Pass-by noise engineering: a review of different transfer path analysis techniques

Transcription

1 ss-by noise engineering: a review of different transfer path analysis techniques K. Janssens 1,. Bianciardi 1, L. Britte 1, P. Van de Ponseele 1, H. Van der Auweraer 1 1 Simulation & Test Solutions, Siemens Industry Software NV Interleuvenlaan 68, B-3001, Leuven, Belgium karl.janssens@siemens.com Abstract The upcoming pass-by noise regulation will highly impact the vehicle development process. The revised standard requires more tests in different conditions and the execution of the tests is also more complex. Secondly, and most importantly, the new directive will force car manufacturers to further reduce the emitted noise levels of their vehicles. The auto OEM s are therefore looking for enhanced engineering techniques that allow quantifying noise contributions of vehicle subsystem components (engine, intake, gearbox, tires, exhaust, etc.) to enable them to validate designs for pass-by noise targets, early in the design process. This paper gives an overview of different source contribution analysis methods for pass-by noise. Three techniques are presented and compared: a traditional pressure inversion method, a power-based approach, and a signal processing technique using a transmissibility approach. The paper reviews the main principles of the techniques, their advantages and disadvantages, and the major selection criteria. Various application examples are presented, including a passenger car with internal combustion engine, a motorbike, and an electric vehicle. 1 Introduction The new pass-by noise regulation ECE-R51.03 will highly impact the vehicle development process, particularly in the vehicle homologation phase. More tests, more test iterations and a more detailed analysis of test results will be required to comply with the new regulation. irstly, the new ECE-R51.03 is based on a revision of the ISO 362 standard [1]. The revised standard requires more tests in different conditions. The execution of the tests themselves will also become more complex. or instance, more precise driver s guidance and instructions will become necessary. Secondly, and most importantly, the new European directive will force car manufacturers to further reduce the pass-by noise levels of their vehicles. Acceptable noise level limits will be lowered as displayed on figure 1. Most passenger cars will need to achieve a reduction of more than 6 dba of their pass-by noise levels. igure 1. Evolution of pass-by noise limits in dba within European Union 3923

2 3924 PROCEEDINGS O ISMA2014 INCLUDING USD2014 To avoid costly and time-consuming design iterations, enhanced engineering techniques are used to quantify and rank the contributions of separate subsystems, such as intake, powertrain, exhaust, tailpipe and tires to the overall pass-by noise level. Those techniques are not only efficient troubleshooting tools for noise reduction; they are also more and more employed for precise target setting, early in the design process. After a general presentation of the pass-by noise regulation for vehicle homologation, this papers presents an overview of different engineering techniques for source contribution analysis in the pass-by noise context. 2 Homologation testing 2.1 New ECE-R51.03 The new ECE-R51.03 is based on a revision of the ISO 362 standard (ISO 362-1:2007). The new standard mainly differs from previous versions in the following: Test approach speed: While the current standard imposes that the driver enters the track with a speed of 50 km/h, the revision imposes a speed of 50 km/h (± 1 km/h) at the zero line (position of the pass-by noise microphones). It is up to the testing team to estimate the test entry speed accordingly. Precise test guidance will be required to perform the test efficiently. Test operating condition: The current standard imposes a wide open throttle (full load) test only. This test does not reflect the reality of urban traffic. Statistically, urban traffic is characterized by slow accelerations and short drives at constant speed. The ISO 362 revision mainly reproduces partial load conditions by combining two pass-by noise tests instead of one: a wide open throttle (full load) test and coast-by (constant speed) test. In other words, the number of required tests doubles. Tested gears: The current standard requires testing a vehicle in 2 nd and 3 rd gear. According to the new standard, the tested gear will depend on the power-to-mass ratio of the vehicle. igure 2 shows research results from GM Europe, based on tests performed on 252 different vehicles, according to the new ISO 362 standard. It clearly shows that the tested gears are no longer by default 2 nd and 3 rd gear. Higher gears often need to be engaged for proper testing. As a general rule, vehicles with high power-to-mass ratio are tested in higher gears. igure 2. Tested gears for 252 different vehicles in function of power-to-mass ratio (analysis GM Europe)

3 TRANSER PATH ANALYSIS AND INVERSE METHODS 3925 When testing according to the new standard and regulations, the relative contribution of the different noise sources of the vehicle to the overall noise level changes. In the past, the powertrain noise used to be the dominant noise source, contributing most to the total pass-by noise level. Experiments have shown that, with the new standard, the relative contribution of the tire noise has increased significantly, as illustrated in figure 3. igure 3. ss-by noise contribution of powertrain and tires for old versus new standard (analysis GM Europe) There are a few explanations for this. irst of all, constant speed tests are now included. In these driving conditions, the engine noise is relatively low. Secondly, as most vehicles will be tested in higher gears, the average engine speed during the test is lower. Lower engine speed means lower engine noise and less contribution to the total pass-by noise. Besides a reviewed testing procedure, ECE-R51.03 also further limits acceptable pass-by noise levels. The limits within the EU will be lowered in 2 phases. Benchmark tests performed by automotive OEMs prove that these new limits will set challenges: only few of the currently available passenger cars (less than 8%) do not exceed the pass-by noise limit of 68 db(a) [2]. To pass homologation tests, it will be insufficient to focus only on reducing the powertrain noise. Innovative pass-by noise engineering techniques will be required. 3 ss-by noise testing environments ss-by noise tests can be performed outdoor on a test track as well as indoor on a chassis-dyno in a semianechoic test facility. Both testing environments present advantages and disadvantages. 3.1 Exterior pass-by noise testing In this test the vehicle is driving on a test track and two target microphones are placed on both sides of the track to measure the noise levels. The microphones are placed in the middle of the track at 7.5 m distance from the vehicle centre line. The height of the microphones is 1.2 m. Dedicated pass-by testing instrumentation is necessary, including light barriers to detect the entrance of the vehicle on the track, a speed radar, a telemetry system, a weather station, etc. Exterior pass-by noise tests are mainly used for vehicle homologation. However, the lack of repeatability between test runs, and the weather conditions represent an important limitation for detailed engineering studies.

4 3926 PROCEEDINGS O ISMA2014 INCLUDING USD In-room pass-by noise testing Instead of measuring pass-by noise levels outdoor, many automotive OEMs prefer executing part of the measurements indoor. In-room tests offer many advantages. irst of all, the test is not influenced by the weather conditions; it can be performed throughout the year. Ambient air temperature can easily be controlled indoor. Secondly, the vehicle is driven in repeatable, controlled conditions, making the measurement sequences much more efficient and effective. Thirdly, no wireless data transfer between vehicle and track is required, which greatly simplifies instrumentation. The principle is simple: the vehicle is positioned on a chassis-dyno in a semi-anechoic room. Linear arrays of microphones are placed at 7.5 m aside the vehicle covering the distance from -10 m to +10 m. The passby noise levels are estimated by interpolating the noise levels measured at different microphone positions along both sides of the vehicle. A schematic representation of a classical setup is shown in figure 4. The semi-anechoic room has to ensure free field conditions in the frequency range of interest; the lower cut-off frequency is defined by the room dimensions. inally, as the vehicle is not moving, a more extensive instrumentation of the vehicle is possible and relatively straightforward. This additional instrumentation will help engineers gaining more insights in the acoustic behavior and contribution of the various subsystem components. Indoor facilities also allow testing prototypes or vehicle mock-ups at an early stage of the design process. igure 4. Setups for exterior and in-room pass-by noise testing A sub-group of the ISO committee has been appointed to propose a new ISO standard for vehicle homologation based on indoor pass-by noise tests. The group s main challenge is to define the proper technique to characterize the tire noise contribution. Indeed, tires running on the rolls of a test bench produce different sounds than tires rolling on an exterior test track. In order to deliver the same results as when measuring outdoor, the work group is investigating different methods to correct this variation. A few different correction techniques have been proposed in the draft version of ISO

5 TRANSER PATH ANALYSIS AND INVERSE METHODS ss-by noise engineering Several engineering techniques can potentially be employed. They are presented and evaluated in the sections below. 4.1 Masking A classical technique consists in testing a vehicle, where potential noise sources are modified or insulated. or example, engineers will fit the vehicle with an additional exhaust damper in order to mask the tailpipe noise. An illustration is given in figure 5. igure 5. Application of an exhaust damper to mask the noise contribution of the tailpipe The analysis of the various masking configurations lets the engineers derive the noise contribution of the different subsystems. A typical work procedure is presented in figure 6. igure 6. Example of vehicle configurations that are to be measured to derive the subsystem contributions from masking tests In practice, the masking technique is a very time-consuming method, as it takes a lot of time to prepare and measure the many different configurations. Additionally, masking might cause slight modifications of operating temperatures, and impact the vehicle pass-by noise level. In consequence, this classical method is often replaced by other techniques.

6 3928 PROCEEDINGS O ISMA2014 INCLUDING USD Array techniques Beamforming arrays can be used to localize and measure the most dominant sources during a pass-by noise test. An example of a beamforming measurement is shown in figure 7. The array employed in this example consists of 54 microphones, which are irregularly spaced to guarantee an optimized dynamic range of 16 db up to 3 khz. The spatial resolution R of the array, i.e. the minimum distance between sources that can be separated, is determined by formula (1) below: With: d = distance between array and vehicle D = diameter of array λ = acoustic wavelength As there is a linear relation between the spatial resolution and the wavelength, source separation becomes difficult for the lower frequencies, especially when distances are large between the array and the vehicle, limiting the accuracy of the beamforming technology for exterior pass-by noise measurements. On larger vehicles like buses and trucks, the technique can be still sufficiently accurate, given the larger distances between the noise sources. The issue with spatial resolution is solved when performing the measurements in a chassis-dyno test room. In this case, the array can be placed as close as desired to the vehicle, making the technique very attractive in combination with in-room pass-by noise testing. When the array measurements are conducted close to the vehicle, additional near-field processing techniques can be applied, besides beamforming, that help localizing the low frequency noise sources with a very high spatial resolution. Examples are Near-ield Acoustic Holography, Inverse Boundary Element Method and ocalization [3,4]. igure 7. Example of acoustic array measurements on outdoor test track (MicrodB & LMS)

7 TRANSER PATH ANALYSIS AND INVERSE METHODS Airborne source quantification (ASQ) Airborne Source Quantification (ASQ) is a technique adopted to separate the airborne noise sources on a vehicle and evaluate their individual contribution to the overall pass-by noise level. The technique is based on a Transfer th Analysis (TPA) approach. The approach follows a four steps procedure as outlined in figure 8. A detailed method description is presented below for an indoor test setup with vehicle on the chassis-dyno and linear arrays of microphones on both sides of the room. A conceptual picture of the approach is shown in figure 9. Step 1: requency-domain ASQ model Step 2: Time-domain synthesis of acoustic loads and partial noise contributions Step 3: ss-by noise synthesis Step 4: Source contribution analysis igure 8. low diagram of ASQ method r k L 2 1 = y k NT ki u j = Q i H ji W = targets (measured and predicted) = indicators (measured) = loads (identified) d x = + x = 0 x = - igure 9. Conceptual picture of ASQ approach

8 3930 PROCEEDINGS O ISMA2014 INCLUDING USD2014 Step 1: requency domain ASQ model Two types of ASQ models can be applied: i) a traditional linear phase-based pressure inversion method and ii) a power-based formulation [5]. Each method will be illustrated for a specific application case. The linear ASQ method consists of discretizing the noise radiating components in coherent point sources, respecting a spacing of less than half the acoustic wavelength for the maximum frequency of interest [6]. The summed contribution of the individual noise sources to each target microphone k can be described by the following equation: with Q i the acoustic loads and NT ki the noise transfer functions between loads and target. The acoustic loads can be identified in two ways: i) by a direct calculation based on operating surface normal velocities, and ii) with an indirect identification procedure using a pressure inversion method. The surface normal velocity method is the most suitable for the lower frequency range, while the pressure inversion method is better suited for the mid and higher frequencies and for components where measuring a surface normal velocity is not possible or practical, e.g. for the tires and exhaust. In the application examples presented in the next sections, the acoustic loads are identified using an inverse estimation approach. A set of indicator microphones is uniformly distributed around the noise sources and a local transfer function matrix is measured between the source points and indicators. The operational loads are then identified from the indicator measurement data by applying the inverse of the local transfer function matrix. Mathematically, this can be expressed as follows: [ ] with uj the operational indicator measurements and Hji the local transfer function matrix between the source points and indicators. The number of indicators is taken higher than the number of source points to ensure stability of the matrix inversion. The local transfer function matrix between the source points and indicators and the transfer functions to the target microphones are measured with a calibrated volume velocity source (VVS). The measurements are usually conducted in a direct way by consecutive excitation tests (one excitation per source point). Note that the above equations 2 and 3 represent a complex, linear formulation of the problem, considering amplitude and phase in the load identification and noise contribution analysis. This complex, linear ASQ formulation is suited for closely spaced, correlated noise sources. An application case will be presented for a motorbike in section More recently, a power-based ASQ approach was developed to overcome the limitations of the phasebased approach at high frequencies. When extending the frequency range to several khz, the discretization into point sources becomes very dense due to the short wavelengths, and moreover the estimation of the operational loads by matrix inversion becomes very sensitive to phase errors. A power-based approach is more suited in that case. The power-based ASQ model assumes uncorrelated loads and omits the phase information in the formulation. With such an approach, larger surface areas or patches can be created that are represented by an average source strength. The target responses can then be formulated as follows:

9 TRANSER PATH ANALYSIS AND INVERSE METHODS 3931 with Q i 2 the acoustic loads expressed as autopower spectra. The noise transfer functions in this equation should be considered as an average transfer from several discrete point sources. Here again, the acoustic loads are identified using a pressure inversion method. This can be expressed as: [ ] The number of indicator responses should exceed the number of loads to have a well-determined system of equations when calculating the pseudo-inverse. An over determination by a factor 2 is typically taken as a rule-of-thumb. Still, one of the remaining drawbacks of the energetic pressure inversion in equation 5 is that it can return negative load power estimations, which is physically not realistic. There can be several reasons for this such as measurement noise on the data, missing sources in the analysis etc. However, this drawback can be overcome by solving the above system of equations with a constrained Least Squares method forcing the load estimates to be positive. The power-based ASQ approach is suited for uncorrelated, broadband type of excitations, allowing an extension of the frequency range up to frequencies well above 1 khz. An application of the power-based method will be shown for a passenger car with IC engine in section Step 2: Time domain synthesis of acoustic loads and partial noise contributions In the second step of the method, IR filter sets are constructed from the frequency-domain ASQ model. These filters allow a time-domain synthesis of the acoustic loads and their partial contributions to the targets starting from the measured indicator time signals [7]. When using a power-based model, scaling factors are applied to transform the pressure indicator measurements to acoustic loads. ast T Convolution is used to do the required time-domain filtering. This T-based implementation makes use of the well-known Overlap-Add method and avoids time-consuming convolutions in the timedomain. A proper filter pre-processing (treatment of delays, etc.) is important, making sure that the filters are causal and do not generate artefacts. The advantage of the time-domain synthesis is that streams of time data (loads, partial contributions, summed contributions) become available, enabling a dedicated pass-by noise processing (time and frequency domain) and sound quality analysis (listening and objective metrics). Step 3: ss-by noise synthesis In the third step of the approach, pass-by sounds are generated for the zero position (x = 0) on the left and right target microphone arrays by mixing the r target microphone responses synthesized in the previous step. The mixing is performed for each source contribution. irst, the time-delay is removed from every synthesized target response signal. or example, when considering the source contribution of the exhaust, the applied time-delays are the travel times of the exhaust noise to reach the different target microphones. These time-delays can be derived from the phase slope in the NT s. By applying the time shifts, the target responses become well aligned in phase, avoiding interference problems when mixing the signals. The shifted signals are then recombined into a pass-by sound by taking into account the speed profile of the vehicle, which is measured from the chassis dyno. The pass-by sound is generated by running through the different targets, from microphone 1 in the front of the vehicle towards microphone r in the back, and meanwhile mixing the sounds of the closest two microphones. In a final stage, the Doppler shift is included by resampling the audio data. This is achieved by applying a non-linear transformation of time, taking into account the time-varying distance or delay between source and pass-by noise target location. Step 4: Source contribution analysis The output of the previous three steps allows several interesting analyses like:

10 3932 PROCEEDINGS O ISMA2014 INCLUDING USD i) analysis and ranking of the different source contributions in a pass-by noise test - ii) comparison of the summed noise contributions with measurements - iii) detailed time-frequency analyses - iv) listening and sound quality assessment The ASQ technique can be applied in in-room test environments as well as outdoor. The two application cases presented in the following sections show two indoor pass-by noise measurements. The first one shows the application of the linear ASQ technique on a motorbike, while the second one presents the application of the energetic technique on a passenger car with internal combustion engine Linear ASQ application case: Piaggio Beverly 350 IE The linear ASQ technique was tested and validated on an indoor motorbike pass-by noise test [8]. The test object was a Piaggio Beverly 350 IE motorbike, shown in figure 10. The noise signature of the motorbike is characterized by many lower frequency harmonics below 1 khz. Intake Exhaust Transmission ` igure 10. Piaggio motorbike pass-by noise engineering application Six main noise sources were considered: engine, transmission, intake, gearbox, exhaust and rear tire. The noise radiating shells were divided in 41 patches with a dimension of 8x8 cm; 59 indicator microphones were mounted close to the noise source components. Sufficient over-determination was required to obtain a well-conditioned matrix ensuring a stable pressure inversion; 19 equally spaced target microphones were mounted on a linear array on the right side of the test room. The transfer functions were measured in a direct way with a calibrated volume velocity source with frequency range from 200 Hz to 8 khz. A total amount of 3198 transfer functions was acquired in half a day. A linear ASQ approach was used to deal with the coherent, low frequency harmonic source excitations of the motorbike; phase relations play an important role and should be respected in the analysis. The acoustic loads were identified by inversion of the full transfer function matrix in equation 3, including all the crossterms between the engine, transmission, intake, gearbox, exhaust and rear tire. Operational measurements were performed using smooth roll surfaces. Various operational conditions were tested including constant speed and wide open throttle tests. The target microphones and near-field indicators were measured synchronously at 25.6 khz sampling frequency. In order to obtain a similar system behaviour as in operational conditions, the transfer functions were measured by placing a weight on the motorbike seat, simulating the presence of a rider. Additionally, the motorbike was firmly fixed at the front wheel to the ground in order to avoid excessive movements during

11 db(a) TRANSER PATH ANALYSIS AND INVERSE METHODS 3933 the operational tests. Depending on the frequency, excessive variations of the motorbike position may introduce considerable deviations from the identified system characteristics. igure 11 presents the pass-by noise source contribution results in function of the motorbike position for the right side of the motorbike. The differences in the source ranking are clearly observable. As expected, the back tire reveals to be the less dominant source, while the transmission has the largest contribution to the right side of the pass-by noise for the complete test. Even if the transmission is positioned on the left side of the motorbike, as shown in figure 10, it has a large contribution to the pass-by noise because the emitted noise radiates through the wheel towards the right without obstacles. While the noise contribution of the intake, mounted as well on the left side of the motorbike and well-shielded toward the right side, shows a small level decrement in the middle of the pass-by noise test. inally the exhaust contribution shows a significant increase when the motorcycle is passing by. The total contribution is clearly very similar to the measured right side overall pass-by level. Small deviations in the total noise synthesis can be attributed to slight movements of the motorbike in the acceleration phase and to other types of errors (measurement errors, model assumptions, etc) inherent to an ASQ analysis Overall level Right Side Virtual:S (A) Overall level Right Side:S/ (A)<Total contribution Overall level Right Side:S/ENGINE (A) Ov erall lev el Right Side:S/EXHAUST (A) Overall level Right Side:S/GEARBOX (A) Overall level Right Side:S/INTAKE (A) Overall level Right Side:S/TIRE (A) Overall level Right Side:S/TRANSMISSION (A) m 8.00 igure 11. OA-level of source contributions in function of the motorbike position for ISO 362 run-up Energetic ASQ application case: 4-cylinder ICE passenger car The power-based ASQ method was tested on a front wheel driven 4-cylinder vehicle with internal combustion engine. A picture of the measurement setup is shown in figure 12. The same test room was used as for the Piaggio motorbike. The target microphone array was also mounted on the right side of the test vehicle.

12 3934 PROCEEDINGS O ISMA2014 INCLUDING USD2014 igure 12. In-room test setup for the 4-cylinder ICE vehicle A reduced ASQ model was developed, considering seven noise sources for the engine (6 engine sides and 1 intake source), two sources for each front tire (1 for the leading edge and 1 for the trailing edge) and one source for the exhaust pipe. Pressure indicators (22 for the engine, 6 for each front tire and 2 for the exhaust pipe) were distributed closeby and uniformly along the different sources, respecting the rule-ofthumb overdetermination by a factor 2. The local R s and noise transfer functions to the 19 target microphones were measured in a direct way with the calibrated mid-frequency VVS (200 Hz - 8 khz). Operational tests were done for various operating conditions including run-up, run-down and constant speed experiments in different throttles and gears, tests on smooth and tarmac road profiles, measurements with engine switched off and tests according to the ISO 362 standard (2007). The engine rotational speed, vehicle speed and pressure response at all target and indicator microphones were measured synchronously at sampling rate of 51.2 khz. The power-based ASQ approach was used for the analysis. or identifying the operational loads, the local transfer function matrix Hji was utilized in a block diagonal form, excluding the cross-terms between the engine, exhaust and tire components. The constrained Least Squares estimation method was applied, forcing the load estimates to be positive. Time-domain IR filters were constructed, generating the source contribution time streams for the 19 target locations and the mixing algorithm was applied to recombine these time signals into pass-by sounds. Some frequency-domain results are shown in figure 13 for three different target microphones. irst of all, a good agreement between synthesized and measured target pressure can be observed over the complete frequency range, which shows that all relevant noise sources were included in the analysis. The figures furthermore provide in-depth information about the importance of each source group in function of the frequency and the relative position of the vehicle. The influence of the relative position on the source ranking is clearly visible. The exhaust contribution obviously becomes more important in microphone 15 towards the back of the car, while the contribution of the engine is more significant at microphone 3 in the front. The right tire clearly dominates mostly at microphone 9 on the side of the car. It clearly exceeds the contribution of the left tire, which is at a further distance and partly masked by the car body and right wheel.

13 db(a) db [ Hz] db(a) db [ Hz] db(a) db [ Hz] TRANSER PATH ANALYSIS AND INVERSE METHODS (a) target microphone 3 (x = m) A L Octave 1/3 Measured Total Right_Tyre Engine Left_Tyre Exhaust (b) target microphone 9 (x = 0.59 m) A L Octave 1/3 Measured Total Right_Tyre Engine Left_Tyre Exhaust (c) target microphone 15 (x = 5.77 m) A L Octave 1/3 Measured Total Right_Tyre Engine Left_Tyre Exhaust igure 13. Source contribution results for different target microphones in ISO 362 run-up test on smooth road profile Another useful analysis consists of tracking the time evolution of the different source contributions during the pass-by test as shown in figure 14. The overall levels are plotted as a function of the car position for a smooth road surface. The variation in the source ranking is clearly observable. The radiation of the right tire changes rapidly and becomes most dominant when the car passes by. The exhaust contribution significantly increases after the passage, while the other source contributions decrease.

14 db(a) 3936 PROCEEDINGS O ISMA2014 INCLUDING USD Measured Total Tyre Right Engine Tyre Left Exhaust m 7.00 igure 14. OA-level of source contributions in function of car position for ISO 362 run-up 4.4 Transmissibility approach The ASQ technique has proven to accurately separate different airborne noise sources and evaluate their contribution to the overall pass-by noise level. The application cases on the motorbike and ICE vehicle showed the reliability of the method. However, the ASQ technique reveals its limitation when dealing with high frequency harmonic noise sources as for example in electric vehicles. In that case, ASQ requires a very fine discretization of the electric motor in a large number of coherent point sources to ensure an accurate prediction of the loads. Huge instrumentation and measurement efforts (large number of microphones, many R excitation tests, etc) are then required, making the ASQ approach impractical. A new method is therefore proposed for this specific electric vehicle application case. The method is based on a transmissibility approach and appears to perform very well as a source separation technique due to the incoherent nature of the electric motor and tire noise sources. The method is based on a MIMO (Multiple Input Multiple Output) transmissibility estimation between multiple references indicators placed nearby the engine and tire noise sources (system inputs) and the target microphones along both sides of the vehicle (system outputs). The transmissibilities are estimated from operational input and output measurement data. R measurements with calibrated speaker source are not required, which is interesting from a practical point of view. or every target microphone k, the transmissibility model can be formulated as follows: with r i the operational references and T ki the transmissibilities between the references and the k th target microphone. Accelerometers (electric motor) as well as microphones (electric motor, tires) can be used as input references. The transmissibility functions between all inputs and outputs are estimated in the following way: [ ] [ ] [ ] with S ii the input autopower matrix (reference auto- and crosspowers) and S ki the crosspower matrix (crosspowers between all references and targets). A pseudo-inversion of the matrix is applied based on Singular Value Decomposition (SVD).

15 rpm Rpm Extr (T1) db(a) rpm Rpm Extr (T1) db(a) TRANSER PATH ANALYSIS AND INVERSE METHODS 3937 The transmissibility approach is conceptually different from ASQ. When using such approach, any possible source of error must be minimized to ensure reliable results [9]: i) the engine and tire references must be uncorrelated to correctly estimate the transmissibility functions, and ii) the crosscoupling between these references must be as small as possible. This is the case for the electric vehicle application example. The electric motor and tire noise sources are by definition incoherent, the former is characterized by sharp, high frequency tonal components while the latter is characterized by broadband type of noise. The cross-coupling between engine and tire references is minimized when accelerometers are used as reference for the electric motor (mainly picking up engine noise, almost no tire noise leaking in) and the motor harmonics are filtered from the tire microphone measurements in a pre-processing phase (for example by using a cross-talk cancellation technique [10]). Similarly as in the ASQ method, a time domain analysis is performed once a good frequency-domain models is obtained. IR filters are derived from the estimated transmissibility functions and applied to the reference measurement data to compute the partial noise contributions of the electric motor and each of the tires. Likewise, ast T convolution is applied to perform the required time-domain filtering Transmissibility application case: fully electric vehicle (EV) The application case was performed on an electrified vehicle with two switched reluctance motors (SRM). SRM s are the noisiest among the electric motors. A pass-by noise engineering study was carried out to evaluate the importance of the electric motor in relation to the road noise. The transmissibility approach was used. The electric vehicle was placed on the chassis-dyno with the front wheels on the rotating drum and the back wheels clamped to the floor to prevent the vehicle from moving forward and backward. Operational measurements were performed on rough roll surfaces. A linear array of 19 target microphones was mounted along the right side of the vehicle; 1 reference accelerometer was placed on the electric motor and 6 reference microphones were mounted on each of the front wheels close to the tire-road contact point. The pass-by noise microphones and reference indicators were measured synchronously at 25.6 khz sampling frequency. Different operational conditions were measured including constant speed and wide open throttle tests. The results are presented for the ISO 362 (2007) run-up case. igure 15 shows a time-frequency analysis performed for one pass-by noise microphone (Mic11). The figure compares the measured pressure response with the summed noise contribution of the electric motor and tires. The time-frequency colormaps are clearly very similar. igure 16 shows the decomposition of the time-frequency colormap into contributions of the engine and road noise, and a further split of the road noise into contributions from the left and right tires. The engine (harmonics) and tire (broadband) noise sources are well separated due to their incoherent nature. a) Measured b) Summed contribution Hz MIC11 (CH11) Hz MIC11 (CH11) igure 15. Summed contribution and measured pressure response for microphone 11 in WOT run-up test (ISO 362, 2007)

16 rpm Rpm Extr (T1) db(a) Rpm Extr (T1) rpm db(a) rpm Rpm Extr (T1) db(a) Rpm Extr (T1) rpm db(a) rpm Rpm Extr (T1) db(a) 3938 PROCEEDINGS O ISMA2014 INCLUDING USD Hz MIC11 (CH11) Mic Hz Electric MIC11 (CH11) motor Hz TIRE_CONTRIBUTION Tires (CH2) Hz MIC11 (CH1) Right tire Hz MIC11 (CH1) Left tire igure 16. Decomposition of pressure response in engine and tire contributions Also the time evolution of the electric motor orders was investigated as a further validation of the transmissibility approach. igure 17 shows a comparison of the dominant order 30 component computed from the measured pressure signal (red curve) and obtained with the transmissibility approach (green curve). The amplitude and phase profiles show a very good match, especially in RPM regions where the order largely exceeds the broadband noise. The differences become significant at low rotation speeds when the order is masked by the tire noise.

17 db(a) Phase db TRANSER PATH ANALYSIS AND INVERSE METHODS 3939 Order MIC11 order_section_measured Order MIC11 order_section_eng_contr rpm Rpm Extr (T1) igure 17. Order 30 evolution for target microphone 11: (red) from measurement; (green) obtained with transmissibility approach igure 18 shows the pass-by noise contribution results in function of the vehicle position for the right side of the car. The total contribution is clearly very similar to the measured overall pass-by noise level. The differences in source ranking are clearly visible. As expected, the front left tire contributes much less compared to the front right tire which is at a shorter distance from the evaluation point. The electric motor contribution ramps up after the zero line (between x = 1 and 4 m) when the order 30 component becomes dominant ( RPM). Overall level Right Side (A) Ov erall level Right Side/Total contribution (A)<Total contribution Overall level Right Side/ENGINE (A) Overall level Right Side/LTIRE (A) Overall level Right Side/RTIRE (A) m igure 18. ss-by noise OA-levels in ISO 362 (2007) test: engine and tire contributions, total sum and measurement. 5 Conclusions Complying with the new pass-by noise regulations will be a challenging task for vehicle manufacturers. The challenge comes from a combination of a new pass-by noise testing standard and lower pass-by noise limits. In order to successfully pass upcoming homologation tests, the automotive OEM will need to reinvent their development process and use new pass-by noise engineering techniques. In this paper, different pass-by noise engineering techniques have been discussed: they help engineers better quantify the noise contribution of the different vehicle subsystems to the overall pass-by noise levels and allow them to set accurate targets for the subsystem components. Three innovative techniques were presented and compared: a traditional pressure inversion method (linear ASQ), a power-based approach (energetic ASQ) and a signal processing technique using a transmissibility

18 3940 PROCEEDINGS O ISMA2014 INCLUDING USD2014 approach. The main principles of the techniques, their advantages and disadvantages, and the major selection criteria were discussed. The techniques were validated with several application cases including a motorbike, a passenger car with internal combustion engine, and an electric vehicle. These techniques have proven to offer several advantages and will likely become instrumental in future pass-by noise engineering processes. Acknowledgements Karl Janssens and abio Bianciardi, authors of this paper, are researchers involved in the P7 Marie Curie IAPP project TYRE-DYN (P7-PEOPLE-IAPP2009, Grant Agreement No ). The financial support of the European Commission is gratefully acknowledged. The research work was also partly carried out in the frame of the IWT project: A new generation of NVH methods for hybrid and electric vehicles (HEV-NVH). The financial support of the lemish Institute for Promotion of Innovation (IWT) is gratefully acknowledged. References [1] ISO 362:2007. Measurement of noise emitted by accelerating road vehicles [2] M.E. Braun, S.J. Walsh, J.L. Horner, R. Chuter, Noise source characteristics in the ISO 362 vehicle pass-by noise test: Literature review. Applied Acoustics 74, p [3] L. Lamotte, B. Beguet, Qualifying the noise sources in term of localization and quantification during flight tests [4] J. Lanslots,. Deblauwe, K. Janssens, Selecting the most appropriate sound source localization techniques for modern-day industrial applications. Sound & Vibration Magazine, April, 2010 [5] P. Van de Ponseele, H. Van der Auweraer, K. Janssens, Source-Transfer-Receiver approaches: a review of methods. Proceedings ISMA Conference, Leuven, Belgium, September 2012 [6] J.W Verheij, Inverse and reciprocity methods for machinery noise source characterization and sound path quantification, part 1: sources. International Journal of Acoustics and Vibration 2 [1], 1997, p [7] K. Janssens, P. Aarnoutse, P. Gajdatsy, L. Britte,. Deblauwe, H. Van der Auweraer, Timedomain source contribution analysis method for in-room pass-by noise. SAE N&V Conference, US, 2011 [8] Motor vehicle exhaust noise - new regulation 2014, [9] P. Gajdatsy, K. Janssens, W. Desmet, H. Van der Auweraer, Application of the transmissibility concept in Transfer th Analysis. Mechanical Systems and Signal Processing (MSSP), Vol. 24, p , [10] K. Janssens,. Bianciardi, Time-domain ASQ method for pass-by noise engineering. Proceedings NOISE-CON 2013, Denver, Colorado, USA, August 2013.