Sound source localization performance improvement method for acoustic wind tunnel test of large civil aircraft

Abstract

The invention discloses a sound source localization performance improvement method for a large civil aircraft acoustic wind tunnel test, and belongs to the technical field of aviation wind tunnel tests. The method comprises the following steps: after microphone array signal acquisition and block filtering noise reduction preprocessing, generating a sound source dirty image by adopting a beam forming algorithm, iteratively eliminating side lobe interference and reconstructing a high-resolution net image by combining a CLEAN-SC technology, and simultaneously, by applying a CPU and GPU cooperative computing architecture, through cross-spectrum matrix optimization, mixed precision floating point operation and a matrix vectorization acceleration strategy, obtaining a high-resolution sound source through a multi-spectral matrix optimization algorithm. On the premise that the sound imaging precision is guaranteed, the calculation efficiency of sound source positioning is greatly improved, and the problems that in a large-scale test, a traditional method is large in calculation resource consumption and low in execution speed are effectively solved.

Description

Technical Field

[0001] The present invention relates to the technical field of image generation, and particularly to a method for improving the sound source localization performance for the acoustic wind tunnel test of large civil aircraft. Background Art

[0002] As the most direct noise testing means, the aeroacoustic wind tunnel test can intuitively reflect the aeroacoustic performance of various equipment and guide the direction of noise reduction design. In the past few decades, various acoustic testing technologies have been developed in countries led by the UK and the US, such as sound source localization and duct mode identification, and applied to various aeroacoustic noise tests such as aircraft engine noise, wing noise, and rotor noise, aiming to improve the acoustic performance of various aircraft and guide the design and optimization direction of the next-generation products. Acoustic imaging is an important tool for identifying the position, amplitude, and shape of sound sources, and often uses near-field acoustic holography and far-field beamforming technologies.

[0003] Beamforming is a non-contact noise source measurement technology based on the sound pressure signals of a microphone array, and is widely used in fields such as aviation, aerospace, navigation, and rail transit. It can effectively identify the noise sources of key components such as the aircraft fuselage and aircraft engine, and provide a visual reference for noise control and reduction. The delay-and-sum beamforming method is the most classic time-domain acoustic imaging method. Because the spatial positions of the microphones in the array are different, the propagation times of the sound waves emitted by the same sound source reaching different microphones are different. This results in the beam after delay and summation operations showing directivity, thus achieving the purpose of localizing the sound source. In order to reduce the error caused by translating the time-domain signal, the time-domain method is often converted to the frequency domain, and the phase difference is used to characterize the delay time. At the same time, the background noise is eliminated by removing the auto-spectrum on the diagonal of the cross-spectrum matrix. In order to identify moving sound sources, a beamforming method in the time-frequency domain has been further derived.

[0004] Traditional beamforming methods can better identify medium- and high-frequency sound sources, but it is difficult to distinguish low-frequency sound sources. This has given rise to the deconvolution beamforming method and many similar derivative methods. Functionally, the deconvolution method aims to extract the local sound intensity maximum; in principle, they aim to identify the point spread function in the acoustic imaging results of traditional beamforming methods and gradually replace it with a single point or a narrow beam. The DAMAS method is based on non-negative constraints and uses the Gauss-Seidel iteration method to solve the inverse problem of the linear equation related to the point spread function. The actual sound source usually has spatial characteristics and is not concentrated at a certain point. In this regard, Pieter Sijtsma proposed the CLEAN method based on spatially coherent sound sources, simply referred to as CLEAN-SC. The core idea of CLEAN-SC is to discard the point spread function and extract the sound source from the traditional beamforming result by analyzing the spatial coherence.

[0005] The CLEAN-SC method is renowned for its stability and intuitive results. It has been verified through wind tunnel tests by many institutions worldwide and is regarded as the gold standard in the industry. Its wide application in the field of acoustic imaging makes it one of the preferred acoustic measurement methods in the industrial sector.

[0006] However, during the operation of the CLEAN-SC program, it is necessary to continuously iterate to extract the local maximum sound intensity, resulting in a relatively high cost for swept-frequency calculations (number of frequencies > 1000). Therefore, there is an urgent need to accelerate this method to meet the actual engineering requirements. By accelerating the algorithm, calculation time and resources can be saved, the data analysis, processing, and decision-making processes can be speeded up, and the efficiency of wind tunnel use can be improved, thus indirectly promoting the development of China's noise control cause. Summary of the Invention

[0007] To solve the technical problems of high computational resource consumption and low execution efficiency existing in the sound source localization method in traditional large aircraft acoustic wind tunnel tests, the present invention provides a method for improving the sound source localization performance for large civil aircraft acoustic wind tunnel tests, including the following steps:

[0008] S1. Receive sound pressure signals through a microphone array, and block and store the received signals;

[0009] S2. Batch process the blocked sound pressure signals;

[0010] S3. Perform Fourier transform on the processed sound pressure signals to convert the processed sound wave signals from the time domain to the frequency domain;

[0011] S4. According to the beamforming algorithm, perform weighting and phase adjustment on the microphone sound pressure signals after Fourier transform, calculate the sound power and cross-spectral matrix, and obtain an image on the scanning plane, which is called the dirty map;

[0012] S5. Suppress the sidelobes in the acoustic imaging results, improve the image resolution, and find the maximum sound power max{A dirty} and its position in the scanning plane;

[0013] S6. Based on the maximum sound power max{A dirty}, change the steering vector h(r,w) of the cross-spectral matrix to obtain the spatial correlation between the maximum point and the remaining points;

[0014] S7. Remove the influence of the maximum point from the cross-spectral matrix to obtain a new cross-spectral matrix, as well as the corresponding dirty map and clean map;

[0015] S8. After i iterations or C i+1 ≥C i ), the finally obtained acoustic imaging is the superposition of the clean beam and the remaining sound sources;

[0016] S9. Define the variables of the cross-spectral matrix, clean image, and dirty image as GPU variables and process them using a graphics processing unit; while process the remaining variables and related operations using a CPU with strong generality;

[0017] S10. Accelerate the processing of steps S1 - S8 through vectorization or matrix operations, and transform the loop processes in S1 - S8 into matrix operations;

[0018] S11. Accelerate the processing of steps S1 - S8 by adjusting the storage precision of floating-point numbers.

[0019] Furthermore, the sound power A is obtained by:

[0020] A = wCw *

[0021] where w is the directivity vector, C is the cross-spectral matrix, and the superscript * represents conjugate transpose;

[0022] The directivity vector is obtained by:

[0023]

[0024] where w is the circular frequency, r is the distance from the microphone to the discrete points on the scanning plane, the symbol |||| represents the two-norm, and G(r, w) is the three-dimensional Green's function in free space;

[0025] The three-dimensional Green's function in free space is obtained by:

[0026]

[0027] where i represents the imaginary number and c is the speed of sound;

[0028] Furthermore, the cross-spectral matrix is obtained by:

[0029]

[0030] where the elements in the lower triangle of the cross-spectral matrix are equal to the complex conjugates of the corresponding matrix elements in the upper triangle;

[0031] A single element of the cross-spectral matrix is obtained by:

[0032]

[0033] where m and n are microphone index numbers, S is the effective subset of microphones, and Y(r, w) is the microphone sound pressure signal after Fourier transform.

[0034] Furthermore, when (m ≠ n) ∈ S, the directivity vector h(r, w) of the cross-spectral matrix is obtained by:

[0035]

[0036] Obtained.

[0037] Furthermore, in S7, the new cross-spectral matrix is obtained by:

[0038]

[0039] Obtained, where is a safety factor, hh * is the outer product of the steering vector and its conjugate transpose.

[0040] The present invention also provides a computer device, which includes a memory and a processor. A computer program is stored in the memory. When the processor runs the computer program stored in the memory, the processor executes the above method for improving the sound source localization performance in the acoustic wind tunnel test of large civil aircraft.

[0041] The present invention also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the above method for improving the sound source localization performance in the acoustic wind tunnel test of large civil aircraft is implemented.

[0042] The beneficial effects of the present invention are as follows: Based on the existing acoustic imaging method based on traditional beamforming, the present invention abandons the point spread function, suppresses the sidelobes in the acoustic imaging result and improves the image resolution by analyzing the spatial coherence; while maintaining the algorithm accuracy, the execution efficiency of the acoustic imaging subject in the acoustic wind tunnel test of large civil aircraft is significantly improved. Description of the Drawings

[0043] Figure 1 is a flowchart of the method for improving the sound source localization performance in the acoustic wind tunnel test of large civil aircraft;

[0044] Figure 2 is a schematic diagram of the microphone array sensor layout in the acoustic wind tunnel test of the landing gear of large civil aircraft;

[0045] Figure 3 is the "dirty map" calculated by the beamforming algorithm;

[0046] Figure 4 is the "clean map" calculated by the CLEAN-SC algorithm. Detailed Embodiments

[0047] To make the technical solutions and advantages in the embodiments of the present invention clearer and more understandable, the following further details the exemplary embodiments of the present invention with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than an exhaustive list of all embodiments. It should be noted that, without conflict, the embodiments in the present invention and the features in the embodiments can be combined with each other.

[0048] Embodiment 1, combined with Figure 1 To illustrate this embodiment, the present invention provides a method for improving the sound source localization performance for the acoustic wind tunnel test of large civil aircraft, including the following steps:

[0049] S1. Receive the sound pressure signal through a microphone array, and block and store the received signal; the wind speed in the acoustic wind tunnel test of the landing gear of a large civil aircraft is 68 m / s, the temperature is 25°, the humidity is 50%, and the atmospheric pressure is 100.43 kPa. The sensor position coordinates of the microphone array are as Figure 2 shown. The total number of microphones is 143, the center point coordinates are (0, 0, 0) m, the sampling frequency is 81920 Hz, the sampling duration is 10 s, and it is divided into 200 blocks.

[0050] S2. Batch process the blocked sound pressure signals; the batch processing includes operations such as filtering and removing background noise to improve the signal quality.

[0051] S3. Perform Fourier transform on the processed sound pressure signal to convert the processed acoustic wave signal from the time domain to the frequency domain;

[0052] S4. According to the beamforming algorithm, perform weighting and phase regulation on the microphone sound pressure signal after Fourier transform, calculate the sound power and cross-spectral matrix, and obtain the image on the scanning plane, which is called the dirty map; the calculation results are as Figure 3 shown. The size of the scanning plane is 4 m × 4 m, the resolution is 100 × 100, and the center coordinates are (0, 0.5, 8.24) m.

[0053] S5. Suppress the sidelobes in the acoustic imaging result, improve the image resolution, and find the maximum sound power max{A dirty} and its position in the scanning plane;

[0054] S6. Based on the maximum sound power max{A dirty}, change the steering vector h(r, w) of the cross-spectral matrix to obtain the spatial correlation between the maximum point and the remaining points;

[0055] S7. Remove the influence of the maximum point from the cross-spectral matrix to obtain a new cross-spectral matrix, and the corresponding dirty map and clean map;

[0056] S8. After i iterations or Ci+1 ≥C i After that, the finally obtained acoustic imaging is the superposition of the clean beam and the remaining sound sources; the calculation results in S8 are as Figure 4 shown.

[0057] S9. Define the variables of the cross-spectral matrix, the net map, and the dirty map as GPU variables and use the graphics processing unit for processing; for the remaining variables and related operations, use the more general-purpose CPU for processing; among them, the central processing unit is suitable for processing complex logical operations and sequential tasks and has strong general computing capabilities and flexibility; while the graphics processing unit is suitable for processing large-scale data and highly parallel tasks.

[0058] S10. Accelerate the steps of S1 - S8 through vectorization or matrix operations, and transform the loop process in S1 - S8 into matrix operations;

[0059] S11. Accelerate the steps of S1 - S8 by adjusting the floating-point storage precision. For acoustic imaging, the precision or the digits after the decimal point in the eighth place have little impact on the final sound pressure level and can be considered to be much less than 1%. Therefore, further improve the performance by reducing the default double-precision data to a single custom mixed precision.

[0060] Specifically, after the acceleration processing of S9 - S11, the comparison between the basic algorithm and the accelerated algorithm is shown in the following table:

[0061] Table 1 Comparison Table of Basic Algorithm and Accelerated Algorithm

[0062] Parallel computing time consumption (CPU) Parallel computing time consumption (GPU) Basic algorithm 1557.27s NAN Acceleration algorithm 308s 232s Acceleration effect 80.2％ 85.2％

[0063] It can be seen from the comparison table that the accelerated algorithm significantly reduces the calculation time of acoustic imaging from 1557.27 seconds of the basic algorithm to 308 seconds on the CPU (80.2% acceleration) and 232 seconds on the GPU (85.2% acceleration), proving that the hardware adaptation and algorithm optimization significantly improve the efficiency.

[0064] Furthermore, the sound power A is obtained through:

[0065] A = wCw *

[0066] where w is the directivity vector, C is the cross-spectral matrix, and the superscript * represents the conjugate transpose;

[0067] The directivity vector is obtained through:

[0068]

[0069] obtained, where \(w\) is the circular frequency, \(r\) is the distance from the microphone to the discrete points on the scanning plane, the symbol \(\|\cdot\|\) represents the two-norm, and \(G(r, w)\) is the three-dimensional Green's function in free space;

[0070] The three-dimensional Green's function in free space is obtained by:

[0071]

[0072] obtained, where \(i\) represents the imaginary unit and \(c\) is the speed of sound;

[0073] Furthermore, the cross-spectral matrix is obtained by:

[0074]

[0075] obtained, where the elements in the lower triangle of the cross-spectral matrix are equal to the complex conjugates of the corresponding matrix elements in the upper triangle;

[0076] A single element of the cross-spectral matrix is obtained by:

[0077]

[0078] obtained, where \(m\) and \(n\) are microphone index numbers, \(S\) is the effective subset of microphones, and \(Y(r, w)\) is the microphone sound pressure signal after Fourier transform.

[0079] Furthermore, when considering \((m\neq n)\in S\), the steering vector \(h(r, w)\) of the cross-spectral matrix is obtained by:

[0080]

[0081] obtained.

[0082] Furthermore, in S7, the new cross-spectral matrix is obtained by:

[0083]

[0084] obtained, where is the safety factor, hh * is the outer product of the steering vector and its conjugate transpose.

[0085] Embodiment 2: The computer device of the present invention may be a device including a processor and a memory, such as a single-chip microcomputer including a central processing unit. Moreover, when the processor executes the computer program stored in the memory, the steps of the above method for improving the sound source localization performance for large civil aircraft acoustic wind tunnel tests are implemented.

[0086] Embodiment 3: Embodiment of computer-readable storage medium.

[0087] The computer-readable storage medium of the present invention can be any form of storage medium readable by the processor of a computer device, including but not limited to non-volatile memory, volatile memory, ferroelectric memory, etc. A computer program is stored on the computer-readable storage medium. When the processor of the computer device reads and executes the computer program stored in the memory, the steps of the above method for improving the sound source localization performance for large civil aircraft acoustic wind tunnel tests can be implemented.

[0088] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any simple substitution or change within the technical idea scope disclosed by the present invention and according to the technical solution of the present invention should be within the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims

1. A method for improving the sound source localization performance for large civil aircraft acoustic wind tunnel tests, characterized in that, Including the following steps: S1. Receive the sound pressure signal through a microphone array, and block and store the received signal; S2. Batch process the blocked sound pressure signals; S3. Perform Fourier transform on the processed sound pressure signal to convert the processed sound wave signal from the time domain to the frequency domain; S4. According to the beamforming algorithm, perform weighting and phase regulation on the Fourier-transformed microphone sound pressure signal, calculate the sound power and cross-spectral matrix, and obtain the image on the scanning plane, which is called the dirty map; S5. Suppress the sidelobes in the acoustic imaging result and improve the image resolution. Find the maximum value of the acoustic power max{A dirty} and its position in the scanning plane from the dirty image; S6. Based on the maximum sound power max{A dirty}, change the steering vector h(r,w) of the cross-spectral matrix to obtain the spatial correlation between the maximum point and the remaining points; S7. Remove the influence of the maximum point from the cross-spectral matrix to obtain a new cross-spectral matrix, as well as the corresponding dirty map and clean map; S8. After i iterations or when C i+1 ≥ C i , the finally obtained acoustic imaging is the superposition of the clean beam and the remaining sound sources; S9. Define the variables of the cross-spectral matrix, clean map, and dirty map as GPU variables and use a graphics processing unit for processing; while the remaining variables and related operations are processed using a general-purpose CPU with stronger versatility; S10. Accelerate the processing of steps S1 - S8 through vectorization or matrix operations, and transform the loop process in S1 - S8 into matrix operations; S11. Accelerate the processing of steps S1 - S8 by adjusting the floating-point storage precision.

2. The method for improving the sound source localization performance for the acoustic wind tunnel test of large civil aircraft according to claim 1, wherein The sound power A is obtained through: A = wCw * , where w is the steering vector, C is the cross-spectral matrix, and the superscript * represents conjugate transpose; The steering vector is obtained through: , where w is the circular frequency, r is the distance from the microphone to the discrete point on the scanning plane, the symbol |||| represents the two-norm, and G(r, w) is the three-dimensional Green's function in free space; The three-dimensional Green's function in free space is obtained through: , where i represents a complex number and c is the speed of sound.

3. The method for improving the sound source localization performance for acoustic wind tunnel tests of large civil aircraft according to claim 1, characterized in that, The cross-spectral matrix is obtained through: , where the elements in the lower triangle of the cross-spectral matrix are equal to the complex conjugates of the corresponding matrix elements in the upper triangle; A single element of the cross-spectral matrix is obtained through: , where m and n are the microphone index numbers, S is the effective subset of microphones, and Y(r, w) is the Fourier-transformed microphone sound pressure signal.

4. The method for improving the sound source localization performance for the acoustic wind tunnel test of large civil aircraft according to claim 1, characterized in that When considering (m ≠ n) ∈ S, the steering vector h(r, w) of the cross-spectral matrix is obtained through: .

5. The method for improving the sound source localization performance for the acoustic wind tunnel test of large civil aircraft according to claim 1, wherein In S7, the new cross-spectral matrix is obtained through: obtained, where, is a safety factor, hh * is the outer product of the pointing vector and its conjugate transpose.

6. A computer device, comprising a memory and a processor, characterized in that, A computer program is stored in the memory. When the processor runs the computer program stored in the memory, the processor executes the method for improving the sound source localization performance for large civil aircraft acoustic wind tunnel tests according to any one of claims 1 - 5.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, it implements the method for improving the sound source localization performance for large civil aircraft acoustic wind tunnel tests according to any one of claims 1 - 5.

Images