Method for positioning abnormal sound under metro vehicle

Abstract

The invention relates to an abnormal sound positioning method under a metro vehicle, which comprises the following steps of: acquiring sound data under the metro vehicle through a microphone array, performing analog-to-digital conversion on the sound data through a data acquirer, transmitting the sound data to a filter, filtering the digitized sound data by the filter, removing background noise, and positioning the abnormal sound under the metro vehicle. The effective data after the background noise is removed are sent to the data processor to calculate the actual relative time delay between the microphones, and the position of a sound source where abnormal sound is generated is located by adopting a gradient descent method according to the actual relative time delay between the microphones. The abnormal sound positioning method with high systematicness and scientificity can be realized, so that the test time is shortened, the test efficiency is improved, the test cost is reduced, powerful support is provided for maintenance and detection of metro vehicles, and the safety and reliability of vehicle operation are improved.

Description

Technical Field

[0001] The invention relates to the technical field of sound source positioning, and in particular to a method for positioning abnormal noise under a subway vehicle. Background Art

[0002] During the operation of subway vehicles, various abnormal noises often occur due to the complex mechanical structure and high-load operating conditions. These abnormal noises not only affect the passengers' riding experience, but may also indicate potential mechanical failures and even affect driving safety. Therefore, timely and accurate positioning and elimination of abnormal noises under subway vehicles is of great significance to ensure the safety and reliability of subway operations. At present, the positioning of abnormal noises under subway vehicles mainly relies on manual experience and conventional detection methods. However, these methods have some limitations. First, manual experience is subjective, and the judgments of different detection personnel may vary greatly. Second, conventional detection methods usually require the vehicle to stop running for inspection, which not only affects operational efficiency, but may also aggravate the fault due to long downtime. In addition, subway vehicles have complex structures and diverse noise sources, and traditional positioning methods are difficult to accurately identify the source of abnormal noise in a complex environment.

[0003] In recent years, abnormal noise location methods based on sensor technology and signal processing technology have gradually attracted attention. These methods install multiple sensors (such as accelerometers, microphones, etc.) at key parts of subway vehicles to monitor the vehicle operation status in real time, collect vibration and acoustic signals, and use advanced signal processing algorithms for analysis and identification, thereby achieving automatic location of abnormal noise. This type of method not only improves the accuracy and real-time performance of positioning, but also can perform detection under normal vehicle operation, minimizing the impact on operations.

[0004] However, the existing sensor-based abnormal noise location methods still have some technical difficulties. For example, how to effectively filter out background noise in a complex undercar environment and extract characteristic signals related to faults; how to process large-scale, multi-dimensional sensor data to achieve efficient and accurate signal analysis and fault diagnosis; how to combine advanced algorithms with existing subway operation and maintenance systems in practical applications to form a complete abnormal noise location solution. These are technical problems that need to be solved urgently. Summary of the invention

[0005] A main purpose of the present invention is to overcome at least one of the above-mentioned defects, and to provide a method for locating abnormal noise under a subway vehicle, which can effectively improve the accuracy and efficiency of locating abnormal noise under a subway vehicle and provide strong support for the operation and maintenance of subway vehicles.

[0006] In order to achieve the above object, the technical solution adopted by the present invention is: The present invention provides a method for locating abnormal noise under a subway vehicle, wherein the sound data under the subway vehicle is collected by a microphone array and transmitted to a filter after analog-to-digital conversion by a data collector, the filter performs filtering processing on the digitized sound data and removes background noise, and the effective data after the background noise is removed is sent to the data processor to calculate the actual relative time delay between each microphone, and the gradient descent method is used to locate the sound source position of the abnormal noise according to the actual relative time delay between each microphone.

[0007] According to one embodiment of the present invention, the microphone array is fixedly mounted on the frame of the subway vehicle, and the microphone array includes a plurality of microphones, and the plurality of microphones are arranged in at least two rows and two columns, and the distance between adjacent microphones is d.

[0008] According to one embodiment of the present invention, the filter uses a wavelet packet unbalanced multi-dimensional noise reduction method to filter the sound data.

[0009] According to one embodiment of the present invention, the data processor calculates the cross-power spectrum of the sound data collected by each microphone in the microphone array by a generalized cross-correlation method to obtain the actual relative time delay between each microphone, constructs a delay matrix based on the obtained actual relative time delay, and processes the delay matrix by a gradient descent method to establish the sound source position.

[0010] According to one embodiment of the present invention, the obtained data of actual relative delays are interpolated and resampled to obtain a delay matrix in which each actual relative delay smoothly transitions in the frequency domain.

[0011] According to one embodiment of the present invention, the process of locating the sound source of the abnormal sound by using the gradient descent method according to the actual relative time delay between each microphone includes the following steps: Step S1: Select a virtual sound source position, calculate the estimated time for the virtual sound source position to propagate to each microphone in the microphone array, and calculate the estimated relative time delay between each microphone based on the estimated time, and then calculate the loss function by combining the estimated relative time delay with the actual relative time delay between each microphone measured and record it; Step S2: calculating the gradient of the loss function with respect to the virtual sound source position by a gradient descent method, and updating the virtual sound source position according to the calculated gradient; Step S3: Repeating the steps S1 and S2 according to the updated virtual sound source position until the loss function converges to a preset value or reaches the number of iterations; Step S4: Statistically determine the minimum value of each of the loss functions, and use the virtual sound source position corresponding to the minimum loss function as the sound source position of the abnormal sound.

[0012] According to one embodiment of the present invention, in step S1, the coordinate position of each microphone is located at a known coordinate position. , where i is a natural number from 1 to N, and N is the number of microphones in the microphone array; Define the virtual sound source position at an unknown coordinate position ; The loss function The calculation formula is as follows , in, is the time delay between the i-th microphone and the j-th microphone.

[0013] According to one embodiment of the present invention, in step S2, the loss function is calculated by the gradient descent method. The location of the sound source Gradient ,gradient The calculation formula is .

[0014] According to one embodiment of the present invention, in step S2, in the process of updating the virtual sound source position according to the calculated gradient, the step length of updating the virtual sound source position is preset to be , the formula for updating the virtual sound source position is as follows: in, is the sound source position at the kth iteration.

[0015] Compared with the prior art, the advantages and beneficial effects of the method for locating abnormal noise under a subway vehicle in the patent application of the present invention are: The method for locating abnormal noise under a subway vehicle of the present application optimizes the structural layout of the microphone as a sound collection sensor, and at the same time improves the subsequent signal processing algorithm, and adopts wavelet packet unbalanced multidimensional noise reduction filtering technology to effectively eliminate background noise and other noise interference, thereby improving the accuracy of sound source positioning. At the same time, the method combines the generalized cross-correlation method to calculate the relative time delay between different microphones and interpolate the time delay matrix, and adopts the gradient descent method to locate the position of the abnormal noise source, so that a systematic and scientific abnormal noise positioning method can be realized, thereby shortening the test time, improving the test efficiency, and reducing the test cost, providing strong support for the maintenance and inspection of subway vehicles, and improving the safety and reliability of vehicle operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Hereinafter, some specific embodiments of the present invention will be described in detail in an exemplary and non-limiting manner with reference to the accompanying drawings. The same reference numerals in the accompanying drawings indicate the same or similar components or parts. It should be understood by those skilled in the art that these drawings are not necessarily drawn to scale. In the accompanying drawings: Figure 1 is a result schematic diagram of a microphone array according to an embodiment of the present invention; Figure 2 is a schematic diagram of a system framework of a method for locating abnormal noise under a subway vehicle according to an embodiment of the present invention; Figure 3 is a schematic diagram of wavelet packet analysis changes on sound data according to one embodiment of the present application; Figure 4 is a simulation diagram of filtering the sound data using a wavelet packet unbalanced multi-dimensional noise reduction method according to one embodiment of the present application; Figure 5 This is a simulation diagram of determining the source of abnormal noise under a subway vehicle obtained according to an embodiment of the present application. DETAILED DESCRIPTION

[0017] The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0018] In addition, the technical features involved in the different embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0019] This embodiment describes a method for locating abnormal noise under a subway vehicle, wherein: Figure 2 As shown, the sound data under the subway vehicle is collected by a microphone array and transmitted to the filter after analog-to-digital conversion by a data collector. The filter filters the digitized sound data and removes background noise. The valid data after background noise removal is sent to the data processor to calculate the actual relative time delay between each microphone. The gradient descent method is used to locate the sound source position of the abnormal sound according to the actual relative time delay between each microphone.

[0020] In this embodiment, the microphone array is fixed on the bogie at the bottom of the subway vehicle. Figure 1As shown, the microphone array is composed of 16 microphones arranged in a 2×8 rectangular shape, and the distance between adjacent microphones is d, which is evenly spaced. The signal collected by each microphone can be amplified by an amplifier and then connected to a data collector, or directly sent to a data collector. The data collector performs digital processing on the collected sound data and converts the analog signal into a digital signal. The filter performs filtering processing on the digitized sound data and removes background noise. The valid data after removing the background noise is sent to the data processor. The filter and the data processor can be set up as functional components separately to realize their functions, or they can be made into software and integrated into a computer or tablet computer, and directly perform subsequent processing after receiving the digitized sound data.

[0021] The filter uses a wavelet packet unbalanced multi-dimensional noise reduction method to filter the sound data.

[0022] During the operation of subway vehicles, background noise and murmurs are important factors affecting the location of abnormal noise under the vehicle. Filtering with wavelet packet unbalanced multidimensional noise reduction method can effectively separate signals and noise. Through multi-scale analysis and frequency band decomposition, the signal is processed at different frequencies and time scales, which can better suppress noise and improve the signal-to-noise ratio.

[0023] Traditional wavelet db10 is often used for noise removal of sound signals. Therefore, wavelet db10 is used to perform wavelet packet decomposition on sound signals to achieve filtering. Performing wavelet transform on sound signal x(t) is equivalent to observing the signal through the change of wavelet scale factor and time shift factor. The localization of wavelet transform is variable. At high frequencies, the time resolution is high and the frequency resolution is low; at low frequencies, the time resolution is low and the frequency resolution is high, that is, it has the property of "zooming", that is, it has the property of adaptive window.

[0024] Since the data collector has converted the analog signal obtained by the microphone into a digital signal and the sampling rate is known, the basic formula of wavelet transform is: Among them, a is the scale parameter, which controls the expansion and contraction of the wavelet; b is the translation parameter, which controls the position of the wavelet; W(a, b) is the continuous wavelet coefficient of the signal at scale a and position b. By using the expansion and translation wavelet function, it can decompose the signal into different scales and positions. This enables it to provide time and frequency information at the same time, and realize the discrete decomposition and reconstruction of the signal through multi-resolution analysis technology.

[0025] The above wavelet transform only decomposes the low-frequency band of the signal, while the wavelet packet transform inherits the time-frequency analysis characteristics of the wavelet transform, further decomposes the high-frequency band signal that is not decomposed in the wavelet transform, and makes different resolution selections for various frequencies at different levels. At various scales, a series of sub-band time domain waveforms are provided within the full frequency band. Wavelet packet analysis is to further subdivide the wavelet subspace in a binary manner to achieve the purpose of improving the frequency resolution. The schematic diagram is shown below. Figure 3 shown.

[0026] In wavelet packet transform, the sound signal is decomposed through a series of filters. Assume f[n] is the original signal, h[n] is a low-pass filter, and g[n] is a high-pass filter. The decomposition and reconstruction of wavelet packet transform can be expressed by the following recursive formula.

[0027] At level j, the signal of the kth node It can be decomposed by the following formula: It can be decomposed by the following formula: After filtering to eliminate the interference of background noise and noise, the signal at the jth level and the kth node is reconstructed: Through the above decomposition and reconstruction formulas, wavelet packet transform can perform detailed analysis and reconstruction of the sound signal in the entire frequency band. The decomposition process involves recursive low-pass and high-pass filtering of the signal, and the reconstruction process restores the signal through the corresponding inverse filtering operation, thereby effectively separating the background noise and noise that affect the location of the sound source.

[0028] The auxiliary design is carried out in Matlab, and the index parameters are determined according to the accuracy requirements of the specific application occasion. Since the center frequency of the filter changes with different signals, it is necessary to set the derived center frequency of the filter as a variable, and finally obtain a universal signal extraction system. The above wavelet packet denoising process is tested, and the results are as follows: Figure 4 As shown, the signal after noise reduction is significantly optimized, making it easier to identify the source of the abnormal noise.

[0029] In one embodiment, the data processor calculates the cross-power spectrum of the sound data collected by each microphone in the microphone array by a generalized cross-correlation method to obtain the actual relative time delay between each microphone, constructs a delay matrix based on the obtained actual relative time delay, and processes the delay matrix by a gradient descent method to establish the sound source position.

[0030] In this embodiment, assuming that there are m single-frequency sound sources, the sampled data obtained after wavelet packet unbalanced multidimensional noise reduction filtering can be represented by the following microphone array model: In the formula, represents the sampled data of the microphone array after filtering, represents the sound source signal to be identified, and D is the arrival time difference of the sound signal at two microphones in the microphone array, for example It is the time difference between the arrival of the sound signal at the second microphone and the first microphone. represents the residual noise after filtering.

[0031] The actual relative time delay between the two microphones in the microphone array is calculated by the generalized cross-correlation method proposed in this embodiment, and the value of the actual relative time delay is obtained by maximizing the cross-correlation function between the filtered sampling signals of the two microphones p and q. The specific calculation formula is as follows: Based on this, the actual relative delay between each microphone can be calculated, and then the delay matrix can be constructed.

[0032] However, considering that the sampling points are discrete, in order to reconstruct the continuous signal between discrete sampling points, Sinc interpolation is used to perform high-quality sound signal resampling so that the delay matrix can transition smoothly in the frequency domain. The formula is: In the process of calculating the relative delay, the reconstruction of the continuous signal is realized by using the sinc interpolation function to the discrete sampling points, which can further avoid the influence of possible high-frequency noise. That is, the initial value of the calculated relative delay matrix.

[0033] The introduction of the Sinc interpolation function can realize further processing of the collected sound signal. This step is mainly carried out in the process of calculating the delay. Because in the processing of speech signals, the sinc difference function can well restore the original continuous signal. Therefore, for the consideration of positioning accuracy, such a sinc interpolation function is added, which can effectively improve the positioning accuracy.

[0034] Finally, the delay matrix T is obtained as follows: The conditions for determining the near and far fields in acoustics can indeed be determined by 2d 2 / λ indicates that this formula is used to calculate the boundary between the near and far fields, where d represents the distance between the microphones in the microphone array, and λ is the wavelength of the sound wave. This formula is applied based on the assumption of a point sound source and takes into account the speed of sound (the speed of sound in air is about 343m / s) and the frequency of the sound source. Specifically, when the distance ρ>2d between the sound source and the microphone array 2 / λ, it can be regarded as a far-field sound source, ρ<2d 2 / λ, it can be regarded as a near-field sound source. In this embodiment, the location where the abnormal sound usually occurs is far away from the location of the microphone array, so it can be regarded as a far-field sound source. In the sound source localization model under far-field conditions, the sound can be regarded as propagating in the air in the form of a plane wave, ignoring the amplitude difference between the sound signals received by each microphone. Since the microphones receive signals in parallel, the sound information collected by each microphone sensor only differs by time delay. Among them, the angles between the transmission direction of a single sound wave and the microphone sensor array are the pitch angle θ and the azimuth angle φ, and the distance between adjacent microphones is d. ‌

[0035] Specifically, the process of locating the sound source of the abnormal sound using the gradient descent method according to the actual relative time delay between the microphones includes the following steps: Step S1: Select a virtual sound source position, calculate the estimated time for the virtual sound source position to propagate to each microphone in the microphone array, and calculate the estimated relative time delay between each microphone based on the estimated time, and then calculate the loss function by combining the estimated relative time delay with the actual relative time delay between each microphone measured and record it; Step S2: calculating the gradient of the loss function with respect to the virtual sound source position by a gradient descent method, and updating the virtual sound source position according to the calculated gradient; Step S3: Repeating the steps S1 and S2 according to the updated virtual sound source position until the loss function converges to a preset value or reaches the number of iterations; Step S4: Statistically determine the minimum value of each of the loss functions, and use the virtual sound source position corresponding to the minimum loss function as the sound source position of the abnormal sound.

[0036] In this embodiment, Figure 1 As shown, the straight line where a row of microphones (8 microphones) are located is taken as the X-axis, the direction perpendicular to the X-axis in the same horizontal plane is taken as the Y-axis, and the direction perpendicular to the X-axis in the same vertical plane is taken as the Z-axis. Generally speaking, the length direction of the subway vehicle is taken as the X-axis, the width direction is taken as the Y-axis, and the height direction is taken as the Z-axis for the convenience of subsequent calculations.

[0037] In one embodiment, in step S1, the coordinate position of each microphone is located at a known coordinate position , where i is a natural number from 1 to N, and N is the number of microphones in the microphone array; Define the virtual sound source position at an unknown coordinate position ; The loss function The calculation formula is as follows , in, is the time delay between the i-th microphone and the j-th microphone.

[0038] In step S2, the loss function is calculated by gradient descent method The location of the sound source Gradient ,gradient The calculation formula is .

[0039] In addition, in step S2, in the process of updating the virtual sound source position according to the calculated gradient, the step length of updating the virtual sound source position is preset to , the formula for updating the virtual sound source position is as follows: in, is the sound source position at the kth iteration.

[0040] The final determined sound source position in this embodiment can be expressed by rectangular coordinates in a three-axis coordinate system, that is, the coordinates as above .

[0041] In another embodiment, the angle between the sound source position and the microphone sensor array can be expressed as the pitch angle θ and the azimuth angle φ, and the sound source position expression coordinates are (θ, φ). The expression of the pitch angle θ and the azimuth angle φ can facilitate physical positioning in the later stage, and this expression method can adapt to the characteristics of the sound signal (far-field signal). However, the rectangular coordinates usually obtained by calculation are , when converting, it is calculated by the following formula: Where r is the polar diameter in the spherical coordinate system, .

[0042] The results of the sound source localization simulation experiment are as follows: Figure 5As shown in the figure, the angular coordinates of the actual position of the sound source are azimuth [10°, 100°, 160°], and elevation [15°, 50°, 80°]. The angular coordinates of the positioning result are azimuth [10°, 100°, 159°], and elevation [15°, 52°, 83°].

[0043] In summary, the method for locating abnormal noise under a subway vehicle of the present application optimizes the structural layout of the microphone as a sound collection sensor, and at the same time improves the subsequent signal processing algorithm, and adopts wavelet packet unbalanced multidimensional noise reduction filtering technology to effectively eliminate background noise and other noise interference, thereby improving the accuracy of sound source positioning. At the same time, the method combines the generalized cross-correlation method to calculate the relative time delay between different microphones and interpolate the time delay matrix, and uses the gradient descent method to locate the position of the abnormal noise source. In this way, a systematic and scientific abnormal noise positioning method can be implemented, thereby shortening the test time, improving the test efficiency, and reducing the test cost, providing strong support for the maintenance and inspection of subway vehicles, and improving the safety and reliability of vehicle operation.

[0044] The above embodiments are only for illustrating the technical concept and features of the present invention, and their purpose is to enable people familiar with this technology to understand the contents of the present invention and implement them accordingly. They cannot be used to limit the protection scope of the present invention. Any equivalent changes or modifications made according to the spirit of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for locating abnormal noise under a subway vehicle, characterized in that: The sound data under the subway vehicle is collected by a microphone array and transmitted to the filter after analog-to-digital conversion by a data collector. The filter filters the digitized sound data and removes background noise. The valid data after background noise removal is sent to the data processor to calculate the actual relative time delay between each microphone. The gradient descent method is used to locate the sound source of the abnormal sound based on the actual relative time delay between each microphone.

2. The method for locating abnormal noise under a subway vehicle according to claim 1, characterized in that: The microphone array is fixedly mounted on the frame of the subway vehicle. The microphone array includes a plurality of microphones. The plurality of microphones are arranged in at least two rows and two columns, and the distance between adjacent microphones is d.

3. The method for locating abnormal noise under a subway vehicle according to claim 1, characterized in that: The filter uses a wavelet packet unbalanced multi-dimensional noise reduction method to filter the sound data.

4. The method for locating abnormal noise under a subway vehicle according to claim 1, characterized in that: The data processor calculates the cross-power spectrum of the sound data collected by each microphone in the microphone array by the generalized cross-correlation method, which is used to obtain the actual relative time delay between each microphone, constructs a delay matrix according to the obtained actual relative time delay, and processes the delay matrix by the gradient descent method to establish the sound source position.

5. The method for locating abnormal noise under a subway vehicle according to claim 4, characterized in that: The obtained data of actual relative time delays are interpolated and resampled to obtain a time delay matrix in which each actual relative time delay smoothly transitions in the frequency domain.

6. The method for locating abnormal noise under a subway vehicle according to any one of claims 1 to 5, characterized in that: The process of locating the source of the abnormal sound using the gradient descent method according to the actual relative time delay between the microphones includes the following steps: Step S1: Select a virtual sound source position, calculate the estimated time for the virtual sound source position to propagate to each microphone in the microphone array, and calculate the estimated relative time delay between each microphone based on the estimated time, and then calculate the loss function by combining the estimated relative time delay with the actual relative time delay between each microphone measured and record it; Step S2: Calculate the gradient of the loss function with respect to the virtual sound source position by a gradient descent method, and update the virtual sound source position according to the calculated gradient; Step S3: Repeating the steps S1 and S2 according to the updated virtual sound source position until the loss function converges to a preset value or reaches the number of iterations; Step S4: Statistically determine the minimum value of each of the loss functions, and use the virtual sound source position corresponding to the minimum loss function as the sound source position of the abnormal sound.

7. The method for locating abnormal noise under a subway vehicle according to claim 6, characterized in that: In step S1, the coordinate position of each microphone is located at a known coordinate position , where i is a natural number from 1 to N, and N is the number of microphones in the microphone array; Define the virtual sound source position at an unknown coordinate position ; The loss function The calculation formula is as follows ， in, is the time delay between the i-th microphone and the j-th microphone.

8. The method for locating abnormal noise under a subway vehicle according to claim 6, characterized in that: In step S2, the loss function is calculated by gradient descent method The location of the sound source Gradient ,gradient The calculation formula is 。 9. The method for locating abnormal noise under a subway vehicle according to claim 8, characterized in that: In step S2, in the process of updating the virtual sound source position according to the calculated gradient, the step length of updating the virtual sound source position is preset to , the formula for updating the virtual sound source position is as follows: in, is the sound source position at the kth iteration.

Images