Vehicle component image fault recognition method and system based on convolutional neural network

Abstract

The present invention belongs to the field of fault technology, and discloses a vehicle component image fault recognition method and system based on convolutional neural network, the method comprising: obtaining a vibration signal of a vehicle bearing within a preset monitoring period, segmenting the vibration signal, and obtaining a plurality of vibration signal segments; performing modal decomposition on each vibration signal segment to obtain a vibration component set; calculating the similarity coefficient between each vibration component and the vibration signal segment in each vibration component set, taking the vibration component with a similarity coefficient greater than a preset coefficient as the target vibration component of the corresponding vibration signal segment, and obtaining at least one target vibration component of each vibration signal segment; constructing a grayscale image of the vibration signal based on at least one target vibration component; inputting the grayscale image of the vibration signal into a fault recognition model constructed based on a convolutional neural network algorithm to obtain a fault recognition result. The present invention reduces the complexity of vibration signal processing and improves the accuracy of fault recognition.

Description

Technical Field

[0001] The present invention belongs to the field of fault technology, and in particular relates to a vehicle component image fault recognition method and system based on convolutional neural network. Background Art

[0002] With the rapid development of the new energy vehicle industry, the number of new energy vehicles in use is increasing, which naturally makes the safety performance of new energy vehicles the focus of public and industry attention.

[0003] The front and rear wheel bearings of a car are one of the important components of the car's driving system. The front and rear wheel bearings mainly play the role of supporting weight, reducing friction, determining wheel position, bearing torque, and transmitting power. When the bearings are damaged or worn, it may cause abnormal noise when the vehicle is driving, steering difficulties, and even serious traffic accidents. Therefore, bearing fault detection and maintenance are crucial to ensure the safe operation of the vehicle.

[0004] In the prior art, a vibration analysis method is often used to collect the vibration signal of the bearing, and whether the bearing has defects is determined based on the vibration signal to achieve bearing fault detection.

[0005] However, the vibration analysis method has at least the following problems in practical application: since the collected vibration signal contains noise and has non-stationary and nonlinear characteristics, the bearing fault recognition rate is low. Summary of the invention

[0006] The purpose of the present invention is to provide a vehicle component image fault recognition method and system based on convolutional neural network, so as to solve the problem that the bearing fault recognition rate is low because the collected vibration signal contains noise and has non-stationary and nonlinear characteristics.

[0007] In order to achieve the above object, the present invention adopts the following technical solutions:

[0008] In a first aspect, the present invention provides a vehicle component image fault recognition method based on a convolutional neural network, the method comprising:

[0009] Acquire the vibration signal of the vehicle bearing within a preset monitoring period, segment the vibration signal, and obtain a plurality of vibration signal segments; wherein the end time of the preset monitoring period is the current time;

[0010] Performing modal decomposition on each vibration signal segment to obtain a vibration component set of each vibration signal segment, wherein the vibration component set includes multiple vibration components;

[0011] Calculating a similarity coefficient between each vibration component in each vibration component set and the corresponding vibration signal segment, taking a vibration component having a similarity coefficient greater than a preset coefficient as a target vibration component of the corresponding vibration signal segment, and obtaining at least one target vibration component of each vibration signal segment;

[0012] constructing a grayscale image of the vibration signal based on at least one target vibration component of each vibration signal segment;

[0013] The grayscale image of the vibration signal is input into the fault recognition model built based on the convolutional neural network algorithm for fault recognition to obtain the fault recognition result.

[0014] Preferably, each vibration signal segment has a sub-monitoring period in a preset monitoring period, and each sub-monitoring period has a plurality of sampling points;

[0015] Calculating the similarity coefficient between each vibration component in each vibration component set and the corresponding vibration signal segment, including:

[0016] For each vibration signal segment, obtaining the signal value of the vibration signal segment at each sampling point in the corresponding sub-monitoring period;

[0017] Calculate the distance between each vibration component and the signal values ​​of the corresponding multiple sampling points and sum them up to obtain the total similarity distance;

[0018] The similarity coefficient between each vibration component and the vibration signal segment is determined according to the similarity total distance.

[0019] Preferably, the function expression of the similarity coefficient between each vibration component and the vibration signal segment is:

[0020] ;

[0021] In the formula, is the mth vibration component The similarity coefficient with the vibration signal segment, is the total similarity distance, is the signal value of the nth sampling point in the vibration signal segment, and N is the total number of sampling points.

[0022] Preferably, the expression of the similarity total distance is:

[0023] .

[0024] Preferably, the fault identification model comprises a convolutional layer, a denoising unit, a fully connected layer and an output layer connected in sequence, and the denoising unit comprises an attention module and a filtering correction module.

[0025] Preferably, the convolution layer is used to extract features from the input grayscale image to obtain a multi-channel feature map;

[0026] The denoising unit is used to perform adaptive denoising on the feature map of each channel in the multi-channel feature map to obtain a denoised multi-channel feature map;

[0027] The fully connected layer is used to fully connect the denoised multi-channel feature map to obtain a classification result;

[0028] The output layer is used to output the classification results.

[0029] Preferably, the noise reduction unit is specifically used for:

[0030] Use multi-channel feature maps as input features of the denoising unit;

[0031] The attention module of the denoising unit maps the input features to obtain a global mapping value of each channel; obtains a first connection output value of each channel according to the global mapping value of each channel and a preset first mapping connection weight; obtains a second connection output value of each channel according to the first connection output value of each channel and a preset second mapping connection weight;

[0032] The attention module of the denoising unit scales the second connection output value of each channel to obtain a scaling weight of each channel;

[0033] The filtering correction module of the noise reduction unit averages the global mapping values ​​of all channels to obtain a global mapping mean; calculates the adaptive filtering threshold of each channel according to the global mapping mean and the scaling weight of each channel; and performs filtering and noise reduction on the global mapping value of each channel based on the adaptive filtering threshold of each channel to obtain a multi-channel feature map after noise reduction.

[0034] Preferably, the method further comprises: training the fault identification model, comprising:

[0035] Acquire a vibration data sample, and expand the vibration data sample to obtain a plurality of expanded samples;

[0036] Encoding a plurality of extended samples to obtain a plurality of vibration image samples;

[0037] Calculate the grayscale distribution feature matrix of a number of vibration image samples, and construct a training set with the grayscale distribution feature matrix of the number of vibration image samples;

[0038] The fault identification model is trained based on the training set until the fault identification model converges to obtain a trained fault identification model.

[0039] Preferably, the vibration data sample is expanded to obtain a plurality of expanded samples, including:

[0040] Based on a preset sampling interval and sampling length, the vibration data samples are overlapped and sampled to obtain a plurality of sampled samples; wherein the sampling interval is the interval between the starting points of two adjacent sampling operations;

[0041] The vibration data samples are divided into a number of sub-samples according to the sampling length, and the sub-samples and the sampling samples are used as a plurality of extended samples.

[0042] In a second aspect, the present invention provides a vehicle component image fault recognition system based on a convolutional neural network, which is used to implement the above-mentioned vehicle component image fault recognition method based on a convolutional neural network, and the system comprises:

[0043] A signal acquisition module is used to acquire the vibration signal of the vehicle bearing within a preset monitoring period, and segment the vibration signal to obtain a plurality of vibration signal segments; wherein the end time of the preset monitoring period is the current time;

[0044] A signal decomposition module, used for performing modal decomposition on each vibration signal segment to obtain a vibration component set of each vibration signal segment, wherein the vibration component set includes multiple vibration components;

[0045] A similarity calculation module is used to calculate a similarity coefficient between each vibration component in each vibration component set and the corresponding vibration signal segment, and take the vibration component with a similarity coefficient greater than a preset coefficient as the target vibration component of the corresponding vibration signal segment, to obtain at least one target vibration component of each vibration signal segment;

[0046] An image construction module, configured to construct a grayscale image of the vibration signal based on at least one target vibration component of each vibration signal segment;

[0047] The fault identification module is used to input the grayscale image of the vibration signal into a fault identification model built based on the convolutional neural network algorithm to perform fault identification and obtain a fault identification result.

[0048] In a third aspect, the present invention provides 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 above-mentioned vehicle component image fault recognition method based on convolutional neural network when executing the computer program.

[0049] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-mentioned vehicle component image fault recognition method based on convolutional neural network.

[0050] Beneficial effects:

[0051] 1. The present invention divides the vibration signal into multiple segments, performs modal decomposition on each segment, and obtains multiple vibration components, which can reduce the influence of noise and the complexity of vibration signal processing. The multiple vibration components have good stability and linearity characteristics;

[0052] 2. The present invention calculates the similarity coefficient between each vibration component in each vibration component set and the corresponding vibration signal segment, and uses the similarity coefficient to screen the target vibration component. This method can enable the constructed grayscale image to retain the fault characteristics in the original vibration signal and avoid noise interference, which is beneficial to improving the accuracy of the fault identification model in identifying bearing faults. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] The accompanying drawings are used to provide a further understanding of the embodiments of the present invention and constitute a part of the specification. Together with the following specific implementations, they are used to explain the embodiments of the present invention, but do not constitute a limitation on the embodiments of the present invention. In the accompanying drawings:

[0054] Figure 1 It is a flow chart of a vehicle component image fault recognition method based on a convolutional neural network provided by an embodiment of the present invention;

[0055] Figure 2 It is a block diagram of a vehicle component image fault recognition system based on a convolutional neural network provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0056] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the present invention will be briefly introduced below in combination with the drawings and the description of the embodiments or the prior art. Obviously, the following description of the structure of the drawings is only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. It should be noted that the description of these embodiments is used to help understand the present invention, but does not constitute a limitation of the present invention.

[0057] Embodiment 1

[0058] Figure 1 FIG. 1 is a flow chart of a vehicle component image fault recognition method based on a convolutional neural network provided by an embodiment of the present invention. Figure 1 As shown, this embodiment provides a vehicle component image fault recognition method based on a convolutional neural network, the method comprising:

[0059] Step S10: Obtain the vibration signal of the vehicle bearing within a preset monitoring period, segment the vibration signal, and obtain a number of vibration signal segments; wherein the end time of the preset monitoring period is the current time, that is, with the current time as a reference, extract the historical data of the vibration signal within a period of time (the preset monitoring period).

[0060] In this embodiment, an accelerometer, a velocity sensor and a position sensor can be used to monitor the vibration signal of the vehicle bearing in real time, wherein the accelerometer can measure the high-frequency component of the vehicle bearing vibration, the velocity sensor can measure the medium-frequency component of the vehicle bearing vibration, and the position sensor can measure the low-frequency component of the vehicle bearing; in the actual monitoring process, the above sensors can be directly installed on the bearing seat or on the structure near the bearing to collect vibration signals; three types of sensors can be used at the same time to collect vibration signals of different frequencies, and then the signals of the three frequencies are synthesized to obtain the final vibration signal; or the signals of the three frequencies can be retained and the signals of the three frequencies can be analyzed separately.

[0061] Step S20: performing modal decomposition on each vibration signal segment to obtain a vibration component set of each vibration signal segment, wherein the vibration component set includes multiple vibration components;

[0062] In this embodiment, a VMD (Variational Mode Decomposition) algorithm is used to perform modal decomposition on each vibration signal segment. The VMD algorithm decomposes a complex multi-component signal into multiple intrinsic mode functions. One intrinsic mode function can output one vibration component. In this embodiment, the number of intrinsic mode functions is 10 to 15. At this time, each vibration signal segment can obtain 10 to 15 vibration components.

[0063] This embodiment divides the vibration signal into multiple segments, performs modal decomposition on each segment, and obtains multiple vibration components, which can reduce the influence of noise and the complexity of vibration signal processing. The multiple vibration components have good stability and linearity characteristics.

[0064] Step S30: Calculate the similarity coefficient between each vibration component in each vibration component set and the corresponding vibration signal segment, and use the vibration component with a similarity coefficient greater than a preset coefficient as the target vibration component of the corresponding vibration signal segment to obtain at least one target vibration component of each vibration signal segment.

[0065] In this embodiment, each vibration signal segment has one sub-monitoring period in the preset monitoring period, and each sub-monitoring period has multiple sampling points; therefore, calculating the similarity coefficient between each vibration component in each vibration component set and the corresponding vibration signal segment includes:

[0066] Step S301: for each vibration signal segment, obtain the signal value of the vibration signal segment at each sampling point in the corresponding sub-monitoring period.

[0067] Step S302: Calculate the distance between each vibration component and the signal values ​​of the corresponding multiple sampling points and sum them up to obtain a similar total distance; wherein the expression of the similar total distance is:

[0068] (1);

[0069] In formula (1), is the total similarity distance, is the signal value of the nth sampling point in the vibration signal segment, N is the total number of sampling points, is the mth vibration component; among them, represents the first characteristic distance, Represents the second characteristic distance.

[0070] In this embodiment, assuming that the length of the vibration signal segment is L, including N sampling points, each sampling point including V eigenvalues, then the vibration signal segment has N*V eigenvalues;

[0071] The vibration component is also a vector of length L, and also corresponds to N sampling points. Each sampling point contains W eigenvalues, and a vibration component has N*W eigenvalues.

[0072] but, , ;

[0073] in, is the vth eigenvalue of the nth sampling point in the vibration signal segment, is the wth eigenvalue of the nth sampling point corresponding to the mth vibration component.

[0074] Step S303: determining the similarity coefficient between each vibration component and the vibration signal segment according to the total similarity distance; wherein the function expression of the similarity coefficient between each vibration component and the vibration signal segment is:

[0075] (2);

[0076] In formula (2), is the mth vibration component The similarity coefficient with the vibration signal segment, is the total similarity distance, is the signal value of the nth sampling point in the vibration signal segment, and N is the total number of sampling points.

[0077] In this embodiment, by calculating the similarity coefficient between each vibration component in each vibration component set and the corresponding vibration signal segment, the target vibration component is screened using the similarity coefficient. This method enables the grayscale image constructed in step S40 to retain the fault characteristics in the original vibration signal and avoid noise interference.

[0078] Step S40: constructing a grayscale image of the vibration signal based on at least one target vibration component of each vibration signal segment.

[0079] In this embodiment, at least one target vibration component of each vibration signal segment is first normalized, and a grayscale image is constructed using the normalized target vibration component.

[0080] Step S50: inputting the grayscale image of the vibration signal into a fault recognition model constructed based on a convolutional neural network algorithm to perform fault recognition and obtain a fault recognition result.

[0081] In this embodiment, since the grayscale image constructed in step S40 retains the fault characteristics of the original vibration signal and eliminates the interference of noise, it can help improve the accuracy of fault identification and avoid missing some faults of vehicle bearings.

[0082] As a further optimization of this embodiment, the fault identification model includes a convolutional layer, a denoising unit, a fully connected layer and an output layer connected in sequence, and the denoising unit includes an attention module and a filtering correction module.

[0083] The convolution layer is used to extract features from the input grayscale image to obtain a multi-channel feature map. Before the grayscale image is input into the convolution layer, the grayscale distribution feature matrix of the grayscale image is calculated, and the grayscale distribution feature matrix of the grayscale image is input into the convolution layer for feature extraction.

[0084] The denoising unit is used to perform adaptive denoising on the feature map of each channel in the multi-channel feature map to obtain a denoised multi-channel feature map; the denoising unit can filter out unimportant features and retain important features, thereby improving the efficiency of fault identification and further improving the accuracy;

[0085] The fully connected layer is used to fully connect the denoised multi-channel feature map to obtain a classification result;

[0086] The output layer is used to output the classification results.

[0087] As a further optimization of this embodiment, the noise reduction unit is specifically used for:

[0088] A multi-channel feature map is used as an input feature of a denoising unit; an attention module of the denoising unit maps the input feature to obtain a global mapping value of each channel; a first connection output value of each channel is obtained according to the global mapping value of each channel and a preset first mapping connection weight; a second connection output value of each channel is obtained according to the first connection output value of each channel and a preset second mapping connection weight.

[0089] Among them, the function expression of the global mapping value of each channel is:

[0090] (3);

[0091] In formula (3), is the global mapping value of the i-th channel, is the input feature of the i-th channel, is the height of the input feature of the i-th channel, is the width of the input feature of the i-th channel, for The coordinate points in .

[0092] The function expression of the first connection output value of each channel is:

[0093] (4);

[0094] In formula (4), is the first mapping connection weight, is the first connection output value of the i-th channel, is the ReLU activation function.

[0095] Among them, the function expression of the second connection output value of each channel is:

[0096] (5);

[0097] In formula (5), is the connection weight for the second mapping, The second connection output value of the i-th channel.

[0098] The attention module of the denoising unit scales the second connection output value of each channel to obtain the scaling weight of each channel; then the function expression of the scaling weight of each channel is:

[0099] (6);

[0100] In formula (6), is the scaling weight of the ith channel, is the Sigmod activation function.

[0101] The filtering correction module of the noise reduction unit averages the global mapping values ​​of all channels to obtain a global mapping mean; calculates the adaptive filtering threshold of each channel according to the global mapping mean and the scaling weight of each channel; and performs filtering and noise reduction on the global mapping value of each channel based on the adaptive filtering threshold of each channel to obtain a multi-channel feature map after noise reduction.

[0102] In this embodiment, for any channel, the global mapping value lower than or equal to the adaptive filtering threshold is set to 0, thereby filtering out unimportant features; the global mapping value greater than the adaptive filtering threshold is retained, so that important features can be retained.

[0103] In this embodiment, an adaptive filtering threshold is used to filter out unimportant features in the global mapping value of each channel, and then the filtered global mapping value of each channel is added to the feature map of each channel to obtain a denoised multi-channel feature map.

[0104] Among them, the expression of the unimportant feature function in the global mapping value of each channel is filtered out by using the adaptive filtering threshold:

[0105] (7);

[0106] In formula (7), is the global mapping value after filtering of the i-th channel, is the global mapping mean.

[0107] As a further optimization of this embodiment, the method further includes: training the fault identification model, including:

[0108] Step a10: Acquire vibration data samples, and expand the vibration data samples to obtain multiple extended samples; wherein the vibration data samples contain labels of known faults, that is, use accelerometers, velocity sensors, and position sensors to collect vehicle bearings with multiple known faults to obtain vibration data samples.

[0109] Since the number of vibration data samples collected from bearings with known faults is small, it cannot meet the subsequent training requirements; therefore, it is necessary to expand the vibration data samples to increase the number of samples; specifically, the vibration data samples are expanded to obtain multiple extended samples, including:

[0110] Step a101: performing overlapping sampling on the vibration data samples based on a preset sampling interval and sampling length to obtain a plurality of sampling samples; wherein the sampling interval is the interval between the starting points of two adjacent sampling operations;

[0111] Step a102: Divide the vibration data sample into a number of sub-samples according to the sampling length, and use the sub-samples and the sampling samples as a plurality of extended samples.

[0112] In this embodiment, the data set is enhanced by overlapping sampling of vibration data samples to enhance the generalization performance of the network.

[0113] Step a20: Encode multiple extended samples to obtain several vibration image samples; in this embodiment, the extended samples must be converted into a two-dimensional image. First, the one-dimensional signal must be reconstructed into a two-dimensional matrix, and then each sampling point in the obtained matrix must be normalized and converted into a grayscale pixel value and encoded into an image to obtain the vibration image samples.

[0114] Step a30: Calculate the grayscale distribution feature matrix of several vibration image samples, and construct a training set with the grayscale distribution feature matrix of several vibration image samples; in this embodiment, the histogram distribution statistics of the vibration image samples can reflect the amplitude characteristic distribution of the signal. The more stable the signal, the more uniform and concentrated its histogram distribution is, and the opposite is true for signals with large fluctuations. After feature conversion, the grayscale distribution feature matrix of the vibration image samples can be obtained.

[0115] Step a40: Train the fault identification model based on the training set until the fault identification model converges to obtain a trained fault identification model.

[0116] The present invention divides the vibration signal into multiple segments, performs modal decomposition on each segment, and obtains multiple vibration components, which can reduce the influence of noise and reduce the complexity of vibration signal processing. The multiple vibration components have good stability and linearity characteristics; and by calculating the similarity coefficient between each vibration component in each vibration component set and the corresponding vibration signal segment, the similarity coefficient is used to screen the target vibration component. This method can enable the constructed grayscale image to retain the fault characteristics in the original vibration signal and avoid noise interference, so as to improve the accuracy of the fault identification model in identifying bearing faults.

[0117] Embodiment 2

[0118] Figure 2 FIG. 1 is a block diagram of a vehicle component image fault recognition system based on a convolutional neural network provided by an embodiment of the present invention. Figure 2 As shown, this embodiment provides a vehicle component image fault recognition system based on a convolutional neural network, which is used to implement the vehicle component image fault recognition method based on a convolutional neural network in Example 1. The system includes:

[0119] A signal acquisition module is used to acquire the vibration signal of the vehicle bearing within a preset monitoring period, and segment the vibration signal to obtain a plurality of vibration signal segments; wherein the end time of the preset monitoring period is the current time;

[0120] A signal decomposition module, used for performing modal decomposition on each vibration signal segment to obtain a vibration component set of each vibration signal segment, wherein the vibration component set includes multiple vibration components;

[0121] A similarity calculation module is used to calculate a similarity coefficient between each vibration component in each vibration component set and the corresponding vibration signal segment, and take the vibration component with a similarity coefficient greater than a preset coefficient as the target vibration component of the corresponding vibration signal segment, to obtain at least one target vibration component of each vibration signal segment;

[0122] An image construction module, configured to construct a grayscale image of the vibration signal based on at least one target vibration component of each vibration signal segment;

[0123] The fault identification module is used to input the grayscale image of the vibration signal into a fault identification model built based on the convolutional neural network algorithm to perform fault identification and obtain a fault identification result.

[0124] This embodiment also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the vehicle component image fault recognition method based on convolutional neural network in Embodiment 1 is implemented.

[0125] This embodiment also provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the vehicle component image fault recognition method based on a convolutional neural network in the first embodiment is implemented.

[0126] The present invention divides the vibration signal into multiple segments, performs modal decomposition on each segment, and obtains multiple vibration components, which can reduce the influence of noise and reduce the complexity of vibration signal processing. The multiple vibration components have good stability and linearity characteristics; and by calculating the similarity coefficient between each vibration component in each vibration component set and the corresponding vibration signal segment, the similarity coefficient is used to screen the target vibration component. This method can enable the constructed grayscale image to retain the fault characteristics in the original vibration signal and avoid noise interference, so as to improve the accuracy of the fault identification model in identifying bearing faults.

[0127] 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 combining 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, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0128] 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 system that specifies the functions of a box or multiple boxes.

[0129] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included within the scope of the claims of the present application.

Claims

1. A vehicle component image fault recognition method based on convolutional neural network, characterized in that: The method comprises: Acquire the vibration signal of the vehicle bearing within a preset monitoring period, segment the vibration signal, and obtain a plurality of vibration signal segments; wherein the end time of the preset monitoring period is the current time; Performing modal decomposition on each vibration signal segment to obtain a vibration component set of each vibration signal segment, wherein the vibration component set includes multiple vibration components; Calculating a similarity coefficient between each vibration component in each vibration component set and the corresponding vibration signal segment, taking a vibration component having a similarity coefficient greater than a preset coefficient as a target vibration component of the corresponding vibration signal segment, and obtaining at least one target vibration component of each vibration signal segment; constructing a grayscale image of the vibration signal based on at least one target vibration component of each vibration signal segment; The grayscale image of the vibration signal is input into the fault recognition model built based on the convolutional neural network algorithm to perform fault recognition and obtain the fault recognition result; Each vibration signal segment has a sub-monitoring period in a preset monitoring period, and each sub-monitoring period has a plurality of sampling points; Calculating the similarity coefficient between each vibration component in each vibration component set and the corresponding vibration signal segment, including: For each vibration signal segment, obtaining the signal value of the vibration signal segment at each sampling point in the corresponding sub-monitoring period; Calculate the distance between each vibration component and the signal values ​​of the corresponding multiple sampling points and sum them up to obtain the total similarity distance; Determine a similarity coefficient between each vibration component and the vibration signal segment according to the similarity total distance; The expression of the similarity total distance is: ； In the formula, is the total similarity distance, is the signal value of the nth sampling point in the vibration signal segment, N is the total number of sampling points, is the mth vibration component; The function expression of the similarity coefficient between each vibration component and the vibration signal segment is: ； In the formula, is the mth vibration component The similarity coefficient with the vibration signal segment, is the total similarity distance, is the signal value of the nth sampling point in the vibration signal segment, and N is the total number of sampling points.

2. The vehicle component image fault recognition method based on convolutional neural network according to claim 1 is characterized in that: The fault identification model includes a convolutional layer, a denoising unit, a fully connected layer and an output layer connected in sequence, and the denoising unit includes an attention module and a filtering correction module.

3. The vehicle component image fault recognition method based on convolutional neural network according to claim 2 is characterized in that: The convolution layer is used to extract features from the input grayscale image to obtain a multi-channel feature map; The denoising unit is used to perform adaptive denoising on the feature map of each channel in the multi-channel feature map to obtain a denoised multi-channel feature map; The fully connected layer is used to fully connect the denoised multi-channel feature map to obtain a classification result; The output layer is used to output the classification results.

4. The vehicle component image fault recognition method based on convolutional neural network according to claim 3 is characterized in that: The noise reduction unit is specifically used for: Use multi-channel feature maps as input features of the denoising unit; The attention module of the denoising unit maps the input features to obtain a global mapping value of each channel; and obtains a first connection output value of each channel according to the global mapping value of each channel and a preset first mapping connection weight; Obtaining a second connection output value of each channel according to the first connection output value of each channel and a preset second mapping connection weight; The attention module of the denoising unit scales the second connection output value of each channel to obtain a scaling weight of each channel; The filter correction module of the noise reduction unit averages the global mapping values ​​of all channels to obtain a global mapping mean; According to the global mapping mean and the scaling weight of each channel, the adaptive filtering threshold of each channel is calculated; based on the adaptive filtering threshold of each channel, the global mapping value of each channel is filtered and denoised to obtain a denoised multi-channel feature map.

5. The vehicle component image fault recognition method based on convolutional neural network according to any one of claims 1 to 4, characterized in that: The method further includes: training a fault identification model, including: Acquire a vibration data sample, and expand the vibration data sample to obtain a plurality of expanded samples; Encoding a plurality of extended samples to obtain a plurality of vibration image samples; Calculate the grayscale distribution feature matrix of a number of vibration image samples, and construct a training set with the grayscale distribution feature matrix of the number of vibration image samples; The fault identification model is trained based on the training set until the fault identification model converges to obtain a trained fault identification model.

6. The vehicle component image fault recognition method based on convolutional neural network according to claim 5 is characterized in that: The vibration data samples are expanded to obtain multiple extended samples, including: Based on a preset sampling interval and sampling length, the vibration data samples are overlapped and sampled to obtain a plurality of sampled samples; wherein the sampling interval is the interval between the starting points of two adjacent sampling operations; The vibration data samples are divided into a number of sub-samples according to the sampling length, and the sub-samples and the sampling samples are used as a plurality of extended samples.

7. A vehicle component image fault recognition system based on a convolutional neural network, used to implement the vehicle component image fault recognition method based on a convolutional neural network as described in any one of claims 1 to 6, characterized in that: The system comprises: A signal acquisition module is used to acquire the vibration signal of the vehicle bearing within a preset monitoring period, and segment the vibration signal to obtain a plurality of vibration signal segments; wherein the end time of the preset monitoring period is the current time; A signal decomposition module, used for performing modal decomposition on each vibration signal segment to obtain a vibration component set of each vibration signal segment, wherein the vibration component set includes multiple vibration components; A similarity calculation module is used to calculate a similarity coefficient between each vibration component in each vibration component set and the corresponding vibration signal segment, and take the vibration component with a similarity coefficient greater than a preset coefficient as the target vibration component of the corresponding vibration signal segment, to obtain at least one target vibration component of each vibration signal segment; An image construction module, configured to construct a grayscale image of the vibration signal based on at least one target vibration component of each vibration signal segment; The fault identification module is used to input the grayscale image of the vibration signal into a fault identification model built based on the convolutional neural network algorithm to perform fault identification and obtain a fault identification result.

Images