized in 5 main sources (needless to say that SEA models need to have averaged quantity also for the input). In figure 4,5 and 6 the experimental layout in three different stages of the experimental campaign are shown.

6

Simulation of exterior noise propagation for the acoustic load estimation of airborne …

Figure 4 First phase experimental campaign

Figure 5 Second phase experimental campaign at vehicle level

7

Simulation of exterior noise propagation for the acoustic load estimation of airborne …

Figure 6 Second phase experimental campaign at vehicle level

The campaign has exploited the high frequency source with smaller orifice with the reciprocity theorem that has allowed to use a reduced set of microphones (see figure 5) and several points of excitation: in this way only 32 microphones were necessary. During the first phase of the experimental campaign some sensitivity for the numerical model description was performed: – Engine bay sealing effect; – Evaluation of normal incidence tube absorption of under bonnet treatment; – Evaluation of normal incidence tube absorption of firewall – engine side. The impact of the sealing between the top of the cowl and the bonnet is reported in figure 7 where the green curve is the transfer function P/Qdot between the engine bay and the windshield when the sealing is removed and the red curve refers to the same transfer function when the sealing is correctly installed. It can be noted that the effect is around 20 dB starting from 400 Hz. This consideration has allowed to set up a transfer admittance for the simulation of the noise abatement due to that rubber element.

8

Simulation of exterior noise propagation for the acoustic load estimation of airborne …

Figure 7 Effect of the sealing on the bonnet (red P/Qdot without sealing, green with sealing)

As the engine bay is covered with many acoustic treatments, the complex impedance of those areas was characterized in terms of complex impedance variable in frequency by means of proper samples cut out from the original acoustic trims and tested in the normal impedance tube. The average value of the different samples was then used (see fig. 8 and 9 for an example of the measures)

Figure 8 Kundt's tube absorption values for different samples of firewall trim

9

Simulation of exterior noise propagation for the acoustic load estimation of airborne …

Figure 9 Kundt's tube absorption values for different samples of under bonnet trim

Creation of the Boundary Element model The target of the project is to be able to forecast the exterior acoustic field of a vehicle between 400 and 8000 Hz, therefore, in order to accomplish this task, three boundary element (BEM) models were realized considering the full vehicle level.

Figure 10 Exploited workflow to build and run the BEM model

10

Simulation of exterior noise propagation for the acoustic load estimation of airborne …

The approach starts from the computational fluid dynamic model (CFD) that is a watertight model widely used. By using wrapping techniques and remeshing phases the CFD model was split into three different size BEM models (see figure 11): – Coarse model with average mesh size of 25 mm for frequencies up to 3200 Hz; – Medium model with average mesh size of 15 mm for frequencies up to 5400 Hz; – Refined model with average mesh size of 7 mm for frequencies up to 8200 Hz.

Figure 11 BEM models with different sizes ranging from 25 mm right to 7 mm left

The change of the mesh size is fundamental to increase the computational efficiency of the octree algorithm beneath the Fast Multipole BEM (FMBEM) method . Nevertheless, itis not trivial since it requires to remove some tiny details (CFD models have mesh of 1 mm for some areas) and to mesh with an almost constant mesh the geometrical domain. . Moreover, some details needed a proper model, different from the CFD one, for acoustic purposes: e.g. the front radiator (figure 12) and the sealing of the bonnet. In particular, the latter was modelled as a transfer admittance that allows to represent a jump of pressure across that part.

11

Simulation of exterior noise propagation for the acoustic load estimation of airborne …

Figure 12 Model of the radiator with intercooler (holes with equivalent area)

Engine bay analysis and validation with the experimental campaign of the first phase The first stage of the research project consisted in simulating the airborne transfer functions from sources located in the front of the vehicle (2 above the windshield, 2 in front of the grille and 2 at 150 mm out of the wheel hub) towards the engine bay. The receivers inside the engine bay were 30 in different areas. The comparison of the P/Qdot transfer functions between the exterior sources and the microphones inside the engine bay (with 20 Hz frequency steps) was good. The first validation of the engine bay was performed exploiting the coarse model of size 25 mm and comparing the results between 300 and 3000 Hz as reported in figure 13. As the most complicated geometrically area – the engine bay – has proven to be sufficiently accurate, the model of that part was frozen and used for the other models.

12

Simulation of exterior noise propagation for the acoustic load estimation of airborne …

Figure 13 P/qdot transfer functions comparison between exterior source and engine bay receiver (experimental blue, numerical brown)

Full vehicle analysis and validation with second phase experimental campaign The final step of the validation consists in a full vehicle simulation exciting different source locations with a monopole. The three FMBEM models have been evaluated separately in different frequency ranges so as to shorten the required elapsed time. Table 1 required elapsed time for the different models Model

# of frequencies

25 mm size (140000 nodes) 15 mm size (400000 nodes) 7 mm size (870000 nodes)

50 (< 3000 Hz)

Required time for covering the range (days) 3 days

9 (> 3000 & < 5400 Hz)

7,5 days

8 (> 5400 Hz)

16 days

The number of evaluated frequencies was defined after a trade-off between accuracy and the necessity to evaluate the smallest number of frequencies. After some trials (estimated with the coarse model), a sufficient reliability was found exploiting the 12th octave band. In the future it will be probably possible to switch to 24th octave band for a more precise evaluation, but the current status is that with those frequency steps the accuracy is within 1.5 dB (see figure 14 for the average of the pressure of the exterior field).

13

Simulation of exterior noise propagation for the acoustic load estimation of airborne …

Figure 14 comparison between average values of pressure on all the output points in fine bandwidth (red line), averaged in third octave with points evaluated with steps of 20 Hz (red diamond shape) and in third octave band with points evaluated in twelfth octave band (gray squared shape)

The results, in terms of pressure on the outer surface, were evaluated as in the experimental process at 40 mm from the wetted surface so as to reliably place the field points in the FMBEM simulation. The huge number of output points was reported in figure 15 like a network of points around the model.

14

Simulation of exterior noise propagation for the acoustic load estimation of airborne …

Figure 15 Field points of the FMBEM model aligned to experimental locations

Then the ASCII files of all the points were processed and the diffuse acoustic field for the SEA model was evaluated. The same integration scheme (spatial average of 4 to 7 points for each panel and frequency average of 12th octave bands) was used both for the experimental data and the numerical pressures. A Matlab routine was then elaborated in order to automate the complete process and to write XML files in the format readable in VAone. This process has dramatically decreased the required time to upload the acoustic load on the airborne model. With those loads, the airborne model was run twice and the noise inside the cabin was evaluated for the final assessment of the virtual analysis. The results were then analyzed in terms of average of absolute differences in third octave bands of the noise at driver’s head cavity with an acceptable mean error of maximum 1.8 dB (see figure 16 for four comparisons). The acoustic loads were also compared for each panel and the maximum error was located in panels that are far from the source. It has then been inferred that after multiple diffractions also the FMBEM method is losing accuracy. However as the acoustic loads decrease with the increase of the distance from the source, this error (around or over in small areas 3 dB) is not affecting too much the estimation of the source.

15

Simulation of exterior noise propagation for the acoustic load estimation of airborne …

Figure 16 Comparison of noise at driver's head cavity for the model loaded with experimental pressures (red) and numerical pressures (blue)

Conclusions The presented work has shown the potentiality of an integrated approach for the minimization of the time required for a fully virtual analysis of the airborne acoustic transfer functions. It is reliable with an average error between numerical and experimental data less than 2 dB, and it has been proven to be industrially suitable for the development process if massive usage of parallel computation is used and if the model is created starting from the already available CFD model. Nevertheless, the capability of a complete virtual analysis allows to assess the different configurations in an early phase of design. Moreover, the developed software tool is capable of creating a seamless approach that could shorten the time required for the translation of the acoustic load from the BEM software into the SEA model. This approach has also the benefit of a quite accurate description of the exterior field and therefore it is an enabler of the pass-by noise simulation in terms of transfer functions.

16

Simulation of exterior noise propagation for the acoustic load estimation of airborne …

Acknowledgements The authors gratefully acknowledge Eng R. Raniolo for the original ideas and Eng. P. Napolitano for the fruitful cooperation in the data automation.

References [1]

Lyon H., Maidanik G., “Power flow between linearly coupled oscillators”, J. Acoustical Society of America, 34, 5, 623-639, 1962

[2]

Danti, M., Raniolo R., Van Genechten B., Rejlek J., Galezia A., Ruschmeyer S., 2012, “Recent advances in the simulation of exterior noise field propagation and its frequency extension”, ISMA 2012 Leuven

[3]

Miccoli, G., Parise G., Silar P., Priebsch H. H., Bertolini C., Nizzoli T., 2012, “Application of the wave based technique for the exterior acoustic field of a simplified car mock up”, ISMA 2012 Leuven

[4]

http://www.lmsintl.com/simulation/virtuallab/acoustics

[5]

Muller, S., Cotoni, V., Connelly, T., 2009, “Guidelines for using Fast Multipole BEM to calculate automotive exterior acoustic loads in SEA models”, SAE paper 2009-01-2220.

Authors Eng. M. Danti Dr. G. Bartolozzi Mr. M. Meneguzzo Vehicle concept & integration / Performance Integration & Validation / Vehicle integration CRF – NVH, Centro Ricerche Fiat, Strada Torino 50, 10043 Orbassano (TO), Italy Dr. C. Campagna Vehicle concept & integration / Performance Integration & Validation/ Noise Vibration & Harshness PCC, Corso Settembrini 40, 10135 Torino, Italy

17

On the role of simulation in accounting for the design complexity of engine encapsulation R. D’Amico1, R. Stelzer1, J. Grebert1, P. Chandler 2, G. Fossaert2 1

Autoneum Management AG, Switzerland

2

Jaguar Land Rover Limited, UK

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 W. Siebenpfeiffer (Hrsg.), Automotive Acoustics Conference 2017, Proceedings, https://doi.org/10.1007/978-3-658-20251-4_12

1

On the role of simulation in accounting for the design complexity of engine …

1 Introduction Driven by more and more stringent CO2 [1] and noise emission regulations [2], the need for advanced engineering in the engine bay has significantly increased over the last few years. One of the most effective solutions involves engine encapsulation, which consists in enclosing as much as possible the engine into a capsule-like assembly. On the one hand, this allows retention of heat and reduced CO2 emission, and, on the other hand, decreasing both interior and exterior vehicle noise [3][4]. Design and material choices are key to develop an effective encapsulation or to improve its performance. The most common procedure to assess a design and its improvement is experimental. The vehicle is equipped with a prototype capsule and measured with respect to its thermal and acoustic performances. However, several build-test-optimize loops are often required to reach further improvements, thus generating longer development times and higher costs. In this context, simulation can streamline the process and drastically reduce the use of resources. With respect to pure testing, simulation additionally allows accounting for a vast amount of design details, bills of materials (BOMs) and – yet very important – encapsulation configurations. However, for simulation to be successful, material and geometry description as well as underpinning assumptions must be carefully chosen and defined. The purpose of this paper is to show the role of simulation in handling the engineering complexity of the problem and its potential within the NVH design process investigated under a collaboration between Autoneum and Jaguar Land Rover. The problem complexity and its variables are presented in Sec. 2, while Sec. 3 provides an overview of the simulation tools to address all specific design questions. Section 4 focuses on the vehicle level simulations and their accuracy with respect to measurements. Section 5 closes the paper by drawing the conclusions.

2 Complexity of the design process The basic idea behind engine encapsulation consists in designing an enclosure around the powertrain with the main function of decreasing as much as possible the noise generated during its operating conditions and storing the heat to achieve reduction of CO2 emissions. Given the complexity of the environment in which the encapsulation is placed, the design process must account for a number of engineering constraints. In general terms, the performance of an encapsulation is dependent on two strongly related aspects of the engineering process, which are the choice of the design of the encapsulating parts and the choice of the optimal BOMs for that specific design.

22

On the role of simulation in accounting for the design complexity of engine …

Starting with the first, encapsulation can be classified in three types depending on where and how the parts are mounted. If the capsule is mounted directly on the engine, the design is usually referred to as engine-mounted, Fig. 1 (left). The vicinity of the engine makes the material requirements particularly high in this case. On the contrary, if the parts are fixed to the surroundings and are not directly attached to the engine, the design is referred to as body-mounted, Fig. 1 (right). Here, to cover a certain engine portion, parts must have a higher surface area with respect to an engine-mounted version. Nevertheless, they are generally located farther from the engine and material requirements are generally lower. Finally, a hybrid design presents a combination of engine- and bodymounted components.

Figure 1 Examples of engine-mounted (left) and body-mounted (right) encapsulations [4].

The choice of the encapsulation strategy very much depends on the overall engine bay architecture, the type of powertrain, the space available, the assembly process and required serviceability. Although in principle none of the aforementioned strategies is more performing than the others, the portion of encapsulated engine surface, known as coverage ratio, is one of the parameters driving the performance. In general, a coverage ratio of about 80% leads to significant achievements, both in terms of CO2 reduction and NVH improvements [3]. However, given the geometrical complexity of the engine bay and of the power train, reaching and preserving high levels of coverage requires significant design effort. As mentioned earlier, choosing the BOM is a complementary part of the design process. In general terms, materials for encapsulations must withstand a series of requirements depending also on the mounting position of the part they are related to. First of all, material choice must comply with the part production process, weight, cost and packaging space available. Given the vicinity with the source, heat resistance is required together with good thermal properties if the encapsulation has heat storage requirements. In this regard, the material must comply with safety requirements as well. Resistance to chemical attack is needed to withstand the highly aggressive environment the encapsulation components are placed in. Additionally, given the high levels of vibration, fatigue resistance and appropriate fixing methods are another requirement that in particular interests engine-mounted parts.

3 3

On the role of simulation in accounting for the design complexity of engine …

When it comes to the choice of the material to achieve the desired acoustic target, one must consider that often the performance of a part is not just reflected by its absorption and transmission loss (TL) characteristics, but rather obtained from a balance of the two. This can be quickly seen by recalling the following mathematical relation that quantifies the insertion loss (IL) of a generic acoustic enclosure [5]: ܵ௪ ሺ ߙ ൅ ߬ௗ ሻ ൅ ͳ ܵ஺ ௗ ൢሾ†ሿ ܵ௪ ߬ௗ ൅ ͳ ܵ஺

‫ ܮܫ‬ൌ ͳͲ Ž‘‰ଵ଴ ൞

(1)

whose symbols are explained in Fig. 2. Equation (1) is very simple, yet it takes into account all macro-variables driving the acoustic performance of an encapsulation, namely coverage ratio, openings and acoustic properties of the materials. Equation (1) also reveals that using either purely absorbing or insulating materials alone is not enough to reach the best performance. While high TL is necessary to retain the noise, high absorption allows dissipating it and reducing the leaks through the openings. Thus, both acoustic characteristics are necessary and must be well-balanced.

Figure 2 Simple model of engine encapsulation.

Equation (1) can be very handy for an early assessment of the capsule performance. Nevertheless, more accurate and specific tools are necessary to account for the full complexity of the problem, which includes several additional aspects. For instance, the part geometry and thickness distribution play an important role, as well as its position with respect to the engine. Furthermore, when assessing the vehicle level performance, the acoustic engine bay package must be considered. In such a complex context, the flexibility of simulation tools comes to the forth, as they can support the engineering process efficiently and effectively, and they can shortcut the time-demanding process involving prototyping for optimization. Additionally, simulation tools allow a much wider spectrum of data post-processing and analysis, which in many cases is hardly feasible and costly to be carried out experimentally. As described in the next section, NVH engineers can rely on a set of tools allowing decision making throughout the design process.

44

On the role of simulation in accounting for the design complexity of engine …

3 NVH simulation tools in context Moving from material- to vehicle-level performance assessment (Fig. 3), the first question to answer is related to the quantification of the behaviour of different BOMs and their ranking. Here the Transfer Matrix Method (TMM) represents a very flexible and efficient tool to predict the absorption and TL of a multi-layer encapsulated part [6]. For part level simulation, the TMM is implemented in the software VisualSISAB, developed by Autoneum, see Fig. 4. Starting from the CAD of the component, VisualSISAB allows extracting its thickness distribution and acoustic properties. These information are then used to rank several BOMs and determine the aforementioned desired balance between absorption and TL, calculated from flat sample performance data.

Figure 3 Simulation tools to support different design phases.

Figure 4 Workflow for acoustic part performance assessment with VisualSISAB.

Depending on the location of the part, different excitations can be used for the assessment [7]. Within the TMM, the material is assumed to be of infinite extent. No edge,

5 5

On the role of simulation in accounting for the design complexity of engine …

boundary or fixation effects are taken into account in the simulation. For this reason, when considering the effect of finite size and boundary conditions, the Finite Element Method (FEM) provides a more suitable framework that can be used up to vehicle level performance assessment. Thanks to the discretization into small elements, the FEM allows a high-fidelity representation of the component geometry. Additionally, the FEM belongs to the class of the so-called deterministic approaches, that allow describing the acoustic field directly in terms of pressures and capturing resonant behaviours accurately. Previous studies have shown the effectiveness of these approaches in dealing with engine bay problems [8][9]. Opposite to deterministic approaches are energetic methods [10][11], of which underpinning assumptions include the diffusivity of the acoustic field. It is worth noting that, given the geometrical size of the problem and the acoustic wavelengths travelling in the engine bay, such an assumption may not be valid over the whole frequency range of interest, that commonly reaches 2.5 to 3 kHz. This makes energetic approaches less reliable over that frequency span. One additional limitation of the energetic approaches is that they are not made to accurately describe the problem geometry. Thus, thickness distributions, packaging spaces and small design modifications may not be taken into account with a satisfying level of detail. Besides acoustic FE, porous elements are used to model the acoustic treatments and engine encapsulation. The choice of adopting a full Biot formulation with respect to an impedance boundary condition is related to the possibility of accurately capturing the waves travelling through the material [6]. This comes at the price of larger models, of which common size is around 1 to 2 Mio. nodes at vehicle level. However, the size of the model is highly dependent on the number of components accounted for, and in turn, on the level of accuracy to be reached. Generally, key components are the hood liner, outer dash and under engine shield. However, the model refinement should always be chosen based on the physical understanding to gain. This is important to balance the modelling effort to the accuracy level. Main hurdle of the FEM is the time required to mesh the model, which takes a large portion of the overall simulation process. This becomes even more limiting when several design-simulation loops are necessary. For example, in the case of a thickness distribution change, the user has to perform almost a new meshing process. To overcome this time consuming operation, Autoneum is currently developing material homogenization approaches allowing material and thickness changes without requiring re-meshing. Such procedures result in a significant speed-up when it comes to design optimization loops [7]. Another aspect to take into account during the simulation process is the source modelling, which is important and yet challenging. Generally, intensity and pressure meas-

66

On the role of simulation in accounting for the design complexity of engine …

urements are a common way to assess an engine radiation pattern. However, a number of difficulties arise when translating them into velocity profiles for the FE model, for example, through inverse characterization approaches. Such a refined piece of information may not be readily available and very demanding to obtain. In case the simulation target is the assessment of a performance improvement with respect to an initial design, using a simplified source provides a valid alternative, although not applicable to assess absolute noise levels. In the reminder of this paper, a uniform surface vibration is used as an excitation and to calculate the difference in performance between two configurations. As shown later, such a simplification allows efficient modelling choices without losing control on the problem at hand. The tools described so far allow tackling most of the key design aspects presented in Sec. 2. They are tailored to answer specific questions, but their synergies can strongly support engineers during the design and development process.

4 Encapsulation design by means of simulation tools As mentioned in the Introduction, the work presented in this paper is part of a concept development project. The vehicle under consideration is a Jaguar XE mounting a 2.0l longitudinal petrol engine. Three vehicle states are taken into account in the following analysis: ●

State 0 / Bare. The vehicle implements the classic acoustic engine bay package including absorbing hood liner, outer dash, under engine shield and under floor. No encapsulation is present in this state.

●

State 1 / Baseline. Vehicle engine bay package is equal to State 0 but an enginemounted encapsulation is added. Its geometrical features are represented on the left side of Fig. 1.

●

State 2 / Developed concept. Also in this case, the acoustic engine bay package is the same as in the previous cases. The new encapsulation concept is here implemented and can be classified as a hybrid. While the gearbox and oil sump covers are unchanged with respect to State 1, a box-like capsule is placed around the engine block. The total surface of the capsule is larger than in State 1 and its final coverage is about twice as much as the baseline. Given the use of a lightweight material like ThetaFiber-Cell, the weight of the capsule is slightly lower than the one in State 1, although the higher coverage. Precise details of the design cannot be reported on this document.

Moving from State 1 to 2 required several intermediate CAD steps to account for geometrical, technical, service and manufacturing constraints. The performance assessment at material level was supported by VisualSISAB, that allowed ranking the different

7 7

On the role of simulation in accounting for the design complexity of engine …

BOM and selecting the most promising combination. These are already examples of how CAE tools can support the design process and be the starting point for prototyping. Following these phases, a FEM model of the vehicle, including all acoustic components were built and used for all preliminary predictions and design assessments. Even though, the thermal aspects are not discussed here, thermal simulations were also ran to predict the heat storage capability of the designed capsule. The next section discusses the accuracy of the vehicle model to capture the key acoustic paths. Successively, the model is used to assess the performance improvement from State 1 to State 2 and to compare the simulated data with measured ones.

4.1 Assessment of the simulation accuracy In this section the accuracy of the FE model is assessed by comparing the results of the simulations with measurements. For this validation State 0 and 2 are considered and Acoustic Transfer Functions (ATF) are used as a measure of the contribution from either the top or bottom engine faces to the exterior field. The details of the measurement and simulation procedures to quantify them are now discussed.

4.1.1 Measurement and simulation setup During the measurement process, the vehicle is placed into a semi-anechoic chamber, and three volumetric source positions are chosen on the left-hand side of the vehicle at a height of 0.3m and 1.5m distant from the vehicle side, see Fig. 5. Ten flat microphones are located on the engine top, below the engine top cover. This set of microphones allows assessing the transfer between the engine top face and the points outside the vehicle. Ten additional microphones are located in the bottom area of the engine bay, including some points over the oil sump, lower gearbox and the surroundings. They allow assessing the acoustic transfer between the lower part of the engine and the outside points. The aforementioned measurements are carried out in bare conditions as well as

Figure 5 Source location for vehicle ATF measurements.

88

On the role of simulation in accounting for the design complexity of engine …

with encapsulation with encapsulation gine bay. with encapsulation gine with bay. encapsulation gine bay. gine bay.

State State State State

2. 2. 2. 2.

Figure Figure Figure Figure

6 6 6 6

shows shows shows shows

the the the the

locations locations locations locations

of of of of

the the the the

microphones microphones microphones microphones

in in in in

the the the the

enenenen-

Figure 6 Microphone locations on engine top (left) and engine bottom (centre and right). Figure 6 Microphone locations on engine top (left) and engine bottom (centre and right). Figure 6 Microphone locations on engine top (left) and engine bottom (centre and right). Figuresimulation 6 Microphoneset locations on engine top (left) and engine bottomconditions (centre and right). The up reproduces the measurement accurately.

Point source The simulation set up reproduces the measurement conditions accurately. Point source and receiver locations are chosen the in agreement with the aforementioned upsource in all The simulation set up reproduces measurement conditions accurately. set Point and receiver locations are chosen the in agreement with the aforementioned upsource in all The simulation set up reproduces measurement conditions accurately. set Point configurations. While for vehicle CAD models are used, 3D-scanning and receiver locations arethe chosen in original agreement with the aforementioned set up of in the all configurations. While for vehicle CAD models are used, 3D-scanning and receiver locations arethe chosen in original agreement with the aforementioned set up of in the all prototyped encapsulation is performed to ensure geometrical details are accurately configurations. While for the vehicle original CADall models are used, 3D-scanning of the prototyped encapsulation is performed to ensure geometrical details are accurately configurations. While for the vehicle original CADall models are used, 3D-scanning of the captured. encapsulation is performed to ensure all geometrical details are accurately prototyped captured. encapsulation is performed to ensure all geometrical details are accurately prototyped captured. All engine bay acoustic treatments are modelled with porous elements implementing the captured. All engine bay acoustic treatments are modelled with porous elements implementing the u-p engine formulation [6]. Specifically, included the model are the hood All bay acoustic treatments the are components modelled with porous in elements implementing the u-p engine formulation [6]. Specifically, included the model are the hood All bay acoustic treatments the are components modelled with porous in elements implementing the liner,formulation the outer dash, the under engine shield and the under floor, indicated 7. u-p [6]. Specifically, the components included in theasmodel are in theFig. hood liner,formulation the outer dash, the under engine shield and the under floor, indicated 7. u-p [6]. Specifically, the components included in theasmodel are in theFig. hood It is worth mentioning the engine capsuleshield is almost entirely liner, the outer dash, thethat under and the undermade floor,ofasTheta-FiberCell indicated in Fig.but 7. It is worth mentioning the engine capsuleshield is almost entirely liner, the outer dash, thethat under and the undermade floor,ofasTheta-FiberCell indicated in Fig.but 7. over a limited portion ofthat surface presentsisaalmost sandwich heat shield whose outer sheets but are It is worth mentioning the capsule entirely made of Theta-FiberCell over a limited portion ofthat surface presentsisaalmost sandwich heat shield whose outer sheets but are It is worth mentioning the capsule entirely made of Theta-FiberCell madea of perforated These details are properly accounted for in the are FE over limited portionAluminium. of surface presents a sandwich heat shield whose outer sheets made perforated These details are properly accounted for in the are FE over a of limited portionAluminium. of surface presents a sandwich heat shield whose outer sheets model.ofAllperforated structural Aluminium. components that aredetails included the model are treated made These areinproperly accounted forasinfully the rigFE model.ofAllperforated structural Aluminium. components that aredetails included the model are treated made These areinproperly accounted forasinfully the rigFE id, theseAll include for instance, the that bonnet, the dash,inthe body-inmodel. structural components are included theengine modelbay are walls, treatedthe as fully rigid, theseAll include for instance, the that bonnet, the dash,inthe body-inmodel. structural components are included theengine modelbay are walls, treatedthe as fully rigwhite butinclude also thefor wheel outer liners, that are thiswalls, application. These id, these instance, the bonnet, thehighly dash, reflective the engineinbay the body-inwhite butinclude also thefor wheel outer liners, that are thiswalls, application. These id, these instance, the bonnet, thehighly dash, reflective the engineinbay the body-inassumptions by liners, the factthat thatarethehighly size of the engine bayapplication. openings isThese such white but alsoare thesupported wheel outer reflective in this assumptions by liners, the factthat thatarethehighly size of the engine bayapplication. openings isThese such white but alsoare thesupported wheel outer reflective in this that the acoustic transfer from source thesize receiver miassumptions are supported by the fact thattothe of thethrough engine the baystructure openingsisisofsuch that the acoustic transfer from the source to the receiver through the structure is of miassumptions are supported by fact that the size of the engine bay openings is such nor relevance. Totransfer model the field car, thetheAutomatic that the acoustic fromfree theacoustic source to the around receiverthe through structure Perfectly is of minor relevance. Totransfer model the field car, thetheAutomatic that the acoustic fromfree theacoustic source to the around receiverthe through structure Perfectly is of miMatched LayerTo (APML) is used. This consists in additional layers of 3D FE Perfectly elements nor relevance. model the free acoustic field around the car, the Automatic Matched LayerTo (APML) is used. This consists in additional layers of 3D FE Perfectly elements nor relevance. model the free acoustic field around the car, the Automatic that provide strong artificial damping, that noinsound is reflected infinity, as it is Matched Layer (APML) is used. Thisso consists additional layersfrom of 3D FE elements that provide strong artificial damping, that noinsound is reflected infinity, as it is Matched Layer (APML) is used. Thisso consists additional layersfrom of 3D FE elements the case in the radiation to adamping, free field.soFinally, fully is reflective placed as below that provide strong artificial that no asound reflectedplane fromisinfinity, it is the case in the radiation to a free field. Finally, a fully reflective plane is placed below that provide strong artificial damping, so that no sound is reflected from infinity, as it is vehicle model to reproduce effect of theafloor. The software Actranbelow 16 is the case in the radiation to a freethe field. Finally, fully reflective planeFFT is placed vehicle model to reproduce effect of theafloor. The software Actranbelow 16 is the case in the radiation to a freethe field. Finally, fully reflective planeFFT is placed the vehicle model to reproduce the effect of the floor. The software FFT Actran 16 is the vehicle model to reproduce the effect of the floor. The software FFT Actran 16 is

Figure 7 Components included into the vehicle model. Figure 7 Components included into the vehicle model. Figure 7 Components included into the vehicle model. Figure 7 Components included into the vehicle model.

9 9 9 9 9

On the role of simulation in accounting for the design complexity of engine …

used to run the simulations that account for all abovementioned features. Excluding the APML elements, the bare model consists of about 7 Mio elements and 1.5 Mio nodes (fluid and porous). The simulation sweeps frequencies from 400 to 2500 Hz with 15Hz step, leading to a solution in about 12 hours.

4.1.2 Result analysis In order to summarize the results, ATFs are averaged over the three source positions and in turn either over the top or bottom of the engine. Figure 8 shows the narrow band comparison between the measured and simulated ATFs for State 0, where no encapsulation is mounted on the engine. Left and right pictures show the top and bottom engine ATFs, respectively. Figure 9 reports the same data calculated in third octave bands. For both ATFs, the trend is correctly captured and with good accuracy. In the case of the bottom engine ATF, simulation results are always within 3dB from the measured data, which can be considered a reasonable bound to assess the accuracy of the predictions [8]. Errors are slightly larger when the top engine ATF is considered, indicating a little underestimation of the absorption within the engine bay. Given the complexity of the engine bay, such a contribution could also come from details that are not taken into account within the numerical model. Nevertheless, their influence on the accuracy of the response can be considered still within reasonable bounds.

Figure 8 State 0. Engine top ATF in narrow band. Comparison between measurements and simulations.

Figure 9 State 0. Engine top ATF in third octave bands. Comparison between measurements and simulations.

10 10

On the role of simulation in accounting for the design complexity of engine …

Analogously Fig. 10 and 11 present the data for State 2, where the capsule is mounted on the engine. The simulated bottom engine ATF well predicts the measurements and provides an accuracy below 3dB over the full frequency range. Also the top engine ATF shows a good accuracy. As can be seen on the left picture, the trend is very well captured at low frequencies. At higher frequencies, there is probably a slight underestimation of the absorption in the engine bay, which leads to higher ATF level.

Figure 10 State 2. Engine top ATF in narrow bands. Comparison between measurements and simulations.

Figure 11 State 2. Engine top ATF in third octave bands. Comparison between measurements and simulations.

Figure 12 summarizes the results by displaying the absolute errors. The right picture shows the absolute difference between the simulation prediction and the measurements over third octave bands. For each configuration data are averaged and reported on the right bar graph. On the left it can be observed that the two states share similar error trends in some frequency regions. This is very clear for the bottom engine ATF curves. For the top engine ATFs, this is evident mainly at low and high frequencies over the range. Figure 13 condensates the absolute error for all configurations in a histogram together with its cumulative. Most recurring error values are included into the interval ranging from 0.5 to 1 dB, and about 70% of the predictions lead to an absolute error below 2dB [8]. The overall level of accuracy is quite remarkable, given the complexity of the problem. This also allows concluding that the key geometrical features of the model are correctly represented within the FE framework and that material models are accurate enough over the frequency range of interest. Even more important is that this analysis confirms that the modelling procedure allows reliably capturing the key acoustic paths and phenomena taking place inside the engine bay but also towards the exterior field.

11 11

On the role of simulation in accounting for the design complexity of engine …

Figure 12 Absolute error calculated over third octave bands (left) and averaged (right).

Figure 13 Histogram of absolute error and cumulative.

4.2 Assessment of the encapsulation design The following section focuses on the performance improvement moving from State 1 to State 2. In order to assess such an improvement it may be helpful understanding on which regions of the engine the encapsulation is more effective. To this end, the surface of the engine model is split into several areas. Uniform vibration is applied to each of these regions separately to capture the transfer functions to the exterior microphones. As shown later, besides being very practical, the application of uniform velocity over the source leads to results that are comparable to the performance measured when the engine is running.

4.2.1 Measurement and simulation set up The measurement setup used for the following analysis is geometrically similar to the one showed in Fig. 5. Exterior sources are replaced by microphones and mirrored on the right hand side. Engine run-up’s in neutral is the operating condition during measurements, and data are taken from the range between 2500 and 5000 rpm.

12

12

On the role of simulation in accounting for the design complexity of engine …

The overall simulation procedure is not modified with respect to what described in the previous section, as well as the frequency range of analysis. Different materials are used in the State 1 encapsulation, like foam, and they are taken into account in the numerical model. One difference with the previous model is the excitation. The engine surface is split in regions, to which a unit velocity is applied in separate simulation load cases, see Fig. 14. This excitation condition is chosen for two main reasons. First of all, by splitting the engine in several surfaces it is possible to assess their respective contribution to the exterior noise, as they were engine ATFs to the exterior field. Secondly, the uniform vibration condition leads to a unitary radiation efficiency, namely the surface radiates very well. While in reality this is not the case, in simulation it defines a critical scenario to excite the acoustic treatment. From a computational perspective, solving the same model for several load cases does not increase the calculation time drastically. This is due to the mathematical structure of the FEM matrices and their advantageous features when it comes to the solving phase.

Figure 14 Engine surface split.

4.2.2 Result analysis Figure 15 shows the SPL improvements reached by moving from State 1 to 2 for the engine regions that are most interested by the State 2 capsule. The pressures at the left and right microphones are averaged to estimate an overall SPL level. The delta is calculated by subtracting the SPL of State 1 to the one of State 2. Consequently, the lower the delta SPL, the higher the benefits of the new concept are. Figure 15 shows that the contributions from almost all regions of the engine are reduced by adopting the State 2 encapsulation. This is especially true for regions that experience a higher coverage like the rear and the top. When considering the whole engine vibration, the entire surface is equally excited and the contributions from untreated regions also influence the overall delta performance.

13 13

On the role of simulation in accounting for the design complexity of engine …

Figure 15 Delta SPL for uniformly vibrating engine surfaces.

It is important to notice that as long as the sound radiation from a certain surface cannot be quantified nor weighted with respect to the others, the information reported in Fig. 15 is relative and not absolute. Thus, it does not reveal the absolute performance of the capsule when the engine is running. Nevertheless, the minimum and maximum bounds of the previous chart can also be interpreted as performance bounds for the case of running engine. As the weightings are unknown, let us assume they are all equal, thus a uniform and constant velocity is applied over the whole engine surface. This boundary condition should ensure that all key acoustic paths are excited in both states, allowing the calculation of the performance improvement that is expected to be comparable to the one measured when the engine is running. In Fig. 16 this approach is verified by comparing the three states by taking as reference the State 1 (0 dB). For State 2, using a uniform velocity leads to results very close to the measurements when the engine is running, with the low frequency being the main exception. When applied to State 0, the delta

Figure 16 Delta SPL calculated by simulation with uniformly vibrating engine and measurements with running engine. State 1 is the reference (0 dB line).

14 14

On the role of simulation in accounting for the design complexity of engine …

performance is overestimated. However the global trend and growth rate is captured over the frequency range under analysis, except again at low frequencies. Since the exact source vibration is unknown, it is difficult to provide a reason behind such differences. Additionally, when the engine is running many specific noises are present that would require dedicated characterization to be modelled. Nevertheless, Fig. 16 shows that the application of a uniform velocity is an efficient way of supporting the performance assessment, especially in an early design stage.

5 Conclusions Tightening regulations on CO2 and noise emissions require greater need for high performance engine encapsulations, of which design must account for a high degree of complexity already from its early stages. This paper shows how simulation allows accounting for such complexity when it comes to NVH design by considering a real concept development case study. While TMM and VisualSISAB are reliable tools to support the choice and ranking of the suitable materials, the FEM is proposed to tackle vehicle level assessment. As shown in the paper, the latter can predict the acoustic performance with a good degree of accuracy at engine ATF level – lower than 2dB in 70% of the analysed cases. When assessing the performance with running engine, the inclusion of the source might prove cumbersome. However, as shown in the paper, using a uniformly vibrating surfaces can be an efficient – yet effective – starting point for an overall performance assessment, even when considering a running engine. Another relevant point to address concerns the use of the present simulation tool to predict the effects of the capsule on the interior acoustics. In fact, the tools discussed so far allow assessing the airborne contribution to the exterior, and thus the pressure loading over the whole vehicle surface, being this the outer shell, but also the relevant components in the engine bay, like the outer dash or outer tunnel. These pressures can be seen as interface information to be used as input for an interior FE model or even Statistical Energy Analysis (SEA) calculation, and that come out of exterior FE simulations at no additional calculation cost. However, while tackling such a problem in its full complexity may be very challenging from a simulation perspective, focusing on intermediate quantities allows assessing the overall benefits without putting in place a full modelling for the vehicle interior. This is an interesting outlook for the presented methodology.

References [1] Regulation EC No 443/2009 (2009).

15 15

On the role of simulation in accounting for the design complexity of engine …

[2] Regulation ECE R51.03 (2014). [3] T. Bürgin, C. Bertolini, D. Caprioli, C. Müller, Engine encapsulation for CO2 and noise reduction, ATZ worldwide, 116(3), 16-21 (2014). [4] C. Bertolini, D. Caprioli, J. Horak, T. Bürgin, D. Petley, W. Jansen, B. Wicksteed, Acoustic benchmark of different CO2 encapsulation strategies: application examples and methods for the design of powertrain encapsulation, Automotive Acoustic Conference, Zürich (2013). [5] F. Fahy, Foundations of Engineering Acoustics, Elsevier, 2nd edition (2005). [6] J.-F. Allard, N. Atalla, Propagation of Sound in Porous Media: Modelling Sound Absorbing Materials, John Wiley & Sons, 2nd edition (2009). [7] R. D’Amico, R. Stelzer, A. Bihhadi, T. Courtois, Engine encapsulation: impact on vehicle NVH and CAE challenges, International Conference on Noise and Vibration Engineering (ISMA2016), Leuven (2016). [8] A. Bihhadi, C. Bertolini, C. Locqueteau, Simulation of exterior powertrain ATFs on an engine bay mock-up with complex trim configurations, Automotive Acoustic Conference, Zürich (2015). [9] G. Miccoli, C. Bertolini, A. Bihhadi, Comparative analysis of different deterministic methods for the simulation of exterior noise acoustic transfer functions, AIA-DAGA Conference, Merano (2013). [10] B. Andro, S. Chaigne, T. Schmitt, A. Shah, Simplified integral energy method application to pass by noise, ICSV13, Vienna (2006). [11] M. Thivant, A. Cloix, C. Clerc, N. Blairon, C. Braguy, Boundary Element Energy Method: an efficient tool for acoustic computation, 10ème Congrès Francais d’Acoustique, Lyon (2010).

16 16

Pass-by noise synthesis from frequency domain exterior acoustic simulations Alexis Talbot, Gregory Lielens Free Field Technologies, Mont-Saint-Guibert, Belgium

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 W. Siebenpfeiffer (Hrsg.), Automotive Acoustics Conference 2017, Proceedings, https://doi.org/10.1007/978-3-658-20251-4_13

1

Pass-by noise synthesis from frequency domain exterior acoustic simulations

The evolution of the Pass-by Noise regulation put Automotive OEMs under pressure to reduce the emitted noise from vehicles. To be able to decrease the emitted noise level while meeting other constraints such as weight, space or sound quality, OEMS and their suppliers have to come up with innovative solutions to reduce the different noise sources and study the noise propagation around the vehicle. The experimental study of the acoustic propagation around a complete vehicle during a pass-by noise event occurring late in the development process, the need for methods to investigate solutions to control the acoustic propagation around the vehicle arises. This paper presents a method for predicting a vehicle pass-by noise based on a Finite Element exterior acoustic calculation in the frequency domain. With an approach similar to what is proposed for experimental indoor pass-by noise measurement, the simulated pass-by noise is synthesized from multiple acoustic transfer functions between each noise source and a line of virtual microphones located 7.5m on the side of the vehicle. The method is demonstrated on a car model using Actran commercial software. The contribution from several noise sources to the pass-by noise is evaluated up to 2 kHz and expressed both in terms of sound pressure level and frequency content. Finally the performances and computational times are reported for integration in a design process.

Introduction The evolution of the Pass-by Noise regulation [1] put Automotive OEMs under an increasing pressure to reduce the emitted noise from vehicles. Noise emission constraints are becoming of increasing importance and add a layer of complexity to the process of designing a car. To be able to decrease the emitted noise level while meeting other constraints such as weight, space or sound quality, OEMs and their suppliers have to come up with innovative solutions to reduce the different noise sources and study the noise propagation around the vehicle. During the past decades, numerical modelling methods have emerged as powerful tools for the development and optimization of safer, more efficient and more comfortable vehicles. The use of simulation complements the different experiments conducted on prototypes and allows engineers to test design changes more efficiently and earlier in the design process. Thanks to the improvements brought to the meshing algorithms and the numerical methods on one hand and the increase in computational resources on the other hand, the Finite Element Method is now widely used to solve complex acoustic radiation problems for large systems and relatively high frequencies. The acoustic scattering around a complete car for various type of noise sources can be computed up to 2 kHz in approximately 10h using a frequency response analysis. In order to provide engineers with results coherent with the indicators used in the passby noise measurement standard [2], a post-processing methodology using frequency domain numerical results is presented. The approach used is similar to what is proposed

22

Pass-by noise synthesis from frequency domain exterior acoustic simulations

for experimental indoor pass-by noise measurement [3]. The simulated pass-by noise is synthesized from multiple acoustic transfer functions between each noise source and a line of virtual microphones located 7.5m on the side of the vehicle.

Pass-by noise numerical simulation Pass-by noise reciprocity Noise emissions of ground vehicles are regulated in different regions in order to limit the disturbance introduced by the different vehicles circulating on road networks [1]. The methodology to measure noise emissions is defined by the International Organization for Standardization (ISO) [2] and commonly referred to as “pass-by noise”. It consists in measuring the noise emission of a vehicle under acceleration condition, with various gear positions and in very specific measurement setup described in Figure 1.

Figure 1 - Pass-by noise measurement setup as described in [2]

The tested vehicles is driven through a 20m test zone and two microphones P and P’ are located on both side of the track, at 7.5m on the side and at 1.2m height. The Aweighted sound pressure level is recorded during the event and the maximum value reached during the test is evaluated. In a frequency domain simulation, the vehicle cannot be set in motion since harmonic state of the studied problem is assumed. In order to obtain pass-by noise results using the indicators described in the ISO standard, the approach presented in Figure 2 is used.

3 3

Pass-by noise synthesis from frequency domain exterior acoustic simulations

Figure 2 - Numerical pass-by noise calculation setup

The vehicle is located in the middle of the test zone and two lines of microphones are defined, one on each side of the vehicle at 7.5m from the center of track and at 1.2m from the ground. In order to reconstruct the sound pressure level signal recorded by the test microphone, A-weighted sound pressure level results from the different microphones are used successively starting from the farther microphone in front of the car. The appropriate signal processing that must be applied in order to reconstruct the final signal in a consistent way is presented later in this paper.

Finite Element model description A finite element model is built to study the exterior acoustic propagation around a complete car using the Actran Acoustic commercial software, see Figure 3. The model consists of a wrap surface mesh created based on a detailed car body structure finite element model [4].

Figure 3 - Finite Element Model wrap surface mesh

The wrap surface mesh includes the geometric representation of the car body, the wheels and suspensions, the powertrain, and other parts such as the fuel tank or the battery. A finite element mesh is built around the wrap surface mesh by the solver during the calculation as shown in Figure 4. The surface wrap mesh is considered perfectly rigid in this model.

4

Pass-by noise synthesis from frequency domain exterior acoustic simulations

The exterior acoustic mesh consists of a two volumes of quadratic finite elements. The near field volume mesh (in blue) supports the accurate computation of the acoustic scattering around the vehicle body. The second volume mesh (in red) is designed to damp progressively the outgoing acoustic waves using the Perfectly Matched Layers technology in order to create a non-reflecting boundary condition at the interface between the two mesh volumes. The exterior acoustic finite element model is excited by a constant mass flow rate monopole source located at the tailpipe exit of the exhaust system (in yellow). The acoustic pressure computed at a particular microphone location therefore corresponds to an acoustic transfer function between the source and this microphone. Different noise sources could be considered within the same calculation using several load cases which enable the computation of the exterior acoustic propagation from multiple sources in a single calculation. The floor is considered as perfectly reflecting in this example but a complex admittance value could be considered for the near and far field propagation.

Figure 4 - Finite Element Model volume mesh and acoustic source

From the acoustic field computed in the near field, results at pass-by noise microphone locations are obtained using the integration method of Ffowcs Williams and Hawkings (FWH) based on the pressure and particle velocity information stored on the surface located at the interface between the two mesh volumes. The use of FWH integration methods brings flexibility to the process provided that the pressure and particle velocity can be stored on the surface used for the integration. Results can be computed at different microphone locations in the far field without running the complete calculation.

5 5

Pass-by noise synthesis from frequency domain exterior acoustic simulations

Computational performance The computational performance of the finite element exterior acoustic calculation is studied in order to ensure the process is compatible with application in an industrial process. The calculation is run from 5Hz to 2000Hz with a 5Hz frequency step. Hadaptivity is used to reduce the computation time for low frequencies. Computational times presented in this paper are obtained on a high performance computing node (Intel(R) Xeon(R) CPU E5-2699 v4 @ 2.20GHz, with 396GB of RAM). Four parallel processes are run, each process using 11 threads. All meshes required for the computation using h-adaptivity are created by the solver in a preliminary step. This preliminary step is performed on a single process and takes one hour. The computation of all frequencies using the previously described configuration takes 10 hours and 8 minutes when the calculation only contains one acoustic source. When 10 different loadcases are present, the calculation lasts 10 hours and 30 minutes which confirms the benefit of running multiple loadcases calculations.

Signal processing for pass-by acoustic transfer function Let us consider 𝑝𝑝𝑛𝑛1,𝐴𝐴 the A-weighted acoustic transfer function between the acoustic source delivering unit amplitude mass flow rate Q and the microphone 𝑃𝑃𝑛𝑛 , 𝑛𝑛 designating the position of the microphone on the line of microphone described in Figure 2. Provided the vehicle speed 𝑣𝑣𝑐𝑐𝑐𝑐𝑐𝑐 , a virtual vehicle position along the track direction 𝑥𝑥𝑛𝑛 can be associated to each microphone. The A-weighted overall sound pressure level 𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑛𝑛 can be computed for each vehicle position as: 𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝐿𝐿𝑛𝑛 = 20 log

𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓

√∫𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓 𝑝𝑝𝑛𝑛1,𝐴𝐴 𝑑𝑑𝑑𝑑 𝑝𝑝𝑟𝑟𝑟𝑟𝑟𝑟

with 𝑝𝑝𝑟𝑟𝑟𝑟𝑟𝑟 the standard reference sound pressure equal to 20μPa.

(1)

The evolution of the OASPL with the displacement of the vehicle can be plot as well as the frequency content of the acoustic transfer function at each location. Results for the example case are shown on Figure 5 and give an overview of the behaviour of the system. The problematic vehicle positions can be identified precisely on the OASPL plot and the corresponding acoustic transfer function can be studied. Large scale resonance phenomena occurring under the car can also be identified as vertical lines on the waterfall diagram showing the frequency content of the signal. These indicators do not account for the Doppler Effect caused by the vehicle movement.

66

Pass-by noise synthesis from frequency domain exterior acoustic simulations

Figure 5 - Evolution of acoustic transfer function OASPL and frequency content with respect to vehicle location

During pass-by noise test, the maximum A-weighted sound pressure level is measured using the "fast" response of a precision sound-level meter according to [5], with a reading made at a time interval not greater than 30ms. This time-weighted sound pressure level at an instant 𝑡𝑡 is noted 𝐿𝐿𝐴𝐴𝐴𝐴 and is expressed as 𝐿𝐿𝐴𝐴𝐴𝐴 (𝑡𝑡) = 20 log

√1 ∫𝑡𝑡 𝑝𝑝𝐴𝐴2 (𝜉𝜉 )𝑒𝑒 −(𝑡𝑡−𝜉𝜉)/𝜏𝜏 𝑑𝑑𝑑𝑑 𝜏𝜏 −∞ 𝑝𝑝𝑟𝑟𝑟𝑟𝑟𝑟

(2)

Where: - 𝜏𝜏 is the exponential time constant (equal to 125ms in “fast” mode) - 𝑝𝑝𝐴𝐴 is the A-weighted instantaneous sound pressure - 𝑝𝑝𝑟𝑟𝑟𝑟𝑟𝑟 is the standard reference sound pressure equal to 20μPa

Since a time 𝑡𝑡𝑛𝑛 can be associated to each vehicle position 𝑥𝑥𝑡𝑡 , the same post-processing can be applied to the simulation results in order to obtain a time-weighted sound pressure level consistent with the measurement method. The resulting curves shown in Figure 6 show a smoother trend and a change in maximum value. With this post-processing, the acoustic transfer functions between the different noise sources and the pass-by noise microphones can be evaluated. This can provide guidelines regarding the critical sound directivities around the car and the critical frequencies at which a given sound source would cause high sound pressure level during the pass-by noise test. Investigations on the influence of the geometry or acoustic treatments can be performed using simulations.

7

Pass-by noise synthesis from frequency domain exterior acoustic simulations

Figure 6 - Calculated time-weighted Sound Pressure Level

Signal processing for transient input signals The dissociation between the acoustic source and acoustic propagation is an interesting approach to study the pass-by noise ISO indicator in details. On a computational point of view it is convenient since it allows to compute the acoustic transfer functions of various sources in a single frequency response analysis and then recombine the sources with the acoustic transfer functions during post-processing. On the engineering point of view it is also very powerful since both source and propagation can be studied independently in details. In order to replicate accurately the ISO indicator, transient source signals must be associated to the different acoustic transfer functions. Transient data for the different acoustic sources can come from various environments: 3D structural Finite Element Analysis (FEA) results for vibrating parts such as engine or gearbox, 1D Computation Fluid Dynamic (CFD) results for the mass flow rate at exhaust tail pipe exit or measured pressure signals for other sources such as tire noise. The method used to post-process the calculation results using the transient source signal for the exhaust tail pipe exit mass flow rate is described in Figure 7. The frequency results output by Actran are used as filters for the Mass Flow Rate transient signal and the final results are assembled from all the different parts. The complete process is implemented in a Python script compatible with Actran Graphical User Interface.

8

Pass-by noise synthesis from frequency domain exterior acoustic simulations

Figure 7 - Overview of pass-by noise reconstruction method

Transient signal pre-processing and conversion in frequency domain The processing of the transient mass flow rate signal through the different frequency filters requires a conversion of the signal in the frequency domain. Before using a Discrete Fourier Transform (DFT), the signal is pre-processed in order to reduce ringing effect and to ensure that the signal is long enough to cover the time delay introduced by the acoustic propagation between the source and the microphone at the end of the signal. Ringing effect is prevented by de-trending the input signal so that the first and last values of the signal are zeros, see Figure 8. The bias introduced by this operation is limited to very low frequencies (inverse of signal length) that are not of interest in this process.

Figure 8 - Transient signal pre-processing

The frequency spectrum of the pre-processed signal 𝑄𝑄𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝𝑝 (𝑡𝑡𝑡 is computed using a DFT. The resulting spectrum 𝑄𝑄𝑄𝑄𝑄𝑄 has a frequency step corresponding to the inverse of the signal length and should be multiplied by the calculation results.

9

Pass-by noise synthesis from frequency domain exterior acoustic simulations

Acoustic transfer functions interpolation The frequency domain FEM calculation is run for a finite number of frequencies. The number of frequencies solved is directly proportional to the calculation time since each frequency is solved one after the other (considering sequential solving). There is therefore an interest in limiting the number of frequencies to compute by having a frequency step larger than the frequency step of the input signal spectrum, and obtaining the missing values for the acoustic transfer functions through an interpolation process. In order to have an accurate interpolation, it is important that the interpolated signals have variations spanning more than the computation frequency step. When looking at the real and imaginary parts of the acoustic transfer functions on Figure 9, one can observe that the variations have very important when the frequency increases.

Figure 9 - Acoustic Transfer Function as Real/Imaginary part

In order to get a function better suited for the interpolation, the acoustic transfer functions are divided by a normalization function. The goal is to obtain a smoother function easier to interpolate. Provided that the acoustic propagation problem solved by the FEM calculation consists in computing the scattering of a source by the car body, the analytical expression of the incident pressure (without any diffraction) is used as normalization function. The obtained signal (Figure 10) is smoother both for its real and imaginary part which facilitates its interpolation.

10 10

Pass-by noise synthesis from frequency domain exterior acoustic simulations

Figure 10 - Normalized Acoustic Transfer Function used for interpolation

The computed acoustic transfer functions are divided by the analytical expression of the pressure created by a unit monopole source considering the source and microphone location. The normalized function is interpolated and multiplied again by the pressure computed analytically to obtain the final interpolated signal. The interpolated acoustic transfer functions are shown in Figure 11 and can be used for further data processing.

Figure 11 - Interpolated Acoustic Transfer Function

Once interpolated, each acoustic transfer function 𝑝𝑝𝑛𝑛1,𝐴𝐴 (𝑓𝑓𝑓 is multiplied by the source spectrum 𝑄𝑄𝑄𝑄𝑄𝑄 to obtain the pressure spectrum at the corresponding microphone 𝑝𝑝𝑛𝑛𝐴𝐴 (𝑓𝑓𝑓. The pressure spectrum at each microphone is then converted back to the time domain using an Inverse Discrete Fourier Transform (IDFT). The obtained signals correspond to the time domain A-weighted pressure at each microphone 𝑝𝑝𝑛𝑛𝐴𝐴 (𝑡𝑡) when the car is not moving. Normalization coefficients for the DFT and the IDFT are chosen carefully to ensure that for any signal 𝜑𝜑𝜑𝜑𝜑𝜑, the reciprocity property is verified: 𝜑𝜑(𝑡𝑡) = 𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼(𝜑𝜑(𝑡𝑡))).

11

Pass-by noise synthesis from frequency domain exterior acoustic simulations

Pass-by noise ISO indicator reconstruction Starting from 𝑝𝑝𝑛𝑛𝐴𝐴 (𝑡𝑡), the Pass-by noise ISO indicator 𝑝𝑝 𝐴𝐴 (𝑡𝑡) recorded by a moving microphone is reconstructed by assembling contributions from the different static microphones as depicted in Figure 12.

Figure 12 - Pass-by noise microphone pressure reconstruction

The pass-by noise ISO signal 𝑝𝑝 𝐴𝐴 (𝑡𝑡) is recomputed from the different pressure signal 𝑝𝑝𝑛𝑛𝐴𝐴 (𝑡𝑡) computed for each microphone 𝑛𝑛 using a linear interpolation in their respective range of validity. 𝑝𝑝 𝐴𝐴 (𝑡𝑡) =

𝑡𝑡𝑛𝑛𝑛𝑛 − 𝑡𝑡 𝐴𝐴 𝑡𝑡 𝑡 𝑡𝑡𝑛𝑛 𝑝𝑝 (𝑡𝑡) + 𝑝𝑝 𝐴𝐴 (𝑡𝑡) 𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑛𝑛 , 𝑡𝑡𝑛𝑛𝑛𝑛 [ 𝑡𝑡𝑛𝑛𝑛𝑛 − 𝑡𝑡𝑛𝑛 𝑛𝑛𝑛𝑛 𝑡𝑡𝑛𝑛𝑛𝑛 − 𝑡𝑡𝑛𝑛 𝑛𝑛

Where 𝑡𝑡𝑛𝑛 is the time at which the moving microphone is at the same position as microphone 𝑛𝑛. In between each discrete microphone location, the pressure is obtained through linear interpolation of the two nearest microphone signals. The microphones spacing should be sufficiently small to reach convergence of the solution. This corresponds to 6-8 microphones per smallest acoustic wavelength in this case. The number of microphones used could be reduced using a higher order interpolation to reconstruct the signal between the different microphones or by introducing time shifts between the interpolated signals to better account for the difference in acoustic propagation time between the source and the different microphones.

12 12

Pass-by noise synthesis from frequency domain exterior acoustic simulations

Different reconstructed signals can be summed to obtain the pass-by noise from various noise sources. If the input source signals are in-phase, the phase information is not lost during the process.

Results post-processing Once the pass-by noise ISO signal is recomputed, it can be thoroughly post-processed and studied to get directions for the improvements of problematic noise sources. Starting from the recomputed signal 𝑝𝑝 𝐴𝐴 (𝑡𝑡) the time-weighted sound pressure level 𝐿𝐿𝐴𝐴𝐴𝐴 (𝑡𝑡) can be computed from (2). 𝐿𝐿𝐴𝐴𝐴𝐴 (𝑡𝑡) corresponds to the sound pressure level as measured during pass-by noise test which means that the maximum level from the curve is the indicator used for pass-by noise assessment.

Figure 13 - Pass-by Noise ISO signal and frequency content

The frequency content of the signal can also be visualize to identify problematic frequencies or engine rotation speed, or to get the sound pressure level for each engine order.

Conclusions This paper has presented an efficient method for the prediction of the acoustic propagation of noise source in the framework of vehicle pass-by noise. A single frequency response finite element model is used to compute the acoustic transfer functions between the different vehicle sources and the pass-by noise ISO microphone. The reconstruction of the pass-by noise signal from the different frequency domain acoustic transfer functions is achieved through a Python script with a method adapted from indoor pass-by noise measurement method. Provided input noise source signals originating from measurements or prior calculations are filtered by the different acoustic transfer functions in the frequency domain and then

13 13

Pass-by noise synthesis from frequency domain exterior acoustic simulations

assembled in the time domain using a method accounting for the Doppler Effect. The reconstructed signal reproduces the pass-by noise ISO indicator and can therefore be used to assess the absolute noise level produced by different designs. The complete simulation and post-processing process is carried out in the Actran acoustic simulation environment. Efficient methods such as h-adaptivity are used to run the calculation. Acoustic transfer functions for all noise sources around a complete car are computed up to 2000Hz in approximately 10 hours using a high performance computing node.

References [1] “Regulation (EU) No 540/2014 of the European Parliament and of the Council of 16 April 2014 on the Sound Level of Motor Vehicles,” Official Journal of the European Union, 27 May 2014. [2] “Measurement of Noise Emitted by Accelerating Road Vehicles,” Engineering Method - Part 1: M and N Categories, ISO 362-1:2007. [3] A. Schuhmacher, Y. Shirahashi , M. Hirayama and Y. Ryu, “Indoor pass-by noise contribution analysis using source path contribution concept,” ISMA2012-USD2012, 2012. [4] “2012 Toyota Camry Detailed Finite Element Model Version 5a,” Center for Collision Safety and Analysis & George Mason University, June 2016. [5] “IEC 61672-1:2002: Precision sound level meters,” International Electrotechnical Commission, 2002.

14 14

An exploration study of automotive sound package performance in the mid-frequency range Nicolas Schaefer, Bart Bergen, Tomas Keppens, Toyota Motor Europe, Vehicle Performance Engineering; Prof. Wim Desmet, KU Leuven, Dept. of Mechanical Engineering, Member of Flanders Make.

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 W. Siebenpfeiffer (Hrsg.), Automotive Acoustics Conference 2017, Proceedings, https://doi.org/10.1007/978-3-658-20251-4_14

1

An exploration study of automotive sound package performance in the mid-frequency …

Abstract The different sound packages surrounding the interior cavity have a significant contribution to the interior noise performance of a car. The design of the trim for the highfrequency range is well established. However, its design for the mid-frequency range (100-1000 Hz) is more difficult, because of the complex inputs involved, the lack of representative performance metrics and the high computational cost of the simulations. In order to make early decisions for trim design, performance maps, describing the performance metrics of a silencer against the different design variables, are desired for the mid-frequency range. A framework has been developed to retrieve those maps from an exploration of designs1: it constructs those maps based on a surrogate model built on a Gaussian process regression. This surrogate model is iteratively enriched with new designs to increase the accuracy of that model. Each new design consists of a finite element simulation of a sound package in mid-frequencies. In order to gain computation time, it is sped up using a reduced order model which is iteratively improved, until a convergence is achieved on all the metrics of interest. This paper first summarizes the main aspects of this framework. In a second part, it presents an application of that framework on two typical automotive trim cases under a structure-borne excitation. Both material and geometrical parameters are considered to build performance maps in the mid-frequency range.

1 Schaefer, N., Bergen, B., Keppens, T., Desmet, W., “A Design Space Exploration Framework for Automotive Sound Packages in the Mid-frequency Range”, SAE Technical Paper 2017-01-1751, 2017.

2

An exploration study of automotive sound package performance in the mid-frequency …

Introduction The panels surrounding the car interior cavity such as firewall, door or floor panels are of key importance to the interior noise performance of a vehicle. In order to control their contribution, those panels are covered with sound packages. They can typically comprise porous layers, a damping layer, a heavy layer, a foil or an air flow resistivity layer. In a usual V-shape development, targets are first set at vehicle level then broken down to panel level in relevant panel metrics: mass, cost, packaging space and NVH performance. The panels are then developed iteratively according to those targets and final performance is confirmed back at vehicle level. Ideally, this process should be able to: 1. Confirm the feasibility of targets early in the development 2. Understand the performance sensitivity to the design parameters, to support the detailed design 3. Confirm the final design and its dispersion Those requests can be achieved based on performance maps according to the sound package design variables. The bounds of those maps evolve through the development process. At the earliest stage (1), it is the widest possible, drawing a feasible space on which the targets can be confirmed and the technology decided. (2) As the design evolves and more bounds are set, it is possible to zoom in to see the detailed performance sensitivity (2). Finally, the map focusses until only the dispersion of the converged design is represented (3). Design exploration methodologies are a common tool to build those performance maps, especially in the case of expensive simulations or experiments. Design space exploration methodologies have already been used from a process viewpoint in many fields. Related examples include works of Briceño and Mavris2 (balancing noise, emissions and fuel consumption in a jet engine) and Bennur et al.3 (balance ride and handling and NVH performances of vehicle bushings). Related to sound package development, Dinsmore and Unglenieks4 have used quasi-Monte Carlo 2 Briceño, S. and Mavris, D., "Quiet Supersonic Jet Engine Performance Trade-off Analysis Using a Response Surface Methodology Approach," SAE Technical Paper 2002-01-2929, 2002. 3 Bennur, M., Hogland, D., Abboud, E., Wang, T. et al., "Multi-Disciplinary Robust Optimization for Performances of Noise & Vibration and Impact Hardness & Memory Shake," SAE Technical Paper 2009-01-0341, 2009. 4 Dinsmore, M. and Unglenieks, R., "Acoustical Optimization Using Quasi-Monte Carlo Methods and SEA Modeling," SAE Technical Paper 2005-01-2431, 2005.

3

An exploration study of automotive sound package performance in the mid-frequency …

methods to SEA models to understand which combinations of car components are the most effective to reduce interior noise. In this contribution, the design space exploration framework to construct the performance maps is based on Schaefer et al.¹. In its first section, this paper summarizes the framework used. In a second section, this paper illustrates the application of the framework on two typical automotive sound package cases under a structureborne excitation. Both material and geometrical parameters are considered to build a parametric response surface of the performance in the mid-frequency range.

Design space exploration framework Sound packages consist of multiple layers and can typically include a host layer (representing the body shell), a damping layer, one or two porous layers, an air flow resistivity layer or an elastic layer. This leads to many possible arrangements of stacks, with many design parameters available. In order to effectively develop the midfrequency performance of those sound packages, it is necessary to have a tool to support the decision making at different stages of the development process: 1. Early stage: confirm feasibility of targets and decide technology based on global performance map 2. Support the detailed design by showing performance sensitivity to the design parameters 3. Confirm the final design and its dispersion To meet those requests, maps of performance against the different design parameters are desired. Such performance map can be used over the different phases by selecting an appropriate ‘zoom level’. Those performance maps are constructed based on the framework developed by Schaefer et al.1. It consists in building a Kriging surrogate model of the different performances from an initial set of design points in the parameter design space. This model is then iteratively enriched with new design points to increase its accuracy. Each design point consists in a finite element simulation of a sound package in midfrequencies. In order to gain computation time, the FE simulations are run using a reduced order model based on a Krylov projection matrix-free algorithm. In this section, the two main aspects of this framework are recalled: the surrogate modelling technique and the matrix-free reduced order model.

4

An exploration study of automotive sound package performance in the mid-frequency …

Kriging surrogate model Surrogate modelling, also called metamodelling, consists in approximating the responses of computationally intensive simulations with analytical models —called surrogate models or metamodels5. In the current case, those surrogate models represent the different performance responses of interest (third-octave band performance metrics) against the different design parameters . They are fitted by simulating the expensive finite element model of the sound package in a limited number of design points. Once generated, those surrogate models are inexpensive to evaluate and allow therefore mathematical operations such as optimum search, sensitivity analysis and differentiation. The analytical formulation of the response makes it easy to zoom in on specific ranges (as the development progresses). As such, one surrogate model can be used throughout the development. Accuracy can be refined locally as needed as the design parameter ranges converge. Many surrogate modelling techniques exist6. In this present case, the Kriging surrogate model is used to build the different third-octave band performance metrics. Kriging surrogate modelling postulate that the predicted response can be written in any point of the design space as7: ( )= ( )+ ( ) Where, ( ) is a global approximation of fit the actual response .

(1) and ( ) creates “localized” deviations to

The error of the generated surrogate model is based on the evaluation of its crossvalidation. The -fold cross-validation8 consists of randomly partitioning the calculated designs in equal sized subsets. Out of the subsets, − 1 are used to build the model, and the remaining subset is kept for the validation of the model. The crossvalidation process is then repeated times (the folds), with each of the subset used exactly once as the validation data. The results from the folds are then averaged to produce a single estimation: the cross-validation measure of the model.

5 Gary Wang, G. and Shan, S., “Review of Metamodeling Techniques in Support of Engineering Design Optimization,” Journal of Mechanical design 129(4):370-380, 2007. 6 Jin, Y., “A Comprehensive Survey of Fitness Approximation in Evolutionary Computation,” Soft Computing, 9(1):3–12, 2005. 7 Simpson T.W., Mauery T.M., Korte J.J., and Mistree F., “Kriging models for global approximation in simulation-based multidisciplinary design optimization,” AIAA journal 39(12):2233-41, 2001. 8 Olson, D.L., and Delen. D., “Advanced Data Mining Techniques,” Springer Science & Business, 2008.

5

An exploration study of automotive sound package performance in the mid-frequency …

If this measure is above the desired error criterion, new design points are selected. Many strategies exist9,10. In the case of the current framework¹, the LOLA-Voronoi11 adaptive sampling technique is used. This technique combines both an exploration criterion (adding points in the regions where there is the least by using Voroni tessellation) and exploitation criteria (adding point in the least smooth regions, “LOLA”).

Matrix-free model order reduction Even with an efficient response surface algorithm such as Kriging, a sizeable number of design points will need to be computed. Each design point will consist in the simulation of sound packages performed with finite element analysis in mid-frequencies. Typically, a simulation of a thick sound package can require up to multi-million degrees of freedom and can take multiple hours on a supercomputer node. Therefore, a suitable model reduction scheme is needed to reduce calculation effort. Model Order Reduction (MOR) methods12 approximate a system (or a part of it) by projecting it on a much smaller subspace which preserves the desired information. In the framework applied in this contribution1, the matrix-free algorithm used is based on Li’s doctoral thesis13. This algorithm consists in a Krylov moment-matching method relying on forced responses. It expands the response functions ( ) into a series of moments around an expansion point. By matching the moments at different expansion points, a reduced order model (ROM) can be constructed for each response. Then, the method can build the reduced transfer response ( ) from the ROM in any frequency of interest . The number of expansion points is iteratively increased by computing

9 Lehmensiek, R, and Meyer, P., Müller, M., “Adaptive Sampling Applied to Multivariate, Multiple Output Rational Interpolation Models with Application to Microwave Circuits,” International Journal of RF and Microwave Computer‐Aided Engineering 12(4):332-340, 2002. 10 Provost, F., Jensen, D., and Oates, T., “Efficient Progressive Sampling,” Proceedings of the Fifth ACM SIGKDD International Conference on Knowledge Discovery and Data Mining 1(1):23–32, New York, 1999. 11 Crombecq, K., De Tommasi L., and Gorissen D., “A Novel Sequential Design Strategy for Global Surrogate Modeling,” Presented at Winter Simulation Conference 1(1):731-742, December 13, 2009. 12 Schilders, W.H., Van der Vorst H.A., and Rommes, J., “Model Order Reduction: Theory, Research Aspects and Applications,” (Berlin, Springer, 2008). 13 Li, X., “Power flow prediction in vibrating systems via model reduction,” Ph.D. thesis, College of Engineering, Boston University, Boston, 2004.

6

An exploration study of automotive sound package performance in the mid-frequency …

four new frequency lines at optimal locations14 until specific convergence criteria are satisfied15.

Application on two structure-borne cases In order to present the application of the above framework, two cases are introduced in a first part, in which design parameters, finite element models and used metrics are detailed. In a second part, the results of the surrogate model are staged and discussed.

Cases In this contribution, the interest is to understand the performance of the two following sound package cases: ● “Single-porous” case: A steel panel backed by a porous layer ● “Decoupler” case: A steel panel backed by porous layer and a heavy layer,

Design parameters Many design variables exist: input, geometrical and material parameters. In order to reduce the number of cases, the panel studied was kept rectangular and of constant dimensions and the structureborne input was set to be uniform. A sensitivity study on the material parameters was first performed in order to reduce the number of parameters, which are listed in table 1.

14 Jonckheere, S., Li, X., and Desmet, W., “A matrix-free Model Order Reduction scheme for vibro-acoustic problems with complex damping treatments,” Proceedings of the 2016 International Conference on Noise and Vibration Engineering 1(1):3521-3534, 2016. 15 Schaefer, N., Bergen, B., Jonckheere, S., and Desmet W., “Application of a matrix-free model order reduction scheme to automotive treated panels,” Proceedings of the 2016 International Conference on Noise and Vibration Engineering 1(1):3863-3872, 2016.

7

An exploration study of automotive sound package performance in the mid-frequency …

Table 1. List of the design parameters. Component

Parameter

Range

Steel panel Porous layer

Thickness Thickness Young’s modulus Air flow resistivity Total density Thickness Young’s modulus

0.5-1 mm 5-25 mm 4-23 kPa 6-325 krayls 40-200 kg/m³ 1-3 mm 0.01-10 GPa

Heavy layer

The parameters not listed such as the characteristic lengths, the tortuosity, the porosity, the Poisson’s ratios, the damping factors and the panel Young’s modulus, the density of the heavy layer were kept constant.

Finite Element Model The sound packages cover a rectangular surface (350 mm × 500 mm) and are excited at their boundaries by a uniform input velocity of =1 m/s. They are backed by an air layer and a perfectly matched layer (PML) which allows free-field radiation16. Those conditions allow the simulation of a specific experimental set-up17, whose goal is to assess the performance of the stacks under a structural boundary input. The frequency range of interest covers the nine third-octave bands in the mid-frequency range, from 100 Hz to 630 Hz. The models are simulated with the finite element method, with a mesh only made of quad and hexa elements; the mesh sizes are refined according to the requested frequencies. The steel plate and heavy layer are modelled with Mindlin-Reissner shells and the porous layer is described by the full Biot u-p formulation. This leads to large matrices (in the order of one million degrees of freedom at 1 kHz). In order to speed up the computation, the third-octave responses are computed with the matrix-free MOR, introduced in the previous section. Due to the symmetry of the model, only a quarter-model is studied, as depicted in figure 1. This symmetry also implies that only the odd modes will be excited—a quarter of a plate modal base. 16 Bérenger, J.P., “A Perfectly Matched Layer for the Absorption of Electromagnetic Waves,” J. Comput. Phys. 114(2):185–200, 1994. 17 Van der Kelen, C., Vivolo, M., Van Genechten, B., Pluymers, B. et al., “Validation of a dedicated test set-up for boundary excitation of trim assemblies,” Proceedings of the 2012 International Conference on Noise and Vibration Engineering 1(1):4041-4050, 2012.

8

An exploration study of automotive sound package performance in the mid-frequency …

Figure 1. Sketch of the stack modelling.

Objective Metrics In order to quantify the transmission performance in the mid-frequency range, the two and the metrics proposed in the framework1 are used: the panel velocity loss package velocity loss . =

(2)

=

(3)

Where, , = ∬| | are respectively the bottom-side and top-side integrated squared absolute velocities, | | in the current case of a uniform structureborne input = is the normal velocity, is the bottom-side surface, is the impedance of air. Those definitions link to the radiation efficiency , which quantifies the relation between the structural vibrations and the transmitted power : =



(4)

In the case of a flat rigid vibrating surface (“piston”), the radiation efficiency is = 1. In other cases, it characterizes how much less a vibrating surface radiates than a piston of the same area. Since both and will exhibit the resonances of the panel, it is expected that will contain those resonances, damped by the back-coupling effect of the package,

9

An exploration study of automotive sound package performance in the mid-frequency …

while will hide them: only the resonances along the thickness of the package are expected in .

Results and Discussion In figure 2, main effect plots are shown for the area mass and the octave-band + performances, for all the parameters and of the two cases. In order to build those plots, first, the surrogate model of each case is computed for 400 thousand design realisations, which are pseudo-randomized in the design space. Then, every parameter is regularly spaced in ten segments and the “centre of gravity” of the performance on each segment is computed. For each parameter, a curve between all the centres of gravity can be drawn, showing the median impact. Porous case 9

325000 23000

6000 4000

7 5.040 6

8

325000 23000

4000 6000

7 5.040 6

0.50

9

25.0200 area mass (kg/m2)

20025.0 area mass (kg/m2)

area mass (kg/m2)

8

1.00

1.00

1.00 9

8

325000 4000 5.0 40

0 PVL+SVL (dB), 250 Hz

0 PVL+SVL (dB), 125 Hz

Por. density (kg/m3) Porous AFR (rayls) Por. thickness (mm) HL thickness (mm) HL YM (Pa)

23000 6000

7

6

0.50

Steel thickness (mm) Porous YM (Pa)

200 25.0

0.50 0 PVL+SVL (dB), 500 Hz

Porous & heavy layer (HL) case

10 23000

325000

1e+10 4000 1e+07 6001

5.0 1.0 40

25.0

12

1.00 200 3.0

11 23000 325000 1e+07

10

9

5.0

1.0

4000 6001 1e+10

40

0.50 8

0 PVL+SVL (dB), 125 Hz

8

25.0

area mass (kg/m2)

200 3.0

11

9

12

1.00

area mass (kg/m2)

area mass (kg/m2)

12

1.00 200 3.0 25.0

11

10

9

1e+07 23000 325000 6001 1e+10 4000 5.0

1.0 40 0.50

0.50 0 PVL+SVL (dB), 250 Hz

8

0 PVL+SVL (dB), 500 Hz

Figure 2. Comparison of the mass and the octave-band SVL+PVL performances for the porous and porous + heavy layer cases. The same x-axis ranges are used for each pair of graph to allow relative comparison.

10

An exploration study of automotive sound package performance in the mid-frequency …

This type of graph allows: ● The relative comparison of different parameters on two performances in the same time. For instance, it can be seen from the bottom three graphs in figure 2 that it is better to increase the thickness of the porous rather than the thickness of the heavy layer (HL) in the case of porous + heavy layer. ● Checking the linearity of the impacting parameter: while the steel thickness increases linearly the mass performance, it has a non-linear impact on the PVL+SVL performance: the projection of the arrow heads on the x-axis is not uniform. ● Obtaining the slopes (dB or kg/m² by unit of parameter) in order to give design recommendations at early stage.

Impact of thicknesses and porous density Increasing packaging mass mainly increases performance. In the absence of top layer, whether it is the density or the thickness of the porous layer which is increased, it will impact in a similar manner the mass and the SBN transmission performance: the performance is driven by the mass. In the presence of a heavy layer, there is a clear ranking between the different parameters on the slope of mass vs. performance: ● The thickness of the porous layer is the “lightest” solution: it increases the performance by lowering the spring stiffness of the decoupling layer. However, this solution can be quickly limited by the available packaging space. ● The thickness of the heavy layer is the second “lightest” solution: it improves the performance by increasing the mass of the mass-spring system. Though this solution is not taking much space, it is not as efficient as the porous thickness, especially in the lower frequencies. ● The density of the porous layer is the third solution: by increasing the total mass of the package, it increases the performance without impacting the packaging space. For the two cases, the thickness of the steel has mostly a negative impact on the + performance, which can be explained because of the combination of the two following causes: ● It moves the modal density to higher frequency while increasing mass. This can be clearly seen in the lower frequency bands for the porous case: by increasing the thickness, some modes are moving to higher frequencies, entering or leaving the frequency band. As a single porous package has small low-frequency insulation, it results in the appearance of an “S-shape” on the performance. ● The steel panel is excited with a constant velocity and the performance metric used: +

=

| |

11

An exploration study of automotive sound package performance in the mid-frequency …

which will not take into account the increase of this thickness on the input side while it will have a negative impact on the passive side .

Impact of Young’s moduli and porous air flow resistivity In low-frequency, the most significant parameter is the porous Young’s modulus for both cases: a lower spring is better for insulation performance, which is according to expectation in the case of a decoupler solution. When moving to higher frequencies, the impact of Young’s modulus lowers. The impact of the AFR follows the opposite trend: its impact in the lower frequencies is minimal and increases in higher frequencies. The stiffness of the heavy layer does not impact much the performance.

Conclusions A framework to retrieve the response surface has been summarized. This framework allows, from a limited number of explicit design points, the creation of multidimensional performance maps according to the different design parameters. In a second section, this framework was applied on two typical automotive application cases: a mono-porous sound package and a decoupler solution. The obtained results confirm the expectations from the literature and also allow further understanding by giving: ● the ranking of the different effects, ● the ranges of impact of each parameter, ● the slopes of performance improvement vs. mass. This framework strengthens the V-shape development of the vehicle in the midfrequency range by giving a multi-scale tool for performance prediction.

Acknowledgements The research of Nicolas Schaefer is funded by a grant from the IWT Flanders. The computational resources and services used in this work were provided by the VSC (Flemish Supercomputer Center), funded by the Research Foundation – Flanders (FWO) and the Flemish Government – department EWI. The authors would like to gratefully acknowledge Norimasa Kobayashi and Masashi Komada from Toyota Motor Corporation for their support.

12

The noise reduction potential of lightweight acoustic metamaterials – a numerical and experimental study Dipl.-Ing. Peter Schrader Research Assistant Otto-von-Guericke-University Magdeburg Dipl.-Ing. Fabian Duvigneau Research Assistant Otto-von-Guericke-University Magdeburg Prof. Dr.-Ing. Hermann Rottengruber Otto-von-Guericke-University Magdeburg Prof. Dr.-Ing. habil. Dr. h. c. Ulrich Gabbert Otto-von-Guericke-University Magdeburg

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 W. Siebenpfeiffer (Hrsg.), Automotive Acoustics Conference 2017, Proceedings, https://doi.org/10.1007/978-3-658-20251-4_15

1

The noise reduction potential of lightweight acoustic metamaterials – a numerical …

Introduction Noise reduction of passenger car engines contributes to an improvement of the driving comfort at low and medium speeds. It also helps to meet the acoustic regulations of the legislative authorities that force lower sound pressure levels of the pass-by-noise of cars in the next years. An option to reduce the noise radiated from the engine is the application of a full or a partial engine encapsulation. Such an encapsulation has to cause a sufficient sound pressure level (SPL) reduction especially for engine concepts with high cylinder pressure gradients. The challenge of any encapsulation development is to obtain a satisfying noise reduction with an additional mass being as low as possible. In the paper material concepts are examined that combine a low density with a better noise absorption than conventional damping materials especially at low frequencies. The first investigated concept improves the energy dissipation within the material by mass inclusions in Polyurethane (PUR) foams, where the inclusions are acting as local massspring-damper system. Investigations in the literature have shown that such foam-masscompounds can change and improve the noise absorption behavior at frequencies lower than 1000 Hz. In the paper, the acoustic effects are determined for configurations that differ in weight, size and the position of the inclusions within the PUR foam. Cellular material configurations using plastic honeycomb structures are investigated as well. The second investigated concept consists of PUR foams with different cavities at the surface of the foam. The cavities are first uncovered and then covered by one or more microperforated membranes. Here, the sound reduction is gained by antiphase membrane oscillations as well as by acoustic shortcuts at the perforations. The acoustical effect of each of the mentioned methods is determined and evaluated by measurements and simulations. The acoustic effects of the different metamaterials are compared with conventional materials.

Numerical analysis In this section the numerical investigations are explained, the used models are presented and some representative results are discussed. The Finite Element Method (FEM) is used for the numerical simulations in this paper.

Numerical model In Fig. 1 the FE-model of the vibrating structure is shown. It contains a rectangular aluminum plate (440 mm x 240 mm x 5 mm) with an attached foam layer (400 mm x 200

2 2

The noise reduction potential of lightweight acoustic metamaterials – a numerical …

mm x 40 mm). For both materials the classical structural damping is used. The plate’s displacements are blocked in all three directions at the four corners of its backside. For introducing the excitation force an aluminum cylinder is attached in the middle of the backside of the plate in analogy to the experimental setup. White noise is used to excite the whole frequency range. The numerical analysis is executed in the frequency domain, as in a related investigation of the authors [2].

Fig. 1: FE-model of the rectangular plate with foam

Fig. 2: FE-model of the surrounding fluid

33

The noise reduction potential of lightweight acoustic metamaterials – a numerical …

InInFig. Fig.22the theFE-model FE-modelofofthe thesurrounding surroundingair airvolume volumewith withaaradius radiusofof330 330mm mmisisshown. shown. The Thefluid fluidisismodeled modeledonly onlyasasaahalf halfsphere spheretotoreduce reducethe thecomputational computationaleffort. effort.Both Boththe the structural structuraland andthe theacoustical acousticaldomain domainare arediscretized discretizedby bytetrahedral tetrahedralelements elementswith withquadratic quadratic shape shapefunctions. functions.The Themaximum maximumedge edgelength lengthofofthe thestructural structuralmodel modelisis88mm. mm.At Atthe thepeperiphery ripheryofofthe thesphere spherethe themaximum maximumedge edgelength lengthisis30 30mm. mm.The Theacoustic acousticproblem problemisiscalcucalculated latedunder underfree-field free-fieldconditions. conditions.For Forthis thisreason, reason,all allboundaries boundariesofofthe theair airvolume volumeare are modeled modeledasasabsorbing absorbingboundaries boundariesexcept exceptfor forthe theinterface interfacebetween betweenstructure structureand andair. air.At At this thisinterface interfacethe thesurrounding surroundingair airvolume volumeisisexcited excitedby bythe thesurface surfacevelocities velocitiesofofthe thevibratvibrating ingstructure, structure,which whichare arecalculated calculatedininadvance. advance.Coincident Coincidentmeshes meshesare areused usedfor forboth bothdodomains mainsatatthe theinterface. interface.To Toreduce reducethe thenumerical numericaleffort effortan anuncoupled uncoupledacoustic acousticsimulation simulation isisused, used,that thatmeans meansthat thatthe thefeedback feedbackofofthe thevibrating vibratingair airtotothe thestructure structureisisneglected. neglected.The The geometrical geometricaldimensions, dimensions,mesh meshparameters, parameters,boundary boundaryand andloading loadingconditions conditionsare arethe thesame same for forall allinvestigated investigatedconfigurations. configurations.InInthis thisway wayaagood goodcomparability comparabilityisisgiven, given,asasonly onlythe the foam foamchanges changesinside insidedue duetotothe theinclusions. inclusions.InInthis thisstudy studyonly onlyspherical sphericalinclusions inclusionsare areininvestigated vestigatednumerically. numerically.They Theydiffer differininvolume, volume,density, density,Young’s Young’smodulus, modulus,position positionand and number. number. Table Table1:1:Material Materialdata dataofofthe thenumerical numericalmodel model

Young’smodulus modulus Young’s Density Density Poissonratio ratio Poisson Structuraldamping damping Structural

aluminium aluminium 70000N/mm² N/mm² 70000 2,7g/cm³ g/cm³ 2,7 0,3 0,3 0,01 0,01

foam foam N/mm² 55N/mm² 0,05g/cm³ g/cm³ 0,05 0,3 0,3 0,05 0,05

inclusions inclusions 5000N/mm² N/mm² 5000 0,5g/cm³ g/cm³ 0,5 0,3 0,3 0,0 0,0

InInTab. Tab.11the thematerial materialdata dataofofthe thealuminum, aluminum,foam foamand andinclusions inclusionsare arelisted, listed,which whichare are used usedwithin withinthe thenumerical numericalsimulations. simulations.

Numerical Numericalresults results InInthis thissubsection subsectionthe theresults resultsofofthe thenumerical numericalstudy studyare arepresented. presented.The TheFig. Fig.33shows showssix six different differentconfigurations configurationsthat thatare areinvestigated. investigated.All Allthese theseconfigurations configurationscontain contain50 50spherispherical calinclusions inclusionswith withaadiameter diameterofof15 15mm mmand anddiffer differonly onlyininthe theposition positionofofthe theinclusions. inclusions. One Onemajor majorquestion questionwas, was,whether whetherthe thepositioning positioningininthe theheight height(z-direction) (z-direction)has hasaalarger larger influence influencethan thanthe thepositioning positioningininthe thex-y-plane. x-y-plane.This Thiswas wassupposed supposeddue duetotothe theexperiences experiences ofofaaprevious previousstudy study[1]. [1].The Theconfiguration configuration(a) (a)isisused usedasasreference referenceconfiguration configurationfor forall all numerical numericalstudies studiesininthis thispaper. paper.InIncomparison comparisontotothe thefoam foamwithout withoutinclusions inclusionsconfiguraconfiguration tion(a) (a)causes causesaareduction reductionofofthe theresulting resultingsum sumlevel levelofofthe theA-weighted A-weightedsound soundpower powerofof 55dB(A). dB(A).The Thesound soundpower powerlevel levelisiscalculated calculatedon onthe thespherical sphericalsurface surfaceofofthe theair airvolume. volume. InInsome somevery verysmall smallfrequency frequencybands bandsthe theamplitudes amplitudesofofthe thesound soundpower powerlevel levelare areincreased increased by byintroducing introducingthe theinclusions, inclusions,but butiningeneral generalaareduction reductionofofthe thesound soundpower powerlevel levelwas was

444

The noise reduction potential of lightweight acoustic metamaterials – a numerical …

observable in in the the whole whole frequency frequency domain domain (mostly (mostly less less than than 55 dB(A)). dB(A)). Between Between observable observable the whole frequency domain (mostly less than dB(A)). Between and kHz appears appears the largest reduction of more more than less 10less dB(A). This frequency band observable thewhole whole frequency domain (mostly than 55 dB(A)). Between inininthe frequency domain (mostly than 5This dB(A)). Between 22observable and 33 kHz the largest reduction of than 10 dB(A). frequency band 2and 3the appears the largest reduction of more than 10 dB(A). This frequency band 2and and 3kHz kHz appears the largest reduction of more than 10 dB(A). This frequency band contains highest sound power levels of the whole frequency range and is consequently 2 3 kHz appears the largest reduction of more than 10 dB(A). This frequency band contains the highest sound power levels of the whole frequency range and is consequently contains the highest sound power levels of the whole frequency range and consequently contains the highest sound power levels of the whole frequency range and isisconsequently the most critical frequency range of the configuration with foam without inclusions. The contains the highest sound power levels of the whole frequency range and is consequently the most critical frequency range of the configuration with foam without inclusions. The the most critical frequency range of configuration with foam without inclusions. The the most critical frequency range of with foam without inclusions. The Fig. shows the calculated calculated sound power levels of the thewith configurations ofinclusions. Fig. andThe conthe critical frequency range ofpower thethe configuration foam without Fig. 44most shows the sound levels of configurations of Fig. 33 and conFig. 4 shows the calculated sound power levels the configurations of Fig. 3 and conFig. 4 shows the calculated sound of the configurations of Fig. 3 and conFig. 4 shows the calculated sound power levels of the configurations of Fig. 3 and contains the the explanation explanation of of these these configurations configurations in in its its legend. legend. Additionally, Additionally, the the sum sum level level of of tains tains the explanation of these configurations legend. Additionally, the sum level tains the explanation of these configurations legend. Additionally, sum tains the explanation these configurations in in its its legend. Additionally, sum level of 4ofof the A-weighted soundof power given in the the legend. legend. The comparison ofthe (a)the (c) inlevel Fig. the A-weighted sound power isis given in The comparison of (a) -- (c) in Fig. 4 the A-weighted sound power is The comparison of(a) -(c) (c) Fig. the A-weighted sound power is given given legend. The comparison -or inin the A-weighted sound power ismiddle given in in thethe legend. The comparison of of (a) -(a) (c) in 4the44 shows that for for this this example position better than position close farFig. toFig. shows that example aa middle position isis better than aa position close or far to the shows that for this example athat position is better better than position close orfar farthe tothe the shows that for this example a middle middle than aa position close or to shows that for this a middle isdistributions better than a (d) position or far to plate. However, isexample obvious theposition random (f) close lead to lower sum plate. However, itit is obvious that the random distributions (d) –– (f) lead to lower sum plate. However, obvious that random distributions (d) (f) lead tolower lower sum plate.However, However, obvious that distributions –– (f) lead sum plate. ititit isisis obvious that thethe random distributions (d)(d) – (f) lead to to lower sum

Fig. 3: Investigated configurations (top and and side side view viewof of thesphere sphereposition positionininthe thefoam) foam) Fig. Investigated configurations (top Fig. 3:3:Investigated configurations (top andand of the thethe sphere position ininthe Fig. 3: Investigated configurations (top side view of sphere position thefoam) foam) Fig. 3: Investigated configurations (top and side view of the sphere position in the foam)

sound soundpower power[dB(A)] [dB(A)] sound power [dB(A)] sound power [dB(A)] sound power [dB(A)]

140 140140 140 140 130 130130 130 130 120 120120 120 120 110 110110 110 110 100 100 100 100 100 90 90 90 90 90 80 80 80 80 80 70 70 70 70 6070 60 60 60 5060 50 50 50 4050 40 40 40 0 0 0 40 0 0

foam without spheres, sum 143.30 dB(A) foam without spheres, sum level level 143.30 dB(A) foam without spheres, level 143.30 dB(A) foam without spheres, sumsum level 143.30 dB(A) (a)foam 5x10 0.5 height, 15 mm diameter, sum level 138.28 dB(A) (a) 5x10 spheres, 0.50.5 height, 15 mm diameter, sum level 138.28 dB(A) (a) 5x10 spheres, height, 15 mm diameter, sum level 138.28 dB(A) without spheres, sum level 143.30 dB(A) (a) 5x10 spheres, 0.5 height, 1515 mm diameter, sum level 138.28 dB(A) (b)(a) 5x10 spheres, 0.25 height, mm diameter, sum level 139.87 dB(A) (b) 5x10 spheres, 0.25 height, 15 mm diameter, sum level 139.87 dB(A) (b)5x10 5x10 spheres, 0.25 height, mm diameter, sum level 139.87 dB(A) spheres, 0.5 height, 1515 mm diameter, sum level 138.28 dB(A) (b) 5x10 spheres, 0.25 height, 15 mm diameter, sum level 139.87 dB(A) (c) (c) 5x10 spheres, 0.75 height, 15 diameter, sum level 138.96 dB(A) (c) 5x10 spheres, 0.75 height, 15mm mm diameter, sum level 138.96 dB(A) 5x10 spheres, 0.75 height, 15 mm diameter, sum level 138.96 dB(A) (b) 5x10 spheres, 0.25 15 mm diameter, sum level 139.87 dB(A) (c) spheres, 0.75 height, 15 mm diameter, sum sum level 138.96 dB(A) (d)5x10 5x10 spheres, 15 mm diameter, random ininplane, level 136.94 dB(A) (d) 5x10 spheres, 150.75 mm diameter, random plane, sum level 136.94 dB(A) (d) 5x10 spheres, 15 mm diameter, random in plane, sum level 136.94 dB(A) (c) 5x10 spheres, height, 15 mm diameter, sum level 138.96 dB(A) (d) spheres, 15 mm diameter, random in plane, sum level 136.94 dB(A) (e)5x10 5x10 spheres, diameter, random in height sum level 136.50 dB(A) (e) 5x10 spheres, 15 mm diameter, random in height sum level 136.50 dB(A) (e) 5x10 5x10 spheres, spheres, 15 mm diameter, random in height sum level 136.50 dB(A) (d) diameter, random in plane, sum level 136.94 dB(A) (e) spheres, 15 diameter, random in height sum level 136.50 dB(A) (f)5x10 5x10 spheres, mm diameter, random distribution, sum level 136.47 dB(A) (f) 5x10 spheres, 15 mm diameter, random distribution, sum level 136.47 dB(A) (f) 5x10 spheres, 15 mm diameter, random distribution, sum level 136.47 dB(A) 5x10 diameter, random in height sum sum level level 136.47 136.50 dB(A) dB(A) (f) (e) 5x10 spheres, 15 mm diameter, random distribution, (f) 2000 5x10 spheres, 15 mm diameter, random distribution, sum level 136.47 dB(A) 1000 3000 4000 5000 6000 7000 8000 1000 2000 3000 4000 5000 6000 7000 8000 1000 2000 3000 4000 5000 6000 7000 8000 1000 2000 3000 frequency 4000 [Hz] 5000 6000 7000 8000 1000 2000 3000 frequency 4000 6000 7000 8000 [Hz] frequency [Hz] 5000 frequency [Hz] Fig. 4: Sound power levels of the configurationsfrequency of Fig. 3[Hz]

4: Sound power levels configurations Fig. Fig.Fig. 4: Sound power levels of of thethe configurations ofof Fig. 33 Fig. 4: Sound power levels of the configurations of Fig. 3 Fig. 4: Sound power levels of the configurations of Fig. 3

5 55 5 5

The noise reduction potential of lightweight acoustic metamaterials – a numerical …

levels. Moreover, Moreover, configuration configuration (e) (e) is better than than (d) (d) and and almost almost the the same same as as (f). (f). This This levels. levels. Moreover, configuration (e) isis better better than (d) emphasizes the thesis that the distribution in the height-direction seems to be more imemphasizes the thesis that the distribution in the height-direction seems to be more imemphasizes the thesis that the distribution in the height-direction portant than the distribution in the plane. For production purposes it is advantageous, ifaa portant than the distribution in the plane. For production purposes it is advantageous, if portant than the distribution in the plane. For production definedrandom randomdistribution distributionhas hasto tobe beproduced producedonly onlyin inone onedirection. direction.In Inadditional, additional,ititisis defined defined random distribution has to be produced only in remarkablethat thatall allconfigurations configurationsshow showaaavery verysimilar similarbehavior behaviorbelow below11kHz. kHz.Only Onlyininthe the remarkable remarkable that all configurations show very similar frequencyrange range1-5 1-5kHz kHzare aresignificantly significantlydifferences differencesvisible, visible,above above66kHz kHzthe thedifferent different frequency frequency range 1-5 kHz are significantly differences curvescome comecloser closerand andcloser closeragain. again. curves curves come closer and closer again. InFig. Fig.555the theconfiguration configuration(a) (a)of ofFig. Fig.333is shownfor fordifferent differentvariations variationsof ofthe thedensity density In In Fig. the configuration (a) of Fig. isisshown shown for and the Young’s modulus. At first, it can be noted, that the Young’s modulus of theininand the Young’s modulus. At first, it can be noted, that the Young’s modulus of the and the Young’s modulus. At first, it can be noted, that clusionshas hasno nosignificant significantinfluence, influence,as asthe thestiffness stiffnessdifference differencebetween betweenthe thefoam foamand andthe the clusions clusions has no significant influence, as the stiffness difference inclusionsisis isso solarge largethat thatthe theinclusions inclusionsact actlike likerigid rigidbodies, bodies,even evenififthe theYoung’s Young’smodulus modulus inclusions inclusions so large that the inclusions act like rigid bodies, iscomparably comparablysmall. small.In Ingeneral, general,itititcan canbe beobserved observedthat thathigher higherdensities densitieshave haveno noinfluence influence isis comparably small. In general, can be observed that forfrequencies frequencieshigher higherthan than4.5 4.5kHz, kHz,only onlythe themuch muchlower lowerdensity densityshows showsdifferences differencesinin for for frequencies higher than 4.5 kHz, only the much lower thewhole wholefrequency frequencyrange. range.In Inthe theinvestigated investigatedexample, example,the thelowest lowestdensity densityleads leadstotothe the the the whole frequency range. In the investigated example, worstresult, result,but butalso alsothe thehigher higherdensities densitiesare areworse worsethan thanthe thereference referenceconfiguration. configuration.This This worst worst result, but also the higher densities are worse than iscaused causedby bythe thefact factthat thatthe theinclusions inclusionswith withthe thehigher higherdensities densitiesincrease increasesome someof ofthe the isis caused by the fact that the inclusions with the higher peaksin inthe thecritical criticalfrequency frequencyrange. range.This Thisresults resultsin inaaahigher highersum sumlevel, level,even evenififthe theamampeaks peaks in the critical frequency range. This results in plitudes of of other other frequency frequency ranges ranges are are reduced reduced significantly significantly (for (for example: example: the the sound sound plitudes plitudes of other frequency ranges are reduced significantly powerlevel levelisis isreduced reducedup upto to35 35dB(A) dB(A)between between1.2-2.2 1.2-2.2kHz kHzby bythe theconfiguration configurationwith with power power level reduced up to 35 dB(A) between 1.2-2.2 ten-timesof ofthe thedensity). density). ten-times ten-times of the density). 140 140 140 130 130 130 120 120 120 sound sound power power [dB(A)] [dB(A)]

110 110 110 100 100 100 90 90 90 80 80 80

foamwithout withoutspheres, spheres,sum sumlevel level143.30 143.30dB(A) dB(A) foam without spheres, sum level 143.30 dB(A) foam 5x10spheres, spheres,15 15mm mmdiameter, diameter,sum sumlevel level138.28 138.28dB(A) dB(A) 5x10 spheres, 15 mm diameter, sum level 138.28 5x10 5x10spheres, spheres,15 15mm mmdiameter, diameter,0.2 0.2xxxdensity, density,sum sumlevel level146.53 146.53dB(A) dB(A) 5x10 spheres, 15 mm diameter, 0.2 density, 5x10 5x10spheres, spheres,15 15mm mmdiameter, diameter,2.0 2.0xxxdensity, density,sum sumlevel level138.99 138.99dB(A) dB(A) 5x10 spheres, 15 mm diameter, 2.0 density, 5x10 5x10spheres, spheres,15 15mm mmdiameter, diameter,10.0 10.0xxxdensity, density,sum sumlevel level141.20 141.20dB(A) dB(A) 5x10 spheres, 15 mm diameter, 10.0 density, 5x10 5x10 spheres, 15 mm diameter, E 500, sum level 138.29 dB(A) 5x10 spheres, 15 mm diameter, E 500, sum level 138.29 dB(A) 5x10 spheres, 15 mm diameter, E 500, sum level 5x10spheres, spheres,15 15mm mmdiameter, diameter,EEE50.000, 50.000,sum sumlevel level138.28 138.28dB(A) dB(A) 5x10 spheres, 15 mm diameter, 50.000, sum 5x10

70 70 70 60 60 60 50 50 50 40 40 40 000

1000 1000 1000

2000 2000 2000

3000 3000 3000

4000 5000 4000 5000 4000 5000 frequency[Hz] [Hz] frequency [Hz] frequency

6000 6000

7000 7000

8000 8000

Fig. 5: Sound power levels of config. (a) of Fig. for density and Fig.5: 5:Sound Soundpower powerlevels levelsof ofconfig. config.(a) (a)of ofFig. Fig.333for fordensity densityand andYoung’s Young’smodulus modulusvariations variations Fig.

6666

The noise reduction potential of lightweight acoustic metamaterials – a numerical …

In In Inthe the thenext next nextstep, step, step,the the thedensity-volume-ratio density-volume-ratio density-volume-ratioisis isfixed fixed fixedfor for forthe the thefive five fiveconfigurations configurations configurationsthat that thatare are are shown shown shownin in inFig. Fig. Fig.6, 6, 6,that that thatmeans means meansthat that thatall all allconfigurations configurations configurationshave have havethe the thesame same sameadditional additional additionalmass mass massof of of 39.6 39.6 39.6g. g. g.The The Theaim aim aimisis isto to toanswer answer answerthe the thequestion, question, question,whether whether whetheronly only onlythe the theadded added addedmass mass masshas has hasan an aninfluence influence influence or or orthe the thevolume, volume, volume,too. too. too.AA Adifferent different differentsphere sphere spherevolume volume volumeleads leads leadsto to toaaadifferent different differentspring spring springstiffness, stiffness, stiffness,ifififthe the the inclusion inclusion inclusionand and andthe the thefoam foam foamunder under underthe the theinclusion inclusion inclusionare are areunderstood understood understoodas as asspring-mass-system. spring-mass-system. spring-mass-system.The The The equation equation equationin in inFig. Fig. Fig.666was was wasused used usedto to tocalculate calculate calculatethe the theresulting resulting resultingdensity density densityfor for forthe the thegiven given givenmass mass massand and andaaa defined defined definedradius. radius. radius.Fig. Fig. Fig.777shows shows showsthe the theresulting resulting resultingA-weighted A-weighted A-weightedsound sound soundpower power powerlevels. levels. levels.ItItItcan can canbe be beseen seen seen that that that the the the qualitative qualitative qualitative behavior behavior behavior isisis similar, similar, similar, but but but there there there are are are significant significant significant differences. differences. differences. Larger Larger Larger

Fig. Fig. Fig.6: 6: 6:Other Other Otherinvestigated investigated investigatedconfigurations configurations configurations(top (top (topand and andside side sideview view viewof of ofthe the thesphere sphere sphereposition position positionin in inthe the thefoam) foam) foam) 140 140 140 130 130 130 120 120 120 sound power power [dB(A)] sound sound power [dB(A)] [dB(A)]

110 110 110 100 100 100 90 90 90 80 80 80 foamwithout withoutspheres, spheres,sum sumlevel level143.30 143.30dB(A) dB(A) foam foam without spheres, sum level 143.30 dB(A) (a)5x10 5x10spheres, spheres,555mm mmdiameter, diameter,sum sumlevel level137.97 137.97dB(A) dB(A) (a) (a) 5x10 spheres, mm diameter, sum level 137.97 dB(A) (b)5x10 5x10spheres, spheres,10 10mm mmdiameter, diameter,sum sumlevel level137.44 137.44dB(A) dB(A) (b) (b) 5x10 spheres, 10 mm diameter, sum level 137.44 dB(A) (c)5x10 5x10spheres, spheres,15 15mm mmdiameter, diameter,sum sumlevel level138.28 138.28dB(A) dB(A) (c) (c) 5x10 spheres, 15 mm diameter, sum level 138.28 dB(A) (d)5x10 5x10spheres, spheres,20 20mm mmdiameter, diameter,sum sumlevel level138.57 138.57dB(A) dB(A) (d) (d) 5x10 spheres, 20 mm diameter, sum level 138.57 dB(A) (e)5x10 5x10spheres, spheres,25 25mm mmdiameter, diameter,sum sumlevel level140.36 140.36dB(A) dB(A) (e) (e) 5x10 spheres, 25 mm diameter, sum level 140.36 dB(A)

70 70 70 60 60 60 50 50 50 40 40 40 000

1000 1000 1000

2000 2000 2000

3000 3000 3000

4000 5000 4000 4000 5000 5000 frequency[Hz] [Hz] frequency frequency [Hz]

6000 6000 6000

7000 7000 7000

8000 8000 8000

Fig. Fig. Fig.7: 7: 7:Sound Sound Soundpower power powerlevels levels levelsof of ofthe the theconfigurations configurations configurationsof of ofFig. Fig. Fig.666with with withconstant constant constantdensity-volume-ratio density-volume-ratio density-volume-ratio

7777

The noise reduction potential of lightweight acoustic metamaterials – a numerical …

spheres seems to be disadvantageously, as they reduce the volume of the absorbing foam significantly and they lead to a stiffer spring, which increases the eigenfrequency of the single spring-mass-system. The spheres with a diameter of 10 mm are the best configuration in Fig. 7. Fig. 8 shows the results of the configurations of Fig. 6 for the case that all configurations have the density of the reference configuration (Fig. 6(c)). Consequently, the added mass 140 130

sound power [dB(A)]

120 110 100 90 80 foam without spheres, sum level 143.30 dB(A) (a) 5x10 spheres, 5 mm diameter, 1.47 g, sum level 145.71 dB(A) (b) 5x10 spheres, 10 mm diameter, 11.7 g, sum level 145.56 dB(A) (c) 5x10 spheres, 15 mm diameter, 39.6 g, sum level 138.28 dB(A) (d) 5x10 spheres, 20 mm diameter, 93.9 g, sum level 142.15 dB(A) (e) 5x10 spheres, 25 mm diameter, 183.3 g, sum level 142.75 dB(A)

70 60 50 40 0

1000

2000

3000

4000 frequency [Hz]

5000

6000

7000

8000

Fig. 8: Sound power levels of the configurations of Fig. 6 with constant density

is different. For this reason, the added mass is also given in the legend of Fig. 8. The configurations (a) and (b) give the worst results due to the very low masses, especially in the critical frequency domain between 2.0 and 2.8 kHz. In contrast, the configurations (d) and (e) show in this frequency range reductions up to more than 20 dB(A) in comparison to the reference (c). However, they cause higher amplitudes in some other frequencies (for example 2.9, 3.3 and 4.1 kHz). This effect can be explained by the fact that the large spheres reduce the volume of the absorbing foam too much. Consequently, the energy dissipation by the foam decreases significantly. For this reason, the spheres with 15 mm diameter seem to be the best compromise for this example. All the numerical investigations show that the position, density, volume and number of inclusions have a significant influence on the resulting sound reduction. Further, it is obvious that too much additional mass respectively too large inclusions can be disadvantageous. For this reason, the mentioned parameters will be analyzed by computational optimizations in further studies to be able to evaluate the optimization potential and the limits of this type of metamaterial with respect to additional mass and achievable sound reduction.

88

The noise reduction potential of lightweight acoustic metamaterials – a numerical …

Experimental investigation In this section the experimentally investigated material concepts are introduced with the aid of the existing literature from which they are inspired. The manufacturing of prototypes of different concepts is outlined as well as the experimental test setup to determine their acoustical effect. Three material concepts are taken into account: (a) mass-springdamper-material consisting of a soft polyurethane (PUR) foam with different mass-inclusions (b) PUR foam with honeycomb structures and masses inside and (c) membrane material with one or more micro perforated membranes covering or subdividing a cavity at the surface of the material, see Fig. 9.

Introduction of the experiments and the material concepts The materials are developed for the purpose of a direct application on the surface of vibrating and sound emitting surfaces such as an oil pan, a crank case, a cylinder head. The aim of this study is to achieve high transmission losses (TL) with small masses. The TL is defined as:  = 10 log 

 ∙

  ∙

 = 10  





 + , − ,

(1)

wherein AS and ADM are the areas of the sound-emitting surface and of a damping material applied on this surface (with ADM ≤ AS). Lp,S and Lp,DM are sound pressure levels radiated from the uncovered surface and from the covered one. The area term 10*log(AS/ADM) increases the TL when the damping material covers only a part of the plate. Usually this term is used in ducts acoustics and represents the cross-sections of the inlet and outlet of a muffler. Here, this term is used to compensate the noise being radiated without damping from the uncovered surface the plate. The plate used in this investigation has an area of AS=400 x 200 mm. Each material prototype has an area of ADM=355 x 155 mm. The area term thus delivers an increase of 1.63 dB for all far field measurements. There are different ways to increase the acoustic effect of a material used for noise reduction purposes. Classic materials used in noise control applications are (a) homogeneous heavy foils consisting of viscoelastic materials with high density such as bitumen or butyl rubber. They are mainly applied at noise emitting metal sheets and increase the TL by mass law and material damping which is described by the loss factor. In (b) heterogeneous materials such as PUR foams, melamine foams or microfiber mats the material damping of the solid material is combined and increased with viscous and thermal dissipation effects in the fluid phase, friction between fluid and solid phase or even, in the case of microfibers, friction between the solid components themselves. While the acoustical ef-

9 9

The noise reduction potential of lightweight acoustic metamaterials – a numerical …

fect of a given homogeneous material can be increased only by its thickness heterogeneous, porous materials offer more influence parameters such as average pore size, porosity and reticulation rate (areal rate of open cells). These parameters can be influenced by chemists and process engineers during the foaming process. The dissipation effects in poroelastic materials are often increased by a mass layer at the material surface. With this layer the material represents a mass-spring-damper system. The effects of such surface mass layers on the transmission loss were studied in [2]. The first group of acoustic metamaterials investigated in this study are locally resonant structures. They are composed of PUR-foam matrix and mass inclusions of different weight and size. Locally resonant materials are well-known from literature [3-9]. In these studies mass inclusions within the elastic matrix cause frequency bands with high transmission losses or absorption coefficients. The frequency of increased transmission losses depends on the mass of the inclusion and its position in the material [5-7] and can be explained by a negative mass density [7-9]. The studies show a general possibility of a low-frequency increase of the vibration energy dissipation by resonant masses in a matrix consisting of a damping material. These studies suggest to use different masses and positions in the matrix to achieve sound energy dissipation in a broader frequency range. In [10] a lightweight honeycomb-metamaterial with a covering membrane was introduced that show high transmission losses at low frequencies due to the membrane. Here, such a honeycomb structure with a surface membrane is first investigated single. After that it was filled with PUR-foam in the first step, while in the second step resonant masses were added to the foam. They were placed in the Honeycombs. The third concept is inspired by room acoustic applications of perforated acoustic panels. These concepts dissipate acoustical energy by friction between moving air particles and solid material in the panel’s perforations that have to be very small [13]. In their room acoustical application the air volume between the perforations and the wall works as a Helmholtz resonator whose effects increase the friction losses in the micro perforations. The theory of noise absorption by micro-perforated panel absorbers (MPA) was developed by Dah-You Maa in [11]. Zhang suggested a double layered MPA structure with two sound absorption maxima in [12]. The application in this study uses commercial MPA’s. The principal design of the combination of one or more membranes and a PURfoam-material is shown in Fig. 10 below. At least one membrane was tensed and fixed across one or two cavities at the surface of the PUR foam. Additional tensed membranes could have been inserted into the cavities in order to obtain a multi-membrane structure.

10 10

The noise reduction potential of lightweight acoustic metamaterials – a numerical …

Material production The PUR foam was made from FlexFoam-iT!® III that is contributed by KauPo as a twocomponent product of fluent isocyanates (component A) and alcohols (component B) which together react quickly under emergence of CO2 that foams up the synthetized PUR. The two components were mixed in fix proportions and casted into a wooden form where they reacted to PUR under expansion up to the fifteen fold of their primary volume. During its expansion process it flows around the mass inclusions that were fixed on steel wires as shown in Fig. 9 (a). For this purpose the form consists of a base plate and four frames of identical size with height of each 10 mm. There are 8 notches at the short side of each frame that allow inserting the steel wires at which the drilled mass inclusions are positioned in regular patterns. The wires can be placed at three levels. If four wooden frames are used, the inner dimensions of the box are 355 x 155 x 40 mm. This was done for the resonant mass structures and for the MPA-structures, while the honeycomb prototypes were realized in a thickness of 20 mm, using only two frames.

Fig. 9: Process of material production shown for locally resonant structures. (a) Form with four frames and wires with mass inclusions, (b) release agent and the two components of FlexFoam it!®III, (c) finished metamaterial with mass inclusions, removed from the form.

The MPA-Materials basic foam required the casting of one or two cavities at the surface of the foam. They were realized with aluminium plates in the thicknesses of 10 and 15 mm and in the size 340 x 140 mm (one cavity) and 165 x 140 mm (two cavities). These plates were placed at the bottom plate before the expansion of the PUR foam started and the top plate with its expansion drills was screwed at the top frame to close the form. The honeycomb-foam structures were realized with semi-open honeycomb plates with the dimensions of the form and a thickness of 15 mm. They were placed at the bottom of the form

11 11

The noise reduction potential of lightweight acoustic metamaterials – a numerical …

with the closed side down. The first prototype was kept without mass inclusions. The liquid foam was casted onto the open honeycombs, from where the foam was expanding filling the remaining space in the form and delivering a material with a thickness of 20 mm. Two Honeycomb-foam structures with mass inclusions were realized. The inclusions were placed in a regular pattern in some of the honeycombs before the the foam was filled in. The variant with the higher mass has small glass and steel inclusions in the dimension 5 mm and a mass of each 0.27 g. The 340 mass particles were put each by each in every second honeycomb. The foam didn’t filled each honeycomb completely so that the inclusions are still able to move between the foam and the closing membrane causing a rustling noise if the material is shaked. In the second variant drilled steel spheres with a mass of 0.71 g and a diameter of 10 mm were jammed into each 8th (each fourth comb in each second row). The drills were positioned in such a way that a foam expansion into the honeycomb space under the steel spheres was possible. So, the mass particles are

Cavity (1) one MPA MPA‘s (2) two MPA‘s Damping Material

(3) three MPA‘s Fig. 10: Design concept of micro perforated panel absorbers (MPA)-materials

fixed in the foam matrix or between foam and the covering membrane. The MPA-materials are designed according Fig. 10. They are basically realized by glueing different micro perforated membranes at the backside of the mounting frame and cover the cavity of the basic foam by mounting the foam with the frame (see Fig. 11(e)). Each membrane was prestressed under a constant force of 11 N as shown in Fig. 11 (c) and is fixed at the frame under this stress. By this all membranes fixed on the basic material were mounted under a defined stress being identic for each membrane. Membranes with three different pore sizes were used (Fig. 11 (d)). They are characterized in Table 2. Table 2: micro perforated membranes used in the investigation (manufacturer’s data [14])

Miniperf Acoperf Nanoperf

12

Membrane thickPerforation ness [mm] number [1/m²] 0.3 30,000 0.18 400,000 0.15 500,000

Perforation diameter [mm] 0.5 0.15 0.1

Perforation rate [%] 5 0.8 1.0

The noise reduction potential of lightweight acoustic metamaterials – a numerical …

For materials with two or three perforated membranes (Fig. 10) the MPA’s were prestressed and fixed at 5 mm high frames made of the rigid plastic foam Rohacell®. The mass of the frames with the membrane was between 17.6 and 21.5 g. The frames were inserted with the fixed membrane the into the 15 mm deep cavity of the prototype shown in Fig. 11 (a) right below.

Fig. 11: Production of micro perforated panel absorber (MPA)-Material, second stage. (a) PUR foams of the thickness 40 mm with two or one cavities at the surface with a depth of 10 and 15 mm, (b) 5 mm thick frames made of Rohacell® with membranes for placement in a cavity, (c) prestressing of the panel with F=11 N, (d) three perforation sizes (Barrisol® Miniperf, Acoperf, Nanoperf), (e) Panel covering the cavities of the foam, fixed at the mounting frame

Experimental setup The material prototypes were investigated in an anechoic room with full absorption above 100 Hz at all walls. Fig. 12 shows the experimental setup where the material prototype is fixed with a steel frame at the surface of an aluminium plate with the dimensions 400 x 200 x 18 mm. The frame is fixed as shown in Fig. 12(c) with screw nuts at thread rods that are fixed at the plate. The screw nuts are tightened at each prototype until the small gap between foam and plate has disappeared at each side. The plate is coupled to an electrodynamic shaker by a rod that is fixed at the backside of the plate. The shaker is excited with a white noise current generated and amplified with constant magnitude by a signal generator and an amplifier. The transduced force between shaker and plate was measured by a force transducer mounted inline with the connection rod. The radiated noise of the plate was measured by a single microphone positioned in a distance of 0.95 m from the plate and in the height of the plate’s center. The microphone was rotated at equal height around the plate between 0° and 180° in steps of 10° in order to obatin a directional

13 13

The noise reduction potential of lightweight acoustic metamaterials – a numerical …

Fig. 12: Experimental setup. (a) plate without material with microphone array (near field) and rotatable far field microphone. The plate’s backside and the shaker is shielded with an absorption material mat from the frontside. (b) plate with mounted MPA-Material and far-field microphone at an angle of 120°. (c) mounting frame and material at the plate

Fig. 13: Averaged SPL of the far field microphone (upper left), directional sound radiation characteristics in the far field of the uncovered Aluminum plate (upper right) and array-measured SPL patterns for characteristic frequencies at the plate (lower row)

characteristic of the SPL in the frequency range 0.1-12.8 kHz (Fig. 13, upper right). The directional characteristic was averaged arithmetically in order to receive the overall SPL characteristics (Fig. 13, upper left). Within the anechoic room, the averaged SPL represents the radiated power of the plate in the far field, i. e. above 0.35 kHz. For the evaluations in the following sections the results are shown as third octave levels of the SPL characteristics. The transmission loss was calculated from the third octave levels of the

14 14

The noise reduction potential of lightweight acoustic metamaterials – a numerical …

uncovered plate and the plate with the acoustic material following Eq. (1). For certain measurements, the 4 x 8 microphone array was placed in a distance of 50 mm from the noise radiating surface (plate or damping material). This is near field up to a frequency of 6.5 kHz. The sound pressure reduction in the nearfield is regarded for the honeycomb structures.

Experimental results Fig. 14 shows the averaged SPL in the far field and the transmission losses for three different resonant structures. They differ in the number and the mass of the inclusions. The arrangement of the inclusions is shown at the right side and related to the graphs. The overall SPL is almost the same for all five materials including the foam without additional masses, which thus has a clear preference due to its smaller mass. Its overall TL is 9.5 dB. The materials with mass inclusions show overall TL above 8.7 dB. Within the observed frequency range the positive effect of the mass inclusions was detected in the frequency range between 0.21 and 0.7 kHz. Here, improvements of the transmission of up to 5 dB were gained in comparison to the simple foam. The prototypes with included masses of

Fig. 14: Sound pressure levels in the far field (upper diagram) and transmission losses (lower diagram) for the basic plate (black), a 40 mm PUR foam without inclusions (red), three different local resonant structures with varied mass inclusions shown at the right side

15 15

The noise reduction potential of lightweight acoustic metamaterials – a numerical …

different sizes (1.4 g, 0.8 g and 0.23 g) show the best effect on the first sound pressure peak at about 560 Hz (see Fig. 13). This effect seems to be caused by the use of different mass inclusions and not simply by the additional mass that these prototypes have. If this would be the case similar effects had to be observed using the material with additional 48 x 1.4 g plastic spheres at the medium layer (Fig. 15, violet graph). This material doesn’t show the low frequency improvements of the multi-mass prototypes that are visible in the TL diagram of Fig. 14. This could be seen in equivalence to the result of the

Fig. 15: transmission losses of the material with 48 inclusions à 1.4 g at the upper layer (blue, dotted) compared with additional and shifted arranged 48 inclusions at the medium layer (violet, dashed)

numerical investigation shown in Fig. 8, where an increase of the diameter of the massive spheres from 15 mm to 20 mm leads to an increase of the overall sound power level of nearly 4 dB. The 48 additional masses in the medium layer could have caused a stiffening of the structure by bridge effects between the closely arranged spheres. As the increase of the mass of the inclusions in the designs c, d, and e in Fig. 8, the additional mass inclusions in the medium layer show an increased sound power radition in comparison to the material prototype with 48 spheres à 1.4 g only at the top layer of the material. Both simulation and experiment show that an “oversaturation” of the damping material structure with mass inclusions or the use of mass inclusions with too high masses and dimensions can cause lower damping effects and has to be avoided. On the other hand the diagrams of Fig. 14 shows that the use of inclusions with different masses and sizes can improve the damping in a certain frequency range compared to mono-mass material design.The left diagram in Fig. 16 compares the TL of the already mentioned mono-mass structure with 48 spheres at the top layer with a commercial and effective damping material characterized by an impregated mass layer and a thin lamination on top. The material with mass inclusions at the top layer was measured again upside down, the masses towards the plate. This and the previous application are similar to case (b) and (c) in Fig. 3. As well as the simulation results in Fig. 4 the experimental results show a shift

16 16

The noise reduction potential of lightweight acoustic metamaterials – a numerical …

of the damping effects up to higher frequencies when the masses are placed close to the plate. This is caused by a higher spring stiffness of the foam due to a shorter distance between plate and mass. The commercial foam has an overall thickness of 15 mm and a mass layer created by inpregnation with a depth of 6 mm leaving 9 mm soft foam as spring and damper. This material shows the highest damping effects above 0.7 kHz, where its TL is higher than the TL of the resonant structures and the PUR-foam without inclusions. In the frequency range between 0.3 and 0.7 kHz the resonant structures show a better damping than the commercial material with almost equal mass. In the right diagram of Fig. 16 the specific transmisson losses i. e. the TL divided by the mass of the material are shown for the resonant structures with the highest overall effects, the commercial damping material and the simple PUR foam without inclusions. Regarding

Fig. 16: Transmission losses (left) and mass-specific TL (right) for different materials. Left diagram: same mass inclusions at different positions in the foam (blue) in comparison with a commercial damping material (grey) with equal mass and a PUR-foam without inclusions. Right diagram: Specific TL for the best resonant-mass material configurations compared with the commercial foam (grey) and the reference-PUR foam without inclusions (red)

the specific TL the advantageous behavior of the resonant structures in a small band between 0.3 and 0.7 kHz is reproduced while the PUR foam without inclusions has the best overall effect especially at higer frequencies. The sound transmission behavior of the honeycomb structures is regarded by the SPL measured with the microphone array in a distance of 50 mm from the noise radiating surface. The SPL reduction ∆Lp delivers the acoustic damping effect. The basic investigation was carried out with a PUR foam without additional elements and a thickness of 20 mm (instead of 40 mm for the resonant mass structures and for the microperforated membrane materials). Its resulting acoustical behavior is shown by the red graphs in the Diagrams of Fig. 17. A 15 mm thick plastic honeycomb plate was applied to the aluminum plate with the closing membrane outside. This arrangement of the honeycomb and its covering membrane was used for the honeycomb-PUR-foam and honeycomb-PUR-foam-mass inclusions as well. All material configurations were mounted with the covering membrane of the honeycombs outside. An experiment where

17 17

The noise reduction potential of lightweight acoustic metamaterials – a numerical …

Fig. 17: Sound pressure level in the near field for honeycomb structures (left) and sound pressure level reduction for different honeycomb-foam-mass structures (right diagram)

the membrane was placed at the side of the plate showed worse results. The grey dashed graphs in Fig. 17 show the effect of the mere honeycomb plate on the SPL radiated from the plate. The light blue graph represents the 15 mm honeycomb plate combined with a PUR-foam filling without mass inclusions. The dark blue and violet graph represent the honeycomb-foam structures with mass inclusions. As the right diagram of Fig. 17 shows, all honeycomb structures combined with PUR foam show significant SPL reductions in a frequency range between 0.5 and 2 kHz. These reductions are significantly higher than these gained with the simple PUR-foam of the same thickness. The honeycomb plate with the surface membrane alone doesn’t show these effects with the exception of the third octave band at 0.6 kHz where the first eigenmode’s noise radiation was reduced about 6 dB. The honeycomb concepts with foam are increasing this damping effect significant and a further improvement is gained by the use of the mass inclusions – especially the concept with 340 small mass inclusions fixed by the foam within the honeycomb shows a high SPL reduction for this resonance. At high frequencies above 5 kHz the honeycombPUR-foam design has the best reductions. Below 0.5 kHz the honeycomb-foam material with mass inclusions of 0.71 g shows a small improvement in comparsison to the other honeycomb concepts. Compared with the resonant mass structures the positive acoustical effects were proved in higher but also broader frequency ranges. The use of mass inclusions was able to improve the noise reduction in low and medium frequencies down to 0.3 kHz. Mass inclusions with higher mass in the honeycombs, it seems, are shifting their acoustic effect down to lower frequencies as in the resonant material structures. Fig. 18 shows the TL for micro perforated membrane-materials with one single membrane tightened across one or two surface cavities. The red graphs are marking the basic foam with the cavities without membrane. It should be mentioned that deeper cavities mean lower effective material thicknesses and thus lower material masses, but not necessarily lower TL as Fig. 18 shows. The membranes were already caracterized in

18

18

The noise reduction potential of lightweight acoustic metamaterials – a numerical …

Fig. 18: Transmission losses for MPA-Materials with different membranes tightened across different cavities. Upper diagrams: 10 mm cavities deepness, lower diagrams: 15 mm; left diagrams: one cavity, right diagrams: two cavities

Table 2. The use of one perforated membrane led to increased TL in the high-, medium-, and low frequency range in all cases. The effects differ between different cavity depths. The remarkable improvement in the frequency range between 0.9 and 2 kHz due to the membranes was observable only at the cavities with a deepness of 15 mm. Here, as well as at frequencies below 0.2 kHz, the Acoperf membrane gained the highest improvements. At frequencies above 2 kHz the effects of the different perforation membrane configuration show only small deviations in comparison to the material

Fig. 19: Sound pressure level (left) and transmission losses (right) of a MPA-Material with and without microfibers in the cavity.

19 19

The noise reduction potential of lightweight acoustic metamaterials – a numerical …

without membrane. In the low frequency range the Miniperf configuration seems to be advantegous as well as the Acoperf membrane. The two-cavity design whose TL were shown in Fig. 18 was developed because of the dominating two-antinode noise radiation at 3.15 kHz that is shown in Fig. 13. The subdivision of the cavity doesn’t show an effect on the transmission losses in this frequency range. Thus for the following experiment the lightest basic material was chosen – the foam with one 15 mm cavity. This cavity was completely filled with light microfibers with a mass of 13.7 g. The microfibers are supposed to cause friction effects that lead to dissipation of acoustic energy and the damping of membrane vibrations. The effect of the Microfibers were tested with the Nanoperf-Membrane. The results are shown in Fig. 19. The Microfibers cause improvements of the TL up to 2 dB in the whole frequency range, but especially in the range between 0.4 and 1.6 kHz. Due to their small density they can be recommended as a simple acoustical improvement of such structures.

Fig. 20: Transmission losses for multi-membrane designs compared with the according single membrane version (solid blue line) and the basic material without membrane (red)

Finally, different MPA-membranes were fixed under the defined prestress of 11 N on the light frames with a thickness of 5 mm. They were inserted and jammed into the single 15mm cavity that was again covered with the different MPA-membranes. Several combinations with two or three membranes of different or equal pore size were investigated. Their transmission losses are shown in Fig. 20. The acoustical improvements of the additional membranes within the cavity are small and mainly located

20 20

The noise reduction potential of lightweight acoustic metamaterials – a numerical …

in the frequency range above 7 kHz that is of little interest. Improvements are also visible for all multi-membrane designs below 0.125 kHz. A little increase of the TL was also gained by all membrane combinations in the frequency range between 2.1 and 5.4 kHz. Here, the three-membrane-designs show the highest improvements of up to 3 dB. At frequencies below 2.1 kHz additional membranes caused a smaller transmission loss in some frequency bands. The additional mass for all multi-membrane designs was not higher than 49 g.

Conclusion In this study three different metamaterial concepts were investigated numerically and experimentally in a test setup designed for an application of these materials at the surface of an aluminium component in order to reduce the radiated noise. Various prototypes of the different concepts were fashioned based on a PUR-foaming process that was identical for all prototypes. The numerical study of the resonant structures revealed their potential to reduce the radiated sound power up to 7 dB. It was shown that a randomized arrangement of the mass inclusions especially in the height direction has a positive influence on the sound absorption. Another important factor revealed by the simulation is the density and the mass of the inclusions. A low density or very small masses can cause even a noise amplification in comparison to the bare plate, while too high masses of the inclusions reduce the damping effect of the damping material. The experimental investigation of different mass-resonant structures show the potential of this concept to reduce the noise radiation of a surface at low frequencies. A possibility to the design the acoustic effect was found parallel to according simulations: arrangement of the masses in smaller distance to the damped surface leads to higher frequencies of the damping effect. A concept with too many mass inclusions in close layers was rejected as inefficient. Positive effects on the noise reduction were shown for concepts with mass inclusions of different size. The combination of plastic honeycomb with a membrane, PUR foam and mass inclusions show acoustical improvements in higher frequency ranges. The application of mass inclusions caused a shift of the damping effect to lower frequencies down to 0.3 kHz that were achieved by an increased mass as well as the advantages of the resonant materials structure. The latest investigated structures – micro perforated-membrane absorbers – gained their acoustic improvements with very small increase of additional mass. The acoustic effect of the micro perforated membranes is broader than the one of the honeycomb concepts and the mass inclusion concepts, and improvements of the radiated noise are gained at low, middle and high frequency range. The small perforations of the Acoperf membrane show the best overall results. A filling of the cavity with light microfibers show an additional positive effect. The effects of additional perforated membranes tightened up within the cavity were small mainly visible at higher frequencies.

21 21

The noise reduction potential of lightweight acoustic metamaterials – a numerical …

Acknowledgements The presented work is part of the joint project COMO III (COmpetence in MObility), which is financially supported by the European Union through the European Funds for Regional Development (EFRE) as well as the German State of Saxony-Anhalt (ZS/2016/04/78118). The micro perforated membranes were made available from barrisol®. Both supports are gratefully acknowledged.

Literature [1] P. Schrader, F. Duvigneau, M. Trübe, H. Rottengruber, U. Gabbert, „Passive Reduktion der Schallabstrahlung von Oberflächen durch Anwendung von Metamaterialstrukturen“, 43. Jahrestagung für Akustik - DAGA, Kiel, 2017. [2] P. Schrader, F. Duvigneau, R. Orszulik, H. Rottengruber and U. Gabbert, „A numerical and experimental study on the noise absorption behavior of functionally graded materials considering geometrical and material influences”, Internoise 2016 conference proceedings, pp. 6451-6462, 2016. [3] Z. Liu, X. Zhang, Y. Mao, Y. Y. Zhu, Z. Yang, C. T. Chan and P. Sheng, “Locally resonant sonic materials”, Science, vol. 289, pp. 1734–1736, 2000. [4] Z. Liu, C. T. Chan and P. Sheng, “Analytic model of phonoic crystals with local resonances”, Physical Review B, vol. 71, pp. 014103 1–8, 2005 [5] C. Fuller and T.-D. Saux, “Sound absorption using poro-elastic acoustic metamaterials, Internoise 2012 conference proceedings, 2012 [6] K. Idrisi, M. E. Johnson, D. Theurich and J. P. Carneal, “A study on the characteristic behavior of mass inclusions added to a poro-elastic layer”, Journal of Sound and Vibration, vol. 329, pp. 4136–4148, 2010. [7] P. A. Deymier (Editor), “Acoustic Metamaterials and Phononic Crystals”, SpringerVerlag Berlin Heidelberg, pp. 159–197, 2013. [8] P. Sheng, J. Mei, Z. Liu and W. Wen, “Dynamic mass density and acoustic metamaterials”, Physica B, vol. 394, pp. 256–261, 2017. [9] M.-H. Lu, L. Feng and Y.-F. Chen, “Phononic crystals and acoustic metamaterials”, Materials today, vol. 12, number 12, pp. 34–42, 2009.

22 22

The noise reduction potential of lightweight acoustic metamaterials – a numerical …

[10] N. Sui, X. Yan, T.-Y. Huang, J. Xu, F.-G. Yuan and Y. Jing, “a lightweight yet sound-proof honeycomb acoustic metamaterial”, Applied Physics Letters vol. 106, pp. 171905-1–4, 2015. [11] Maa, D.-Y., “Potential of microperforated panel absorber”, Journal of the Acoustical Society of America, vol. 104, pp. 2861-2866, 1998. [12] Z. M. Zhang and X. T. Gu, “The Theoretical and Application Study on a Double layer Microperforated Sound Absorption Structure, Journal of Sound and Vibration, vol. 215, pp. 399-405, 1997 [13] H. V. Fuchs, “Schallabsorber und Schalldämpfer”, 2nd edition, Springer-Verlag Berlin Heidelberg New York, pp. 103-124, 2007. [14] Barrisol®, „Acoustics® & Design“, brochure, http://de.barrisol.com/PDF/brochures/barrisol-acoustic-design.pdf, lastly called 05-17-2017.

23 23

The noise reduction potential of lightweight acoustic metamaterials – a numerical …

Authors: Dipl.-Ing. Peter Schrader Research Assistant Otto-von-Guericke-University Magdeburg Universitätsplatz 2, 39106 Magdeburg ++(0)391 67-12831 [email protected] Dipl.-Ing. Fabian Duvigneau Research Assistant Otto-von-Guericke-University Magdeburg Universitätsplatz 2, 39106 Magdeburg ++(0)391 67-52754 [email protected] Prof. Dr.-Ing. Hermann Rottengruber Otto-von-Guericke-University Magdeburg Universitätsplatz 2, 39106 Magdeburg ++(0)391 67-18721 [email protected] Prof. Dr.-Ing. habil. Dr. h. c. Ulrich Gabbert Otto-von-Guericke-University Magdeburg Universitätsplatz 2, 39106 Magdeburg ++391 67-58609 [email protected]

24

Vibration damping behavior of flexible polyurethane foams under low and high strain regimes Mark Brennan, Martino Dossi, Yueqi Wang, Maarten Moesen and Jan Vandenbroeck, Huntsman Polyurethanes

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 W. Siebenpfeiffer (Hrsg.), Automotive Acoustics Conference 2017, Proceedings, https://doi.org/10.1007/978-3-658-20251-4_16

1

Vibration damping behavior of flexible polyurethane foams under low and high strain …

Introduction Polyurethane flexible foam chemistry is a urethane reaction of isocyanate and polyol, to form a soft-like polymer, and a urea reaction of isocyanate and water to provide the blowing agent CO2 which is released during the reaction. The former reaction gives the foam its flexibility, strength and vibration damping properties, while the latter leads to the cellular microstructure and urea groups that phase separate into hard segments and contribute to the load bearing properties of the foam1. A variety of foam microstructures can be formed from polyurethane chemistry including polydisperse spherical cells of higher density foam and anisotropic polyhedral cells of low density foam. The microstructure can be characterised by its porosity, cell size distribution and closed cell content2. The final microstructure will depend on the formulation chosen, including types of catalysts and surfactants used. In Figure 1 and Figure 2, volume renderings of X-ray computed tomography (micro-CT) scans of a low-density sound absorbing and higher density sound insulating foam are shown.

Figure 1 3D reconstruction of X-ray tomography data of a sound absorbing PU foam with a density of 15 kg/m3

Figure 2 3D reconstruction of X-ray tomography data of a sound insulating PU foam with a density of 65 kg/m3

1 Randall, D. and Lee S., eds., (2002) The Polyurethanes Handbook, London: Wiley 2 Montminy, M.D., Tannenbaum A.R. and Macosko, C.W.: The 3D structure of real polymer foams. Journal of Colloid and Interface Science 280 (2004) 202–211

2

Vibration damping behavior of flexible polyurethane foams under low and high strain …

The vibration damping properties of the foam are heavily influenced by the viscoelasticity of the polymer in the foam. This is mainly governed by a nano-sized phase separated structure consisting of polyurethane-rich flexible phase and polyurea-rich hard phase. This is formed during the reaction process and can be controlled by the formulation parameters, the miscibility of the raw materials and the choice of polyisocyanate and polyol3. For sound transmission and vibration damping applications, the material property specifications often require parameters such as density, shear and bulk modulus, compression hardness at different degrees of compression, hysteresis, dynamic elastic modulus and loss factor. In Figure 3, a single typical compression cycle is shown for high resilient (HR) and viscoelastic (VE) sound insulating foams. Both foams display non-linear stress strain behaviour with an initial higher elasticity, followed by a lower elasticity plateau region due to buckling of foam cells, and finally higher elasticity due to densification of the foam. The foams display a hysteresis loss. Typically, these measurements are performed at strain rates of 10-1000 mm/min. In Figure 4 a typical vibration damping measurement on an electromagnetic shaker is shown for the two sound insulating foams. Assuming a mass-spring-damper model of the foam, a dynamic elastic modulus and loss factor can be derived from this measurement. In Figure 4 the sharp resonant peak of the HR foam compared to the more damped viscoelastic foam can be observed. The exact position of the resonant frequency will depend on the material properties, experimental setup and geometry of the sample. The challenge in designing new vibration damping materials can be seen in the uniaxial compression behaviour of PU flexible foam. It is non-linear and hyperelastic due to the buckling of the foam microstructure during compression; viscoelastic or frequency dependent due to the dynamic response and mobility of the polymer in the foam microstructure; and hysteretic due to the air-solid interaction during loading and unloading. This paper will present advanced methods to study the non-linear and viscoelastic behaviour of PU flexible foam and show how lower length scale structures can be linked to macroscale properties. It will be shown that the foam is non-linear even at small strains and therefore it is difficult to ascribe a single elasticity, loss factor or Poisson’s ratio to foam materials, often required for sound insulation specifications.

3 Lee, S.T. and Ramesh, N.S.: Polymeric Foams: Mechanisms and Materials. Boca Raton: CRC Press 2004

3

Vibration damping behavior of flexible polyurethane foams under low and high strain …

Figure 3 One compression cycle of sound insulating foam, made from different PU chemistry, showing non-linear stress strain behaviour and hysteresis loss

Figure 4 Vibration damping measurement of sound insulating foams, made from different PU chemistry, showing mass-spring-damper behaviour

The effect of the microstructure of the foam is studied using micro-CT combined with a compression stage. Modelling microstructures with simple polymer properties can illustrate how the apparent vibration properties can be influenced by compression. Using Stereo Digital Image Correlation, the Poisson’s ratio of the foam at different strains can be measured. Dynamic mechanical analysis is used to study the viscoelasticity of the polyurethane polymer in the foam. Finally, a material model will be proposed for PU flexible foam describing the foam’s hyperelasticity and viscoelasticity and how it can be used to interpret vibration damping measurements.

4

Vibration damping behavior of flexible polyurethane foams under low and high strain …

Material Characterisation Methods Microstructural analysis of flexible polyurethane foam undergoing compression An in-house PU flexible foam sample of 9x9x12 mm3 was prepared for micro-CT scanning with a Bruker Skyscan 12724. The sample was compressed in Bruker’s 44 N in-situ materials testing stage at a speed of 0.24 mm/min in 18 steps of 0.67 mm (5%). Initially and after each compression step, a micro-CT scan was performed at a resolution of 7.4 microns. 3D reconstructions and basic analysis were made using Bruker’s NRecon and CTAn software. Cell size analysis was made using in-house developed code. The force is recorded during compression and is shown in Figure 5 below. The familiar nonlinear form of the compression and hysteresis is obtained even for this very small sample. The curve is irregular due to the foam being relatively soft, which results in relatively low recorded forces in comparison to the sensitivity of the load cell, and due to the ‘step and shoot’ scanning method, which allows stress relaxation during the hour-long scan at each step. An initial vertical cross section of the foam microstructure is shown in Figure 6. As it undergoes axial compression, small transverse expansion can be observed at low strains, particularly for the case of 10% strain and near the compression plates, Figure 6. However, for further axial compression the foam has very little further transverse expansion, Figure 7, with foam cells buckling and foam behaving with almost perfect compressibility. This suggests a higher initial Poisson’s ratio, the ratio of transverse and axial strain, with the value dropping with further compression. Image analysis involves creating fully connected skeleton structures of scanned data of foam samples. This is done by segmenting the data into air pores and solid struts. In Figure 8, the foam microstructure initially and at a compression of 25% and 50% is shown. From these skeleton structures, it is possible to calculate the sizes of the air pores and the solid struts in the microstructures. The size distributions of the pores and struts for these foams are shown in Figure 9 and Figure 10. These figures illustrate that average cell diameter reduces from 450 micron to 150 micron however, the average size of the solid material in the foam remains unchanged with compression. The struts buckle which leads to cell collapse and cell size reduction. This contributes to the non-linear stress-strain relationship of the foam.

4 https://www.bruker.com/products/microtomography/micro-ct-for-sample-scanning/skyscan1272/overview.html; Retrieved 21st May 2017

5

Vibration damping behavior of flexible polyurethane foams under low and high strain …

Figure 5 Recorded force during micro-CT scan of foam undergoing compression

Figure 6 Vertical cross sections from CT reconstructions of a flexible foam undergoing compression, 0% (left), 5% (middle) and 10% (right) strain

6

Vibration damping behavior of flexible polyurethane foams under low and high strain …

Figure 7 Vertical cross sections from CT reconstructions of a flexible foam undergoing compression, 25% (left), 50% (middle) and 75% (right) strain

Figure 8 Individual foam cell reconstruction, coloured with cell size, for the uncompressed foam (left), foam compressed 25% (top right) and foam compressed 50 % (bottom right)

7

Vibration damping behavior of flexible polyurethane foams under low and high strain …

Figure 9 Pore size distribution of the initial foam and at a compression of 25% and 50%

Figure 10 Strut size distribution of the initial foam and at a compression of 25% and 50%

Stereo digital image correlation and estimating Poisson’s ratio Digital Image Correlation (DIC)5 is a full-field optical deformation measurement technique. The method uses digital cameras to capture the deformation of a paint speckled specimen under loading. Deformation is measured by analysing the digital images and compared to classic point-wise measurement methods such as strain gauge or extensometer, DIC provides a full-field measurement which contains far more detailed information of materials like strain concentration and material heterogeneity. DIC is an image registration technique and measures deformation by mapping the image of a deformed specimen compared to the image of its un-deformed state. StereoDIC uses two cameras to build a stereo vision of the sample, Figure 11, so out-ofplane deformation can also be measured. Stereo-DIC does not suffer from misalignment error, compared with traditional single camera setups, which makes experimental setup easier and yields more accurate results.

5 https://en.wikipedia.org/wiki/Digital_image_correlation; Retrieved on May 21st 2017

8

Vibration damping behavior of flexible polyurethane foams under low and high strain …

Figure 11 Stereo image of polyurethane flexible foam with speckled pattern undergoing compression with resulting vertical strain field obtained from digital image correlation (DIC)

The compression tests were performed on a Zwick Z010 tensile bench. Two AVT F201B stingray cameras with 50mm Nikon lenses were deployed to monitor the deformation of the specimens. The test speed was 50 mm/min, and camera speed was 300 ms/frame. The displacement and force measured by the tensile bench were transferred to the DIC system, and synchronized with the images. Experimental control and data analysis were performed using DIC software MatchID6. A single analysis of the strain field of foam at 25% compression is shown in Figure 11. Using Stereo-DIC, the strain fields in different directions were measured in the compression test. The strain field is averaged in a rectangle. Figure 12 and Figure 13 shows the average transverse and axial strains for two types of PU foam, namely, high resilient and viscoelastic foam. Transverse strains initially rise for both foams but quickly flatten to near zero. This results in a high Poisson’s ratio at low strains ~ 0.35, falling to near zero are larger strains, Figure 15.

6 http://www.matchidmbc.be/AboutMID; Retrieved 21st May 2017

9

Vibration damping behavior of flexible polyurethane foams under low and high strain …

Figure 12 Axial and transverse strain measurements of a high resilient PU foam obtained from Stereo-DIC

Figure 13 Axial and transverse strain measurements of a viscoelastic PU foam obtained from Stereo-DIC

Figure 14 True strain – true stress relationship for high resilient and viscoelastic PU foams obtained from Stereo-DIC

Figure 15 Poisson’s ratio at different strains for high resilient and viscoelastic PU foams obtained from Stereo-DIC

Together with the force measurement synchronized with the images, the true strain (average over field) – true stress curves can also be obtained for both foams, Figure 14.

10

Vibration damping behavior of flexible polyurethane foams under low and high strain …

Dynamic mechanical analysis and polymer viscoelasticity Dynamic mechanical analysis (DMA) is a technique used by polymer scientists to study viscoelastic behaviour of polymers. Usually a small sinusoidal stress, to ensure the linear viscoelasticity, is applied and the strain in the sample material is measured. The complex modulus consisting of an in-phase component, called the storage modulus, and an out-of-phase component, called the loss modulus, is determined. The storage modulus measures the ability of the material to store energy and the loss modulus indicates the materials ability to damp energy. Commonly the damping component is represented by the loss factor, a ratio of the loss and storage moduli7. During the measurement, the temperature of the sample and/or the frequency of the stress applied are varied, leading to variations in the storage and loss modulus. From this, the glass transition temperature of the material can be determined, i.e. the temperature above which the material will behave rubbery instead of glassy. These transitions can be linked with different polymer mobility and viscoelasticity. In Figure 16 and Figure 17 the results of a DMA measurement for high resilient (HR) and viscoelastic (VE) sound insulation foams are compared. A peak in the loss factor can be observed at -50˚C and -30 ˚C for all foams indicating the glass to rubbery transition temperature, Tg, which is assigned to the beginning of molecular motion in the soft phase of the polyurethane polymer. Increasing frequency increases the storage modulus, or elasticity, of the foam and affects the loss factor. Note that the viscoelastic (VE) foam is more temperature dependent and has a higher loss factor near room temperature. This is easily observed when touching the foam at room temperature. Reformulating with a different flexible polyol with a different molecular weight, reactivity and/or miscibility with MDI results in other viscoelastic behaviour. The overall hardness can also be influenced by the type and amount of MDI used in the formulation. These phenomena can be observed using DMA. To study foam materials in sound insulation applications, the full frequency dependence of the polymer in the foam is required. Typically, dynamic mechanical laboratory equipment has an upper limit of about 100 Hz, so very high frequencies are not accessible. However, exploiting the observation that the temperature and frequency dependence of polymers is related, the time-temperature superposition principle can be applied. The full frequency dependence of foam materials in the linear viscoelastic range can be obtained. Using this principle, elasticity and loss factor of HR and VE sound insulating foam at room temperature are shown in Figure 18 and Figure 19. At low frequencies both foams have similar elasticity, however, the elasticity of the vis7 Macosko, C.W.: Rheology: Principles, Measurements, and Applications. New York: WileyVCH 1994

11

Vibration damping behavior of flexible polyurethane foams under low and high strain …

coelastic foam rises faster. For all frequencies, as expected, the loss factor of the viscoelastic foam is higher.

Figure 16 Dynamic mechanical analysis showing the influence of temperature, polymer chemistry (HR vs VE) and frequency (1 Hz vs 50 Hz) on the foam storage modulus

Figure 17 Dynamic mechanical analysis showing the influence of temperature, polymer chemistry (HR vs VE) and frequency (1 Hz vs 50 Hz) on the foam loss factor

Figure 18 Estimated elasticity as a function of frequency for two sound insulating foams

Figure 19 Estimated loss factor as a function of frequency of two sound insulating foams

12

Vibration damping behavior of flexible polyurethane foams under low and high strain …

Material Modelling Methods Ogden strain energy function for elastic foams Ogden8 proposed a strain energy function to describe hyperelasticity for slightly compressible rubbers, which has subsequently adapted for highly compressible lowdensity foams. It has the form =∑

+

+

−3 +U ( )

are the principal stretches of deformation, U is a function dependent upon where the volume ratio = and and are material constants. The principal Cauchy stresses are given by = ∑

=

+ U ( ),

= 1,2,3.

to be the principal stretch correspondFor uniaxial compression, and defining = = , to be the lateral stretches and recalling ing to the compression direction and that for foams the lateral deformation is small compared to the compression deformation, Figure 12 and Figure 13, it can be assumed that = 1 and , = 0, therefore, the principal stress becomes = ∑

−1 .

Finally, taking the applied stress to be ing strain, then =

∑

=

= 1 + , where

and

is the engineer-

((1 + ) − 1).

An elasticity, or tangent modulus, as a function of strain can be defined by taking the derivative of the stress, defined above, with respect to strain as follows, =

=−

(1 + )

−

((

)

)

.

In Figure 20 and Figure 21 the stress response and corresponding tangent modulus are shown for examples of high resilient and viscoelastic PU foams.

8 Ogden, R.W.: Large deformation isotropic elasticity: on the correlation of theory and experiment for compressible rubberlike solids, Proceedings of the Royal Society of London A 328 (1972) 567-583

13

Vibration damping behavior of flexible polyurethane foams under low and high strain …

Viscoelastic models of PU flexible foam Viscoelasticity can be modelled by a set of Maxwell elements, namely a spring reprein series with a dashpot representing a viscosity, senting a Young’s modulus, connected in parallel. An additional single elastic spring element, represents the long-time response of the material. The relaxation modulus of this material can be defined mathematically9 as ( )=

/

+∑

.

Its complex dynamic modulus10, or frequency dependence, is defined as ( )=

+∑

(

/

)

;

( )=∑

(

/

)

.

and can be obtained from DMA and the use of the timeThe parameters temperature superposition principle, Figure 18. If the elasticity estimated from the stress strain relationship at low speed represents the long-term behaviour of the material, , then an estimate of the elasticity as a function of strain and frequency can be obtained. Figure 22 and Figure 23 show this for two example PU foams.

Comparison with experimental data Vibration damping experiments on electromagnetic shaker were performed on PU samples of high resilient and viscoelastic foam. To maximise the range of loads, two sets of cubic foams samples were prepared for each type of foam. The dimensions of the foam were 8x8x4 cm3 and 4x4x4 cm3. Assuming a single degree of freedom massspring-damper model, the natural frequency and damping coefficient were estimated for the two foams with the two different sample sizes. A summary of the results is shown in Figure 24 and Figure 25.

9 Mills, N.J.: Polymer foams handbook: engineering and biomechanics applications and design guide, Elsevier 2007 10 Macosko, C.W.: Rheology: Principles, Measurements, and Applications. New York: WileyVCH 1994

14

Vibration damping behavior of flexible polyurethane foams under low and high strain …

Figure 20 Compressive stress response (blue) of high resilient PU foam and corresponding tangent modulus (red) as a function of strain

Figure 21 Compressive stress response (blue) of viscoelastic PU foam and corresponding tangent modulus (red) as a function of strain

Figure 22 Estimated elasticity of high resilient PU foam as a function of strain for different frequencies

Figure 23 Estimated elasticity of viscoelastic PU foam as a function of strain for different frequencies

15

Vibration damping behavior of flexible polyurethane foams under low and high strain …

Figure 24 Natural frequency of high resilient and viscoelastic PU foam for different loads assuming mass-spring-damper behaviour (dark = large sample; light = small sample)

Figure 25 Damping coefficient of high resilient and viscoelastic PU foam for different loads assuming mass-spring-damper behaviour (dark = large; light = small sample)

Note that natural frequencies of all samples, materials and loads are between 10 and 50 Hz with the viscoelastic foam having higher values over all loads. The damping coefficient of the viscoelastic foam is also higher, like what was observed in DMA, Figure 19, although the magnitude difference is higher. This could be explained by the additional air damping experienced by the foam in the electromagnetic shaker which is not experienced for the small displacements in DMA. To compare the results from vibration damping experiments with models described above, a stiffness is calculated for each load and for different frequencies. The measured apparent stiffness, and hence elasticity, for all materials and sample sizes is much higher than the low frequency limit, typical of a compression measurement. The shape of the change in stiffness with load is captured especially for the high resilient foams Figure 26 and Figure 27. For the high resilient foam the behaviour of the material suggests a frequency of between 1 and 10 Hz, lower than the experimental natural frequency. For viscoelastic foam, similar frequency dependence is observed between experiment and prediction, see Figure 28 and Figure 29. Note also the large variation in stiffness and hence elasticity even at low loads and hence low strains.

16

Vibration damping behavior of flexible polyurethane foams under low and high strain …

Figure 26 Comparison of calculated stiffness at different frequencies with measurements for an 8x8x4 cm3 sample of high resilient PU foam

Figure 27 Comparison of calculated stiffness at different frequencies with measurements for an 4x4x4 cm3 sample of high resilient PU foam

Figure 28 Comparison of calculated stiffness at different frequencies with measurements for an 8x8x4 cm3 sample of viscoelastic PU foam

Figure 29 Comparison of calculated stiffness at different frequencies with measurements for an 4x4x4 cm3 sample of viscoelastic PU foam

17

Vibration damping behavior of flexible polyurethane foams under low and high strain …

Microstructure modelling of ideal foam microstructures To further study the influence of foam microstructure and the degree of compression on vibration damping, a simple idealised foam microstructure was built in Comsol Multiphysics11. The system is compressed and a small harmonic perturbation applied and the transmissibility calculated. In Figure 31 the transmissibility for the ideal foam microstructure in Figure 30 is compared for different degrees of compression and different damping coefficients. The elasticity of the polymer is the same for all simulations. For both damping coefficients, the natural frequency of the system decreases when the foam is compressed indicating a change in the apparent stiffness, or elasticity, of the foam. This is similar behaviour to that observed in the experiments shown in Figure 24.

Figure 30 Ideal foam model with predicted first eigenmode for pre-compressed state

Figure 31 Comparison of the predicted transmissibility for an ideal foam microstructure

11 https://www.comsol.com/; Retrieved on 21st May 2017

18

Vibration damping behavior of flexible polyurethane foams under low and high strain …

Conclusions This paper demonstrated the non-linear behaviour of PU flexible foam and how it influences vibration damping properties. Various characterisation techniques to study the non-linear behaviour were presented including CT scanning and digital image correlation to study deforming foam microstructures and dynamic mechanical analysis to study polymer viscoelasticity. Material models were presented linking uniaxial compression data, DMA and vibration damping measurements of PU foam under different loads.

Acknowledgements The authors would like to thank Wim Van Bourgognie for PU flexible foam sample preparation, Nadja Richter for DMA measurements and Kristof Verniers for CT scans.

About Huntsman Huntsman Corporation is a publicly traded global manufacturer and marketer of differentiated chemicals with 2016 revenues of approximately $10 billion. Our chemical products number in the thousands and are sold worldwide to manufacturers serving a broad and diverse range of consumer and industrial end markets. We operate more than 100 manufacturing and R&D facilities in approximately 30 countries and employ approximately 15,000 associates within our 5 distinct business divisions including the Pigments and Additives division that we intend to IPO or spin-off as Venator Materials Corporation. For more information about Huntsman, please visit the company's website at www.huntsman.com.

19

Predicting pass-by noise levels for trucks based on component test bench measurements – by using virtual assembly techniques Patrick Corbeels, Dr. P. Van de Ponseele, M. Choukri Siemens Industry Software NV R. Sinnig, Dr. W. Kerres Daimler AG

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 W. Siebenpfeiffer (Hrsg.), Automotive Acoustics Conference 2017, Proceedings, https://doi.org/10.1007/978-3-658-20251-4_17

1

Predicting pass-by noise levels for trucks based on component test bench …

1 Introduction The goal of this research is to predict how main noise-causing components contribute to the exterior noise level, without having to physically integrate them in the truck. Road transportation noise is a growing source of concern. Legislations define pass-by noise test procedures according to which vehicles need to be tested. In the European Union, the noise emission limit is set by directive ECE-R51.03 [1]. The ISO standard has recently been updated, defining a more constrained test procedures that better reflect the reality of urban traffic noise [2]. At the European level, a plan has been set to lower noise limits in the next years.

Figure 1 Overview of the evolution of PBN limits in the EU for Pass-by Noise – ISO 362 Category N Directive ECE-R51.03 (Left), Exterior PBN test setup (Right)

Manufacturers want to achieve best-in-class NVH performance and, at the same time, speed up their design process at reasonable costs. They are exploring techniques that will allow them to set detailed acoustic targets for components (intake, engine, gearbox, exhaust, ATS, tires, etc.) and frontload NVH performance in the design process. The procedure proposed in this work allows to acquire all operational data on the component test benches (e.g. ATS test bench). Data is analyzed with the purpose of a realistic prediction of expected PBN noise contributions and setting of detailed design targets. A larger number of design variants can be quickly assessed and the optimal design can be selected. This process is called “pass-by noise vehicle synthesis” and allows permanent pro-active control of the NVH performance.

2

Predicting pass-by noise levels for trucks based on component test bench …

Figure 2 The PBN Vehicle Synthesis process shown for the exhaust After-Treatment-System ATS

The approach has been validated by applying the method to the industrial case of the exhaust ATS of a truck.

Figure 3 the After-Treatment-System ATS installed on a Daimler truck

3

Predicting pass-by noise levels for trucks based on component test bench …

2 The Process 2.1 Theory The inherent goal of the PBN vehicle synthesis process is to describe each of the sound sources in an invariant matter, independent of the receiving structure. Airborne sources are divided in a number of monopoles. It is assumed that in most cases the source strength of these monopoles can be considered invariant. Using this kind of source description implies that the internal source mechanism should be kept unchanged, independently of where the source is installed, on the component test rig or on a certain vehicle variant. For example, while measuring on the component test rig, it is important to verify that the source is in the same operating condition as if installed on the vehicle. For airborne sources, several methods exist to experimentally determine the source strength. They are called acoustic source quantification (ASQ) methods [3] [4]. The preferred ASQ method for the exterior PBN sources is matrix inversion [5]. Here one uses the near field acoustical transfer functions between the sources to be identified and a set of indicator microphones. These are combined with operational pressure measurements allowing to apply matrix inversion to estimate the source strength. { ( )} =

( )

∙ { ( )}

1  Q1 ( )   H11 ( ) H 21 ( )  H n1 ( )   p1 ( )        Q2 ( )    H12 ( ) H 22 ( )  H n 2 ( ) . p2 ( )                   Qn ( )  H1v ( ) H 2 v ( )  H nv ( )   pv ( ) 

with: ● ●

: the indicator microphone measured in operational condition. : matrix of near field acoustical transfer functions FRFs measured between source and indicators, using an acoustic source at the source locations. ● : source strength: a volume velocity for each source/monopole.

4

Predicting pass-by noise levels for trucks based on component test bench …

Figure 4 Schematic for ASQ using matrix inversion

For linear ASQ, the number of source patches depends on the maximum frequency band of interest. At higher frequency it is necessary to take more divisions into account. To obtain a stable matrix inversion, the number of indicators usually needs to be higher than the number of sources by a factor 2 to 3. The equation is solved using an unconstrained pseudo-inverse. In the PBN vehicle synthesis process, the idea is to go one step further: once the source is described by a set of monopoles, it can be recombined with a certain receiver to generate a virtual assembly. The receiver will be described by a set of noise transfer functions (NTFs) between the source locations and target microphone response locations . Multiplying these NTFs with the source loads generates the expected target response without having to physically integrate the source in the vehicle. Since the PBN run-up is a fast transient phenomenon, the method will most of the times be executed in the time domain [6]. By convoluting the measured time data with a set of FIR filters one can generate the expected target responds sound trace. ( ) =

( )∙

( )

● yk : represents the summed of the n sources contribution at the k target microphone. ● NTFki : the noise transfer functions between each load and each target.

5

Predicting pass-by noise levels for trucks based on component test bench …

The specific case of pass-by noise prediction requires a method that can handle a wide frequency range in combination with a broad band excitation (e.g. 150Hz – 8kHz). For this particular frequency range, earlier publications [5] show that the energetic ASQ approach is highly suitable. One of the main advantages of the method is that the source has to be divided in only a limited number of patches, which keeps the required instrumentation effort limited. Using instead the linear ASQ method, a very fine division of the source would be required. This method is rather recommended for coherent, low frequency harmonic source excitations. The scope of this paper is limited to energetic ASQ. The source strength can be identified experimentally with energetic ASQ using a constrained Least Squares solver: {

( )} =

( )

∙{

( )}

with: ● ●

: the indicator microphone autopowers measured in operational condition. : the frequency response functions FRFs measured using an acoustic source between the source locations and indicator microphones. ● : the resulting volume velocities for each source.

The method is based on the source-transfer-receiver concept that allows to evaluate the contribution of each noise source toward a target microphone. The pressure response at each pass-by noise target microphone location k can be expressed using this equation: ( ) =

( )∙

( )

● yk : represents the summed of the n sources contribution at the k target microphone ● NTFki : the noise transfer functions between each load and each target. To validate the approach, the energetic ASQ has been applied to identify the strengths of the sources on the test rig (ATS), but also on the full vehicle in the interior PBN facility, allowing to generate a full truck ASQ model (ATS, engine, tires, transmission, …).

6

Predicting pass-by noise levels for trucks based on component test bench …

2.2 Process Steps The PBN vehicle synthesis process consists of following steps: 1. 2. 3. 4.

Source Characterization Receiver / Vehicle Characterization Assembly Configuration Target Response Prediction

In the next chapter, these steps are detailed on an industrial application case. The PBN contributions of an exhaust ATS, an important contributor, is predicted using loads estimated on a component test bench. To validate the approach, each of the component variants has been built in the vehicle. The PBN levels where estimated at a full vehicle interior PBN test facility.

Figure 5 PBN Vehicle Synthesis process

2.2.1 Source Characterization The source is installed on a so-called ‘component test bench’ where it is the only source present and can be operated in a similar condition as if it would be installed on a vehicle. An energetic ASQ is performed on the component for all relevant operating conditions, resulting in a set of volume velocity time traces representing the source. (Alternatively the source can be installed on a certain test vehicle, and can be described in-situ). Ideally, the component test bench measurements are performed using the same operational driving profiles. On most engine test benches, it is not feasible to exactly replicate the fast pass-by noise run-up, hence the need to execute the process in the frequency domain instead of the time domain. Based on the time traces, a spectral load map in function of engine speed is generated. Since the source might behave differently when the run-up speed is changed, it is important to take into account the implications of this procedure: more on this in 3.2.2.

7

Predicting pass-by noise levels for trucks based on component test bench …

In practice, working on a component test rig has the huge advantage that the source strength can be estimated quickly for a large number of source variants before integration of the component in the vehicle. This speeds up and frontloads the development process.

2.2.2 Vehicle Characterization To characterize each vehicle, NTFs between each of the assumed source locations and the arrays of pass-by noise microphones are measured. A source component with the same shape is installed on the vehicle, assuring the invariance of these NTFs. The vehicle is installed in a dedicated interior PBN facility (or alternatively the transfer functions are measured on an outdoor PBN track). The NTF are measured with a mid-high frequency source.

2.2.3 Assembly Configuration To generate an assembly configuration a vehicle (step 2) can be combined with a certain source variant (step 1). In this way, a number of assembly variants can be generated. As a result, the acoustic loads are multiplied with the vehicle NTFs to generate contribution maps for each PBN array microphone. Next, the algorithm will use the expected PBN engine run-up speed to generate the contributions. Knowing also the vehicle position in function of time, one can synthesize the contributions for the left & right PBN microphone. Furthermore, one can even predict the contributions for an arbitrary engine speed profile using the source load maps. For example, how will the contributions be affected when the run-up speed is changed because of a change in the transmission or engine controls? The method allows to easily assess the effect of such a modification.

2.2.4 Target Response Prediction In the last step, one is able to quantify the overall PBN contributions of the component under investigation. Adding this source to the contributions for all other sources (intake, engine, tires, clutch, transmission, rear axle,…) allows to predict the total PBN level. For every source variant, one is able to check the change of the total PBN level. This allows to assess very early in the development cycle for a high number of component variants the PBN performance.

8

Predicting pass-by noise levels for trucks based on component test bench …

Figure 6 Main contributors for exterior noise of a heavy duty truck

3 Validation In this chapter, the complete process will be validated step by step to assess the accuracy, assumptions & practicability for an industrial application case.

3.1 Results: predicted vs. measured To assess the method, the predictions of the virtual assembly were compared with the levels measured after physically integrating the source in the truck. The display below shows the predicted PBN levels for the vehicle synthesis model on the left and the measured interior PBN levels for the full vehicle on the right. The curves are displayed for 4 vehicle assembly configurations, each with a different ATS variants original ATS (red), ATS variant A (blue), ATS variant B (green) & completely encapsulated ATS (brown).

9

Predicting pass-by noise levels for trucks based on component test bench …

Figure 7 Full vehicle predictions for the overall level of the right PBN microphone: 1. Predicted by the PBN Vehicle Synthesis model using the ATS test rig data (left) 2. Measured using the full vehicle on the indoor test facility (right)

Figure 8 ATS installed on the component test rig: 1. Original ATS, 2. ATS Variant A & 3. ATS Variant B

The predictions of the assembly resemble very well the measured responses. Even if we look in more detail to the relative effect of the modifications, the method predicts them in a highly accurate way. For example, the PBN vehicle synthesis predicts a reduction of -0.9 dB of the total PBN level for variant A, where also after applying the modification to the real truck also a reduction -0.9 dB was measured. A similar conclusion can be made for Variant B: -1.0 dB vs. -1.1 dB.

10

Predicting pass-by noise levels for trucks based on component test bench …

Furthermore, we notice that the results from the PBN vehicle synthesis are smoother than the actual measurement results. This can be explained by the use of the method in the frequency domain as explained in 3.2.2. If the predicted levels are deviating from the design targets, the model allows to pinpoint the root cause of the overshoot. The method results in the partial contributions related to the ATS, which allows selecting the optimal countermeasure. For example, it is possible to separate the noise coming from the exhaust outlet and the one coming from the shell (Figure 9). Looking at these partial contributions, it is important to remember the assumption used by energetic ASQ. It is designed to separate well the contributions of incoherent sources [6]. In this application case, this assumption is validated by checking the coherences of the identified loads (Figure 14).

Figure 9 Exhaust After-Treatment-System ATS contributions towards the right PBN microphone: 1. Estimated using the PBN Vehicle Synthesis model using the ATS test rig data (left) 2. Estimated using energetic ASQ for PBN on the full vehicle in the indoor PBN test facility (right)

The virtual total ATS contribution (red, left) matches well the contribution calculated for the ATS in the full vehicle (red, right). The partial contributions are, as expected, less accurate but still give an indication of the relative importance of each ATS subpart. Especially if we look at the effect of the applied modifications (Figure 10), these partial contribution maps give great insight in the efficiency of the applied counter measure. If we compare ‘original ATS’ with variant A, the outlet, which is the main contributor, is affected with a reduction of 7-8 dB. If we look further at the comparison between variants A & B, it is clear that only the modified panels, right and back, are affected by the adaptations. The overall contribution of the ATS for variant A & B is quite similar since the main contributors are not affected.

11

Predicting pass-by noise levels for trucks based on component test bench …

Figure 10 Exhaust ATS contributions towards the right PBN microphone estimated using the VVA model using the ATS test rig data : 1. Original ATS, 2. ATS Variant A: Exhaust take off connected & 3. ATS Variant B: Exhaust take off + Encapsulation of the right/back panels

The contribution results show a deviation starting from 7-8m vehicle position (greyed out zone, Figure 7). This deviation has no significant impact on the maximum overall PBN level. Nevertheless, this is currently subject of a further investigation. Looking into the physical differences between the test rig and the full vehicle setups, there are possible causes for this deviation: ● The rear tires were encapsulated during the full vehicle ASQ to estimate all acting contributions on the truck (engine, tires, intake …) used for the PBN vehicle synthesis model. During the validation measurements, these tires were not covered, causing an increase of the total PBN level from 8m onwards ( Figure 7 -right side).

12

Predicting pass-by noise levels for trucks based on component test bench …

Figure 11 During the measurements on the full vehicle in the indoor PBN test facility for the creation of the full vehicle energetic ASQ model the rear tires were partially covered

● In the ATS component test cell, differences are expected due to potential different loading and suspension of the component. Unexpected resonance behavior was present in the high rpm range: this caused increased excitation levels which generate increased load estimations (Figure 7 (left side)). Overall these results demonstrate the potential of the method: PBN vehicle synthesis allows to predict the PBN level contributions for each ATS variant.

3.2 Energetic Acoustic Source Quantification for heavy duty trucks To build up the PBN vehicle synthesis models used in the previous sections, the results from a full vehicle and component test bench ASQ were used. These are described in detail in the next paragraphs.

13

Predicting pass-by noise levels for trucks based on component test bench …

3.2.1 Full Vehicle Energetic ASQ for PBN A full PBN ASQ was performed for the complete vehicle: the contributions of all relevant contributions towards the left and right PBN microphone array (engine, clutch, transmission, tires, intake, ATS …) were estimated.

Figure 12 Installation of the truck in the indoor PBN facility / LMS Q Source installed at source location for the simultaneous measurement of near & far field transfer functions

A number of indicator microphones were installed around all significant sources. Simultaneously, the PBN microphones arrays were measured during several PBN runups of the vehicle. Between all source locations and indicator / target microphones, the transfer functions were measured using a mid-high frequency LMS Q Source in the range of 100 Hz – 8 kHz (Figure 12). In this way, all data was gathered to apply the energetic ASQ approach. Displayed below, one can find the resulting contributions of each of the sources towards the synthesized left and right target PBN microphone. This approach is generally used for PBN engineering to pinpoint in a reliable and fast way the main contributing sources.

Figure 13 Overview of the contribution estimated using energetic ASQ for PBN – the total contribution matches well the actual measured PBN levels.

14

Predicting pass-by noise levels for trucks based on component test bench …

The match between the measured target responses and the estimated response based on the total contribution of the ASQ model is very good. This validates the quality of the model. It is within the 0.5dB range which proves that the method is accurate for truck PBN engineering (Figure 13). To further fine-tune the energetic ASQ model, the sensitivity of following model parameters were investigated: 1. The number of patches: each of the main contributors is divided in a number of patches. The number of these divisions varied to come to an optimal number of patches for the exhaust ATS. 2. The number of indicator microphones: to have a stable load estimation, it is typically required to have minimum 2 to 3 times more indicators than sources. Specifically for the ATS, the minimum set of indicators was derived without significant loss of accuracy on the load estimations. 3. Energetic ASQ assumption: the method assumes that each of the sources in uncorrelated with the other sources. If this is not the case, one can only trust the total contributions of these correlated sources. This assumption was verified by looking at the coherences between the identified loads. This verification helps to divide the contributors (e.g. ATS, engine…) in the correct number of sub-sources. In the end, the ATS was divided in 7 sources: 6 sides + 1 outlet. A way to fast check the validity of this assumption is to visualize the coherences in a matrix display as depicted in Figure 14.

Figure 14 Matrix display showing the correlation between the estimated loads – energetic average between 100 Hz – 8kHz

Out of these results, estimations were derived for each of the exterior noise contributors. Especially the contributions of the ATS are of interest since these can be later on be compared with contributions calculated based on the ATS test rig measurements.

15

Predicting pass-by noise levels for trucks based on component test bench …

3.2.2 ATS Component Test Rig The ATS is installed on a full anechoic ‘component test bench’ where it can be operated in a similar condition as if it would be installed on a vehicle. An engine, installed on a test rig in the adjacent room, excites the ATS via a special inlet tube (alternatively the component is installed in a test vehicle for in-situ characterization) (Figure 15). An energetic ASQ is performed on the component for all relevant operating conditions, resulting in a set of volume velocity time traces representing the source strength of the ATS. The source strength was used in the PBN vehicle synthesis model to estimate the total PBN level.

Figure 15 Exhaust After-Treatment-System ATS component test rig: engine test rig (left) – full anechoic ATS test bench (right)

Again, a number of indicator microphones were installed around the ATS. To check the accuracy of the source model, 3 extra target microphones were installed in the far field in the room. These were not used to estimate the loads. Several engine run-ups were performed. Afterwards, the transfer functions between source locations and indicator & target microphones were acquired to complete the ASQ model. Using the model, the loads and target contributions were estimated. The measured target and total contribution could be used to validate the acquired model (see Figure 16 -red & green curve).

16

Predicting pass-by noise levels for trucks based on component test bench …

Figure 16 Estimated overall level of the ATS loads in function of rpm

A slow run-up of 15 seconds was performed on the component test bench, since it was not possible to perform the run-up as fast as the real PBN procedure on these type of component test rigs (Figure 17). This is why PBN vehicle synthesis executes the contribution calculation in the frequency domain instead of the time domain.

Figure 17 Engine speed profile in function of time comparison between: full vehicle PBN run-up (green) vs ATS test rig run-up (red)

17

Predicting pass-by noise levels for trucks based on component test bench …

An in-depth analysis of the effect on the excitation is ongoing but first results show that the loss of accuracy is acceptable for most tested conditions, even if the run-up speed is reduced. Still caution is advised, since numerous effects can cause differences in the excitation behavior (engine controls, etc.…). If we limit the analysis to the effects of the processing in time vs. frequency domain, we can see that there is a limited effect on the calculated contributions: using the identical time data set, the contributions were processed (see Figure 18, in time domain (right) vs. frequency domain (left)).

Figure 18 Truck PBN contributions were processed using the same dataset in time domain (right) vs. 1/3 octave frequency domain (left)

In the end, these loads were re-combined with full vehicle NTFs to derive the expected contributions towards the PBN microphones. The process was repeated for the 3 ATS variants. As shown earlier, these test bench-based contributions match well with the full vehicle ASQ contributions although they are based solely on component test bench measurements (Figure 7). Recombining these contributions with the other vehicle sources gives a view on the total expected PBN level for this ‘virtual assembly’.

4 Conclusions Complying with the new, more restrictive pass-by noise regulations will be a challenging task for truck manufacturers. The truck OEM will need to “reinvent” their development process, trying to frontload NVH performance engineering to set design targets early in the development phase. The PBN vehicle synthesis method described in this paper addresses this need and has been validated in-depth for an important contributor to the exterior noise: the exhaust ATS.

18

Predicting pass-by noise levels for trucks based on component test bench …

The method accuracy has been demonstrated by quantifying the source strength on a component test rig and then virtually integrating it in the vehicle using an energetic ASQ model. The model allows to predict the expected PBN levels. These were proven accurate by comparing them to the measured levels after integrating the source in a test vehicle. It also confirms the invariant load characteristic of the acoustical loads identified on test rig and in-situ environment. This allows to use them in any arbitrary assembly to predict their contributions. Executing this process early for several exhaust ATS prototypes (>10) and vehicles limits efforts to select the best design for further optimization. The design can be iteratively improved throughout the development cycle. The process helps keeping control of the expected PBN levels. In the end, it reduces the required testing work and the need for troubleshooting / expensive modifications in the late integration phase.

5 References [1] “UN/ECE regulation 51.03 and EU regulation (EU) 54/2014”. [2] “ISO 362:2015. Measurement of noise emitted by accelerating road vehicles”. [3] J. Verheij, “Inverse and reciprocity methods for machinery noise source characterization and sound path quantification, part 1: sources,” International Journal of Acoustics and Vibration 2, pp. 11-20, 1997. [4] S. Goossens, T. Osawa and A. Iwama, “Quantification of Intake System Noise Using an Experimental Source-Transfer-Receiver Model,” in SAE, Noise and Vibration Conference, Traverse City, Michigan, USA, 1999. [5] K. Janssen, F. Bianciardi, L. Britte, P. Van de Ponseele and H. Van der Auweraer, “Pass-by noise engineering: a review of different transfer path analysis techniques,” in ISMA, Leuven, Belgium, 2014. [6] K. Janssens and F. Bianciardi, “Time-domain ASQ method for pass-by noise engineering,” in Proceedings NOISE-CON 2013, Denver, Colorado, USA, 2013.

19

NVH development strategies for suspensions – challenges and chances by autonomous driving Andreas Schilp, Managing Director, Prof. Dr. Hartmut Bathelt, AZL Technology Center, Germany

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 W. Siebenpfeiffer (Hrsg.), Automotive Acoustics Conference 2017, Proceedings, https://doi.org/10.1007/978-3-658-20251-4_18

1

NVH development strategies for suspensions – challenges and chances …

1 Introduction The safety aspects of autonomous driving, the limiting risks of complex situations or weather conditions and the legal consequences are the present focus of discussions. Once this technique will have become an every days practice in our live, automotive developers will face a new world of customer expectations. After pressing the push bottom “autonomous”, the driver becomes a passenger who is everything else than amused by sporty manner of driving. On a long distance ride, busy with preparing a business presentation or playing games on a tablet with the kids, passengers expect the same comfort of a high speed train. Especially road noise and micro shake on highways make up for a big difference even with the best of our nowadays cars.

2 Dynamic forces and accuracy limitations in their analysis Dominating factor of road comfort are dynamic forces in the frequency range between 20 and 500 Hz introduced to the car body at all connection points of the suspension. The transfer path analysis (TPA) of structural noise as first step of NVH-development is based on the direct or indirect determination of these forces. Table1. Dynamic Force Determination Approaches

2

NVH development strategies for suspensions – challenges and chances …

2.1 Indirect force measurement Success or failure in the force analysis is dependent on reliable measurements of the physical source of information in each of these approaches. By practical reasons we usually do not measure the actual physical quantities the forces are calculated from, but use accelerometers, which are easy to fix, and point accelerations can additionally be used for displaying operating mode shapes. Both indirect force determination methods – stiffness method and matrix TPA – reach their limits when the accelerations caused by local force input are masked by superimposed global mode accelerations. With the stiffness method this is the case when a stiff mount connects two parts with high accelerations. Then the vibration level on both sides of the mount is about the same and the difference is not a reliable measure for the mount deformation. The inertance method –called matrix TPA- using the local deformation of the body structure caused by a force impact gets a poor data offer, when points on a stiff structure show almost the same inertance curve. A paper presented 4 years ago on this conference by Daimler [1] could show, that strain gauge measurements at the force input point would deliver much higher resolution data for the force calculation in such cases. The inevitable scatter of measurements under operational conditions leads to a calculation with an unreliable data basis and finally questionable results. In these situations a direct force measurement is a way out.

2.2 Direct force measurement Installing piezo force transducers into the force flow are the technique with the highest dynamic resolution and delivers precise results when indirect techniques are prone to failure [5]. The only problem is: the contact areas on both sides of the force sensor have to be plane and precisely machined. Irregular pressure distribution by unevenness or deformation can strongly influence the force signal. The application in test cars (driven on the road or rough surface rollers) therefore is only practicable on a few special points between suspension and body, like damper strut mount, or the example below:

3

NVH development strategies for suspensions – challenges and chances …

Picture 1. Installation of triaxial piezo force sensors on a whishbone bush

4

NVH development strategies for suspensions – challenges and chances …

2.3 Suspension test stand with direct force measurement

Picture 2. Suspension Force Rig on 3,2mø- road noise rollers

These difficulties in determining suspension forces finally led to the development of a so called “force rig”, replacing the real car body by a stiff, massive structure with integrated force sensors at the suspension attachment points.

5

NVH development strategies for suspensions – challenges and chances …

Table 2. Force Rig Features Overview

Picture 3 and 4. Mounting of force sensors (yellow) between adaptor plates

The pictures above show the mounting situation of the yellow force transducer at a subframe bush. Specially milled adaptors (blue) care for central positioning and equal pressure distribution at the support surfaces of the sensor.

6

NVH development strategies for suspensions – challenges and chances …

A similar force rig has been presented at this conference two years ago by Vibracoustic [2]. The excitation there is introduced by servo-hydraulic shakers, so basically functioning as an elastomer test stand with bushes being tested under real mounting conditions. Since we are investigating the complete suspension system and need to see a force-TPA, we – in difference to that- use real road excitation with wheels running on 3.2mø- road noise rollers. In picture no. 2 we can see our first version of the force rig operating. The design is determined by the objective to accept different kinds of passenger car suspensions, front or rear, various car models. What could not be realised on test cars, is now possible: a direct validation of the road noise quality of competitors suspensions without the influence of the body. Furthermore an objective separation between suspension and body structure contribution to a road noise problem is possible, an old discussion issue in car development history. One question we frequently are asked is: what`s the eigenfrequency of your force rig structure ? This implements the suspicion, a resonance would influence or change the measured forces. This is not the case so long the accelerations body side, means force rig side, are at least 20dB lower than at the suspension side. Picture no.5 shows a typical example measured at rear subframe mounts.

Picture 5. Vibration level difference between suspension parts (subframe, black) and foundation (Force Rig, green), 3 directions x,y,z (top → down)

7

NVH development strategies for suspensions – challenges and chances …

Above 50 Hz we can see more than 20 dB difference – certainly more than on a real car body would be measured – a guarantee that there is no backlash by coupling of body and suspension. This is the situation on almost all attachment points with rubber bushes. An example where we have to make a critical check will be discussed later (see point 3.4).

2.4 Modified suspension test stand for increased efficiency

Picture 6. Latest version of Suspension Force Rig on 3,2mø-road noise rollers

Next development step to speed up force analyses are separate mounting plates with integrated force transducers which stay in place while suspensions or components are attached. The risk of mounting influences or damage to the sensible transducers is minimised by this concept.

8

NVH development strategies for suspensions – challenges and chances …

The extremely low acceleration level at the rig structure is the justification for neglecting the inertia effect of the mounting plates to the transmitted force. Picture no.6 (video at the conference) shows the latest force rig version under operation,

Picture 7. Sensor position between mounting plate accepting a spring-damper top and force rig

The following measurements have all been made on the force rig rolling on rough road shells. Progress in the exactness of force analyses are one precondition for an optimised suspension tuning. The restrictions in reducing these forces nevertheless are remaining.

9

NVH development strategies for suspensions – challenges and chances …

3 Limits in the reduction of structural road noise forces to the body Why is road noise the dominant interior noise even on premium cars?

3.1 NVH-development of front suspensions Compared to the continuous progress in NVH development the reduction of road noise seems to approach a barrier in the last 10 years.

Picture 8a. Transverse wishbone and control arm for longitudinal elasticity

Looking back historically the finding of the 70`s was the longitudinal elasticity of the wheel suspension as the key to better ride comfort. Therefore a target in suspension design of premium cars was to achieve an elastokinematic concept allowing longitudinal travel of the wheel without any change of the steering angle. The functional separation of cornering forces reaction and longitudinal elastic travel at the front suspension is the consequent realisation of this idea. The transverse wishbone is a stiff connecting rod taking the load of the cornering force and its mounting point is the turning centre of the fore-aft movement of the wheel carrier. The amount of fore-aft movement is determined by the diagonal strut and the elasticity of the big rubber bush bodyside, therefore called control mount and control arm.

10

NVH development strategies for suspensions – challenges and chances …

3.2 Recent change of front suspension concepts While suspension elasticity in longitudinal and vertical direction was exhausted for the reduction of dynamic forces to the car body we could observe a contrary tendency in transversal direction especially on front axles. This can be explained by the fact that two priorities became dominant in the last 15 years: handling precision and weight reduction. Both of them led to abandoning the concept of a rubber mounted front subframe. It was replaced by a rigidly mounted structure eliminating the elasticity of the rubber bushes and additionally stiffening the front end in transversal direction. The change in front suspension design is obvious looking at picture no.9: the tube frame with big rubber bushes in the upper picture is replaced by a pressed sheet connection of the longitudinal front end beams, stiffening the front end in transversal direction. The steering has been shifted from the firewall down to wheel centre level resulting in radically increased steering stiffness.

11

NVH development strategies for suspensions – challenges and chances …

Picture 9. Change of front subframe and steering

12

NVH development strategies for suspensions – challenges and chances …

This is only one example for a general tendency. The current front wheel guidance concept of all three German premium car manufacturers basically show common features: ● The front subframe is rigidly mounted to the front body without any rubber bushes. ● The lower wishbone triangle is built by a diagonal strut (blue arrow) with a soft “control mount” and a transverse wishbone, see picture 8a. The control mount determinates the elastic movement of the wheel in longitudinal direction. The transverse wishbone takes the major part of the cornering forces of the wheel. This wishbone bush therefore is one of the stiffest in the suspension.

Picture 8b. Longitudinal guidance of wheel carrier

● Transverse wishbone and track rod form a parallelogram piloting the wheel carrier. Driving straight on they allow an elastic longitudinal movement without distracting steering angle- a crucial feature for low road noise. It is obvious that any elasticity in the arms of that parallelogram would have negative influence on steering precision.

13

NVH development strategies for suspensions – challenges and chances …

Picture 10. Longitudinal wheel guidance with curved control arm (routed backwards)

Picture 11. Longitudinal wheel guidance: Daimler C-class

14

NVH development strategies for suspensions – challenges and chances …

3.3 Consequences to road noise comfort Now we come close to the core of the road noise problem: with the rigidly mounted front subframe and the extreme stiffness of the transverse wishbone bush we end up with a transmission path without any insulation to the body. The same situation at the other arm of the steering linkage, the track rod connecting the wheel rigidly with the steering gear, which is rigidly bolt to the subframe. The consequence can be seen comparing accelerations wheelside and forces bodyside in 3 directions: although the excitation in transverse direction y is not higher than in z and x-direction, the forces to the body are almost 20dB higher at frequencies up to 200Hz

Picture 12. Acceleration at the active side (black curve) of the transverse wishbone bush representing the excitation introduced from the wheel, left and right, longitudinal (x), transversal (y) and vertical (z) direction (top → down) green: acceleration at the body side ////////: difference resulting from lower stiffness in axial (y) and high stiffness in radial bush directions (y,z)

15

NVH development strategies for suspensions – challenges and chances …

Picture 13. Forces acting to the body at the transverse wishbone attachment point, longitudinal (x), transversal (y) and vertical (z) direction (top → down)

16

NVH development strategies for suspensions – challenges and chances …

3.4 Stiffness of transverse wishbone bush Stiffness measurements of the wishbone bush in radial direction confirm these results.

Picture 14. Transverse wishbone bush

With radial stiffness values of 5000 to 20.000 N/mm it comes close to the usual structural body stiffness. Together with the rigidly connected subframe we see a direct force transmission from wheel hub to body.

17

NVH development strategies for suspensions – challenges and chances …

Picture 15. Effect of transverse bush stiffness reduction from 20.000N/mm to 6.000N/mm

The spring-mass-model, applicable to soft suspension bushes “half stiffness = half transmission force” is not valid in this case. A stiffness reduction from 20.000 to 6.000 N/mm does not change the force, only the local acceleration level is increased. This is the transmission behaviour of a chain of springs with no reduction of the induced force.

3.5 Barrier for further improvements of other transmission paths: With one transmission path dominating and physically at its limits, the progress in reducing forces in longitudinal and vertical direction or at the other transmission points only can lead to marginal improvements which cannot really be felt by an average customer on the road. By the same reason a lower road noise contribution of the rear suspension with rubber mounted subframe is masked by the noise of the front suspension, see picture no. 16. The excitation front to rear – represented by the wheel acceleration, picture no. 17 – is almost the same and would not explain the difference in interior noise.

18

NVH development strategies for suspensions – challenges and chances …

The high priority of direct steering response and the race to outperform the competition in handling features are a simple explanation why NVH-engineers have been unable to break this sound barrier up to now.

Picture 16. Comparison of interior road noise contribution of front and rear axle

19

NVH development strategies for suspensions – challenges and chances …

Picture 17. Wheel hub accelerations in 3 directions, comparison of front and rear axle

4 Change of customers priorities on autonomous driving Once step 5 in autonomous driving is in common use, the subjective feedback to the drivers steering input is not an issue anymore. The handling- “sound barrier” cease to exist in the autonomous mode. The computer controlling the ride has a so much shorter reaction time than a human driver and will easily cope with elasticity in the action line. A completely different suspension tuning will be the adequate use of this new scope. Now many of the NVH-solutions developed in the past, but not realized because of handling drawbacks, can be retrieved from the treasury box.

4.1 Switchable suspension bushes Active driving nevertheless will still keep its attractiveness for the customer in all situations where it is enjoyable or necessary. Passenger cars for out of city traffic will have to offer nowadays sportive handling additionally in the active mode.

20

NVH development strategies for suspensions – challenges and chances …

The key to provide two different suspension tunings in the same car are switchable suspension bushes. Already 3 decades ago switchable engine mounts were the key to open the way for the direct injection diesel engine from the truck into passenger cars. In the example of the front subframe the solution would be to go back to a rubber mounted subframe with mechanically lockable subframe mounts.

4.2 Consequences to front subframe design This does not mean to copy an old subframe design. The future subframe will have to cope with two different requirements: a structural stiffener for the front body with locked mounts and a decoupled subsystem with maximum insulation. The first function needs just stiffness in the horizontal plane taking compression tension and shear stress.

Picture 18. Example for a front end stiffening subframe

Its not only the increased static stiffness of the front end in transverse direction for reacting cornering forces. Downsized combustion engines with much higher excitation of low frequency vibrations also need a higher dynamic stiffness of the body structure below 50Hz. A rigid connection of the longitudinal front end beams helps to prevent or reduce a breathing mode of the beams excited by engine mounting.

21

NVH development strategies for suspensions – challenges and chances …

Picture 19. Example for a tube frame (BMW)

The second function needs maximum 3dimensional dynamic stiffness to raise the first torsional and bending mode frequency to 130 Hz upwards. None of the present or past designs will fulfil both specifications because they have been developed for only one of these targets. A new design can be expected as a cast aluminium lightweight structure that combines a tube frame with a shear field in the middle part. With the chance to use much softer subframe bushes as in former times the weight of a steel frame is not necessary anymore to achieve sufficient insolation even at low frequencies.

22

NVH development strategies for suspensions – challenges and chances …

4.3 Rear subframe Nowadays rubber mounted rear subframe structures will basically not change. Since the vertical elasticity of the mounts in this case does not have a critical influence on steering precision, switchable mounts will mainly change the stiffness in the horizontal plane. This can be achieved by valves in an internal hydraulic system or by activating mechanical switches limiting the elastic deformation.

4.4 Upper wheel suspension Adaptive dampers already offer a variable ride behaviour and will have to widen the spread between “comfort” and “sport”-adjustment. Depending on the effectiveness of this development on reducing transmission forces, the damper top mount could become an adaptive element too. The same consideration could be applied to the upper wishbone plane on multilink front suspensions.

5 Chances for improved body shake Switching to softer suspension bushes in the autonomous mode will disclose new chances for an optimum tuning of the coupled system wheel – powertrain mass. A research work already 17 years ago [3] proved, that reducing the resonance frequency of the wheel mass to that of powertrain-engine mounting, can lead to a force cancellation with the powertrain acting against the wheel mass. The necessary modifications were reduced stiffness of the wheel suspension combined with less damping of the hydraulic engine mounts. This research result could not be transferred into production cars at that time because of the well-known restrictions.

23

NVH development strategies for suspensions – challenges and chances …

Picture 20. Force effects of front suspension and powertrain mass to body (series condition).

The first picture shows the force interaction in the initial condition. The black curve representing the powertrain force is antiphase in the first period, but too small to cancel the blue force of the front suspension. After 2 oscillations the powertrain mass acts in phase with the suspension because of its lower frequency.

24

NVH development strategies for suspensions – challenges and chances …

Picture 21. Force effects of front suspension and powertrain mass to body with reduced stiffness of front suspension and stiffer engine mounting with less damping.

By reducing the suspension stiffness powertrain and suspension are vibrating with the same frequency and the antiphase constellation is kept permanently. In order to increase the amplitudes of the black curve the damping of the powertrain mounting had to be reduced. The effect is the breakdown of the red oscillation of the body after 1 ½ periods.

25

NVH development strategies for suspensions – challenges and chances …

Picture 22. Body shake acceleration: improvement by powertrain mass acting antiphase to front suspension. Excerpt from [3].

In the frequency domain the breakdown of the red curve of the body acceleration looks like better damping but is due to force cancellation. The driver on real road describes his impression as a very “compact feeling” of the car rolling over an unevenness.

6 Further possibilities with electric vehicles Pure electric vehicles are less prone to body shake problems due to the lower powertrain mass and stiffer engine mounting.. Electric engines – especially permanently excited synchronous engines as frequently used – do not produce low frequent vibrations like combustion engines. This can be the basis for an optimised electric powertrain mounting as demonstrated by the paper of Vibracoustic [4] in section “NVH of EV”. A comparatively stiff connection of the electric engine to the subframe can additionally be used for an increased insulation of structural road noise, if soft subframe mounts are activated on autonomous driving. In the frequency range where the engine mass is coupled to the mass of a subframe it could multiply the resistance to road induced vibrations.

26

NVH development strategies for suspensions – challenges and chances …

7 Conclusion While with current passenger cars considerable improvement of road noise comfort cannot be expected due to targets like steering precision and weight reduction as a priority, the special conditions of autonomous driving are the chance to enter a new NVH-comfort level, when adaptive suspension bushes enable the switch from current crisp handling to autonomous cruising. On the example of the subframe an outlook can be estimated, how this expected development will influence future chassis design. This forecast cannot claim detailed future design features, but is based on predictable clients demands, and the fact, that basically necessary components like switchable rubber mounts or the electronic control of adaptive dampers are on the market and need only adaption to a new application. The comfort potential does not only include structural road noise, but also body shake, as the example of tuning the wheel mass resonance and powertrain mounting could demonstrate.

Index [1]

Accuracy of inverse load identification techniques for transfer path analysis Daniel Sachse, NVH Test Engineer, Daimler AG, Germany, T. Geluk, T. v. Wayenberge, LMS International, Belgium 2nd International ATZ Acoustics Conference, 26 and 27 June 2013

[2]

Chassis system NVH – advanced analysis methods Dr. Christoph Rambacher, Director Vehicle Testing and NVH, Dr. H. Sell, TrelleborgVibracoustic GmbH, Germany; B. Reff, J. Palandri, Ford Motor Company, Germany 3rd International ATZ Conference, 23 and 24 June 2015

[3]

Ohlendorf, J., Pfeffer, P., Bathelt, H., Wodtke, H.W.: Komfortsteigerung durch optimierte Kopplung von Aggregat- und Fahrwerkslagerung. ATZ 102 (2000) 5, S. 326-332

[4]

Powertrain mounting systems for electric vehicles David Roth, System Simulation Specialist, Dr. T. Ehrt, C. Bultel, H. Kardoes, Vibracoustic GmbH & Co. KG, Germany 4th International Automotive Acoustics Conference, 11 and 12 July 2017

[5]

Bukovics, J.; Vogel, F.; Bathelt, H.; Wodtke, H.-W.: Improving Ride Comfort by Direct Measuring of Vibration Forces ATZ 100 (1998) 8, S. 532-539

27

NVH development strategies for suspensions – challenges and chances …

Authors Andreas Schilp, Managing Director, Prof. Dr. Hartmut Bathelt, AZL Technology Center, Germany Zürich / Rüschlikon, Switzerland, 11 and 12 July 2017

28

A framework for the sensitivity analysis of transfer paths combining contribution analysis and response modification analysis Dr.-Ing. Dejan Arsić, Müller-BBM VibroAkustik Systeme GmbH Matthias Pohl, Müller-BBM VibroAkustik Systeme GmbH

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 W. Siebenpfeiffer (Hrsg.), Automotive Acoustics Conference 2017, Proceedings, https://doi.org/10.1007/978-3-658-20251-4_19

1

A framework for the sensitivity analysis of transfer paths combining contribution …

1 Introduction The Transfer Path Analysis (TPA) has been thoroughly investigated in the past, and a wide range of different approaches has already been implemented [1]. Each approach has its own properties and possible application cases. These range from a physically complete description of the entire assembly, including contributions and forces, a description of individual components for dynamic sub-structuring, to traditional troubleshooting applying operational measurements only. All approaches have their own place in the R&D or production process, and are fully integrated into the processes at most automotive OEMs. The overall goal is to fulfil predefined targets, which might be based on regulations, design decisions or comfort. One of the key points of the TPA is to determine the individual contributions of active sources and the quantification of transfer paths. This insight is subsequently used to modify sources and paths in order to reach the previously defined targets. Depending on the complexity of the structure, this can be an iterative process, as depicted in figure 1a). Experience has shown, this process can be troublesome, as it is not always clear how the entire structure reacts on a modification. A rather dominant source may be the largest contributor at the response position of interest, but be rather insensitive to modifications, resulting in small to none changes of the observed signal. Furthermore, it is not always possible to modify every source or path. Hence, a method is desired which guides the engineer in the modification process to avoid unnecessary and time consuming iterations. In order to cope with this challenge, the so-called Response Modification Analysis (RMA) is applied. The general idea is to manipulate the original response signal and adjust it this way that it fulfils the defined target. This can be lowering or raising of, for example, a bandpass level or a specific order. The modified response is fed into the previously defined TPA network, which will provide the sensitivity to changes of sources and transfer paths. The sensitivity analysis is a reliable indicator for the effect of changes of individual components in the overall system at the response position. Integrating the RMA into the R&D process, results in less iterations and subsequently in faster results.

Figure 1: The iterative development process with the OTPA in the loop

2

A framework for the sensitivity analysis of transfer paths combining contribution …

The present treatise will start with a brief introduction to the Operational Transfer Path Analysis (OTPA), including some general thoughts about the transfer path problem itself. Subsequently, the principle of the Response Modification Analysis will be explained, containing some remarks on the use and hints for practical usage. Examples from the automotive industry will provide practical insight of the method, followed by a conclusion and an outlook to further developments and improvements.

2 Operational Transfer Path Analysis 2.1 The Transfer Path Problem A simple model of a dynamic assembly AB, as depicted in figure 2, can easily describe the transfer path problem. The active component A is rigidly connected to B at the interface node and induces the contained excitation into the passive system. In order to describe the system, the following points have to be considered: ● The degrees of freedom (DOF), creating the excitation in the active component, which are usually not measurable ● The interface DOFs in-between the active and passive component ● The DOFs observed at the response position at the passive side It is assumed that the excitation at the active component, which cannot be measured directly, is transferred via the interface node to the response positions at the passive side. In order to describe the observations and responses, a set of transfer functions, containing both airborne and structure-borne sound, is required. Considering the assembled system, the spectra at the receiving position , created by excitations at node 1, are of interest.

Figure 2: The transfer path problem depicted with the assembly AB of a passive and active component, connected at a rigid interface

3

A framework for the sensitivity analysis of transfer paths combining contribution …

This is simply obtained by superposition of the individual contributions, convolved with the transfer functions contained in matrix ( ) = ∑

⟹

=

(1)

It should be noted that the responses might contain any quantities, including sound pressure, velocity, acceleration or displacement.

2.2 Operational Transfer Path Analysis A wide range of applications does not require any knowledge on the interface forces, but is solely relying on the individual contributions of paths and sources. These can be computed from most of the known TPA approaches [1], where the Transfer Path Synthesis (TPS) is used to calculate the contribution of each source to the response position. Transfer functions, characterizing the propagation of both sound and vibration from the source to the response position, are required for the synthesis. Nevertheless, it is desired to compute these from operational measurements only, where the OTPA is used [2]. Multiple operational measurements in different loads, where excitation and response are measured simultaneously in operational mode of the object, are the input for the OTPA. The relation between reference signals ( ) and response signals ( ) is described as a linear system (

) (

)= (

)

(2),

where ( ) is the transfer function matrix and the frequency dependency is denoted by ( ). The so-called time-synchronous short-term Fourier transformation (STFT) is used for the representation of the measurement. The transfer function matrix ( ) can be determined by solving eq. 1 for ( ), where the inverse of matrix ( ) has to be computed. With observations and response positions, eq. 2 can be solved for H with: ℎ( ⋮ ℎ(

( )

)

= )

⋮

( )

⋯ ⋱ ⋯

( )

⋮

( )

( )

⋮

( )

(3)

In order to be able to solve this equation, the number of SFTFs needs to be larger than the number of reference signals. The overdetermined least squares problem is solved by using a Singular Value Decomposition (SVD), which is able to remove some of the crosstalk between the reference positions, which usually occurs when a source influences the reference signal of another source. Noise is typically removed by neglecting small singular values. The determined transfer functions are used to design finite impulse response filters (FIR filters). These can be applied on the time raw data of the measured reference signal, resulting in the synthesized contribution of each reference position to the response signal. As the convolution is done in the time domain, further

4

A framework for the sensitivity analysis of transfer paths combining contribution …

analysis of the contribution can be performed and a playback of sounds is possible. A detailed description of the estimation of transfer functions is given in [3]. The OTPA, as described by Yoshida in [4], is usually considered as troubleshooting tool. Its aim is usually the quantification of individual sources and the information provision on potential improvement. Nevertheless, the OTPA has been already fully implemented into the development process of exterior noise measurements at indoor test benches in the automotive industry. As previously described in [5], it is used to rank the contribution of tires, intake, engine, and exhaust system to the overall noise perceived at the pass-by microphone.

3 Response Modification Analysis 3.1 Combining OTPA and RMA The RMA can be considered as extension of the OTPA, as described in section 2.2. By adding a modification Δ to the response signal , i.e. lowering or raising the original observation, a slightly altered set of transmissibilities = + Δ can be computed with ( ) ( )

=

⋮

( )

⋯ ⋱ ⋯

( )

⋮

( )

( )

⋮ +Δ ⋮

(4)

( )

The slightly different set already shows that not every path in the structure reacts equally to the modification. Hence, small changes, here a small Δℎ , are indicators that modifications of the path or component do not have any impact on the observed response and are rather insensitive to changes. Once more, it is evident that countermeasures at a source with a high contribution to the response signal are not necessarily promising. With the RMA, it is now possible to estimate the probability of successful modifications of sources and paths, which would be low in the case that the energy induced into a path is simply rerouted through another path. The RMA is consequently often also regarded as sensitivity analysis. One of the main advantages of this approach is that it is a rather efficient method, adding negligible effort to the OTPA, which is conducted anyways. It uses an extended TPS network with additional modified channels. This means that only the step of modifying the responses is actually added to the process, while the rest remains the same. In addition to the path contributions, an information on path sensitivities is gathered by applying virtual modifications of the original signal.

5

A framework for the sensitivity analysis of transfer paths combining contribution …

Figure 3: Exemplary modification of the 4th order of an exterior noise measurement. The original signal is shown in red, the modification in green. The blue line depicts the rpm.

3.2 Signal Modifications Prior to the sensitivity analysis, the response channels have to be modified. While the classic troubleshooting approach usually suggests lowering the levels of sound or vibration in specific frequency bands or orders, the RMA also allows raising levels. This has two advantages. On the one hand, it allows the fitting of the measured response to a specified target, which may be considered as a kind of audio design. On the other hand, experience has shown that lowering the level in the specific frequency range provides a fitting virtual response, but results in a signal too close to the noise floor. This may provide misleading results. Therefore, it seems reasonable to increase the level in order to compute the sensitivity of a path, as its sensitivity to modifications is computed no matter how the response is altered. Currently, two different modification methods have been implemented: ● Modification of band pass levels in a specific frequency range ● Modification of selected orders, which allows an analysis in the rotational domain For both approaches a graphical interface, as shown in figure 3, has been implemented. After computing the order of interest or the detector channel in the frequency range of interest, the level can be modified. FIR filters are used to modify the original signal in the time domain. This allows a further analysis and playback of the signal.

4 Application Examples and Results The previous sections have briefly described the theory of both OTPA and RMA. As it is rather difficult to show general results, three selected applications from the automotive domain will be presented. They show the potential of the combined approach, while illustrating the acceleration of the development process by limiting the amount of required iterations until the target is achieved.

6

A framework for the sensitivity analysis of transfer paths combining contribution …

Figure 4: On the left, an exemplary setup of indoor pass-by measurements. On the right, OTPA contribution analysis results and sensitivity of individual sources. The original is shown in blue, the modified one in light blue.

4.1 Indoor Pass-by As stated earlier, one of the OPTA’s first process integrations has been the simulated exterior noise measurement. The measurement procedure for the simulated pass-by is equal to the real pass-by, but conducted on an (all-wheel) dyno in a semi-anechoic chamber. Although such a dyno is already installed at most of the test facilities, regulations require a dedicated setup of the test bench, which is illustrated in figure 4. The chamber’s dimensions have to represent the measurement corridor of the real test track. This results in a width of at least 15 m, which is given by the distance of the microphones from the vehicle centre line. In order to simulate the approaching and receding of a vehicle on a fixed dyno, microphone line arrays are installed on both sides of the virtual test track. Consequently, the microphones have to be in line with the virtual position of the vehicle, resulting in test benches with a length of usually at least 20 m. Common setups consist of roughly 36 microphones on each side of the vehicle. In the case that the chamber is not as wide as required, a one-sided setup is possible. Here, the vehicle needs to be rotated by 180° to cover both sides. All microphones are captured synchronously and provide a static noise, as the distance between the vehicle and each microphone is constant. In order to be able to hear and measure the effect of the approaching and respectively receding car, an interpolation based on both the vehicle speed and time is performed, where each microphone is only used for a fraction of time.

7

A framework for the sensitivity analysis of transfer paths combining contribution …

Figure 5: Results of the modified vehicle for indoor measurements. The component, responsible for the high overall level, was probably the exhaust system. After confirmation by the RMA, the exhaust has been changed, resulting in lower overall level.

The basic idea is now to compute the contributions of engine, intake, tires, and exhaust system to each of the exterior noise microphones. Therefore, both microphones and accelerometers are placed close to the sources of interest. By applying the OTPA on the acquired signals and using the microphone at the virtual zero line as response signal, a ranking of the contributions, as illustrated in figure 4b) can be performed. In this case, it seems that the exhaust system is the component with the highest contribution. Combining the OTPA with the RMA, where the 4th order has been selected, provides a sensitivity analysis of the observed components. Applying the RMA adds the information that the exhaust is the most sensitive source to modifications. As shown in figure 5, the overall level could be lowered by this countermeasure without any effect on all the other components.

4.2 Comfort Measurements for the Whole Vehicle Development Customer satisfaction has priority during the whole vehicle development. Both comfort and quality are essential – especially for premium vehicles. Disturbing noises and vibrations within the cabin should be avoided at all cost. Therefore, comfort measurements are usually carried out with comparatively high effort. Nevertheless, modifications of components, due to availability or cost improvement, are frequently performed within the production process. Quality has to be assured even after changing the originally approved configuration. In the selected example, a vibration could be perceived at the driver seat’s seat rail, which could be also shown in the spectrum as 1.5th order. Hence, only accelerometers have been used for the operational measurements. These have been placed at engine mounts, the gearbox mount, the torque mount, front wheels, and the exhaust system. Applying the OTPA with the mentioned sensor positions and the seat rail as response, as shown in figure 6, indicates the ex-

8

A framework for the sensitivity analysis of transfer paths combining contribution …

haust system as the cause for the undesired behavior. The response channel has been modified at the 1.5th order. It became obvious that the largest contribution, the exhaust system, is rather insensitive to modifications. Surprisingly, as figure 6 illustrates, the torque mount is the source with the highest sensitivity. Additionally, an Operational Deflection Shape (ODS) analysis of the entire system has been performed, using high principal components from the crosstalk cancellation as input. This finally proves that the tilting movement of the engine induces vibration to the exhaust system via the coupling of the engine and exhaust system. By changing this coupling, the vibration can be completely avoided.

Figure 6: Exemplary measurement at a seat rail. The contributions and sensitivities are shown in b), where it becomes clear that the actually rather low contributing torque mount is highly sensitive to changes. The exhaust mounts are obviously not sensitive to modifications.

4.3 Engine Development Engines or respectively the powertrain are usually investigated extensively on specialized test benches. Nevertheless, some problems are only perceivable in the overall assembly of chassis and powertrain. A vibration has been perceived at the steering wheel only at coast downs and with a very low intensity even in idle state. The engine itself has been considered as inconspicuous on the test bench. The question is now, if there is any chance for countermeasures without changing the engine, as it is at a late development state. Therefore, an OTPA has been performed; using accelerometers on

9

A framework for the sensitivity analysis of transfer paths combining contribution …

the passive side of the interfaces between engine and chassis, namely the mounts, and an accelerometer has been positioned on top of the steering wheel. The TPS has shown that the largest contribution to the perceived vibration is transmitted via one of the engine mounts, as shown in figure 7. Applying the RMA, again with a modification of an order, as the vibration is definitively created by a rotating part, shows that the paths sensitivity to changes is not necessarily depending on the path contribution itself. In this case, the largest contributors would be the left engine mount into Z and Y direction and the gearbox mount in Z and Y direction. The RMA now shows that the paths most sensitive for modifications would be the left engine mount and the exhaust, although being rather inconspicuous in the TPS. Hence, the solution to this issue is a modification of the paths connected to these mounts.

Figure 7: Results for the RMA investigating vibrations at the steering wheel. The highest contributions can be observed at the left engine mount and the gearbox mount. Nevertheless, exhaust and right mount seem most sensitive to modifications.

5 Conclusion and Outlook The combination of OTPA and RMA has been successfully integrated in the R&D process in order to speed it up by avoiding unnecessary iterations. In contrast to the sole application of the OTPA, the RMA additionally addresses the sensitivity of modifications to sources and transfer paths. This is inevitable, as high contributing paths are not necessarily sensitive to modifications, as the energy within the system has to be taken into account. Nevertheless, it has to be noted that the RMA does not consider expert knowledge, and is suggesting sensitivities based on the transfer path synthesis. There is no prediction on the actual possibility in changing the component. This is not always possible. By combining this knowledge and the RMA, a rather conclusive picture can be achieved.

10

A framework for the sensitivity analysis of transfer paths combining contribution …

6 References 1. van der Seijs M., de Klerk D., Rixen D.: General framework for transfer path analysis: History, theory and classification of techniques, Mechanical Systems and Signal Processing, Volumes 68–69, February 2016, Pages 217-244, ISSN 08883270 2. Putner J., Lohrmann M., Fastl H: Contribution analysis of vehicle exterior noise with operational transfer path analysis, Proceedings of Meetings on Acoustics, 2013, Acoustical Society of America 3. Putner J., Fastl H., Lohrmann, M., Kaltenhauser A., Ullrich F.: Operational transfer path analysis predicting contributions to the vehicle interior noise for different excitations from the same sound source, Inter-Noise 2012, Institute of Noise Control Engineering 4. Noumura K., Yoshida J.: Method of transfer path analysis for vehicle interior sound with no excitation experiment, Proceedings of the FISITA 2006 World Automotive Congress, Yokohama, Japan 5. Arsic. D, Bock F., Becker S. An Integrated R&D Approach for Exterior Noise Development Applying Contribution Analysis and Response Modification Analysis, Proceedings Internoise 2016, Hamburg, Germany, 2016

11

Tagungsbericht Mathias Heerwagen

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 W. Siebenpfeiffer (Hrsg.), Automotive Acoustics Conference 2017, Proceedings, https://doi.org/10.1007/978-3-658-20251-4_20

1

Tagungsbericht

4.InternationaleATZ-Fachtagung Fahrzeugakustik Am 11. und 12. Juli 2017 trafen sich mehr als 240 Experten aus 19 Ländern auf der Automotive Acoustics Conference in Zürich/Rüschlikon (Schweiz), um über die Heraus-forderungen der Automobilindustrie beim Thema Fahrzeugakustik zu diskutieren. AKUSTIK GEWINNT ZUNEHMEND AN BEDEUTUNG

Wie wird die Elektromobilität das Geräuschverhalten im Fahrzeug beeinflussen? Was muss sich in Zukunft ändern, um Geräusche, die vorher durch den Verbrennungsmotor erzeugt oder überdeckt wurden, zu reduzieren? Wie wichtig sind Akustiksimulationen während der Fahrzeugentwicklung? Über diese und weitere Themen diskutierten die Teilnehmer der ATZliveTagung. Die Fahrzeugakustik ist ein wichtiges Thema, herstellerübergreifend und weltweit. Stark vertreten waren in diesem Jahr Teilnehmer aus Asien – jeder fünfte Gast kam aus Japan, Südkorea oder China. Die unter der wissenschaftlichen Leitung von Autoneum stehende und gemeinsam mit ATZlive organisierte Konferenz gilt als eine der bedeutendsten internationalen Akustik-Fachtagungen für Automobilhersteller, Zulieferer und Forschungsinstitute. Mehr als 20 Fachpräsentationen wurden ergänzt durch Workshops zu neuen Mobilitätstrends. Eine umfangreiche Fachausstellung präsentierte aktuelle Innovationen aus dem Kfz-Akustikbereich, darunter neue leichte Dämmstoffe. MOTORKAPSELUNG BRINGT NICHT NUR VORTEILE

Im Keynote-Vortrag zeigte Masashi Komada von Toyota die Probleme der Hersteller auf: Wo kann man mit welchen Materialien Motor-, Wind- und Straßengeräusche reduzieren? Der Fahrzeuginnenraum lasse sich mit isolierenden oder absorbierenden Akustikteppichen auslegen, auch spezielle Türdichtungen tragen dazu bei, das Innenraumgeräusch zu senken. Doch nicht nur die Geräusche im Innenraum spielen eine Rolle. Komada wies darauf hin, dass beim Pass-by-Test bei 50 km/h fast drei Viertel des Geräuschs von den Reifen stammt. Gab es vor einigen Jahren noch Sicherheitsbedenken, sind heute herstellerübergreifend spezielle Reifen im Einsatz, die Laufgeräusche absorbieren. Auch zur Reduzierung des Motorengeräuschs gibt es verschiedene Möglichkeiten. Die Kapselung des Motors bringt in Sachen Geräuschverhalten

2

Tagungsbericht

viele Vorteile mit sich, aber auch einige Nachteile. Durch höhere Temperaturen in der Kapselung können Teile Schaden nehmen, zudem ist eine Kapselung teuer und technisch aufwendig. Am Beispiel des Brennstoffzellenfahrzeugs Toyota Mirai erklärte Masashi Komada die Herausforderungen für die Akustikingenieure bei E- und FCVFahrzeugen. Zwar fällt das Motorgeräusch weg, aber Pumpen, Ventile, Injektoren und vor allem der Kompressor des Mirai erzeugen ebenfalls Geräusche, die es einzudämmen gilt.Perry Gu von Geely erwähnte in seinem Vortrag, dass kleine, hochaufgeladene Motoren ganz andere Vibrationen und Geräusche erzeugen als groß-volumige Motoren. Mittlerweile werden absorbierende Materialien und Strukturen ins Ansaugsystem eingebracht, um die Geräusche zu dämmen. Das Thema Akustik sei in China wichtiger als in Europa, chinesische Käufer seien sehr sensibel und achteten vor dem Kauf explizit darauf, dass der Motor leise läuft und das Fahrgeräusch niedrig ist. GERÄUSCH- UND GEWICHTSREDUZIERUNG

Nicht nur die Geräusch-, sondern auch die Gewichtsreduzierung spielt eine zunehmend wichtige Rolle. Beides ist eng miteinander verbunden: Dünnere Bleche können höhere Frequenzen erzeugen, was jedoch durch Verstärkungen im Blech oder andere Strukturen in gewissem Umfang ausgeglichen werden kann. Auch in der Präsentation von Yuksel Gur von Ford ging es um Gewichtsreduzierung und den Einfluss auf die Akustik. Im MMLV (Multi Material Lightweight Vehicle) kamen unter anderem Komponenten aus Aluminium, Magnesium und Carbon zum Einsatz. Das Fahrzeug ist rund 20 % leichter als das vergleichbare Serienfahrzeug, unter anderem durch den Einsatz dünnerer Scheiben, die statt 4 nur noch 2,4 mm dick sind. Durch eine besondere Unterbodenisolation, eine Motorkapselung und Akustikteppiche ist das Geräuschniveau dennoch auf einem ähnlichen Level wie beim Serienfahrzeug. Die Antwort auf die Frage aus dem Publikum, was solche Maßnahmen in der Großserie kosten würden, umschiffte Yuksel Gur geschickt. Ford nennt keine Zahlen, weiß aber was es kostet, ein Kilogramm Gewicht zu sparen oder das Geräuschniveau um einige Dezibel zu reduzieren. MIT SOFTWARE DEN LÄRM AUSLÖSCHEN

Während die meisten Unternehmen auf die Optimierung der Hardware setzen, beschäftigen sich die Experten bei Harman Becker Automotive Systems mit der digitalen Geräuschreduzierung. Hierbei werden Sensoren und Mikrofone an verschiedenen Stellen am Fahrzeug installiert, die Fahrgeräusche wahrnehmen.

3

Tagungsbericht

Dr. Nikos Zafeiropoulos erklärt, dass anschließend eine Software, ähnlich wie bei Noise-Cancelling-Kopfhörern, gegenläufige Schallwellen errechnet und über Lautsprecher erzeugt, die den Lärm auslöschen. So können die Fahrgeräusche je nach Frequenz um bis zu 8 dB reduziert werden – ein deutlicher Unterschied, der auch im Demo-Video zu hören war und im Publikum rege diskutiert wurde. Der Einsatz der Hard- und Software sei in allen Fahrzeugklassen möglich, die Geräuschreduzierung funktioniere auf allen Fahrbahnbelägen. Alexis Talbot von MSC Software aus Belgien verdeutlichte in seinem Vortrag, wie wichtig es ist, bereits in einem frühen Stadium der Fahrzeugentwicklung an die später entstehenden Geräusche zu denken. So sei die frühe Simulation von Pass-byNoise enorm wichtig, da sich Änderungen am fertigen Fahrzeug später nur noch schwer umsetzen lassen. In weiteren Vorträgen ging es unter anderem um die Geräuschentwicklung von Elektromotoren, den Einfluss von Motoraufhängungen auf die Vibrationen im Fahrzeug, neue Teppichmaterialien und darum, wie ein Elektrofahrzeug klingen muss, um der Marke gerecht zu werden. Experten aus den Akustik- und NVH-Abteilungen der großen Hersteller, Zulieferer und Entwicklungsdienstleister werden auch 2019 wieder in Zürich/Rüschlikon zusammenkommen, zur dann 5. Internationalen ATZliveFachtagung Fahrzeugakustik. Die Tagung wird erneut mit dem Kooperationspartner Autoneum veranstaltet. [Quelle: ATZ 119 (2017), Nr. 10, S. 76f]

4