Assessment of a Statistical Energy Analysis model to perform automotive acoustic comfort subjective evaluation

Valentin MIQUEAU 1

Laboratoire Vibration Acoustique INSA-Lyon / Saint-Gobain Research Compiègne 25 bis Avenue Jean Capelle, 69100 Villeurbanne FRANCE / 1 rue de Montluçon, 60777 Thourotte FRANCE

Etienne PARIZET 2

LVA - INSA Lyon 25 bis Avenue Jean Capelle, 69100 Villeurbanne FRANCE

Sylvain GERMES 3

Saint-Gobain Research Compiègne 1 rue de Montluçon, 60777 Thourotte FRANCE

ABSTRACT Saint-Gobain develops Statistical Energy models of the interior noise of automotive vehicles in order to help the design of Saint-Gobain products. Our objective is to assess the possibility of using the predictions of such models to perform subjective evaluations of the acoustic comfort inside the cabin. Thus, as a tier-one supplier, Saint-Gobain could introduce psycho-acoustic evaluations earlier in its design process. A car equipped with several sets of glazing was exposed to a di ff use acoustic field to record an experimental set of sounds. Simultaneously, those car configurations were simulated using the model so a numerical set of sounds was auralized. Both the experimental and numerical sets were compared through two jury testing campaigns. The campaigns highlighted that the participants gave similar evaluations of sound regardless of its origin set. However, the ratings were not exactly the same between the measurement and the simulation especially for the car configurations using tempered glazing. The definition of the glazing inside the models and / or its interaction with the elements linked to it, like the window seals, might be the cause of those rating di ff erences. Therefore, some investigations were conducted to estimate the influence of the window seals on the transmission loss of glazing.

1. INTRODUCTION

Due to recent technological advances, the evaluation of acoustic comfort has become one of the fundamental steps in the design of a vehicle. The acoustic insulation foams or the use of PVB

1 valentin.miqueau@insa-lyon.fr & valentin.miqueau@saint-gobain.com

2 etienne.parizet@insa-lyon.fr

3 sylvain.germes@saint-gobain.com

a slaty. inter.noise 21-24 AUGUST SCOTTISH EVENT CAMPUS O ¥, ? GLASGOW

interlayers in windows are, for instance, helping to make cars more and more silent. The quality of the sound itself has become one of the criteria for a consumer to make his choice. It will eventually make him develop an a ff ect for the vehicle or entirely reject it [1]. As a result, car manufacturers must constantly improve the acoustic environment of their vehicles [2]. They have, during the last three decades, worked in order to integrate the psycho-acoustic aspect as early as possible in their design process. Many studies were conducted to reinforce psycho-acoustical knowledge and to adapt them to automobile issues [3–5]. Some of them attempted to develop new metrics to characterize the subjective evaluation of the sound [6] while some others tried to link all those indicators with each other [7]. However, all those studies present the main inconvenience of mostly relying on the use of prototypes. In this paper, the focus will be put on the possibility to use the output of a Statistical Energy Analysis (SEA) models [8] developed at Saint-Gobain to realize psycho-acoustical evaluations. This paper is a follow-up to the conference paper [9]. The first chapter will therefore briefly recall the important aspects of the SEA approach with respect to our problem. Then, we will expose the subjective comparison by a jury between the experimental and the predicted (SEA model) audio signals. Following the conclusions brought by this first test run, it was decided to study the impact of the window seals on the Transmission Loss (TL) of the glazing. In the last part of this paper, this campaign will be presented.

2. REMINDER ON THE SEA APPROACH

As was previously evoked in the article [9], the SEA method was chosen by Saint-Gobain as it is, by comparison to other approaches such as the finite element method or the boundary element method, especially suitable to address high frequencies aero-acoustic issues. Between all the sources to which cars are subjected, the airborne noise is one of the main component in the mid-high frequency domain. It is important to note that this is the frequency domain where the critical frequency of the glazing tends to happen (around 3.5 kHz for tempered glazing). This frequency occurs when the wavelength of the acoustic wave in the air matches the wavelength of the bending wave. It results in a significant local drop of the acoustic insulation around this frequency. In this chapter, the main characteristics of the considered model are reminded. The geometry is the one of a C-segment car. The following subsystems are defined and an energy balance is performed between them :

– The glazing shown in figure 1.

– Structural panels of the car shown in figure 1. Those panels, are defined by a Transmission Loss (TL) coe ffi cient.

– The acoustic cavities inside (see. figure 2) and outside the car cabin. The acoustic cavity we focus on in this paper is the one representing the space around the head of the driver. It is highlighted in red in figure 2.

Figure 1: Illustration of the car panels inside the model

Figure 2: Acoustic cavities as represented in the model inside the car cabin

Those energy exchanges are represented in the diagram 3. It is important to note that the glazing and the panels surrounding them are not exchanging energy in the present model. This hypothesis will be discussed later in this paper. From the outputs of the SEA model, it is possible to obtain the energy inside each acoustic cavity and then to calculate the acoustic pressure using equation 1.

< E > = < p 2 > V

ρ c 2 (1)

where < p 2 > is the averaged quadratic pressure ( Pa ) in the acoustic cavity, V is the volume of the cavity ( m 3 ), ρ is the air density ( kg . m − 3 ) and c is the speed of sound inside the air ( m . s − 1 ).

Figure 3: Diagram of the energy balance in the Saint-Gobain SEA model.

Pin,2 | carbody Pane! |_, 2 113227 Acoustic Cavity N23

3. JURY TESTING COMPARISON

The goal of this experiment was to study if the sounds generated from the SEA output receive a subjective evaluation similar to the one of the sounds measured on a vehicle. For this purpose, sounds for di ff erent car configurations were on one hand measured and one the other hand auralized from the outputs of the SEA model. All sounds were given an unpleasantness score by a jury (free rating). Afterwards, the ratings were treated using an analysis of variance.

3.1. Creation of stimuli In the previous phase of our study presented in the article [9], the measurement of acoustic pressures have been carried out at the level of the driver’s headrest on a vehicle exposed to a di ff used acoustic field generated from a "white noise" source. The pressures were acquired using a third octave bands analysis and a narrow bands analysis. The acoustic field inside the car cabin was manually modified using acoustic isolating masks directly applied on the glazing. The initial glazings were also changed from tempered glazing to laminated ones. All this allowed to obtain nineteen car configurations with di ff erent yet similar acoustic pressures and thus to generate the first set of sounds. Those car configuration are all listed in the previous paper [9]. In this paper, those pressures, as well as the auralized signals generated from them, will be refereed to as "measured". The same car configurations were reproduced in parallel by simulation via the SEA model to obtain a set of simulated pressures in third octave bands. It should also be noted that the TL which defined the energy exchanges through the glazing in the SEA model was, on the one hand, calculated using an in-house software and, on the other hand, determined using an experimental approach. Therefore, it was possible to study a potential impact of the definition of this parameter in the model. In this paper, those pressures, as well as the auralized signals generated from them, will be refereed to as "simulated". The steps of simulation and auralization are more deeply detailed inside the conference paper [9] and the whole process is summarized in figure 4.

These sets of sounds were then used to perform a jury testing campaign.

Figure 4: Production of the stimuli used for the subjective evaluation study

vets anay USTENING TEST

3.2. Free rating campaign 24 volunteers, aged between 20 and 30 years old, were asked to listen to the sounds with the help of headphones in the sound-proof booth of the Laboratory of Vibration and Acoustics pictured in figure 5. They were instructed to imagine themselves inside a car cruising on a highway and to evaluate the discomfort they felt listening to the audio signals. In order to note their perception, the participants had an interface with a cursor to move on a scale going from "not at all" unpleasant to "extremely" unpleasant (see. figure 6). Those subjective evaluations were converted into scores between 0 and 1000 in order to then apply statistical post-treatments. All the four sets of sounds evoked sooner were presented one after the other. The presentation order was modified between each candidate to avoid any influence from it. The pressure level was calibrated to correspond to a realistic level measured by Saint-Gobain during track tests.

ee rt

Figure 5: Sound-proof booth at the Laboratory of Vibration and Acoustics

Figure 6: Interface used for the free rating campaign

3.3. Results Firstly, the representation of the averages of the scores given by the participants to the sound of each "car configuration" (see. figure 7) for each "origin of set" highlights the similar evolution of judgment regardless of the origin of the sound.

Notat al Ame Is this noise unpleasant ? AL Extremely SOGs Was

Figure 7: Mean value and 95 % confidence interval of disagreement for car configuration for all origin sets

Secondly, the table of appreciation scores given by the participants was used to perform a two-factor ANalysis Of VAriance (ANOVA). The scores were the dependent variable while the "origin of the sets" and the "car configuration" were the factors studied. The results of this ANOVA are presented in table 1. In this table, a p-value smaller than the threshold value of 5 % is considered significant, meaning that the influence of the factor studied on the means of the dependent variable is itself significant.

Table 1: Summary of the results for a Two Factor ANOVA

Source Sum of Squares Df Mean Square F Ratio P Value

Set origin 288975 3 96325 0.995 0.4

Car configuration 25285405 18 1404745 57.99 < 2e-16 ***

Set origin*Car configuration 1557320 54 28839 2.039 9.32e-6 ***

Signif. codes : 0 ’***’ 0.001 ’**’ 0.01 ’*’ 0.05 ’.’ ” 1

Therefore, we can determine that the "car configuration" is a highly significant factor (F = 57.99 and p ≪ 1e-3). It confirms that the participants do perceive the acoustic modifications that were realized on the vehicle. The ANOVA also showed that the factor "origin of the sets" in itself is not a significant parameter (F = 0.995 and p = 0.4). However, its interaction with the "car configuration" factor is influential (F = 2.039 and p ≪ 1e-3). Therefore, the origin of the sound only seems to impact the rating of the annoyance felt for some particular configurations of the vehicle. By looking at figure 7, we can determine that the configurations impacted by this phenomenon are those for which the glazing is made out of tempered glass. For these configurations, we studied the pressure level at the driver’s head cavity. Spectral comparison as the one presented in figure 8 were made. They show a slight overestimation of the pressure level at the critical frequency of the glazing from the SEA model.

This means that participants hear a higher frequency component more loudly, which may be more disturbing to them. It is important to mentioned that all this does not happen for configurations using laminated glazing.

Figure 8: Sound Pressure Level comparison at the driver’s head cavity for the car configuration 17 (All glazing contributing)

Two hypotheses were put forward to explain the origin of this phenomenon:

– Lack of exchange of energy between the glazing and the body panels.

– Window seals procure an additional dissipation of energy.

The second hypothesis, which has already been the subject of a first series of tests, is developed more in detail by the section below.

4. INFLUENCE OF THE WINDOW SEALS ON THE TRANSMISSION LOSS OF THE GLAZING

As mentioned earlier in this document, the description of the glazing is achieved by the use of a transmission loss index which can either be numerically evaluated or by experiments on the vehicle. Figure 7 shows a spectral comparison between the sound pressure level of the model output using the analytic TL and the one of the model output using the measurement method. This brings to light the impact of this choice of TL to define the glazing. Moreover, we know that the TL at the critical frequency of a structure is mainly influenced by the structural damping of the material composing it (see. figure 9). In addition to that, according to C. O. Serna [10], window seals have been highlighted that glazing as a significant source of dissipation by an internal study of the PSA group. As a result, window seals could be at the origin of dissipation phenomena acting on the edges of the glass, thus influencing the TL characterizing the acoustic isolation of the glazing. As the assembly between the seal and the glazing become more dissipative than the glass alone, then the TL at the coincidence frequency would be more important than what is actually implemented in the SEA model. The

laminated glazing being really dissipative by nature, they would be less sensitive to this phenomenon than the tempered glazing. That would explain that only the car configurations with tempered glazing are significantly impacted by the window seals.

Figure 9: Figure from the book of F.J.Fahy [11]. "Damping indicator for a infinite plate exposed to a di ff used field with two structural damping η = 1% et η = 0 . 1%". The dotted line is the mass law of the plate.

4.1. Measurement To explore this research avenue, a measurement campaign of the TL was carried out in the acoustic laboratory of SGR Compiègne on a simplified prototype fixed between two reverberant rooms as shown on the figure 10. It was a flat glass made of either tempered or laminated glass. For each type of glazing (tempered, laminated), the TL was measured a first time with the usual boundary conditions of the installation, the glass being hold by compression between corner blocks shown figure 11 which are themselves fixed in the wall by screws. These corner blocks could be slid into the wall to ensure that the glass was as tightly fixed as possible (cf. figure 12). The assembly was sealed with a thin rigid foam of low thickness. In a second time, the boundary conditions of the flat glass were changed. Cutouts of window seals illustrated on figure were introduced between the glass and the corner blocks in order to reproduce, as well as possible, boundary conditions like the ones applied to the glazing of a car. To avoid any influence from the pressure imposed by the corner blocks, the distance between the saures and the flat glass was measured using millimeter scale bands directly fixed on them.

trequeree (He)

Figure 10: TL measurement. The glazing is fixed between two reverberant rooms

Figure 11: Corner blocks used for the installation

Yes 2 2 3

Figure 12: Mounted tempered glass in window seals cutouts

The TL measurement were made accordingly to the standard [12]. To do so, the pressure level was measured on each side of the wall as well as the reverberation time in the reception room. The TL (in

dB) could then be estimated using equation 2.

TL = L 1 − L 2 + 10 log ( S

A ) (2)

where L 1 is the average sound pressure level inside the emission room ( Pa ), L 2 is the average sound pressure level inside the reception room ( Pa ), S is the opening area inside which the studied element is installed ( m 2 ) and A is the equivalent sound absorption area inside the reception room ( m 3 ). A is extracted from the reverberation time measurement using equation 3.

A = 0 . 16 V

Tr 60 (3)

where Tr 60 is the reverberation time measured in the reception rooms ( s ) and V is the volume of the same room ( m 3 ).

Those tests have just been realized and the conclusions are still to be synthesized. Therefore, they do not figure in this paper but they will be exposed during the oral presentation.

5. CONCLUSION

A first study of comparison of the subjective perception between sounds generated from SEA model and sounds produced from measurements was carried out with a jury. The robustness of the auralization methodology employed has been verified. The model allows the auralization of sounds close to what is produced by the measurement for the majority of the studied vehicle configurations. Nevertheless, for the configurations where all the glazing are made of tempered glass, an overestimated acoustic pressure at the level of the driver’s headrest, mainly for frequencies close to the critical frequency, is detected on the output of the SEA model. This over pressure is perceived by the participants. Several hypotheses have been formulated as to its origin. One of them concern the impact of the window seals on the transmission loss of the glazing. It has already been tested at Saint-Gobain’s facilities and the results will be presented at the congress.

REFERENCES

[1] D. Västfjäll, M.-A. Gulbol, M. Kleiner, and T. Gärling. A ff ective evaluations of and reactions to exterior and interior vehicle auditory quality. Journal of Sound and Vibration , 255(3):501–518, 2002. [2] F. Penne, M. de Munck, and P. Sas. Shaping the sound of the next-generation BMW, International Conference, International conference on noise and vibration engineering. In ISMA , pages 25–39, Leuven:, 2004. [3] E. Zwicker and H. Fastl. Psychoacoustics: Facts and Models . Springer Series in Information Sciences. Springer Berlin Heidelberg, 2013. [4] Y. S. Wang, G. Q. Shen, H. Guo, X. L. Tang, and T. Hamade. Roughness modelling based on human auditory perception for sound quality evaluation of vehicle interior noise. Journal of Sound and Vibration , 332(16):3893–3904, 2013. [5] S. Lee. Objective evaluation of interior sound quality in passenger cars during acceleration. Journal of Sound and Vibration , 310:149–168, 2008. [6] M. J. Nor, M. H. Fouladi, H. Nahvi, and A. K. Ari ffi n. Index for vehicle acoustical comfort inside a passenger car. Applied Acoustics , 69(4):343–353, 2008.

[7] R. Bisping. Car interior sound quality : Experimental analysis by synthesis. Acustica , 83:813– 818, 1997. [8] Richard H Lyon, Richard G DeJong, and Manfred Heckl. Theory and application of statistical energy analysis, 1995. [9] Valentin Miqueau, Etienne Parizet, and Sylvain Germes. Psycho-acoustic evaluation of the automotive acoustic comfort using vibro-acoustic prediction methods. INTER-NOISE and NOISE-CON Congress and Conference Proceedings , 263(6):703–714, 2021. [10] Clara Oliver Serna. Vibroacoustic Analysis of Car Door and Window Seals. PhD thesis, 2016. [11] F. Fahy. Sound and structural vibration. Sound and Structural Vibration , January 2007. [12] AFNOR. Norme NF-EN-10140-2, 2016.