Optimization method for fault location of transmission based on multi-layer convolutional sparse coding

Abstract

The present application relates to a kind of based on multilayer convolution sparse coding transmission device fault location optimization method, belong to mechanical fault diagnosis field.The method includes the following steps: according to the measured transmission device design suitable microphone array;Utilize microphone array to gather the acoustic signal of measured transmission device;According to the acoustic array signal of collection and imaging algorithm generates the fault location chart of measured transmission device;Construct the transmission device fault location optimization model based on multilayer convolution sparse coding;The fault location chart generated is as test set, input into the transmission device fault location optimization model trained, obtains transmission device fault location optimization result.The present application optimizes the problem that originally acoustic detection method is not accurate enough in transmission device fault location, realizes the transmission device fault location with higher accuracy.

Description

Technical Field

[0001] This invention belongs to the field of mechanical fault diagnosis and relates to an optimization method for fault location of transmission devices based on multi-layer convolutional sparse coding. Background Technology

[0002] Existing methods for fault diagnosis and location in rotating machinery transmission systems often utilize vibration sensors to collect operational data. However, this contact-based measurement method is unsuitable for high-temperature, high-pressure, and highly corrosive environments. Due to the limitations of vibration signal acquisition, many researchers have explored using other signal types, such as sound signals, for diagnostic analysis of rotating machinery transmission systems. Compared to vibration signals, sound signals offer the advantage of non-contact measurement, preventing relative movement between the sensor and the rotating machinery itself due to severe vibrations. Furthermore, sound signals can more easily identify early fault characteristics in rotating machinery transmission systems, demonstrating promising application prospects.

[0003] A microphone array is a group of sound collectors arranged in a specific way. Typical applications of microphone array technology include video conferencing, sound source localization, speech recognition, and human-computer interaction. Sound source localization based on microphone arrays is one of its main functions. Simply put, microphone array sound source localization uses a microphone array to determine the spatial location of a sound source. However, in actual working conditions, the sound signals generated by the rotating machinery transmission system and its surrounding environment produce a large amount of interference noise, resulting in a low signal-to-noise ratio of the acquired sound signal. Therefore, the accuracy of fault location results for rotating machinery transmission systems generated using imaging algorithms based on microphone array signals often fails to meet the requirements of industrial inspection. Although many studies are currently dedicated to improving signal preprocessing methods to filter out the influence of environmental noise and other factors from the acquired acoustic signals, these processes typically involve a large amount of complex calculations and offer limited improvement in the accuracy of the results.

[0004] In recent years, deep learning methods have made significant progress in the field of image processing. The Unet network, proposed in the 2015 ISBI competition, achieved good semantic segmentation results with only a small number of training samples, providing another approach to optimizing acoustic localization results of transmission device faults. However, compared to traditional signal processing methods, deep learning algorithms are more complex in their computation and training processes, and require large amounts of data that are difficult to collect. Summary of the Invention

[0005] In view of this, the purpose of this invention is to provide an optimization method for fault location of transmission devices based on multi-layer convolutional sparse coding. This method overcomes the limitations of traditional vibration detection in rotating machinery transmission systems, such as inconvenient installation, vibration along with the equipment, and unsuitability for high-temperature, high-pressure, and highly corrosive environments. It optimizes the vibration acoustic detection location results of traditional rotating machinery transmission systems to improve positioning accuracy, reduces the number of parameters in deep learning algorithms used in acoustic detection of traditional rotating machinery transmission systems, lowers computational costs, and provides a certain degree of interpretability.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A fault location optimization method for transmission devices based on multi-layer convolutional sparse coding, the method includes the following steps:

[0008] S1: Design a microphone array based on the transmission device under test. The microphone array should be able to cover the transmission device under test in space, and each microphone sensor should be able to collect the acoustic signal emitted by the transmission device under test. The microphone array should be perpendicular to the ground.

[0009] S2: Based on the designed microphone array, construct an acoustic signal acquisition system for traditional rotating mechanical transmission devices, and use the microphone array to acquire the acoustic signals of the transmission device under test;

[0010] S3: Design multiple fault locations based on the actual working conditions of the transmission device under test. For each fault location, generate a fault location map of the transmission device under test based on the collected acoustic array signals and imaging algorithms.

[0011] S4: Construct a fault location optimization model for transmission devices based on multi-layer convolutional sparse coding. Use the fault location maps of different fault locations produced in S3 as the training set and input them into the network model for training to obtain the trained fault location optimization model for transmission devices.

[0012] S5: Repeat S1, S2 and S3 for the actual transmission device being tested as experimental verification. Use the generated fault location map as a test set and input it into the trained transmission device fault location optimization model to obtain the transmission device fault location optimization result.

[0013] Optionally, in step S2, the acoustic signal acquisition and processing method is as follows:

[0014] S21: The attenuation ratio of a sound signal as it travels through the air to the microphone is related to the signal center frequency r. (f) The signal is directly related to the distance s between the reflecting surfaces, the sound absorption coefficient ξ of the reflecting surface, the signal attenuation ratio γ per unit distance in the air, and the number of reflections i. Specific parameters can be found in the *Acoustics Handbook*. The signal r collected by the microphone sensor is composed of the direct signal r0 and the reflected sound, expressed as:

[0015]

[0016] The direct signal r0, which contains information about the transmission device fault, can be obtained by subtracting the superimposed reflected sound signal from the acquired signal r.

[0017] Optionally, S3 specifically includes the following steps:

[0018] S31: Let the position (x, y) of each fault location in the sensor array be the location of the transmission device under test. The pixel value P(x, y) generated by the imaging algorithm is represented as:

[0019]

[0020] Among them, (x i ,y i ) and (x j ,y j The coordinates of the microphone sensors at the two endpoints forming a detection path are shown below; C represents the propagation speed of the acoustic signal, approximately 340 m / s in air; M represents the total number of microphone sensors; D represents the horizontal distance between the microphone array plane and the fault point of the tested transmission device; to avoid the complexity of near-field signals, the microphone array is arranged in the far-field position, and D satisfies the following condition:

[0021]

[0022] Where L represents the array length and λ represents the signal wavelength;

[0023] S32: After normalizing the pixel value of each point, the damage image of each damage sample is obtained.

[0024] P(x,y)=P(x,y) / max(P(x,y)).

[0025] Optionally, in S4, the samples used as the training set are fault location maps of the same transmission device under test at different speeds and loads. The training set is divided into 80% samples for training and 20% samples for verification.

[0026] Optionally, in step S4, the construction of the transmission device fault location optimization model based on multi-layer convolutional sparse coding specifically includes the following steps:

[0027] S41: Construct a damage contour prediction network model based on a U-shaped convolutional network and containing an encoding layer and a decoding layer. The encoding layer path is responsible for capturing detailed information of the input image, and the decoding layer path is responsible for localization. In order to keep the size and content information of the high-dimensional feature vector generated by the decoding layer from being lost, a skip connection is added between each downsampling block of the encoding layer and each upsampling block of the decoding layer.

[0028] S42: By replacing the convolutional layers in the U-shaped convolutional network with a multi-layer convolutional sparse coding (ML-CSC) module, and using the damage imaging map as the input image y of the damage contour prediction network model, it can be represented as:

[0029] y=D1Γ1,

[0030] Γ1=D2Γ2,

[0031]

[0032] Γ L-1 =D L Γ L .

[0033] Among them, Γ i D represents the result of the i-th convolutional layer, where i = 1, 2, ..., L, and L represents the layer number; i Represents a special dictionary; each D i Represents the convolution matrix W i The transpose of, i.e.:

[0034] D i =W i T

[0035] S43: Based on the base pursuit algorithm, construct an alternating hierarchical pursuit algorithm, which involves solving the following equations:

[0036]

[0037]

[0038]

[0039]

[0040] in, This represents the input to the alternating hierarchical tracking algorithm and has This represents the sparse solution of the i-th layer of the alternating hierarchical pursuit algorithm; λ denotes the square of the L2 norm; ||||1 denotes the sparse regularization constraint term; i Indicates control Γ i Regularization parameters for sparsity;

[0041] S44: The sparse solution is approximated using a form similar to the forward propagation of a convolutional neural network (CNN). The approximate solver of the alternating hierarchical tracking algorithm uses a multi-layer iterative soft thresholding algorithm to solve the sparse solution.

[0042]

[0043]

[0044]

[0045]

[0046] in, Indicates; μ i and θ i W represents the first and second parameters of the training, respectively. i W represents the convolution operation; i T This represents the deconvolution operation; k represents the number of expansions.

[0047] S45: Use the Dissloss function as the loss function of the network model, with the following expression:

[0048]

[0049] Among them, Y i P represents the label at the i-th pixel of the actual damage location map in the labeled image; i N represents the predicted probability that the i-th pixel of the fault location optimization map of the transmission device at the same layer in the same network model is defective; P This indicates the total number of pixels in the labeled and optimized image.

[0050] Optionally, in S5, the degree of optimization of the final result is determined based on the distance between the fault center point and the actual fault location of the unoptimized fault location map (as test input) and the optimized fault location map (as output).

[0051] The beneficial effects of this invention are as follows: This invention overcomes the limitations of traditional vibration detection methods for rotating machinery transmission systems, such as inconvenient installation, vibration along with the equipment, and unsuitability for high-temperature, high-pressure, and highly corrosive environments. It uses a microphone array to collect acoustic signals during operation, achieving acoustic fault location for traditional rotating machinery transmission devices. Furthermore, this method combines multi-layer convolutional sparse coding and a U-net network, using a fault location map of the transmission device generated by an imaging algorithm as input. This enables the extraction and accurate reconstruction of fault location features from the acoustic signals, resulting in an optimized fault location map with higher accuracy than the original. Simultaneously, this network model construction reduces the complexity of deep learning network models to some extent, decreases parameter computation, and provides a degree of interpretability.

[0052] Other advantages, objectives, and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination, or may be learned from practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description

[0053] To make the objectives, technical solutions, and advantages of the present invention clearer, the preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings, wherein:

[0054] Figure 1 This is a flowchart of the acoustic fault localization optimization method for transmission devices based on multi-layer convolutional sparse coding according to the present invention.

[0055] Figure 2 This is a schematic diagram of an acoustic testing platform for a transmission device.

[0056] Figure 3 A schematic diagram illustrating the construction of an optimization model for fault localization of a transmission device based on multi-layer convolutional sparse coding;

[0057] Figure 4 The specific shape of the microphone array;

[0058] Figure 5 Before and after comparison of the damage localization imaging map using the optimized network model for acoustic fault localization of the transmission device based on multi-layer convolutional sparse coding. Figure 5 (a) is the damage localization image obtained by the time-delay overlay algorithm. Figure 5 (b) is the optimized damage localization map. Detailed Implementation

[0059] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0060] The accompanying drawings are for illustrative purposes only and are schematic diagrams, not actual pictures. They should not be construed as limiting the invention. To better illustrate the embodiments of the invention, some parts in the drawings may be omitted, enlarged, or reduced, and do not represent the actual product dimensions. It is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.

[0061] In the accompanying drawings of the embodiments of the present invention, the same or similar reference numerals correspond to the same or similar components. In the description of the present invention, it should be understood that if terms such as "upper," "lower," "left," "right," "front," and "rear" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, they are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, the terms used to describe positional relationships in the drawings are only for illustrative purposes and should not be construed as limiting the present invention. For those skilled in the art, the specific meaning of the above terms can be understood according to the specific circumstances.

[0062] Please see Figures 1-3 A method for predicting composite material damage profiles based on multi-layer convolutional sparse coding and U-net includes the following steps:

[0063] S1: Design a suitable microphone array based on the transmission device under test. The design and installation requirements of the microphone array are that the entire array area can cover the transmission device under test in space and each microphone sensor can effectively collect the acoustic signal emitted by the transmission device under test. The microphone array is perpendicular to the ground.

[0064] In one specific embodiment, the transmission device under test is 1.4m long, with all transmission shafts averaging approximately 0.5m above the ground. The motor speed is 600rpm, and the transmission ratio is 1:1. The sequence of components is: motor, coupling, gears, and load, which is a magnetic powder brake. Considering the basic 60° field of view of the camera at the center of the microphone array and site constraints, the center of the microphone array is fixed at the same horizontal height as the average transmission shaft, i.e., 0.5m. Simultaneously, the sensor array is kept perpendicular to the ground and directly facing the transmission device under test, positioned laterally in the center of the transmission device, i.e., 0.7m from both ends of the transmission device in a parallel direction. The horizontal distance D between the sensor array and the transmission device under test is set to 1m. Verification shows that the image acquisition range of the camera at the center of the microphone array can completely cover the entire area of ​​the transmission device under test.

[0065] The specific configuration and shape of the microphone array are as follows: Figure 4 As shown, the entire array is a 420mm x 420mm square, with a 2-megapixel camera at the center and eight GRAS 40PH-10 microphones evenly distributed around it, with the closest pair of microphones spaced 130mm apart. All the microphones and cameras are fixed to a plastic plate, which is supported by a custom tripod to ensure it is perpendicular to the ground and facing the transmission device under test. The center of each microphone is 15mm from the edge of the plastic plate.

[0066] In one specific embodiment, the highest frequency of fault signals in the entire transmission device is a broken tooth fault. Considering that the motor speed is 600 rpm, the faulty gear has 60 teeth, and the transmission ratio is 1:1, the near-field and far-field distance discrimination formula is used:

[0067]

[0068] It can be seen that when D = 1m, the far-field condition is satisfied. Here, L represents the array length, and λ represents the signal wavelength.

[0069] S2: Based on the designed microphone array, construct a traditional rotating mechanical transmission device acoustic signal acquisition system and use the microphone array to acquire the acoustic signals of the transmission device under test.

[0070] In one specific embodiment, all eight microphones in the array are connected to an eight-channel oscilloscope via an ATA-4012 signal method amplifier. The oscilloscope, the camera, and the host computer are then connected to the host computer. The signal acquisition software in the host computer enables real-time synchronous acquisition of the eight-channel audio signals and the camera image signals.

[0071] S21: The attenuation ratio of a sound signal as it travels through the air to the microphone is related to the signal center frequency r. (f)The signal is directly related to the distance s between the reflector surfaces, the sound absorption coefficient ξ of the reflector surface, the signal attenuation ratio γ per unit distance in the air, and the number of reflections i. Specific parameters can be found in the "Acoustics Handbook". Therefore, the signal r collected by the microphone sensor is composed of the direct signal r0 and the reflected sound, which can be expressed by formula (1).

[0072]

[0073] The direct signal r0, which contains information about the transmission device fault, can be obtained by subtracting the superimposed reflected sound signal from the acquired signal r.

[0074] In one specific embodiment, the 8-channel audio signal was filtered using the method described above.

[0075] S3: Design multiple fault locations based on the actual working conditions of the transmission device under test. For each fault location, generate a fault location map of the transmission device under test based on the collected acoustic array signals and imaging algorithms.

[0076] S31: Assuming that the position of each fault location of the tested transmission device in the sensor array is (x, y), then the pixel value P(x, y) generated by the imaging algorithm is expressed as:

[0077]

[0078] Among them, (x i ,y i ) and (x j ,y j ) represent the coordinates of the microphone sensors at the two endpoints that form a detection path; C represents the propagation speed of the acoustic signal, which is approximately 340 m / s in air; M represents the total number of microphone sensors; and D represents the horizontal distance between the microphone array plane and the fault point of the transmission device under test.

[0079] In one specific embodiment, bearing inner and outer ring faults, bearing roller faults, gear tooth breakage faults, gear wear faults, and shaft misalignment faults, as well as corresponding healthy operating modes, were set at 300rpm, 480rpm, 600rpm, 720rpm, and 900rpm, respectively. These operating modes were conducted by placing the same microphone array in the same position under the same load to carry out experimental acoustic signal acquisition.

[0080] In one specific embodiment, in order to increase the quantity and diversity of training samples, white noise of different degrees was added to the experimentally acquired signals, and simulated signals at different positions were added to each signal according to the actual experimentally acquired signals and the fault mechanism model. Finally, 2400 sets of signals under different working conditions were obtained as training samples, of which 80 sets of signals for each speed and each fault mode were selected. 70 sets were randomly selected from these as the training set and 10 sets were selected as the test set.

[0081] S32: After normalizing the pixel value of each point, the damage image of each damage sample is obtained.

[0082] P(x,y)=P(x,y) / max(P(x,y)) (4)

[0083] In one specific embodiment, for the final damage imaging image, the area occupied by the device under test in the image is selected as the input of the transmission device fault location optimization model network based on multi-layer convolutional sparse coding, so as to improve the computational efficiency and accuracy of the model.

[0084] S4: Construct a fault location optimization model for transmission devices based on multi-layer convolutional sparse coding. Use the fault location maps of different fault locations produced in S3 as the training set and input them into the network model for training to obtain the trained fault location optimization model for transmission devices.

[0085] Furthermore, the construction of the transmission device fault location optimization model based on multi-layer convolutional sparse coding in S4 specifically includes the following steps:

[0086] S41: Construct a damage contour prediction network model based on a U-shaped convolutional network and containing an encoding layer and a decoding layer. The encoding layer path is responsible for capturing detailed information of the input image, and the decoding layer path is responsible for achieving accurate localization. To ensure that the size and content information of the high-dimensional feature vector generated by the decoding layer are not lost, a skip connection is added between each downsampling block of the encoding layer and each upsampling block of the decoding layer.

[0087] S42: By replacing the convolutional layers in the U-shaped convolutional network with a multi-layer convolutional sparse coding (ML-CSC) module, and using the damage imaging map as the input image y of the damage contour prediction network model, it can be represented as:

[0088]

[0089] Among them, Γ i D represents the result of the i-th convolution layer, where i = 1, 2, ..., L, and L represents the layer number; i Represents a special dictionary; each D i Represents the convolution matrix W i The transpose of, i.e.:

[0090] D i =W i T (6);

[0091] S43: Based on the base pursuit algorithm, construct an alternating hierarchical pursuit algorithm, which involves solving the following equations:

[0092]

[0093] in, This represents the input to the alternating hierarchical tracking algorithm and has This represents the sparse solution of the i-th layer of the alternating hierarchical pursuit algorithm; λ denotes the square of the L2 norm; ||||1 denotes the sparse regularization constraint term; i Indicates control Γ i Regularization parameters for sparsity;

[0094] S44: The sparse solution is approximated using a form similar to the forward propagation of a convolutional neural network (CNN). The approximate solver of the alternating hierarchical tracking algorithm uses a multi-layer iterative soft thresholding algorithm to solve the sparse solution.

[0095]

[0096] in, Indicates; μ i and θ i W represents the first and second parameters of the training, respectively. i W represents the convolution operation; i T This indicates a deconvolution operation; k represents the number of expansions.

[0097] In one specific embodiment, in order to reduce the computational load of the algorithm model while obtaining accurate results, L is set to 3 and k is set to 2.

[0098] S45: The Dissloss function is used as the loss function of the network model, and its expression is:

[0099]

[0100] Among them, Y i P represents the label at the i-th pixel of the actual damage location map in the labeled image; i N represents the predicted probability that the i-th pixel of the fault location optimization map of the transmission device at the same layer in the same network model is defective; P This indicates the total number of pixels in the labeled and optimized image.

[0101] In one specific embodiment, the constructed damage contour prediction network model based on a U-shaped convolutional network and containing encoding and decoding layers has 2400 damage images as input as a dataset, of which 2100 are used as the training set and 300 are used as the test set. The minimum batch size is 4, the training period is 20, and the ADAM strategy of gradient descent and learning rate ratio reduction is adopted.

[0102] S5: Repeat S1, S2 and S3 for the actual transmission device being tested as experimental verification. Use the generated fault location map as a test set and input it into the trained transmission device fault location optimization model to obtain the transmission device fault location optimization result.

[0103] In one specific embodiment, the laboratory test device, microphone array design, and sound signal acquisition system are completely consistent with the steps described above, with the motor speed set to 600 rpm and the fault type being a broken tooth fault in a 60-tooth gear. Figure 5 The damage localization imaging is compared before and after using an optimized network model for acoustic fault localization of the transmission device based on multi-layer convolutional sparse coding; the comparison is made by... Figure 5 (a) and Figure 5 (b) As can be seen, the damage localization image obtained by the delay-overlay algorithm is passed through a trained transmission device acoustic fault localization optimization network model based on multi-layer convolutional sparse coding. The resulting optimized damage localization image significantly removes other possible sources of interference and some false localization results in the imaging process, thus achieving more accurate identification and spatial localization of the broken tooth fault mode. In the figure, the elliptical area centered on white represents the damage imaging region; the whiter the color, the higher the probability of damage at that location.

[0104] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A fault location optimization method for transmission devices based on multi-layer convolutional sparse coding, characterized in that: The method includes the following steps: S1: Design a microphone array based on the transmission device under test. The microphone array should be able to cover the transmission device under test in space, and each microphone sensor should be able to collect the acoustic signal emitted by the transmission device under test. The microphone array should be perpendicular to the ground. S2: Based on the designed microphone array, construct an acoustic signal acquisition system for traditional rotating mechanical transmission devices, and use the microphone array to acquire the acoustic signals of the transmission device under test; S3: Design multiple fault locations based on the actual working conditions of the transmission device under test. For each fault location, generate a fault location map of the transmission device under test based on the collected acoustic array signals and imaging algorithms. S4: Construct a fault location optimization model for transmission devices based on multi-layer convolutional sparse coding. Use the fault location maps of different fault locations generated in S3 as the training set and input them into the network model for training to obtain the trained fault location optimization model for transmission devices. The construction of the transmission device fault location optimization model based on multi-layer convolutional sparse coding specifically includes the following steps: S41: Construct a damage contour prediction network model based on a U-shaped convolutional network and containing an encoding layer and a decoding layer. The encoding layer path is responsible for capturing detailed information of the input image, and the decoding layer path is responsible for localization. In order to keep the size and content information of the high-dimensional feature vector generated by the decoding layer from being lost, a skip connection is added between each downsampling block of the encoding layer and each upsampling block of the decoding layer. S42: A multi-layer convolutional sparse coding (ML-CSC) module is used to replace the convolutional layers in the U-shaped convolutional network, and the damage imaging map is used as the input image for the damage contour prediction network model. y Then it is represented as: in, Indicates the first i Layer convolution results, , L Indicates the number of layers; Represents a special dictionary; each Represents the convolution matrix The transpose of, i.e.: S43: Based on the base pursuit algorithm, construct an alternating hierarchical pursuit algorithm, which involves solving the following equations: in, This represents the input to the alternating hierarchical tracking algorithm and has ; The alternating hierarchical tracking algorithm represents the first... i Layer sparse solution; Represents the square of the L2 norm; This represents a sparse regularization constraint term; Indicates control Regularization parameters for sparsity; S44: The sparse solution is approximated using a form similar to the forward propagation of a convolutional neural network (CNN). The approximate solver of the alternating hierarchical tracking algorithm uses a multi-layer iterative soft thresholding algorithm to solve the sparse solution. in, and These represent the first and second parameters of the training process, respectively. This represents the convolution operation; This represents the deconvolution operation; k Represents the number of expansions; S45: Use the Diceloss function as the loss function of the network model, with the following expression: in, The image showing the actual damage location as a labeled image. A label of 1 pixel; This represents the first fault location optimization graph of the transmission device at the same layer in the same network model. The predicted probability of a defective pixel; This represents the total number of pixels in the labeled and optimized image. S5: Repeat S1, S2 and S3 for the actual transmission device being tested as experimental verification. Use the generated fault location map as a test set and input it into the trained transmission device fault location optimization model to obtain the transmission device fault location optimization result.

2. The transmission device fault location optimization method based on multi-layer convolutional sparse coding according to claim 1, characterized in that: In S2, the acoustic signal acquisition and processing method is as follows: S21: The attenuation ratio of a sound signal as it travels through the air to the microphone is related to the signal's center frequency. , the distance s between reflecting surfaces, the sound absorption coefficient ξ of the reflecting surface, the signal attenuation ratio γ per unit distance in air, and the number of reflections. i Directly related; signals collected by the microphone sensor From direct signal The superposition of reflected and superimposed signals is represented as follows: Among them, the collected signals Subtracting the superimposed reflected sound signal yields the direct signal containing transmission device fault information. .

3. The fault location optimization method for transmission devices based on multi-layer convolutional sparse coding according to claim 2, characterized in that: S3 specifically includes the following steps: S31: Let the position of each fault location of the tested transmission device in the sensor array be... The image pixel values ​​generated by its imaging algorithm Represented as: in, and These represent the coordinates of the microphone sensors at the two endpoints that form a detection path; This indicates the speed of acoustic signal propagation, which is approximately 340 m / s in air; M Indicates the total number of microphone sensors; D This indicates the horizontal distance between the microphone array plane and the fault point of the tested transmission device; to avoid the complexity of near-field signals, the microphone array is positioned in the far-field location. D The following conditions must be met: Where L represents the array length. Indicates the signal wavelength; S32: After normalizing the pixel value of each point, the damage image of each damage sample is obtained. 。 4. The transmission device fault location optimization method based on multi-layer convolutional sparse coding according to claim 3, characterized in that: In S4, the samples used as the training set are fault location maps of the same transmission device under different speeds and loads at different locations. The training set is divided into 80% of the samples for training and 20% of the samples for verification.

5. The transmission device fault location optimization method based on multi-layer convolutional sparse coding according to claim 1, characterized in that: In S5, the degree of optimization of the final result is determined based on the distance between the fault center point and the actual fault location of the unoptimized fault location map (as test input) and the optimized fault location map (as output).

Images