Gearbox fault positioning method and device based on acoustic image, equipment and medium

By combining microphone line arrays and CNN models with principal component analysis and near-field acoustic holography, precise location of gearbox faults was achieved, overcoming the limitations of vibration signal analysis in existing technologies and improving the accuracy of fault identification.

CN118918920APending Publication Date: 2024-11-08AVIC SHANGHAI AERONAUTICAL MEASUREMENT CONTROLLING RES INST
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410964449.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-18
Publication Date
2024-11-08

AI Technical Summary

Technical Problem

In existing technologies, gearbox fault diagnosis mainly relies on vibration signal analysis, which has limitations. In particular, it is difficult to accurately locate faults when vibration characteristics are not obvious, and the value of acoustic imaging technology in fault diagnosis has not been fully utilized.

Method used

A microphone line array is used to collect the gearbox acoustic signal. The dominant frequency band is obtained through noise reduction and spectrum analysis. The sound field is visualized by combining principal component analysis and near-field acoustic holography. A CNN model is used for fault identification, and precise positioning is achieved through sound-image matching.

Benefits of technology

It improves the accuracy of gearbox fault location by enhancing fault identification capabilities through acoustic field spatial tomography and feature vector analysis, thus achieving precise fault location.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118918920A_ABST
    Figure CN118918920A_ABST
Patent Text Reader

Abstract

The invention relates to a gearbox fault positioning method, device, equipment and medium based on a sound image, and the method comprises the steps: collecting a gearbox operation sound signal through a microphone line array, carrying out the noise reduction of the sound signal, carrying out the spectrum analysis, intercepting the noise-reduced sound signal corresponding to an interested frequency, and carrying out the noise reduction of the noise-reduced sound signal, a dominant frequency band region is obtained; performing sound field visualization on the dominant frequency band region by a sound image graph acquisition method based on principal component analysis and near-field acoustical holography to obtain a sound field space fault slice with dominant frequency; inputting the sound field space fault slice into a trained CNN model to obtain a fault identification result of the gearbox; and fault sound image matching is carried out according to the fault identification result of the gearbox, and accurate fault positioning is realized. According to the invention, accurate fault positioning can be realized.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of gearbox fault diagnosis, and in particular to a method, device, equipment and medium for locating a gearbox fault based on an acoustic image. Background Art

[0002] In recent years, as the acoustic imaging technology of array measurement has become increasingly mature, the acoustic signal of the sound field can be measured by a microphone array and the source image can be reproduced, that is, the surface vibration of the sound source and its radiated sound field can be fully reproduced, which has created conditions for the development of new acoustic fault diagnosis technology. At present, acoustic imaging technology is mainly used to identify and locate the sound source through the distribution of sound pressure or vibration velocity in the sound image. It is only an important preliminary work for mechanical noise analysis and control, and the engineering utilization value of acoustic imaging has not been fully utilized. If it is possible to locate the sound source or vibration source of the machine based on the sound image, it is also possible to monitor the state of the machine operation, diagnose the fault of the machine and determine the location of the fault, introduce acoustic imaging technology into the field of fault diagnosis, and develop it into a fault diagnosis method based on sound image that integrates sound source identification and fault diagnosis, it will have great engineering application value.

[0003] The transmission shaft and gear failures inside the gearbox can only be analyzed and diagnosed by measuring the vibration signal at the bearing seat. The acquisition of vibration signals is limited by contact measurement, and single-channel signal analysis is limited to local diagnosis. Fault diagnosis based on vibration signal processing has certain limitations. In addition, due to the variety of gearbox failure modes, the characteristics of their performance are also different. In some fault conditions, the vibration characteristics are not obvious, but the acoustic characteristics are more obvious. Therefore, it is very necessary to develop fault diagnosis methods based on acoustic signals. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide a gearbox fault locating method, device, equipment and medium based on an audio-visual image, which can realize accurate fault locating.

[0005] The technical solution adopted by the present invention to solve the technical problem is: to provide a gearbox fault location method based on an acoustic image, comprising the following steps:

[0006] The gearbox operation sound signal is collected by using a microphone line array, and the sound signal is subjected to noise reduction processing, and then the noise reduction sound signal corresponding to the frequency of interest is intercepted through spectrum analysis to obtain a dominant frequency band area;

[0007] The acoustic image acquisition method based on principal component analysis and near-field acoustic holography realizes the visualization of the acoustic field in the dominant frequency band area, and obtains the spatial tomographic slice of the acoustic field of the dominant frequency;

[0008] Inputting the acoustic field spatial tomographic slice into the trained CNN model to obtain a gearbox fault identification result;

[0009] According to the gearbox fault identification result, fault sound and image matching is performed to achieve accurate fault positioning.

[0010] The sampling frequency f of the microphone line array satisfies: f>2f max ; The spacing d of the elements of the microphone linear array satisfies: d<λ / 4; The measurement dynamic range D of the microphone linear array satisfies: Among them, f max is the maximum frequency of interest, λ is the wavelength corresponding to the maximum frequency of interest, and R is the reconstruction resolution, expressed as:

[0011] The method for acquiring an acoustic image based on principal component analysis and near-field acoustic holography realizes the visualization of the acoustic field in the dominant frequency band area and obtains a spatial tomographic slice of the acoustic field at the dominant frequency, specifically comprising:

[0012] Selecting k dominant frequencies with a fixed step size in the dominant frequency band, and arbitrarily selecting k1 dominant frequencies from the k dominant frequencies for combination, to obtain all k1 dominant frequency combinations;

[0013] The near-field measured holographic surface complex sound pressure of k1 dominant frequency combinations is obtained using the sound field reconstruction formula;

[0014] Calculating the reference cross-spectrum matrix and the transfer cross-spectrum matrix of the near-field measurement of the complex sound pressure of the holographic surface, and combining the principal component analysis to obtain the principal component matrix of the holographic surface;

[0015] Performing a sound field spatial transformation on the main hologram of the main component matrix of the holographic surface, using a two-dimensional Fourier transform to achieve a rapid transformation between the spatial domain and the beam domain, and reconstructing the sound field quantity on the source surface;

[0016] The transformation results of each main hologram are accumulated to obtain the holographic surface sound pressure map of each dominant frequency, forming a spatial tomographic slice of the sound field of the dominant frequency combination.

[0017] The trained CNN model is established in the following way:

[0018] Collect typical single fault sound signal test samples of gearbox, and randomly select 80% of the test samples as training set, and the remaining 20% ​​of the test samples as test set;

[0019] All test samples are subjected to noise reduction processing, and the noise-reduced acoustic signals are subjected to spectrum analysis to intercept the fault frequency band of interest, that is, to obtain the dominant frequency band area of ​​the test samples;

[0020] A sound image acquisition method based on principal component analysis and near-field acoustic holography realizes sound field visualization of the dominant frequency band of the test sample, and obtains a sound field spatial tomographic slice of the test sample;

[0021] Extract texture features and singular value features from the sound field space slices of the test sample, and combine the features of k1 sound and image slices in sequence to form a feature vector;

[0022] The feature vectors of the training set are input into the CNN model for training. After the training is completed, the test set is used for testing. If the test results meet the requirements, the CNN model training is completed; if the test results do not meet the requirements, the learner parameters are optimized again until the test results meet the requirements.

[0023] The singular value feature is composed of the first several larger values ​​of the singular value vector; the texture feature is to normalize each sample image, divide the sound image into grayscale levels according to the size of the grayscale level to obtain a grayscale image, and construct identification features of different grayscale levels based on the texture feature coefficient based on the grayscale co-occurrence matrix.

[0024] The fault sound and image matching is performed according to the fault identification result of the gearbox to achieve accurate fault location, specifically including:

[0025] Constructing a sound source plane coordinate system according to the moving distance of the guide rail and the distance between the elements of the microphone linear array;

[0026] Constructing equal sound pressure lines according to the sound source surface coordinate system and the reconstructed holographic surface sound pressure map;

[0027] The image and calibration template are read to learn the image calibration information, and the fault sound and image are matched according to the fault identification result of the gearbox. The sound source surface coordinate system is converted to the image coordinate, the isoacoustic pressure lines are converted to the image coordinate, and the colors and coordinates corresponding to the contour lines are overlapped on the image to complete the precise fault positioning.

[0028] In the gearbox fault location method based on acoustic image: k1 is determined based on the idea of ​​exhaustive method, and different grayscale and fault combinations are identified using CNN model, and the number of fault frequency combinations with the best accuracy is found in different grayscale and fault combinations, and the number of combinations is used as k1.

[0029] The technical solution adopted by the present invention to solve the technical problem is: to provide a gearbox fault locating device based on an acoustic image, comprising:

[0030] The acoustic signal collection and processing module is used to collect the gearbox operation acoustic signal using a microphone line array, and perform noise reduction on the acoustic signal, and then perform spectrum analysis and intercept the noise-reduced acoustic signal corresponding to the frequency of interest to obtain a dominant frequency band area;

[0031] A sound field space tomographic slice acquisition module is used to realize sound field visualization for the dominant frequency band area based on the principal component analysis and near-field acoustic holographic acoustic image acquisition method, and obtain the sound field space tomographic slice of the dominant frequency;

[0032] A fault identification module is used to input the acoustic field spatial tomographic slice into the trained CNN model to obtain a gearbox fault identification result;

[0033] The matching and positioning module is used to perform fault sound and image matching according to the fault identification result of the gearbox to achieve accurate fault positioning.

[0034] The technical solution adopted by the present invention to solve its technical problem is: to provide an electronic device, including a memory, a processor and a computer program stored in the memory and executable on the processor, and the processor implements the steps of the above-mentioned gearbox fault location method based on the acoustic image when executing the computer program.

[0035] The technical solution adopted by the present invention to solve its technical problem is: providing a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, the steps of the gearbox fault locating method based on the acoustic image are implemented.

[0036] Beneficial Effects

[0037] Due to the adoption of the above-mentioned technical scheme, the present invention has the following advantages and positive effects compared with the prior art: the present invention utilizes principal component analysis and near-field acoustic holography technology to obtain the sound field spatial tomographic slices of the dominant frequency of the acoustic signal. Since the fault information and redundancy degree contained in the fault sound pressure distribution at different positions are different, the characteristic vector of the combination of the sound field spatial tomographic slices is used to enhance the fault identification capability, thereby improving the fault location accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 It is a technical framework diagram of a gearbox fault location method based on an acoustic image in a first embodiment of the present invention;

[0039] Figure 2 It is a flow chart of the method for acquiring an acoustic image based on principal component analysis and near-field acoustic holography in the first embodiment of the present invention. DETAILED DESCRIPTION

[0040] The present invention will be further described below in conjunction with specific embodiments. It should be understood that these embodiments are only used to illustrate the present invention and are not intended to limit the scope of the present invention. In addition, it should be understood that after reading the content taught by the present invention, those skilled in the art can make various changes or modifications to the present invention, and these equivalent forms fall within the scope limited by the appended claims of the application equally.

[0041] The first embodiment of the present invention relates to a gearbox fault location method based on an acoustic image, such as Figure 1 As shown, the following steps are included:

[0042] Step 1: Use a microphone line array to collect the gearbox operation sound signal, and perform noise reduction on the sound signal, then perform spectrum analysis and intercept the noise-reduced sound signal corresponding to the frequency of interest to obtain a dominant frequency band.

[0043] In this step, when selecting a microphone array to collect the gearbox fault sound signal, the entire measurement grid is required to basically envelop the gearbox. To obtain a higher reconstruction accuracy, the requirements for the measurement equipment and the selection of measurement parameters are relatively stringent. To avoid aliasing errors caused by Fourier transform, the sampling frequency f>2f max , f max is the maximum analysis frequency of interest; within the range allowed by the measurement equipment and array aperture, try to make the array element spacing d < λ / 4, λ is the wavelength corresponding to the highest frequency of interest; when the holographic measurement distance d1 is the distance between the holographic measurement surface and the source surface, when it exceeds the minimum wavelength, the evanescent wave information is lost, and only the plane wave information is measured, so the holographic measurement distance is taken as (generally less than or equal to half of the minimum wavelength); in holographic reverse reconstruction, the evanescent wave information of the sound source needs to be used to improve the resolution of the sound source reconstruction. The measurement of the evanescent wave directly affects the reliability of the reverse reconstruction. Therefore, the dynamic range of the measurement system determines the effect of the reverse reconstruction. The measurement dynamic range The closer the microphone array is to the sound source, the higher the reconstruction resolution is. The minimum holographic resolution refers to the fastest fluctuating sound field that can be quantitatively measured in a certain direction in space. Since the evanescent wave decays with the measurement distance d1, and the effective evanescent wave that may be received is also limited by the dynamic capability D of the measurement system, the reconstruction resolution is When the measurement distance d<<λ and D is large, R≈27.27d / D; the selection of the holographic measurement array size is generally based on experience that the length L of the two coordinate axes of the holographic surface x ,L y At least it should be greater than the corresponding length I of the area of ​​interest x ,I y 1.5 times of.

[0044] In this step, when the acoustic signal is subjected to noise reduction processing, the noise reduction algorithm used may be filtering, wavelet analysis, wavelet packet analysis, EMD, blind source separation, etc. After the noise reduction processing is completed, the noise-reduced acoustic signal is subjected to spectrum analysis to intercept the fault frequency band of interest, that is, to obtain the dominant frequency band area.

[0045] Step 2, based on the principal component analysis and near-field acoustic holography acoustic image acquisition method, the sound field visualization is realized for the dominant frequency band area to obtain the sound field spatial tomographic slice of the dominant frequency. Figure 2 The figure is a schematic diagram of a method for obtaining an acoustic image based on principal component analysis and near-field acoustic holography, which specifically includes the following sub-steps:

[0046] Step 1), selecting k dominant frequencies with a fixed step size in the dominant frequency band, and arbitrarily selecting k1 dominant frequencies from the k dominant frequencies for combination, to obtain all k1 dominant frequency combinations;

[0047] Step 2), using the sound field reconstruction formula to obtain the near-field measured holographic surface complex sound pressure of k1 dominant frequency combinations, the specific implementation method is as follows:

[0048] For a closed vibrating body, located in a fluid medium with a density of ρ and a sound speed of c, its surface is denoted by S, the internal area surrounded by the vibrating body is D′, and the external area is E, S ∞ Surrounding the three-dimensional space E' of interest, n is the normal vector, r is a point in the three-dimensional space E', r s is a point on the surface S; It is a holographic surface S h The sound pressure data (frequency is ω), It is a holographic surface S s The sound pressure data, d = (z h -z s ) is a constant. The basic relationship between the surface sound field of the vibrating body and the external sound field is:

[0049]

[0050] For smooth surfaces, the solid angle coefficient C(r) is: The sound pressure in space r is Among them, r represents any point in the three-dimensional sound field space, is the complex sound pressure of the sound wave with angular frequency ω at the sound field r, is the sound field r of the sound wave with angular frequency ω on the surface of the vibrating body s The complex sound pressure at, k = ω / c is the wave number; Green's function formula G D for:

[0051]

[0052] In practical applications, whether it is holographic data measurement or reconstruction data calculation, the data processed is finite and discrete. Therefore, logical assumptions and approximations should be used to reasonably process the planar near-field acoustic holographic reconstruction formula in a finite and discrete manner, and the final sound field reconstruction formula becomes:

[0053]

[0054] Where: n1, n2 represent the grid sub-block numbers in the x and y directions of the reconstruction surface, l1, l2 represent the grid sub-block numbers in the x and y directions within the holographic aperture, and their value ranges are 0 to (N-1).

[0055] Step 3) Calculate the reference cross-spectrum matrix and the transfer cross-spectrum matrix of the near-field measurement of the holographic surface complex sound pressure, combine the principal component analysis to obtain the holographic surface principal component matrix, and define the holographic surface principal component matrix as follows:

[0056]

[0057] Among them, each column a of A i It is called the main hologram. The main hologram is in the form of The diagonal matrix D1 contains the J non-zero eigenvalues ​​d of the reference cross-spectral matrix i ; S1 contains the first J columns of S, and the singular value decomposition of the reference cross-spectral matrix is C RR Similar to the diagonal matrix D, S is a unitary matrix of the standard orthogonal matrix L×L; the transfer cross-spectrum matrix Among them, E[] represents the time average, P R ,P H are the complex sound pressure vectors of the reference surface and the holographic surface respectively.

[0058] It should be noted that the contribution ratio of the sound source corresponding to each main holographic sound pressure component to the sound field is determined by d i / (d1+d2+…+d J ), so the main hologram with a very small contribution can be ignored.

[0059] Step 4), the sound field space transformation is performed on the main hologram of the main component matrix of the holographic surface, and a two-dimensional Fourier transform is used to realize the rapid transformation between the spatial domain and the beam domain, and the sound field quantity on the source surface is reconstructed. The sound pressure at any position in the reconstructed space can be obtained by the following formula:

[0060]

[0061]

[0062]

[0063]

[0064] Where F represents the two-dimensional Fourier transform, F -1 represents the inverse two-dimensional Fourier transform; k x and k y are the spatial wave numbers in the x and y directions respectively; for The inverse of represents the propagation wave component; Represents the evanescent wave component.

[0065] Step 5), after all main holograms are transformed into near-field acoustic holograms, the accumulation of the transformation results of each main hologram is the total transformation result. Therefore, by accumulating the transformation results of each main hologram, the holographic surface sound pressure map of each dominant frequency can be obtained, forming a spatial tomographic slice of the sound field of the dominant frequency combination.

[0066] Step 3: Input the acoustic field spatial tomographic slice into the trained CNN model to obtain a gearbox fault identification result.

[0067] The trained CNN model in this step is obtained by:

[0068] Collect typical single fault sound signal test samples of the gearbox. The collection time is 10s. M test samples are collected, and 80% of the test samples are randomly selected as the training set, and the remaining 20% ​​of the test samples are used as the test set;

[0069] All test samples are subjected to noise reduction processing, and the noise-reduced acoustic signals are subjected to spectrum analysis to intercept the fault frequency band of interest, that is, to obtain the dominant frequency band area of ​​the test samples;

[0070] A sound image acquisition method based on principal component analysis and near-field acoustic holography realizes sound field visualization of the dominant frequency band of the test sample, and obtains a sound field spatial tomographic slice of the test sample;

[0071] The texture features and singular value features are extracted from the sound field space tomographic slices of the test sample. The k1 dominant frequencies correspond to k1 sound field space tomographic slices. The size of each tomographic slice is the same as the size of the sound source reconstruction surface (source image). The feature order of the k1 sound and image slices is merged together to form a feature vector.

[0072] Singular value decomposition is stable when describing an image. Singular values ​​as identification features mainly depend on larger singular values. Therefore, in this embodiment, the singular value features are composed of the first several larger values ​​of the singular value vector, that is, the first several larger values ​​of the singular value vector are selected and the remaining part is deleted; the texture feature is to normalize each sample image, divide the sound and image into grayscale levels according to the size of the grayscale to obtain a grayscale image, and then construct a group of identification feature quantities for different grayscale levels according to the texture feature coefficients based on the grayscale co-occurrence matrix, as shown in Table 1.

[0073] Table 1 Texture feature table

[0074]

[0075] The combined fault feature vector of the sound pressure map of the training set is input into the CNN model, and the convolution kernel is used to perform convolution operation on the feature map of the previous layer, and the output feature map is constructed using the activation function. The output of each layer is the convolution of multiple input features, and this cycle repeats. A CNN model learner is trained for each type of training set; after the training is completed, the test set is used for testing. If the test results meet the requirements, the CNN model training is completed; if the results do not meet the requirements, the learner parameters are optimized again until the test results meet the requirements.

[0076] Step 4, perform fault sound and image matching according to the fault identification result of the gearbox to achieve accurate fault positioning. This step is mainly implemented in the following three steps. The first step is to construct the X-axis and Y-axis coordinate system of the sound source surface (i.e., the gearbox scanning plane) through the moving distance of the guide rail and the distance of the array elements of the microphone linear array; the second step is to construct isoacoustic pressure lines according to the sound source surface coordinate system and the reconstructed holographic surface sound pressure map, which is equivalent to calculating the sound pressure distribution of the sound source surface under the world coordinate (actual object sound source distribution); the third step is to read the image and calibration template to learn the image calibration information, perform fault sound and image matching according to the fault identification result of the gearbox, convert the sound source surface coordinate system to the image coordinate, convert the isoacoustic pressure lines to the image coordinate, and overlap the colors and coordinates corresponding to the contour lines on the image to complete the accurate fault positioning. The widely used image calibration methods in this step are Tsai two-step method and Zhang calibration method. Commonly used calibration tools include OpenCV, Camera Calibration Toolbox in Matlab and calibration training in Labview. For example, calibration training in Labview provides five calibration methods, namely point distance calibration, point coordinate calibration, distortion mode (grid), camera mode (grid), and micro plane (grid).

[0077] It is worth mentioning that in this implementation, the highest accuracy of fault location effect of all combinations within the specified number k1 is used as the recognition rate when the number of combinations is k1, and based on the idea of ​​exhaustive method, the CNN model is used to identify different grayscale and fault combinations, and the number of fault frequency combinations with the best accuracy is found in different grayscale and fault combinations, and the number of combinations is determined as the specified number k1.

[0078] It is not difficult to find that the present invention uses principal component analysis and near-field acoustic holography technology to obtain the sound field space tomographic slices of the dominant frequency of the acoustic signal. Since the fault information and redundancy degree contained in the fault sound pressure distribution at different positions are different, the characteristic vector of the combination of the sound field space tomographic slices enhances the fault identification capability, thereby improving the fault location accuracy.

[0079] A second embodiment of the present invention relates to a gearbox fault location device based on an acoustic image, comprising:

[0080] The acoustic signal collection and processing module is used to collect the gearbox operation acoustic signal using a microphone line array, and perform noise reduction on the acoustic signal, and then perform spectrum analysis and intercept the noise-reduced acoustic signal corresponding to the frequency of interest to obtain a dominant frequency band area;

[0081] A sound field space tomographic slice acquisition module is used to realize sound field visualization for the dominant frequency band area based on the principal component analysis and near-field acoustic holographic acoustic image acquisition method, and obtain the sound field space tomographic slice of the dominant frequency;

[0082] A fault identification module is used to input the acoustic field spatial tomographic slice into the trained CNN model to obtain a gearbox fault identification result;

[0083] The matching and positioning module is used to perform fault sound and image matching according to the fault identification result of the gearbox to achieve accurate fault positioning.

[0084] The sampling frequency f of the microphone line array satisfies: f>2f max ; The spacing d of the elements of the microphone linear array satisfies: d<λ / 4; The measurement dynamic range D of the microphone linear array satisfies: Among them, f max is the maximum frequency of interest, λ is the wavelength corresponding to the maximum frequency of interest, and R is the reconstruction resolution, expressed as:

[0085] The acoustic field space tomographic slice acquisition module comprises:

[0086] A selection unit is used to select k dominant frequencies with a fixed step size in the dominant frequency band, and arbitrarily select k1 dominant frequencies from the k dominant frequencies for combination to obtain all k1 dominant frequency combinations;

[0087] A near-field measurement holographic surface complex sound pressure acquisition unit is used to obtain the near-field measurement holographic surface complex sound pressure of k1 dominant frequency combinations by using a sound field reconstruction formula;

[0088] A holographic surface principal component matrix acquisition unit is used to calculate the reference cross-spectrum matrix and the transfer cross-spectrum matrix of the near-field measurement of the holographic surface complex sound pressure, and obtain the holographic surface principal component matrix in combination with principal component analysis;

[0089] A reconstruction unit, used for performing a sound field spatial transformation on the main hologram of the main component matrix of the holographic surface, using a two-dimensional Fourier transform to realize a fast transformation between the spatial domain and the beam domain, and reconstructing the sound field quantity on the source surface;

[0090] The accumulation unit is used to accumulate the transformation results of each main hologram to obtain the holographic surface sound pressure map of each dominant frequency, forming a spatial tomographic slice of the sound field of the dominant frequency combination.

[0091] The trained CNN model is established in the following way:

[0092] Collect typical single fault sound signal test samples of gearbox, and randomly select 80% of the test samples as training set, and the remaining 20% ​​of the test samples as test set;

[0093] All test samples are subjected to noise reduction processing, and the noise-reduced acoustic signals are subjected to spectrum analysis to intercept the fault frequency band of interest, that is, to obtain the dominant frequency band area of ​​the test samples;

[0094] A sound image acquisition method based on principal component analysis and near-field acoustic holography realizes sound field visualization of the dominant frequency band of the test sample, and obtains a sound field spatial tomographic slice of the test sample;

[0095] Extract texture features and singular value features from the sound field space slices of the test sample, and combine the features of k1 sound and image slices in sequence to form a feature vector;

[0096] The feature vectors of the training set are input into the CNN model for training. After the training is completed, the test set is used for testing. If the test results meet the requirements, the CNN model training is completed; if the test results do not meet the requirements, the learner parameters are optimized again until the test results meet the requirements.

[0097] The singular value feature is composed of the first several larger values ​​of the singular value vector; the texture feature is to normalize each sample image, divide the sound image into grayscale levels according to the size of the grayscale level to obtain a grayscale image, and construct identification features of different grayscale levels based on the texture feature coefficient based on the grayscale co-occurrence matrix.

[0098] The matching and positioning module includes:

[0099] A sound source plane coordinate system construction unit, used to construct a sound source plane coordinate system according to the moving distance of the guide rail and the distance between the elements of the microphone linear array;

[0100] An isoacoustic pressure line construction unit, used to construct isoacoustic pressure lines according to the sound source surface coordinate system and the reconstructed holographic surface sound pressure map;

[0101] The positioning unit is used to read the image and calibration template to learn the image calibration information, perform fault sound and image matching according to the fault identification result of the gearbox, convert the sound source surface coordinate system to the image coordinate, convert the isoacoustic pressure line to the image coordinate, and overlap the color and coordinate corresponding to the contour line on the image respectively, so as to complete the accurate positioning of the fault.

[0102] The determination of k1 is based on the idea of ​​exhaustive method. The CNN model is used to identify different grayscale and fault combinations, and the number of fault frequency combinations with the best accuracy is found among different grayscale and fault combinations, and this number of combinations is used as k1.

[0103] A third embodiment of the present invention relates to an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the gearbox fault location method based on acoustic image of the first embodiment when executing the computer program.

[0104] A fourth embodiment of the present invention relates to a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of the gearbox fault location method based on acoustic image of the first embodiment are implemented.

[0105] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage and optical storage, etc.) that contain computer-usable program code.

[0106] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0107] These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product including an instruction method, which is implemented in the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0108] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

Claims

1. A method for locating gearbox faults based on acoustic images, characterized in that: The following steps are involved: The gearbox operation sound signal is collected by using a microphone line array, and the sound signal is subjected to noise reduction processing, and then the noise reduction sound signal corresponding to the frequency of interest is intercepted through spectrum analysis to obtain a dominant frequency band area; The acoustic image acquisition method based on principal component analysis and near-field acoustic holography realizes the visualization of the acoustic field in the dominant frequency band area, and obtains the spatial tomographic slice of the acoustic field of the dominant frequency; Inputting the acoustic field spatial tomographic slice into the trained CNN model to obtain a gearbox fault identification result; According to the gearbox fault identification result, fault sound and image matching is performed to achieve accurate fault positioning.

2. The method for locating a gearbox fault based on an acoustic image according to claim 1, characterized in that: The sampling frequency f of the microphone line array satisfies: f>2f max ; The spacing d of the elements of the microphone linear array satisfies: d<λ / 4; The measurement dynamic range D of the microphone linear array satisfies: Among them, f max is the maximum frequency of interest, λ is the wavelength corresponding to the maximum frequency of interest, and R is the reconstruction resolution, expressed as:

3. The method for locating a gearbox fault based on an acoustic image according to claim 1, characterized in that: The method for acquiring an acoustic image based on principal component analysis and near-field acoustic holography realizes the visualization of the acoustic field in the dominant frequency band area and obtains a spatial tomographic slice of the acoustic field at the dominant frequency, specifically comprising: Selecting k dominant frequencies with a fixed step size in the dominant frequency band, and arbitrarily selecting k1 dominant frequencies from the k dominant frequencies for combination, to obtain all k1 dominant frequency combinations; The near-field measured holographic surface complex sound pressure of k1 dominant frequency combinations is obtained using the sound field reconstruction formula; Calculating the reference cross-spectrum matrix and the transfer cross-spectrum matrix of the near-field measurement of the complex sound pressure of the holographic surface, and combining the principal component analysis to obtain the principal component matrix of the holographic surface; Performing a sound field spatial transformation on the main hologram of the main component matrix of the holographic surface, using a two-dimensional Fourier transform to achieve a rapid transformation between the spatial domain and the beam domain, and reconstructing the sound field quantity on the source surface; The transformation results of each main hologram are accumulated to obtain the holographic surface sound pressure map of each dominant frequency, forming a spatial tomographic slice of the sound field of the dominant frequency combination.

4. The method for locating a gearbox fault based on an acoustic image according to claim 1, characterized in that: The trained CNN model is established in the following way: Collect typical single fault sound signal test samples of gearbox, and randomly select 80% of the test samples as training set. The remaining 20% ​​test samples are used as the test set; All test samples are subjected to noise reduction processing, and the noise-reduced acoustic signals are subjected to spectrum analysis to intercept the fault frequency band of interest, that is, to obtain the dominant frequency band area of ​​the test samples; A sound image acquisition method based on principal component analysis and near-field acoustic holography realizes sound field visualization of the dominant frequency band of the test sample, and obtains a sound field spatial tomographic slice of the test sample; Extract texture features and singular value features from the sound field space slices of the test sample, and combine the features of k1 sound and image slices in sequence to form a feature vector; The feature vectors of the training set are input into the CNN model for training. After the training is completed, the test set is used for testing. If the test results meet the requirements, the CNN model training is completed. If the test results do not meet the requirements, the learner parameters are optimized again until the test results meet the requirements.

5. The method for locating a gearbox fault based on an acoustic image according to claim 4, characterized in that: The singular value feature is composed of the first several larger values ​​of the singular value vector; the texture feature is to normalize each sample image, divide the sound image into grayscale levels according to the size of the grayscale level to obtain a grayscale image, and construct identification features of different grayscale levels based on the texture feature coefficient based on the grayscale co-occurrence matrix.

6. The method for locating a gearbox fault based on an acoustic image according to claim 3, characterized in that: The fault sound and image matching is performed according to the fault identification result of the gearbox to achieve accurate fault location, specifically including: Constructing a sound source plane coordinate system according to the moving distance of the guide rail and the distance between the elements of the microphone linear array; Constructing equal sound pressure lines according to the sound source surface coordinate system and the reconstructed holographic surface sound pressure map; The image and calibration template are read to learn the image calibration information, and the fault sound and image are matched according to the fault identification result of the gearbox. The sound source surface coordinate system is converted to the image coordinate, the isoacoustic pressure lines are converted to the image coordinate, and the colors and coordinates corresponding to the contour lines are overlapped on the image to complete the precise fault positioning.

7. The method for locating a gearbox fault based on an acoustic image according to claim 3, characterized in that: The determination of k1 is based on the idea of ​​exhaustive method. The CNN model is used to identify different grayscale and fault combinations, and the number of fault frequency combinations with the best accuracy is found among different grayscale and fault combinations, and this number of combinations is used as k1.

8. A gearbox fault location device based on an acoustic image, characterized in that: include: The acoustic signal collection and processing module is used to collect the gearbox operation acoustic signal using a microphone line array, and perform noise reduction on the acoustic signal, and then perform spectrum analysis and intercept the noise-reduced acoustic signal corresponding to the frequency of interest. Get the dominant frequency band area; A sound field space tomographic slice acquisition module is used to realize sound field visualization for the dominant frequency band area based on the principal component analysis and near-field acoustic holographic acoustic image acquisition method, and obtain the sound field space tomographic slice of the dominant frequency; A fault identification module is used to input the acoustic field spatial tomographic slice into the trained CNN model to obtain a gearbox fault identification result; The matching and positioning module is used to perform fault sound and image matching according to the fault identification result of the gearbox to achieve accurate fault positioning.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the steps of the gearbox fault location method based on acoustic image as described in any one of claims 1-7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the gearbox fault location method based on acoustic image as claimed in any one of claims 1 to 7 are implemented.