Sound source distribution of high-speed trains and reduction of aerodynamic bogie noise Toki Uda 1 Railway Technical Research Institute 2-8-38 Hikari-cho, Kokubunji-shi, Tokyo, 185-8540 Japan Mariko Akutsu 2 Railway Technical Research Institute 2-8-38 Hikari-cho, Kokubunji-shi, Tokyo, 185-8540 Japan Tsugutoshi Kawaguchi 3 Railway Technical Research Institute 2-8-38 Hikari-cho, Kokubunji-shi, Tokyo, 185-8540 Japan Yukie Ogata 4 Railway Technical Research Institute 2-8-38 Hikari-cho, Kokubunji-shi, Tokyo, 185-8540 Japan ABSTRACT The distribution of noise sources around a Japanese Shinkansen train was measured using a micro- phone array. By applying a deconvolution method to improve the spatial resolution of the microphone array signal processing, it was found that the main aerodynamic noise sources of the entire train distribute in the lower part of the leading vehicle, the pantographs, the bogies, and the inter-car gaps. Wind tunnel tests were conducted to verify measures for reducing aerodynamic noise generated from the bogie section. It was found that the combination of noise insulation measures by flat undercover and cavity noise reduction measures by rounding the corners of the cavity can reduce the bogie- section aerodynamic noise by approximately 3 dB. 1. INTRODUCTION

Wayside noise is one of the most important issues in a railway environment. The Japanese high-speed trains known as Shinkansen are subject to the environmental noise standard, 70 dB in residential areas and 75 dB in commercial and industrial areas (both at the maximum A-weighted level with time- weighting S). The evaluation point is usually 25 m away from the nearest track. Railroad operators apply various noise reduction techniques, such as noise barrier, to vehicles and ground equipment to

1 uda.toki.92@rtri.or.jp 2 akutsu.mariko.82@rtri.or.jp 3 kawaguchi.tsugutoshi.00@rtri.or.jp 4 ogata.yukie.96@rtri.or.jp

‘linter.noise 21-24 AUGUST SCOTTISH EVENT CAMPUS ‘GLASGOW

comply with this environmental noise standard. In general, vehicle improvement is less expensive, so the priority is to consider techniques applied to the vehicle side. In considering noise reduction measures for the train vehicle, it is necessary to understand the contribution of each noise source to the total noise level and to reduce larger noise sources. Here, a noise prediction model is proposed for a Japanese Shinkansen that places four sources around the vehicle and structure, as shown in Figure 1 (left). With this model, the contribution of each sound source is estimated at 320 km/h [1], as shown in Figure 1 (right). The blue color in the figure indicates aerodynamic noise, and the green color indicates rolling noise. The contribution of aerodynamic noise reaches 60% of the total noise at 320 km/h and approximately 70% at 360 km/h. This is because the power of aerodynamic noise generated from bogies and pantographs is proportional to the 6th power of the train speed, while the power of rolling and bridge noise is proportional to the 2nd to 3rd power of the train speed; the impact of aerodynamic noise on wayside noise increases at high-speed. There- fore, reducing the aerodynamic noise of vehicle is necessary, which is investigated by verifying low- noise measures using wind tunnel experiments.

(b) Figure 1: (a) Noise source model of a Shinkansen. (b) Predicted contribution ratio of noise sources for a train running at 320 km/h on the wayside total noise at 25-m away from the nearest track with noise barrier of R.L.+2 m. 2. NOISE SOURCE IDENTIFICATION BY FIELD TEST

vara noise = oN -_ Upper-part — aerodynamic noise peter te Lower-part aerodynamic noise \ Rolling noise Bridge noise

(a)

To reduce the noise level of vehicle, it is necessary to determine the noise source distribution, which indicates the levels and locations of sound sources. The Railway Technical Research Institute (RTRI) has designed and fabricated a two-dimensional microphone array that is portable and easy to install in a variety of field conditions. This section describes the details of this array.

2.1. Microphone Array Design Concept A two-dimensional microphone array was designed and fabricated. As shown in Figure 2, the array has 72 microphones (GRASS 40PH) arranged on a plane to enable two-dimensional noise source identification and is circular with an outer diameter of 3 m and an inner diameter of 0.6 m. Eight microphones are arranged at equal intervals from the center in the radial direction along nine spiral arms from the inner circle (multi-arm spiral array [2]). The array consists of six sections, making it possible to install the array in a limited space and at various field locations. The microphone cables are bundled with LEMO cables and input into a data recorder (TEAC WX-7096). The passing train speed was determined using a ground vibration meter by autocorrelation analysis of the vibration acceleration level. The error of speed calculation by this method was approximately within 0.3%. Optical sensors detected the position of the train.

2.2. Microphone Array Directivity Characteristics Fixing the outer diameter and the number of microphones, array simulations were performed by changing parameters such as the inner diameter, the number of spiral arms, and the spiral angle. The

Bridge noise Pantographs Upper-part aerodynamic Rolling noise noise Lower-part aerodynamic noise

array performance was evaluated with the maximum sidelobe level for the directivity characteristics and spatial resolution in the 250 Hz to 6.3 kHz band as shown in Figure 3. The blue lines show the directivity characteristics along the y = 0 line, and the orange lines show the maximum level along with concentric circles from the array center. The spatial resolution width (3 dB per side) at 500 Hz is approximately 2 m and at 1kHz it is approximately 1 m, indicating that the system can separate sound sources near the front and rear axles of a bogie (2500 mm).

Center of the rail 3400 5000 Unit mm Microphone array 3000 {Ballast track

(a) (b) Figure 2: (a) Schematic of the cross section at the measurement point. (b) Microphone arrangement of the multi-arm spiral array developed by RTRI.

y (m) -2 #1 #2 #9 #8 #3 #6 -2

Figure 3: Simulation results for the microphone array spatial resolution in the frequency range of 500 Hz to 4 kHz.

500.0 Hz 1000.0 Hz 2000.0 Hz 4000.0 Hz = 90 90 +34 90 90 3 3dB @ 80 80 80 80 & “g 70 70 70 70 5 60 60 60 60 y(m)

3. IMPROVEMENT OF RESOLUTION

3.1. Signal Processing Technique The most common signal processing for noise source identification is called beamforming, which is based on the delay and sum (DAS) method shown in Equation 1.

ே ೘೔೎ ∑ 𝑟 ௠௝ 𝑝 ௠ ቀ𝑡൅ ௥ ೘ೕ ሺ௧ሻ

𝑏 ௝ ሺ𝑡ሻൌ ଵ

௖ బ ቁ ே ೘೔೎ ௠ୀଵ , (1)

where 𝑏 ௝ ሺ𝑡ሻ is the sound pressure after beamforming at focus point j , 𝑁 ௠௜௖ is the number of micro- phones, 𝑟 ௠௝ ሺ𝑡ሻ is the distance between focus point j and microphone m , 𝑝 ௠ is the sound pressure observed at microphone m , and 𝑐 ଴ is the sound speed. The beamformed sound pressure level is equiv- alent to a distance of 1 m apart from the sound source. Note that in this formula, the effect of fre- quency modulation due to the Doppler effect is taken into account because the sound pressure signal is upsampled when the train is approaching and vice versa when it leaves. In the DAS method, the sound source after array processing is widened over a finite width even for an ideal point source due to the finite array deployment area and the finite number of microphones. This distribution is called the point spread function (PSF). On the other hand, the deconvolution (DC) method has been developed to greatly improve the spatial resolution. This method removes the PSF defined by the array arrangement and improves the spatial resolution of the noise sources. Several deconvolution algorithms [3][4] with different characteristics have been developed. We used PULSE acoustics post-processing software (HBK division of Spectris), wherein the non- negative least-squares deconvolution method applicable to moving noise sources is implemented to identify the noise sources. However, this software is specialized for the HBK array system and is not capable of reading arbitrary arrays and obtaining time-series data. Therefore, we converted the time- series data of sound pressure into a database that can be read by the software so that signal processing calculations for source identification could be performed with our newly developed array system.

3.2. Comparison Between DAS and Deconvolution Noise source identification was conducted at a site without sound barriers. The array with a maximum diameter of 3 m developed by RTRI was used, as described in the previous sections, and was installed at the location shown in Figure 2 (left). Figure 4 compares the results of noise source identification using DAS and DC methods. The DC method is shown to improve the source resolution and clarify detailed sources. The spatial resolution width is approximately 1/3 times narrower in both the rail direction and the vertical direction, resulting in a spatial resolution improvement of approximately 10 times.

Figure 4: Difference between the DAS method (top) and the deconvolution method (bottom).

SPL La rge rg HodB. Small

3.3. Noise Source Distribution for a High-Speed Train Figure 5 shows the results of noise source identification for an inbound commercial train running at about 300 km/h. After applying A-weighting in the 160 Hz to 6.3 kHz band, the partial overall values are shown here, separated by every two cars from the leading car. It is known that the noise sources around Shinkansen trains are aerodynamic noise at the leading vehicle, lower-part noise, pantograph noise, and upper-part aerodynamic sound. The detailed source locations of these noise sources are indicated in Figure 5. It is clear that the aerodynamic noise generated from the bogie section of the leading vehicle is the largest in the vehicle, because high-speed airflow enters under the vehicle floor. However, lower- part noise at bogies except the leading vehicle should give great contribution to the total noise even if noise barriers are installed. The noise sources in the lower-part between cars are believed to be generated from the gaps and the yaw dampers between cars. The noise source of the pantographs is located at the panhead above the sidewall of car No. 5.

Figure 5: Noise source identification for a high-speed train running at about 300 km/h. (Partial overall values in the 160 Hz band to 6.3 kHz band) 4. MEASURES FOR REDUCING AERODYNAMIC BOGIE NOISE

From the field test described in the previous section, Shinkansen train bogies are presumed to generate a certain level of aerodynamic noise. Therefore, a wind tunnel test was conducted in the large-scale low-noise wind tunnel of RTRI for a bogie section, and measures to reduce aerodynamic bogie noise were verified.

4.1. Wind Tunnel Test for Aerodynamic Bogie Noise As shown in Figure 6, the aerodynamic noise radiating to the side of the bogie section was measured using a microphone array and a 1/7th scale model of an actual Shinkansen train vehicle. The main flow velocity was set at 320 km/h, the maximum speed of a Shinkansen. The microphone array was a wheel array with a diameter of 1 m consisting of 66 microphones. To improve the spatial resolution, the microphone array was placed 2060 mm away from the center of the wind tunnel (as close as possible to the model train in an enclosure that prevents the airflow from impinging on the micro- phone array). The front of the enclosure was covered with a porous metal material to allow penetra- tion of sound waves from the bogie section.

Inter-car gap Leading vehicle "_—— > Bogie noise (aerodynamic + rolling) Large 10dB Small

The sound pressure level was determined by the spatial integration of the obtained source distri- bution. Note that the gain of the microphone array depends on the frequency, so comparisons across frequencies do not necessarily reflect differences in sound source levels. The frequencies are con- verted to the scaled frequency by multiplying by the model scale (1/7).

Unit: mm

Nozzle Width: 3 m Height: 2.5 m

Bogie centerline

750 3000

Array enclosure

1/7scale train model

z

500

o

x Base stage

Schematic of a bogie

Microphone array

Enclosure of microphone array

Microphone array

Porous metal material

Figure 6: Side view of the wind tunnel test using a 1/7th scale train model and microphone array.

4.2. Noise Reduction Design Concept According to the literature [5], the noise sources in the bogie section are locally distributed at com- ponents exposed to the high-speed airflow between the vehicle bottom and the ground, such as the main motors, gear units, and wheels. Measures applicable for noise reduction have been previously discussed. However, it is not always easy to change the bogie design due to safety reasons. Therefore, in this study, we propose changing the shape of the vehicle near the bogie for noise reduction. The specific measures used herein are flat undercover and inner corners as shown in cases (a) and (b) in Figure 7. The primary purpose of the flat undercover is to improve the acoustic insulation effect. It should prevent the flow from winding at the bottom edge of the side cover and impinging along the downstream side of the cavity. The inner corners should suppress flow impinging on the downstream side of the cavity.

Basic conditions Measures

Baseline (a) Flat undercover (b) Embedded cavity inner corners Cavity Wheels without cavity

Figure 7: List of wind tunnel test cases (top: bottom view, bottom: side view)

4.3. Reduction of Aerodynamic bogie noise Figure 8 shows the reduction effect of changing the vehicle shape. The baseline test case is the same configuration as the standard vehicle equipped with side covers and a detailed bogie model. The cavity case is a condition with side covers wherein the bogie is removed from baseline condition. The wheels without cavity case is a configuration wherein the cavity space is flattened and only the lower- part of the wheels are installed. (1) Basic case Comparing baseline and cavity cases, the baseline case is about 4 dB greater than the cavity case over the entire frequency range (125 Hz to 1.25 kHz). Since the difference between these two cases is the presence or absence of the bogie, the results indicate that the aerodynamic noise generated by high-speed airflow impinging on the bogie is much greater than that generated at cavity. Next, com- paring wheels without cavity and baseline cases, there is no difference in frequencies ranging from 150 to 200 Hz. This frequency range should correspond to the aerodynamic noise radiating from wheels. Comparing wheels without the cavity case and with the cavity case at 300 Hz and above, the observed difference is cavity sound brought about by the presence or absence of the cavity. Since the major frequency of the cavity tone is on the order of tens of Hz, this is believed to represent the broad vortex sound caused by the cavity. (2) Noise Reduction Measures In Figure 8, we compare the results for case (a) flat undercover, (b) embedded cavity inner cor- ners, and (a)+(b) against baseline condition. The case of (a) is found to reduce the aerodynamic noise at bogie by 1 to 2 dB over a frequency range of 200 Hz to 700 Hz. This is mainly because the sound generated by bogie is prevented from being transmitted to the side, i.e., an acoustic sound insulation effect occurs. Only about 1 dB reduction of aerodynamic sound is observed for the embedded cavity inner corners case, which is smaller than that in the flat undercover case. This may be because the embedded cavity inner corners case reduced the cavity noise by about 3 dB. For the combination of these cases, sound insulation from the flat undercover case reduces aerodynamic noise generated by the bogie, and the embedded cavity inner corners reduce the cavity noise, resulting in a reduction of about 3 dB compared to baseline case. On the other hand, no effective noise reduction measures have been obtained for aerodynamic noise generated from the wheels. It would be effective to install a cowl just before the wheels or to reduce the flow velocity under the vehicle’s floor, but this is an issue to be addressed in future work.

70

Baseline

Spatially integrated SPL, dB

Cavity

65

Wheels without cavity

60

(a) Flat undercover

(b) Embedded cavity inner corners

55

(a)+(b) Embedded cavity inner corners with flat undercover

50

100 1000

Scaled frequency, Hz

Figure 8: Comparison of sound pressure levels under basic and countermeasure conditions (spatial integration of noise source identification by microphone array) 5. CONCLUSIONS

Using a microphone array, we measured the distribution of sound sources around a Japanese Shinkan- sen train. Measures to reduce the aerodynamic noise generated from the bogie section were verified using the large-scale low-noise wind tunnel at RTRI. The results obtained are as follows. (1) We designed and fabricated a microphone array suitable for noise source identification for Shinkansen trains, and successfully measured the noise source distribution around a train running at about 300 km/h. We applied a deconvolution method for signal processing to improve spatial resolution. (2) The result of introducing the deconvolution method for signal processing clarified that the main aerodynamic sources of the entire train are distributed in the lower part of the leading vehicle, pantographs, bogies, and inter-car gaps. (3) The aerodynamic bogie noise consists of the aerodynamic noise generated by the bogie and the cavity noise, the latter being about 4 dB lower than the former. (4) The aerodynamic bogie noise can be reduced by approximately 3 dB by simultaneously applying sound insulation using a flat undercover and cavity noise reduction by rounding the cavity inner corners. 6. REFERENCES

1. Nagakura, K. Recent studies on wayside environmental problems. Quarterly Report of RTRI , 58(2) , 88-92 (2017). 2. Mueller, T.J. Aeroacoustics Measurements , Springer, 2002. 3. Ehrenfried, K. & Koop, L. Comparison of iterative deconvolution algorithms for the mapping of acoustic Sources. AIAA Journal , 45(7) , 1584-1595 (2007). 4. Sijtsma, P. CLEAN based on spatial source coherence. International Journal of Aeroacoustics. 13th AIAA/CEAS Aeroacoustics Conference (28th AIAA Aeroacoustics Conference), AIAA Pa- per 2007-3436, 6(4) , 357-374 (2007). 5. Uda, T., Akutsu, M. & Kitagawa, T. Sound sources and measures of aerodynamic noise generated from bogies of Shinkansen train. Proceedings of Inter-Noise2020 , 15(6) , 294, Seoul, Korea(e- congress), August2020.