Automotive Acoustics Conference 2015: 3. Internationale ATZ-Fachtagung [1. Aufl. 2019] 978-3-658-27647-8, 978-3-658-27648-5

Table of contents :
Front Matter ....Pages I-X
Time reversal applications: source detection, defect localization and perceptive structures (Jean-Louis Guyader, Quentin Buisson, Guillaume Guyader)....Pages 1-17
Acceptance of synthetic driving sounds in the interior of electric vehicles (André Fiebig, Brigitte Schulte-Fortkamp)....Pages 18-34
Interior noise effects of active vibration cancellation for a 4-cylinder engine (Niklas Müller, Enrico Kruse, Andrew Harrison)....Pages 35-48
Simulation of exterior powertrain ATFs on an engine bay mock-up with complex trim configurations (Abdelkader Bihhadi, Claudio Bertolini, Christophe Locqueteau)....Pages 49-83
Acoustic source detection for climate systems via computational fluid dynamics for improved cabin comfort (Jan Biermann, Barbara Neuhierl, Adrien Mann, Min-Suk Kim)....Pages 84-97
Vehicle validation of the structure-borne noise of a lightweight body and trim design solution obtained with new integrated FE optimization (J. W. Yoo, Théophane Courtois, J. Horak, Francesca Ronzio, S.-W. Lee)....Pages 98-127
Diagnostics and reparation of customers’ NVH complaints – a strategy of a carmaker (Léon Gavric, Cyril Peronnet, Guillaume Catusseau)....Pages 128-142
Enhanced lightweight NVH solution based on vibro-acoustic metamaterials (C. Claeys, E. Deckers, Bert Pluymers, W. Desmet)....Pages 143-157
Sound field control in the automotive environment (Jordan Cheer, Stephen J. Elliott, Woomin Jung)....Pages 158-175
Importance of the evaluation of structure-borne NVH performance for lightweight trim design (Théophane Courtois, Marco Seppi, Francesca Ronzio, L. Sangiuliano, T. Yano)....Pages 176-217
How to avoid annoying rolling bearing noises (Cédric Geffroy, Hannes Grillenberger, Carsten Mohr)....Pages 218-239
Engine NVH performance improvements with polymer gears (Björn Fink, Ralf Weidig, Frank J. Ferfecki, Tony Whitehead, Justin Salisbury)....Pages 240-254
New road noise testing techniques (Hartmut Bathelt, Heiko Kolm, Kay Schammer)....Pages 255-271
Extension of acoustic holography to cover higher frequencies (Jørgen Hald)....Pages 272-289
Consideration of the influences of the modal sound field with respect to the sound source localization results of the beamforming process in a vehicle interior (Clemens Nau, Rob Opdam, Werner Moll, Michael Vorländer)....Pages 290-300
Tagungsbericht (Jonathan Walker)....Pages 301-304

Citation preview

Proceedings

Wolfgang Siebenpfeiffer Hrsg.

Automotive Acoustics Conference 2015 3. Internationale ATZ-Fachtagung

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 2015 3. Internationale ATZ-Fachtagung

Hrsg. Wolfgang Siebenpfeiffer Stuttgart, Deutschland

ISSN 2198-7440  (electronic) ISSN 2198-7432 Proceedings ISBN 978-3-658-27648-5  (eBook) ISBN 978-3-658-27647-8 https://doi.org/10.1007/978-3-658-27648-5 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

Alongside roadholding, handling and driveability, acoustics and NVH are a prime indicator of vehicle quality and refinement. They hence enjoy an increasing focus in automotive research and development, at a time of fundamental changes in powertrain configurations. With the Automotive Acoustics Conference, Autoneum and ATZlive provide an essential global forum for the exchange of information in these technologies. Further, the two-day conference provides opportunities for personal interaction and networking among an international community of experts with a shared passion and common goals. Starting with keynote speeches from leading industry figures, the 3rd International Automotive Acoustics Conference will cover its customary broad spectrum of themes, chosen and categorized to give participants a comprehensive, wellstructured overview. As such, the conference proceedings represent a valuable, bi-annual snapshot of the stateof-the-art in acoustics and NVH. Among more than 20 scheduled presentations, the 2015 program includes presentations on noise attenuation in the latest hybrid powertrains, new lightweight acoustic materials and the generation of synthetic sound. Complementing this breadth of technology is the conference’s global reach, with contributors from both established and emerging centers of vehicle development. The setting for the conference is again ETH Zurich, Switzerland’s renowned university of technology and natural sciences. We look forward to meeting you in Zurich in June 2015! On behalf of the Scientific Advisory Board Dr. Davide Caprioli Scientific Director of the Conference Autoneum Wolfgang Siebenpfeiffer Editor-in-Charge ATZ | MTZ | ATZelektronik

V

Inhaltsverzeichnis

Time reversal applications: source detection, defect localization and perceptive structures Prof. Dr. Jean-Louis Guyader, Quentin Buisson und Guillaume Guyader Acceptance of synthetic driving sounds in the interior of electric vehicles André Fiebig und Prof. Dr. Brigitte Schulte-Fortkamp Interior noise effects of active vibration cancellation for a 4-cylinder engine Niklas Müller, Enrico Kruse und Andrew Harrison Simulation of exterior powertrain ATFs on an engine bay mock-up with complex trim configurations Abdelkader Bihhadi, Claudio Bertolini und Christophe Locqueteau Acoustic source detection for climate systems via computational fluid dynamics for improved cabin comfort Dr. Jan Biermann, Dr. Barbara Neuhierl, Adrien Mann und Min-Suk Kim Vehicle validation of the structure-borne noise of a lightweight body and trim design solution obtained with new integrated FE optimization J. W. Yoo, Dr. Théophane Courtois, J. Horak, Francesca Ronzio und S.-W. Lee Diagnostics and reparation of customers’ NVH complaints – a strategy of a carmaker Dr. Léon Gavric, Cyril Peronnet und Guillaume Catusseau Enhanced lightweight NVH solution based on vibro-acoustic metamaterials C. Claeys, E. Deckers, Dr. Bert Pluymers und W. Desmet Sound field control in the automotive environment Dr. Jordan Cheer, Prof. Stephen J. Elliott und Woomin Jung Importance of the evaluation of structure-borne NVH performance for lightweight trim design Dr. Théophane Courtois, Dr. Marco Seppi, Francesca Ronzio, L. Sangiuliano und T. Yano How to avoid annoying rolling bearing noises Cédric Geffroy, Dr. Hannes Grillenberger und Dr. Carsten Mohr Engine NVH performance improvements with polymer gears Ralf Weidig, Dr. Björn Fink, Frank J. Ferfecki, Tony Whitehead und Justin Salisbury New road noise testing techniques Prof. Dr. Hartmut Bathelt, Heiko Kolm und Kay Schammer Extension of acoustic holography to cover higher frequencies Jørgen Hald

VII

VIII

Inhaltsverzeichnis

Consideration of the influences of the modal sound field with respect to the sound source localization results of the beamforming process in a vehicle interior Clemens Nau, Rob Opdam, Dr. Werner Moll und Prof. Dr. Michael Vorländer Tagungsbericht Jonathan Walker

Autorenverzeichnis

Prof. Dr. Hartmut Bathelt  Akustikzentrum GmbH, Stuttgart, Deutschland Claudio Bertolini  Autoneum Management AG, Winterthur, Schweiz Dr. Jan Biermann  BMW AG, München, Deutschland Abdelkader Bihhadi  Autoneum France SASU, Aubergenville, Frankreich Quentin Buisson  SONORHC Technologies, Charnoz-sur-Ain, Frankreich Guillaume Catusseau  PSA Peugeot Citroën Automobiles, La Garenne-Colombes, Frankreich Dr. Jordan Cheer  University of Southampton, Southampton, England C. Claeys  KU Leuven, Leuven, Belgien Dr. Théophane Courtois  Autoneum Management AG, Winterthur, Schweiz E. Deckers  KU Leuven, Leuven, Belgien W. Desmet  KU Leuven, Leuven, Belgien Prof. Stephen J. Elliott  University of Southampton, Southampton, England Frank J. Ferfecki  Victrex Europa GmbH, Hofheim, Deutschland André Fiebig  HEAD acoustics GmbH, Herzogenrath, Deutschland Dr. Björn Fink  Victrex Europa GmbH, Hofheim, Deutschland Dr. Léon Gavric  PSA Peugeot Citroën Automobiles, La Garenne-Colombes, Frankreich Cédric Geffroy  LuK GmbH & Co. KG, Bühl, Deutschland Dr. Hannes Grillenberger  Schaeffler Technologies AG & Co. KG, Herzogenaurach, Deutschland Prof. Dr. Jean-Louis Guyader  National Institute of Applied Sciences (INSA), Lyon, Frankreich Guillaume Guyader  SONORHC Technologies, Charnoz-sur-Ain, Frankreich Jørgen Hald  Brüel & Kjaer SVM A/S, Naerum, Dänemark Andrew Harrison  TrelleborgVibracoustic GmbH, Morganfield, USA J. Horak  Autoneum Management AG, Winterthur, Schweiz Woomin Jung  University of Southampton, Southampton, England Min-Suk Kim  Exa Corp., San Francisco, USA Heiko Kolm  AUDI AG, Ingolstadt, Deutschland Enrico Kruse  TrelleborgVibracoustic GmbH, Morganfield, USA S.-W. Lee  Hyundai Motor Group, Seoul, Südkorea Christophe Locqueteau  Renault S.A., Boulogne-Billancourt, Frankreich Adrien Mann  Exa Corp., San Francisco, USA

IX

X

Dr. Carsten Mohr  LuK GmbH & Co. KG, Bühl, Deutschland Dr. Werner Moll  Daimler AG, Sindelfingen, Deutschland Niklas Müller  TrelleborgVibracoustic GmbH, Weinheim, Deutschland Clemens Nau  Daimler AG, Sindelfingen, Deutschland Dr. Barbara Neuhierl  Exa GmbH, München, Deutschland Rob Opdam  RWTH Aachen University, Aachen, Deutschland Cyril Peronnet  PSA Peugeot Citroën Automobiles, La Garenne-Colombes, Frankreich Dr. Bert Pluymers  KU Leuven, Leuven, Belgien Francesca Ronzio  Autoneum Management AG, Winterthur, Schweiz Justin Salisbury  Victrex Europa GmbH, Hofheim, Deutschland L. Sangiuliano  Autoneum Management AG, Winterthur, Schweiz Kay Schammer  AUDI AG, Ingolstadt, Deutschland Prof. Dr. Brigitte Schulte-Fortkamp  TU Berlin, Berlin, Deutschland Dr. Marco Seppi  Autoneum Management AG, Winterthur, Schweiz Prof. Dr. Michael Vorländer  RWTH Aachen University, Aachen, Deutschland Jonathan Walker  Kaufbeuren, Deutschland Ralf Weidig  Victrex Europa GmbH, Hofheim, Deutschland Tony Whitehead  Victrex Europa GmbH, Hofheim, Deutschland T. Yano  Nihon Tokushu Toryo C., Ltd, Tokio, Japan J. W. Yoo  Hyundai Motor Group, Seoul, Südkorea

Autorenverzeichnis

Time reversal applications: source detection, defect localization and perceptive structures Prof. Jean-Louis Guyader, Former Director of the Laboratoire Vibrations Acoustique, INSA de Lyon and SONORHC Technologies, France. Quentin Buisson and Guillaume Guyader, SONORHC Technologies, France

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 W. Siebenpfeiffer (Hrsg.), Automotive Acoustics Conference 2015, Proceedings, https://doi.org/10.1007/978-3-658-27648-5_1

1

Time reversal applications: source detection, defect localization and perceptive …

1 Introduction Time Reversal (TR) concept has been proposed in the 1980s by Mathias Fink and colleagues, it is based on the symmetry of vibrations governing equations for positive and negative time, when damping effect is negligible. The main consequence of this property is the possibility of creating time reversed waves travelling back to the location where the primary waves were created. Several applications in bio mechanics, military purposes, room acoustics, structural vibrations and ultrasonics for SHM applications have been developed in literature. We just give here few papers for illustrating the different treated aspects [1, 2, 3, 4, 5, 6] This paper focuses on engineering applications in the field of acoustics and structural vibration. After a general presentation of the TR method, we give examples of possible applications in vibro acoustics problems. The phenomenon of focalization is presented experimentally and it is shown that contrary to standard inverse problems for source detection, structural complexity renders Time Reversal technique more efficient. A first application of TR, is defect detection: when a structure has no defect, playing recorded time reversed signals produce focalization of vibrations at primary source locations, when the structure has defects the wave propagation travel is modified and the focalization is no more observed, indicating the presence of defect. Results obtained on water ducts are presented. Based on time reversal a new concept of acoustics antenna that only needs a very small number of sensors has been developed (the butterfly antenna) and application to source localization in cars is shown. A major point is that anechoic conditions are not needed when using time reversal approach. The concept of perceptive mechanical media is lastly presented; these media can detect and react if a modification is occurring in their environment. This opens a gate to a lot of applications in particular the communication with tactile structures.

2 Basis of Time Reversal To introduce the time reversal concept we consider the basic equation of vibration of a mechanical system in a matrix form: (M)

⁄

+(C)

⁄

+(K) x = F

The initial conditions applied to the structure are: ( )=

, and

⁄

( )=

The resulting vibrations produced at an observation time servation time displacement and velocity vectors:

2

can be described as final ob-

Time reversal applications: source detection, defect localization and perceptive … =

⁄

, and

=

Let us make a change of variable in these equations by setting equations are: (M)

⁄

-(C)

+(K) x = F

⁄

(− ) =

, and

⁄

=

, and

⁄

−

= − , the resulting

(− ) = −

=

These equations are identical to the previous ones except for the damping term of the equation of motion; however damping is generally small in practice and can be ignored in first approximation. The important property associated to these two sets of equations is that in the reversed time configuration placing the system in the final conditions will produce the same response of the structure ending in the initial conditions of the standard time. A particular case is the shock excitation, at location of the structure. The equations describing the motion are: (M)

⁄

+(C)

(0) = 0 , and

+(K) x = δ (t)

⁄ ⁄

( 0) = 0

Where δ (t) represent the shock created at time = 0 and at location The structural response is a huge vibration at the excitation at location, just after the shock then the vibrations propagate all over the structure. In the classical application of time reversal the shock excitation is not present and the equations in the reversed time are: (M)

⁄ −

=

-(C) , and

+(K) x = 0

⁄ ⁄

−

=

The final conditions are producing waves, which propagate toward the origin of their creation. Namely focalization will be observed at location corresponding to the high vibration level produced just after the shock in the standard time experiment. Because the shock is not applied in the reversed time experiment a divergent signal is observed after focalization instead of a blocked mechanical system. If the shock is applied the structure vibrations are stopped, this is known as ‘time reversal well concept’.

3

Time reversal applications: source detection, defect localization and perceptive …

3 A first experiment of focalisation A semi complex structure (see figure 1) was considered and a shock was applied in a point. Vibration where recorded in two points and plaid in reversed time for placing the structure in the final conditions observed at two points.

Figure 1. Experiment on a semi complex structure

In figure 2, the experiment is described, first a shock is applied on the structure and resulting vibrations are measured in two points, then signals are reversed and re-emitted in the system, the vibration response at location where the shock was applied is measured. One can see as expected the focalisation (figure2). It is the demonstration of the robustness of the method because theoretically all the structure must be in final conditions, we will see later an interpretation of this properties where only two sources can lead to the focalization. In this type of experiment one can see at the focalisation point first a convergence of the waves, then a focalisation and lastly the divergence of the signal figure 3.

4

Time reversal applications: source detection, defect localization and perceptive …

Figure 2 Shock excitation and time reversed experiment

Figure 3 Typical focalization signal

A second phenomenon is of interest: the space focalisation spot as shown in figure 4. One can see in this figure that vibrations level is decreasing when the distance to the excitation shock is larger, creating a focalisation spot. The spot size is related to the frequencies contained in the vibration signal for high frequencies the spot is small and inversely it is large for low frequencies.

5

Time reversal applications: source detection, defect localization and perceptive …

Figure 4. The focalization spot

4 The influence of structural complexity Theoretically all points of a structure have to be put in the final conditions to ensure the perfect time reversed motion. This is quite impossible to do as it requires a huge number of re-emission transducers. However focalization is observed with only two transducers and sometimes just one. This is due to the fact that reflection on boundary and obstacles can be seen as virtual sources, thus, for complex structures the number of reflection is huge and few real sources are sufficient to produce focalisation. The figure 5 gives an illustration of the phenomenon.

Figure 5 Direct field and virtual sources associated to waves reflections

6

Time reversal applications: source detection, defect localization and perceptive … This result is important for industrial applications where structural complexity is generally achieved. Contrary to all others inverse method for source localisation complexity helps the method of TR to work.

5 Application to source detection by the butterfly antenna. 5.1 The Butterfly antenna The Butterfly antenna is a thin structure with high modal complexity (with a large modal density) that is placed in an acoustic field in order to detect the direction of an incoming noise or to detect the mechanical sources locations responsible for the observed noise. In figure 6 the photograph of the antenna is shown, only two sensors are used to localisation of sources because structural complexity increase considerably the information measured, thus it is an antenna with low number of sensors. It works in a two steps method first a learning step and then a detection step. The learning step consists in exciting at different locations where the sound could be generated in order to get the transfer functions, then by virtual time reversal through the transfer function, the source is localised among the learned possibilities.

7

Time reversal applications: source detection, defect localization and perceptive …

Figure 6 The Butterfly antenna

5.2 Direction of incoming sound An experiment was done (in cooperation with Vibratec) using an anechoic chamber to learn the direction of incoming sound wave. The butterfly antenna was placed in the chamber and acoustic sources placed at different locations all around the antenna were used for learning the direction of incoming acoustic waves (see figure 7). In this experiment sources are produced by white noise emitted through a loudspeaker

8

Time reversal applications: source detection, defect localization and perceptive …

Figure 7 locations of sources for learning direction of incoming waves on the antenna.

First sources are created in the anechoic chamber in order to localize them. The locations of the sources correspond exactly to learning points. As can be seen in figure 8, the localization is perfect.

Figure 8 Indicator of source location for 7 experiments when the source is placed successively on position 7, 6, .., 1.The localisation is perfect.

9

Time reversal applications: source detection, defect localization and perceptive … In a second experiment the source is moved from a learning point to another one, thus, passing through points not included in the learning set. The different positions of the source are described in figure 9. It is clear in figure 10 that the indicator of localisation describes well the situation starting from a clear localisation of the first learned point to a localisation of both points and finally to a clear localisation on the second point.

Figure 9 the two learned points of emission and intermediate positions

10

Time reversal applications: source detection, defect localization and perceptive …

Figure 10 Indicator of localisation for two learned points and intermediate localisations

Can the learning done in an anechoic chamber be used in non treated room? This question is interesting in the sense that learning can be done once and then used in all acoustic cavities. The reported experiment was done in a standard room and four positions of excitation were tested. For 3 positions the detection of incoming sound clearly indicates the exact location of the source in the fourth case two opposite directions are indicated this can be explained by a reflexion of sound on the opposite panel resulting in a mirror source associated to the primary one.

Figure 11 source locations in a non treated room

11

Time reversal applications: source detection, defect localization and perceptive …

5.3 Application to mechanical source detection in a car The reported experiment were done in cooperation with Vibratec. The problem we were interested in is the localisation of a force applied on a car suspension. The first question is about the possibility of the antenna placed in the car, at the driver position, to detect the location of the mechanical excitation producing the noise measured by the antenna. The second question is more difficult; is it possible to detect also the direction of excitation? Here again the first step is to learn the transfer from potential excitation point.

Figure 12 Points of potential mechanical excitation

Figure 13 Source indicator, scan 1, 2, 3 different positions, scan 4, position 3 but vertical excitation

12

Time reversal applications: source detection, defect localization and perceptive … Results are presented in figure 13 for shock excitations, it can be seen that the localization is very good for the four experiments changing point of excitation and direction of excitation. In a second experiment multiple excitation are studied. The excitation points are located on four points as indicated in figure 14 the fifth case is again a different excitation points. Points one and two are excited by shakers with white noise.

Figure 14 excitation points

Figure 15 show three experiments (scans). In the first; point 1 and 2 are excited with similar level excited by shakers driven by white noise in the band from 20 Hz to 500 Hz., then in experiment 2 the level of shaker in point 1 is increased whereas the one in point 2 is decreased and in experiment 3 the inverse is done . The indicator of source location describes well this situation. However the point 5 contribution is not negligible indicating coupling between directions of excitation, because reference 5 is located in point 2 but in the horizontal direction. .

13

Time reversal applications: source detection, defect localization and perceptive …

Figure 15 Source location indicator when two sources are acting at point 1 and 2.

6 Defect detection Time reversal can be used for detecting the modification of structures by virtually creating shocks in different points of structures and then observing the effect of local modifications on the divergent signal. If the virtual shock is produced on a modified point the divergent signal will be highly affected, if the virtual shock is created on a non modified point the divergent signal will be only lightly affected. This was used in the 3S project in cooperation with Metravib for detecting defects in structures located inside a shell with sensors placed outside the shell. Figure 16 shows the experimental set up for the 3S experiment, sensors are located on the top plate and defects in the internal structures. Four sensors are used to focalize on 8 points in the internal structure where a defect can be present and four sensors are used for analyzing the divergent signal. With this method it was possible to detect added masses on the internal structure and even a crack in one internal plate. In figure 17 the indicator of defect as a probability of presence at a learned point is presented, it shows that point one is detected and corresponds to a crack on an internal plate.

14

Time reversal applications: source detection, defect localization and perceptive …

Figure 16 the experimental set up for the 3S experiment to detect defects

Figure 17 indicator of defect as probability of presence on a reference point (point one is detected) and the detected crack on an internal plate.

15

Time reversal applications: source detection, defect localization and perceptive …

7 Perceptive structures Using Time reversal it is possible to render mechanical media perceptive to changes of their environment. The technique used is not described as it is in the process to be patented. However an example of perceptive structure is shown in figure 18, depending on the contact zone the structure understand different messages and can react. Demonstration of this will be done during the conference.

Figure 18 A perceptive animal

16

Time reversal applications: source detection, defect localization and perceptive …

8 References [1] M. Fink. Time reversed acoustics. Physics Today, 50(3) :34–40, 1997. [2] D. Cassereau and M. Fink. Time reversal of ultrasonic fields-part III: theory of the closed Time Reversal cavity. IEEE Trans. Ultrason. Ferroelec. Control, 39(5):579– 592, 1992. [3] G.F. Edelmann, T. Akal, W.S. Hodgkiss, S. Kim, W.A. Kuperman, and H.C. Song. An initial demonstration of underwater acoustic communication using time reversal. IEEE J. Oceanic Eng., 27:602–609, 2002. [4] S. Yon, M. Tanter, and M. Fink. Sound focusing in rooms: The time reversal approach. J. Acoust. Soc. Am., 113(3):1533–1543, 2003. [5] J. De Rosny. Milieux réverbérants et réversibilité. thèse de doctorat de l’Université Paris 6,2000. [6] D. Vigoureux, J-L Guyader. A simplified Time Reversal method used to localize vibrations sources in a complex structure. Applied Acoustics, 73, p491-496, 2012

17

Acceptance of synthetic driving sounds in the interior of electric vehicles André Fiebig, M.A. [email protected] HEAD acoustics GmbH Germany Prof. Dr. Brigitte Schulte-Fortkamp Brigitte.Schulte-Fortkamp@tu-berlin Technische Universität Berlin Germany

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 W. Siebenpfeiffer (Hrsg.), Automotive Acoustics Conference 2015, Proceedings, https://doi.org/10.1007/978-3-658-27648-5_2

1

Acceptance of synthetic driving sounds in the interior of electric vehicles

1 Introduction Meanwhile the concept of electric vehicles is well known, but still consumers have little personal experiences about driving electric vehicles. In order to improve the perceived quality of an electric vehicle, the optimization of the interior noise and vibration is highly relevant. But, how must an electric vehicle sound, what is the target sound of a certain electric vehicle, can NVH knowledge and experiences from the past be used for the optimization of modern electric vehicles as well? In order to find answers to such questions it seems necessary to clarify the notion of product sound quality first. Product sound quality is a descriptor of the adequacy of the sound attached to the product [1]. Whereby the term suitability is underlining the strong relation of the sound attached to the technical object emitting it [2]. It means in effect that the concept of sound quality comprises an understanding that modifying factors beyond the acoustical stimulus are incorporated by the product users into their final sound quality judgment [3]. As the perception of sound quality is context-dependent, therefore the quality of an acoustical stimulus referring to sound quality cannot be determined without addressing contextual variables [4], like the way an electric vehicle is driven. In the end, the assessment of sound quality evolves from a process, in which recognized features are compared to some kind of reference [5]. This means that a consumer develops an impression based on a relative evaluation of the sound compared to the expected sound of the product, and often is not based on technical merit [6]. But, the challenge is that an established frame of reference relying on the concept of expected and desired features of the respective object, does not exist for the majority of users. How do subjects assess the sound quality of an electric vehicle, if they compare the recognized features to some kind of expected “quality” without pre-knowledge? Customers cannot rely on an established set of expectations to express their wishes and needs reliably. This means that they hypothetically construct experiences to specify what is needed and what is unnecessary. This might be a reason, why surveys addressing the sound quality of electric vehicles frequently state a variety of customer preferences [7] and the development and conceptual orientation of sound design in electric vehicle seems very much open. On the one hand the sound character of the electric motor could be kept and refined [8], or on the other hand there are ideas for synthetic sound playback of sporty vehicle interior noises [9]. In general, it is obvious that certain cognitive processes that convert sounds to feelings and emotions are not simply captured by physical parameters and metrics. The sound character is related to the basic attributes of an auditory event without considering context, action and higher level of cognitive processing and is ideally devoid of any contextual conditions. The sound character can be described by physical parameters [10]. In contrast to it, sound quality describes the perception of sound adequacy affected by di-

2

Acceptance of synthetic driving sounds in the interior of electric vehicles verse factors such as context, cognition, interaction. Thus, it is imperative that people experience and evaluate product sound in context to collect reliable data about the assessment of product sound quality and to be able to make predictions about needed sound characteristics of a future product. It is necessary to learn about how customers respond to new products and sounds and how they create new frames of reference for (sound) quality assessment in electric vehicles. This will go much beyond to simply ask somehow what a customer think, in case he/she would drive an electric vehicle, an appropriate interior sound would be. In other words it does need research in case of hybrid vehicles, noise and vibration aspects in the complex interplay between combustion engine and electric motor have to be considered, since certain phenomena are “unfamiliar to most drivers and often judged as disturbing” [11]. In the context of exterior noise of electric vehicles, due to a popular fear of “silent” cars [12] and expected legal actions [13], sound design is mostly discussed from the perspective of pedestrian safety [14].

2 Requirements for Investigating the Acceptance of Synthetic Sounds in an Electric Vehicle Since sound quality is highly context-dependent, any investigation neglecting the assessment-relevant context cannot provide meaningful data. It is clear that any assessment of sound quality must take place in an electric vehicle usually while driving the vehicle. Thus, data must be collected during a real drive. In this regard it is assumed that the acceptance of the sound or its quality is driven by the decision whether it fits to the functional properties of the experienced product. This notion implicates an act of interpretation referring to the assigned meaning of an acoustical input. This means that evaluative behaviour of product users is not guided by the acoustical signals that we provide with them, but rather by the “meaning” which is transferred via these signal [15]. This transfer of meaning is only possible in realistic contexts, where a product is experienced, whereas in laboratory experiments using artificial test situations in order to control confounding variables the acoustical stimulus and its meaning is more abstract. Then, subjects must associatively create (hypothetical) contexts in their mind to be able to decide whether a sound is adequate or not. Unfortunately, such potentially created hypothetical contexts are not investigated. By means of comprehensive instructions it is attempted to construct an inter-individually constant hypothetical context (please image that …). However, the adequacy of a provided context is often not deliberately considered. To overcome the drawback of laboratory experiments, tests can be carried out in more realistic contexts ensuring high ecological validity of test results. This principle is used by the Explorative Vehicle Evaluation method (EVE) [16], where a test subject creates stimuli for judgment by itself in most realistic context and can even use its own vocabulary to judge the vehicle and its acoustics. These conditions increase the perceived naturalness of the test situation. It is evident that the influence of uncontrolled

3

Acceptance of synthetic driving sounds in the interior of electric vehicles confounding variables, the individual vocabulary and complex interplay of senses must be discussed in detail with respect to the generality level of results. However, the vehicle evaluation method aims to explore the perception and assessment process in close to reality contexts in order to detect relevant variables and to “explain” variance by illuminating the individual assessment strategies. The EVE method was adapted by Sellerbeck to automotive sound development processes [17]. In order to keep the naturalness of the (test) situation during a test drive only one sound concept was presented. This should promote an impression of an every-day drive, where usually the acoustics of the vehicle is not directly changed by pressing a button.

3 Case Study about the Acceptance of Synthetic Sounds in an Electric Vehicle In order to investigate needs and requirements regarding sound quality of electric vehicles, different sound proposals were developed and presented in an original context. The development of sounds and the collection of consumer reactions were done in a pilot study [18]. The motivation of study was to collect consumer reactions to different sound design concepts while driving a series-production electric vehicle (Opel Ampera).

3.1 Generation of Synthetic Sounds for a Case Study and Sound Development In order to create synthetic sounds depending on operational conditions of an electric vehicle in real-time, a sound synthesis tool (a parametric synthesizer for vehicle auralization) was developed and applied in the study. The sound synthesis tool allows for developing sounds by means of several synthesizers providing harmonics at certain intervals, Shepard tones, noise, modulation or roughness [19]. Figure 1 shows the available features which could be used for the sound synthesis. According to Küppers synthesizer programming is superior to any sample based algorithm and by means of several oscillators many possibilities exist to create complex spectral sounds [20]. Sounds were created for different operational conditions to generate sound characteristics and patterns needed for distinctive areas in the driving condition map. Sound design has to consider different levels of requirements on driving conditions [20]. It is realized by the definition of sounds for diverse operational conditions, which allows for implementing different sound characters in the driving condition map. In addition to the “static” sounds for discrete operational conditions, the change of conditions is considered introducing an additional level of sound dynamics by means of adaptive gain. This is required to emphasize additionally rapid load changes; in case of a rapid positive load change, a strong loudness increase of the synthetic sound can be additionally implemented. After load change the loudness of the synthetic sound decreases with an adjustable decay. The fea-

4

Acceptance of synthetic driving sounds in the interior of electric vehicles ture provides, if desired, a stronger impression of load feedback, on top of the possibly designed loudness increase with increasing speed and load of the respective synthetic sounds in the operating map. On the basis of different developed sounds distributed over the entire vehicle map (speed and load map ranging from zero to maximum), for all driving conditions the respective sounds is provided. Between defined sounds within the driving condition map the respective sound is virtually created by interpolation.

Figure 1: Sound development via a parametric synthesizer

The degree of detail of the sound synthesis is up to the sound designer and the needs to implement different sound characters and sound dynamics. Frequently, the use of Shepard tones is proposed [20]; the phenomenon was discovered by Shepard in 1964 [21]. Shepard tones have the advantage that they can create an auditory impression of continually increasing or decreasing frequency, although the spectral content remained constant. This allows for creating an impression of a sound permanently moving upwards (or downwards) without the need for changing the spectrum of the sound significantly. This avoids the occurrence of disturbing and unpleasant high frequency content with in-

5

Acceptance of synthetic driving sounds in the interior of electric vehicles crease of the reference variable like speed. Figure 2 shows exemplarily the spectrum of a Shepard sound creating an impression of a sound permanently moving upwards.

Figure 2: Sound consisting of a superposition of sine waves creating the auditory illusion of a sound that continually ascends in pitch

Finally, the synthesis also included a function to generate additional noise to reduce the perceived artificiality of any synthesized sound. In order to collect data with respect to the acceptance of certain synthetic driving noises in the context of an electric vehicle, three sound concepts were generated. The different sound concepts cover conceivable sound design approaches. The created sound concepts were (1) a sound resembling a combustion engine, (2) a modern, rather unconventional sound, and (3) an inconspicuous, modest sound. To compare the reactions to these sound concepts with reactions evoked by the original sound of the test vehicle, the original vehicle sound of the test vehicle was evaluated as well. Table 1 summarizes some acoustical differences between the evaluated sound concepts. Table 1: General properties of sound concepts considered in the pilot study Concept

General character

Idle noise

Roughness

Intervals (relative to fundamental frequency)

Sound concept 1

like combustion engine

yes

yes

Sound concept 2

modern, unconventionally

no

no

lower quart, minor third, quint

Sound concept 3

inconspicuous

yes

no

tonic, quint, higher octave

6

lower octave, quint, tonic

Acceptance of synthetic driving sounds in the interior of electric vehicles Figure 3 displays the design of the different sound concepts regarding loudness with respect to different driving conditions. The total loudness increases in the sound concepts 1, 2 and 3 compared to the original sound only scenario (sound concept 4) due to adding synthetic sound to the original sound. However, the amount of loudness increase was designed differently. Sound concept 1 and 2 are more prominent in terms of loudness (see Figure 2) and temporal as well as spectral patterns (see Figure 3). Sound concept 1 used main engine orders of an eight-cylinder combustion engine. In case of sound concept 3 engine orders were chosen with higher frequencies causing a different sound character. Sound concept 2 was realized by using the lower quart, minor third and quint in relation to the fundamental frequency, which is related to vehicle speed. In addition, the synthetic sounds possess different levels of adaptive gain related to load changes. Figure 5 illustrates the influence of loudness adaption level on perceivable patterns. The applied Relative Approach analysis identifies changes in the short time spectrum in the frequency and time domain [22]. It can be seen that in case of a tip-in within the “original sound only scenario” no prominent patterns occur, whereas in case of a load change in the sound concept 3 noise patterns occur suddenly. In case of sound concept 1 the amount of noise patterns is higher compared to sound concept 3. Sound concept 2 has also a lot of noise patterns in the frequency domain, but the load change is not indicated well by a considerable increase of patterns. In summary, the sound concepts vary in spectral content, sound character, total loudness, modulation, loudness gain (sound dynamics), implementation of idle sound, level of variation of sound in the driving condition map and their specific magnitude-load-speed relationship.

7

Acceptance of synthetic driving sounds in the interior of electric vehicles

Figure 3: Loudness over time of different sound concepts and driving conditions according to the DIN 45631/A1. Top left to bottom right: Idle, constant speed at 50km/h, acceleration from 0 to 50 km/h with partial load, tip-in at a speed of 50 km/h (only left channel shown).

Figure 4: FFT (A-weighted) vs. time of different sound concepts for the driving condition “acceleration from 0 to 50 km/h with partial load”

8

Acceptance of synthetic driving sounds in the interior of electric vehicles

Figure 5: Relative Approach Analysis vs. time of sound concepts for the driving condition “tipin at a speed of 50 km/h” (only left channel shown). The tip-in event occurred at about 3 s.

3.2 Performance of Explorative Study in Realistic Test Environment The sound concepts described above were assessed by test subjects in a compact class series-production plug-in hybrid electric vehicle (Opel Ampera). The test procedure, data acquisition and questioning of subjects were performed according to the method described in [16], [17]. At first, the test subject is instructed in a written and oral form regarding test procedure, evaluation task and test route. After the instruction, the test subjects familiarize with the test vehicle. Then, the subjects drive the test course, where they comment on vehicle characteristics including acoustics. After completing the test drive, a semi-structured interview is carried out in the test vehicle, where given comments during test drive can be explained more in detail, additional questions using an interview guideline are discussed and finally personal data is collected. The test subject answered to some questions raised by the experimenter regarding relevant in-situ judgments and comments and judged the perceived overall quality and sound quality. The possibility to explain in-situ judgments have two advantages: (a) the subjects can add detailed descriptions to the comments, and (b) the experimenter can collect further information to understand the given comments in detail. To avoid any memory effects, a test subject had to wait several working days until the next test drive was performed. Thus, test subjects could not recall all acoustic details and were gradually reset.

Test route The test subjects drove the electric vehicle for 25 to 30 minutes on a defined test route in the area around Aachen, Germany (see Figure 6). The test route was chosen to provoke relevant driving situations. Figure 6 (right) illustrates that (a) all relevant speed

9

Acceptance of synthetic driving sounds in the interior of electric vehicles ranges were covered and specific electric vehicle relevant speed ranges were pronounced (typical urban road speeds). In total 29 test drives were performed providing approximately 15 h test drive data for subsequent analysis. Figure 7 displays the frequency of occurrence of driving situations based on all test drives in terms of a color map. The urban speed range with low accelerations and idling occurred most frequently.

Speed distribution

35

frequency of occurence in %

30 25 20 15 10 5 0

120

speed in km/h

Figure 6: Test route (left) and speed distribution based on 29 test drives (right)

Figure 7: Frequency of occurrence of driving situations by means of colors from black to white as result of test route selection and driving behavior of test sample

10

Acceptance of synthetic driving sounds in the interior of electric vehicles

Instruction During the test drives the subjects were requested to spontaneously express their associations, emotions, feelings and thoughts with respect to the vehicle, its general comfort and its acoustics in their every-day life language. (“Bitte äußern Sie spontan und direkt Ihre Assoziationen, Eindrücke und Gedanken.“)

Test subjects The subjects were not informed whether they are exposed to additional synthetic sound or not while driving the test vehicle. The order of the test drives with their different sound concepts was balanced over the test sample in order to reduce the influence of order effects on the test result. In total, ten subjects (six male, four female) with different background concerning experiences with electric vehicles and NVH background knowledge took part in the case study. Five subjects evaluated all four sound concepts.

3.3 Analysis and Results of the Case Study Data For each test drive vehicle interior noise, comments, vehicle speed and throttle position, GPS data were recorded. In addition, the experimenter took notes about test drive related aspects, relevant comments, which could be addressed in the post experimental interview, and used a trigger to mark relevant comments in the data stream of the recording. This allows for a quick identification of comments relevant for categorization and further analysis. The verbal data was subject to an extensive text analysis [23] based on a qualitative content analysis [24]. All given comments were analyzed, clustered and assigned to driving condition the comment referred to. Based on the simple classification of comments into positive, neutral and negative, and their number, first conclusions can be already drawn. All sound concepts provoke positive and negative comments. Figure 8 shows the specific distribution of comments over the provided sound concepts.

11

Acceptance of synthetic driving sounds in the interior of electric vehicles 120

Connotation of comments over sound concepts provided

100 Number of comments

Sound concept 1 negative neutral positive

Sound concept 2 17%

33% 44%

45%

80

40%

22%

60

Sound concept 3

Sound concept 4

26%

40

32% 45%

42%

20 29%

0

Sound concept 1

Sound concept 2

Sound concept 3

Original sound only

Negative comments Neutral comments Positive comments

26%

Figure 8: Left: Number of comments for the different sound concepts with respect to their connotation (positive, neutral, negative), Right: Ratio between positive, neutral and negative comments for the different sound concepts

First of all, the playback of synthetic driving noises led to more comments compared to the drives without any sound playback. Obviously, synthetic sounds can additionally stimulate emotions and feelings. Most comments were provoked by sound concept 2, whereas sound concept 3 and the original sound scenario clearly achieved a lower number of comments. It can be seen that the ratio between positive and negative comments is different over the sound concepts. The best ratio between positive and negative comments has sound concept 3. For sound concept 3 more positive than negative in-situ comments were observed. This is a remarkable result, since the majority of subjects explained in interviews that they prefer an electric vehicle without any “unnecessary” sound playback. Moreover, the highest percentage of comments referring to the original sound of the electric vehicle while an additional synthetic sound was presented, was observed in the context of sound concept 3 (36 %). This illustrates the modest character of this sound, since comments (still) frequently refer to the original sound.

12

Acceptance of synthetic driving sounds in the interior of electric vehicles

14 12

Comments regarding constant speed at 50 km/h positive comments neutral comments negative comments

10 8 6 4 2 0

Sound concept 1

Sound concept 2

Sound concept 3

Original sound only

Figure 9: Number of comments for the different sound concepts with respect to their connotation referring to the driving condition “constant speed at 50km/h”

Figure 9 underlines the benefit of the modest sound in case of sound concept 3 providing a better acoustical feedback than the original sound. Although the overall loudness is almost similar in both cases (see Figure 3), the introduced acoustical patterns improve the sound quality during cruising, although almost all subjects explained that they preferred that no additional sound is played back. For sound concept 1 and 2 the introduction of additional sound deteriorates the assessment. Obviously, the synthetic sound was too prominent. The original sound condition leads to a balanced result. The sound concepts 1 and 2 provoked more negative than positive comments, which means that due to the additional sound the sound quality was gradually reduced. Figure 10 displays the frequency of occurrence of driving situations based on all 29 test drives in terms of a color map and the comments given by the test subjects while driving. The rectangles show the respective driving conditions, where comments were given, and the connotation of the comments indicated by color. It must be noted that repeated comments by a subject, who replicates comments, are not shown. This avoids that a subject gains in importance due to repeated comments.

13

Acceptance of synthetic driving sounds in the interior of electric vehicles

Figure 10: Distribution of comments (classified as positive, negative or neutral) referring to the different sound concepts over speed and acceleration

Clusters of comments were found in the mid-speed range with moderate acceleration and in the low speed range with positive and negative acceleration. This reflects the most relevant driving conditions in perceptual sense. In the low speed range with low or negative acceleration, mainly negative comments occurred. Sound concept 3 stimulated obviously several positive comments in the mid-speed range with moderate acceleration. By means of a text-analytic investigation of the comments and interview data further information is achieved. In cruising situations, subjects requested a “quiet”, “gliding”, “adequate” and “discreet” sound. Sound concept 1 achieved several negative comments in the low speed range with low or medium acceleration. The diverse comments referred to a “relative loud”, “obtrusive” and salient acoustic feedback, which did not fit to the respective driving situation at all. Although, it was frequently stated that sound character of concept 1 is acceptable and pleasant, the sound was perceived as “inappropriate”, “too low frequent” and “not suitable” for distinct driving situations. A subject even concluded “I feel deceived”. These comments make clear that the adequacy check of a developed sound has to be determined within the original context. Moreover, it strikes that the ratio of positive, neutral and negative comments of sound concept 1 is similar to the original sound situation. It allows for drawing two conclusions: (a) the original vehicle sound has some sound quality deficits, and (b) the introduced sound concept 1 produces less negative comments related to the original sound of the

14

Acceptance of synthetic driving sounds in the interior of electric vehicles vehicle due to masking (lowest number of comments still referring to the original sound for the three sound concepts) and at the same time provokes (new) negative comments related to the presented synthetic sound. Sound concept 2 received the lowest relative number of positive comments provoking mostly comments in driving situations with low acceleration. Subjects stated that the sound was “uniform”, “monotonous” and “artificial”. The “omnipresence of sound” with a relative constant sound character was clearly rejected, since the sound becomes “exhausting” and “tedious” after a certain time. Sound concept 3 was perceived as “inconspicuous” leading to the best assessment of sound. The sound was described as “pleasant” and “suitable”. A few subjects even did not recognize the presence of additional interior sound, but gave more positive comments than in the original sound only test condition. In the original sound test condition, some negative comments were collected, like “howling”, “whistling” or “annoying tones”. Obviously, the original vehicle sound has some sound quality deficits. This illustrates the possibility to improve the sound quality by energetically or attentionattracting masking.

4 Conclusions and Outlook Based on a context-sensitive, explorative method the acceptance of different sound concepts in an electric vehicle was studied. It turned out that offering synthetic driving noises in the interior of the test vehicle led to more comments and did not necessarily foster positive evaluations or perceptions of the car and its acoustics [25]. The higher number of negative comments in the synthetic sound scenarios shows the high sensitivity of customers regarding perceived (sound) quality. Adding synthetic sound, which does not match the customer preferences well, leads to a reduction of sound quality. Moreover, the study has shown that target conflicts occur. Test subjects expressed their general preference for a quiet electric vehicle, but demand an adequate acoustic load feedback. It appears that a well implemented adaption of loudness in case of load changes is of utmost relevance and can improve perceived quality. Moreover, test subjects were inclined to favor inconspicuous, discreet sounds, which in turn lead to an increase of felt acoustic transparency of the vehicle, which was negatively connoted in the study [23]. Moreover, although even few subjects were not aware of the presence of a synthetic sound at all during a test drive, this sound concept achieved a slightly better assessment than the original sound only condition. It illustrates that a discreet sound character can lead to positive comments. But, this sound does still not fully mask disturbing noises within the original vehicle sound. In general, it is known that the general acceptance of a technology with its inherent properties like sound is reached through different steps. However, the acceptance is not only determined by technological progress and innovation, it is negotiated in society. Diverse actors like politicians, local authorities, researchers, manufacturers interact in social debates to shape function and

15

Acceptance of synthetic driving sounds in the interior of electric vehicles form (like the implementation of synthetic sound) of a new technology supporting or impeding the spread of a technology. This means that the electric vehicle technology is not automatically successful, because it is technically feasible. Aspects like ecology, consumer needs, industrial strategies and interests moderate the success of a new technology and its “design”. Thus, the customer with its appreciation of a technology and its realization is influenced by diverse aspects, which might also change the requested sound quality over the years.

References [1]

Blauert, J., Jekosch, U. (1997). Sound-quality evaluation – a multi-layered problem, Acustica & Acta Acustica 83 (5), Hirzel Verlag, 747-753

[2]

Guski, R. (1997). Psychological methods for evaluating sound quality and assessing acoustic information, Acustica & Acta Acustica, Vol. 83 (5), 765-774

[3]

Västfjäll, D. (2004). Contextual influences on sound quality evaluation, Acta Acustica united with Acustica, Vol. 90, Hirzel Verlag, 1029-1036

[4]

Guski, R., Blauert, J. (2009). Psychoacoustics without psychology, NAG/DAGA 2009, Proceedings, Rotterdam, Netherlands

[5]

Blauert, J., Jekosch, U. (2007). Auditory quality of performance spaces for music – the problem of the references, ICA 2007, Proceedings, Madrid, Spain

[6]

Pietila, G.M. (2013). Intelligent system approaches to product sound quality analysis, Doctoral thesis, Michigan, USA

[7]

Wyman, O. (2009). Elektromobilität 2025 – Powerplay beim Elektrofahrzeug, München, Germany, 2009

[8]

Pletschen, B. (2009). Comfort quo vadis? In conflict between crisis and CO2 challenge, Aachen Acoustics Colloquium 2009, Proceedings, Germany, 2009

[9]

Hofmann, M., Lauer, M., Engler, O. (2012). Sound design at Mercedes AMG, Aachen Acoustics Colloquium 2012, Proceedings, Germany, 2012

[10] Zeitler, A. (2007). Kognitive Faktoren bei der Skalierung von Höreindrücken.

DAGA 2007, Proceedings, Stuttgart, Germany [11] Sellerbeck, P., Nettelbeck, C. (2010). Enhancing Noise and Vibration Comfort of

Hybrid/Electric Vehicles Using Transfer Path Models, Aachen Acoustics Colloquium 2010, Proceedings, Germany, 2010 [12] The voice of blind and partially impaired people in Europe (2014). EBU Newslet-

ter No. 99, September – October 2014

16

Acceptance of synthetic driving sounds in the interior of electric vehicles [13] European Commission (2014). Proposal for a regulation of the European parlia-

ment and of the council on the sound level of motor vehicles, Procedure 2011/0409/COD [14] Singh, S., Payne, S.R., Mackrill, J.B., Jennings, P.A. (2014). Evaluation of electric

vehicle exterior sounds in virtual and real-world environments – a comparative study, Aachen Acoustics Colloquium 2014, Proceedings, Germany, 2014 [15] Blauert, J. (2005). Analysis and synthesis of auditory scenes, In: Blauert, J. (Ed.)

(2005). Communication acoustics, Springer-Verlag, Berlin, Heidelberg, Germany [16] Schulte-Fortkamp, B., Genuit, K., Fiebig, A. (2006). New approach for the devel-

opment of vehicle target sounds, In: 35th International Congress and Exposition on Noise Control Engineering, Internoise 2006, Proceedings, Hawaii, USA, 2006 [17] Sellerbeck, P. (2014). Eine innovative Methode zur Fahrzeugbewertung und Ziel-

geräuschdefinition, Magdeburger Symposium, Proceedings, Germany, 2014 [18] Kerkmann, J. (2014). Untersuchungen zur Akzeptanz synthetischer Fahrgeräusche

im Innenraum von Elektrofahrzeugen, Master thesis, TU Berlin, Germany, 2014 [19] Genuit, K., Fiebig, A. (2014). Sound design of electric vehicles – Challenges and

risks, Internoise 2014, Proceedings, Australia, 2014 [20] Küppers, T. (2012). Results of structure development process for electric vehicle

target sounds, Aachen Acoustics Colloquium 2012, Proceedings, Germany, 2012 [21] Shepard, R. N. (1964). Circularity in Judgments of Relative Pitch, J. Acoust. Soc.

Am. 36, 2346, 1964 [22] Genuit, K. (1996). Objective evaluation of acoustic-quality based on a Relative

Approach, Internoise 1996, Proceedings, UK, 1996 [23] Diechmann, M. (2014). Das Geräusch der Zukunft auf dem Prüfstand der Gegen-

wart Eine qualitative Untersuchung der Akzeptanz von synthetischen Innenraumgeräuschen bei einem Elektrofahrzeug, Master thesis, TU Berlin, Germany, 2014 [24] Mayring, P. (2010). Qualitative Inhaltsanalyse. In: Mey, G., Mruck, K. (eds.).

Handbuch Qualitative Forschung in der Psychologie, Wiesbaden, VS Verlag für Sozialwissenschaften und Springer Fachmedien GmbH, Germany, 2010 [25] Kerkmann, J, Schulte-Fortkamp B, Genuit K, Fiebig A. (2014). Acceptance of

synthetic driving noises in electric, Forum Acusticum, Proceedings, Poland, 2014

17

Interior noise effects of active vibration cancellation for a 4-cylinder engine M.Sc. Niklas Müller Dipl.-Math. Enrico Kruse, Andrew Harrison, MBA

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 W. Siebenpfeiffer (Hrsg.), Automotive Acoustics Conference 2015, Proceedings, https://doi.org/10.1007/978-3-658-27648-5_3

1

Interior noise effects of active vibration cancellation for a 4-cylinder engine

Abstract Electric vehicles with range extending internal combustion engines represent a particular noise and vibration challenge. The relatively quiet and smooth operation in electric mode is contrasted by the rather loud and harsh running range extender. This situation is often exacerbated by the need for weight reduction throughout the vehicle to maximize the electric driving range. The omission of passive noise reduction measures like insulation material and engine balance shafts reduces weight but further increases perceived noise and vibration. In this paper the application of an active vibration control system to such a vehicle with a gasoline range extender without balance shafts will be presented. The effects of error sensor choice and actuator inclination on interior noise and vibration levels will be addressed.

Kurzfassung Elektrische Fahrzeuge mit reichweitenverlängernden Verbrennungsmotoren stellen eine besondere Herausforderung bezüglich Innengeräuschen und Schwingungskomfort dar. Der relativ leise und gleichmäßige elektrische Antrieb steht in Kontrast zum eher lauten und rauen Lauf des Verbrennungsmotors. Die Situation wird oft noch verschärft durch den Zwang zur Gewichtsreduktion, um die Reichweite im elektrischen Betrieb zu maximieren. Der Verzicht auf passive Geräuschminderungsmaßnahmen wie Dämmung und Motorausgleichswellen reduziert Gewicht, erhöht aber zusätzlich den empfundenen Geräusch- und Vibrationspegel. In diesem Artikel wird die Anwendung eines Systems zur aktiven Schwingungskontrolle in einem solchen Fahrzeug mit benzinbetriebenem Reichweitenverlängerer ohne Ausgleichswellen präsentiert. Die Auswirkungen verschiedener Sensorvarianten und Aktuatoranstellwinkel werden dargestellt.

2

Interior noise effects of active vibration cancellation for a 4-cylinder engine

1 Introduction and task On a global scale, 4-cylinder engines are and will continue to be used in the majority of all passenger cars. They function as their main means of propulsion (diesel, gasoline, flex fuel, etc.) or work in hybrid drives in combination with electric motors. Nowadays turbo charging, downsizing, downspeeding and the removal of balancer units, to name just a few, are commonplace measures in 4-cylinder engines to reduce fuel consumption and CO2 emissions while increasing the specific output of said engines. However, all these measures typically also generate or exacerbate NVH (noise, vibration, harshness) issues inside and outside the cabin. The goal of this project was to use active vibration cancellation measures to find a better balance between CO2 reduction and noise and vibration requirements. The target vehicle for this investigation is a range extended hybrid electric car from a major OEM. The internal combustion range extender is a 4-cylinder gasoline engine without balance shafts. The hybrid powertrain was mounted in the vehicle using a standard pendulum style setup for laterally installed engines: load carrying mounts above the powertrain center of gravity (right hand side and left hand side) and a single torque link at a lower, more central position.

Fig. 1: powertrain and mounting system

3

Interior noise effects of active vibration cancellation for a 4-cylinder engine The original setup showed several areas of high interior noise levels (“booming”) with a significant contribution from the 4-cyliner engine’s 2nd order – the order that would otherwise have been canceled by the balance shafts. An active vibration control system was applied to tackle these 2nd engine order NVH issues. The challenging 2nd engine order interior noise target was derived from a sameclass competitor vehicle also sporting a 4-cylinder gas engine but with balance shafts and with active noise cancellation through loudspeakers. In comparison to balance shafts an AVC solution is ● ● ●

Lightweight and practically frictionless, i.e. less fuel consuming More adaptive to environmental and production variations Smaller in size

All vehicle tests were carried out at the customer site in an anechoic chamber on a dynamometer to ensure controlled test conditions. Slight variations in the “AVC off” baseline results were still observed. Possible reasons besides the natural variation: The vehicle was a prototype, and the installed actuator had a small mass effect even when it wasn’t operated.

2 Active Vibration Control System An active vibration control solution is comprised of these main components: ● ● ●

4

Actuator to generate a force output to be transmitted to the vehicle structure Input sensor (either microphone or accelerometer) to measure the vehicle response, typically either in g (or m/s2) or dB (or dB(A)) Electronic control unit (ECU) including the control software to process input signals, to determine the appropriate output response to the vehicle, to feed back information about failure states and to tune and monitor the system during installation

Interior noise effects of active vibration cancellation for a 4-cylinder engine

Fig. 2: Active Vibration Control (AVC) system

2.1 Actuator The actuator is the system component that generates the counter vibrations. It has to be small enough to fit yet strong enough to provide sufficient force output over the entire operating frequency range. There are different constructions of actuators available: piezoelectric, magnetostrictive, electrodynamic or linear proportional (variable reluctance / solenoid type). The first two are not suitable for consideration as they do not offer enough stroke of the mass required to generate an appropriate level of force. This leaves both electrodynamic and linear proportional as actuator types for consideration. TrelleborgVibracoustic has chosen the uni-axial electrodynamic shaker design. It provides the following advantages over linear proportional / variable reluctance type constructions: ● ● ● ●

Linear output response Smaller package possible Response more stable at high temperatures Better response at medium and high frequencies, i.e. mid and high engine speed range

5

Interior noise effects of active vibration cancellation for a 4-cylinder engine The actuator, along with the control hardware and software, determines the reaction speed of the AVC. Electrodynamic shakers show a much faster displacement response than variable reluctance devices. Figure 3 shows how the electrodynamic shaker almost instantaneously responds to the downshift from 3rd to 2nd gear.

Fig. 3: interior noise variation during downshift from 3rd to 2nd gear

2.2 Sensor Both the microphone and accelerometers used for the AVC were standard automotive components. This ensures compatibility and lowers the threshold from demo car prototype to serial industrial application of the AVC. The microphone was placed inside the cabin so it didn’t need any special protection from the elements. The accelerometer is a completely sealed design suitable for under hood usage in cars: ● ● ●

Meets IP6K9K standard (sealed against dust, jet wash, submersion) Wide range of operating temperatures (-40 to +125 °C) Reverse battery & overvoltage protection

The accelerometer could be integrated into the shaker housing. But this only makes sense for a serial application, and only for local control setups. (It is not always optimal to place the sensor right next or to or inside the actuator.)

6

Interior noise effects of active vibration cancellation for a 4-cylinder engine

2.3 Control hardware The electronic control unit (ECU) is fully contained in a sealed casing meeting the necessary automotive standards for under hood installation. It comprises: ● ● ● ● ●

Main microprocessor running the source code Interface circuitry for the vehicle CAN bus Drive circuit to operate the actuator Onboard memory Input circuit (if required) for the error sensor

Besides the source code which includes the main Active Vibration Controlling algorithms, the ECU also contains interface software to allow a hookup via PC to support vehicle development activities. Dependent upon vehicle packaging environment (temperature!), actuator size and system type (open or closed loop), the controller is located remotely (e.g. near the firewall) or within the actuator assembly. In a serial application the electronics could also be integrated into other, already existing, vehicle controllers.

2.4 Control algorithm Active vibration control solutions can be roughly classified into open loop and closed loop systems. Closed loop systems use a continuous feedback loop employing an accelerometer or microphone as the error sensor. The engine’s unbalanced 2nd order shaking forces are the target in this project. These excitations are purely mass and inertia generated, i.e. they depend on the engine speed but not on the engine load. (E.g. at constant engine speed it doesn’t matter if the vehicle is going uphill or downhill.) In such a case a closed loop feedback is not necessary and an open loop system may suffice. In open loop systems the determination of the drive to the actuator is performed simply by cross checking a lookup table which has listed the output amplitude and phase shift required by the actuator at any given point in the engine rpm range. In theory this lookup table can indeed be quite complex, and include other determining factors available on the vehicle CAN (controller area network) bus, such as ambient temperature, vehicle loads, battery voltage and so on.

7

Interior noise effects of active vibration cancellation for a 4-cylinder engine Benefits of this approach over a closed loop system are: ●

Significant cost reduction: ● ● ●

● ●

Removal of input sensor and associated input circuitry Small and cheaper electrical connector Simplification of ECU and wiring bundle

Smaller ECU PCB (printed circuit board) resulting in smaller housing or easier integration within the actuator Complete removal of EMC (electromagnetic compliance) concerns with sensor input line

It should be noted that an error sensor (accelerometer or microphone) is still needed at the position of issue to train the system and populate the lookup table, just not to operate the system once it is initiated.

3. Approach and Vehicle Test Results 3.1 Placement of actuator In principle, the electrodynamic shaker can be placed anywhere in the vehicle provided there is a flat enough surface and sufficient room. It can also work under any inclination angle. Obviously, in this project it needs to be placed where it results in the biggest reduction in interior noise. It is also evident to place the actuator on the body side rather than on the powertrain itself. That way the engine mounting system can provide the first layer of isolation and the AVC only needs to deal with the remaining vibrations. There were significant time constraints in this project, so a comprehensive transfer path analysis from the powertrain mounts to the driver sensor points (namely the driver’s ear, but also seat track, steering wheel, …) could not be carried out. However, it was quite clear that the unbalanced 4-cylinder engine would generate a significant amount of 2nd order engine shaking. The acting point of this shaking force is the middle of the crankshaft, i.e. closer to the right hand side (RHS) mount than to the left hand side (LHS) one. Vibrations at the powertrain mounts in all directions under several operating conditions were checked. As can be seen in the example in figure 4, vibrations in RHS Z proved to be dominant, as anticipated. Vibrations in other directions or at other mount locations were much lower.

8

Interior noise effects of active vibration cancellation for a 4-cylinder engine

Fig. 4: 2nd order powertrain side vibration levels during engine run-up

Removing the RHS mount completely (and supporting this side of the powertrain to ground to eliminate this vibration path) dropped the 2nd order content of the interior noise in some engine speed ranges by up to 5-15 dB. This served as another indicator that placing the actuator upright and on the rail close to the RHS powertrain mount should be a good start.

3.2 Selection of error signal The main target in this project was the reduction of interior noise, or more specifically the 2nd engine order content thereof. So selecting a microphone as the error signal source seems a natural choice. However, this solution would mean additional equipment in the cabin and wiring between the microphone and the controller. Experience showed that in some cases it is enough to focus on the cancellation of the local vibration levels right at the location of the actuator. Positioning an accelerometer next to or even integrated within the actuator also makes for a very compact “plug and play” solution. Of course, this assumes the sensor can withstand the sometimes harsher conditions there, such as increased temperatures and generally higher vibrations levels (road shake).

3.3 Open loop control with local accelerometer So in a first set of investigations an accelerometer was placed right next to the shaker, on the rail close to the RHS powertrain mount. The vertical vibrations at that point were taken as the input signal for the electronic controller. With this setup a broadband reduction of the local vibrations was achieved.

9

Interior noise effects of active vibration cancellation for a 4-cylinder engine

Fig. 5: vibration reduction at actuator position with local error sensor at actuator

However, this improvement did not materialize in the cabin. The vertical seat rail vibrations were somewhat lowered in certain speed ranges but overall the improvement was negligible and not perceivable by the driver.

Fig. 6: 2nd order vertical vibrations at seat rail with local error sensor at actuator

The most important signal, the 2nd order driver side interior noise, remained largely unaffected. The major “booms” still existed. In fact, around 2500 rpm the interior noise even deteriorated.

10

Interior noise effects of active vibration cancellation for a 4-cylinder engine

Fig. 7: 2nd order interior noise at driver’s ear with local error sensor at actuator

Possible reasons for this discrepancy could be: ● ● ●

Cancelling only the RHS Z transfer path neglects potential interactions with other transfer paths down the line. The achieved local reduction was not high enough. The local accelerometer was not placed on a suitable part of the vehicle structure.

It should be noted that a further optimization (e.g. through algorithm tuning, shaker inclination or sensor repositioning) of this setup would have been possible but it was deemed more effective to use a microphone next.

3.4 Open loop control with microphone Eventually, after a driver side microphone was used as the error signal to control the actuator. The actuator itself was still placed near the RHS powertrain mount in pure vertical direction. This setup achieved a broadband reduction in interior noise levels for the driver; the results met the given target over the entire engine speed range. The 2nd order interior noise at the “booming spots” was diminished by up to 20 dB(A). Even at the co-driver side the noise could be reduced for engine speeds above 3800rpm; so this side now also met the target. For the rear passengers the noise levels remained more or less unchanged. Rear bench noise had not been an issue before, so this result is also well acceptable.

11

Interior noise effects of active vibration cancellation for a 4-cylinder engine

Fig. 8: 2nd order seat vibration levels during engine run-up

12

Interior noise effects of active vibration cancellation for a 4-cylinder engine

3.5 Effect of actuator inclination Finally, the influence of inclining the actuator about the X axis (i.e. top side inboard) was investigated. This should direct some of the cancelling force into the Y direction which was showing the second highest vibration levels at the RHS mount for higher engine speeds (compare fig. 4).

Fig. 9: 2nd order interior noise at driver’s ear with varying actuator inclination

Analysis of the test results: ● ● ● ●

Range A: target met even without AVC; no major difference between AVC setups Range B: all AVC setups meet the target; 0° is optimal Range C: target barely met even without AVC; all AVC setups reduce noise but no clear winner Range D: all AVC setups reduce noise but only 0° shaker inclination meets the target

Consequently, the shaker was kept at the original installation angle of 0° (pure vertical) for optimal noise reduction.

13

Interior noise effects of active vibration cancellation for a 4-cylinder engine

4 Conclusions and outlook It was shown that a single actuator AVC system with open loop control is already sufficient to meet the customers NVH requirements. It is able to reduce interior noise over a wide range of frequencies to levels otherwise only achievable through balance shaft units or the like. This is especially remarkable since the AVC has no direct impact on the air borne noise paths between the engine and the driver’s ear. Since the completion of this project, TrelleborgVibracoustic has also successfully applied ● ● ● ●

Closed loop systems Two and three AVC systems in parallel on the same vehicle Multi-order cancellation Cancellation of full body resonances, steering wheel vibrations, etc. etc.

With the increasing number of hybrids, the use of downsized engines (3- and 4-cylinder engines) and ever more lightweighting efforts throughout the automotive industry, active solutions are bound to become a part of the standard NVH solution toolbox.

14

Simulation of exterior powertrain ATFs on an engine bay mock-up with complex trim configurations Abdelkader Bihhadi (Autoneum France), Claudio Bertolini (Autoneum Switzerland), Christophe Locqueteau (Renault)

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 W. Siebenpfeiffer (Hrsg.), Automotive Acoustics Conference 2015, Proceedings, https://doi.org/10.1007/978-3-658-27648-5_4

1

Simulation of exterior powertrain ATFs on an engine bay mock-up with complex …

1 Introduction This article is a follow-up of companion articles published at the Automotive Acoustics Conference in 2011 and 2013 [1]-[2], documenting a numerical/experimental activity aimed at assessing the capabilities of deterministic numerical methods in relation to the simulation of powertrain exterior Acoustic Transfer Functions (ATFs). At introductory level, it is useful to briefly recall (more details and related bibliography can be found in [2]) that the main motivation for this activity comes from the changes that during the last years took place in the legislation about the exterior noise generated by transportation vehicles. These changes were the result of a rather long process, initially triggered by the realization of the fact that the old ISO-362 measurement method (introduced in 1975) and SPL limits were not representative of the actual urban driving conditions any longer and that, as a consequence, such regulation was ineffective in reducing the noise disturbance perceived by the population. It is in view of these changes that, in the forth-coming years, OEMs might be forced to a very aggressive approach to the reduction (if not elimination) of powertrain noise and that, as a consequence of this, reliable simulation tools in this field of automotive NVH might become again relevant, after many years during which OEMs have been able to achieve pass-by targets and solve issues related to the acoustic radiation from the engine bay simply on the basis of experience and good practices. As it will be recalled in more detail in section 2, the concrete test-case considered in the activity is constituted by an engine-bay mock-up (the size is approximately that of the engine bay of a C-segment vehicle) made with stiff plywood and presenting, -in a geometrically simplified way- all the features that make from the numerical standpoint the simulation of powertrain exterior ATFs challenging: a pseudo-closed acoustic environment with narrow gaps (the engine bay) inside which acoustic treatments are applied and that is coupled with the exterior space through localized apertures. All this combined with the need of simulating ATFs on a rather wide frequency range (at least up to 3.5kHz), a condition that potentially leads to big models and then high computational effort. The activity documented in [1]-[2] was a re-assessment of the main deterministic numerical methodologies commercially available for the simulation of powertrain exterior ATFs. Namely, the Boundary Elements Method (BEM), the Infinite Finite Elements Method (IFEM), the Automatically Matched Layer Method (AMLM) were used to calculate exterior ATFs in different engine-bay mock-up configurations. Such configurations differed one from the other for the opening/closing of one or more apertures in the engine-bay mock-up walls and for the application/non-application of acoustic treatments on these same walls. The different simulation methodologies were compared in terms of results quality (assessed by means of correlation with test data) and computational performance. It turned out that basically all 3 methodologies, when properly applied, could guarantee a

2

Simulation of exterior powertrain ATFs on an engine bay mock-up with complex … more than acceptable correlation with test data. Only IFEM and AMLM, though, could provide a computational performance compatible with industrial applications, being BEM by far too slow. Unfortunately, a really direct comparison between IFEM and AMLM in terms of computational performance was not possible, since different machines were used to run the calculations with these 2 methods. From the results, though, it appeared that AMLM could guarantee a slightly shorter computational time. The novelty of the work presented in this article -in comparison with [1] and [2]- consists in the complexity of the acoustical treatments considered as well as in the complexity of the numerical technologies used to represent them. Namely, in [1] and [2], the acoustical treatments considered were very simple, consisting in a 20mm thick doublelayer felt that could be numerically represented as an impedance boundary condition applied on the inner face of the engine-bay walls. In this article, the acoustical treatments considered, while being still geometrically simple, involve materials and multilayers that are more realistic for treatments in the engine bay and for whose representation poroelastic Finite Elements are in principle required. This ‘extension to more complex acoustic treatments represents a necessary step towards the application of the analysed simulation methodologies to actual industrial test-cases. At the same time such extension represents also, from the physical and numerical standpoint, a new challenge compared to what was presented in [1] and [2]. From the physical standpoint, the mechanisms that lead to the dissipation of energy inside the treatments that will be analysed in this article are more difficult to model than those involved in a simple two-layer felt wall treatment and that consist –basically- just in the friction between the acoustic waves and the felt fibres. From the numerical standpoint, poroelastic FE involve the use of non-compatible meshes and are, furthermore, generally computationally expensive. From all this it appears that what will be presented in the article is more than just a bare repetition of what was presented in [1] and [2]. The paper is organized as follows. In the next 2 sections, the experimental activity is presented and the numerical model(s) used for the simulations are described in detail. After this, the correlation between test and simulation data is reported and analysed, both through single-point-ATF examples and through a more statistical approach (given the amount of test data available, this seemed to be the most sensible approach here). Eventually, some final conclusions are drawn.

3

Simulation of exterior powertrain ATFs on an engine bay mock-up with complex …

2 Experimental Activity A car engine bay mock-up was built using stiff plywood (see figure 1), trying to keep the dimensions and the shape of the mock-up as close as possible to those of a real Csegment engine bay. The presence of the engine was taken into account by putting another stiff plywood structure (a kind of “engine block mock-up”, see figure 2) inside the engine-bay mock-up. The shape and the dimensions of this latter structure roughly reproduced those of an engine block (including ancillaries), leaving a relatively small airgaps (in the order of 5-10 cm) in between its walls and the side walls.

Figure 1 – Engine Bay Mock-Up in plywood. Side and rear apertures are visible.

4

Simulation of exterior powertrain ATFs on an engine bay mock-up with complex …

Figure 2 – Plywood structure representing the engine block. This was positioned inside the engine bay mock-up shown in figure 1.

Furthermore, a few apertures were made on the external walls of the engine bay mockup, in order to simulate the main apertures normally present on the boundary of a real vehicle engine bay. A first aperture was put in the front wall (radiator), one in the back wall (driving shaft), two small apertures in the bottom wall (to simulate the apertures normally present in the under-engine shield for ventilation purposes) and two in the side walls (apertures for wheel axles). Two of these apertures (right and rear) are visible in figure 1. The overall arrangement of the apertures is shown in figure 3.

5

Simulation of exterior powertrain ATFs on an engine bay mock-up with complex …

Figure 3 – Three different views of the engine-bay mock-up, where the apertures on the walls are visible

For ATF tests, 21 microphone positions were distributed around the engine mock-up. These are shown in figure 4.

6

Simulation of exterior powertrain ATFs on an engine bay mock-up with complex …

Figure 4 – Microphone positions on engine mock-up faces

As far as source positions are concerned, 12 source positions were considered, arranged on 2 rows as shown in figure 5. The first row is at a distance of 1m from the outer-most corner of the engine-bay mock-up, while the second row is at a distance of 3m from the

7

Simulation of exterior powertrain ATFs on an engine bay mock-up with complex … same corner. Both rows are on the right side of the engine bay mock-up. Furthermore, all sources are positioned at 1.2m from the ground. On each row, the distance between 2 adjacent source positions is 1m.

Figure 5 – ATF source positions relative to engine-bay mock-up

8

Simulation of exterior powertrain ATFs on an engine bay mock-up with complex … Figure 6 gives an overall view of the acoustical treatments that were applied on the engine bay mock-up. These resembled the typical “wall” acoustic treatments applied inside an engine bay (a hoodliner, a dash outer and an engine under-shield), with the addition of 2 engine side panels. These latter parts were added in order to shield the side faces of the engine bay mock-up from its side apertures. As a matter of fact, previous tests (see [1] and [2]) demonstrated how classical ‘wall’ treatments are rather ineffective in reducing ATFs relative to engine side faces and to side source positions (frequency-averaged Insertion Loss of the treatments was in the order of 2-3dB). This can be somehow expected because of the existence –for such a case- of a direct path between source and microphones that is not directly affected by the application of acoustic treatments. Beyond this, engine side panels are also interesting in relation to the fact that they are more and more frequently seen as part of body-mounted engine encapsulations [3]. An analysis of their impact on exterior ATFs can then be of interest both for suppliers and OEMs. It is important to remark that, differently from the other more traditional treatments, these engine side panels are not applied directly on the surface of the engine block. Rather, they are applied in between the side surfaces of the engine block and the side walls of the engine bay.

Figure 6 – Overview of acoustical treatments applied inside the engine-bay mock-up

The acoustical multilayer used to manufacture the engine side panels is called ThetaFiberCellTM and includes 2 materials: Theta-Fiber™ and Theta-Cell™. Theta-Fiber™ is a fibre based material, which acts as a carrier and guarantees high stiffness and good mechanical properties. The high thermal stability (200 °C in peak and 160 °C at 1000 h)

9

Simulation of exterior powertrain ATFs on an engine bay mock-up with complex … allows the application of the material very close to heat sources and makes it then suitable for engine encapsulations (see, e.g. [3]). Theta-Cell™ is a semi-rigid PU foam with low density (~15kg/m3) and high absorbing properties. The combination of the two materials leads to a lightweight multilayer (area weight between 1000gsm and 1500gsm) with excellent acoustic and thermal properties. Theta-FiberCellTM was used for the engine side panels since this multilayer, while keeping a very good absorption on both sides, it guarantees also a relatively good insulation (insertion loss up to 15dB at high frequencies) and this feature is of course important for a treatment that is meant to somehow isolate the engine sides’ acoustic radiation from the engine bay side apertures. For the hoodliner and the dash outer, being the acoustic function of these parts purely absorptive, a 20mm layer of Theta-Cell™. Light PU foams are actually nowadays used by many car makers for this kind of engine-bay absorbers, due to the fact that they feature a unique combination of very good acoustic absorption and very low weight. Eventually, for the engine under-shield Autoneum Ultra-Silent was used. This is a 100% polyester fibrous material normally used by Autoneum to manufacture engine undershields and –more in general- vehicle under-body parts. The part installed on the enginebay mock-up was cut from an 8mm flat blank having an area weight of 1200gsm. Similarly to what was done in [1] and [2], ATFs were measured on the engine-bay mock-up in different configurations. These are summarized in table 1 and differ one from the other for what concerns application of treatments. In this paper, both for brevity reasons and for time reasons, only results related to configurations 1 and 4 will be reported and analysed. Table 1 – List of configurations measured

all open

Treatment

10

no treatment

Configuration 1

dash outer + hoodliner

Configuration 2

dash outer + hoodliner + under engine shield

Configuration 3

all treatments

Configuration 4

Simulation of exterior powertrain ATFs on an engine bay mock-up with complex …

Simulation Activity To simulate the exterior ATFs on the engine-bay mock-up the APML (Adaptive Perfectly Matched Layer) simulation technique available in Actran 14.0 [4] was adopted. The method is called ‘adaptive’ because the PML mesh is automatically (and implicitly) generated at each frequency without the need of any action from the user and this allows an efficient modeling of the acoustic radiation problem, resulting is a reduction of the meshing work and in a gain in terms of calculation time. Acoustical treatments were modeled by means of poroelastic FE domains that were meshed independently from the acoustic domains surrounding them, both for what concerns the element type and for what concerns the mesh size. The coupling between poroelastic FE domains and surrounding acoustic cavities is managed by means of suitable incompatible mesh interface conditions. The Adaptive Perfectly Matched Layer (APML) method is a derivation of the classical Perfectly Matched Layer (PML) method. In the PML, the (theoretically unbounded) acoustical domain is meshed with standard finite elements up to a certain distance from the radiating or scattering structure. Standard fluid material properties are assigned to these elements. A suitable portion of the acoustical domain immediately outside this meshed region is then also meshed, but its elements are assigned strongly absorbing properties, in such a way that any reflection back to the radiating or scattering structure is avoided. In practice, this latter portion of the acoustical domain has often the shape of a “layer” that wraps the meshed region close to the structure and its acoustical properties are defined in such a way to reproduce (as much as possible) point by point the Sommerfeld radiation condition that guarantees the absence of acoustical reflections. It is for this reason that the method is called “Perfectly Matched Layer”. One practical difficulty of this method is the fact that, in principle, the mesh of the PML is frequencydependent: as it is intuitive, the lower the frequency, the thicker the PML has to be. In the APML implementation, the generation of the PML is automated (thus the name “Adaptive PML”) and all the user has to do is to build the standard FE mesh of the nearfield region. Acoustic quantities outside of the near-field meshed domain are evaluated via a post-processing step that is based on standard boundary-integral techniques that make use as an input of the acoustic pressure and velocity on the nodes belonging to the outer surface of the same near-field meshed domain (something similar to what happens for traditional Boundary Elements techniques). Figure 7 shows a section of the acoustical domains included in the simulation model: the ‘Interior Fluid’ represents the air inside the engine bay, while the ‘Exterior Fluid’ represents the near-field portion of the external acoustic space that has actually been meshed. For these acoustic domains, linear tetrahedral elements were used with a size of 14mm. This allows running simulations up to about 3500Hz having 7 elements per

11

Simulation of exterior powertrain ATFs on an engine bay mock-up with complex … wavelength. Figure 8 shows the mesh used for the acoustical treatments. For these, linear hexahedral elements were used, with a size of 4mm in the thickness direction and a size of 8mm in the in-plane direction. Two different options were investigated for what concerns the underlying formulation of Biot equations. In a first case, the classical u-p formulation of Biot equations was considered [5]-[7]. In a second case, a simplified formulation based on the assumption of a rigid skeleton was considered [8]. This second formulation basically assumes that the vibration of the solid phase of all porous materials involved is negligible. In such a situation, it is possible to reduce porous materials to ‘equivalent fluids’ having a frequency dependent complex density and speed of sound and this should, in turn, lead to a substantial saving in terms of computational efficiency. For all materials (Theta-Cell, Theta-Fiber and Ultra-Silent), Biot parameters were taken from available and validated material models. The mesh of the acoustic domains (figure 7) and of the poroelastic domains (figure 8) were assembled in one single model as shown in figure 9. The coupling between acoustic and poroelastic domains (that are meshed in an independent way) is managed through suitable interface conditions based on geometrical projection algorithms [4].

Figure 7 – Simulation model – Mesh of acoustic domains

12

Simulation of exterior powertrain ATFs on an engine bay mock-up with complex …

Figure 8 – Simulation model – Mesh of poroelastic domains

Figure 9 – Simulation model – Assembly of acoustic and poroelastic domains

The total number of elements, nodes and degrees-of-freedom in the whole model is presented in table 2.

13

Simulation of exterior powertrain ATFs on an engine bay mock-up with complex … Table 2 – Number of elements and nodes in the acoustic and poroelastic domains of the simulation model

In order to perform ATFs calculations, 21 receiver points were defined around the engine mock-up as shown in figure 10. The positions of these receiver points corresponded of course –as much as possible- to those considered for the microphones during the tests. For computational reasons, though, the receiver points in the model are not taken directly on the rigid surface of the engine mock-up, but at a distance of about 10mm (thus anyway much lower than the acoustic wavelength) from it. Similarly, 12 different load cases were defined, corresponding to 12 monopole sources located at positions corresponding to those considered during the tests.

14

Simulation of exterior powertrain ATFs on an engine bay mock-up with complex …

Figure 10 – Positions of receiver points in simulation model. These correspond to microphones position used in ATF tests

15

Simulation of exterior powertrain ATFs on an engine bay mock-up with complex … Table 3 – Calculation Frequencies 200 500 830 1250

Hz Hz Hz Hz

Frequency Band to to to to

500 830 1250 3500

Hz Hz Hz Hz

2 3 4 5

Step Hz Hz Hz Hz

Figure 11 – Position of acoustic sources in simulation model, corresponding to those used in ATF tests.

Eventually, table 3 reports the detailed list of the calculation frequencies used. An increasing frequency step was used, in such a way to guarantee an even (and sufficient) number of calculation frequencies for each 1/3 octave band between 250Hz and 3150Hz. Al computations were carried out in a Linux Machine Intel(R) Xeon(R) CPU E5-2650 v2 @ 2.60GHz, equipped with 8 cores having a total of 193 GB of RAM. The frequency response calculation was accomplished using the frequency parallelism on 4 cores, and allocating 45 GB per core. This led to the computational times shown here below in table 4.

16

Simulation of exterior powertrain ATFs on an engine bay mock-up with complex … Table 4 – Summary of computation times Case 1 Case 4 FEM rigid-porous Case 4 FEM porous-elastic

33 Hours 33 Hours 74 Hours

Correlation between measured and simulated ATFs This section includes a detailed analysis of the correlation between measured and simulated engine-bay mock-up ATFs. Given the relevant amount of data available, this analysis can be carried out at different levels. A first level consists of course in checking the correlation of “single-point ATFs” (i.e. one microphone position/source position combination). Three examples of numerical/experimental correlation at this level are shown in figures 12 to 14 both for case 1 (all apertures open – no treatments) and for case 4 (all apertures open – all treatments applied). These examples were chosen following 2 criteria. First of all, they all refer to non-trivial source-receiver transmission paths, where source and receiver do not ‘see each other’ through one of the apertures in the walls of the engine bay mock-up. Secondly, they were chosen in such a way to be representative of the ‘average’ correlation level found. For some of the 12x21=252 source/microphone combinations the correlation found is better than what is shown in these figures, for other it is substantially worse. Furthermore, it has also to be considered that the assessment of the level of correlation involves not only source and microphone positions, but also the frequency. Given the type of simulation methodology used, it is expectable that the level of correlation is somehow frequency-dependent and tends to be worse at lower frequencies.

17

Simulation of exterior powertrain ATFs on an engine bay mock-up with complex …

Figure 12 – Comparison between measured and simulated ATF for microphone 5 (front face) and source position 19. Left: case 1 (all apertures open, no acoustic treatment). Right: case 4 (all apertures open, all acoustic treatments applied)

Figure 13 – Comparison between measured and simulated ATF for microphone 19 (left face) and source position 23. Left: case 1 (all apertures open, no treatments applied). Right: case 4 (all apertures open, all treatments applied)

18

Simulation of exterior powertrain ATFs on an engine bay mock-up with complex …

Figure 14 – Comparison between measured and simulated ATF for microphone 2 (bottom face) and source position 5. Left: case 1 (all apertures open, no treatments applied). Right: case 4 (all apertures open, all treatments applied)

These 3 examples seem to indicate that, in general, correlation between test and simulation is more than acceptable, being the discrepancies within 3-3.5dB (with some exceptions, like in the case of figure 13-right at 400Hz or at 2kHz). At the same time, though, it is clear that by means of just 3 (or even 5 or 10) examples –even when they are carefully chosen- it is difficult to draw clear objective and quantitative conclusion. For this reason, it was decided to check whether, by using simple statistical indicators, it was possible to obtain a clearer global picture of the level of numerical/experimental correlation. Namely, the “numerical/experimental discrepancy” was first calculated as the absolute value of the difference in dB between measured and calculated ATFs. For a fixed frequency band, in this way 252 (= number of source positions times number of microphone positions) delta-dB values are obtained and their statistical distribution is analysed. Figures 15 to 18 show the results obtained for 500Hz and for 2500Hz, both for case 1 (all apertures open, no treatments) and for case 4 (all apertures open, all treatments applied). In each figure, there are 2 sub-plots: on the left sub-plot the differential distribution is reported, while on the right sub-plot the cumulative distribution is reported, both calculated using a resolution of 0.25dB. Here, the differential distribution is evaluated in the following traditional way: given an error interval [x, x+0.25]dB, the value of the differential distribution corresponding to such interval is simply given by the number of simulations (i.e. source position/microphone position combinations) for which the discrepancy between the measured and the simulated ATF is included in such interval, i.e. it is between xdB and (x+0.25)dB. Correspondingly, the value of the cumu-

19

Simulation of exterior powertrain ATFs on an engine bay mock-up with complex … lative distribution relative to such interval is equal to the number of simulations for which the value of discrepancy between measured and simulated ATF is lower than xdB. Furthermore, in figures 15 to 18, on the right sub-plot a vertical red-line is also drawn: this line defines a specific value xcf of the discrepancy between measured and simulated data, such that 70% of the simulations (i.e. source position/microphone position combinations) have a discrepancy lower than xcfdB. In practice, this value defines a kind of 70% confidence interval. Eventually, figure 19 shows a chart of the 70% confidence interval values as a function of the frequency, both for case 1 and for case 4. All this gives an idea of the overall level of correlation between measured and simulated single-point ATFs that is able to embrace –in a statistical sense- all the available data and not just a few selected examples.

Figure 15 – Statistical distribution of dB-delta between measured and simulated ATFs Case 1 at 500Hz

20

Simulation of exterior powertrain ATFs on an engine bay mock-up with complex …

Figure 16 – Statistical distribution of dB-delta between measured and simulated ATFs Case 4 at 500Hz

Figure 17 – Statistical distribution of dB-delta between measured and simulated ATFs Case 1 at 2500Hz

21

Simulation of exterior powertrain ATFs on an engine bay mock-up with complex …

Figure 18 – Statistical distribution of dB-delta between measured and simulated ATFs Case4 at 2500Hz

Figure 19 – Numerical/Experimental correlation on single ATFs – 70% confidence limit

22

Simulation of exterior powertrain ATFs on an engine bay mock-up with complex … Figure 19 shows that the 70% confidence limit is in general lower (with the exception of the 800Hz band) for case 1 than for case 4, by about 1-1.5dB. This means that, on average, correlation for case 4 is worse than correlation for case 1 by about 1-1.5dB. This is likely to be due both to the higher numerical complexity of the simulations (and tests) for case 4 and also to uncertainties in the measurement and definition of Biot parameters for the materials constituting the various trim parts. Always from figure 19 it appears that the 70% confidence limit tends to decrease with frequency, but this is just a general tendency that is not really respected in a monotonic way. As a matter of fact, while it is true that PML methodologies tend to be intrinsically more accurate for high frequencies, it is also true that at high frequencies the discretization error tends to increase and these 2 conflicting tendencies are partially balancing each other. In any case, one can conclude that, on average, the accuracy expected for case 1 (no treatments, all apertures open) ranges from 3.5dB at low frequencies to 2dB at high frequencies while for case 4 (all treatments applied, all apertures open), it ranges from 4.5dB at low frequencies to 3.5dB at high frequencies. This level of accuracy can generally be considered acceptable but, again, it has to be considered with care because it has a statistical meaning. On specific single ATFs, much higher discrepancies can occur, up to 10-12dB on single frequency bands. All analyses considered so far were related to single-point ATFs, (i.e. ATFs for one source position/microphone position combination). A second level of analysis is the one that involves “face-average ATFs” where the averaging is carried out on groups of microphones positioned on the same face of the engine mock-up. Figures 20 to 22 show 3 examples of numerical-experimental correlation obtained for this kind of ATFs. The criteria adopted for the choice of these 3 examples (out of the 72 = 6 faces x 12 sources possible choices) were the same as those already mentioned for the case of single-point ATFs (i.e. typical average correlation and non-trivial transmission path). The level of accuracy found can be considered generally good, again typically in the order of 3dB. On face-average ATFs, statistical analyses similar to those considered for single-point ATFs were also carried out. Final results are shown in figures 23 to 25. Namely, in figure 23 the 70% confidence limit for the discrepancy between measured and simulated face-average ATFs is reported, for case 4 and for case 1. The chart in this figure confirms that, also for faceaverage ATFs, the discrepancies found for case 4 are on average bigger than those found for case 1. But for face-average ATFs the gap between the 2 cases is smaller than the one found for single-point ATFs. Furthermore, and differently from the case of single-point ATFs, the 70% confidence limit for the discrepancy between measured and simulated face-average ATFs seems to be relatively independent from the frequency. In figure 24, the 70% confidence limit for the discrepancy between measured and simulated singlepoint ATFs is compared with the one found for face-average ATFs for case 1. A similar comparison is shown in figure 25 for case 4. These figures show that, as one has to expect, correlation is generally better for face-average ATFs by 1-1.5dB.

23

Simulation of exterior powertrain ATFs on an engine bay mock-up with complex …

Figure 20 – Comparison between measured and simulated average ATF for front face and source position 19. Left: case1. Right: case 4

Figure 21 – Comparison between measured and simulated average ATF for left face and source position 15. Left: case1. Right: case 4

24

Simulation of exterior powertrain ATFs on an engine bay mock-up with complex …

Figure 22 – Comparison between measured and simulated average ATF for top face and source position 4. Left: case1. Right: case 4

Figure 23 – Numerical/Experimental correlation on face-average ATFs – 70% confidence limit

25

Simulation of exterior powertrain ATFs on an engine bay mock-up with complex …

Figure 24 –70% confidence limit on numerical/experimental correlation: single-point ATFs vs. face-averaged ATFs. Case 1

Figure 25 –70% confidence limit on numerical/experimental correlation: single-point ATFs vs. face-averaged ATFs. Case 4

26

Simulation of exterior powertrain ATFs on an engine bay mock-up with complex … Eventually, a third and last level of analysis involves ATFs obtained by averaging over all 21 microphones distributed over the surface of the engine. This kind of averaging is actually quite common when measuring interior ATFs. Figures 26 and 27 report the correlation between measured and simulated engine-averaged ATFs for 2 source positions. As one can see, the correlation is quite good and generally better than the one found for single-point ATFs and face-average ATFs. For engine-average ATFs it is clearly not possible to carry out statistical analyses similar to those previously considered for single-point and face average ATFs. The reason for this is the fact that the number of source positions is rather small and statistically not really meaningful. Just to have a quantitative idea of the numerical/experimental correlation for engine-average ATFs, one can simply take an average over the sources. Figure 28 shows the average discrepancy between measured and simulated engine-average ATFs both for case 1 and for case 4. By visually comparing the values reported in this figure with those reported in figures 24 and 25, it is possible to realize that by averaging over all the microphones distributed on the surface of the engine a higher accuracy is obtained above 500Hz. In making this comparison, though, some care has to be used because the two quantities (the 70% confidence limit on the numerical experimental discrepancy reported in figures 24 and 25 and the source-averaged discrepancy reported in figures 28) are not directly comparable from the statistical standpoint. In any case, from figure 28 one can fairly say that, on average, by means of APML it is possible to simulate engine-average ATFs with an accuracy of 2-3dB at low frequency and 1-2dB at high frequency both for case 4 and for case 1.

Figure 26 – Comparison between measured and simulated engine-averaged ATF for source position 19

27

Simulation of exterior powertrain ATFs on an engine bay mock-up with complex …

Figure 27 – Comparison between measured and simulated engine-averaged ATF for source position 4

Figure 28 – Source-averaged delta-dB discrepancy between measured and simulated engine ATFs.

28

Simulation of exterior powertrain ATFs on an engine bay mock-up with complex … Before leaving this section, it is interesting to have a look at the correlation between measured and simulated Insertion Loss (IL) curves. Basically, for a given ATF (single point or averaged), the corresponding IL represents the effect that acoustic treatments have on it, i.e. the difference between the ATF values measured for case 1 and for case 4. For a sound package designer, it is of course crucial that simulation methods are able to properly predict this quantity, in order to be able to rely on such methods to make proper decisions. Similarly to the case of ATFs, also the analysis of the correlation between measured and simulated IL curves can be done at single-point level, at faceaverage level and at engine-average level. In the case of IL, though, the single-point level is generally less interesting because normally the target of the application of a certain treatment is the reduction of the acoustic radiation from a certain area/face of the engine. Figures 29 to 31 show the IL relative to all source averaged face-ATFs. As one can see, the correlation is generally quite acceptable (within a couple of dB). For most faces, simulation shows a certain tendency to overestimate IL in the mid-frequency range. It is also interesting to observe that simulation seems to be able to respect the IL variations that take place from face to face. This is particularly visible in figure 31, where the IL for the top face is much higher (reaching levels of 30dB) than that for the bottom face (in the order of a few dB maximum) both for simulation and for testing.

Figure 29 – Comparison between measured and simulated source-averaged face-ATFs for engine mock-up front face (figure on the left) and rear face (figure on the right).

29

Simulation of exterior powertrain ATFs on an engine bay mock-up with complex …

Figure 30 – Comparison between measured and simulated source-averaged face-ATFs for engine mock-up right face (figure on the left) and left face (figure on the right).

Figure 31– Comparison between measured and simulated source-averaged face-ATFs for engine mock-up top face (figure on the left) and bottom face (figure on the right).

Eventually, figure 32 shows the comparison between the measured and the simulated source-averaged engine ATF. This is the quantity that is normally used to assess the impact of a certain engine bay package. The correlation is rather good, generally within 11.5dB.

30

Simulation of exterior powertrain ATFs on an engine bay mock-up with complex …

Figure 32– Comparison between measured and simulated source-averaged engine ATFs

Analysis of simulation results obtained with rigidskeleton formulation for porous materials Simulation results reported in previous section were obtained using, for all poroelastic materials involved, finite elements based on the classical u-p formulation of Biot equations [5]-[7]. Given the characteristics and the application conditions of the acoustic trim parts involved (either suspended or applied on a rigid wall, but always single layer and relatively “open”), it is very likely that at least in the mid-high frequencies the structural behaviour of the porous materials’ solid phase does not play a substantial role in their vibro-acoustic behaviour. For this reason, it was decided to investigate the possibility of using, for the simulation of all poroelastic materials, finite elements based on a rigidskeleton formulation of Biot equations [8]. Such formulation, while being in principle simplified (the vibrational behaviour of porous materials’ skeleton is simply ignored), allows substantial computational savings since it can reduce poroelastic materials to ‘equivalent fluids’ with suitable frequency dependent and complex sound propagation characteristics (density and speed of sound). In practical terms, this means that for each node belonging to the acoustic trim parts, the number of degrees-of-freedom is reduced from 4 to 1. As already reported in a previous section, the computational saving obtained in this way is very substantial given that calculation time is reduced from 74hours to 33hours.

31

Simulation of exterior powertrain ATFs on an engine bay mock-up with complex … Statistical analyses similar to those reported in previous section for what concerns the correlation between calculated and measured ATFs were then repeated on the results obtained from simulations with rigid-skeleton porous materials. Figure 33 compares the 70% confidence limit for the correlation between measured and simulated single-point ATFs resulting from the 2 different FE formulations for the simulation of porous materials. Results at face-average level and at engine-average level are quite similar. As one can see from figure 33, the use of a rigid-skeleton model for porous materials does not seem to involve –statistically speaking- any loss in the level of correlation with test data. Even, on a good part of the frequency range the rigid-skeleton simulations seem to guarantee –always on average- a slightly better correlation. Also for what concerns IL, the use of a rigid-skeleton model for porous materials does not seem to lead to any loss of correlation with test data. As a representative example, figure 34 shows the comparison between the measured IL for the source-averaged ATF relative to the engine left face and the simulated one obtained both by using the upformulation for porous materials and by using the rigid-skeleton formulation for porous materials. As one can see, while it is true that in the high frequencies the two simulated ILs assume different values, both correlate in a more than acceptable way with the curve coming from testing. One general tendency that has been observed is that IL curves obtained by means of rigid-skeleton porous materials tend to be lower than those obtained by means of standard up-poroelastic materials.

Figure 33 – 70% confidence limit for discrepancy between measured and simulated single-point ATFs.

32

Simulation of exterior powertrain ATFs on an engine bay mock-up with complex …

Figure 34 – Comparison between measured and simulated source-averaged ATF for engine left face

Conclusions The objective of this paper consisted in assessing the capabilities of the APML method in relation to the simulation of exterior powertrain ATFs, using as a meaningful test case a simplified but not trivial engine-bay mock-up. Acoustic treatments could be applied in the engine-bay mock-up and special importance was given to the possibility of simulating, by means of the APML method, the impact that such acoustic treatments have on ATFs. While the geometrical shape of the acoustic treatments was relatively simple (flat, constant thickness), the materials used and the application conditions were substantially more complex than those already used in the past on similar activities [1]-[2]. Given the important amount of data available (12 source positions, 21 microphones), it was possible (and it turned out to be important) to analyse the correlation between test and simulation data in a statistical way and at different levels. Results indicate that: ●

●

For what concerns single-point ATFs, the average level of correlation that one can expect for a case without treatment is generally decreasing from 3.5dB at low frequency to 2.5dB at high frequency, while for a case with treatment it is generally decreasing from 4.5dB at low frequency to 3.5dB at high frequency; The level of correlation between measured and simulated ATFs tends to improve when face-average ATFs and engine-average ATFs are considered. For engineaverage ATFs, the level of correlation obtained is generally around 2-2.5dB at low frequency and 1.5-2dB at high frequency.

33

Simulation of exterior powertrain ATFs on an engine bay mock-up with complex … Again, it is worth stressing here that the conclusions mentioned here above have a purely statistical value. On single-point ATFs and/or on single frequency bands, the discrepancy between measured and simulated ATFs can reach values much higher than those mentioned here above (even 10-12dB). To be confident to get a good correlation level, it appears necessary to consider average ATFs. In a way, this goes in parallel with the fact that even at purely testing level to get a stable a repeatable ATF curve it is necessary to average over many points. The use of rigid-skeleton porous finite elements for modelling acoustic trim parts does not seem to lead to any loss of correlation with data coming from testing and allows a very substantial cut of calculation times. This fact, while being interesting, should not be generalized. There can be cases (e.g. when spring-mass systems are used) where the use of full poroelastic finite elements turns out to be necessary. Nevertheless, the possibility to use rigid-skeleton porous finite elements (or, more in general, simplified formulations that allow reducing porous materials to equivalent fluids) should be always considered as a possibility and assessed –maybe through some initial simplified simulations- before starting long and complex analyses that involve fully poroelastic materials. Eventually, simulated IL curves correlate in an acceptable (though not exceptional) way with measured ones. Typically, IL can be simulated within 2dB. Also for the simulation of IL curves, no substantial loss of correlation with test data is observed by switching from a u-p formulation to a rigid-skeleton formulation of porous finite elements. In general, though, it was noticed that simulated IL curves obtained by means of a full u-p formulation tend somehow to overestimate the measured IL curves, while the contrary is true for simulated IL curves obtained by means of a rigid-skeleton formulation.

34

Simulation of exterior powertrain ATFs on an engine bay mock-up with complex …

References [1] – G. Miccoli, G. Parise, N. Totaro – Simulation of the effect of the absorbing treatments in the engine compartment on the acoustic transfer functions from the powertrain surface to exterior receivers, Proceedings of the 2011 Automotive Acoustics Conference, Zürich, 2011 [2] – A. Bihhadi, G. Miccoli, C. Bertolini – Simulation of exterior powertrain Acoustic Transfer Functions using IFEM and other deterministic simulation methods, Proceedings of the 2013 Automotive Acoustics Conference, Zürich, 2013 [3] – C. Bertolini, D. Caprioli, J. Horak, T. D. Petley, W. Jansen, B. Wiksteed – Acoustic benchmark of different CO2 encapsulation strategies: application examples and methods for the design of powertrain encapsulations, Proceedings of the 2013 Automotive Acoustics Conference, Zürich, 2013 [4] – ACTRAN 15.0 Users Guide Vol1. 2014 [5] – N. Atalla, R. Panneton, P. Debergue, A mixed displacement-pressure formulation for poroelastic materials, Journal of the Acoustical Society of America, 104(3), 1998 [6] – P. Debergue et al., Boundary conditions for the weak formulation of the mixed (u,p) poroelasticity problem, Journal of the Acoustical Society of America, 106(5), 1999 [7] – N. Atalla et al., Enhanced weak integral formulation for the mixed (u,p) poroelastic equations, Journal of the Acoustical Society of America, 109(6), 2001 [8] – J. F. Allard and N. Atalla – “Propagation of Sound in Porous Media”, Second Edition, John Wiley and Sons, 2009

35

Acoustic source detection for climate systems via computational fluid dynamics for improved cabin comfort Jan Biermann, BMW AG Barbara Neuhierl, Adrien Mann, Min-Suk Kim, Exa Corp.

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 W. Siebenpfeiffer (Hrsg.), Automotive Acoustics Conference 2015, Proceedings, https://doi.org/10.1007/978-3-658-27648-5_5

1

Acoustic source detection for climate systems via computational fluid dynamics for …

1 Introduction and Summary In order to address the increased relevance of vehicle cabin acoustics as a constituent of the overall passenger’s comfort in automobiles, precise and predictive methodologies approaches are required to identify and reduce the relevant noise sources. In the present case this problem is addressed to the HVAC (Heating, Ventilation, Air Conditioning) systems, including blowers, mixing unit, ducts or registers. The necessary prediction capabilities for such an aeroacoustics approach have been validated both against subsystems of HVAC units on testing rigs and full systems of to-bebuilt vehicles, see [1], [2] and example below. In close collaboration between BMW AG and EXA GmbH, acoustic comparisons for characteristic sub-units of an HVAC system were completed under operating conditions such as defrost, fresh air ventilation or air recirculation mode, obtained by alteration of the flap positions. The variation of the blower rotation speed also played an important role in the resulting operating conditions in order to better understand the system behaviour. This paper presents an example of analysis of the flow-induced noise sources present in the HVAC system. The noise sources are detailed as contributions per component, corrected by the acoustic transfer function between the source and a defined target point or target space (driver’s ear – passenger’s ear). This unique type of analysis provides both an assessment of the performances of a given HVAC system and a ranking of the noise sources according to the perceived noise at the passenger’s ears, providing desired guidance for an efficient design and optimization process.

2 Motivation After considering advances in the reduction of car interior’s engine, wind and rolling noise, other sound emitting systems become more predominant. One of these systems is the HVAC- system, which is particularly crucial, since its noise is acting constantly on the passenger. In addition, OEM’S are facing strong competition, shorter development cycles, challenging cost targets, and steadily increasing customer expectations on the acoustic comfort, especially in the premium segment. From a technical point of view, the HVAC system is distributed throughout a huge part of the vehicle, and its packaging interacts with many other systems. Changes in the hardware stage are difficult, costly and usually insufficient. It is therefore crucial to ensure a robust design in an early development stage. In order to meet all these requirements, computational methods are needed since they allow access to the functional and acoustical properties.

2

Acoustic source detection for climate systems via computational fluid dynamics for … Nowadays, incompressible CFD Methods based on finite volumes that solve the NavierStokes-Equations, such as LES or DES are widely spread. For the aeroacoustic assessment, a hybrid approach is often employed, applying a separate acoustic solver that solves the wave equation (Helmholtz equation) with source terms that are extracted from the CFD solution (based on aero-acoustic analogies). Taking into account the computational costs for this approach in combination with the restrictive time requirements, this usually leads to computations on subsystem or even component level. Several problems arise with this strategy though. One is the handling of the boundary conditions at the subsystem interfaces. Since the systems response, especially the prediction of aero-acoustic sources, is sensitive to the fluid mechanic quantities at the inlet of e.g. a duct, having just a subsystem at scope can lead to insufficient accuracy. A second problem is the consideration of the acoustic propagation in the duct system. When solely focusing on the aero-acoustic sources on a subsystem level and neglecting the representation of the acoustic propagation in the duct system as a whole, an integral and important part for the assessment of the overall acoustic performance remains unconsidered. All this together asks for a transient computation of the HVAC-System as a whole or even in an overall-vehicle context. This in turn is demanding for computationally very efficient methodology. Even if all the aforementioned requirements were met, one important step for an efficient problem solution would still be missing. The desired methodology is not only supposed to identify acoustic sources and describe the distribution of the acoustic energy properly, but it should be capable of quantifying the contribution of each source region with respect to the passenger’s ears. Only that way a meaningful successive elimination of weak spots can be conducted and hence, the system optimized. In the following a transient, compressible method will be proposed that is capable of delivering all the aforementioned and desired features, functionalities and performance measures.

3 The Lattice Boltzmann Method for Computational Aeroacoustics The simulation results presented in this paper were performed with PowerFlow (EXA Corporation), a CFD (Computational Fluid Dynamics) software based on the LatticeBoltzmann method (see [3], [4]). The fluid flow is represented by a discrete kinetic equation, the so-called Lattice-Boltzmann equation. It is describing the dynamics of particle distribution which are moving over time on a regular Cartesian lattice. The continuous velocity contribution of particles in reality is replaced by a set of discrete particle velocities in the simulation. During each elementary time interval, these particle distributions move from one voxel to a neighboring voxel, according to their velocity, where

3

Acoustic source detection for climate systems via computational fluid dynamics for … they collide with other particle distributions or solid structures. The so-called collision operator is modelling in detail the particle interactions, defining a meaningful physical behavior, thus approximating the Navier-Stokes equations. The evolution of the particle distribution movement over time results in inherently transient flow behavior. The boundary geometry is resolved exactly, enabling detailed modeling of details. Due to the methods inherently transient and compressible behavior, fluid flow and acoustics are represented simultaneously. So, there is no need to apply analogies of any kind in order to be able to predict wave propagation. Details are described e.g. in [5]. In order to capture small scale fluid structures, all available geometry details are incorporated in the simulation model providing an accurate representation of the surfaces of the HVAC systems In addition, boundary conditions such as induced temperatures, heat transfers, absorption and impedance can be taken into account, providing a complete and exhaustive representation of the real system. For simulations containing arbitrary shaped rotating geometry, like the blower in an HVAC system, a rotating reference frame is defined within the flow domain, covering all the parts rotating around a fixed axis [6]. The Lattice-Boltzmann method hence simulates very efficiently transient, compressible fluid flows, meeting the requirement of representing both the flow and the flowgenerated acoustic field propagating within the fluid. Interactions that may occur between flow structures and acoustic field are represented inherently. Its efficiency and scalability as well as the fully automated grid generation enable a short turnaround time, making the method suitable for productive usage. In addition, the inherent low dissipation of the Lattice-Boltzmann method is of crucial importance for the accurate resolution of the acoustic field. The noise sources are flow induced, and are thus directly captured implicitly by the simulation. For a better understanding, and to help derive measures for the improvement of systems, the acoustic sources generated by the fluid flow can be determined and visualized (patent pending).

4 A Validation study example: HVAC Component The final goal is to determine the aeroacoustic performance of an HVAC-system on an overall-verhicle level, meaning the acoustic effect of a functionally driven mass flow. For the validation of this entire chain it was chosen to stay on a component level since the boundary conditions, such as the acoustic properties of the receiving cavity, are easier to control.

4

Acoustic source detection for climate systems via computational fluid dynamics for … Below the results of a validation study based on a single HVAC duct is documented. The HVAC duct was connected to an inlet supplying a fixed mass flow (see figure 1) and was placed in an anechoic environment. Microphone signals were taken at defined locations relative to the outlet register. The simulation model covered the situation accordingly. Figure 2 shows the comparison of simulated (black curve) and experimentally measured (red curve) acoustic pressure in a third-octave plot. The overall magnitude as well as the spectral property shows good correlation.

Figure 1: HVAC duct, connected to inlet, in anechoic test environment

Figure 2: Comparison test – experiment

The conclusion that can be derived from this analysis is solely that the model is capable of representing aeroacoustic noise mechanisms and the transmission of the acoustic energy correctly. The relevance for the product development is limited so that the transition to the full-system consideration is required in the next step

5 Simulation of full HVAC Systems The acoustic performance of the overall HVAC-system is to a large extent driven by the integration of the single components in the full vehicle and their interaction. That is why the model has to represent the integration environment as realistically as possible (e.g. see [7] for the effect of blower integration). Here, the system boundaries have been extended from fresh air inlet via fan and porous media filters to manifolds and registers to the complete vehicle cabin. In this model, no mass flow has been imposed but is driven by the rotating centrifugal fan.

5

Acoustic source detection for climate systems via computational fluid dynamics for …

Figure 3: Generic Vehicle with generic HVAC, flow entering the cabin

Figure 4: Transparent model of real cabin with real HVAC system (including blower, ducts, registers….)

As the intake conditions are influenced by the mounting conditions of the HVAC system, the simulation domain was extended towards the assemblies located in the vicinity of the fresh air intake region, typically adjacent to the engine bay close to the cowl (see figures 5, 6).

Figure 5: Detail: intake area, with battery (green outline) sitting in front of the intake grid

Figure 6: Detail: Cutting plane through blower and intake box, including intake grind

Downstream, the simulation regime covers the complete passenger cabin including seats and interior components such as seats, steering wheel etc. The flow can leave the cabin through openings at the rear end. The complete model is placed inside an anechoic environment, where radiated sound is absorbed at the walls.

6

Acoustic source detection for climate systems via computational fluid dynamics for … Figure 7 shows in principle the interior modelling, using the example of a generic HVAC system together with an idealized cabin interior. At the location of the driver’s ears, microphone points have been located where pressure signals are recorded. It has to be mentioned that in the current stage no acoustic properties of the cabin have been considered other than the geometry of the cavity.

Figure 7: Example of generic vehicle cabin, with visualized acoustic sources in the HVAC duct; acoustic power (left), clustered acoustic sources (right)

6 Results For the study presented here, two blower speeds were simulated: Blower speed “high” is a typical value for the full cooling mode when the air inside the cabin needs to be cooled down. Blower speed “low” is smaller, a typical value for the steady state mode where the temperature is to be kept constant in the cabin. The choice has been made on the one hand for the sake of having a sanity check by comparing the computed delta values of the sound pressure from experience based on measurements, and on the other hand to get insight into the different system behavior. The first step is to assess the system by its fluid dynamic behaviour. Exemplary, the flow through a BWM sedan HVAC system, and the vehicle cabin is shown in figure 8 below. The blower is rotating with high speed.

7

Acoustic source detection for climate systems via computational fluid dynamics for …

Figure 8: Flow Results: Vortex Structures

The visualization demonstrates the presence of highly turbulent areas inside the HVAC system and HVAC ducts, next to locations with hardly any vortex structures generated. Within the turbulent areas, noise sources are generated by the flow. The radiated Acoustic Power generated by these sources is shown below (figure 9): Areas with high density of acoustic sources are shown in red. These are the regions that need to be optimized in order to make the system overall quieter. As mentioned before, the noise sources are obtained via a specific post-processing method on the fluid flow results (patent-pending).

Figure 9: Radiated Acoustic Power

8

Acoustic source detection for climate systems via computational fluid dynamics for … In order to simplify the acoustic analysis, the acoustic sources are shown clustered below. Clustering simply groups high Acoustic Power measurement volume cells spatially close to each other, and consequently identifies regions of space where strong noise sources are particularly present. (see figure 10 and 11 below).

Figure 10: Low blower speed

Figure 11: High blower speed

By the evaluation of the overall acoustic power radiated per cluster, which is depicted in figure 12 and 13 for the integrated levels and in figure 14 and 15 for the spectral characteristic, a ranking can be introduced. This ranking is indicated by the numbering in figures 10 and 11.

Figure 12: Integrated Clustered Acoustic Power, Low blower speed

Figure 13: Integrated Clustered Acoustic Power, High blower speed

Obviously, and as expected, most acoustic sources in both examined cases – low and high blower speed – occur in the blower region. Significant acoustic sources can also be found within the center ducts, before the flow leaves the system through the registers into the cabin, whereas the registers themselves and also the ducts leading to the side openings are comparatively quiet.

9

Acoustic source detection for climate systems via computational fluid dynamics for …

Figure 14: Radiated Acoustic Power over frequency, Low Blower Speed

Figure 15: Radiated Acoustic Power over frequency, High Blower Speed

Comparing the two load cases it can be seen, that the ranking of the different source regions within the ducting changes, but both times the blower region is by far the dominant one. This can give a distorted picture because the location of the source regions and thus the transfer behaviour is not taken into account. To do so is of major importance when it comes to the actual target value for the system optimization, namely the sound pressure perceived by the passenger. Thus the relevance of a source region for acoustic result at the ear position has to be determined. Therefore, the acoustic transfer functions between the respective acoustic sources and the target position, here the drivers left ear, have to be computed in an intermediate step. These are strongly dependent on the geometric boundary condition as well as the presence of sound absorbing materials along the transfer path. The result of this step is illustrated in figures 16 and 17. It shows that the transfer functions differ significantly.

10

Acoustic source detection for climate systems via computational fluid dynamics for …

Figure 16: Low blower speed, calculated acoustic transfer function

Figure 17: High blower speed, calculated acoustic transfer function.

By applying these transfer functions to the sound power of the respective source cluster, the relative contributions to the sound at the target position can be identified. The result from this procedure is depicted in figures 18 and 19.

11

Acoustic source detection for climate systems via computational fluid dynamics for … This result gives some interesting insight: In the case of low blower speed, the blower noise is indeed the predominant noise source (figure 14), and also clearly the highest contributor to the sound observed at the target location (figure 18). For higher blower speeds though, the behaviour of the system changes and the ranking of the noise sources differs over frequency (figure 19). Looking at the noise sources only, also here the blower region would be identified as the area to be targeted for aeroacoustic optimization. The success would be very limited, because here the center ducts contribute almost to the same extent for frequencies from 800 Hz to 2 kHz, especially in the 1,25 kHz octave band. No global improvement of the system would be achieved, as this frequency range is dominating the overall sound pressure level and also is of major importance for the perceived quality of the sound. The aforementioned procedure, taking into account the transfer behaviour together with the noise sources, thus allows to target the right location for system improvement.

Figure 18: Low blower speed, Output (Input X Transfer Function)

Figure 19: High blower speed, Output (Input X Transfer Function)

12

Acoustic source detection for climate systems via computational fluid dynamics for …

7 Conclusions and Outlook In this paper, a holistic approach for aeroacoustic evaluation of complete HVAC systems has been motivated by pointing out that the analysis of flow through standalone subcomponents like e.g. HVAC ducts can only give fragmentary information on the overall acoustic performance of such system. The procedure described here is based on the Lattice-Boltzmann method which is able to accurately predict the transient and compressible the flow through the system, also representing the generation and propagation of sound. The sophisticated postprocessing method introduced can give enhanced insight by localizing and quantifying the generated acoustic power determined from fluid variables and enables a more complex view on noise reduction in a system. The spatial localization of noise sources in conjunction with the consideration of the transmission behaviour of sound from the source to the passenger’s ear location are the key ingredient for an effective system improvement. This is because taking into account the sound propagation behaviour has a non-negligible effect and making a decision on a sub-system to be improved solely based on the sound generation behaviour of this subsystem would be misleading. This has been shown by a contribution analysis of the critical subsystems that have been identified in an initial step. It was also shown that the ranking of the subsystem contribution changes depending of the operating mode, such as different blower speeds. Hence, a tool has been presented to properly address the parts of the overall system that are most critical for sound perceived by the customer and herewith allows for an efficient optimization process in the earliest stages of the vehicle development. Future work will focus on the definition of an optimization strategy, the implementation of this strategy into an automated process and finally the validation of this process.

13

Acoustic source detection for climate systems via computational fluid dynamics for …

Bibliography 1. Pérot, Meskine, Le Goff, Vidal, Gille, Vergne, Dupuy, “Flow-induced noise predictions of complete HVAC systems using LBM HVAC simulation”, AIAA 2013-2265, 2013 2. Pérot, Meskine, Ocker, “Direct flow-induced noise prediction of a simplified HVAC duct using a Lattice Boltzmann Method”, AIAA 2013-2265, 2014 3. H. Chen, S.Chen, Matthaeus, "Recovery of the Navier-Stokes equations through a lattice gas Boltzmann equation method", Physical Review A, vol.45, 5339, 1992 4. Chen, Doolen, “Lattice Boltzmann Method for Fluid Flows”, Ann. Rev.Fluid n. Mech., Vol. 30: p. 329-364, 1998. 5. Brès, Pérot, Freed, “Properties of the Lattice-Boltzmann Method for Acoustics” AIAA Paper 2009-3395, 15th AIAA/CEAS Aeroacoustics Conference, 2009 6. Pérot, Kim, Moreau, Henner, Neal, “Direct Aeroacoustics prediction of a low speed axial fan” AIAA 2010-3887, 2010 7. Neuhierl, Felföldi , “Computational Analysis of Noise Generation and Propagation Mechanisms Using the Example of an HVAC Blower”, Fan2015, fan2015-88, 2015

14

Vehicle validation of the structure-borne noise of a lightweight body and trim design solution obtained with new integrated FE optimization J.W. Yoo2,T. Courtois1, J. Horak1, F. Ronzio1, S.-W. Lee2 1

Autoneum Management AG, Schlosstalstrasse 43, 8406 Winterthur, Switzerland

2

NVH CAE Team, Hyundai Motor Group, Korea

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 W. Siebenpfeiffer (Hrsg.), Automotive Acoustics Conference 2015, Proceedings, https://doi.org/10.1007/978-3-658-27648-5_6

1

Vehicle validation of the structure-borne noise of a lightweight body and trim …

Abstract In the context of the development of lightweight and vehicle-NVH-refined sound package solutions, car manufacturers have identified that it is essential to promote new dedicated CAE design methods which can be applied at a very early stage, before body and trim designs are frozen. In the case of body NVH design, simulation methods based on FE modelling are increasingly used in order to influence the weight reduction of damping products by finding their optimal layout on the body structure, while meeting panel vibration targets. Due to the availability of validated FE models even before the shape and stiffness of the different panels of the vehicle body are fixed, it is also becoming practically possible to perform optimizations of the damping pads simultaneously with smart panel local stiffening, thus offering further weight savings in the low and medium frequency range [1]. Some efforts are also spent in this context, to quantify the design performance with respect to panel radiated noise [2]. On the other hand in the case of trim NVH design, the parts are traditionally optimized with respect to airborne noise transmission in the medium and high frequency range, without systematically considering their side influence on the panel vibration. However, with the growing importance of rolling noise contribution to vehicle NVH, and with the introduction of light weight trim parts on body structures, it is shown that the vibroacoustic interaction of body structure and soft poro-elastic trim has increasing importance at medium frequencies, where structure borne noise propagation is dominant in vehicles. In this case, it becomes even more important to assess the design of the body NVH not only at the panel vibration level, but more at the panel radiated noise level. This paper is bringing one major advancement in this CAE problematic, with the release and application at vehicle level of a methodology based on FE optimization for the design of the body NVH including damping and local stiffeners, while taking into account the influence of the trim, to be assessed on the SPL performance improvement, thus offering additionally the integration of the trim in a fully integrated optimization. First of all, the FE representation of the trim can be included into the vehicle FE models that are traditionally used for structural optimization, so that the body vibration target can be substituted by a more realistic interior SPL target. Then the concept is extended to the inclusion of the trim variables (like poro-elastic materials) into the optimization variables, so that the body and trim designs can be accomplished simultaneously, taking into account the mutual influences. The technical aspects of this “integrated” simulation-based optimization have already been presented for a simple validation case [3]. In the present paper, the method is applied on an existing vehicle body in white, demonstrating the interactive influence of damping, stiffening and trim on the structure borne

2

Vehicle validation of the structure-borne noise of a lightweight body and trim … noise of a vehicle. The demonstration is completed by the execution of a simultaneous optimization, leading to an increased weight advantage, and by an interesting outlook towards future design strategies. Finally, the different intermediate optimization results, including the effect of both body and trim design changes, have been validated against test results on two different BIW and trim variants of the vehicle under consideration, using an experimental SPL contribution analysis based on the utilization of PU probes.

1 Introduction In the context of body NVH design, simulation methods based on Finite Elements (FE) are traditionally used to improve panel vibration level. Different approaches may be applied for setting the targets, for example first mode higher than a certain frequency, vibration level lower than a fixed value. Also the definition of panel can vary from OEM to OEM. The region of interest can be a full component (e.g. floor, dash) or a more limited region (e.g. the front floor between the reinforcement). Optimization is usually exploited to reduce the weight of damping coatings by finding their optimal distribution on the body structure. Whilst many OEMs traditionally fix the shape (and stiffness) of the body in the early phase of pre-development projects and then design the damping pads layout, it has been showed that a simultaneous optimization of both damping and shape may lead to lighter and more effective design [1]. In the latest years many efforts have been dedicated to assess vehicle body NVH performance with respect to panel radiated noise and to study how structural modifications can impact on the development of lightweight vehicle NVH-refined sound package solutions. While so far the design of trim parts, even lightweight solutions, has been mainly driven by air-borne considerations (e.g. power contribution analysis), nowadays there is high interest in the quantification of the impact that lightweight trim parts may have on the low frequency performance in terms of interior Sound Pressure Level (SPL). From a methodological point of view, this trend translates into the need to support the design process traditionally based on high frequency methodology (Statistical Energy Analysis or Transfer Matrix Method), with FE simulations able to cover the lowmedium frequency region (100-800 Hz). New dedicated CAE design methods must be developed and validated so that they can support engineering design activities at a very early stage, before body and trim designs are frozen. This paper investigates further the application of a new dedicated design tool which aims at enhancing structure-borne performance of vehicle by simultaneously optimising damping package, body stiffness and acoustic trim with regards to the internal cavity SPL in presence of a structural excitation [3].

3

Vehicle validation of the structure-borne noise of a lightweight body and trim … In this paper a methodology for the efficient design of panel stiffness, damping layout and lightweight technology treatment is presented. The proposed method is based on the combined use of porous material FE simulation techniques, i.e. Treasuri2/FE, and genetic algorithms. The simultaneous design of all these variables represents a big challenge given the large number of design parameters. So Autoneum proposes a unique approach that is able to identify the shape/damping/trim configuration with the best performance to weight ratio. The performance is measured in terms of sound pressure level of one or more control nodes (e.g. microphone at driver head position) while a weight constraint may be provided. The methodology has been exploited here for the optimization of a floor component belonging to a real vehicle test-case. The validation of the results has been made possible thanks to the cooperation and support from Hyundai Motor Group (HMC), which has provided Autoneum Vehicle Testing Team with two prototyped body in white (BIW): the reference and the optimized one. The paper is organized as follows. Theoretical aspects and the potential of the integrated optimization methodology are summarized in paragraph 2. In paragraph 3, the setup of the optimization strategy for a real vehicle test-case is summarized, followed by the description of the measurement setup used to validate the optimization results. In paragraph 4 the outcomes of the optimization and the measurement validation are shown. Further applications are presented in the 5th paragraph.

2 Development of the methodology 2.1 Poroelastic material FE simulation methodology Due to their dissipative properties, porous materials (felts, foams, etc.) are extensively used in automotive acoustics for passive noise control. The possibility to efficiently include such materials in FE simulation has been a real challenge for the technical community. Nowadays, many numerical porous material models are available. The majority of them is based on the system of equations initially developed by Biot [4][5]. The method used here implements a more recent formulation of the Biot equations [6][7], which uses the displacement of the solid phase and the acoustic pressure of the fluid phase as field variables with a significant reduction of degrees of freedom with respect to the original Biot formulation. As extensively explained in [8], by means of MSC.Nastran DMAP routines, it is possible to compute the matrices Hij describing the trim boundary impedance so that the effect of the trim on a structure/fluid system can be included by just using standard solutions (the standard SOL111 available in Nastran).

4

Vehicle validation of the structure-borne noise of a lightweight body and trim …

2.2 Genetic algorithms for vehicle NVH optimisation Optimisation algorithms are applied to find the maximum/minimum of a given objective function by linking a combination of the design variables with the target (objective) of the optimisation. Among optimisation technics, the need for robustness and efficiency in searching complex multidimensional spaces, makes the Genetic Algorithms (GA) technic very suitable for the problem to be analysed here. GA [9] are iterative and operate on a set of potential solutions called population, and utilize concepts such as selection (survival of the fittest), crossover (information exchange), and mutation (adding the genetic diversity). Principles of evolution and heredity are used for the optimization of an objective function (OF). The core process of the genetic code is represented by the following steps, which are repeated for each population. ● ● ● ●

Evaluation of the objective function for each individual (one combination of variables) of the population; Generation of the next population accordingly to the Darwin’s evolution scheme by means of Crossover/Permutation/Mutation functions; Evaluation of the objective function for each individual of the new population; Comparison of the best individual's objective function with the one of the previous population. If the best individual is constant through a relevant number of populations then the algorithm has converged.

Usually the first generation is simply defined by a random selection of individuals. In the cases under study here, the objective function is directly linked to the performance of each individual solution in terms of SPL or panel mobility vibration and to the weight of the damping package plus the trim. Every individual solution is described by a genetic code in which all the information regarding damping variables (damping pads location, weight, size, and material), shape variables (ribs, soap-film, etc.), and trim variables (area coverage, different bills of material) is represented. The entire variable space has to be described in such a way that is compatible with a GA structure. The individuals of a genetic search are encoded as strings (chromosomes). The genotype will constitute the set of possible value that each variable (gene) can assume. The phenotype will be a unique map of genes. The most commonly used representation in GA’s is the binary alphabet (0,1). For example a problem with 5 variables can be mapped as: U = 1010 | 1001 | 1101 | 0010 | 0010 where for each variable 4 bits are used to describe the 16 possible configurations. The minimum value of the function U for each gene will be 0, whilst the maximum will be 2^0+2^1+2^2+2^3=15.

5

Vehicle validation of the structure-borne noise of a lightweight body and trim … An example is presented assuming that only damping is concerned by the optimization. Each gene will represent one damping pad location (4 pads in figure 1) and the number of bits to be used for a certain gene will be function of the number of thicknesses and different materials to be investigated (for 4 variants – 2 thickness per 2 materials – 2 bits are required).

Figure 1: Example of an eight digits individual

Integration of the trim in the multi-parametric optimization Traditionally, when only body structure variables are considered (damping and shape), any optimization or performance assessment is carried out by targeting panel vibration level. The possibility of including the trim as a further variable in the optimisation process is adding a lot of complexity to the problem and brings out the necessity of considering an acoustic target in place of a structural one [3,10]. It also affects the optimization strategy to be used, creating different possible application scenarios. The optimization of body stiffness and damping package is normally exploited during advanced pre-development projects, when the BIW design process is still on going. In this phase the technology to be used for the insulation parts may or may not be yet defined. For optimization of the trim parts, two different scenarios are to be expected. In the first scenario, weight reduction is the main goal: the reference is an existing shape of the body, defined damping package and existing heavy carpet, usually inherited from a previous vehicle. The target is the weight reduction of the full package (e.g. 15%), by keeping the same performance. In the second application scenario, the main target is performance improvement: in case of the introduction of a new carpet technology, the carpet design may need to be improved within the available design variables (for example recovery improvement in case of replacement to light weight). The reference is existing shape and damping combination with the new carpet definition (for example coming from airborne design).

6

Vehicle validation of the structure-borne noise of a lightweight body and trim …

2.3 Multi-physics optimization example on a simple test case The test case considered here is represented by the test-system shown in figure 2. In this case, the acoustic domain is made of a rectangular box-shaped cavity of 0.7m x 0.8m x 1.6m size, while a flat aluminium plate of 0.7m x 0.8m size and 1.25 mm thick constitutes the elastic domain (i.e. the plate is the only flexible wall of the box cavity, and all the other walls can be well assumed as rigid). The vibro-acoustic behavior of the system is rather complex because even if the system is conceptually simple, the plate and cavity size lead to a relatively high number of modes. The FE model counts 85700 degrees of freedom and 18057 elements.

Figure 2. Mock-up plate box experimental set-up and virtual model

In both testing and simulation conditions, the plate is excited (experimentally by means of an electro-dynamic shaker) at a point close to one of its corners. The SPL inside the cavity is recorded at 3 different points. The average SPL represents the target performance for the optimisation. In order to consider a significant optimisation variable space, the following set-up is defined: the total number of combinations explored is 2^48 (respectively 2^18 for the damping, 2^12 for the shape, and 2^18 for the trim). For the damping package optimisation, 9 different possible design areas are identified (fig. 3), each of them with 4 different possible alternatives: no damping or coated with

7

Vehicle validation of the structure-borne noise of a lightweight body and trim … three different materials, Autoneum lightweight damping (2.5 kg/m2), standard bitumen (3.6 kg/m2 or 4.2 kg/m2).

Figure 3. Damping and shape design area and shape variables.

For the structural shape optimisation, 4 different possible design areas are identified, each with 8 different variants (singular, double, diagonal, with different orientations) of a basic beading shape (250x50 mm with a height of 5 mm). The beading shape has been chosen for prototyping reasons, for the sake of the correlation work [3]. For the trim optimisation, 9 different possible design areas are identified (the same as for the damping variables of figure 2). For each of them, 4 different possible trim alternatives are investigated.

Figure 4. Trim design areas and variables

In addition to a foam-heavy layer trim, the effect of a possible technology change is considered, while keeping the packaging space constant: lightweight foam-felt configurations are set as a possible alternative. This choice is motivated by its wide range of variability of air-borne TL and absorption, which are the main driver for the high frequency design of the trim. Nevertheless they are also extremely suitable for a structureborne optimization problem, since they are characterized by a different transmissibility and wide weight variability. A first optimisation is performed by investigating only damping and shape variables with the main purpose of understanding the possible convenience of introducing an acoustic target (SPL) instead of a more traditional structural target (r.m.s. average velocity of the plate).

8

Vehicle validation of the structure-borne noise of a lightweight body and trim … As well-documented in [3], the main outcome of this optimization is that both optimizations provide very satisfying results in terms of performance, the use of an acoustic target leading to lighter damping package solution (fig. 5).

Figure 5. Optimum for damping and shape optimisation

The introduction of the trim in the set of variables provides further interesting results. Not only is reinforced the statement that an acoustic target is more effective than a structural target in terms of weight reduction, but when comparing the optimum performance with the non-designed solution, improvements reach up to 10-15 dB in the full frequency range with a weight increase of only 4%. By comparing the optimum with the non-designed solution of trim and damping, the optimum performance reaches improvements up to 5-8 dB up to 550 Hz with a weight decrease of 22% (fig. 6). The full range of results is available in [3] together with the experimental validation of the most relevant cases.

Figure 6. Optimum for damping, shape, and trim optimisation

9

Vehicle validation of the structure-borne noise of a lightweight body and trim …

3 Application to a vehicle test-case The application of this CAE methodology to a real vehicle test-case for early phase vehicle development has been carried out in cooperation with Hyundai Motor Corporation. Autoneum has validated and applied the new tool for the optimization of the floor components of a C-segment vehicle, including body shape, damping design and acoustic treatment, with respect to the interior SPL. For the testing, HMC has produced the floor panel as resulted from the optimization and mounted it on an existing BIW.

Figure 7. BIWs prototype – (a) Reference (b) Optimized

The excitation is of structure borne noise type, as it is the most dominant contributor at low-medium frequency. The frequency range of interest is from 100 Hz to 1000 Hz. The numerical optimization has been performed up to 600 Hz, even though the SPL will be evaluated up to 800 Hz and the testing validation will be performed up to 1000 Hz.

3.1 Description of the reference BIW and set up of the optimization The vehicle test case is a C-segment vehicle. The related FE models of the structure counts more than one million nodes for more than eight million DOFs The FE model of the acoustic internal cavity counts more than twenty thousand nodes (fig. 8).

10

Vehicle validation of the structure-borne noise of a lightweight body and trim …

Figure 8. FE model C-segment vehicle

The area selected for the optimization is the front floor including the heel-kick region. During the full optimization, the fluid-structural coupling between the BIW and the cavity is limited to the area to be optimized. The goal is to highlight the contribution from this area to the NVH and to avoid pollution originated from other areas (dash, trunk, etc.) which for certain frequencies may or may not overcome the contribution of the region of interest. The risk here-hence is to optimize a response which, as matter of fact, is not the most relevant one: a careful analysis of the contribution levels helps to avoid this issue. Moreover, the decision to limit the coupling area to the design region offers the possibility to use the super-element reduction technique [1]. The acoustic domain is not included in the super-element. The reduction technique is providing a reliable representation of the full vehicle behaviour, both in terms of structural response and acoustic response (fig. 9).

Figure 9. Super-Element Reduction Technique

A fundamental step for the definition of the correct optimization condition is the definition of the load conditions, which means the identification of the most relevant input forces. To obtain the necessary ranking different approaches are possible (e.g. Transfer Path Analysis). Through the analysis of the mobility response of the area to be opti-

11

Vehicle validation of the structure-borne noise of a lightweight body and trim … mized by means of SILVER (Autoneum CAE tool based on a DMAP modification of a standard Nastran Solution, the SOL111), more than thirty unit load-cases corresponding to the engine mounts and to the suspension systems are examined and only 3 are selected for the optimization (fig. 10). The input-force points are localized at the front mounting of the front axle lower both left and right side (force in Z direction) and the attachment point of the right engine mount (force in Y direction).

Figure 10. a) Area of the optimization, colour-map of the vibrational level b) Excitation point

3.2 Variables description Damping and Shape Variable Setup The optimization area has been divided in different design regions. While for the damping each pad may be considered as a variable, for the shape different designed regions have been identified. Up to 55 areas correspond to the possible locations for the coating (fig. 11). For each of them, only one material choice is possible. A conventional bitumen coating with average performance and average weight has been used, the maximum possible weight of the damping package being 3.57 kg. In all the following optimizations, this is considered to be the weight of the reference damping configuration.

12

Vehicle validation of the structure-borne noise of a lightweight body and trim …

Figure 11. Damping and Shape Variables

For each of the 16 design variables (fig. 11), 8 different configurations are possible. A wide range of beadings and bubble soap skin combinations is investigated. The optimization space counts 2^55 x 2^48 possible solutions. The baseline treatment is a needle punch carpet consisting by a low weight backing latex and PE. PET fibre blanks are considered for the decoupler. The overall packaging space varies from 5 to 60 mm.

Figure 12. 3D FE model of the carpet divided in 10 areas

Up to10 trim design areas have been distinguished: front floor left/right, middle floor left/right, rear floor left/right, cross beam region left/right, rear tunnel centre area, heelkick region. The toe-board areas have been excluded from the optimization, since the dash trim has not been modelled and the coupling of the toe-board to the firewall can not be correctly represented in simulation. The total weight of the carpet simulated as such is 4.1 kg. The optimization strategy for the trim can focus on different aspects: ● ● ●

Replacement of the decoupler technology Absorption of the carpet Weight behind the carpet.

13

Vehicle validation of the structure-borne noise of a lightweight body and trim … For this test case, due to the constraint inherent to hand prototyping of the optimized carpet, only the first and third aspects have been considered. In an optimization prospective, felt technology is valued to be more suitable than a foam technology. In particular, the idea of investigating the use of an IFP (injected fibre process) solution allows to control the density of the fibres. That means that the first variable of the optimization will be the trade-off between decoupler density and its stiffness, compared to PET fibre blanks. The IFP bulk densities chosen for the optimization are 40 – 80 – 120 kg/m3. Using an IFP technology has also another advantage: it ensures the possibility to reach high level of coverage, even where the packaging space is not enough for injected foam and other technologies (cross beam and side-sill regions). Hence, a second variable for the optimization is the enlargement of the area coverage of the decoupler in the areas where a limited packaging space is available (maximum 5mm – fig. 14).

Figure 14. Example of trim variables

The third parameter to be investigated is the backing weight. The possibility to insert a light heavy layer (1650 gsm) behind the carpet will be explored. This is practicle because it can be easily prototyped, even though it may not be considered for a real design, where the goal would rather be driven by the selection of low weight solutions. Overall, the combination of these three parameters (decoupler technology, area coverage and backing weight) allows exploring a wide range of solution in terms of weight: for the front floor right side, for example, the heaviest configuration weights 1.36 kg, the lightest 0.54 kg (reference weight is 0.74 kg)

14

Vehicle validation of the structure-borne noise of a lightweight body and trim …

Target of the Optimization To complete the set-up of the optimization, targets have to be set. In the case of body local stiffness and damping pad optimization, 9 subcases, namely 3 microphone positions for 3 input positions have been used. For the trim optimization only 4 subcases are targeted because the spectra at the driver head position and passenger have been proved to be mainly sensitive to the same modification and therefore they can be averaged. The less contributing input point have been also excluded. Nevertheless, the final results of the optimization have been also compared by considering the 9 subcases, in order to examine the robustness of the optimization. The definition of the optimization target takes into account more aspects: ● ● ● ●

desirable improvement. reference individual status. rough evaluation of possible improvement (function of the variety in terms of shape and possible areas of modification for the variables). frequency range of main interest.

3.3 Validation Technique and measurement set up PU sensors application The aim of the testing is to demonstrate the SPL improvement reached by the optimization. Since the simulation represents transfer function p/F, where F is a structural point force, the testing could easily provide the same quantity. However, in simulation the coupling is restricted to the floor area, which is not the case when measuring p/F transfer functions. Thus there is a real challenge to reproduce in testing the condition of the simulation, that is to say the contribution to the Sound Pressure Level (p/F) of the floor only. The most traditional approach consists in using the masking technique, where the contribution of the floor only could be measured by artificially reducing the acoustic radiation originated by the rest of the vibrating structure with heavy acoustic treatment. This windowing method is very pragmatic and intuitive, but has some drawbacks. The presence of the maximum package can alter the dynamics of the structure represented by the vehicle body especially in the low frequency region since the weight of the maximum package can go up to 150kg. Secondly, the acoustics of the cavity represented by the passenger compartment can be modified because the top surface of the maximum package is often reflective and it has then an acoustic impedance that is substantially higher than that of the cavity walls in theory (rigid). These factors make the evaluation of the desired contribution necessarily approximate. A better solution could be a direct measurement of the SPL contribution by using the p-u probes techniques (“velocity-base reconstruction meth-

15

Vehicle validation of the structure-borne noise of a lightweight body and trim … od”). The contribution of a certain surface is calculated by the surface integral of the product of the panel velocity times the Acoustic Transfer Function (ATF) between the panel velocity itself and the acoustic pressure at the receiver microphone [11]: p op (P )   unop  p ATF  p op  unATF d 

Among the different devices available, the PU sensors proves to be extremely suitable because they provide a pretty accurate reconstruction of the SPL contribution, and they also provide a localized description of this contribution area by area. The possibility to acquire the position of each probes allows the reconstruction of the measurement geometry: this will be used to resolve the surface integral, but can be also used to create a mesh where the integral is calculated for each elements or for a group of them, directly quantifying the contribution area by area, or representing a distribution of contribution.

Figure 15. a) Colour-map of SPL contribution at 259 Hz (reference BIW) – b) Contribution chart averaged in band of 100 Hz

Such colour-maps of the SPL contributions, as shown in figure 15.a, highlight the most contributing area and could be exploited to show how the optimizer is impacting the contribution of each area at a certain frequency. Moreover, the reconstruction by area can provider a ranking among the different areas of contribution (fig. 15.b). To validate the results of an optimization the flexibility in post-processing the PU probes data is extremely helpful. In the end, the output of the testing campaign will be: ● ● ●

16

a vehicle SPL (measured without any masking) a floor contribution SPL localized SPL contribution information (colour-map or frequency band contribution ranking).

Vehicle validation of the structure-borne noise of a lightweight body and trim …

Prototyping of the carpet For the prototyping of optimized carpet, the same top layer of the reference carpet has been used to avoid discrepancies either in terms of weight or shape. Since the decoupler is an air-laid felt of different density at different areas, the prototype has been realized by hand: locally gluing additional heavy layer to the original carrier, using flat felt material samples and by cutting and gluing of felt die-cut pieces together (fig. 16).

Figure 16. Carpet Prototyping

Testing set up The exploitation of the PU sensors does not require the creation of an heavy max package. Nevertheless few precautions have to be taken. All the cavity openings have been closed by using either bigger patches of HL (9 kg/m^3) or smaller patches of damping (aluminum constrained layer, 3kg/m^2) to close to small holes on the floor body. The body has been suspended/isolated/decoupled by means of 4 air springs, near the jack support points. The rigid mode resonance frequency of the body sitting on 4 air springs was very low, only a few Hz (fig. 17).

Figure 17. Testing set up

17

Vehicle validation of the structure-borne noise of a lightweight body and trim … Two shaker excitation points have been used (fig. 17): ● ●

Right engine mount, Y-direction Left lower arm front mount, Z-direction

Force and Acceleration were measured at each input points. 19 Microflown PU-sensors have been used, each sensor being placed on the floor surface using a decoupling support. The acquisition chain used consists in 5 Response Microphones: driver head, external and internal ear (FL outer, FL inner), front passenger, external and internal ear (FR inner, FR outer), rear passenger center (Rear Center). A volume velocity source with a nozzle reference sensor (at 5 microphone positions) has been used. For these microphones positions the vehicle SPL (p/F) has been also directly measured. The area covered by the PU sensors has been defined in order to match as much as possible the coupling area used in the FE simulations (fig. 18). This leads to a large number of measurements to be executed: 472 PU positions for the damped case and 301 PU positions for the bare case.

Figure 18. PU sensors coverage

4 Results and Validation For the validation against testing the optimized body has been prototyped. To cope with the technical time necessary for the manufacturing, a first optimization involving only damping and panel shape has been performed, thus freezing as early as possible the BIW optimized design. Once the BIW has been frozen, the trim optimization was performed on top of the optimized body. In paragraph 5 the results of such a simultaneous optimization of damping, shape and trim will be shown.

4.1 Optimization of the BIW During the optimization of the BIW, the GA has investigated 1846 individuals, among possible combinations of damping pads and panel shapes, 26 generations have evolved

18

Vehicle validation of the structure-borne noise of a lightweight body and trim … in 20 days of computational time on an 8 cores 192 GB memory server. A weight constraint is used: at least 10% damping weight reduction has to be achieved. The results is an optimized body (fig. 19) better performing in terms of SPL and with a 20% weight reduction of the damping package. The performance improvement is clear: between 200Hz and 450Hz there is always improvement, while between 500-550Hz, the combination of shape and damping is deteriorating the performance for the driver microphone position, but is also improving the front passenger sound pressure level.

Figure 19. Optimized BIW

Between 450Hz and 500Hz, there is for the driver microphone position a deterioration of the performance, which was small during the optimization (~2dB) but it is highlighted when the full body is simulated instead of the reduced one with the super-element databases. For sake of clarity, hereafter only the spectra corresponding to one excitation point (the most contributing one) are shown (fig. 19). The other load-cases are showing similar trends and improvement levels.

Figure 20. Performance Improvement

It is of extreme interest also to evaluate how the best solution behaves above 600Hz (upper limit of the frequency range of the optimization) and how the optimization is affecting the radiation of the floor system.

19

Vehicle validation of the structure-borne noise of a lightweight body and trim … Test data (fig. 21) are highlighting that vehicle SPL and the SPL floor contribution are often showing the same behavior (in terms of general trends and levels). This is not surprising, since the floor is one of the most contributing areas of the body in this testing condition (no carpet and focused selection of the excitation positions). The most important outcome is that in both simulation and testing the optimum is better performing than the reference, even in the higher frequency range. More in details: ● ● ● ● ● ● ●

200 Hz: simulation improvement is corresponding to a slight improvement on testing vehicle SPL. 300 Hz: simulation improvement and improvement on testing at around 250 Hz 350 – 400 Hz: simulation improvement is translated in vehicle SPL improvement. 450 Hz: simulation deterioration becomes deterioration on testing SPL floor contribution but not on vehicle SPL. 500 Hz: simulation improvement is highlighted on testing vehicle SPL. 650 – 750 Hz: simulation improvements mirror high improvements (higher than 5dB) on testing vehicle SPL. Above 750 Hz, the optimum SPLs are equal or slightly worse than the reference one, until 950 Hz.

Figure 21. Validation damping and shape optimization

The color-maps of the SPL contribution provide additional interesting input. At 259 Hz, for example, there is a dramatic drop in the measured SPL. The optimization of shape and damping manages not only to decrease the overall contribution (max level), but also to modify the distribution of radiation areas. Avoiding extended radiating areas seems to be effective to decrease the overall SPL value.

20

Vehicle validation of the structure-borne noise of a lightweight body and trim …

Figure 22. Colour-map SPL contribution

4.2 Optimization of the trim To perform the optimization of the trim, the weight saved with the damping optimization (0.7 kg) has been used as possible weight increment for the trim. Hence, the weight of the trim can vary between 3 and 5 kg. The decision to increase the weight of the trim is clearly contra-intuitive from an optimization point of view, but it has been driven by the two following considerations: ● ●

in the considered optimization strategy, weight increase for each trim pad is possible through the insertion of light heavy layer patches behind the carpet. using the weight of the initial configuration (trim plus damping) as reference will make it possible to compare the results of this sequential optimization (damping and shape and successively trim) with the outcome of the integrated optimization (damping, shape and trim simultaneously optimized).

Figure 23. Performance Improvement

21

Vehicle validation of the structure-borne noise of a lightweight body and trim … During the trim optimization, the GA has investigated 90 individuals over 22 days of computational time. The result is an optimized trim (fig. 24) which shows a very different weight distribution in comparison to the reference one. The SPL performance have been improved slightly (2-3 dB) between 200 Hz and 300 Hz, and remarkably (higher than 5 dB) after 430 Hz (fig. 23). Similar results have been achieved for the other load-cases.

Weight [kg] HMC reference

Optimum

Percentage Variation

Trim

4.1

4.7

14%

Damping

3.6

2.9

-20%

Total

7.7

7.5

-2%

Figure 24. Optimized carpet

The weight distribution between the different trims is deeply changed by the optimizer. The front left and the cross beam left regions become heavier than the reference while on the front right side the selection of the lightest IFP as decoupler allows to locally save 35% of the weight. Overall, the optimized trim is 12% heavier than reference, but if one considers the total weight of trim and damping package, the reference configuration and the optimum have similar weight, but the latter has improved performance. Test results are always confirming that the optimum solution is better than the reference solution. They are sometimes showing improvements in slightly different frequency ranges than simulation results. In general the FE results are showing lower absolute level than testing results. The prediction of relative improvement is still valid. The deltas and more generally the trends characterizing different treatment configurations are always well predicted in simulation. The most important outcome is that in both simulation and testing the optimum is better performing than the reference, even in the higher frequency range. More in details (fig. 25): ●

22

250 Hz: even if there is no simulation improvement or deterioration, testing SPL floor contribution shows a 10dB improvement and more than 5dB improvement is evident on testing vehicle SPL.

Vehicle validation of the structure-borne noise of a lightweight body and trim … ● ● ● ●

350 Hz: simulation small deterioration corresponds to a deterioration lower than 5 dB on testing vehicle SPL and testing SPL floor contribution data. 450 – 550 Hz: simulation improvement is highlighted on testing SPL floor contribution data and also on testing vehicle SPL (improvement higher than 5 dB) 750 – 900 Hz: improvement higher than 3 dB on testing SPL floor contribution. 600 – 1000 Hz: on vehicle SPL, the optimum’s results are equal or slightly worse than the reference one.

Figure 25. Validation damping, shape and trim optimization

Overall, also in testing the optimum performance is better than the reference: the optimization has been proved to be successful. A detailed analysis of the SPL contribution may provide information about the effect on the different contributions of the floor area to the SPL. For this purpose, the pressure values have been averaged in frequency bands of 100 Hz (fig. 26). The relative contribution between the different area of trim is substantially changing.

Figure 26. Trim by trim percentage contribution to the SPL Exc. Front Axle Lower Left – Driver Microphone

In particular this is more evident if we consider the frequency range between 500 Hz and 600 Hz (fig. 27). In this frequency range, the lower SPL is a direct consequence of the lower contribution of all the different areas, especially of the beam region on the

23

Vehicle validation of the structure-borne noise of a lightweight body and trim … right and the front left part of the carpet. In the beam region on the right, the optimum solution is composed with heavy layer and extended decoupler. In the front left part of the carpet the optimizer has chosen to insert the light heavy layer patch and to use the decoupler with higher density/stiffness ratio.

Figure 27. Trim by trim percentage contribution to the SPL Exc. Front Axle Lower Left – Driver Microphone – 500-600 Hz

5 The “Integrated optimization” The next challenge is the simultaneous optimization of damping, panel shapes and trim. In the following paragraph two different optimizations are described. In the first one, the same set of variables previously defined is used so that those may be directly compared with the previous one. There is only one difference: in the integrated optimization a “symmetric approach” has been adopted for the shape and damping variables. This means that all the damping and shape configurations have been symmetrically rearranged. As a matter of fact, using a symmetric approach is penalizing the optimization since it is limiting the number of possible combinations. In paragraph 5.2, the optimization will follow a strategy more likely to be used in a real application case. The insertion of heavy layer patches behind the carpet has been ex-

24

Vehicle validation of the structure-borne noise of a lightweight body and trim … cluded from the range of possibilities. Also in this case a symmetric approach has been adopted for the damping and shape variables.

5.1 Integrated vs. Sequential Optimization During the integrated optimization, the GA has investigated 136 individuals over 32 days of computational time. The outcome of this optimization is of high interest, especially when compared to the results of the sequential optimization (where the trim was designed on top of an already optimized structure). The first consideration concerns the computational time. On the same workstation, whilst for the sequential optimization 42 days were necessary (22 for the shape and damping optimization and 20 for the trim), the integrated optimization required 32 days (25% less computational time).

Figure 28. Performance Improvement

In terms of performances (fig. 28), both sequential and integrated optimizations provide good results, but the latter is slightly better in the full frequency range. Furthermore, the integrated solution is 11% lighter than the sequential one. The weight trade-off between damping and trim is clearly shifted towards the trim, but still the damping pads constitute 13% of the available weight. No constraint regarding the weight split between the trim and the damping is included in the optimizer. In principle even a solution with no damping could have been selected as the best one. Instead the optimizer always tries to exploit the damping propriety of the coating, by searching the best trade-off between stiffness, damping and insulation. In few areas, for example the middle floor, the presence of more ribs and stiffening seems to be linked to the little amount of damping applied or vice versa. Moreover, when a trim with the heavy layer patch is selected, then the corresponding panel shape comes out with less beadings and reinforcements.

25

Vehicle validation of the structure-borne noise of a lightweight body and trim …

Weight [kg] HMC reference

Optimum Sequential

Optimum Integrated

Trim

4.1

4.7

5.6

Damping

3.6

2.9

1.0

Total

7.7

7.5

6.6

Variation

--

2%

13%

Figure 29. Weight distribution

Figure 30. Optimization Sequential vs. Integrated

5.2 Integrated light weight efficient optimization By excluding the heavy layer patches from the set of variables for the trim, the optimization strategy is steered to a more realistic application case. The idea is to see how the damping optimization may support the design of the trim, in case of change of technology, driven by the reduction of weight.

26

Vehicle validation of the structure-borne noise of a lightweight body and trim … The outcome of this optimization shows that by saving more than 20% of weight, it is still possible to improve the structural borne response of the vehicle. By comparing the SPL spectra (fig. 31) of the respective solutions with and without the inclusion of HL patches, the performance of the two solutions are pretty similar. The former solution gives a better (around 3 dB) performance mainly in the frequency range between 250 Hz and 350 Hz, but it is also 13% heavier (fig. 32-33).

Figure 31. Performance Improvement

Optimum Weight [kg] HMC reference Integrated with HL

Optimum Integrated without HL

Trim

4.1

5.6

4.1

Damping

3.6

1.0

1.9

Total

7.7

6.6

6.1

Variation

--

13%

21%

Figure 32. Weight distribution

27

Vehicle validation of the structure-borne noise of a lightweight body and trim …

Figure 33. Optimization with and without HL variable

6 Conclusions The methodology revealed in this publication aims at supporting floor component design, with a focus on the multi-physics interaction of all design variables that influence or contribute to the improvement of the structure borne performance of the vehicle (the body shape modification, the damping pads design as well as the trim design). This combination is done in such a way, that the design becomes more efficient, overcoming the problematic of mutual interaction of different countermeasures that influence each other and that make conventional compartmented design approaches less effective (at least from the engineering point of view). At the same time we paid attention, thanks to the exploitation of the GA technic, that the method is compatible with the type of engineering variable, available when such a problem is solved: the design variables are discrete, and include in such a way implicitly all the constraints of the design (localization, patch size, manufacturing limitations, applicability, and so on). Another key factor addressed in this article for the success of the design, is the definition of a clear optimization strategy. Indeed, not only the methodology has been applied and validated in a real vehicle test case (the collaboration with HMC and the application of the PU probes methodology have allowed, for the first time, to fully validate the optimization on a real vehicle); the methodology has been also exploited to demonstrate the influence and the importance of the optimization strategy. The comparison between integrated and sequential optimi-

28

Vehicle validation of the structure-borne noise of a lightweight body and trim … zation is a very good example: in both cases, the low frequency structural borne acoustic performance of the vehicle has been improved without increasing the weight of the design solution, but with a considerable advantage in terms of efficiency with the integrated strategy. This is the proof, that digital data availability and integration in the early development stage, is the winning scenario for NVH design engineers, given that an appropriate tool is able to take them into account in a rationalized way. In addition to this, the final “light weight” optimization has shown that, by providing the appropriate variables associated to a technological strategy, it is possible to achieve a very competitive light weight design solutions, without compromising (too much) with the NVH performance. While it is clear that a trade-off between weight reduction and performance improvements exists, this last result is the proof that lightweight treatments do not necessarily mean poor structural borne performance. Their design is a new and appealing challenge for NVH engineers.

Acknowledgments The present work was made possible and performed with the support and direct contribution of Hyundai Motor Group, Noise&Vibration CAE Team and the Production Control Department 3 and Body Department 3 in HMC Ulsan plant. The authors would like also to thank the colleagues of the Vehicle Acoustic Testing group (Autoneum – Winterthur) for the great help and support received during all the experimental activity documented in this article.

Bibliography 1 D. Caprioli, C. Gaudino, L. Ferrali and L. Hao, Shape and damping automatic vibroacoustical optimisation of automotive panels by means of GOLD, Rieter Automotive Conference, 2005. 2 R. Stelzer, T. Courtois, K. Chae, D. SEO, et al., FE Simulation of the Transmission Loss Performance of Vehicle Acoustic Components at Low and Medium Frequencies, SAE Int. J. Passeng. Cars – Mech. Syst. 7(3):2014, doi:10.4271/2014-01-2081. 3 F. Ronzio, T. Courtois and J.W. Yoo, Integration of acoustic trim in FE structureborne noise optimization with SPL target, Proceedings of SAPEM, 2014. 4 M. A. Biot, Theory of Propagation of Elastic Waves in a Fluid-Saturated Porous Solid – High Frequency Range, Journal of the Acoustical Society of America, 28(2), 1956.

29

Vehicle validation of the structure-borne noise of a lightweight body and trim … 5 P. Göransson, A 3-D symmetric finite element formulation of the Biot equations with application to acoustic wave propagation through an elastic porous medium, International Journal of Numerical Methods in Engineering, 41, 167-192, 1999. 6 S. Rigobert, F. Sgard and N. Atalla, Investigation of the convergence of the mixed displacement-pressure formulation for three-dimensional poroelastic materials using hierarchical elements, Journal of the Acoustical Society of America, 114(5), 26072617, 2003. 7 N. Atalla, R. Panneton and P. Debergue, A mixed displacement-pressure formulation for poroelastic materials, Journal of the Acoustical Society of America, 104(3), 1998. 8 C. Bertolini, L. Guj, F. Avenati Bassi, K. Misaji and F. Ide, Treasuri2/FE: A Tool for the FE Simulation of Sound Package Parts Fully Integrated in Nastran, SAE Int. J. Passeng. Cars – Mech. Syst., vol. 2, no. 1, pp. 1511-1529, 2009. 9 D. E. Goldberg, Genetic Algorithms in Search, Optimization e Machine Learning, Addison-Wesley Publishing Company, 2002. 10 L. Guj, T. Courtois and C. Bertolini, A FE based procedure for optimal design of damping package, with presence of the insulation trim. Proceedings of SAE Noise & Vibration Conference, 2011. 11 C. Bertolini, J. Horak, G. Lo Sinno, Interior panel contirbution based on pressureveloctiy mapping and acoustic transfer functions combined with the simulation of the sound package, Proceedings of the 1st Automotive Acoustics Conference, 2011. 12 F. Allard, Propagation of Sound in Porous Media, Elsevier Applied Science, 1993. 13 C. Bertolini, C. Gaudino, D. Caprioli, K. Misaji and F. Ide, FE Analysis of a Partially Trimmed Vehicle using Poroelastic Finite Elements Based on Biot's Theory, SAE Technical Paper 2007-01-2330, 2009. 14 L. Guj, C. Bertolini, T. Courtois, FE and genetic algorithm optimization od a vehicle sould package with respect to the interior SPL under weight constraints, Proceedings of the 1st Automotive Acoustics Conference, 2011. 15 C. Bertolini, Numerical Investigations on Structure-Borne and Air-Borne Insertion Loss of Acoustical Multilayers used in the automotive industry, in AIA-DAGA 2013 Conference on Acoustics, Merano, 2013.

30

Diagnostics and reparation of customers’ NVH complaints - a strategy of a carmaker Léon GAVRIC, Cyril PERONNET, Guillaume CATUSSEAU PSA – Peugeot Citroën Automobiles

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 W. Siebenpfeiffer (Hrsg.), Automotive Acoustics Conference 2015, Proceedings, https://doi.org/10.1007/978-3-658-27648-5_7

1

Diagnostics and reparation of customers’ NVH complaints - a strategy of a carmaker

1 – Introduction In automotive industry, the “diagnostics-ability” can be defined as “capacity to identify the defective part to be repaired or replaced on a vehicle using the customers’ complaints as input”. This is not an easy task when noise and vibrations are considered, due to the complexity of the excitation and transfer phenomena involved.

It comes out that it’s often difficult to identify the right root cause of a NVH complaint, when a customer brings his vehicle to be repaired. This may result in an inappropriate intervention of mechanics inducing important costs and customer insatisfaction.

The statistic analysis of the aftersales repair expenses shows that nearly quarter of the costs induced by the NVH complaints can be considered as useless. Such costs can reach several million of euros per year for middle size carmakers. Since the unwanted NVH phenomenon is still present, such intervention can even create a supplementary frustration of customer. Generally, in the automotive industry, the NVH item is one of the principal warranty costs with suspected inappropriate repairs

Figure 1 – Left: NVH diagnostic issues; Right: Typical diagnostic test sequence Let’s summarize this silly situation and figure its consequences! • On the one hand: the mechanics who have engaged reparation costs unnecessarily, while thinking to do the best for car producer image and customer’s satisfaction. • And on the other hand: the customer whose vehicle is still not repaired, who has spent his time pointlessly during one or several garage interventions and who is even

2

Diagnostics and reparation of customers’ NVH complaints - a strategy of a carmaker more frustrated by the NVH phenomenon still present, while blaming lack of competence of carmaker’s mechanics. The most obvious way out from such embarrassing situation would be to describe accurate diagnostic instructions, aimed to guide the mechanics or analysts from the customer complaint to the component(s) to be repaired using specific test procedure(s). Unfortunately, such a never-failing procedure does not exist when NVH is considered. The main difficulty relies on the fact that neither objective nor subjective description of noise and vibration phenomena can be explicit / accurate enough to be used by a nonspecialist. Following facts can be observed during the intervention related to NVH complaints in our workshops: • Inacceptable long time laps needed to inform mechanics of the known NVH defaults generated during the development of product, which generates recurrent and useless repairs’ costs. • Insufficient information enhancement concerning costumer NVH complaints (vehicle operating conditions, location of noise source, temperature…) leading to the incapacity to reproduce the defect in order to repair and to verify the non-recurrence. • Lack of specific tools for the noise and vibration diagnostics appropriate for NVH non-specialists (mechanics in the garages). As response to these facts, the R&D and the aftersales engineers of PSA Peugeot Citroën Group have decided to develop a specific methodology of treatment of NVH complaints/defects for passenger cars. The procedure includes: the powertrain, the chassis, and the cockpit.

Figure 2 – Left: Scope of the NVH diagnostic; Right: The PSA diagnostic system

3

Diagnostics and reparation of customers’ NVH complaints - a strategy of a carmaker

The diagnostic system (or procedure), described in the paper comprises: methodological framework and specific tools designed to facilitate the diagnostic procedure and NVH defect handling. The diagnostic system of PSA Group consists of: • A garage related procedure with specific requirements to the noise and vibration enhancement. • A unique semantic Classification of noise vibrations phenomena aimed to improve the communication between R&D engineers, production plants and aftersales people. • A Customer Incident Form needed to enhance the information of the NVH complaint, which is based on the noise and vibration semantic classification. • A prototype tool for Ranking of diagnostic procedures relevant to NVH symptoms (complaints or defects). • A diagnostic tool - AudioBox ® - needed for diagnostic of the most delicate cases by experts. • A Training activity including both the face-to-face course and a simplified e-learning module.

2 – Garage related procedure Procedure to be applied in the garage describes the chronological sequence of the different diagnostic activities. They can be defined as a set of instructions which guarantees a structured and unique approach. The objective is to improve costumer’s satisfaction and minimize aftersales costs, by: • Diminishing or even eradicating unjustified part deposited and reparation activity. • Repairing immediately the known NVH defects with authorized counter-measure. • Guiding the diagnostic to the most probable cause of NVH complaint, while conditioning the deposit and repair to central support’s technical assistance.

4

Diagnostics and reparation of customers’ NVH complaints - a strategy of a carmaker

Figure 3 – Left: NVH diagnostic basis; Right: NVH diagnostic procedure This procedure is thus based on these 3 NVH types, which are then associated with the customer complaints and related to the most relevant NVH sources. After mandatory reproduction of the NVH default in the presence of the customer, the procedure implies to execute the process-oriented diagnostic procedure adapted to the noise type (accepted, failure, other): • The accepted product noises are associated with known NVH phenomena, which are judged as acceptable during the development and which correspond to the actual design and production process. Since, any part exchange or repair would not resolve the NVH complaint, any garage activity is therefore not allowed. The customer is informed of the constancy and predictability of the phenomenon and on the lack of impact on the reliability of his vehicle. The recordings corresponding to the accepted product noises are integrated into a specific NVH aftersales data basis under the flag "Diagnostic Assistance”. • The failure product noises correspond to known NVH phenomena, too. They are different from the accepted ones because they are caused by inconformity of the production process or by the premature degradation of the parts. Since they are identified during the development of the product, the specific procedures of one-shot repair are already defined. The procedures are then applied without any supplementary and more complex diagnostic activity. The recordings of failure noises are integrated into the aftersales NVH data basis under the specific flag “Right to the target”. • The other product noises correspond to all phenomena which are not inventoried as known on the given car, and request an in-depht analysis. The instruction is then to ask the central support for technical assistance. The aim is to ensure an accurate identification of the root cause before any repair or part exchange. The set of specific diagnostic procedures is used, proposing a sequence of tests to be executed before any intervention of mechanics. The idea is to authorize the repair activity only once the problem is accurately identified.

5

Diagnostics and reparation of customers’ NVH complaints - a strategy of a carmaker

Figure 4 - Synopsis of the garage related procedure

6

Diagnostics and reparation of customers’ NVH complaints - a strategy of a carmaker The efficiency of the procedure is maximized if all the complaints corresponding to the known NVH phenomena (accepted or failure type) can be detected, which rapidely brings the most appropriate response to customer.

3 – Classification of noise and vibrations According to the garage related procedures, it is crucial to reproduce the NVH complaint in the presence of the customer before any further intervention. Once the NVH phenomenon is identified and validated by customer, it has to be documented in order to be shared (garage, aftersales, R&D, production) and handled. When a written document is to be established the description of the noise is one of the most difficult tasks. Descriptive onomatopoeia as “ou-ou-ou” noise or “ie-ie-ie” noise are definitely not the best choice to characterize a NVH customer’s complaints. It is therefore necessary to identify unique semantic classification of noise vibrations phenomena. In PSA Peugeot Citroën Group a reference terminology is defined, to be applied to whole vehicle (powertrain, chassis, and cockpit) whenever NVH complaints are to be described. These noise and vibration phenomena are classed according to principal frequency content of basic acoustic pattern and its recurrence. The final result is 6 families of NVH complaints in automotive industry: • 1 single family of vibration - according to tactile sensation • and 5 familles/types of noises – according to subjective noise sensation

Figure 5 – Left: Noise classification diagram; Right: Blind test validation menu Thus was born the PSA Peugeot Citroën classification of noise and vibration, designed to help choosing the family which most accurately describes the customer’s complaint. The choice is done using frequency-recurrence diagram/scale.

7

Diagnostics and reparation of customers’ NVH complaints - a strategy of a carmaker Such reference classification becomes a company standard, used in R&D, production and aftersales departments as well as in garage by mechanics during diagnostics. The classification is validated by blind tests involving more than 100 people, both NVH experts and non-specialists. The use of the classification methodology leads to standardized NVH vocabulary, resulting in better identification and description of customer’s complaints. This allows more accurate processing and quality assessment concerning the NVH.

4 – Customer incident form – CIF In order to identify the NVH event leading to the customer’s complaint, the accurate description of the phenomenon using the classification of noise and vibration is not sufficient. For complete identification, an accurate assessment of driving and environmental conditions is needed, too. The Customer Incident Form is designed to enhance all this information. The idea is to use 5 WH-words to ask the right questions: • Who: identification of the client, history of the vehicle. • What: description of the complaint using the standard classification of noise and vibration. • Where: subjective perception of localization of the NVH phenomenon. • When: subjective temporal description of occurrence of NVH phenomenon. • How to: conditions inducing NVH phenomena.

Figure 6 – Left: Five “wh-words” concept; Right: Peugeot and Citroën brands CIF

Such formalism allows the mechanics to associate rapidly the complaint with a known NVH phenomenon. When NVH incident is difficult to identify, the mechanics can con-

8

Diagnostics and reparation of customers’ NVH complaints - a strategy of a carmaker tact the central technical support. In such a case, the exhaustive description of complaint via Customer Incident Form is of precious help.

5 – Ranking of diagnostic procedures Even when complaints are described using Classification of noise and vibrations and Customer Incident Form, the NVH diagnostics remains a hard task. It is basically a kind of trial and error procedure, based on the verification of the root cause hypothesis by an appropriate test/experiment. It requires NVH skills, analysis capacity and rigor. The relevance of supposed root causes will strongly influence the success of diagnostic procedure. In order to improve the success rate of diagnostic, PSA Peugeot Citroën intends to develop a software tool, which automatically proposes the most relevant hypotheses and related validation experiments. The methodology is based on the comparison of data enhanced in CIF and the databases containing the known NVH phenomena. • When CIF data match to those corresponding to accepted product noise, the repair is not authorized, while the customer is informed that such defect hasn’t any impact on the reliability of his car. • When CIF data are recognized as failure product noise, the software will orient the mechanics to the one-shot repair. • If CIF data don’t match to any known phenomenon the software recommends to ask help to the central technical support, which can suggest the appropriate tests, aimed to clearly identify the defect before any part exchange or repair.

Figure 7 - Left: Tool for automatized diagnostic ranking; Right: AudioBox tool.

9

Diagnostics and reparation of customers’ NVH complaints - a strategy of a carmaker For the moment, the tool exists only in a form of prototype homemade software. Its completion remains one of the major challenges for future development

6 – AudioBox One of the major requests of PSA Peugeot Citroën mechanics, dealing with noise and vibration complaints, is to have access to appropriate tools dedicated to NVH assessment. Such a tool, conceived by PSA Peugeot Citroën to help localizing and assessing noise and vibrations, is named AudioBox®. The development is done in collaboration with One-Too, a designer and manufacturer of aftersales automotive solutions. AudioBox® is aimed to increase efficiency and expertise of the mechanics and flying specialists from technical central support. AudioBox® is equipped with a hand-held microphone for near field listening and 4 collets furnished each one with three-axis accelerometer and small size removable microphone. It is designed to serve as extended and focusing human ears, able even to hear the structure noise and vibrations in the both static and running vehicle conditions. In order to allow an accurate subjective analysis and recording of structure borne sound, a particular attention is paid to the design of the collets, in order to avoid any disturbing resonance phenomenon. AudioBox® enhances microphone and accelerometer signals and enables real time listening of the airborne and structure borne sound without any parasite noise. Consequently the capacity to recognise the noise generating component is amplified, which largely improve the quality of analysis and rapidity of diagnostic procedure. The AudioBox® is a user friendly tool. It enables recording, play back and data export, wich simplfies the back-office analysis and the data exchange between aftersales workshops, central technical support, and R&D teams. The final result is an accurate, flexible and easy to use tool, particularly appropriate for automotive NVH diagnostics. AudioBox® tool is awarded during the EquipAuto show held on October 2013 in Paris, and largely promotted by the press, radio, TV.

10

Diagnostics and reparation of customers’ NVH complaints - a strategy of a carmaker

Figure 8 - Presentation of AudioBox® during 2013 EquipAuto show In use in R&D, production and aftersales departments, AudioBox® is intended to become a common standard for measurement, real time subjective analysis, recording and replay of the customers’ NVH complaints. It is furthermore designed to be easily updated by the new features as, for example, sound recognition. This last topic represents the probably most important challenge of future developments.1

1 More information here: http://www.one-too.com/fr/produits/audiobox

11

Diagnostics and reparation of customers’ NVH complaints - a strategy of a carmaker

7 – Training The previously described tools help the aftersales operational network and garage mechanics to ensure an efficient diagnostics of the NVH complaints, and subsequently a suitable repair. To insure an efficient application of the methodology, the developed procedures and tools have to be used in an appropriate way. Therefore, some training is required. For this purpose an on-line module which explains methods and tools is created. It also contains the sound recognition tests using NVH complaints as examples. This on-line training is particularly focussed on identification of NVH defect types (accepted / failure / other product noises), and on the classification of noise and vibration (6 families), which are the two pillars of the methodology.

Figure 9 - Left: NVH Diagnostic e-training module; Right: face-to face NVH course Moreover, when putting a new vehicle on the market, classical training course is organized (face-to-face of teacher and students). It comprises: practical exercises with dedicated tools (blind test on NVH classification, completion of Customer Incident Form, use of AudioBox ®…) and exhaustive presentation of all known NVH phenomena which can potentially cause the customer’s complaints for the given car.

8 – Concluding remarks The first ambition of this PSA NVH diagnostic system is to develop our aftersales network expertise by:

12

Diagnostics and reparation of customers’ NVH complaints - a strategy of a carmaker • Structuring the analysis and diagnostic procedures. • Standardizing lexical and phenomenological descriptions and terminology. • Training the mechanics to the dedicated equipment.

Figure 10 – Left: Summary of diagnostic expertise; Right: NVH diagnostic benefits

The first results are impressive: improved diagnostic accuracy (80% as target) and several million euros annual savings due to reparation and exchange of parts being real root cause NVH complaints. As a consequence an increase of customer satisfaction concerning NVH quality items has been noticed. The most efficient vectors of this success are the “right to the target” and the “diagnostic assistance” medias. The key part of the two sequences consists in preparation and documentation of NVH phenomena observed during product development. Nowadays, the establishing of database needed for One-Shot and no repair instructions becomes a standard work package for powertrain, chassis and cockpit development.

Figure 11 – Left: NVH database framework; Right: Project development procedure.

13

Diagnostics and reparation of customers’ NVH complaints - a strategy of a carmaker At the end of project, NVH identity card is then created, and delivered to the aftersales department in a form of such a database. It contains the sound samples recorded with Audio Box by R&D teams at the end of development. As the AudioBox is also used in everyday NVH diagnostics in garages, it becomes possible to listen and compare R&D and garage recordings. By such a subjective analysis, one can confirm the defect responsible for the NVH phenomenon. The final target would be to develop an appropriate signal processing (i.e.sound recognition) in order to automatically detect the root cause of unwanted NVH phenomenon. Such a tool, similar to a kind of “Shazam” application for automotive NVH complaints, represents the most challenging part of future developments.

Figure 12 – Next step - future automatized diagnostic tools

Acknowledgement: The authors acknowledge the contribution of their colleagues to this work: O. Bouttaz for benchmarking and project launch, C. Seite and S. Geslin for their support to this project management, W. Raguenet for AudioBox development, O. Marck and P. Le for

14

Diagnostics and reparation of customers’ NVH complaints - a strategy of a carmaker test sequences validation, and many other people for their partipation to this venture (F. Croc, EM. Bertomeu, JL. Ferre, V. Vairac, X. Salles, R. Da Silva, A. Regnier, JP. Fernandes, T. Duparchy, A. Duvivier, R. Spehner, C. Royer Mladenov, GP. Martin, M. Deshais, K. Ramon, R. Chhun, D. Ledoux, J. Deak, G. Mahajoub, L. Barre, E. Rolland, P. Rousseaux).

References: [1] C. Peronnet: "Système de Diagnostic Bruits Vibrations PSA", « vendredi de l’expertise », 2011. [2] C. Peronnet: "Système de Diagnostic Bruits Vibrations ", « 25ans Vibratec », 2011. [3] F. Croc: "Maintenabilité, Km0 pour politique technique ", PSA Peugeot Citroën technical references, 2011. [4] G. Catusseau: "Classification bruits-vibrations PSA", PSA Peugeot Citroën technical references, 2009. [5] O. Bouttaz, G. Catusseau: "Fiche Incident Client Acoustique Vibrations", PSA Peugeot Citroën technical references, 2009. [6] G. Catusseau, W. Raguenet, C. Peronnet: "AudioBox, dotation des réseaux AC/AP", PSA Peugeot Citroën aftersales tool deployment, 2013. [7] G. Catusseau: "Processus de co-conception du diagnostic des bruits vibrations", PSA Peugeot Citroën technical references, 2011. [8] C. Seite, X. Salles: "Processus diagnostic bruits et vibrations", PSA Peugeot Citroën technical references, 2014. [9] C. Seite, L. Blampain: "Questionnaire diagnostic bruit pour les visites en garage", PSA Peugeot Citroën technical references, 2014. [10] G. Catusseau, C. Seite, X. Salles: "Standard des données d’Aide au Diagnostic", PSA Peugeot Citroën technical references, 2014. [11] X. Salles: "Standard des Aides au Diagnostic", PSA Peugeot Citroën technical references, 2014. [12] C Peronnet, L Gavric, G Catusseau: " The PSA NVH diagnostic system", SIA conference « Automotive NVH comfort », 22-23 October 2014 Le Mans, France.

15

Enhanced lightweight NVH solution based on vibro-acoustic metamaterials C. Claeys, E. Deckers, B. Pluymers, W. Desmet KU Leuven, Noise and Vibration Research Group

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 W. Siebenpfeiffer (Hrsg.), Automotive Acoustics Conference 2015, Proceedings, https://doi.org/10.1007/978-3-658-27648-5_8

1

Enhanced lightweight NVH solution based on vibro-acoustic metamaterials

Abstract The NVH performance of conventional panels and structures is mainly driven by their mass. Vibro-acoustic metamaterials with stopband behaviour are looked upon as a possible solution for combing NVH and lightweight requirements for engineering structures. This contribution discusses a metamaterial concept based on sandwich structures with embedded resonant structures that exhibit vibro-acoustic stopband behaviour. This potential is shown through the design and measurement results of two demonstrators. First a metamaterial encapsulation is designed and produced to showcase the potential of this type of metamaterials to improve acoustic insulation performance in targeted frequency ranges. Second a graded metamaterial strip is designed; this strip is built as a combination of metamaterials with different design frequencies, opening up the potential for broadband vibration reduction in a low frequency zone. These metamaterials can be achieved through a variety of designs or production processes, depending on the application, and can offer technological benefits such as possible integration in structural parts, use in harsh environments and ease of design of the beneficial frequency ranges, paving the way for a new class of light and compact NVH solutions with ample applications in automotive for low and mid frequency NVH mitigation.

1 Introduction Increasing customer expectations and more restrictive legal requirements turn the Noise and Vibration Harshness (NVH) behaviour of products into an important design criterion in the machine and transportation industry as well as in the construction and consumer goods sector. Ecological trends and the associated run for efficiency, however, increase the importance of lightweight design and reduce the applicability of classical (heavy) solutions to improve NVH behaviour. In view of this challenging and often conflicting task of merging NVH and lightweight requirements novel solutions are required. Ideally these novel solutions are easy to design and are characterised by a low mass and compact volume along with a high reliability at an affordable cost. Vibro-acoustic metamaterials come to the fore as possible candidates for lightweight material systems with superior noise and vibration insulation, be it at least in some targeted and tuneable frequency ranges, referred to as stop bands. Contrary to phononic crystals, stop bands in metamaterials do not rely on periodicity or Bragg scattering and work on spatial scales much smaller than the wavelength [9]. The stop bands induced in metamaterials result from resonant cells arranged on a subwavelength scale and can be described based on the Fano-type interference between incoming waves and the waves re-radiated by the resonant cells [5] [7].

2

Enhanced lightweight NVH solution based on vibro-acoustic metamaterials Previous papers of the authors explain the working principles of stop bands based on resonant metamaterials and list the driving parameters for stop band design [2]. Furthermore, the authors introduced a novel metamaterial concept based on sandwich structures with embedded resonant structures that exhibit vibro−acoustic stopband behaviour [3]. This paper continuous the research by using the same metamaterial concept and investigating the potential for broadband vibration reduction in a low frequency zone. This paper is structured as followed. Section 2 discusses the rationale behind the proposed metamaterial concept, followed by a section describing the numerical prediction of the stop band behaviour of this concept. Section 4 shows the potential of this concept for acoustic reduction, while section 5 discusses a demonstrator for broadband vibration reduction. The paper ends with the conclusions.

2 Metamaterials by inclusion of resonant structures Metamaterials with stopband behaviour are obtained through the inclusion of resonant cells on a scale smaller than the structural wavelengths to be influenced [2] [8]. Stop band behaviour can thus be achieved through the introduction of any system that introduces local resonant behaviour. In view of applications, the goal is to find resonant systems which do not jeopardise other requirements such as structural integrity, light weight, applicability in contaminated environment, fire-resistance, … . The kinds of resonant systems which are eligible heavily depend on the structure to which the resonant systems have to be added. Inspiration for a high potential structure to introduce stop band behaviour is sought in the class of periodic lightweight structures, such as honeycomb core sandwich panels. They are becoming attractive for application in transport and machine design due to the combination of excellent mechanical properties with a low mass. Figure 1 shows examples of sandwich structures; a core acts as spacer to create distance between the skins such that a light structure with excellent stiffness properties in bending is obtained. The core has as main role to create distance between the skins as well as to resist forces perpendicular to the structure while the skin is designed to show a high in plane strength. Given the different requirements for both, often the skin is made of a different material as the core. Different core layouts are possible and two typical layouts are hexagonal and rectangular cores.

3

Enhanced lightweight NVH solution based on vibro-acoustic metamaterials

Figure 1: Examples of sandwich structures; left a hexagonal core, right a rectangular core [10].

The internal cavities of these periodic core sandwich panels are an ideal host structure to include resonant systems. In such configuration the resonant systems can be freely designed for optimal acoustic behaviour in a targeted frequency zone while safeguarding the excellent mechanical properties of the sandwich panels. The only requirement for the resonant structures is the presence of a resonance mode; a huge number of designs could be proposed, leaving room for optimisation towards e.g. minimal weight, maximal attenuation or broadband attenuation. For this paper a straightforward design is chosen that resembles a mass-spring system; two thin legs are used to connect a heavy mass to a host structure (Figure 2). The connection legs will determine the stiffness while the thick part of the resonator will determine the mass of this resonant structure.

Figure 2: Resonant structure used in the acoustic demonstrator to introduce stop band behaviour.

Adding the resonant structures to the cavities of a periodic sandwich core (Figure 3) will introduce a stop band, and thus improved acoustic behaviour, for frequencies in the vicinity of the resonance frequency of the resonant structure.

4

Enhanced lightweight NVH solution based on vibro-acoustic metamaterials

Figure 3: Rectangular core without (left) and with (right) resonant structures.

3 Stop band prediction From literature it is known that wave propagation through infinite periodic structures can be investigated through unit cell modelling [1] [6]. Based on an undamped Finite Element (FE) model of the unit cell and the application of periodicity boundary conditions, dispersion curves for freely propagating waves in an infinite periodic structure can be derived. Frequency zones for which no solutions are found, correspond to frequency zones without free wave propagation and thus a stop band region.

Figure 4: Dimensions in millimeter of the unit cell of the acoustic demonstrator. The resonator leg thickness (L) equals 1 mm, the resonator mass (M) equals 4 mm.

5

Enhanced lightweight NVH solution based on vibro-acoustic metamaterials Figure 4 shows the geometry of the unit cell of the acoustic demonstrator of section 4. The resonant structure makes up 30% of the weight of the unit cell. Figure 5 depicts the FE model of this unit cell. Linear Quad4 elements are used to represent the resonant structure, core and skin. The resonator mass has a larger thickness than the connection legs; different properties are assigned to both element groups. The resonator mass element group is depicted in grey on the right side of Figure 5.

Figure 5: FE model of the unit cell of the acoustic demonstrator. The grey elements in the top view (right) represent the resonator mass.

The material characteristics of the acoustic enclosures are given in Table 1. The enclosure is made through Selective Laser Sintering (SLS) of Polyamide, the material and production characteristics are discussed in more detail in next section. Since the main goal of the simulations is to get an indication whether stop band behaviour is present rather than obtaining a detailed model, isotropy and linear behaviour of the material are assumed. Table 1: Material characteristics of the material of the unit cell in the numerical simulations. Name

Value

Young’s modulus

1.10 Gpa

Density

950 kg/m3

Poisson’s ratio

0.4

The resonant structure is designed to have a pronounced low frequent bending mode followed by subsequent modes higher in frequency. Figure 6 shows the first two modes of the free unit cell which can be related to modes of the resonant structure. On the left the bending mode is shown which occurs at a frequency of 889 Hz; the next mode,

6

Enhanced lightweight NVH solution based on vibro-acoustic metamaterials shown at the right, occurs at 4370 Hz. This large shift in frequency between the modes allows to correlate stop band behaviour with a certain resonance mode. Convergence of the model was validated against a refined unit cell model of 6766 nodes and 6576 elements. The used model with 198 nodes and 168 elements has an accuracy of 1 % on the first mode and 1.5 % on the second mode.

Figure 6: Undeformed mesh (light) and the deformation of the first modes (dark) of the resonant structures. Left: bending mode at 889 Hz. Right: torsional mode at 4370 Hz.

The unit cell allows derivation of the dispersion curves of the metamaterial design. Figure 7 shows the bending wave dispersion curves for this unit cell in comparison to the dispersion curves of the host structure without resonant structures. The dispersion curves are similar except for a region around the resonance frequency of the resonant system for which no dispersion curves exists; a stop band opens up from 886 Hz up to 999 Hz.

Figure 7: Dispersion curves of the metamaterial with resonant structures (—), which can be seen as the equivalent counterpart of the right side of Figure 3, in comparison with the dispersion curves of the host structure without resonant structure (—), left side of Figure 3.

7

Enhanced lightweight NVH solution based on vibro-acoustic metamaterials

4 Acoustic enclosure demonstrator To prove the potential of the introduced metamaterials concept to reduce acoustic transmission, an acoustic enclosure making use of a resonant structure based metamaterial is designed. A variety of demonstrators can be thought of to determine the acoustic transmission characteristics of a material. In the choice for a suitable demonstrator, some determining factors are: unambiguous proof of concept, quantifiable effect and engineering relevance. It was chosen to design acoustic enclosures as boxes with one open side which can be placed over a small speaker. The acoustic transmission loss is then determined by comparing sound radiation with and without enclosure. Additive manufacturing is chosen as production process since it allows producing complex parts without the need of an expensive mould, making it a suitable production process for prototype design. Within the range of additive manufacturing processes, Selective Laser Sintering (SLS) is selected to create the demonstrator; this process allowed most freedom in design.

Figure 8: Nominal demonstrator design; side (left) and bottom (right) view.

Figure 8 shows a side and bottom view of the demonstrator design. One side of the enclosure contains 8 x 8 unit cells. It should be noted that the corners of the demonstrator are hollow and no resonators are added. The size of the enclosure is a balance between obtaining a light demonstrator and still being relevant; the enclosure is designed such that the inner dimensions make up a cube of 100 x 100 x 100 mm. Figure 9 shows a produced version of this demonstrator.

8

Enhanced lightweight NVH solution based on vibro-acoustic metamaterials

Figure 9: Pictures of the acoustic demonstrator.

The test set-up for acoustic evaluation consists of a small loudspeaker placed on a wooden plate (Figure 10). The wooden plate is placed on an iron support in the centre of a semi-anechoic chamber. Between the wooden plate and the loudspeaker, a small trimlike piece of fabric is placed; the trim covers the cable of the loudspeaker which runs in a small split in the plate. The sound power is evaluated on the surface of a surrounding box of 250 x 250 x 190mm centred on the loudspeaker; the black dots on the wooden plate (Figure 10) indicate the width and length of the surrounding box. The acoustic power is evaluated based on 5 intensity measurements, one on each side of the box, with a scanning intensity probe.

Figure 10: Picture of the test set up (left) and an intenisty measurement of an acoustice metamaterial enclosure on the test set up (right).

9

Enhanced lightweight NVH solution based on vibro-acoustic metamaterials Figure 11 compares the measured insertion loss of the metamaterial demonstrator to the measured insertion loss of a regular enclosure with the same mass but build from flat panels of 3.5 mm thickness. Between 700 and 1000 Hz the demonstrator enclosure clearly outperforms the regular enclosure. The frequency zone of improved acoustic behaviour is a bit wider than predicted by the stop band simulation (886 – 999Hz), but this corresponds well with previous simulations of the author [2]. There it was shown that the start of the predicted stop band corresponds with the point of maximal attenuation while the end of the stop band corresponds to the next resonance and point reduced insertion loss; in general the zone of strong attenuation is thus always centred around the starting frequency of the stop band zone predicted.

Figure 11: Comparison between the measured insertion loss for the metamaterial demonstrator and a demonstrator with flat side panels and an equivalent weight to the metamaterial demonstrator.

5 Broadband vibration reduction demonstrator To prove the potential of metamaterials for vibration reduction, a strip of metamaterial is designed to investigate how force is transmitted from one side to the other side of the structure. A simple strip of material has the advantage that the transmission path is straight forward and that the structure can be freely hinged, reducing the effect of boundary conditions. From the acoustic demonstrator, it is clear that a stop band leads to a zone of reduced vibrations and hence reduce acoustic transmission. Therefore, in this realisation, not just one stop band is designed, but throughout the strip different resonant structures are used such that, when the wave is transmitted from one side to the structure to the other, different stop bands will be encountered and can reduce the transmitted energy in different frequency bands.

10

Enhanced lightweight NVH solution based on vibro-acoustic metamaterials Figure 12 shows a sketch of the demonstrator: the demonstrator is built up out of 6 zones, A to F, which consist of 3 by 3 resonant cells. The resonators in each of these zones have a different leg width or mass thickness, Figure 13, in order to change the resonance frequency and thus the associated stop band. The different widths and thicknesses of the resonant structures are listed in Table 2. A section of 3x3 cells is chosen for each zone since previous research has indicated that wave attenuation is clearly visible as soon as a wave has travelled past 3 resonant cells [4].

Figure 12: Sketch of the structure used for vibration reduction testing. The numbers on the sketch indicate both the measuring points. The capital letters indicate the different metamaterials zones.

Figure 13: Dimensions in millimeter of the unit cell of the metamaterial strip for vibration testing. The resonator leg width (w) and the resonator mass thickness (t) equals are chosen according to Table 2.

11

Enhanced lightweight NVH solution based on vibro-acoustic metamaterials Table 2: Dimensions of the leg width and mass thickness for each metamaterials zone in Figure 12 together with the predicted stop band zones. Zone A B C D E F

w [mm] 1.5 2 2.5 3 3.5 4

t [mm] 3 4 5 5 5 5

Stop band [Hz] 283-316 311-351 331-376 374-421 413-481 452-532

Table 3: Material characteristics of the material of the unit cell in the numerical simulations. Name

Value

Young’s modulus

3.30 Gpa

Density

1190 kg/m3

Poisson’s ratio

0.37

Figure 14 shows the produced structure. The structure is designed through laser cutting of Plexiglas. As can be seen on Figure 15, by gluing different layers together, a metamaterial can be derived. Based on the material parameters of Plexiglas, given in Table 3, the stop band predictions for each metamaterial zone can be performed. The results are listed in Table 2.

Figure 14: Picture of the structure for vibration testing

12

Enhanced lightweight NVH solution based on vibro-acoustic metamaterials

Figure 15: Layerwise production merthod of a metamaterial

Given the lightweight nature of the structure, impact testing is chosen as testing method for this structure. Impact testing has as advantage that no excitation device is attached to the structure, but nevertheless the input force can be measured by using an impedance head (type PCB 086C03). For the response measurement lightweight accelerometers (type PCB 352A24 with a weight of 0.8 gram a piece) are used. These accelerometers are very small and are light enough to have a negligible influence on the structure, overcoming the main hurdle of accelerometer measurements. To measure the wave propagation through the structure, the structure is hinged by 2 bungees, through holes in the skin close to points 1 and 4 on Figure 12, and the response in the accelerometers based on the input with an impact hammer is measured. Point 3 is used as an input point, while points 3, 6, 11, 14, 19, 22 and 27 (Figure 12) are used as response points. Figure 16 shows the resulting frequency response curves. Throughout the structure a strong wave reduction is achieved. Comparison of the direct response (3/3) versus the response of the point opposite of the impact point (27/3), indicates a strong improvement between 250 and 500 Hz. Furthermore, it can be seen that the zone of vibration reduction become wider in the direction of higher frequencies every time a metamaterial zone is passed, as is expected based on the design of the structure. This design thus serves as a first proof of concept how vibration reduction can be obtained in a wider frequency range using a graded design of a metamaterial structure.

13

Enhanced lightweight NVH solution based on vibro-acoustic metamaterials

g/N dB

30.00

3/3 6/3 11/3 14/3 19/3 22/3 27/3

-30.00 10.00

Hz

600.00

Figure 16: Measured acceleration in the response points 3, 6, 11, 14, 19, 22 and 27 for excitation in point 3 (Figure 12).

6 Conclusions This paper introduces a novel method to create resonant metamaterials for NVH materials; the inclusion of resonant structures within the core of a sandwich structure. Through the design of an acoustic demonstrator it is shown that this results in a frequency zone of increased acoustic insertion loss with respect to equivalent materials of the same weight. To see and hear this potential in a movie, the interested reader is referred to a movie which can be seen on the following link (http://youtu.be/tOch_GsGaXg). Also a graded design has been proposed, combining different unit cells, with different stopbands to reduce vibrations in a wider frequency range. Promising first results are presented, showing a clear vibration reduction in a low-frequency band from 250 to 500Hz. The design of this metamaterials is aided by unit cell modelling: this tool allows a quick estimation of the location of the stop band frequencies and can be used as to assess changes in resonant structure design. This metamaterial concept allows the combination of light weight, compact mass and good NVH behaviour along with other technological benefits such as integration in structural parts, use in harsh environments and easy designable beneficial frequency ranges.

14

Enhanced lightweight NVH solution based on vibro-acoustic metamaterials

Acknowledgments The authors would like to acknowledge the European Commission for their support through the ENLIGHT-project (http://www.project-enlight.eu/) and the KU Leuven Research Fund for their support through an IOF-Leverage project. Elke Deckers is a Postdoctoral Fellow of the Fund for Scientific Research-Flanders (F.W.O.), Belgium. Furthermore, Cédric Peeters and Roeland Pollet are gratefully acknowledges to produce and measure the graded metamaterial design in the scope of their Ma thesis project.

References [1] L. Brillouin, Wave propagation in periodic structures, 2nd ed., McGraw-Hill Book Company, 1946. [2] C.C. Claeys, K. Vergote, P. Sas, and W. Desmet, On the potential of tuned resonators to obtain low frequency vibrational stop bands in periodic panels, Journal of Sound and Vibration 332 (2013), no. 6, 1418–1436. [3] C.C. Claeys, Design and analysis of resonant metamaterials for acoustic insulation, PhD Thesis, KU Leuven, 2014. [4] C.C. Claeys, M.Vivolo, P. Sas, and W. Desmet, Study of honeycomb panels with local cell resonators to obtain low-frequency vibrational stopbands, Dynacomp 2012, Arcachon, France. [5] C. Goffaux, J. S´anchez-Dehesa, A.L. Yeyati, P. Lambin, A. Khelif, JO Vasseur, and B. Djafari-Rouhani, Evidence of fano-like interference phenomena in locally resonant materials, Physical Review Letters 88 (2002), no. 22, 225502. [6] R.S. Langley, A note on the force boundary conditions for two-dimensional periodic structures with corner freedoms, Journal of Sound and Vibration 167 (1993), 377–381. [7] F. Lemoult, N. Kaina, M. Fink, and G. Lerosey, Wave propagation control at the deep subwavelength scale in metamaterials, Nature Physics 9 (2012), 55–60. [8] Z. Liu, X. Zhang, Y. Mao, YY Zhu, Z. Yang, CT Chan, and P. Sheng, Locally resonant sonic materials, Science 289 (2000), no. 5485, 1734–1736. [9] J B. Pendry, D.J. Holden, A J.and Robbins, andW.J. Stewart, Magnetism from conductors and enhanced nonlinear phenomena, Microwave Theory and Techniques, IEEE Transactions on 47 (1999), no. 11, 2075–2084. [10] H.Wadley, Multifunctional periodic cellular metals, Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences 364 (2006), no. 1838, 31–68.

15

Sound field control in the automotive environment Jordan Cheer, Stephen J. Elliott and Woomin Jung

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 W. Siebenpfeiffer (Hrsg.), Automotive Acoustics Conference 2015, Proceedings, https://doi.org/10.1007/978-3-658-27648-5_9

1

Sound field control in the automotive environment

1 Introduction Sound field control involves the generation and manipulation of acoustic fields using electroacoustic transducers [1]. This encompasses spatial audio reproduction [2], zonal sound control [3], active absorption or reflection of sound [4], and active noise control [5]. In all of these technologies either loudspeakers or structural-acoustic actuators are used to produce a sound field with a controlled spatial and temporal distribution. For example, in the context of spatial audio an array of loudspeakers may be driven to reconstruct a pre-defined sound field, such as that experienced in a concert hall during an orchestral performance [6]. In the automotive environment, sound field control has been used to provide additional functionality and to overcome traditional performance limitations in both the audio system and in the control of vehicular noise. For example, in the context of the car audio system, sound field control methods have been used to improve the typical stereo imaging problems caused by off centre listening [7] and to provide spatial audio reproduction capabilities [8, 9]. In the context of the noise, vibration and harshness (NVH) performance, active noise control has been employed to control both engine and road noise [10]. The application of active noise control to the automotive environment has been investigated within the automotive industry for in excess of 20 years [10]. This work has generally focussed on reducing the interior engine [11] or road noise [12] in the vehicle cabin using a number of loudspeakers. This process is based on generating a sound field that is out of phase with the unwanted noise produced by the vehicle, and thus reducing it through destructive interference. The performance of active noise control systems is thus dependent on both the spatial and temporal accuracy with which the sound field produced by the control loudspeakers matches the unwanted, disturbance sound field [4]. In the context of the automotive audio system, sound field control has recently been employed to generate multiple independent listening zones in the car cabin [13]. This system allows the occupants of the car cabin to listen to individualised audio entertainment without the need for headphones. This system essentially relies on the same physical principles as active noise control, however, in this case the unwanted noise in each listening zone which must be cancelled is due to the audio content being reproduced in the other listening zones in the car cabin. This paper presents a review of the limitations of sound field control within the automotive environment. In Section 2 the limitations of global active control in the automotive environment are first outlined, and then practical examples are used to highlight the limitations for both engine and road noise control. In Section 3 it is then shown how some of these limitations may be overcome by using a local active control strategy and, once again, a practical automotive application is used to demonstrate the potential advantages

2

Sound field control in the automotive environment of a local control strategy. In Section 4 the physical limitations of active noise control will then be linked to the limitations on the generation of localised audio sound zones within the car cabin and, once again, this will be achieved using the results from a practical sound zone generation system.

2 Global Active Noise Control Global active control of noise in an enclosure, such as a car cabin, aims to minimise the sound pressure level throughout the enclosure. This is usually achieved in theory by minimising the total acoustic potential energy within the enclosure, although in practice this has to be approximated by minimising the sum of the squared pressures measured at a number of error microphone locations. If the control system is implemented as a feedforward controller, as shown in Fig. 1, and consists of L error microphones located in the enclosure, M control loudspeakers and K reference signals, the vector of error signals, e, at a single frequency can be expressed as

e d  Gu  d  GWx

(1)

where d is the vector of disturbance signals measured at the error microphones without control, G is the (L×M) matrix of transfer responses between the inputs to the control loudspeakers and the pressures measured at the error microphones, u is the vector of M control signals which drive the control loudspeakers, x is the vector of reference signals and W is the control filter matrix. In general, global active control aims to minimise the sum of the squared error signals, which will approximate the total acoustic potential energy in the enclosure provided that a sufficient number of microphones are employed. The cost function that should be minimised in this case is given at a single frequency as

J  eH e uH G H Gu uH G H d  dH Gu dH d.

(2)

The optimal vector of control signals which minimises this cost function is then given in the overdetermined case by [5] 1

uopt   G H G  G H d.

(3)

It can be seen from eq. (3) that this optimal solution to the multi-input multi-output active noise control problem assumes that disturbance signals are known in advance. In the automotive environment this is unlikely to be the case and instead it is necessary to implement an adaptive algorithm, which iteratively converges towards the optimal solution given by eq. (3). For example, this can be achieved using the filtered-x LMS algorithm [14]. However, eq. (3) is useful to determine the physical limitations on active noise control in the automotive environment without considering the additional complications due to the use of adaptive algorithms.

3

Sound field control in the automotive environment

Figure 1 – Multichannel feedforward controller.

In order to understand the physical limitations on global active noise control, the interior automotive noise control problem will be represented by the control of the noise produced by a single primary acoustic source located in a rigid walled rectangular enclosure with similar dimensions to a small car interior. Fig. 2 shows the total acoustic potential energy in the rectangular enclosure before control (thick black line) and after control using either a single control source located in one corner of the enclosure (red dot-dashed line) or eight control sources located in the corners of the enclosure (blue dashed line). From this plot it can be seen that with a single control source the system is able to achieve significant attenuation of the compliant mode and the first longitudinal enclosure mode at 72 Hz. Control at higher frequencies, however, is rather limited and this is due to the increasing number of acoustic modes that are excited in a three dimensional enclosure as the frequency of excitation is increased. From the blue dashed line in Fig. 2 it can be seen that when 8 control sources located in the corners of the enclosure are used to minimise the acoustic potential energy, global attenuations are achieved up to around 240 Hz. This increase in the control performance is because the control sources are able to couple into, and therefore control more modes of the enclosure. However, the number of control sources required rapidly increases to an impractical level, since the number of acoustic modes in a three-dimensional enclosure approximately increases with the cube of the excitation frequency.

4

Sound field control in the automotive environment

Total acoustic potential energy, dB

90 95 100 105 110

Primary source alone Single secondary source

115 120

Eight secondary sources 0

100

200

300 400 500 Frequency, Hz

600

700

Figure 2 – The total acoustic potential energy in a three-dimensional rectangular enclosure before control (thick black line) and after control using a single secondary control source located in one corner (red dot-dashed line) and eight secondary control sources located in the corners of the enclosure (blue dashed line).

2.1 Active Control of Engine Noise Feedforward control of engine noise was first demonstrated in the late 1980s [11] and has since been used in various configurations to control the increase in noise level due to lightweight vehicle design [10], to reduce the variation in the engine noise characteristic due to the use of economical engine designs such as variable displacement [15], and to improve the perceived sound quality of the engine noise [16]. A cost effective feedforward engine noise control system can be implemented using an engine speed reference sensor, low-cost microphone error sensors, the car audio loudspeakers and their amplifiers as control sources and the car audio digital signal processing (DSP) capabilities. Such active engine noise control systems have been implemented by a number of manufacturers.

5

Sound field control in the automotive environment In order to demonstrate the performance limitations of an active engine noise control system, Fig. 3 shows the attenuation achieved by a practical system implementation applied to a small city car with a 2-cylinder engine [17]. The control system in this case employs 8 error microphones mounted on the seat headrests, the four low-frequency car audio loudspeakers as control sources, a reference signal obtained from a tachometer and a filtered-x LMS adaptation algorithm. The results shown in Figure 3 show the attenuation in the sum of the squared pressures measured at 16 microphone locations at the frequency of the first engine order during an engine run-up. From these results it can be seen that, in general, the level of control decreases with increasing frequency and, as shown in the previous section, this is due to the increasing number of acoustic modes excited in the car cabin.

1st Order Attenuation, dB

20

15

10

5

0 1500 2000 2500 3000 3500 4000 4500 5000 Engine speed, rpm Figure 3 – The attenuation in the sum of the squared pressures measured at 16 microphone locations at the first engine order during an engine run-up.

6

Sound field control in the automotive environment

2.2 Active Control of Road Noise Although engine noise control has seen a number of commercial implementations, road noise control has been less commercially successful. This is due to the higher costs and challenges of implementation. However, active road noise control is an important technology to allow lightweight vehicle design, which tends to result in an increase in the low frequency broadband interior noise due to the interaction of the tires and the road. Feedforward active noise control systems have previously been developed to reduce the noise levels in the car cabin using reference signals obtained by direct measurement of the vibration due to road excitation [12,18,19]. Although these feedforward road noise control systems could again be implemented using low-cost microphone error sensors, the car audio loudspeakers as control sources and the car audio DSP capabilities, it is also necessary to employ at least six accelerometers mounted to the vehicle's suspension and bodywork to obtain reference signals with sufficient coherence with the error signals in order to obtain reasonable levels of control [12,19]. In order to overcome the high cost associated with employing a number of accelerometer reference signals, road noise control systems employing only microphone sensors have been investigated [20, 21]. One approach in this case is to employ headrest mounted error microphones in a feedback control configuration [20]. However, a higher level of performance has been achieved by employing both headrest mounted error microphones and floor mounted microphones to provide a time-advanced signal to the controller, which can be implemented using either a feedforward or internal model control feedback structure [21]. Fig. 4 shows the performance of this road noise control system implemented in a small city car using the four low-frequency car audio loudspeakers, four headrest error microphones and four additional microphone sensors mounted on the floor of the cabin. Fig. 4 shows the sum of the squared pressures measured at the headrest error microphones and it can be seen that the broadband peak in the road noise between 80 and 200 Hz has been attenuated to a similar level by both the feedforward and feedback controller implementations and a maximum attenuation of 8 dB has been achieved at 115 Hz. At higher frequencies it can be seen from these results that the performance of the road noise controller is again rather limited and this can again be linked to the increase in the number of acoustic enclosure modes that are being excited. However, in the feedforward controller architecture there is also a limitation on the control bandwidth due to the low multiple-coherence between the floor microphone signals and the headrest microphone signals at higher frequencies. And, in the case of the feedback implementation, there is a bandwidth limitation due to the delay in the response between the control loudspeakers and the headrest error microphones.

7

Sound field control in the automotive environment

J, dB re. arbitrary reference

0 5 10 15 Uncontrolled FB Controlled FF Controlled

20 25

0

100

300 200 Frequency, Hz

400

500

Figure 4 – The sum of the squared pressures measured by the headrest error microphones while the small city car is driven at 50 km/h on a pave road surface. The lines show the results before control (thick black line), after control using the optimised feedforward (thin black line) and feedback (thin light line) controllers. The results are plotted in decibels with respect to an arbitrary reference level.

3 Local Active Noise Control In the previous section it has been shown that the bandwidth of global active control in a car cabin sized enclosure is physically limited by the rapid increase in the number of acoustic modes with increasing frequency. Additionally, it has been highlighted that, although a practical feedback road noise control system overcomes some limitations of feedforward implementations in terms of coherence between reference and error signals, it is limited by the delay in the plant response. In order to attempt to overcome these limitations and extend the high frequency limit of active noise control, there has been interest in using local active noise control to create zones of quiet using loudspeakers in the headrest, for example [22, 23]. The performance of such a local control system de-

8

Sound field control in the automotive environment pends on the local geometry of the head and headrest [24], but also on the spatial properties of the primary sound field [25, 26]. Previous work has conducted investigations of the effect of these properties on local active control through calculating the performance for multiple realisations of the primary sound field and averaging the results [24, 27, 28], however, recently a closed-form solution for the expectation of the controlled sound field has been presented [25]. In the following section this formulation will be used to demonstrate the influence of the primary sound field on the performance of a simple local active control system. Subsequently, some experiments are described in a car, in which the spatial properties of the road noise were measured at an array of microphones. This data is then used to predict the levels of active control that can be achieved with two loudspeakers mounted in the headrest.

3.1 Freefield Simulations of Local Active Control In order to investigate the effects of the spatial properties of the primary sound field on a local active control system using the formulation presented in [25], we initially consider the case of a single-input single-output control system located in the freefield. The control system consists of a single control source located at (L, 0) and an error sensor located at (0, 0), as shown in Fig. 5. The influence of the spatial distribution of the primary field on the shape of the 10 dB zone of quiet is illustrated in Fig. 5 at a normalized excitation frequency of kL equal to 0.5, where k is the wavenumber. The shape of the zone of quiet for the simulations of the diffuse sound field, which was achieved with 408 uncorrelated sources forming a sphere in the far field, is the same as that in [27] where the results have been calculated by averaging over multiple random primary sound fields. Also shown in this figure is the shape of this zone when only 21 uncorrelated primary sources are operating, either above or to the right or to the left of the quiet zone. The zone of quiet is greatest when the primary field is mainly from above, since in this case the primary pressure field is almost uniform in the plane shown in Fig. 5, so that reductions at the control point will result in similar reductions at all positions which are a similar distance from the secondary source. When the primary field is mainly from the right, the zone of quiet is somewhat broader in the x-direction than that achieved with a diffuse primary field, since the phase variation of the primary field then more nearly matches that of the secondary field in this direction. Conversely, when the primary field is mainly from the left, the phase variations of the primary and secondary field match more accurately in the y-direction, and so the extent of the quiet zone has been extended in this direction.

9

Sound field control in the automotive environment kL=0.5 2 1.5 Diffuse

1

From above

y/L

0.5

From right From left

0

Control Sources

0.5

Control Sensors

1 1.5 2

1

0

1 x/L

2

3

Figure 5 – The extent of the 10 dB zone of quiet, for a SISO local control system at a normalized excitation frequency of kL=0.5, when the simulation field is diffuse (thin solid black line), mainly coming from above (thick solid black line), mainly from the right hand side (thin dash grey line), and mainly from the left hand side (thick solid grey line).

3.2 Local Active Control of Road Noise in a Car To evaluate the potential performance of a local active noise control system in an automotive environment, a series of measurements have been conducted in a Ford S-Max [26, 29]. The experimental setup consisted of an array of 25 microphones mounted on a 0.4 by 0.4 m grid positioned in front of the headrest on the front passenger seat and two control loudspeakers mounted in the headrest, as shown in Figure 6. The microphone array comprised 12 microphones in an upper grid and 13 in a lower grid, which were separated by 75 mm. Initially, the transfer responses between the voltage inputs to the control loudspeakers and the pressures measured at the microphone grid positions have been measured whilst the vehicle was stationary. The pressures were then measured at the microphones when the vehicle was driven at a variety of speeds on different road surfaces [29].

10

Sound field control in the automotive environment

Figure 6 – The geometry of the microphone array, its installation on the front passenger seat and the headrest containing the two control loudspeakers.

Using the in-car measurements the active control performance has been calculated using the theory presented in [25] and assuming perfect reference signals. Figure 7 shows the performance of the local control system when the car was driven at 50 mph on a relatively smooth road and the two control loudspeakers have been used to minimize the pressures measured at four microphones in the second row of the microphone array, which approximately correspond to locations of the occupant’s ears. The black line in

11

Sound field control in the automotive environment Figure 7 shows the A-weight power spectrum averaged over all 25 microphones before control and the red dot-dashed line shows the result of control. From these results it can be seen that attenuation over the full microphone array is achieved up to around 300 Hz, which is an increase from the typical 200 Hz limit for global active control in a car. However, also shown in Figure 7 is the average A-weighted power spectral density at the 4 error microphones located close to the expected ear positions of the occupant before and after control. In this case it can be seen that significant attenuation is achieved up to 500 Hz and this clearly demonstrates the potential performance increase that may be achieved by focusing the active control at the occupant’s ears.

50

See, dBA ref. 20Pa

45 40 35 30 25 20

200

400 600 Frequency, Hz

800

1000

Figure 7 – The A-weighted averaged power spectral density before and after control of road noise using the local control system with two headrest loudspeakers and 4 error microphones for a 50 mph smooth road driving condition. No control averaged over 25 mics (black line); control averaged over 25 mics (red dot-dashed line); no control averaged over 4 mics (grey dashed line); control averaged over 4 mics (blue dashed line).

12

Sound field control in the automotive environment

4 Sound Field Control for Audio Zone Generation In addition to the use of sound field control to attenuate unwanted automotive noise, it can also be used in the automotive environment to enhance the functionality of the audio system, as discussed in the introduction. One particular area of interest is the generation of independent listening zones in the car cabin. This problem has been approached from two principle directions including directional loudspeakers [30, 31] and sound field control [13, 32, 33]. A commercial system has recently been presented which relies on both of these operating principles [34], however, these ideas were first presented in [13]. In the following section a practical sound zone control system is described, which utilises both active sound control principles, based on controlled constructive and destructive interference between sources, and directional loudspeakers. The performance of this system is assessed in terms of the generation of independent listening zones in the car cabin and the physical limitations are discussed.

4.1 Practical Sound Zone Control System The generation of independent listening zones in a car cabin would ideally be achieved using the standard car audio loudspeakers. However, similarly to the limits on achieving active control presented in the previous sections, it has been shown that using the standard car audio loudspeakers can only achieve separation between listening zones at low frequencies [13, 32]. Therefore, similarly to the implementation of local active noise control systems, the standard car audio system must be supplemented with loudspeakers mounted in closer proximity to the desired listening zones. In the presented practical example the car cabin personal audio system installed in a Ford S-Max consists of the four standard low-frequency car audio loudspeakers and an additional array of eight phase-shift loudspeakers, with one loudspeaker mounted on each side of each headrest, as shown in Fig. 8. The individual phase-shift loudspeakers have been designed to achieve a hypercardioid directivity, as detailed in [35].

13

Sound field control in the automotive environment

Figure 8 – The headrest mounted loudspeakers used in the practical car cabin personal audio system [13].

In addition to the design of the physical loudspeaker array, the implementation of personal audio systems requires the optimisation of the signals driving the loudspeakers. In practice this requires a bank of filters to be designed which adjust the magnitude and phase of the audio signals being sent to the loudspeakers. A number of different methods of implementing this filter optimisation process have been proposed [36-38], however, here we will use the least squares design method due to its ability to produce high quality audio [13, 37]. The performance of personal sound zone systems is generally characterised by the difference in reproduced sound level between the independent listening zones. In the considered car cabin application the system aims to produce one listening zone in the front seats and one listening zone in the rear seats. The ability of the sound zone system to produce these independent listening zones can be assessed using the acoustic contrast [36], which is given by the ratio between the sum of the squared pressures measured in the listening zone to the sum of the squared pressures measured in the quiet zone. Fig. 9 shows the acoustic contrast for the implemented car cabin sound zone control system when a listening zone is produced in the front seats Fig. 9a and in the rear seats Fig. 9b. The performance is shown for the standard car audio loudspeakers (thin lines) and for the headrest loudspeaker array (thick lines) when it is measured at 4 microphones at each headrest. From these results it can be seen that in both configurations the car audio loudspeakers are able to achieve significant acoustic contrast up to around 200 Hz, which is similar to the limit on global active control in the automotive environment discussed in Section 2. At higher frequencies the car audio loudspeakers are not able to achieve significant separation between the two listening zones, due to the rapid increase

14

Sound field control in the automotive environment in the number of acoustic modes excited within the enclosure. At higher frequencies it can be seen from Fig. 9 that the headrest loudspeaker arrays are able to improve the acoustic contrast, although there are some limitations when generating a rear listening zone due to the fixed directivity of the individual headrest loudspeakers and there orientations relative to the listening zones [13]. Nevertheless, the increase in the performance bandwidth is comparable to that achieved in the context of local active noise control, as discussed in Section 3.

Figure 9 – The acoustic contrast between the listening and quiet zones achieved by the car cabin sound zone control system when producing (a) a listening zone in the front of the car and (b) when producing a listening zone in the rear of the car. The thick lines show the performance achieved by the headrest loudspeaker array and the thin lines show the performance achieved by the standard car audio loudspeakers.

5 Conclusions The use of sound field control in the automotive environment offers the potential to improve the acoustic experience of the vehicle’s occupants. This can include both refinement of the vehicle NVH through active noise control and additional functionality in the car audio system. This paper has presented a review of the limitations of sound field control within the automotive environment in the context of both active control of engine and road noise, and the generation of independent audio listening zones. In both of these areas, sound field control using the standard car audio loudspeakers is limited to frequencies below around 200 Hz due to the rapid increase in the number of acoustic modes with frequency. In both applications of sound field control it is necessary to resort to local control systems, with the control loudspeakers located in close proximity to the listening positions, in order increase the upper frequency at which control can be achieved.

15

Sound field control in the automotive environment

References [1]

F. Rumsey, Sound Field Control, J. Audio Eng. Soc., 61(12), December 2013.

[2]

F. Rumsey, Spatial Audio, Focal Press, Oxford, 2001.

[3]

T. Betlehem, W. Zhang; M.A. Poletti, T.D. Abhayapala, Personal Sound Zones: Delivering interface-free audio to multiple listeners, IEEE Signal Process. Magazine, 32(2), pp.81-91, 2015.

[4]

H. Zhu, R. Rajamani, and K. A. Stelson. Active control of acoustic reflection, absorption, and transmission using thin panel speakers. J. Acoust. Soc. A., 113(2), pp.852-870, 2003.

[5]

P.A. Nelson and S.J. Elliott, Active control of sound, Academic Press, 1991.

[6]

F. Fazi, P. Nelson, J.E. Christensen and J. Seo, Surround system based on threedimensional sound field reconstruction. In Proc. of Audio Eng. Soc. Convention 125, 2008.

[7]

M.J. Smithers, Improved stereo imaging in automobiles, in Proc. of Audio Eng. Society Convention 123, 2007.

[8]

M.R. Bai, and C.-C. Lee, Comparative study of design and implementation strategies of automotive virtual surround audio systems. J. Audio Eng. Soc., 58(3), pp.141-159, 2010.

[9]

S. Brix, C. Sladeczek, A. Franck, A. Zhykhar, C. Clausen and P. Gleim, Wave field synthesis based concept car for high-quality automotive sound. In Proc. of 48th Audio Eng. Soc. Conf., 2012.

[10] S.J. Elliott, Active noise and vibration control in vehicles, in: X. Wang (Ed.), Vehicle Noise and Vibration Refinement, Woodhead Publishing, Cambridge, 2010, pp. 235–251. [11] S.J. Elliott, I.M. Stothers, P. Nelson, M.A. McDonald, D.C. Quinn, T.J. Saunders, The active control of engine noise inside cars, in Proc. of Inter-noise 88, 2, pp. 987–990, 1988. [12] T.J. Sutton, S.J. Elliott, M.A. McDonald, T.J. Saunders, Active control of road noise inside vehicles, J. Noise Control Eng., 42, pp.137–146, 1994. [13] J. Cheer S.J. Elliott and M.F. Simón Gálvez, Design and implementation of a car cabin personal audio system, J. Audio Eng. Soc., 62(6), pp. 412-424, 2013. [14] S.J. Elliott, I. Stothers, and Philip A. Nelson. A multiple error LMS algorithm and its application to the active control of sound and vibration, IEEE Trans. Acoust., Speech, Signal Process., 35(10), pp.1423-1434, 1987. [15] R. Schirmacher, R. Kunkel, M. Burghardt, Active Noise Control for the 4.0 TFSI with Cylinder on Demand Technology in Audi's S-Series, Society of Automotive Engineers, Graz, no. 2012-01-1533, 2012.

16

Sound field control in the automotive environment [16] Y. Kobayashi, T. Inoue, H. Sano, A. Takahashi, K. Sakamoto, Active sound control in automobiles, in Proc. of the Inter-noise, 2008. [17] J. Cheer, Active control of the acoustic environment in an automobile cabin, PhD Thesis, University of Southampton, U.K., 2012. [18] R. Bernhard, Active control of road noise inside automobiles, in Proc. of ACTIVE 95, pp. 21–32, 1995. [19] S.-H. Oh, H.-S. Kim, Y. Park, Active control of road booming noise in automotive interiors, J. Acoust. Soc. Am., 111(1), 180–188, 2002. [20] J. Cheer and S.J. Elliott, The design and performance of feedback controllers for the attenuation of road noise in vehicles, Int. J. Acoust. Vib., 19(3), 2014. [21] J. Cheer and S.J. Elliott, Multichannel control systems for the attenuation of interior road noise in vehicles, Mech. Sys. Signal Process., 60-61, pp.753-769, 2015. [22] B. Rafaely, S.J. Elliott, and J. Garcia-Bonito, Broadband performance of an active headrest, J. Acoustic. Soc. Am., 106(2), pp.787-793, 1999. [23] M. Pawelczyk, Adaptive noise control algorithms for active headrest system, Control Engineering Practice, 12(9), pp.1101-1112, 2004. [24] J. Garcia-Bonito, S.J. Elliott, and C.C. Boucher,. Generation of zones of quiet using a virtual microphone arrangement. J. Acoust. Soc. Am., 101(6), 3498-3516, 1997. [25] S.J. Elliott and J. Cheer, Modelling local active sound control with remote sensors in spatially random pressure field, J. Acoust. Soc. Am., 137(4), 2015. [26] S.J. Elliott, W. Jung and J. Cheer, The spatial properties and local active control of road noise, in Proc. of Euro-noise, 2015. [27] P. Joseph, S. J. Elliott, and P. A. Nelson, Near field zones of quiet, J. Sound Vib. 172, 605–627, 1994. [28] J. Garcia-Bonito and S. J. Elliott, “Local active control of diffracted diffuse sound fields,” J. Acoust. Soc. Am. 98(2), 1017–1024, 1995. [29] W. Jung, Local active sound control with remote loudspeakers and microphones, Master’s thesis, University of Southampton, U.K., 2014. [30] F. P. Thigpen, “Vehicle audio system with directional sound and reflected audio imaging for creating a personal sound stage.” United States Patent, US 2004/0109575 A1, June 2004. [31] S. Goose and F. Arman, “System and method for creating personalized sound zones.” United States Patent, US 2006/0262935, November 2006. [32] S. Berthilsson, A. Barkefors, L.-J. Brännmark, M. Sternad, Acoustical Zone Reproduction for Car Interiors Using a MIMO MSE Framework, in Proc. of 48th International Audio Eng. Soc. Conf., 2012. [33] W.F. Druyvestseyn and J. Garas, Personal Sound, J. Audio Eng. Soc., 45(9), pp.685-701, 1997.

17

Sound field control in the automotive environment [34] D. Shewchuk, “Harman unleashes new potential for in-car listening experiences and sound management”, http://news.harman.com/Press-Releases/HARMANUnleashes-New-Potential-for-In-Car-Listening-Experiences-and-SoundManagement-212.aspx, January 2015, (accessed 06/04/2015). [35] M.F. Simón Gálvez, Loudspeaker Arrays for Family Television, Master’s thesis, University of Southampton, UK, 2011. [36] J.-W. Choi and Y.-H. Kim, Generation of an Acoustically Bright Zone with an Illuminated Region Using Multiple Sources, J. Acous. Soc. Am., 111, pp.1695– 1700, 2002. [37] M.F. Simón Gálvez, S. J. Elliott, and J. Cheer, A superdirective array of phase shift sources. J. Acous. Soc. Am., 132(2), pp.746-756, 2012. [38] P. Coleman, P.J. Jackson, M. Olik, and J.A. Pedersen, Personal audio with a planar bright zone, J. Acoust. Soc. Am., 136(4), 1725-1735, 2014.

18

Importance of the evaluation of structure-borne NVH performance for lightweight trim design Dr. Théophane Courtois, Head Products and Systems Simulation Team, M. Seppi, F. Ronzio, L. Sangiuliano, Autoneum Management AG, Switzerland; T. Yano, Nihon Tokushu Toryo Co., Ltd, Japan.

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 W. Siebenpfeiffer (Hrsg.), Automotive Acoustics Conference 2015, Proceedings, https://doi.org/10.1007/978-3-658-27648-5_10

1

Importance of the evaluation of structure-borne NVH performance for lightweight …

Abstract The design of automotive parts for the control of interior NVH involves the analysis of the noise contribution from interior carpet and dash insulator systems. The interior trim is usually designed with respect to insulation and absorption performance, for the control of air-borne noise that is commonly admitted to dominate the overall vehicle NVH for frequencies above approximately 800 Hz. Below this frequency range, it can be shown that the interior noise level is more dominated by structure borne phenomena that are transmitted through the vibration of the body structure, and for which the influence of the insulators is usually considered in a simplified way. In this low frequency range, it is preferred to exploit stiffness and damping countermeasures that act directly on the panel vibration, assuming that the vibration reduction will result in interior noise reduction. From this stand point, it sounds natural to privilege the selection of insulators, which on top of their air-borne primary function can also influence the panel vibration reduction at low and medium frequencies by means of added stiffness and/or damping. This is driving to favour stiffer, thus heavier porous decouplers. In this paper, we first of all show by means of FE simulations on simple flat samples the influence of decouplers on the panel vibration. Concurrently, it can be shown that the interior structure borne noise performance – that is the relevant customer annoyance – is also directly influenced by the insulator transmissibility. Due to the spring-mass nature of insulators, this may lead to the selection of softer and therefore lighter decouplers. Therefore, in the context of light weight insulator products, it is crucial to demonstrate the influence of material selection on the transmission of structure borne noise from the structural vibration to the interior acoustics. In this paper, the flat sample test case is extended to a vibro-acoustical plate-box case. By means of FEM simulation, we show the relation between the decoupler mechanical properties and the final structure borne noise performance. The results on flat samples offer an overview of different technical solutions, including felt, light foam or viscoelastic foam. The validation on a plate-box test rig confirms the simulated tendencies. At the same time, due to the manufacturing limitations of the decoupler materials, it is shown that the mechanical properties driving the final performance can come into conflict with other functional requirements, such as the density, or static compression. For light weight technologies, a proper analysis leads to comprehensive strategies for the design of 3D parts. In this article, we propose a particular focus on injected fibre technologies, where the simulation can be used to find the best material density selection versus thickness, with a final demonstration on a vehicle floor application.

2

Importance of the evaluation of structure-borne NVH performance for lightweight …

1 Introduction The design of automotive parts for the control of interior NVH involves the analysis of the noise contribution from interior carpet and dash insulator systems. The interior trim is usually designed with respect to insulation and absorption performance, for the control of air-borne noise that is commonly admitted to dominate the overall vehicle NVH for frequencies above approximately 800 Hz [1, 2]. Below this frequency range, it can be shown that the interior noise level is more dominated by structure borne phenomena that are transmitted through the vibration of the body structure [3], for which the influence of the insulators is usually considered in a simplified way. Several studies conducted by the Autoneum Benchmarking have already demonstrated the influence of the trim on structure borne performance [4]. The influence of the trim becomes relevant from 150 Hz upward, and therefore it should not be underestimated [5]. Beside this, developments of simulation methods based on Finite Element Method [6] for the prediction of the effect of the trim, including porous materials, have been an important area of research since the last decade [7-12]. Those methods are still not used in a systematic way to support the structure borne design of trim parts due to a lack of definition of what are the necessary requirements for the structure-borne performance of the trim. Indeed in this low frequency range, it is common to address the structure borne performance, by exploiting stiffness and damping countermeasures that act directly on the panel vibration, assuming that the vibration reduction will result in interior noise reduction [13]. The relation from the body vibration and body stiffness to the acoustics is sometimes tackled in the automotive, from experimental point of view [14], or by considering the radiation efficiency [15-18], but without including explicitly the acoustic as a design measurable. Therefore, from this stand point concentrating on the body vibration, it sounds natural to privilege the selection of insulators, which on top of their airborne primary function can also influence the panel vibration reduction at low and medium frequencies by means of added stiffness and/or damping [19]. More lately, in the context of the evaluation of light weight trim for the structure borne, Autoneum started to exploit in a more systematic way the structure-borne SPL prediction by means of FEM trim simulation tools [20]. In this paper, we exploit such a simulation method to demonstrate the importance of the structure borne evaluation to support an enlightened material decision for insulator. We first of all present a numerical study carried out on a flat plate, showing the influence of decouplers on panel vibration. Then the numerical study is extended to a vibroacoustical plate-box case, showing how the interior structure borne noise performance is also influenced by the insulator transmissibility. In the context of light weight insulator products, the material properties are indeed crucial for the transmission of structure borne noise from the structural vibration to the interior acoustics. The results on flat samples offer an overview on the influence of the decoupler material properties on the

3

Importance of the evaluation of structure-borne NVH performance for lightweight … acoustic. Then we present an experimental study carried out on a plate-box test rig and flat samples, thus confirming and explaining the simulated tendencies. At the same time, due to the manufacturing limitations of the decoupler materials, it is shown that the mechanical properties driving the final performance can come into conflict with the corresponding technological constraints or other functional requirements, such as the density, or static compression. For light weight technologies, a proper analysis leads to comprehensive strategies for the design of 3D parts. A final demonstration is presented at the end of the article on a realistic vehicle floor design application, where the simulation can be used to find the best material density selection, taking into account the available design package size. In this article, we propose a particular focus on a certain type of felt decoupler based on an injected fibre technology, which is offering to the material a rather low density and stiffness. One particular property of this technology, in comparison to other “air-laid” felt solutions, is the homogeneity of the density of the material that is more or less independent of the thickness of the decoupler. In this article, we will refer to this type of felt with the acronym IFP. The corresponding density of the material will be represented by a number after the acronym in kg/m3 (for example IFP40, for an IFP of density 40 kg/m3).

2 Numerical study on the influence of decoupler on flat structure vibration As introduced before, the structure borne performance of a component can be related in a first approach to the level of vibration of its metal structure. In this paragraph, we propose to evaluate the influence, of the trim insulator that lies on the structure, on the vibration level of the body structure. This assessment is performed by means of FE simulation on a simple steel plate, covered by flat samples. After a short description of the numerical set-up and of the simulation method, we will show the calculated vibration results of the plate covered by different insulation treatments involving either foam or felt, against the vibration of the plate when bare, or simply damped by means of damping pads.

2.1 Simulation model and method For this purpose, a flat steel plate having dimensions 0.7 m x 0.8 m and a thickness of 1.25 mm constitutes the simple body structure representation to be simulated, and modelled with PSHELL elements, as shown by Figure 1. The plate has clamping boundary conditions on its contour. It is worth remarking that, while the configuration of the system is conceptually simple, the dimensions of the plate lead to a relatively high number

4

Importance of the evaluation of structure-borne NVH performance for lightweight … of modes (approximately 50 structural modes, up to 500 Hz). The plate is excited by a unit force at one point situated on one corner area, at a dissymmetric location, and the result of interest is the quadratic average (rms) of the calculated velocity on the entire area the plate as frequency response to the normalized excitation. The FE simulation of the frequency response function is performed in NASTRAN by using a SOL111.

F

Figure 1: FE model of the plate. Excitation (force) and response (velocity) location.

When the plate is represented covered by damping pads, the problem is solved in the same way, but the plate is modelled by means of equivalent PSHELL elements, which properties correspond to the composite steel covered by damping. The bending and membrane equivalent properties of such a composite can be obtained by a numerical method [21] that is searching for bending equivalence or simply in a more approximate way by using the Oberst theory [22]. When the plate is covered by an acoustic trim (Figure 2), the simulation is performed by means of a FE integration of 3D poro-elastic elements in NASTRAN. This element is based on a formulation of the Biot theory [23] with explicit pressure and displacement variables [24] and has been implemented and validated by Autoneum. The method relies on a pre-computation of the surface boundary impedance of the trim FE model, which is subsequently integrated into a classical fluid-structure SOL111 solution of NASTRAN. The method is presented and detailed in [25]. The application of such a trim FE simulation requires an appropriate 3D mesh that must respect the degree of refinement necessary for the representation of the relevant wavelengths travelling in the porous material. This is important for representing correctly both the effect that the trim has on the vibration reduction of the plate, and the effect that the trim has on the acoustic pressure, when the plate is coupled to an acoustic cavity (as we will see in a next part of the article).

5

Importance of the evaluation of structure-borne NVH performance for lightweight …

Figure 2: FE model of the plate covered by damping and trim

2.2 Calculated treatments, results We compare the vibration reduction of the plate, when treated alternatively by damping, foam insulation and then felt insulation treatment. The damping treatment consists here in bitumen layer of 1.7 mm thickness and corresponds to the state of the art in terms of weight and performance-effective material solution, commonly used on automotive bodies, with an area weight of 2.5 kg/m2 and a damping loss factor of 0.43. The insulation trim has been chosen, so that main focus is given to the effect of the decoupler. The decoupler is flat with 20 mm thickness, and covered by a 2 kg/m2 mass layer. The mass layer is important in order to provide to the trim the fundamental spring mass behaviour of classical insulation solutions in the automotive. The Table 1 summarizes the material properties used for the simulation. The felt has been chosen with material properties corresponding to a soft injected fibre type solution (IFP). This material has the advantage to feature a controllable density, with an associated dynamic stiffness. The lowest densities (like for IFP40) correspond to rather low stiffness. Table 1: material properties of insulation material for damping comparison. The term “IFP” is used for injected fibre solution. Thickness [mm]

Density [kg/m3]

Stiffness [kPa]

Loss factor [-]

Foam

20

87

73

0.2

IFP40

20

40

12.5

0.1

IFP80

20

80

20

0.1

IFP140

20

140

31

0.1

6

Importance of the evaluation of structure-borne NVH performance for lightweight …

2.2.1 Vibration reduction by means of damping The Figure 3 shows the plate vibration response when treated with different damping strategies, against the vibration of the plate when totally bare (untreated). It is worth observing that when optimally treated with damping (0.8 kg material), the vibration level is close to the level with full damping coverage (1.4 kg), but definitely lighter. On the other hand, when not optimized, or let’s say not properly designed, a damping solution with similar weight of 0.8 kg can lead to vibration levels that are much worse than those with the full coverage treatment or with a well-designed one. -20

RMS Velocity [dB]

-30 -40 -50 -60

Bare plate Damped plate (full coverage = 1.4kg)

-70

Damped plate (optimized coverage = 0.8kg) Damped plate (non optimal design = 0.8kg)

-80 0

100

200

300

400 500 600 Frequency [Hz]

700

800

900

1000

Figure 3: plate rms velocity with different damping treatment layouts

2.2.2 Vibration reduction due to the effect of acoustic treatment The Figure 4 shows the vibration level of the plate when covered by insulation treatment. The effect of the trim on the plate vibration is compared to those obtained with damping treatment. Acoustic treatments have a damping effect on the plate, but with different respective effects. The foam shows a better damping effect than the light felt (IFP40). However the results must be considered together with the respective weight of each solution, where the foam solution is the heaviest, and the damping solution the lightest. From this point of view, the damping remains the most weight effective solution, when one wants only to damp the structure. In this example, the foam remains better than the damping between 150 and 500 Hz and the conclusion may be mitigated.

7

Importance of the evaluation of structure-borne NVH performance for lightweight … The Figure 5 shows additionally the vibration level of the plate when covered by insulation treatment but with also the damping treatment. In this case, we are using the “optimal coverage” damping layout. It is interesting to see that the vibration level of the structure with respectively felt and foam treatment decreases due to presence of damping material more in the case of felt than in the case of foam. Differently said, the light felt solution is naturally more sensitive to damping treatment than the foam solution. Moreover, it is worth stressing out that the felt with damping solution with 2.3 kg has the same weight order of magnitude than the foam solution without damping with 2.1 kg, and a vibration performance that is better on a rather large frequency range. But still, the foam remains better between 150 and 450 Hz for the vibration performance. -20

RMS Velocity [dB]

-30 -40 -50 -60

Bare plate Damped plate (optimized coverage = 0.8kg)

-70

Foam treated plate ( = 2.1kg) Light felt treated plate (= 1.5kg)

-80 0

100

200

300

400 500 600 Frequency [Hz]

700

800

Figure 4: plate rms velocity with different insulation treatments (20 mm thick)

8

900

1000

Importance of the evaluation of structure-borne NVH performance for lightweight … -40

RMS Velocity [dB]

-45 -50 -55 Damped plate (optimized coverage = 0.8kg)

-60

Foam treated plate ( = 2.1kg) Light felt treated plate (= 1.5kg)

-65

Foam and damping treated plate (= 2.9kg) Light felt and damping treated plate (= 2.3kg)

-70 0

100

200

300

400 500 600 Frequency [Hz]

700

800

900

1000

Figure 5: plate rms velocity with different insulation treatment layouts, on top of an optimally damped structure

2.2.3 Influence of the decoupler properties on the vibration performance The performance that the trim has on the vibration of the plate influences the selection of the insulation solutions, and may lead to the selection of stiffer material. The Figure 6 shows the vibration level of the plate with a variation of different felt densities. The Table 1 summarizes the stiffness values of the different fibre densities. We can see that heavier is the felt solution, and better is the vibration performance. From this example, we can see that focus on the structural vibration might lead to the selection of a heavier or stiffer insulation solution.

9

Importance of the evaluation of structure-borne NVH performance for lightweight … -40

RMS Velocity [dB]

-45 -50 -55 IFP40

-60

IFP 80 IFP 140

-65

Foam -70 0

100

200

300

400 500 600 Frequency [Hz]

700

800

900

1000

Figure 6: plate rms velocity with 20 mm thick felt insulation treatment, with variation of different felt densities

3 Numerical study on the influence of decoupler on the acoustic performance of a flat structure In the previous paragraph, it was shown on a virtual flat plate case how the selection of the insulation treatment can influence the vibration of the base structure (the body structure). From this stand point, it was shown that the most weight- and performanceeffective solution resides in the usage of a well-designed damping treatment. However, while a damping treatment reduces the panel vibration and is expected to reduce somehow proportionally the radiated noise level, the insulation treatment has a much more complex relation to the radiated noise. Due to its spring mass nature, the relation from the panel vibration reduction up to the noise radiated by the insulator is not systematically translating the vibration reduction into a reduction of noise. In some cases, it can even be the opposite. Therefore we show in this paragraph by means of FE simulation the concurrent influence that the different treatment solutions have on the acoustical performance, under structure borne excitation.

3.1 Simulation model and method For this purpose, the same flat plate FE model is coupled to an acoustic domain, made of a rectangular box-shaped cavity of 0.7 m x 0.8 m x 1.6 m size. In this fluid structure system, the plate is the only flexible wall of the box cavity, and all the other walls are rigid. While the system is conceptually simple, the plate and cavity size lead to a rela-

10

Importance of the evaluation of structure-borne NVH performance for lightweight … tively high number of modes, making the vibro-acoustic behavior of the system rather complex. The system FE model has 85700 degrees of freedom and 18057 elements. The Figure 7 shows the FE model of the plate-box. The acoustic domain is modelled with 3D elements. The only non-rigid boundary is the bottom side, where there is the vibrating plate. The structural excitation applied on the plate remains the same as before. The FE model is solved by means of a fluid structure SOL111 in NASTRAN. The observed result is still the quadratic average of the calculated velocity on the entire area the plate, and simultaneously the average of the sound pressure level (SPL) at 3 different positions inside the cavity, as shown by the Figure 7.

Figure 7: FE model of the acoustic box, featured with the flexural plate at its bottom

In the case where the plate is covered by an insulation treatment, the fluid structure coupling is performed through the same trim surface impedance matrix as before [25]. Thanks to the projection of the surface impedance matrix of the trim onto the modal bases of both the plate structure and the acoustic cavity, the problem can be still solved by a similar SOL111. The presence of the trim is formulated in such a way that the result takes into account the effect that the trim has on the structure and the effect that the trim has on the acoustics.

3.2 Calculated treatments and results 3.2.1 Comparison of foam and felt trim solutions We are using exactly the same damping and insulation treatments than in the previous paragraph, but the result is shown in terms of SPL. First of all, we are showing the SPL results for both the foam and the light felt (IFP40) with properties shown in Table 1. The curves of Figure 8 are the averages of the SPL calculated at the 3 microphone locations.

11

Importance of the evaluation of structure-borne NVH performance for lightweight … 110 100

SPL [dB]

90 80 70

IFP40

60

IFP 80 IFP 140

50

Foam 40 0

100

200

300

400 500 600 Frequency [Hz]

700

800

900

1000

Figure 8: plate-box SPL with 20 mm thick foam and felt insulation treatments, with variation of different felt densities

In opposition to the conclusions obtained for the vibration performance, we can see in this example that the felt solution has a better acoustic performance than the foam solution. It is likely to be understood here, that despite a higher vibration level of the structure, the felt solution has a lower transmissibility due to its decoupling properties. When looking at comparisons between different felt densities, it appears that up to 600 Hz, the lightest and softest solution is the best for the acoustic performance, despite poorer base plate vibration levels. Above 600 Hz and up to 900 Hz, the tendency seems to inverts, but the felt solution remains better than the foam solution; the later tend to equal each other with the increase of the frequency. This interesting observation may lead to favour the felt solution against the foam one.

3.2.2 Influence of the thickness When making observations on flat samples, the conclusions must be drawn at different thickness in order to anticipate the question of packaging space for 3D parts. Therefore, it would be interesting to understand how the comparison between foam and felt can be confirmed at other treatment thickness, as the observation was done so far for 20 mm thickness. Indeed the Figure 9 shows the comparison between foam and felt for two different density values in the case of insulation treatments of 5 mm thick (thickness of the decoupler). It is interesting to see that the tendency remains the same but is shifted in frequency in same proportions to the shift of the spring mass resonance frequency of the trim. As a consequence, in the case of 5 mm thickness, the lightest felt solution remains the best solution over almost the full frequency range of observation (values of stiffness for 5 mm remain in theory the same as for 20 mm, as in Table 1).

12

Importance of the evaluation of structure-borne NVH performance for lightweight … 110 100

SPL [dB]

90 80 IFP40

70

IFP 140

60

Foam 50 0

100

200

300

400 500 600 Frequency [Hz]

700

800

900

1000

Figure 9: plate-box SPL with 5 mm thick foam and felt insulation treatments

3.2.3 Influence of the foam properties on the acoustic performance As it was said in the previous paragraph, the foam decoupler has a stronger influence on the vibration level of the body structure than a very light felt. However, we have just shown that this is not systematically reflected in a better acoustic performance. Therefore, it is important in this discussion to add some information about the effect of different foam properties on the final acoustic performance. For this, we have performed a parameter variation study on the stiffness, the density and the loss factor of the foam, at different thickness. The Table 2 summarizes the parameter variation. We will report here only the results at 20 mm, considering that the conclusions remain the same at other thickness, apart for the fact that the structure borne effects are shifted on the frequency scale. All results are presented under the form of SPL spectra averaged on 100 Hz frequency bands on Figure 10 and Figure 11. Table 2: Parameter variation for the foam property influence Loss Factor [-] Young Modulus [kPa]

0.2 50

Density [kg/m3]

50

0.35 50

100 80

50

80

50

0.5 50

100 80

50

80

50

100 80

50

80

13

Importance of the evaluation of structure-borne NVH performance for lightweight … 95

SPL [dB]

90 85 80 75

Density 50 kg/m3 Density 80 kg/m3

70 65 0

100

200

300

400 500 600 Frequency [Hz]

700

800

900

1000

Figure 10: plate-box SPL with parameter variation of 20 mm foam decoupler; Full line E = 100 kPa ; Dashed line E = 50 kPa ; Colour for density

95 90

SPL [dB]

85 80 Loss factor = 0.2

75

Loss factor = 0.35

70

Loss factor = 0.5

65 0

100

200

300

400 500 600 Frequency [Hz]

700

800

900

1000

Figure 11: plate-box SPL with parameter variation of 20 mm foam decoupler; Full line E = 100 kPa ; Dashed line E = 50 kPa ; Colour for loss factor

Those results show the order of magnitude of the influence of the different properties of the foam. Most important is the “softness” of the foam with a sensitivity of more than 5dB on the final result, while the loss factor, which is known to have a strong influence on the structural vibration, has a secondary influence on the acoustic performance of the trim. It seems that the stiffness plays a role together with the loss factor on a limited frequency range corresponding to the region of spring mass resonance effect of the trim. Same conclusion can be done for the density of the foam that seems to play a more important role on a region linked to the resonance of the trim.

14

Importance of the evaluation of structure-borne NVH performance for lightweight …

3.2.4 Influence of the damping treatment Unlike the insulation treatment, the damping treatment should have a direct consequence on the acoustics, as it is applying on the panel vibration reduction, but doesn’t influence or modify the transmissibility property of the trim. The damping acts on the vibration reduction, but is totally transparent from insulation point of view (at least in the frequency range of concern for the structure-borne problematic). Therefore, the usage of damping becomes very judicious when combined to light insulation treatments. In the Figure 12, we show the effect of damping on the acoustical performance of the plate covered by different felt and foam treatments. 110 100 90

SPL [dB]

80 70 60

IFP40 IFP40 with optimal damping Foam Foam with optimal damping

50 40 0

100

200

300

400 500 600 Frequency [Hz]

700

800

900

1000

Figure 12: plate-box SPL with optimal damping treatment and 20 mm thick foam and felt insulation treatment

It appears clearly that the felt solution benefices more from the damping treatment than the foam treatment, due to the very natural reason that in the case of the foam, the plate is already damped by the decoupler, making it less sensitive to additional damping treatment. Moreover, the weight trade-off between insulation treatment and damping must be taken into account. In the case of this plate box, the weight of the damping is very high (0.8 kg) but still in the order of magnitude of the weight difference between the foam and the felt solutions. But we can imagine that in the case of a real vehicle, the damping weight can become small in comparison to the trim.

15

Importance of the evaluation of structure-borne NVH performance for lightweight …

4 Experimental study, and validation of performance tendencies The results on flat samples offer an overview of different technical solutions, including felt, light foam or viscoelastic foam. Those results set against each other the different solutions with respect not only to the body vibration, but also to the acoustic performance. In this paragraph, we are presenting the last results of tests performed on a platebox test-rig, in order to verify the simulated tendencies. The aim of those results, is to see how relevant are the physical phenomena observed through FE simulations, in the complex case of a real flat sample laying on a vibrating plate, so that the simulation can be used in the future to drive with a certain degree of confidence the choices and decisions for technological insulation solutions.

4.1 Test description and method For the reproduction of the conditions of the simulations, we use an academic test-rig (shown on Figure 13) made of one steel plate structurally excited and radiating into a rigid box. The plate is excited by means of an electro-dynamic shaker at a point close to one of its corners. The vibration of the plate is recorded at 180 points (distance between adjacent points ~50 mm) with a laser vibrometer and the SPL inside the cavity is recorded at 3 different microphone positions. The average SPL represents the target performance for the evaluation. It is worth remarking that, while the configuration of the system is conceptually simple, the dimensions of the plate and of the cavity lead to a relatively high number of modes (approx. 50 modes for the plate and 30 modes for the cavity, up to 500 Hz, making the vibro-acoustical behaviour of the system substantially more complex than other similar test-cases that can be found in the literature [26, 27].

Figure 13: plate-box test rig used for evaluation of structure borne performance of flat samples

16

Importance of the evaluation of structure-borne NVH performance for lightweight … In order to ensure the measured SPL to be well representative of the different trim influences, a certain number of testing verifications have been carried out, such as the measurement of the background noise (>20 dB dynamic), the contribution of the rigid walls of the box (>20 dB below the plate vibration level) and the contribution of the sample frame (>20 dB below plate vibration level). The accuracy of the correlation between test and simulation in the case where the testing conditions are well under control has been already demonstrated in previous publications [28]. In the present case, we concentrate on the demonstration of several observed tendencies. For this purpose, we have used existing flat samples with respectively foam and felt available decouplers. The material properties of the decouplers were characterised, so that a “one-to-one” comparative simulation can be also performed. Both foam and felt decouplers were covered by a 2 kg/m2 mass layer. Both samples are about 20 mm thickness. The Table 3 summarizes the major material properties of the samples. Like for IFP, the felt sample is referenced with its nominal density, for example Felt40 for 40 kg/m3. It is interesting to remark that the stiffness of the heaviest felt equals the stiffness of the foam. In general, the stiffness of the felts is higher than the stiffness of the injected fibre materials considered in the previous numerical analysis, and the thickness of the felt samples appear to be lower than the thickness of the foam. This given configuration is particularly very conservative, when investigating on felt performance. Table 3: overview of material properties of the flat samples for testing

Foam sample Felt sample 40 Felt sample 100

Thickness [mm] 20 17.5 18.5

Density [kg/m3] 65 49 103

Stiffness [kPa] 91 37 89

Loss factor [-] 0.13 0.06 0.09

Additionally, we have also performed the measurement of the effect of damping on the trimmed plate-box. For this, we have used existing damping samples that slightly differ from those considered in the simulation. The bitumen material is 2 mm thick, with an area weight of 3.7 kg/m2 (density = 1.8) and a loss factor of 0.25. The damping solution tested on the plate is shown Figure 14. In this case, the overall weight of the damping solution is 0.5 kg, and we paid attention that it is not covering the area of the plate, where is placed the shaker excitation. The design of the damping layout has been obtained by simulation, ensuring that it corresponds to an optimal and weight efficient design, by using an Autoneum methodology [29].

17

Importance of the evaluation of structure-borne NVH performance for lightweight …

Figure 14: damping package used on the plate for testing purpose (Weight = 0.5 kg)

4.2 Experimental results 4.2.1 Results of measured vibration and SPL in the plate-box The Figure 15 shows the measured effect that the trim has on the vibration level of the flat plate. From the test, we definitely verify what was already demonstrated for the vibration performance, namely the fact that the foam shows a stronger influence than the felt on the plate vibration level than the felt. Then, the performance of the felt on the vibration also depends on its stiffness formulation, the stiffer (or the denser), the better performing, excepted at low frequencies, where the softer, the better. The order of magnitude of the level difference between the foam and the lightest felt is in this case similar or even more pronounced (about 10dB) than what was observed in the case of the simulation. This can be explained by the fact that the material characteristics of the measured samples differ from what was considered in the previous section in simulation.

18

Importance of the evaluation of structure-borne NVH performance for lightweight … -35 -40

Velocity rms

-45 -50 -55 -60

Felt 40 Felt 100

-65

Foam -70 0

100

200

300

400

500

600

700

800

900

1000

Frequency

Figure 15: measured flat plate vibration, when covered by different insulation treatments

Concurrently, we show the measured effect of the trim on the SPL. In this case, we have been also recalculating the simulated acoustic performance, in order to take into account the correct material properties of the sample. This allows performing a convincing confrontation of the testing results against what was demonstrated in the numerical study. The Figure 16 shows the measured effect that the trim has on the SPL inside the box, in the case of the three different trim material cases. In this case, the results are a little bit different from the simulation, namely that the SPL of the three solutions are very close to each other. First of all, we can say that despite a very strong difference that was observed on the structural vibration at the advantage of the foam, the SPL performance is perfectly recovered thanks to the different nature of the transmissibility of the different constructions. At the end, the felt acoustic performance is very close, and in some cases even better than the foam. This is explained by the fact that the measured samples have different properties from those used in the numerical study, especially the real felt samples have higher stiffness values.

19

Importance of the evaluation of structure-borne NVH performance for lightweight … 105

SPL [dB]

95 85 75 Felt 40 65

Felt 100 Foam

55 0

100

200

300

400 500 600 Frequency [Hz]

700

800

900

1000

Figure 16: measured box SPL, with the plate covered by different insulation treatments

In order to clarify this discrepancy to the numerical study, we have been re-computing the SPL results with the correct material properties, and comparing those new results with the test results. For this, a particular attention was given to the correct characterization of the samples, especially for what concerns the softest one (the felt 40). The characterization was done using the Autoneum measurement system called ELWIS [30, 31] that can provide the dynamic Young’s Modulus in function of the frequency. Considering that the simulation of the SPL is very sensitive to the stiffness value, it was important to take into account this dependency to the frequency. The Figure 17 shows the Young’s Modulus of the felt 40, as well as the values used for the simulation. For the other materials for which the simulation is less sensitive to the stiffness value, we use an average over the frequency range [50-400] Hz, like shown in the Table 3.

20

Importance of the evaluation of structure-borne NVH performance for lightweight … 70000

Young's Modulus [Pa]

60000 Measured value

50000 40000

Interpolated for simulation

30000 20000 10000 0 0

100

200

300

400

Frequency [Hz]

Figure 17: frequency dependent dynamic Young’s Modulus used for the simulation of the softest decoupler material (Fetl40).

The Figure 18 shows the SPL validation, first of all by comparing the calculated SPL of the three different trims to each other. We confirm here that the simulation provides the same general conclusion as for the test, namely that the SPLs of the different trim solutions tend to equal each other. 105

SPL [dB]

95 85 75 Felt 40 (with freq. dependent E) 65

Felt 100 Foam

55 0

100

200

300

400 500 600 Frequency [Hz]

700

800

900

1000

Figure 18: simulated box SPL, with the plate covered by different insulation treatments

A particular effort has been spent on the numerical correlation of the most difficult trim to be calculated, that is to say the softest felt. The Figure 19 shows the SPL comparison

21

Importance of the evaluation of structure-borne NVH performance for lightweight … test versus FE for the felt 40, confirming a very good correlation. This correlation demonstrates the relevance of the numerical study performed in the first part of the article, and the conclusions drawn in the case of a very soft felt solution, where the low stiffness leads to even better acoustic performance levels. 105

SPL [dB]

95 85 75 Test result

65

Calculated (FEM) 55 0

100

200

300

400 500 600 Frequency [Hz]

700

800

900

1000

Figure 19: validation of the SPL obtained with the measured flat samples, against simulation

4.2.2 Influence of the damping In a similar way to what was performed in numerical study, we have measured the influence of an optimal damping treatment on the vibro-acoustical performance in the case when the plate is also covered by respectively a foam and felt insulation treatment. The Figure 20 shows the measured influence of the damping treatment on the vibration level of the plate, and the Figure 21 shows the tested influence of the damping treatment on the SPL inside the box for respectively two different trim decouplers, namely the Foam and the Felt40 of the Table 3, still covered with 2 kg/m2 of mass layer. In the case of the structural vibration, the damping is more effective at low frequency (below 150 Hz) on the foam configuration than on the felt one. This was not particularly observed in the numerical study. Around 250 Hz, the foam case highlights the effect of the trim spring-mass resonance on the vibration, where the presence of damping is counterproductive (due to its mass). This is not observed in the felt case, where the spring mass resonance in case of structure-borne excitation seems to be mitigated. At higher frequencies, we confirm that the felt configuration remains much more sensitive to the damping than the foam configuration. In the case of the SPL, we can simply comment a translation of the performance already observed on the vibration level, but to a much lesser extent, in a similar way to what can be observed on vehicle.

22

Velocity [dB]

Importance of the evaluation of structure-borne NVH performance for lightweight … -30 -35 -40 -45 -50 -55 -60 -65 -70

Foam Foam - With damping

Velocity [dB]

0

100

200

300

400 500 600 Frequency [Hz]

-30 -35 -40 -45 -50 -55 -60 -65 -70

700

800

900

1000

800

900

1000

Felt40 Felt40 - With damping

0

100

200

300

400 500 600 Frequency [Hz]

700

Figure 20: influence of an optimal damping treatment on the measured plate vibration, when also covered by insulation treatment, respectively foam and felt.

110 SPL [dB]

100 90 80 70

Foam

60

Foam - With damping

50 0

100

200

300

400 500 600 Frequency [Hz]

700

800

900

1000

SPL [dB]

110 100 90 80 70 60

Felt40 Felt40 - With damping

50 0

100

200

300

400 500 600 Frequency [Hz]

700

800

900

1000

Figure 21: influence of an optimal damping treatment on the SPL inside the box, when the plate is also covered by insulation treatment, respectively foam and felt.

23

Importance of the evaluation of structure-borne NVH performance for lightweight …

4.3 Verification performed on standard test (RTC-III like) for the automotive industry We propose here to present the results of a flat sample acoustic validation performed on a less academic measurement device, but which is used to perform standard structure borne validations in the automotive industry. The test consists in a 1.6 mm thick flexural plate clamped on its perimeter to a rigid frame that is rigidly excited by a shaker, similar to the RTC-III measurement system. The vibration level of the plate is assessed by means of one accelerometer. The plate can be covered by a flat acoustic trim. The evaluation of the acoustic performance of the trim can be done by measuring the SPL with a microphone placed above the plate in an anechoic environment, but also by measuring directly the vibration of the top layer of the trim (the mass layer) with a laser vibrometer. The measurement of the vibration level of the top layer of the trim is a good indicator of the overall NVH performance of the trim, because it corresponds to the ultimate interface variable between the waves propagating in the trim, and the waves excited in the receiving acoustic environment. This measurable is considered more robust than the measurement of the SPL, given that the microphone test, in the case of this test setup, could be influenced by secondary noise bridges, or flanking paths such as the radiation of the plate frame (this was most probably eliminated in the case of the academic platebox used for the tests presented previously). The Figure 22 shows the description of the testing setup.

Figure 22: description of the RTC-III-like measurement system

24

Importance of the evaluation of structure-borne NVH performance for lightweight … The plate has been covered alternatively by two different insulation trims, consisting of 20 mm of respectively foam and felt decouplers, covered by a 2 kg/m2 mass layer. The Figure 23 shows the two different materials used for this experiment. The felt (called IFP60) is a hand prototyping of a version of injected fibre technology, offering to the material a rather low density and stiffness. The foam (called PUR57) corresponds to a polyurethane light foam solution existing in current production. We must stress out that in this particular case of sample availability, the felt solution is as heavy, and almost as stiff as (but still softer than) the foam solution. This could be considered as a very interesting case of foam versus felt assessment at iso-weight, keeping in mind the higher technological potential for felt, especially in the field of injected fibre.

Figure 23: Decoupler PUR57 foam and IFP60 felt samples used for the RTC-III flat sample structure borne test

The Figure 24 shows the vibration level of the plate when alternatively untreated (bare), and covered by the foam and felt trims. We can comment here the slightly better or equal performance of the foam for what concerns its effect on the plate vibration reduction (compared to bare case). This can be explained by the fact that both felt and foam solutions have in that case very similar density values, and the felt is rather stiff for an injected fibre type. On the opposite, when looking at the vibration of the trim upper surface (the Figure 25 shows the vibration level of the upper surface of the trim, against the vibration of the uncovered plate.), the performance of the felt is definitively better than the one of the foam, excepted in the frequency range of 400 to 500 Hz. It is not shown on the figure, but the felt is again better than the foam above 500 Hz. This difference between the two trim surface vibration levels is representative of the difference of the respective radiated acoustics, and therefore to their respective contribution to the SPL. This indicates clearly that the contribution of the felt insulation sample (IFP60) to the SPL must be lower (means better), than the contribution of the foam insulation sample (PUR57) to the SPL. And this is expected to be even more important in the case of even lighter injected fibre solution. The relation between the vibration of the top layer of a partition and its acoustic contribution to the SPL can be demonstrated in the case where the acoustic receiver is a free field, by means of the Rayleigh integral [32].

25

Panel center Acc/Frame Acc dB[ms-2/ms-2]

Importance of the evaluation of structure-borne NVH performance for lightweight …

Bare plate PUR57 IFP60 0

100

200 300 Frequency[Hz]

400

500

Trim surface center vel/Frame Acc dB[ms-1/ms2]

Figure 24: plate vibration measurement on the RTC-III measurement system, treated with different insulation flat samples

Bare plate PUR57 IFP60 0

100

200 300 Frequency[Hz]

400

500

Figure 25: trim upper radiating surface vibration measurement on the RTC-III measurement system, treated with different insulation flat samples

5 Decision process for insulation trim design We have seen that the decision for trim insulation strategy with respect to the structureborne performance can be strongly motivated by the damping put on the structure and the softness of the decoupler. It seems that if foam solutions tend to be more effective for panel vibration reduction, the felt solutions can provide a superior acoustic comfort, even at lower weight, given that the low dynamic stiffness offered by the fibre solution can be properly achieved. We have seen also that those behaviours do not have the same relevance on the whole frequency range, and at all the thickness. Moreover, all those considerations have been driven by observations done by means of simulation and testing on flat samples. It is to be expected that the complexity of the packaging space must be taken into account. This aspect is tackled in a later part of the article.

26

Importance of the evaluation of structure-borne NVH performance for lightweight …

5.1 Technological and conceptual variables of the decision process Beside decisions that are driven by the acoustic performance, it is shown that the mechanical properties driving the final performance can come into conflict with other functional requirements, such as the density, or static compression, but also with the manufacturing limitations of the decoupler materials. A first very important aspect is the requirement on the static compression. The acoustic parts that are built in the form of mass-spring insulators are mainly applied to the inner dash area (below and behind the instrumentation panel) and to the floor area. In case of floor application, that is rather common, requirements exist on the static compression hardness of the material, to avoid that the floor is felt too soft when stepping in the car. Normally, OEMs’ requirements are expressed in terms of compression hardness and recovery. There is no standard definition of the two parameters; however the different methods that are defined by the OEMs are conceptually very similar. For compression hardness, we have considered the force necessary to compress a sample to 50% of its thickness, the sample having a diameter of 60 mm. Different decoupling materials have been considered: felt with different composition (representative also of the IFP technology) and foam. The results have been grouped by type of material as function of the density (Figure 26). As it is clear from the picture, being the compression hardness the higher the better, the trend of a felt decoupler is better than the one of a foam decoupler. 16

Hardness [kPa]

14 12

Felt

10

Foam

8 6 4 2 0 0

10

20

30 40 Density [kg/m3]

50

60

70

Figure 26: compression hardness as function of the density for felt and foam decouplers

27

Importance of the evaluation of structure-borne NVH performance for lightweight … For the recovery, we have considered the quantity (in %) represented by the ratio of the difference between the final thickness (after recovery) and the compressed thickness and the difference between the initial thickness and the compressed thickness, for a compression with a force of 35 kg. The results have been grouped by type of material as function of the density (Figure 27). As it is clear from the picture, being the recovery percentage the higher the better, the trend of a foam decoupler is better than the one of a felt decoupler. 100

Recovery [%]

95

felt

90

foam

85

80 0

10

20

30 40 50 Thickness [mm]

60

70

80

Figure 27: recovery as function of the density for felt and foam decouplers

The choice of the decoupler type based on the requirements of compression hardness and recovery is linked to the targets set by the OEMs, however, it is evident that the functional requirements of a floor insulator might influence heavily the choice of the density of the decoupler used. It must be observed as well that, in general (Table 3, Table 4) for the same density the Young’s modulus of felt materials is lower than the Young’s modulus of a foam material. The observations made in the previous Paragraphs might suggest the choice of an IFP material with low density (and consequently low Young’s modulus), however the limitations on compression hardness and recovery might reduce the applicable range of density. Other important technological limits to be taken into account concern the foam loss factor and the foam densification. As observed in 3.2.3, the effect of the loss factor on the structure-borne noise is not negligible at selected frequency ranges, even though it has to be put in perspective together with the effect of the stiffness. It is important to ob-

28

Importance of the evaluation of structure-borne NVH performance for lightweight … serve that different values of the loss factor cannot be easily obtained with the same type of foam. For values around 0.2, a typical formulation of High Resilient foam can be used: for these materials, the density can be lowered down to values between 45 and 55kg/m3 to obtain a favourable low Young’s modulus. However, for higher values of loss factor (0.35 and above), it is necessary to use visco-elastic formulations, for which it is very difficult to obtain densities below 60kg/m3. As a consequence, the Young’s modulus is not an independent parameter from the loss factor in the case of foam, and the two parameters can be chosen in reality only between a more limited matrix of potential combinations. For the foam, it must be also considered that the average density on a 3D part can be controlled within reasonable limits. However, there are areas where the material cannot flow easily during the foaming process because of the reduced thickness of the part. This effect is depicted in Figure 28, where the measured density versus thickness is plotted for foam samples taken from 3D shaped parts. For thicknesses below 15 mm, there is a densification of the material which would influence the Young’s modulus of the foam, as shown on Figure 29. In fact, the foam Young’s modulus increases with the density. The area of “stable” foam density is the one with a thickness higher than about 15 mm. Therefore, on real 3D parts, it must be taken into account that the effective structure-borne behaviour might be influenced especially at low thickness areas.

Volume density [kg/m3]

120 100 80 60 40 20 0 0

10

20

30 Thickness [mm]

40

50

60

Figure 28: example of measured density vs. thickness for foam samples taken from 3D parts

29

Youngs Modulus [Pa]

Importance of the evaluation of structure-borne NVH performance for lightweight …

0

20

40

60

80

100

120

Foam Density [kg/m3]

Figure 29: example of measured Young’s modulus of foam samples, related to Figure 28 (distance between the horizontal bars is 50 kPa).

5.2 Taking into account the packaging space for design of real parts For the design of real parts, we must be able to take into account simultaneously the complex structure-borne NVH behaviors of the bill of material over the full frequency range of interest and within the packaging space offered by the drawings, together with the constraints of the available technologies. This is a challenging need that requires the establishment of clear guidelines that can offer a simple relation between the packaging space, the frequency, the choice of material, and the NVH performance trend. In this article, we propose a particular focus on injected fiber technology (called here IFP), where the simulation can be used to find the best material density selection versus thickness. For this purpose, we exploit the academic plate-box virtual model that has been used in the first part of the article to put in evidence the concept. Here, we exploit in a systematic way the trends given by the simulation, and calculate the vibration and SPL level of flat samples for each available density of the IFP material, and at each of the possible thickness of an insulation part, by steps of 5 mm up to 40 mm thickness. Then the results under the form of frequency response spectra are averaged by range of frequencies and finally condensed under the form of indicator matrix. Those results are a very first attempt to summarize the physical tendencies, but the principle behind this could be improved, so that it fulfills in an optimal way the needs of the design engi-

30

Importance of the evaluation of structure-borne NVH performance for lightweight … neers. The Table 4 summarizes the ranges of frequencies, thickness and densities considered for the simulation. Table 4: parameter ranges for the definition of NVH design guidelines for felt injected fiber type

Density [kg/m3] Young Modulus [kPa] Thickness [mm] Frequency range [Hz]

Material IFP30 IFP40 IFP50 IFP60 IFP80 IFP140 30 40 50 60 80 140 11 13 14 16.5 20 31 5 10 20 30 40 20-300 300600600 1000

Foam 87 73

The Figure 30 and Figure 31 present the results respectively for the vibration of the structure and the acoustic, under the form of abacuses, which show the best density of IFP to be used for each thickness, in order to reach the best performance. Lower is better. The results are also compared to the performance of a standard foam. 95

SPL index [dB]

90

300 - 600 Hz

20 - 300 Hz

600 - 1000 Hz

85 80 75

5mm 10mm 20mm 30mm 40mm

70 65 20

good

70 120 Density [kg/m3]

good 20

70 120 Density [kg/m3]

20

70 120 Density [kg/m3]

Figure 30: abacuses of the SPL performance index of IFP w.r.t. density and thickness, on three different frequency ranges

31

Importance of the evaluation of structure-borne NVH performance for lightweight … -40

20 - 300 Hz

600 - 1000 Hz

300 - 600 Hz

5mm 10mm

Vibration index [dB]

-45 good

20mm 30mm

-50

40mm

-55 good -60 20

70

120

Density [kg/m3]

20

70 120 Density [kg/m3]

20

70

120

Density [kg/m3]

Figure 31: abacuses of the vibration performance index of IFP w.r.t. density and thickness, on three different frequency ranges

The performance of the decoupler depends strongly on the density of the IFP. In the case of the SPL, the lower density is in general the better, but not systematically. This observation is in line with conclusions of previous simulations, which have shown that softer is better. In the highest frequency range of investigation (600 – 1000 Hz), and when the decoupler is thick (40 mm), this dependency tends to be reversed, and a denser IFP can become acoustically better than a light one. In the case of the plate vibration, the situation is quite the opposite. Generally, the denser (or the stiffer) is clearly better, excepted at low frequencies where the softer is better, especially for low thickness. Those observations give a very good insight of how the available thickness of complex part geometry will have to be taken into account in the definition of the bill of material. When the available thickness is small, a soft fiber should favor the structure borne performance of the part. This is also in line with the air borne high frequency requirements for thin areas. The statement remains true when the thickness increases, but with a decreasing relevance of the statement for what concerns the final performance of the part. When the thickness is really large, it can become difficult to place a soft decoupler due to the requirements for respecting a minimum compression set for carpet parts. In this case, a stiffer IFP decoupler means heavier, and that has to be taken into account when the design focuses on light weight solutions. Conversely, a stiffer IFP solution can be beneficial for the structure borne performance in the highest frequency range of structure borne investigation.

32

Importance of the evaluation of structure-borne NVH performance for lightweight …

5.3 Virtual demonstration on vehicle floor design In this last paragraph, we want to show an example of part design, aiming at getting the best structure borne performance, and taking into account the available part thickness. In the case of the IFP, we exploit the statements on density dependency established in the previous paragraph. We show how the final decision for the material of the decoupler influences the overall part performance.

5.3.1 Definition of the test case We propose to evaluate the performance of a 3D part by means of FE simulation. For this purpose, we use a vehicle floor mockup, as shown by Figure 32, featuring a floor body cut-out coupled to a simplified passenger acoustic compartment. The floor is clamped on its contour. As shown by the Figure 33, the vibration of the clamped floor is above 150 Hz similar to the vibration of the floor if coupled to the rest of the vehicle body. The floor appears only stiffer at low frequency (50-100 Hz) due to clamping condition, but this is not compromising our virtual experiment. Above 500 Hz, the structural response doesn’t depend any more on the boundary conditions. The floor is excited structurally at one point situated on a stiff region of the tunnel, by a reference force. The floor is covered by an insulation part consisting in a mass layer and a decoupler, like shown Figure 34. The thickness map of the decoupler is also shown, in order to illustrate the choice of bill of material with respect to the packaging space. As we can see, some areas, named “uncovered areas” are covered by the mass layer only, without decoupler. The SPL acoustic response to the floor excitation is calculated at the driver head position. This response can be assimilated to the contribution of the floor component to the passenger cavity SPL. In a first instance, the simulation is carried out without the contribution of the uncovered areas, so that the contribution of the decoupler to the part performance is magnified in the virtual demonstration. The simulation is performed by using the same FE technic as for the flat sample simulations, but the insulation part must be meshed to respect the 3D geometry of the drawing.

33

Importance of the evaluation of structure-borne NVH performance for lightweight …

Figure 32: FE model of floor cut-out coupled to a simplified passenger cavity

20

Velocity [dB](m.s-1/N)

10 0 -10 -20 -30

Floor clamped

-40

Floor coupled to body

-50 0

200

400

600

800

Frequency [Hz]

Figure 33: influence of the cut-out boundary conditions on the response of the floor

34

1000

Importance of the evaluation of structure-borne NVH performance for lightweight …

a)

b)

Figure 34: a) Insulation part (Green = mass layer; Yellow = decoupler). b) Thickness map of the decoupler

5.4 Simulations and results The floor carpet’s decoupler is simulated respectively with foam and felt materials, by using the same materials as in the previous paragraph (Table 4), covered by a 2 kg/m2 mass layer. In the case of foam, we assume that the foam is iso-density throughout the full volume of the part, independently of the thickness. This is a conservative assumption (see paragraph 5.1 on technology description), considering that a foam would highlight areas of densification in the thin areas for example. In the case of the felt, we exploit the fact that the IFP technology can offer local density design. Therefore we can apply in a more refined way design rules, in function of the available thickness. Like shown by the Figure 35, the proposed design tries to favor low density (here IFP40) for the low and medium thickness (such as tunnel or middle floor), aiming at favoring the best possible performance. In areas with large thickness such as the front floor that features more than 50 mm space, the IFP must have higher densities (here IFP140) due to compression set requirements. This is done at the cost of weight increase, but this should be conversely beneficial for the acoustic performance in the highest frequency range. As a mid-range solution, the rear floor featuring rather thick decoupler (20 to 30 mm) is designed with an IFP80 for acoustic reasons. According to the abacuses discussed in paragraph 5.2, this decision favor the 600-1000Hz region at the expense of the acoustic in the low frequency range, but with an expected minor consequence, considering that the thinner regions are predominant. The geometrical precision of the design and the density choices are compatible with the technological manufacturing limitation.

35

Importance of the evaluation of structure-borne NVH performance for lightweight …

Figure 35: decoupler design of the floor mock-up in the case of variable density IFP

The Figure 36 shows the SPL results inside the passenger compartment, with foam and IFP carpet. In the case of IFP, the result is also compared for the sake of the demonstration to the performance of the lowest weight solution with iso-IFP40 material. The figure puts also the respective results in comparison to their corresponding part weight. First of all, we verify on a 3D part the better structure borne performance of the light felt compared to standard heavier foam, already observed on flat samples. Beside this, the most interesting resides in the relevance of the design of a realistically engineered IFP solution with variable density (the proposed design) compared to a “nonengineered” light solution, with iso-density (IFP40). Up to 350Hz, the performance of the designed solution is not affected by the high density choice on the front floor. On the 350–450 Hz range, there is deterioration, but it corresponds to a range of low response levels, and there might still be room for improvement. Above 450 Hz the designed solution is, as expected, better than the lightest one, due to the contribution of the highest thickness range. The simulation has not been performed above 600Hz, but we would logically expect from what we have learnt an improvement, at least up to 1000 Hz. This example demonstrates also that the acoustical and technological optimum leads to a certain weight compromise, with a total weight of 12 kg (including the mass layer). For comparison, this weight corresponds to an equivalent decoupler of about 1500 g/m2.

36

Importance of the evaluation of structure-borne NVH performance for lightweight … -40 Foam solution - 13kg IFP variable density - 12kg

-50 SPL (dB)

IFP40 iso-density - 10kg -60

-70

-80 100

200

300 400 Frequency (Hz)

500

600

Figure 36: SPL (structure-borne) performance of the floor mock-up with different insulator solutions, without contribution of the uncovered areas. Insulation part weight.

This illustrates very well the implications and the added value of an enlighten design, especially when it comes to light weight acoustic solutions. At the same time, it is necessary to discuss those results in an honest way when it comes to the overall vehicle design performance. The Figure 37 shows the same results, but simulated in a more realistic way, where the influence of the uncovered areas is taken into account – from a certain point of view, the uncovered areas are part of the full packaging space picture. In a natural way, the acoustic performance is less sensitive to the bill of material differentiation, but the IFP solution remains still better than the foam solution, excepted between 450 and 550 Hz. Indeed, according to considerations seen at the beginning of this article, the vibration level of the body with the felt solution is higher than with foam. And particularly this is true not only on covered areas, but also on uncovered areas. The Figure 38 shows the vibration level of body areas “covered” and “uncovered” in case of foam and felt. We have been able to demonstrate with additional virtual tests not presented here, that on some very limited frequency range, the higher vibration of the uncovered areas can penalize the felt solution against other solutions. This conclusion must definitely be mitigated and put in the context of floor insulator design. In the case of foam, the uncovered areas are very often there due to manufacturing restrictions in thin areas, rather than by packaging space restrictions. On the other hand in the case of a felt solution, especially of IFP, the overall amount of uncovered area can be considerably reduced by a rather thin (down to 3 mm) decoupler with a fair density objective of 100 kg/m3 in those regions. This scenario has not yet been simulated, but we expect a

37

Importance of the evaluation of structure-borne NVH performance for lightweight … sensitive improvement in the region of 500 Hz, and even an inversion of the foam and felt levels. Beside this, we can also remark that there is almost no more difference between the different IFP designs (variable versus iso-density). The design solution is slightly better than the IFP40 solution on the highest peak at 210 Hz. This is confirming that, as long as the density design is coherent with the acoustic recommendations, some local increase of densities motivated by the manufacturing constraints can only improve the overall part performance. -30 Foam solution - 13kg IFP variable density - 12kg

-40 SPL (dB)

IFP40 - iso-density - 10kg -50

-60

-70 100

200

300 400 Frequency (Hz)

500

600

Figure 37: SPL (structure-borne) performance of the floor mock-up with different insulator solutions, including contribution of the uncovered areas. Insulation part weight. 10

20

«Covered» area (Tunnel)

“Uncovered” area (Tunnel Driver Side) Velocity (dB)

Velocity (dB)

0 10

0

-10 100

200

300 400 Frequency (Hz)

500

600

-10 -20 -30 100

200

300 400 Frequency (Hz)

500

Figure 38: vibration level of covered versus uncovered areas, in case of felt and foam.

38

600

Importance of the evaluation of structure-borne NVH performance for lightweight …

6 Conclusion To conclude, we can say that light weight trim solutions, such as injected fiber technologies, can be exploited without necessarily deteriorating the structure borne NVH performance, given that their performance is properly evaluated, while taking into account all the variables and limitations of the design engineering problem. The numerical study shows the structure borne performance tendencies of injected fiber in comparison to foam flat insulators, by using material properties obtained on hand prototyped samples. The original idea of the approach is to focus not only on the vibration of the structure, but also on the radiated noise level. This is allowing putting in evidence the role of the acoustic trim transmissibility to the final performance, thus showing the potential of felt decouplers and the importance of the definition of an adequate structure borne performance measurable. This is also bringing the trim onto the same level of comparison than the damping pads, which is very important when looking for a global optimal weight solution. The experimental study carried out on available flat samples was able to demonstrate and confirm the tendencies that rose from the numerical study. In one case, we are also showing a very good numerical correlation. The final test performed on an automotive validation standard test rig confirms unequivocally the acoustic advantage of injected fiber versus foam standard solution, by looking at the result of the trim top layer vibration. We will show during the conference presentation associated to this article that the same result is observed on testing of vehicle floor, comparing IFP carpet against light foam carpet. The validation of the numerical approach for performance evaluation leads to conclusions driving the choice of decoupler properties, mainly associated to the “softness” of the material. Conversely, we explain that this has to be considered together with the limitations of the associated technologies (manufacturing and material limitations such as the thin area limitations), and with other requirements than acoustic (such as the compression hardness). Finally we made an exercise, which was to create a design rule that can drive the design of injected fiber with respect to the structure borne performance and the reduction of the weight. This has been done considering challenging material properties, but still realistic for such technology. This was applied on the case of the design of 3D-shape floor insulation part, considering also the limitations of the insulation technology. The added value of the design was demonstrated virtually by means of simulation, even in a very conservative scenario, where the influence of the uncovered areas on the overall part performance are taken into account.

39

Importance of the evaluation of structure-borne NVH performance for lightweight … As an outlook, we should extend the later exercise to the evaluation of the interaction with the damping, by looking at the influence of an optimal damping layout versus weight advantage (taking into account the overall acoustic package weight including insulator and damping). From a design methodology perspective, we see here the importance and the potential of multi-parameter optimization for structure borne. For more information, we refer to the paper on that topic [33] and presented at the same conference.

7 References [1] [2]

L. Beranek & al: Noise and Vibration Control, INCE publication B. Brouard et al., A general method of modelling sound propagation in layered media, Journal of Sound and Vibration, 183(1), 1995 [3] L. Cremer, M. Heckl and E. E. Ungar: Structure-Borne Sound, Springer-Verlag Berlin 1988 [4] J. Grébert, L. Mazzarella, Reciprocal Transfer Functions Synthesis Method for Rolling Noise and NVH Floor Treatment Investigations, Proceedings of SAE 2009, Traverse City, 2009 [5] C. Zhang, M.A. Hamdi, L. Mebarek, B. Mahieux, Influence of porous elastic components on structure and airborne noise in low and medium frequency ranges, Proceedings of Rieter Automotive Conference 2005 [6] O.C. Zienkiewicz, R.L. Taylor, The Finite Element Method, McGraw-Hill Companies, 1986 [7] R. Panneton, N. Atalla, An efficient finite element scheme for solving the threedimensional poroelasticity problem in acoustics, Journal of the Acoustical Society of America, 101(6), 3287-3298, 1997 [8] P. Göransson, A 3-D symmetric finite element formulation of the Biot equations with application to acoustic wave propagation through an elastic porous medium, International Journal of Numerical Methods in Engineering, 41, 167-192, 1999 [9] M.A. Hamdi, L. Mebarek, A. Omrani, N. Atalla, An efficient finite element formulation for the analysis of Acoustic and Elastic Waves propagation in sound packages, Proceedings of the SAE Conference 2001, Traverse City, 2001 [10] S. Rigobert, F. Sgard, N. Atalla, A two-field hybrid formulation for multilayers involving poroelastic, acoustic and elastic materials, Journal of the Acoustical Society of America, 115(6), 2786-2797, 2004 [11] M. A. Tournour, F. Kosaka, H. Shiozaki, Fast acoustic trim modelling using transfer admittance and Finite Element Method, Proceedings of SAE 2007, Traverse City, 2007

40

Importance of the evaluation of structure-borne NVH performance for lightweight … [12] Courtois, T., Bertolini, C., and Ochs, J., A Procedure for Efficient Trimmed Body FE Simulations, Based on a Transfer Admittance Model of the Sound Package, SAE Int. J. Passeng. Cars – Mech. Syst. 3(2):1-13, 2010 [13] D. Caprioli, C. Gaudino, L. Ferrali, L. Hao, Shape and damping automatic vibroacoustical optimisation of automotive panels by means of GOLD, Rieter Automotive Conference, 2005 [14] B. Faverjon, C. Soize, Equivalent acoustic impedance model. Part 1: experiments and semi-physical model, Journal of Sound and Vibration, Vol. 276, 571-592, 2004 [15] G. Maidanik, Response of Ribbed Panels to Reverberant Fields, J. Acou. Soc. Am., 34(6), 809–826, (1962). [16] L. Cremer, and M. Heckl, Structure-Borne Sound : Structural Vibrations and sound radiation at audio frequencies, Springer-Verlag, 1988 [17] T. Onsay, A. Akanda and G. Goetchius, Vibro-Acoustic Behaviour of BeadStiffened Flat Panels: FEA, SEA, and Experimental Analysis, Proceedings of SAE 1999, Traverse City, 1999 [18] F. Fahy, P. Gardonio, Sound and Structural Vibration, Elsevier/Academic Press, 2007 [19] Guj, L., Courtois, T., and Bertolini, C., A FE Based Procedure for Optimal Design of Damping Package, with Presence of the Insulation Trim, SAE Int. J. Passeng. Cars – Mech. Syst. 4(2):1291-1303, 2011 [20] F. Ronzio, T. Courtois, J.W. Yoo, Integration of acoustic trim in FE structure borne noise optimization with SPL target, Proceeding of Symposium on the Acoustics of Poro-Elastic Materials, Lyon, 2014 [21] P. Strasser, L. Ferrali, D. Caprioli: Emerald: an FE-tool for the optimal design of damped vehicle bodies with respect to low-middle frequency range, Rieter Automotive Conference, 2001 [22] D. Ross, E.E. Ungar, and E.M. Kerwin; Damping of plate flexural vibrations by means of viscoelastic laminate, volume Stuctural Damping, Chap 3. Pergamon Press, New-York, 1950 [23] M. A. Biot, Theory of Propagation of Elastic Waves in a Fluid-Saturated Porous Solid – High Frequency Range, J. Acou. Soc. Am.,28(2), (1956) [24] M.A. Hamdi, N. Atalla, L. Mebarek and A. Omrani, Novel mixed finite element formulation for the analysis of sound absorption by porous materials, Proceedings of Internoise 2000, Nice, 2000. [25] C.Bertolini, L.Guj, F.A.Bassi, K. Misaji and F. Ide, Treasuri2/FE: a tool for the FE simulation of sound package parts fully integrated in Nastran, Proceedings of SAE 2009, Traverse City, 2009 [26] C. Glandier, R. Lehmann, T. Yamamoto, Y. Kamada, Vibroacoustic FEA modeling of two layer trim systems, Proceedings of SAE 2005, Traverse City, 2005

41

Importance of the evaluation of structure-borne NVH performance for lightweight … [27] N. Kobayashi, M. Habuchi, H. Yamaoka, FEM system development for dynamic response analysis of acoustic trim, Proceedings of SAE 2009, S. Charles, 2009 [28] T. Courtois, C. Bertolini, A new procedure integrated in Nastran for trimmed body FE simulation, based on a transfer admittance model of the sound package. Implementation and validation, SIA Comfort Congress October 2010 [29] K. Trdak, T. Courtois, Characterization and Simulation of Sandwich Damping For Optimal Automotive Applications, SIA Comfort Congress October 2010 [30] T. Courtois, T. Falk, C. Bertolini, An acoustical inverse measurement system to determine intrinsic parameters of porous samples, Proceeding of Symposium on the Acoustics of Poro-Elastic Materials, Lyon, 2005 [31] C. Bertolini, T. Courtois, C. Gaudino and L. Marotta, Transfer function based method to identify frequency dependent Young’s modulus, Poisson’s ratio and damping loss factor of poroelastic materials, Proceeding of Symposium on the Acoustics of Poro-Elastic Materials, Lyon, 2008 [32] A. D. Pierce, Acoustics – An Introduction to its Physical Principles and Applications, Acoustical Society of America, 1989 [33] J.W. Yoo, T. Courtois, J. Horak, F. Ronzio, S.-W. Lee, Validation on a vehicle of the structure borne noise of a light-weight body and trim design solution obtained with new integrated FE optimization, Proceedings of the Automotive Acoustics Conference, 2015.

42

How to avoid annoying rolling bearing noises Dipl.-Ing. Cedric Geffroy Dr. Hannes Grillenberger Dr. Carsten Mohr

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 W. Siebenpfeiffer (Hrsg.), Automotive Acoustics Conference 2015, Proceedings, https://doi.org/10.1007/978-3-658-27648-5_11

1

How to avoid annoying rolling bearing noises

1 Introduction The long lasting trend to minimize CO2 emissions leads to several changes in the powertrain. Besides downsizing, high ratio spreads, or hybrid technologies for example, the number of rolling bearings is increasing. Their benefit is the lower friction loss compared to plain bearings and they suit better to start-stop systems as well with reduced wear when starting from standstill. Furthermore rolling bearings need less oil pump power so that the fuel consumption decreases again. On one side rolling bearings have already replaced existing plain bearings in many applications, as for example on balance shafts. Ball bearing turbochargers are becoming more commonplace and several feasibility studies on rolling bearing cranktrains and camshafts have also been performed. On the other side the increased complexity of powertrains leads to new auxiliary systems and thus new bearing applications. In a first step, this paper illustrates how the bearing geometry affects the excitation frequencies. These correlations are well-known in the field of bearing diagnostics and are used for condition monitoring, too. Secondly a few examples of bearing noises which appeared during development in the last years will be detailed: – – –

2

Whistling noise, CVT: how a minimal waviness leads to strong excitation. Whistling noise, Clutch Release Bearing: when kinematic excitation meets modal properties. Mosquito noise, Cover Bearing: when the cage becomes instable.

How to avoid annoying rolling bearing noises

2 Excitation of rolling bearings Even by using the most advanced manufacturing technology, vibrations still occur naturally in rolling bearings. As they do not degrade bearing performance, they are more or less accepted as normal bearing characteristics. If the internal geometry of the bearing is known it is possible to calculate these usual “rollover frequencies” as a function of speed. This is commonly used for diagnostics on machines for fault detection or condition monitoring. The formulas are listed in the standard VDI 3238 and it is easy to build a calculation table based on it. After entering the required data such as the number of rolling elements or the pitch circle diameter, the following frequencies can be determined: – – – –

BPFI (ball passing frequency inner ring) BPFO (ball passing frequency outer ring) BSF (ball spin frequency) FTF (cage rotational frequency)

Figure 1: Schaeffler frequency calculation table based on VDI 3238

3

How to avoid annoying rolling bearing noises

3 Whistling noise – CVT 3.1 Noise phenomenon A few CVTs (Continuously Variable Transmission) have been rejected during the EndOf-Line-Test in the production because of a high amplitude corresponding to the 61th order. This characteristic could also be reproduced on the acoustic test bench: an audible whistling noise occurred (figure 2).

Figure 2: Structure borne noise on rejected CVT

3.2 Root cause The CVT is composed of two pulley sets with a steel chain running between them. Each pulley set is composed of a fixed and a moveable sheave. The gear ratio is changed by axial displacement of the moveable sheave on each pulley set simultaneously which causes the variation of the effective running radius of the chain.

4

How to avoid annoying rolling bearing noises To find out which part is responsible for the noise the different parts were changed step by step. The primary pulley set seemed to be the root cause: after replacing it by another part the noise disappeared. When it was assembled into another CVT, the whistling noise appeared. In other words: the noise was directly linked to the primary pulley set itself (without its bearings). No geometrical irregularity was found on the pulleys, so the focus was set on the bearings. There are two bearings (in the following named “C” and “D”) having their inner raceway directly on the primary pulley shaft (no inner ring). In contrast to bearing “C” the inner raceway of bearing “D” showed a waviness of order 118 with a very low amplitude of 0.1 µm (see figure 2). The position of bearing “D” is shown in figure 3.

Figure 2: Measurement of waviness of inner raceway (bearing “D”)

5

How to avoid annoying rolling bearing noises

Figure 3: CVT-Transmission and bearing “D”

The amplitude of 0.1 µm is very low but on the other side the waviness order of 118 unfortunately matches with the number of 17 rolling elements. If the number of waves divided by the number of rolling elements is a whole number (in our case 118 / 17 = 6.941… = 7) the rolling elements go up and down in phase.

Figure 4: Rolling elements move in phase

6

How to avoid annoying rolling bearing noises A quotient which is exactly between two whole numbers (6.5 for example) would create a perfectly dephased situation, where every second rolling element is in a valley while the others are on a hill. This theoretically reduces the excitation, especially at low waviness order. Because when the number of waves is high compared to the number of rolling elements (it is also the case here) the quotient doesn’t need to be exactly a whole number to create the unwanted “in-phase” situation: the clearance between the cage and the rolling elements permits each rolling element to “find” its own valley. To prove if the measured waviness of only 0.1 µm really causes the claimed whistling noise at 61th order, the frequency generated by the waviness order of 118 has been calculated: when the input shaft speed is 1000 rpm, it causes a frequency of 1025 Hz that corresponds to the 61th order. Thus the waviness of order 118 on the inner raceway of bearing “D” could be identified as the root cause of the 61th order whistling noise. Further investigations took place to evaluate the influence of the amplitude on the subjective rating in the car. The rejected CVT was mounted in a test car and rated 3 (according to VDI 2563 and ATZ scale) in the passenger room. Then the inner raceway of bearing “D” has been finished to 0.05 µm. After reassembling in the car it was rated 7. So the finishing from 0.1 µm to 0.05 µm improved the subjective rating from 3 to 7. This means, that the whistling noise was still present albeit with a lower level.

Figure 5: Influence of waviness amplitude (0.1 to 0.05 µm) on subjective rating

7

How to avoid annoying rolling bearing noises

3.3 Solution The target was not only to understand why the waviness occurred, but also why it occurred only in a few parts. During the grinding process of the inner raceway of bearing “D”, the primary pulley shaft rotates and the grinding wheel rotates as well. Last one gradually becomes smaller because of wear. To assure a constant cutting speed the rotational speed of the grinding wheel is automatically increased while the primary pulley still runs with constant speed. The grinding wheel produces approx. 11000 parts before it reaches its minimum diameter and has to be replaced by a new one. Furthermore the grinding wheel has to be trued by every 10 parts. This is made by a diamond truing wheel, which is rotating, too. The grinding machine is detailed on figure 6.

truing wheel

grinding wheel

primary pulley shaft

Figure 6: The grinding machine

Several investigations on the grinding machine proved that the waviness only occurred when the speed ratio (grinding wheel to machined part) is a whole number. As shown in figure 7 this occurs two times during the lifetime of the grinding wheel: at 1143 rpm (x 9) and 1270 rpm (x 10). The few parts, machined during this brief moment, had waviness because the same surface of the grinding wheel meets the same surface of the part every 9 respectively 10 rotations. So there is no phase shift at these operating points.

8

How to avoid annoying rolling bearing noises

Figure 7: Parameters of grinding process during lifetime of grinding wheel and critical operating points

Furthermore the truing process showed a similar behavior: the diamond wheel printed waviness on the grinding wheel at whole number speed ratios. As a counter measure the automatic speed adjustment was improved to avoid these whole number ratios during grinding and truing.

9

How to avoid annoying rolling bearing noises

4 Whistling noise – clutch release bearing 4.1 Noise phenomenon A whistling noise at 1570 Hz (see figure 8) was detected at standstill in some pre-series cars with a specific type of manual transmission. Some prerequisites were necessary in order to reproduce it: – – –

cold conditions (< 3 °C) and cold engine start higher idle speed (~1150 rpm) due to cold engine start defined clutch pedal travel (500 N axial load respectively 7.5 bar hydraulic pressure)

Figure 8: Whistling noise – clutch release bearing

10

How to avoid annoying rolling bearing noises

4.2 Root cause The noise was generated by the release bearing itself which is located between the diaphragm spring of the clutch and the slave cylinder as shown in figure 9. When the clutch pedal is actuated, the clutch release bearing is pushed towards the diaphragm spring. This removes the load from the pressure plate and liberates the clutch disc.

Figure 9: Clutch release system

Some measurements of structure borne noise (without the whistling noise) showed a structural resonance correlating with the axial load applied to the bearing. When crossing the excitation frequencies (horizontal lines in the Campbell diagram) the amplitude rose (figure 10). A general image of whistling noise mechanism is detailed in figure 11, when a kinematic excitation of the actuator meets modal properties of the resonator.

11

How to avoid annoying rolling bearing noises

Figure 10: Structure borne noise, change of axial load at constant speed

12

How to avoid annoying rolling bearing noises

Figure 11: General image when actuator meets resonator

So the goal was to find out the actuator and the resonator. As there were no stick-slip effects, only the ball passing frequencies could play the role of the actuator. Concerning the resonator, the focus was set first on the diaphragm spring, because it can be easily excited and due to its geometry (thin plate) it could act like a loudspeaker. But measurements without diaphragm spring (replaced by a dummy mass) still showed whistling noise with the same frequency. A modal analysis helped to find out the resonator. The inner ring has a tilting mode with a frequency matching the whistling noise: 1570 Hz at 515 N axial load which corresponded exactly to the operating point in the cars (7.5 bar respectively ~500 N).

13

How to avoid annoying rolling bearing noises

Figure 12: Modal analysis, tilting of inner ring (1570 Hz at ~500 N)

So the higher harmonics of the ball passing frequencies excite the tilting mode of the inner ring. As shown in the figure 10 this is not enough to cause the whistling noise – something else is required. Thanks to numerous measurements on the test rig, the database grew and delivered following information: the whistling noise at 1570 Hz was always present and easy to reproduce. But sometimes there was whistling also at 1350 Hz and 1790 Hz, which was very hard to reproduce and never reproducible in the test cars. A look at the BPF-table helped for understanding the phenomenon: the overlap of higher harmonics of the BPFs played a significant role. Figure 13 shows on the left the BPF-table, in the middle the BPFs along a unique frequency axis to better recognize the overlaps and on the right a measurement with constant speed and variable axial load. On the first load ramp the three frequencies mentioned above are present, on the second ramp only two frequencies and on the third ramp only the main frequency of 1570 Hz which correspond to a perfect overlap of the BPF. The BSF seemed to be less important compared to the overlap of BPFI and BPFO.

14

How to avoid annoying rolling bearing noises

Figure 13: Influence of the overlap of the higher harmonics of ball passing frequencies

4.3 Solution Given that the ball passing frequencies meet the tilting mode, there are two ways to solve the problem: either changes to the actuator or changes to the resonator. Changes to the resonator signifies shift-up/down the tilting-frequency (whistling noise will be still present, but at another rpm) or damping of the tilting mode (difficult and without guaranteed results). On the other side, improvements regarding the actuator can be done quite simply by changing the cage: a randomized pitch sequence of the pockets is sufficient to erase the ball passing frequencies. Figure 14 shows that equally distributed balls excite a single ball passing frequency and their harmonics. By changing the angular position of the balls this single frequency is spread into several sidebands. This causes more frequencies but with clearly lower amplitudes. A similar principle is used at tyres, ventilators or in the CVT-chain pitch sequence.

15

How to avoid annoying rolling bearing noises

Figure 14: Theoretical spectra of symmetrical and randomized cage pitch

This solution has been implemented. As shown in figure 15 the differences between the two cages are not visible at first glance because of the small pitch-differences (see bottom of the new cage to identify a small and a big gap near together), but the aim was reached: the whistling noise never occurred with the randomized cage.

16

How to avoid annoying rolling bearing noises

Figure 15: Comparison of standard cage and new cage with randomized pitch

5 Rattle and mosquito noise – cage instability 5.1 Noise phenomenon A so called mosquito noise was noticed at some pre-series cars with dry double clutch. It appeared at standstill and cold temperature (~0 °C). The Campbell diagram below shows the mosquito noise measured on a test rig where rattle noise was also noticed.

Figure 16: Rattle and mosquito noise

17

How to avoid annoying rolling bearing noises The noises were caused by the cover bearing. As the slave cylinder is supported by the clutch cover and not by the gearbox, it includes a cover bearing to assure the rotation of the clutch while the slave cylinder stands still (figure 17).

Figure 17: Location of the cover bearing

5.2 Root cause The visualization of the bearing using a stroboscope showed a trembling cage during the rattle noise. During the mosquito noise the cage deforms itself in form of waves. The use of a high speed cam led to better records (figure 18). Thus the noises were caused by cage instabilities.

18

How to avoid annoying rolling bearing noises

Figure 18: Cage deformation during mosquito noise (using a high speed cam)

The cage behavior could be analyzed by using a simulation tool developed in house at Schaeffler called CABA 3D. This is a rolling bearing multi-body simulation program that allows to carry out realistic dynamic simulations of all roller bearings. A total of four cage states were found as shown below:

Figure 19: Cage states

The first state is due to the gravity: the cage is turning but its position still stays at the bottom, especially at low speeds and low friction.

19

How to avoid annoying rolling bearing noises The second state is the rigid instability. Figure 20 details how it proceeds from A to D: as the cage presses down against the two horizontally balls due to gravity (A), the friction between balls and cage pushes the cage to the right, which is now touching the two vertically balls (B), then the friction pushes the cage upwards (to C) and so on… The third state (elastic instability) is also due to the friction as shown in figure 21. The deformation of the cage induces new friction points which pull the cage in or out depending on the position around the cage. The last state is a mix of the elastic and the rigid instability.

Figure 20: Rigid instability

Figure 21: Elastic instability

Further simulations showed the states as a function of speed and friction at constant axial load (figure 22), and as a function of axial load and friction at constant speed (figure 23). As long as the friction keeps low, the states 2, 3 and 4 can not occur.

20

How to avoid annoying rolling bearing noises

Figure 22: States as function to friction and speed at constant load

Figure 23: States as function to friction and axial load at constant speed

5.3 Solution As the root cause is the friction between the balls and the cage, several solutions were investigated concerning the grease, the pocket geometry and the pocket clearance without losing sight of the cage guidance function.

21

How to avoid annoying rolling bearing noises The final solution could avoid the rattle and the mosquito noise at the complete temperature range from -30 °C to 120 °C and has also passed the endurance tests.

6 Summary Due to their benefits in terms of fuel consumption rolling bearings are entering new applications by replacing plain bearings on one side, and by new bearing applications on the other side. This paper resumes some rolling bearing noises, which occurred during the development process of different applications. The goal was to give some helpful techniques or ideas when facing bearing noise issues: – Calculation of rollover frequencies (BPF) If the internal geometry of the bearing is known it is possible to calculate the rollover frequencies (including higher harmonics) as a function of speed. – Considering waviness on raceways A short waviness with low amplitude can be sufficient to cause a loud whistling noise, if the waviness order matches with a factor of the number of rolling elements, or if the waviness order is high compared to the number of rolling elements. Be sure your grinding process avoids the creation of waviness by avoiding whole number speed ratios. – When ball passing frequencies meet eigenfrequencies Eigenmodes can be excited by the BPFs, especially the overlap of the higher harmonics. Reducing the amplitude of the BPFs can be done quite simply by a stochastic cage with randomized pitch. – Cage instability Cage instability can occur due to the friction between the rolling elements and the cage. The cage vibrations can be visualized using a stroboscope or a high speed camera. Optimized grease, pocket geometry and pocket clearance can solve the problem.

7 References / Literature [1] Grillenberger, H.: „Elastische Käfiginstabilität in Wälzlagern – Simulation und Versuch“, Antriebstechnisches Kolloquium, Aachen, 7 – 8 March 2015. [2] Lösch, J.: LuK internal reports, whistling noise CVT.

22

Engine NVH performance improvements with polymer gears Dipl.-Ing. Ralf Weidig, Dr.-Ing. Björn Fink, Frank J. Ferfecki, Tony Whitehead, Justin Salisbury

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 W. Siebenpfeiffer (Hrsg.), Automotive Acoustics Conference 2015, Proceedings, https://doi.org/10.1007/978-3-658-27648-5_12

1

Engine NVH performance improvements with polymer gears

Abstract Legislation drives the automotive megatrend to reduce fuel consumption and CO2 is resulting in small high output 3 and 4 cylinder engines. While customers have raised expectations on driving comfort, the smaller high output engines have inherent vibration noise, vibration and harshness (NVH) issues. Thus gear rattle in mass balance systems, cam gears, pump and accessory drives is an NVH issue. The most common approach is to use a split (scissor) gear assembly; the solution is complex, high in cost and heavy. High performance polymer injection molded net shape gears made from VICTREX® PEEK have gained growing attention and acceptance as cost effective, low mass and highly efficient solution to significantly reduce NVH in engine related application areas like mass balancer, cam and pumps. This paper presents test data done with validation customers and independent research institutes. It also references and discusses technical publications at SAE and VDI. With a successful case study it further it demonstrates the benefits of a net shape polymer gear solution offering a 3 dB (~50%) improvement in NVH, a 68% weight reduction resulting in 78% reduction of the moment of inertia and therefore 9% less torque required to operate compared to a cast iron gear set.

Introduction Over many years polymer gears already have established a strong position within various automotive application areas contributing to the overall requirements for cost reduction, supporting weight reduction and improving system performance and functionality. In areas like window lifters, door locking mechanisms, light adjustment and various actuation functions (turbo charger, EGR, electrical power steering, etc.) under-the-hood polymer gears and gear systems made from technical and engineering polymers have led to new, innovative applications delivering quiet, efficient, precise and durable performance. The automotive megatrend, driven by global legislation, to reduce the overall CO2 has opened up new demanding application spaces around and inside the powertrain. Due to the extreme environment with temperatures up to 150°C, high loads, longer lifetime requirements and aggressive engine or transmission fluids engineering polymers no longer deliver the required performance in this application space. The need to meet the CO2 requirements lead to small, high powered, highly charged 3- and 4-cylinder engines and therefore add a further increased need to reduce NVH. This smaller high output engines mostly have inherent vibration issues, while customers have increased expectations on quality. Gear noise in mass balance systems, cam gears, pump and accessory drives became an eminent NVH issue. The most common approach is to use a split (scissor) gear assembly or secondary damping measures; these solutions are complex, high in cost and add weight. High-performance polymer injection molded net shape gears have gained growing attention and acceptance as cost effective, low mass and highly efficient solu-

2

Engine NVH performance improvements with polymer gears tions to significantly reduce NVH in engine related application areas like mass balancer, cam and pumps. This paper is focusing on reviewing several studies done with high performance polymer (PEEK) gears to clearly show the potential for NVH and efficiency improvement for applications in modern downsized combustion engines. With a successful case study it further it demonstrates the benefits of a net shape polymer gear solutions offering a 3 dB(~50%) improvement in NVH, a 68% weight reduction resulting in 78% reduction of the moment of inertia and therefore 9% less torque required to operate compared to a cast iron gear set for a mass balancer module. Additional statements will be made for cam gears and replacing metal scissor gears.

Testing High-performance polymer PEEK is meeting all requirements While metal gears handle loads better than comparably sized polymer gears and have less dimensional and property variations with temperature and potential creep, polymer gears offer many cost, design, processing and performance advantages over metal combined with lower NVH, reduced weight and lower inertia. Further the molding process offers more design freedom enabling more efficient gear geometries and increased productivity over metal. In terms of design the aim should not be to copy the incumbent metal geometry, as this would not lead to an optimum usage of above mentioned benefits. Polymer gears usually have a higher contact ratio and special profiles leading to reduced contact and root stresses to improve performance and lifetime. In order to overcome the assembly challenges to fix a full polymer gear onto a shaft a proven solution would be to overmold metal inserts. While above comments are generic for most polymer gears there are more demanding requirements in an engine system where temperatures range from -40°C to 150°C. The gears in engine systems rotate as high as 14,500 rpm with a peak torque of 25 Nm. Chemical resistance is an important attribute required for the gears as they reside in an oil bath or mist and are also subject to small amounts of gasoline and diesel fuel.

3

Engine NVH performance improvements with polymer gears The injection moldable polymer material needs to meet several key engineering requirements in the application: ● ● ● ● ●

Mechanical strength and stiffness in high-heat environment (operating temperatures between -40°C to over 150°C) High wear resistance – in sliding applications Resistance to chemicals, solvents, fuels and lubricants (no chemical attack or degrading of properties) Dimensional Stability (no change in dimension due to moisture absorption) Needs to process on standard injection molding equipment

While there are many polymer materials that can meet one or two of the material requirements the only material that met all of them was a high performance polymer PEEK (polyetheretherketone), a member of the PAEK (polyaryletherketone) family. VICTREX® PEEK 450G and VICTREX® PEEK 650G polymers are used in the described evaluations.

Transmission gear noise reduction – gear rattle and efficiency Scope The study did focus on evaluating the potential for polymer gears in comparison to metal gears for noise reduction simulating typical automotive conditions. Two aspects were investigated: the impact on gear rattle due to speed fluctuations and the impact of transmission error over a load range at typical engine operating temperatures and speeds. The tests were conducted on a test rig with parallel shaft transmission (Picture 1, 2). The polymer test gears were supplied by Victrex molded in VICTREX® PEEK 650G representing a typical automotive engine application. The iron gears were machined to identical dimensions. Two types of tests were conducted using both the PEEK and the iron gears: a gear rattle test to quantify the noise properties during speed fluctuations and a transmission error test to compare the efficiency and noise under constant load [1].

4

Engine NVH performance improvements with polymer gears

Picture 1: Schematic for test fixture to measure rattle [1]

Picture 2: Schematic for test fixture to measure load efficiency and transmission error

[1]

5

Engine NVH performance improvements with polymer gears

Rattle tests For the rattle response testing 1.5th and 2nd order was chosen to simulate the characteristics of 3- and 4-cylinder engines. The speed ranges were extended above and below the typical speeds where rattle usually is observed in automotive. Table 1: Rattle test input speeds and fluctuations [1]

The results did indicate that the iron gears produced a sharp sound, while the PEEK gears did sound softer and slightly hollow. The recorded sound files at 500 and 3.000 rpm with 330 rpm fluctuations clearly underline this observation. In picture 3 and 4 the sound pressure levels have been plotted for all conditions further indicating the significant performance improvement of the polymer gears.

Picture 3: Sound pressure levels vs. speed fluctuations for iron gears [1]

6

Engine NVH performance improvements with polymer gears

Picture 4: Sound pressure levels vs. speed fluctuations for PEEK gears [1]

Transmission error and efficiency tests The transmission error and efficiency tests were run at 500, 1.000, 2.000 and 3.000 rpm and load steps up to 10Nm were applied. Picture 5 shows the percentage improvement in efficiency of PEEK gears to iron gears. The PEEK gears were up to 4% more efficient except at 500 rpm.

7

Engine NVH performance improvements with polymer gears

Picture 5: Percentage improvement in efficiency of PEEK gears vs. iron gears [1]

In picture 6 the sound levels at various loads and speeds were plotted showing a strong difference in sound pressure with increased speed, while the variation of torque didn’t show much impact. The lack of variation of the noise levels of the polymer gears indicate that the transmission error didn’t change much during the test due to load sharing with increased contact ratio.

8

Engine NVH performance improvements with polymer gears

Picture 6: Sound level comparison between iron and PEEK gears at various loads and speeds [1]

NVH evaluation on machined and molded PEEK gears Scope The scope of the program was to test and develop molded net shape PEEK gears suitable to operate under typical mass balancer conditions with temperature ranges of -40 – 150°C, speeds up to 14.500 rpm, peak torques up to 25 Nm and engine oil lubrication. The PEEK gear solution was expected to deliver required NVH improvement and durability. The chosen gear materials were VICTREX® PEEK 450G and 450CA30 (30% carbon fibre) machined from stock shape and molded with a slightly optimized geometry [2]. In order to evaluate the influence of material choice, gear geometry and backlash on NVH performance a detailed Design of Experiments (DOE) was conducted in a calibrated NVH test cell. Picture 7 shows the used production balancer shaft module and picture 8 details of the test stand. Table 2 shows the details of module configurations for the NVH tests.

9

Engine NVH performance improvements with polymer gears

Picture 7: Production balancer shaft module for NVH test [2]

10

Engine NVH performance improvements with polymer gears Table 2: DOE – module configuration for NVH test [2]

Picture 8: NVH test stand with accelerometer and microphone positions [2]

11

Engine NVH performance improvements with polymer gears All the tested gears, which have been optimized for strength, durability and moldability, had 48 teeth. As the gears rotate at twice the engine speed the data have been analysed at 96th order. Using Minitab the accelerometer and microphone data were analysed and showed similar patterns for the various combinations, but all within the range of 1dB. Picture 9 shows the main effects of the microphone data for the machined gears.

Picture 9: Main effects plot for 96th order microphone data for machined gears [2]

● ● ● ●

450G and 450CA30 are comparable and offer a 1dB improvement over iron High and nominal backlash are equal, low the worst, all within 1dB range Iron driver mating 450G is best PEEK mating PEEK didn’t provide any benefits

With an identical test set up molded VICTREX® PEEK 450G gears with 5° helix have been run varying backlash and mating gear. Again the iron driver mating with a 450G gear with nominal backlash was best now showing a 3 dB improvement in the microphone data over the production gears (picture 10).

12

Engine NVH performance improvements with polymer gears

Picture 10: Interaction plot for 96th order microphone data for molded gears [2]

Additional durability testing with the molded 450G gears has been done using a mass balancer module similar to the NVH testing. Test conditions varying the backlash and the mating gear material were ● ● ● ●

Temperatures of 100 and 150°C Engine oil – 5W-20 6.000 rpm engine speed (12.000 rpm gear speed) Duration 300 h (432 mio. cycles)

All tests did survive the durability test showing no signs of wear and change in backlash. A further test with a PEEK-PEEK combination at 150°C under the same conditions did not show any wear after the full test sequence either (picture 11). As previously noted there was no difference in NVH compared to an iron-PEEK gear pair but it is believed, that a PEEK-PEEK solution might have benefits in long-term durability on the flank due to lower contact pressure. Further the PEEK-PEEK molded gears were on average 2dB quieter in NVH performance on the microphone readings vs. the production gears.

13

Engine NVH performance improvements with polymer gears

Picture 11: VICTREX® PEEK 450G driven gear after durability test at 150°C [2]

Picture 12: VICTREX® PEEK 450G driver gear after durability test at 150°C [2]

Additional testing was done at 80°C to measure the torque required to drive the modules at various revolutions. Here the polymer solution did provide an improvement of ● ● ●

9% at 2.000 rpm 4% at 4.000 rpm 3% at 12.000 rpm

over the iron gear system. This performance is a result of to the 69% lower weight of the PEEK gears, which is equal to in a 78% reduction of the moment of inertia.

14

Engine NVH performance improvements with polymer gears

Conclusions Based on this work Metaldyne, a leading USA mass balance shaft system supplier, has developed a module achieving smoother running and higher system efficiency by replacing ground metal gears with VICTREX® PEEK gears. The enhanced NVH performance and improved fuel efficiency of this module put the company ahead of their competition. The solution was developed in close collaboration with design engineers of Metaldyne, the US gear designer, manufacturer Kleiss Gears and Victrex resulting in a novel gear design. The solution offers a 3 dB (~50%) improvement in NVH compared to a cast iron gear set, a 68% weight reduction resulting in 78% reduction of the moment of inertia and therefore 9% less torque required to operate the mass balancer module and is currently validated at several OEMs. First efforts in replacing metal gears in cam drives have shown similar NVH reductions and led to a better sound quality while passing the tough requirements of the engine durability test on the dynamometer and in the field. Compared to metal scissor gears PEEK gears can provide similar noise levels and are offering significant weight benefits and cost reduction potential. High-performance polymer (PEEK) gears have proven the potential for significant noise reduction while offering the required durability for mass balancer modules used in today’s downsized combustion engines. Various ongoing validations in engine related gear applications underline the increased interest and need of the automotive industry to develop and commercialize a viable solution. These efforts will lead to increased acceptance of high performance PEEK gear solutions in a so far traditional metal domain.

References [1] John Morton, Drive Systems Design Ltd., George Bailey, Southwest Research Institute: “The use of plastic in automotive gears”, VDI Transmission Conference, Germany 2012 [2] Alan Hale, Metaldyne Inc., Frank J. Ferfecki, Victrex USA: „Polymer gear development to improve efficiency and NVH performance of an engine mass balance system, SAE paper 2011-01-0405, SAE Conference 2011.

15

New road noise testing techniques Hartmut Bathelt Heiko Kolm Kay Schammer

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 W. Siebenpfeiffer (Hrsg.), Automotive Acoustics Conference 2015, Proceedings, https://doi.org/10.1007/978-3-658-27648-5_13

1

New road noise testing techniques

1 Introduction Reviewing the last 20 years of automotive development we realize a convincing progress in NVH behaviour with each model generation. Balancing shafts and reduction of engine speed make engine noise almost inaudible even with 4 cylinder engines and acoustic treatment of soft tops bring down wind noise of cabriolets to limousine level. Nevertheless there is one discipline with the least amount of progress, that’s road noise. It seems that a big part of the progress in sus-pension development has been invested into crisp handling and comfort improvements have almost been eaten up by the fashion trend to 20 inch rims and extreme low section tires. As a result road noise still is the dominant annoyance in long distance driving. At least the manufacturers of premium cars seem to face this problem. In the last 2 years almost every second of new roller benches ordered is a special road noise dyno with 3,2m-rollers or at least a multi-purpose NVH dyno with rough road shells for road comfort tests.

2 Rough surface on rollers: a calibration tool for road noise The interior noise measured on rough road surfaces always is a relative quantity related to that specific surface. This leads us to the issue with every new road noise test facility how to make results comparable to existing rough surface rollers or test tracks on a proving ground.

2.1 Traditional technique: casted road surface print 30 years ago mold making was the standard technique to take a print of the road surface and cast an aluminum ring. This ring was pressed on a roller as one piece or cut into sections for detachable road shells. A major drawback of this technique is evident when we look at the roughness profile along the roller surface:

2

New road noise testing techniques

Picture 1. Elevation profile of rough roller circumference: colour coded (top) cross section (below)

With the print of the road surface also long wave unevenness was copied, which leads to an unrealistic periodic excitation by the repetition of the track sample with roller rotation. The induced shake of the test car and the modulation of the sound intensity are a disturbance of the subjective impression driving on the rollers. The machining of the ring castings inside even may add an additional first order amplitude to the rollers.

2.2 Road shells with gravel stones A completely different technology –developed 1998- eliminated this problem: the rough surface was produced in a similar way as a Macadam road: gravel stones were glued on 8mm purely cylindrical aluminum plates. Weight reduction and easier handling was a further advantage.

3

New road noise testing techniques

Picture 2. Stone covered shells mounted on 3,2mØ-rollers

With this technique we do not get a 100% copy, just a similar surface by using the same kind of gravel like on the test road. The same shape of the excitation spectra is achievable, but the absolute level may deviate compared to the road.

4

New road noise testing techniques

2.3 Precision milled road shells Today`s CNC milling and laser measurement technology open up new ways. Laser scanning creates a digital picture of the rough surface. These data can be edited in a similar way like vibration signals. AZL has developed a filtering code to reduce or eliminate long wave unevenness up to an arbitrary wavelength, as you can see in this example:

Picture 3. Elevation profile of rough roller surface along 10m circumference

The filtered surface data now can be fed to a milling machine, which precisely works a copy of the road surface into curved aluminium plates, but without copying disturbing unevenness. The new Audi road noise dyno was the first one, where the rough surface of 10 m length was directly milled into the steel barrel of a 3,18mØ-roller:

5

New road noise testing techniques

Picture 4. Direct milling of digital surface data into rollers steel

The surface unevenness in this case was not completely eliminated but reduced to 30% of the original one. This compromise was chosen between the objective to achieve comparable test data to the existing 20 years old rollers and the wish to reduce the annoying periodic bounce of the vehicle. Above all it is unknown, if a total elimination of unevenness is realistic in comparison to a drive on a real road. It could lead to a condition with locked shock absorbers – when the excitation amplitudes are too small to brake away the friction in the damper tube. The damper acting like a rigid rod could change the structural noise transmission on this path. The comparison of the interior noise (Picture 5) and wheel hub accelerations (Picture 6) measured in the same car immediately after each other shows that the objective number one –same road noise excitation- has been achieved. The differences in wheel hub acceleration of up to 10 dB at frequencies below 30 Hz make the impact of filtering the long wave unevenness apparently. For the road noise above 30 Hz no difference is visible.

6

New road noise testing techniques

Picture 5. Comparison of interior road noise measured on existing and new dyno

Picture 6. Wheel hub accelerations (left wheel, vertical) on existing and new road noise dyno

7

New road noise testing techniques

3 Coherence problem left wheel to right wheel Once we make the step from measuring the current state of the suspension comfort to the mission of improvement, we usually go for a TPA and/or an ODS-analysis. With all these advanced tests an exact phase relation between measurement points is crucial. This leads to the question: which signal should be taken as a phase reference? Usually the input signal –force or acceleration- is the first choice. On real road we have two inputs, left wheel and right hand wheel, and these two inputs are incoherent or just partially coherent. On a dyno we have the option to use the same surface on both rollers.

3.1 Does the arrangement of random roughness influence measured road noise data? In the present case the existing 3,2mØ-rollers used identical road prints left and right, but the surface rings were shifted 90 degrees relative to each other. Since the new dyno has independent rollers we are able to investigate the influence of this offset by measuring wheel hub accelerations and interior noise in original roller condition (90 degree offset) and zero offset (both tracks in phase).

Picture 7. Coherence hub accelerations left wheel to right wheel (z direction)

8

New road noise testing techniques Although the tracks are not 100% identical, the coherence in the low frequency range up to 70Hz is almost 1 with rough surfaces in phase. We could assume a mirror image arrangement of the surfaces left to right would lead to a coherence value of 1,0 up to higher frequencies, means identical excitation of both wheels.

3.2 Does the coherence between left wheel and right wheel excitation have influence on the measured interior noise?

Picture 8. Road noise measured with left to right roller surfaces in phase and 90 degree offset Excitation of front axle

We can see a 3dB higher level at frequencies where the coherence is almost 1,0.

9

New road noise testing techniques When we calculate the transfer function microphone signal to wheel hub acceleration the difference turns out up to 10dB:

Picture 9. Transfer function left microphone related to left wheel hub acceleration (vertical)

10

New road noise testing techniques The coherence function again is significantly higher with in phase excitation (red):

Picture 10. Coherence interior noise (left front mic) to vertical wheel hub acceleration left side

3.3 What are the implication for measurements and data analysis? 3dB higher interior noise with in phase excitation of both wheels rather than random phase can be explained by usual energy considerations. On transfer functions a high value of coherence usually is seen as a quality feature. This could be an argument to strive for identical excitation on both wheels as it is practice with impact bar excitation. On rough road rollers it now can be achieved with the new technique of using identical surface data mirrored about the longitudinal axis and aligning the vehicle exactly in the middle. Another way to avoid a multiple coherence problem is excitation of just one wheel with the other wheel running on a smooth roller. In the past this approach gave good results in a minimum of test time. Of course this is just an option with removable road shells. Practical experience in the future will show which kind of excitation gives the best results.

11

New road noise testing techniques

4 Repetition of signal on rollers: harmonic orders A basic difference to real road is the repetition of the random excitation after each rotation of the rollers. One way to at least reduce this effect is the use of 3,18mØ-rollers giving 10m of driving length rather than the usual 6m of standard rollers. Nevertheless we do not achieve a pure random excitation. We will always detect harmonic orders of roller speed in our measured signals especially at frequencies lower than 50Hz. Therefore it is usual practice not to collect road noise data at constant speed, but to average data of a uniform speed ramp. Averaging has the effect of “smearing” the roller-harmonic components over the frequency range, so all resonance peaks in the measured spectra are vehicle related and do not contain roller frequencies. All spectra shown in picture 5 to 10 were measured during a run up from 20 to 120kph within 20 sec, averaged with 50% overlap using Hanning windows and analysed with 1 Hz frequency resolution.

4.1 Closed ring better than road shells? Sometimes the assumption is expressed that removable road shells will increase the harmonic content by the repetition order of the plates. A direct comparison of the order lines fan out in the pictures 11 and 12 below does not really confirm this suspicion: the 14th order (ending at about 40Hz at 95kph) corresponding to 14 road shells on the circumference seems not to be higher measured on the rollers with removable shells (picture 2), than on closed ring rollers (color maps below).

12

New road noise testing techniques

Picture 11. Roller orders with 14 road shells on 10m circumference Wheel hub accelerations in horizontal and vertical direction

Picture 12. Roller orders on 10m closed ring surface Wheel hub accelerations in horizontal and vertical direction

13

New road noise testing techniques

5 Flexibility against Speed: road shells or fixed surface? When different kinds of rough surfaces like on a proving ground should be available for subjective assessment in the lab, removable road shells are the obvious solution. As an example we see a choice of surfaces in the following pictures:

Picture 13. Road shells with different surfaces

The price of the availability of different surfaces is the installation time of 40 to 60 minutes mounting two times 14 shells on a set of 3,2mØ-rollers. The objective for the new AUDI test bench is a multiple purpose dyno with 2 driveways on each roller: a smooth one for engine noise tests and a rough one for road noise.

14

New road noise testing techniques Rapid change cycles in this case are the dominant priority. Two fixed surfaces on each roller lead to a roller width of 820mm:

Picture 14. Two driveways on new AUDI road noise dyno

The asymmetric position of the test car on the rollers allows good access to just one side of the cars underbody. Extensive design of the dyno structure led to a maximum width of the working pit in spite of the limited space between the rollers (picture 15). Another contribution to maximum efficiency of test work is the one point car fixation, which has proven to achieve the fastest set-up time by self-aligning of the test car.

15

New road noise testing techniques

Picture 15. Pit between rollers: a comfortable working place

Picture 16. One point car fixation with traction force gauge

16

New road noise testing techniques

6 Conclusion The new technique of producing rough surfaces by precision milling opens up ways to tailor the road noise excitation on rollers especially for sophisticated analysis methods. ● ●

●

The removal of annoying bounce and sound modulation of a test car by long wave unevenness improves subjective assessment. Digital data processing of a laser scanned surface offers a mirror image alignment of left and right roller surface roughness. With independent driven rollers and exact angular control an identical excitation of both wheels can be chosen, or a rather incoherent excitation can be achieved with an offset left to right. The micro structure of a rough surface now can directly be milled into a steel roller.

Aluminium rings are needless.

17

Extension of acoustic holography to cover higher frequencies Jørgen Hald, Brüel & Kjær SVM A/S, Denmark

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 W. Siebenpfeiffer (Hrsg.), Automotive Acoustics Conference 2015, Proceedings, https://doi.org/10.1007/978-3-658-27648-5_14

1

Extension of acoustic holography to cover higher frequencies

1 Introduction Near-field Acoustical Holography (NAH) is based on performing 2D spatial Discrete Fourier Transforms (DFT), and therefore the method requires a regular mesh of measurement positions. To avoid spatial aliasing problems, the mesh spacing must be somewhat less than half of the acoustic wavelength. In practice, this requirement sets a serious limitation on the upper frequency limit. Some Patch NAH methods, for example the Equivalent Source Method (ESM) [1] and Statistically Optimized NAH (SONAH) [2-3], can work with irregular microphone array geometries, but still require an average array element spacing of less than half the wavelength. As described by Hald [4], this allows the use of irregular arrays that are actually designed for use with beamforming. Typically, good performance with beamforming can be achieved up to frequencies where the average array inter-element spacing is two to three wavelengths. A practical consequence with such a solution is the fact that the Patch NAH method requires measurement at a small distance to provide good resolution at low frequencies, while beamforming requires a medium-to-long distance to keep sidelobes at low levels. So for optimal wide-band performance, two measurements must be taken at different distances, and separate types of processing must be used with the two measurements, making it difficult to combine the results into a single result covering the combined frequency range. The rather new Compressive Sensing (CS) methods have started making it possible to use irregular array geometries for holography up to frequencies where the average array inter-element spacing is significantly larger than half of the wavelength, see for example [5] and [6]. In general, these methods allow reconstruction of a signal from sparse irregular samples under the condition that the signal can be (approximately) represented by a sparse subset of expansion functions in some domain, i.e., with the expansion coefficients (amplitudes) of most functions equal to zero. The underlying problem solved is that at high frequencies the microphone spacing is too large to meet the spatial sampling criterion, and thus there is no unique reconstruction of the sound field. A reconstruction must therefore have a built-in “preference” for specific forms of the sound field. Doing just a Least Squares solution will typically result in reconstructed sound fields with sound pressure equal to the measured pressure at the microphones, but very low elsewhere. By building in a preference for compact sources, a smoother form of the reconstructed sound field is enforced. The present paper describes a new method called Wideband Holography (WBH), which was introduced in reference [6], and which is covered by a pending patent [7]. The method is similar to the Generalized Inverse Beamforming method published by T. Suzuki [8]. However, instead of applying a 1-norm penalty to enforce sparsity in a monopole source model, WBH uses a dedicated iterative solver that enforces sparsity in a dif-

2

Extension of acoustic holography to cover higher frequencies ferent way. The main contribution of the present paper is a comparison of WBH results and performances with those of a method that solves an optimization problem with 1-norm penalty. Section 2 outlines the basic theory. After an introduction to the applied array designs in section 3, results of different simulated measurements are presented in section 4, and finally section 5 contains the conclusions.

2 Theory Input data for patch holography processing is typically obtained by simultaneous acquisition with an array of M microphones, indexed by m = 1,2…,M, followed by averaging of the M × M element cross-power spectral matrix between the microphones. For the subsequent description, we arbitrarily select a single high-frequency line f with associated cross-power matrix G. An eigenvector/eigenvalue factorization is then performed of that Hermitian, positive-semi-definite matrix G:

G  VSV H ,

(1)

V being a unitary matrix with the columns containing the eigenvectors vµ, µ = 1,2….M, and S a diagonal matrix with the real non-negative eigenvalues s on the diagonal. Based on the factorization in formula (1), the Principal Component vectors p can be calculated as:

p   s v  .

(2)

Just like ESM and SONAH, the WBH algorithm is applied independently to each of these principal components, and subsequently the output is added on a power basis, since the components represent incoherent parts of the sound field. So for the subsequent description we consider a single principal component, and we skip the index , meaning that input data is a single vector p with measured complex sound pressure values. WBH uses a source model in terms of a set of elementary sources or wave functions and solves an inverse problem to identify the complex amplitudes of all elementary sources. The source model then applies for 3D reconstruction of the sound field. Here we will consider only the case where the source model is a mesh of monopole point sources retracted to be behind/inside the real/specified source surface, i.e., similar to the model applied in ESM [1]. With Ami representing the sound pressure at microphone m due to a unit excitation of monopole number i, the requirement that the modelled sound pressure at microphone m must equal the measured pressure pm can be written as: I

pm   Ami qi .

(3)

i 1

3

Extension of acoustic holography to cover higher frequencies Here, I is the number of point sources, and qi, i = 1,2…,I, are the complex amplitudes of these sources. Equation (3) can be rewritten in matrix-vector notation as:

p  Aq ,

(4)

where A is an M × I matrix containing the quantities Ami, and q is a vector with elements qi. In Compressive Sensing terminology the matrix A is called the Sensing Matrix. When doing standard patch holography calculations using ESM, Tikhonov regularization is typically applied to stabilize the minimization of the residual vector − . This is done by adding a penalty proportional to the 2-norm of the solution vector when minimizing the residual norm: 2

2

Minimize p  Aq 2   2 q 2 . q

(5)

A very important property of that problem is the fact that it has the simple analytic solution:



q  A H A   2I



1

AHp ,

(6)

where I is a unit diagonal matrix, and H represents Hermitian transpose. A suitable value of the regularization parameter  for given input data p can be identified automatically, for example by use of Generalized Cross Validation (GCV), see Gomes and Hansen [9]. When using a specific irregular array well above the frequency of half wavelength average microphone spacing, the system of linear equations in formula (4) is in general strongly underdetermined, because the monopole mesh must have spacing less than half of the wavelength, i.e., much finer than the microphone grid. During the minimization in formula (5), the undetermined degrees of freedom will be used to minimize the 2-norm of the solution vector. The consequence is a reconstructed sound field that matches the measured pressure values at the microphone positions, but with minimum sound pressure elsewhere. Estimates of, for example, sound power will therefore be much too low. Another effect is ghost sources because available measured data is far from determining a unique solution. This will be illustrated by simulated measurements. If the true source distribution is sparse (with a majority of elements in q equal to zero), or close to sparse, the above phenomena can be alleviated by replacing the 2-norm in the penalty term of formula (5) by a 1-norm: 2

Minimize p  Aq 2   2 q 1 , q

(7)

see for example references [5] and [8]. Important problems related to this formulation are the lack of an analytic solution and the fact that no good tool is available to identify an optimal value of the regularization parameter  for given input data p. An equivalent problem was solved by Chardon et al. [5]:

4

Extension of acoustic holography to cover higher frequencies

Minimize q 1 q

subject to

p  Aq 2   .

(8)

Here, however, the parameter  is difficult to determine. In cases where the applied source model cannot represent the full measured sound field (for example due to reflections), a rather large value of  may be needed for the problem to be solvable. Instead of ‖ , we can alternatively require a solution close to a requiring a small residual ‖ − minimum of that residual, which will be characterized by the gradient vector ,

w (q )  A H p  Aq 

(9)

of the squared residual function 12 ‖ −

Minimize q 1 q

subject to

‖ being small:

w (q) 2  A H p  Aq    . 2

(10)

The optimization problem of formula (10) is convex and can be solved quite efficiently by available Matlab libraries. In the present paper the CVX library has been used, see references [5] and [10], so the method will be called just “CVX”. Still, the computational demand is significantly higher than for the Tikhonov problem in formula (5) because no analytic solution exists. The minimization of the 1-norm of the solution vector will have the effect of favouring sparse solutions. According to experience, a good way of defining the parameter  in formula (10) is

   w (q  0) 2   A H p , 2

(11)

where  is a small number. Too small values will, however, make the method very sen‖ sitive to noise/errors in the measured data. The same condition ‖ ( )‖ ≤ ‖ occurs also in the stopping criterion of the iterative solution method implemented in Wideband Holography (WBH), but only as one of several conditions that will imply stopping. A main idea behind the WBH method is to remove/suppress the ghost sources associated with the real sources in an iterative solution process, starting with the strongest real sources. In the following we will just highlight some of the most important points – a detailed mathematical description is given in reference [6]. ‖ WBH applies a Steepest Descent iteration to minimize the squared residual ‖ − of formula (4). In the first step a number of real as well as ghost sources are introduced/identified in . When using irregular array geometries, the ghost sources will in general be weaker than the strongest real source(s). We can therefore suppress the ghost sources by setting all components in below a certain threshold to zero. The threshold is computed as being a number of decibels below the amplitude of the largest element.

5

Extension of acoustic holography to cover higher frequencies The dynamic range of retained sources, Dk, is updated during the iteration steps, k, in such a way that an increasing dynamic range of sources will be retained, typically:

Dk 1  Dk  D .

(12)

In the limiting case when   for  , the dynamic range limitation is gradually removed. However, the iteration is stopped when:

Dk 1  Dmax

or

w(q k 1 ) 2   w(q  0) 2   A H p , 2

(13)

where Dmax is an upper limit on Dk and  is a small number. The following values have been found to work in general very well: D0 = 0.1 dB, ΔD = 1.0 dB, Dmax = 60 dB and  = 0.01. Due to slow final convergence of the steepest descent method, the first of the two criteria in formula (13) will usually be first fulfilled. The upper limiting dynamic range Dmax can be changed to match the quality of data, but the choice does not seem to be critical. Dmax = 60 dB has been found to support the identification of weak sources, even when measurements are slightly noisy. Larger values do not seem to improve much. Smaller values may be required for very noisy data. Starting with only 0.1 dB dynamic range means that only the very strongest source(s) will be retained, while all related ghost sources will to be removed. When we use the dynamic range limited source vector as the starting point for the next iteration, the components of the residual vector related to the very strongest source(s) have been reduced, and therefore the related ghost sources have been reduced correspondingly. Increasing the dynamic range will then cause the next level of real sources to be included, while suppressing the related ghost sources, etc. Another aspect is the fact that a minimum number of the point sources of the model will be assigned an amplitude different from zero, enforcing effectively a sparse solution. After the termination of the above algorithm based on steepest descent directions, a good estimate of the basic source distribution has been achieved. Final convergence is, however, very slow, exhibiting so-called zigzagging. Good progress can often be ‖ in a direcachieved by an extrapolation step, which minimizes the residual ‖ − tion defined as the average of the two latest steepest descent steps. A few Conjugate Gradient iterations without dynamic range limitation can then optionally be performed ‖ . Usually, the to ensure convergence to a point very close to a minimum of ‖ − effect on the source model and the modelled sound field is very small, because the primary Steepest Descent algorithm has already reduced the residual to be close to a minimum, but it ensures that full convergence has been achieved. The stopping criteria used with the conjugate gradient method are:

6

Extension of acoustic holography to cover higher frequencies

w(q k 1 ) 2   A H p

2

or

w(q k 1 ) 2  w(q) 2 .

(14)

In comparison with the CVX method defined in formulae (10) and (11) notice that selection of a too small value of  will not destabilize the conjugate gradient algorithm of WBH, since it will then stop, when the gradient norm starts increasing. When that happens, the last step is discarded. The WBH algorithm, which enforces a maximum degree of sparsity in the source distribution, has been found to work well at high frequencies, when a suitable array is used at a not too small measurement distance. However, at low frequencies WBH easily leads to misleading results, when two compact source are so closely spaced that available data does not support a resolution of the two with beamforming. In that case, the WBH algorithm will often identify a single compact source at a position between the two real sources, so the user might be drawing wrong conclusions about the root cause of the noise. Use of the traditional Tikhonov regularization of formula (6), i.e. a standard ESM algorithm, will in that case typically show a single large oblong source area covering both of the two real sources. To minimize the risk of misleading results, it is recommended to use the standard ESM solution up to a transition frequency at approximately 0.7 times the frequency of half wavelength average array inter-element spacing (i.e. spacing ≈ 0.35λ), and above that transition frequency switch to the use of WBH. See reference [6] for details. The CVX method shows similar behaviour, so it should also be supplemented by a different algorithm at low frequencies.

3 Array Design As described in the introduction, the method of the present paper follows the principles of Compressive Sensing, being based on measurements with a random or pseudorandom array geometry in combination with an enforced sparsity of the coefficient vector of the source model. The array geometry used in the simulated measurements of the present paper is shown in Figure 1. It has 12 microphones uniformly distributed in each one of five identical angular sectors. The average element spacing is approximately 12 cm, implying a low-to-high transition frequency close to 1 kHz (where 0.35λ is close to 12 cm). The geometry has been optimized for beamforming measurements up to 6 kHz as described in reference [4].

7

Extension of acoustic holography to cover higher frequencies

Figure 1: Geometry of the applied planar pseudo-random 60-element microphone array with 1 m diameter

An important finding from simulated measurements with the chosen array design is that the measurement distance should not be shorter than approximately a factor two times the average microphone spacing for the method to work well at the highest frequencies. A factor of three is even better, and distances up to typically 0.7 times the array diameter work fine. When the measurement distance is increased, each source in the WBH source model will expose the microphones over a wider area, increasing the ability of the irregular array to distinguish different sources. To get acceptable low-frequency resolution, however, the measurement distance should not be too long either, so overall the best distance seems to be two to three times the average array inter-element spacing.

8

Extension of acoustic holography to cover higher frequencies

4 Simulated Measurements All CVX and WBH calculations in the present paper were performed using D0 = 0.1 dB, ΔD = 1.0 dB, Dmax = 60 dB and  = 0.01.

4.1 Single monopole point source The aim of single-monopole simulated measurement is to demonstrate: i) What happens if Tikhonov regularization is applied above the frequency of half wavelength average array element spacing. ii) How much and which kind of improvement is achieved by applying the sparsity enforcing CVX and WBH algorithms.

Figure 2: Arrangement for simulated measurement on a single monopole point source

As shown in Figure 2, we consider a setup with a monopole point source located on the array axis at 28 cm distance from the array plane, while the source-model mesh is at 27 cm distance, and the sound field in reconstructed in a “Source Plane” 24 cm from the

9

Extension of acoustic holography to cover higher frequencies array plane. The reconstruction mesh has 51 columns and 51 rows with 2 cm spacing, covering a 1 m × 1 m area centred on the array axis, and the source-model mesh is similar, i.e. with 2 cm spacing, but it is extended by 6 rows/columns in all four directions. In total, 63 × 63 = 3969 complex point-source amplitudes must be determined from the 60 measured complex sound-pressure values. No measurement errors/noise was simulated in the single-monopole measurements. Figure 3 shows from left to right the true 4 kHz sound intensity map on the Source Plane, followed by three reconstructed maps calculated using (i) Tikhonov regularization with 20 dB dynamic range (formula (6)), (ii) the CVX algorithm (formulae (10) and (11)) and, (iii) the WBH algorithm. The CVX and WBH maps are both very close to the true intensity map, as could be expected in the present case, although the source-model plane is 1 cm from the real monopole point source. The sound intensity reconstruction based on Tikhonov regularization shows a small low-level peak at the true source position, but in addition there are quite a lot of ghost sources. These ghost sources are responsible for the focusing of the radiation towards the microphones that can be seen in Figure 4. Figure 4 shows from left to right the true 4 kHz sound pressure level (SPL) on the array plane followed by corresponding reconstructed SPL maps calculated from the source model using (i) Tikhonov regularization with 20 dB dynamic range, (ii) CVX and, (iii) WBH. Looking at the Tikhonov result it is clear that the 2-norm minimization has used the heavily underdetermined nature of the problem to focus sound radiation towards the microphones to produce a sound pressure close to the measured pressure, while in all other directions the radiated sound is minimized. Figure 5 shows the implied underestimation of sound power. Using CVX or WBH to get source model amplitudes, the reconstructed array-plane SPL is close the true SPL map, although it has some small ripples.

Figure 3: Contour plots of sound intensity in the “Source Plane”, see Figure 2. Display range is 30 dB with 3 dB contour interval, and the same scale is used in all four plots

10

Extension of acoustic holography to cover higher frequencies

Figure 4: Contour plots of sound pressure in the array plane, see Figure 2. Display range is 15 dB with 1 dB contour interval, and the same scale is used in all four plots

As shown in Figure 5, the sound power is predicted accurately across the full frequency range, when CVX or WBH is used. When Tikhonov regularization is used, sound power underestimation increases quickly with increasing frequency above 1 kHz, since the ability of the source model to focus radiation only towards the microphones increases. Calculation times for the 64 frequencies represented in Figure 5, using Matlab implementations of the CVX and WBH methods, were 829 sec for CVX and 32 sec for WBH.

Figure 5: Relative sound power spectra from area-integration of the sound intensity maps in Figure 3. The sound power from the True intensity map is taken as the reference

11

Extension of acoustic holography to cover higher frequencies

4.2 Two monopole point sources with 10 dB level difference

Figure 6: Side view illustration of the simulated measurement on two point sources

A main purpose of this section is to demonstrate the ability of the CVX and WBH methods to identify weak sources in the presence of strong ones, even though by nature the methods will maximize the number of sources with amplitude equal to zero. We use a setup with two coherent in-phase monopole point sources located 29 cm in front of the array plane. Source 1 is at (x,y) coordinates (15,15) cm and source 2 is at (-15,-15) cm relative to the array axis, with source 1 excited 10 dB stronger than source 2. Figure 6 illustrates the setup as seen from the side, with the source-model mesh 25.5 cm from the array and the “Source Plane” (sound-field reconstruction plane) 24 cm from the array plane. Thus, in this case, the real sources are 3.5 cm behind the source model. The reconstruction mesh has 51 columns and 51 rows with 1 cm spacing, covering a 0.5 m × 0.5 m area centred on the array axis, and the source-model mesh is similar, i.e., with

12

Extension of acoustic holography to cover higher frequencies 1 cm spacing, but it is extended by 6 rows/columns in all four directions. In all the simulated measurements of this section, random noise was added to the complex microphone pressure data at a level 30 dB below the average sound pressure across the microphones.

Figure 7: Contour plots of sound intensity in the Source Plane, see Figure 6. Display range is 20 dB with 2 dB contour interval. The same scale is used in all four plots. Source 1 is the stronger source in the upper right corner, while source 2 is in the lower left corner.

Figure 8: Sound power spectra for source 1 and 2, obtained by area-integration of sound intensity maps as those in Figure 7.

Figure 7 shows the true and the reconstructed sound intensities at 5 kHz with a 20 dB display range. Clearly, the two sources are well identified by both CVX and WBH, and the two methods show very similar results. Actually, the maps look much the same at all frequencies between 1 kHz and 5 kHz. Sound power integration areas are shown with line style corresponding to the sound power spectra in Figure 8. Except for the weakest

13

Extension of acoustic holography to cover higher frequencies source 2 at the lowest frequencies, the two reconstruction methods estimate almost the same sound power spectra for the two sources. As mentioned at the end of section 2, a different solution method based on Tikhonov regularization should be used anyway at the lowest frequencies – for the present array up to 1 kHz. The small over-estimation of the source 1 sound power up to around 4 kHz is probably due to the stronger concentration of the intensity in the reconstructed intensity maps, implying that less power will be outside the integration area. Apart from a 2.5 dB dip around 4 kHz in the estimated source 2 power, accuracy is good up to around 5 kHz and above that frequency, an increasing under-estimation is observed. The maximum frequency of the present array (with 12 cm average microphone spacing) in connection with the SONAH and ESM algorithms is 1.2 kHz, so apparently the CVX and WBH methods extend the frequency range by a factor around 4. The calculation times for the 32 frequencies represented in Figure 8 were 490 sec for CVX and 16 sec for WBH, so again WBH is faster by approximately a factor 30. Another advantage of WBH is the already mentioned smaller sensitivity of WBH to the specified target reduction in the gradient norm, see formulae (10) and (11) for CVX, and (13) and (14) for WBH: A too small value of  causes the CVX method to become unstable.

14

Extension of acoustic holography to cover higher frequencies

4.3 Plate in a baffle 3 kHz

4 kHz

WBH Intensity

CVX Intensity

True Intensity

2 kHz

Figure 9: Contour plots at 2, 3 and 4 kHz of sound intensity in the reconstruction plane 1 cm above the plate. Display range is 20 dB with 2 dB contour interval as in Figure 7. For each frequency, the true sound intensity and the reconstructed maps use the same scale.

The aim of the simulated plate measurements is to show that the WBH method can give quite good results, even when the true source distribution is not sparse. As an example of a more distributed source, a baffled, centre-driven, simply supported, 6 mm thick, 40 cm × 40 cm aluminum plate has been used. The coincidence frequency for the plate is at 2026 Hz. The vibration pattern was calculated using the formulation by Williams [11], and subsequently the radiated sound field was obtained using the discretized Rayleigh integral, approximating the plate velocity distribution by 161 × 161 monopole point sources. This allowed the microphone sound-pressure values and the “true” pressure and particle velocity in a reconstruction plane 1 cm above the plate to be calculat-

15

Extension of acoustic holography to cover higher frequencies ed. As for the simulated measurements on two monopole point sources, random noise was added to the complex microphone pressure data at a level 30 dB below the average sound pressure across the microphones. The reconstruction mesh had 41 × 41 points with 1 cm spacing, covering exactly the plate area, and the array was placed 24 cm above the plate. For the WBH sound field reconstruction a source model comprising 53 × 53 monopole point sources with 1 cm spacing was located 1 cm below the plate. Figure 9 shows the true sound intensity and the corresponding CVX and WBH reconstructions at 2, 3 and 4 kHz with a 20 dB display range. Overall the reconstruction is good, with a bit too high weight on the central area, the two methods performing again very equal. At 4 kHz the reconstructed intensity patterns start getting distorted, because the amount of information in the vibration pattern becomes too large in relation to the data provided by the array. As mentioned earlier, the reconstruction accuracy at the highest frequencies can be improved by an increase of the measurement distance up to three times the array inter-element spacing, but of course at the expense of slightly poorer low-frequency resolution.

Figure 10: Reconstructed sound power relative to true sound power in decibels.

Figure 10 shows the relative sound power spectrum of the CVX and WBH reconstructions: At each frequency, the reconstructed and true sound intensity maps (as shown in Figure 9) have been area-integrated, and the ratio between the estimated and the true sound power values have been plotted in decibels. There is a consistent small underestimation, but up to 5 kHz it remains within 2 dB. Above 5 kHz the underestimation increases rapidly, in particular for the CVX based algorithm. The calculation time for the 32 frequencies represented in Figure 10 was 238 sec for CVX and 9 sec for WBH.

16

Extension of acoustic holography to cover higher frequencies

5 Conclusions An iterative algorithm has been described for sparsity enforcing near-field acoustical holography over a wide frequency range based on the use of an optimized pseudorandom array geometry. The method, which is called Wideband Holography (WBH), can be seen as an example of Compressed Sensing. The algorithm has been tested by a series of simulated measurements on point sources and on a plate in a baffle. Very good results were in general obtained at frequencies up to four times the normal upper limiting frequency for use of the particular array with holography. The focus has been on the ability to locate and quantify the main sources (source areas) in terms of sound power within approximately a 10 dB dynamic range. The method was found to work surprisingly well with distributed sources, such as vibrating plates. Typical application areas could be engines and gearboxes, where measurements at close range are often not possible, and the method seems to work very well at the distances that are typically realistic in such applications. The iterative WBH algorithm was shown to provide sound field reconstructions almost identical to those of a conventional Compressed Sensing algorithm, where an optimization problem must be solved, involving a 1-norm of the solution vector. In the present work, such optimization problems have been solved using the CVX Matlab toolbox. For all the considered examples, the computation time of the CVX-based solution were approximately 30 times longer than those of WBH. In addition, the stopping criteria of the iterative WBH algorithm support the reconstruction of a large dynamic range without the risk of introducing numerical instability. Effectively, the optimal amount of regularization in applied. This is not possible in the CVX-based approach, where a fixed dynamic range must be specified. Engine and gearbox measurements are characterized by having sources at different distances. The sensitivity of the WBH algorithm to sources located outside the assumed source plane was therefore investigated in reference [6]. In general, the estimation of sound power was found to be not sensitive to sources situated outside the assumed source plane. To check the sound power estimation, a scanned measurement with a sound intensity probe was performed on a loudspeaker setup. It was argued in the present paper that it is advantageous to supplement the WBH algorithm with a Tikhonov regularized solution at the lowest frequencies. This was confirmed by simulated measurements in reference [6].

17

Extension of acoustic holography to cover higher frequencies

6 References [1] Sarkissian, A., Method of superposition applied to patch near-field acoustical holography. J Acoust Soc Am. 2005, 118(2), 671–678. [2] Hald, J., Basic theory and properties of statistically optimized near-field acoustical holography. J Acoust Soc Am. 2009, 125(4), 2105-2120. [3] Hald, J., Scaling of plane-wave functions in statistically optimized near-field acoustic holography. J Acoust Soc Am. 2014, 136(5), 2687-2696. [4] Hald, J., Array designs optimized for both low-frequency NAH and highfrequency beamforming. Proc InterNoise 2004. [5] Chardon, G., Daudet, L., Peillot, A., Ollivier, F., Bertin, N., Gribonval, R., Nearfield acoustic holography using sparse regularization and compressive sampling principles. J Acoust Soc Am. 2012, 132(3), 1521-1534. [6] Hald, J., Wideband acoustical holography. Proc InterNoise 2014. [7] International patent application no. PCT/EP2014/063597. [8] Suzuki, T., Generalized Inverse Beam-forming Algorithm Resolving Coherent/Incoherent, Distributed and Multipole Sources. Proc. AIAA Aeroacoustics Conference 2008. [9] Grant, M., Boyd, S., “CVX: Matlab software for disciplined convex programming, version 2.1,” http://cvxr.com/cvx, 2014 [10] Gomes, J., Hansen, P.C., A study on regularization parameter choice in Near-field Acoustical Holography. Proc Acoustics’08 (Euronoise 2008), 2875-2880. [11] Williams, E.G., Fourier Acoustics. Academic Press, 1999.

18

Consideration of the influences of the modal sound field with respect to the sound source localization results of the beamforming process in a vehicle interior Clemens Nau, Rob Opdam, Werner Moll, Michael Vorländer

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 W. Siebenpfeiffer (Hrsg.), Automotive Acoustics Conference 2015, Proceedings, https://doi.org/10.1007/978-3-658-27648-5_15

1

Consideration of the influences of the modal sound field with respect to the sound …

Introduction In the field of automotive engineering there is a lively international competition within each vehicle development sub-discipline. Depending on the vehicle segment, the requirements and customer profile, the development priorities differ for each competitor. In the premium segment, the part NVH, which covers all areas of the acoustic and vibration properties of an automobile has moved closer to the focus of the customers and therefore also the focus of the companies in recent years. Consequently, the effort of the developers is constantly increasing and they take advantage of new, improved analytical methods for practical application. It is the same with the beamforming, the localization and classification of sound sources. The beamforming offers the opportunity to locate sound sources both outside and inside the vehicle, which is why it is often termed source localization. The localization of sound sources within the vehicle turns out to be much more difficult than locating them outside. This is mainly due to the fact that the basic assumptions, under which the beamforming is applied are affected in a closed space. These are in particular the assumption of an acoustic free field and the assumption of a monopol characteristic for the sound sources. In addition, a vehicle interior has a highly complex sound field. On the one hand it is a "small room" (volume approximately 3 m3), which has a complex geometry. This leads to a complex modal structure. On the other hand, the surface is covered with materials with different absorption properties, which can cause very different diffraction and attenuation effects. In this acoustically demanding environment, the application of beamforming is still a challenge. The best known beamforming algorithm “delay and sum”, which is often referred to as Classical Beamforming (CBF), performs very poor under these conditions with respect to parameters such as dynamics and resolution. In the past, therefore, a sound source mapping in a vehicle compartment was limited in performance. However, over the past years the beamforming in modal sound fields, especially in the vehicle interior, has reached a new level. The scope in which the beamforming inside a room is performed must be first differentiated into two frequency ranges. These frequency ranges are separated by the so-called Schroeder frequency, which in room acoustics represents the transition from a modal sound field (below) to a statistical sound field (above). Studies have shown that modern beamforming algorithms as the MUSIC algorithm, the Functional beamforming or variants of Robust Adaptive Beamforming (RAB) methods provide a reliable detection inside a room from around the Schroeder frequency and above (statistical sound field), particularly in a vehicle interior [1]. The frequency range below the Schroeder frequency is consequently dominated by modal sound field effects (modes). Despite the use of modern beamforming algorithms, limitations of the localization precision below the Schroeder frequency are to be expected.

2

Consideration of the influences of the modal sound field with respect to the sound … In this regard, the combination of generalized cross correlation techniques with modern beamforming algorithms has significantly improved the localization precision [2]. The automated mode detection, which will take into account the modal sound field influences embedded in the beamforming process, is able to achieve a better performance below the Schroeder frequency.

Beamforming and Processing The conventional beamforming (CBF) is a valid method for the spatial localization of acoustic signals in free field. The output of the CBF can be expressed in the frequency domain, its performance can be formulated as: 𝒃(𝒈) = 𝒈′ 𝑪 𝒈

(1)

where 𝒈 is the array steering vector (𝒈′ conjugate-complex) and 𝑪 the cross spectral matrix (CSM). Due to the formulation of the CBF it is readily apparent that any components which are correlated to the sound source contained in the CSM are considered equally [3]. Adaptive beamforming algorithms show significant advantages over the CBF in terms of dynamics and resolution. Furthermore, studies show that especially some advanced beamforming algorithms are able to provide accurate localization results under the influence of a reactive sound field [1]. In order to enhance the performance of these algorithms and use them as a valid tool for the detection of acoustic signals in reactive sound fields, an additionally modification of the applied signal processing is advantageous. This is realized through the combination of advanced beamforming algorithms with generalized correlation techniques used as a preprocessing step for the cross spectral matrix (CSM). From the structure-borne sound acoustics and speech processing methods are known, this techniques may detect the presence of a radiating source and estimate the signal travel time difference at physically separated sensors, when energy of this source is received at the sensors. A wellknown type of the generalized cross correlation techniques (GCT), called Smoothed Coherence Transform (SCOT), is formulated as: 𝑪𝒏𝒌 𝑺𝑪𝑶𝑻 =

𝑪𝒏𝒌 √(𝑨𝒌 ∗ 𝑨𝒏 )

(2)

with the cross correlation C, row vector A with the diagonal entries of C (the autocorrelation) [4]. Originally developed for two sensors the transformation applied to the cross spectral matrix acts as a kind of correlation filter. By appropriate weighting of the matrix entries with regard to their correlation to the desired signal, this technique suppresses the correlated components and can therefore increase the robustness of the beamforming towards modal sound field influences significantly.

3

Consideration of the influences of the modal sound field with respect to the sound …

Beamforming in modal sound fields using GCT These signal processing techniques are an essential part of the improvement potential of the beamforming in modal sound fields. A general investigation of the potential of this technique, which is later to be expanded to the case of a vehicle interior, is therefore initially shown using the example of a scale model room.

GCT-Beamforming inside a scale model room The sound field of an enclosed space is characterized by its distinctive modal sound field. Below the Schroeder frequency (fs) modal influences dominate the sound field, thus ensuring a strong, dynamic sound pressure distribution [5]. The conventional fs

Fig. 1: Scale model room (left), reverberation time (right)

beamforming Delay and Sum without using GCT is particularly disturbed by these distinct modes below the Schroeder frequency, since the transfer functions of the room boundaries to the respective microphone positions assume complex shapes [5] Fig. 1 shows a scale modal room with a volume of 1.67 m3, in which the subsequent investigation is performed. Via the cylindrical loudspeaker shown in the front a sweep signal between 20 Hz and 20.000 Hz over a period of 23.78 s is radiated. This signal impinges at the channels of the microphone array showing a spectral sound pressure distribution which can be characterized by the reverberation time of this room. For the analyzed room, a Schroeder frequency of 1463 Hz can be calculated from the measured reverberation time of 0.892 s (see Fig 1). Under these conditions the performance with respect to the localization accuracy of advanced beamforming algorithms in combination with the GCT compared to the CBF is investigated. Fig. 3 shows the results of this comparison.

4

Consideration of the influences of the modal sound field with respect to the sound … The sound source (loudspeaker) is located at the marked position. Here the threedimensional mapping (relative sound pressure distribution) of the CBF and the mapping of the MUSIC algorithm in combination with the SCOT method between 20 Hz and 500 Hz and a dynamic range of 3 dB is exemplary compared. Algorithm CBF, frequency range: 20 – 500 Hz

Algorithm MUSIC + SCOT, frequency range: 20 – 500 Hz

Fig. 2: 3D-beamforming map (20 – 500 Hz, dynamics 3 dB), Localization result of the CBF (left), Localization result of the MUSIC + SCOT (right)

Advanced beamforming algorithms like the MUSIC algorithm combined with SCOT show far better results in terms of source localization precision compared to CBF when applied in a reactive modal sound field. In the case of the Standard beamforming (left), it is no longer possible to identify the source position inside the room, whereas in the right image, the source is clearly located in one place, namely the loudspeaker in the corner. This example illustrates the effectiveness of this technique in conjunction with the beamforming and further leads to the conclusion that this could also result in an improvement of the localization quality inside a vehicle interior.

GCT-Beamforming inside a vehicle interior Initially, the spatial conditions of the passenger compartment are to be compared with

Fig. 3: Comparison of the reverberation time of the studied vehicle (left) and the scale model room (right)

5

Consideration of the influences of the modal sound field with respect to the sound … those of the scale model room. This comparison is conducted based on the frequency dependent reverberation times of these rooms. Fig. 3 illustrates the reverberation times of the two rooms. Comparing the conditions under which the beamforming has been performed inside the scale model room with the conditions of the beamforming in a vehicle interior, it is clear that the beamforming of the examined vehicle interior performes comparable. The much shorter reverberation time of the considered passenger

Loudspeaker position

Loudspeaker position

Fig. 4: Comparison of the beamforming result using the functional beamforming inside a vehicle without GCT (left) and with GCT (right)

compartment of approximately 0.111 s (factor 8 less than in the model room) illustrates the relatively very rapid energy loss, which is caused by properties of the surface materials. This vehicle compartment has a volume of 3,210 m3 with a Schroeder frequency of about 372 Hz, which is about a factor of 4 lower than that of the scale model room. In addition, the number of modes of the vehicle interior can be estimated theoretically from [5] at 15 modes compared to the scale model room with theoretically up to 588 modes. None the less, in the relevant frequency range below the Schroeder frequency such a distinctive modal distribution is to be expected that this will adversely affect the beamforming result. The following comparison illustrates the influence of the GCT on the advanced beamforming algorithms, here carried out by the example of the functional beamforming (𝑣 = 300). From the loudspeaker which is integrated at the marked position in the vehicle, again a sweep signal from 20- 20.000 Hz and a duration of 23.78 s is radiated. The beamforming maps show an analyzed frequency range of 20- 500 Hz. It can be seen that with the GCT not only the location of the radiation are localized more precisely, but also the signal components correlated to the windscreen, the door and the A-pillar can be signifi-

6

Consideration of the influences of the modal sound field with respect to the sound … cantly reduced with this technique. The previously mentioned room acoustical conditions not only provide a lower Schroeder frequency but also, that in these results compared with the ones of 3 dB dynamic range achieved in the scale model room, the dynamics of the beamforming maps is achieves a dynamic range of 7 dB.

Mode Detection The procedure of the (automatic) mode detection and its consideration in the beamforming process is intended to improve the results of the source location in a vehicle interior as a physical approach. Unlike the signal theoretical approach of the GCT, the approach of mode detection for the improvement of the beamforming demands knowledge of the wave propagation in the investigated area. Recognizing this characteristic room property of the respective mode distribution this approach is ideally independent of the excitation signal and location of excitation. To excite the modal sound field of the vehicle interior it is sonicated again with a slow sweep signal. By detection of the sound field at different positions on the microphone array (Fig. 5) the sound pressure is sampled at a total of 128 positions for the duration of the signal and an additional, sufficient decay time.

Fig. 5: Positioning of the microphone arrays inside the vehicle interior

From these data indicators for the detection and determination of the frequency position of the modes must be defined. For this purpose generally two approaches are pursued. On one hand, the sound pressure level in the interior will be evaluated in terms of strong fluctuations, since each mode, depending on the order, describes a characteristic profile (level dips and spikes) inside the room. In the selected vehicle interior it can be assumed

7

Consideration of the influences of the modal sound field with respect to the sound … that no distinct comb filter effects arise. In addition, the sound field of the vehicle is excited only with one source.

Fig. 6: Histograms of the automatic mode detection using 48 channel array (front) and a 48 (front) and 80 channel array (back) simultaneously

The second, independent method is on the other hand to use a so-called pole-zero-fitting [6]. Based on the model after Kaneda the transfer functions of the data of the room are approximated by poles and zeros. The transfer functions of the room are approximated by poles and zeros. The corresponding poles of an acoustic transfer function and thus their resonance frequencies indicate the modal frequencies of a room. As an indicator for modal frequencies the accumulation of poles at certain frequencies is used. After a selected classification the pole frequencies are arranged in their respective frequency in-

71 Hz

72 Hz

73 Hz

74 Hz

Fig. 7: Evaluation of the results of the automatic mode detection through simulation (Direct FEM, hard boundaries)

8

Consideration of the influences of the modal sound field with respect to the sound … tervals and the frequencies with the highest pile densities are identified as mode frequencies. The identified modal frequencies are checked against the determined level of the frequency evaluation step. The diagrams (Fig. 6) show a determined mode distribution for the studied vehicle interior. On the one hand the mode distribution is evaluated using data only from the 48-channel microphone array (front), on the other hand, using data from the 48-channel arrays (front) and 80-channel arrays (back) together. In comparison of the two histograms it can be seen that despite the significantly higher number of sensors of the two arrays together in comparison to the single one the pole distribution remains stable. Only in the area towards the Schroeder frequency deviations are evident. Through additional simulation results, the determined mode frequencies can be evaluated. As simulation model a direct FEM model with 792.892 elements and hard boundaries with a comparable excitation location is used. An exemplary simulation result showing the first longitudinal mode of the vehicle at a frequency from 71 to 74 Hz (Fig. 7).

Beamforming Results The following study shows the influence of the automatic mode detection (AMD), according to the described procedure on the same vehicle and under identical conditions

Fig. 8: Frequency response filter

as in the previous investigations. Well aware that the localization result could be further

9

Consideration of the influences of the modal sound field with respect to the sound … improved by the application of advanced Beamforming algorithms and the application of the GCT, the performance of the automatic mode detection is demonstrated on the basis of CBF to show the capabilities of AMD. Based on the described mode frequency detection, a corresponding FIR filter is created. This filter is applied to the original time data. Such a filter is shown in Figure 8. Figure 9 shows the result of the comparison of the beamforming without using the AMD (left) and with AMD (right) based on the CBF. By applying the AMD a significant improvement in localization quality is recognizable. In contrast to the application of GCT,

Fig. 9: Comparison of the beamforming result using the CBF (20 – 500 Hz, dynamics 3 dB) inside a vehicle without AMD (left) and with AMD (right)

which improvement essentially results in the sharpening of the main lobe and an improvement of the dynamic range. When applying the AMD, as expected there is no sharpening effect to the main lobe. In the case of AMD the result of the beamforming benefits of the desired reduction of modal interfering influences, which can be seen among others by the strong reduction of the reflections on the windshield and the instrument panel.

Summary and future work The present study addresses two approaches for the improvement of the beamforming and especially the localization quality of acoustic sources in a vehicle interior. A signal theoretical approach, which is based on the processing of the cross spectral matrix using generalized cross correlations and a physical approach, which takes into account the modal characteristics of the corresponding room on an automated filter function in the beamforming process. Both are able to improve the beamforming result inside a room significantly. The signal theoretical approach is especially advantageous when dealing

10

Consideration of the influences of the modal sound field with respect to the sound … with highly correlated signals from multiple sources or, as shown here for the suppression of correlated shares of one source. By the integration of the mode detection in the beamforming process, it is possible to locate sources inside a room better, regardless of the type of beamforming. By the presented approaches it has been shown that it has become possible to repeal two major limitations of the beamforming, namely the failure of a reliable localization regarding correlated signal components and the inadequate localization quality in a range below the Schroeder frequency. The consideration of mode detection in the beamforming process offers further potential for improvement. For instance the realization of a proper localization quality while exciting the passenger compartment with several sources, as in this case comb filter-like effects are to be expected. A key development will also be the implementation of improved filters. Furthermore, factors such as the number of sensors and the sensor positioning, as well as the robustness of the method have currently not been adequately investigated.

References [1] Nau C., Vorländer M., “Comparison and evaluation of localization results of synthetic and real acoustic excitations using various beamforming algorithms in a vehicle interior”, Aachen Acoustics Colloquium, 2014 [2] Nau C., Vorländer M., “Analysis of the robustness of various advanced beamforming algorithms in comparison to the classical beamforming method when applied in reactive sound fields”, 41. Jahrestagung für Akustik, Nürnberg, 2015 [3] Benesty J., Chen J., “Microphone Array Signal Processing”, Springer Topics in Signal Processing, 2008 [4] Kuhn J. P., ”Detection Performance of the Smooth Coherence Transform (SCOT)“, Acoustics, Speech, and Signal Processing, IEEE International Conference on ICASSP, 1978 [5] Kuttruff H., ”room acoustics“ Fourth edition, Institute of Technical Acoustics, RWTH Aachen University, Spon Press London, 2000 [6] Y. Haneda, S. Makino, Y. Kaneda, „Common Acoustical Pole and Zero Modeling of Room Transfer Functions”, IEEE Transactions on Speech and Audio Processing, Vol. 2, No. 2, April 1994

11

Tagungsbericht Jonathan Walker

© Springer Fachmedien Wiesbaden GmbH, ein Teil von Springer Nature 2019 W. Siebenpfeiffer (Hrsg.), Automotive Acoustics Conference 2015, Proceedings, https://doi.org/10.1007/978-3-658-27648-5_16

1

Tagungsbericht

Automotive Acoustics Conference 2015 Auch wenn der Einzug der elektrischen Antriebstechnik noch allmählich vonstattengeht, ist er nicht mehr aufzuhalten. Dementsprechend waren NVHAspekte von Hybrid- und Elektroautos ein neues Schwerpunktthema unter den 22 Vorträgen auf der 3. Internationalen ATZlive Automotive Acoustics Conference. Unter der wissenschaftlichen Leitung des Konferenzpartners Autoneum, dem führenden Automobilzulieferer für Akustik- und Wärmemanagement bei Fahrzeugen, fand sie am 23. und 24. Juni 2015 an der ETH Zürich statt. HERKÖMMLICHES TRIFFT AUF NEUES Neben der Akustik von Elektroautos waren weitere Schwerpunkte Gewichtsreduzierung ohne Klangeinbuße und Komfortverlust sowie neue computerunterstützte Werkzeuge, die die Forschung und Entwicklung in der Akustik beschleunigen können. Die völlig unterschiedlichen Antriebselemente in Elektro- und Hybridfahrzeugen bringen zahlreiche neue Herausforderungen bezüglich NVH, wobei auch der Aspekt komprimierter Entwicklungszeiten zu bedenken ist. Das Akustikdesign traditionell angetriebener Fahrzeuge beruht auf jahrzehntelanger Forschung und Entwicklung, und der Käufer eines Elektroautos möchte keineswegs auf dieses hohe Niveau verzichten. Abgesehen von rein elektrischen Fahrzeugen ohne Range Extender arbeiten in Elektro- und Hybridautos aber weiterhin Verbrennungsmotoren, die aufgrund anderer Lastprofile kein herkömmliches Akustikverhalten mehr aufweisen. Bei der diesjährigen Konferenz durften ATZ-Herausgeber Wolfgang Siebenpfeiffer, Autoneum-CEO Martin Hirzel und Prof. Paolo Ermanni von der ETH Zürich 225 Teilnehmer willkommen heißen. Erfreulich seien, betonte Hirzel, der immer größere Anteil an internationalen Teilnehmern und die wachsende Akzeptanz der Tagung seitens der Fahrzeughersteller. Rund 20 % der Teilnehmer kamen aus den wichtigsten Zentren der Automobilentwicklung in Asien, und es waren insgesamt 18 Automobilhersteller durch 50 Teilnehmer und Redner vertreten. CHANCEN UND WAHRNEHMUNGEN In seinem Einführungsvortrag betonte Christoph Meier, Senior Manager NVH Powertrain bei Daimler, dass, obwohl die Entwicklung neuer Antriebs2

Tagungsbericht

konzepte hauptsächlich auf die Reduktion von Stickoxiden und Treibhausgasen zielt, sich daraus auch eine große Chance ergibt, die Lärmbelästigung durch den Straßenverkehr zu minimieren. Durch intensives Forschen, Entwickeln und Testen erreichen die neuen Elektroautos von Mercedes-Benz nicht nur für Fahrer und Passagiere exzellente Akustikwerte. Er wies weiter darauf hin, dass diese hart erarbeiteten allgemeinen Vorteile auch von Befürwortern von künstlichen Fahrgeräuschen zum Fußgängerschutz in Betracht gezogen werden sollten. Im Einklang mit weiteren Vorträgen erklärte Meier, dass Kraftfahrer in Zukunft ihre Wahrnehmung bezüglich des Geräuschs im Innenraum anpassen müssten. Die durch hochdrehende Elektromotoren verursachten Frequenzen werden oft als aufdringliches Jaulen wahrgenommen. Deren Dämpfung oder Modifikation ist zu einer Hauptbeschäftigung der NVH-Entwickler geworden. Die Weichen für Beiträge zu innovativen Mess-, Test-, Berechnungs- und Simulationsmethoden im Bereich der Fahrzeug-akustik stellte Prof. Jean-Louis Guyader, Direktor von SonoRHC Technologies. Er präsentierte als zweiter Keynote-Redner eine zeitinvertierende Lokalisationsmethode für NVH-Quellen. Basierend auf der Symmetrie von Vibrationsgleichungen können Lärmausstrahlungen aufgenommen und zeitinvertiert zurück in einen Körper geleitet werden. Hier fokussieren sie sich auf den Ursprungspunkt zurück. Er trug mehrere Beispiele für die Anwendung des Zeitinvertierens bei der Lösung von vibro-akustischen Problemen vor. NATÜRLICH GEGEN SYNTHETISCH André Fiebig von Head Acoustics referierte über Forschungen, die sich darauf konzentrieren, wie Fahrer von Elektroautos auf Geräusche reagieren. Er stellte Methoden vor, um herauszufinden, wie Elektroautos klingen sollten und in welchem Umfang synthetischer Sound erwünscht ist. Zu den bisher wichtigsten Erkenntnissen gehöre, dass durch selektive Geräuschbeimischung eine angenehmer wahrgenommene Geräuschkulisse erschaffen werden könne. Dies steht jedoch im Konflikt mit dem Wunsch nach einem leiseren Innenraum und einem angemessenen akustischen Feedback bezüglich der jeweiligen Fahrsituation.

3

Tagungsbericht

AKUSTIK IM KUNDENDIENST Großes Interesse fand ein Vortrag, der nicht aus dem Bereich der Forschung und Entwicklung von neuen Produkten stammte. Dr. León Gavric von PSA Peugeot Citroën Automobiles bot einen faszinierenden Einblick in den Unternehmensansatz, NVH-Probleme als Kundendienstleistung zu beheben. Hauptziel sei es, die oft hohen Kosten zu eliminieren, die durch erfolgloses Suchen in den Vertragswerkstätten entstünden. Der erste Schritt in der neuen Vorgehensweise bei PSA besteht darin, die Werkstätten darüber zu informieren, welche NVH-Phänomene während der Entwicklungsphase als normal eingestuft wurden. Danach wird mithilfe eines speziell entwickelten Aufnahme- und Diagnosegeräts eingegrenzt, ob es sich um Geräusche aus dem Antriebsstrang, der Karosserie oder der Innenausstattung handelt. Erste Ergebnisse zeigten schon sowohl eine beeindruckende Reduzierung der Garantie- und Kulanzkosten als auch eine verbesserte Kundenzufriedenheit. Außerdem, so Gavric, gelten die Rückmeldungen aus dem Prozess als äußerst wertvolles objektives Feedback für die PSAEntwickler. Zum Thema kraftstoffsparender Leichtbau referierte Francesca Ronzio von Autoneum über eine integrierte FE-Methodik, um gewichtsreduzierte Karosserie- und Verkleidungsteile akustisch besser zu gestalten. Im Bereich traditioneller Antriebstechnik präsentierte Jaemin Jin von Hyundai Motor Group eine Verbesserung der Innenraumakustik durch die Optimierung von Dieseleinspritzgeräuschen. Beim Abschließen der Konferenz bedankte sich der neue Tagungsleiter Dr. Davide Caprioli von Autoneum – er folgte auf Dr. Maurizio Mantovani – bei allen Rednern für das durchgehend hohe Niveau der Vorträge, das der Zuhörerschaft einen tiefen Einblick in den der zeitigen Stand der Technik der Fahrzeugakustik bot. Er wünschte sich für 2017 eine Fortsetzung des Erfolgs der Automotive Acoustics Conference 2015, besonders hinsichtlich der wachsenden Internationalität von Rednern und Hörern. Walker, J. ATZ Automobiltech Z (2015) 117: 86. https://doi.org/10.1007/s35148-015-0110-x

4