Rolling bearing fault diagnosis method and system based on WSETCNNLBKALSSVM

Abstract

The invention discloses a rolling bearing fault diagnosis method based on a WSETCNNLBKALSSVM, and the method comprises the steps: S1, carrying out the modal decomposition of a rolling bearing vibration signal collected by a sensor through employing the high time-frequency resolution characteristic of wavelet synchronous extraction transformation, obtaining a decomposed time-frequency image, and carrying out the lossless compression processing of the time-frequency image; s2, dividing the time-frequency image data after lossless compression processing into a training set and a test set according to a set proportion; s3, inputting the generated time-frequency image data in the training set into a two-dimensional convolutional neural network, and performing adaptive fault feature extraction; and S4, taking an output result of a full connection layer of the two-dimensional convolutional neural network as an input of a least square support vector machine, optimizing hyper-parameters of the LSSVM by adopting an improved black-wing optimization algorithm, and completing classification of fault types through the LSSVM. The rolling bearing fault diagnosis method has higher classification precision and diagnosis accuracy in rolling bearing fault diagnosis.

Description

Technical Field

[0001] The invention relates to the technical field of rolling bearing fault diagnosis, and in particular to a rolling bearing fault diagnosis method based on WSET_CNN_LBKA_LSSVM. Background Art

[0002] Rotating motors are widely used in the industrial field. The core components of rotating motors are prone to failure in long-term high-speed operation environments, affecting equipment safety and causing property losses. Therefore, monitoring and diagnosing the working status of bearings is of great value in engineering applications. In order to realize rolling bearing fault diagnosis, the collected signals mainly include vibration, acoustic, electrical and temperature signals. Among them, vibration signals are widely used because they can effectively characterize the characteristics of bearing faults. Signal processing is one of the key tasks in machine fault diagnosis. With the introduction of preventive maintenance and equipment health monitoring, a variety of vibration signal processing technologies have been developed to meet the needs of bearing fault diagnosis. The signal processing technologies commonly used in mechanical fault diagnosis can be divided into three categories: vibration analysis based on direct processing of time series data, vibration analysis based on machine learning (ML), and vibration analysis based on deep learning (DL).

[0003] Signal processing techniques for vibration analysis can be further divided into time domain, frequency domain and time-frequency domain analysis. Traditional ML-based methods include artificial neural network (ANN), principal component analysis (PCA), support vector machine (SVM), etc., while traditional DL-based bearing fault diagnosis algorithms include convolutional neural network (CNN), autoencoder, deep belief network (DBN), recurrent neural network (RNN), generative adversarial network (GAN) and transfer learning methods based on deep learning.

[0004] Time-frequency domain analysis has greater advantages than time-domain analysis and frequency-domain analysis. It can expand the frequency-domain analysis of non-stationary signals, and is especially suitable for waveforms under variable speed conditions. Common time-frequency analysis methods include short-time Fourier transform (STFT), Wigner-Ville distribution (WVD), reduced interference distribution (RID), wavelet analysis (such as continuous wavelet transform (CWT), discrete wavelet transform (DWT) and wavelet packet transform (WPT)), etc. These methods can process non-stationary signals for machine fault diagnosis. However, the Heisenberg uncertainty principle imposes limitations on the time-frequency distribution of these methods, resulting in fuzzy time-frequency representation and the inability to accurately locate the instantaneous characteristics of the signal, which may weaken or mask key features such as instantaneous frequency and instantaneous amplitude in fault diagnosis applications. Summary of the invention

[0005] The technical problem to be solved by the present invention is: in the process of rolling bearing fault diagnosis, how to accurately locate the instantaneous characteristics of the signal to avoid weakening or covering up key characteristics such as instantaneous frequency and instantaneous amplitude.

[0006] In order to solve the above technical problems, the present invention provides a rolling bearing fault diagnosis method based on WSET_CNN_LBKA_LSSVM, comprising:

[0007] S1. Using the high time-frequency resolution characteristics of wavelet synchronous extraction transformation, the rolling bearing vibration signal collected by the sensor is modally decomposed to obtain the decomposed time-frequency image, and the time-frequency image is losslessly compressed;

[0008] S2, dividing the time-frequency image data after lossless compression processing into a training set and a test set according to a set ratio, wherein the training samples are used for training the neural network model, and the test samples are used for evaluating the performance of the neural network model;

[0009] S3, inputting the generated time-frequency image data in the training set into the two-dimensional convolutional neural network to perform adaptive fault feature extraction;

[0010] S4. The output result of the fully connected layer of the two-dimensional convolutional neural network is used as the input of the least squares support vector machine, and the improved black kite optimization algorithm is used to optimize the hyperparameters of LSSVM, and then the fault type is classified through LSSVM.

[0011] The aforementioned rolling bearing fault diagnosis method based on WSET_CNN_LBKA_LSSVM, in step one, obtains the vibration signal data of the rolling bearing at the same sampling frequency, load and speed through a public data set, classifies the vibration signal data according to the fault state of the rolling bearing, including roller fault, inner ring fault, outer ring fault, compound fault, and normal state, takes multiple data points of the same category as a sample data, and forms a sample data set through multiple sample data, and finally divides the sample data set into training data and test data set.

[0012] In the aforementioned rolling bearing fault diagnosis method based on WSET_CNN_LBKA_LSSVM, in step 2, the acquired vibration signal data is subjected to modal decomposition by the WSET method, and the frequency characteristics of different time periods are extracted, thereby generating a high-resolution time-frequency image;

[0013] Performing data compression on the generated time-frequency image;

[0014] The obtained compressed data is divided into training set and test set according to the set ratio.

[0015] In the aforementioned rolling bearing fault diagnosis method based on WSET_CNN_LBKA_LSSVM, in step three, the specific steps are as follows:

[0016] S31: The data first passes through a convolution kernel of size 3×3 to scan the local area of ​​the image to extract preliminary features, including edges and basic textures;

[0017] S32: After preliminary feature extraction, the data flows into the first pooling layer, and the redundant information in the image feature map is compressed and filtered through a 2×2 maximum pooling operation;

[0018] S33: The image feature map flows into the second group of convolutional layers and the second pooling layer after being downsampled by the first pooling layer;

[0019] S34: After two rounds of convolution and pooling, the feature map is highly compressed and integrated into the key pattern. After the feature map is flattened into a one-dimensional vector, it is further processed through multiple fully connected layers in sequence.

[0020] In the aforementioned rolling bearing fault diagnosis method based on WSET_CNN_LBKA_LSSVM, in step S31, the convolution kernel generates 16 feature maps, each of which contains response information of the image under different convolution kernels;

[0021] The convolution operation uses the "same" padding strategy to make the output feature map size consistent with the original input image. Batch normalization is performed after convolution to standardize the value of the feature map;

[0022] Then the ReLU activation function is used to transform the linear combination features into nonlinear ones.

[0023] In the aforementioned rolling bearing fault diagnosis method based on WSET_CNN_LBKA_LSSVM, in step S33, the second group of convolution layers uses a 5×5 convolution kernel to generate 32 feature maps to capture more complex and deep feature structures, including the following operations:

[0024] Perform batch normalization;

[0025] Use the ReLU activation function to perform nonlinear mapping on the output of the convolutional layer;

[0026] Then the dropout operation is added to randomly remove the outputs of some neurons.

[0027] In the aforementioned rolling bearing fault diagnosis method based on WSET_CNN_LBKA_LSSVM, in step S34, a fully connected layer of 64 units is first used to map the pooled features to a smaller feature space, retain the core information of the image, and a dropout operation is performed again to obtain a fully connected feature vector;

[0028] Subsequently, a 32-unit fully connected layer is used to integrate the fully connected feature vectors to further optimize the feature expression of fault classification. Finally, the output layer maps the features to the dimension consistent with the fault category, so that the output of the CNN module meets the classification requirements.

[0029] In the aforementioned rolling bearing fault diagnosis method based on WSET_CNN_LBKA_LSSVM, in step 4, the specific steps are as follows:

[0030] S41: The process of the LBKA algorithm is as follows:

[0031] Initialize the population randomly and assume that the positions of black kites are as follows:

[0032]

[0033] in, and are the lower and upper bounds of the black-winged kite search, respectively, and rand is a random value between [0,1];

[0034] The Levy flight calculation formula is as follows:

[0035]

[0036]

[0037]

[0038] in , S is the step length of the Levy flight, , , , represents the Levy distribution, is a random number generated by normal distribution, t is the current iteration number, represents the gamma function, , is to generate random variables The standard deviation of is to generate random variables The standard deviation of

[0039] The attack behavior formula introduced by Levi Flight is as follows:

[0040]

[0041] in is the scaling factor used to control the scale of the step size, is the probability parameter of attack behavior, , , is the number of current iterations, is the total number of iterations, , represents the new position of the i-th individual in the j-th dimension at the t+1-th iteration;

[0042] Adding Levi flight to the migration behavior, the formula is as follows:

[0043]

[0044] in, represents the Cauchy mutation, represents that as of the tth iteration, the black-winged kite is the leading scorer, ;

[0045] S42: Use the LBKA algorithm to optimize the hyperparameters gam and sig2 of LSSVM, find the optimal values ​​and input them into the LSSVM model;

[0046] S43: The feature vector output by the last fully connected layer of the CNN network is used as the input of the LSSVM model, and the LSSVM model finally completes the classification task.

[0047] A computer device / equipment / system comprises a memory, a processor and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the above method.

[0048] A computer-readable storage medium having a computer program / instruction stored thereon, characterized in that the computer program / instruction implements the steps of the above method when executed by a processor.

[0049] The beneficial effects achieved by the present invention are as follows: the method of the present invention uses the WSET algorithm, combined with the multi-resolution characteristics of the wavelet transform, and has good time resolution in high-frequency signals and good frequency resolution in low-frequency signals, so that it can more accurately analyze complex multi-component non-stationary signals. By extracting only the signal time-frequency spectrum corresponding to the instantaneous frequency and removing the ambiguous time-frequency energy, the concentration of the time-frequency representation can be effectively enhanced, the clarity of the signal features in the time-frequency domain is improved, and the resolution limitations caused by the fixed STFT window are overcome. When analyzing complex signals containing fast-changing or slow-changing modes, better time-frequency resolution can be obtained. By using a diagnostic method that integrates WSET and CNN_LBKA_LSSVN, the vibration data collected by the sensor in real time is used to diagnose the type of rolling bearing fault, which can be done with high accuracy and good generalization ability on a small sample training set. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 is a flow chart of the method in Embodiment 1 of the present invention;

[0051] Figure 2 This is a flowchart of LBKA_LSSVM in Example 1 of the present invention;

[0052] Figure 3 It is a framework diagram of WSET_CNN_LBKA_LSSVM in Example 1 of the present invention;

[0053] Figure 4 This is the confusion matrix diagram of the WSET_CNN_LBKA_LSSVM test set in Example 1 of the present invention. DETAILED DESCRIPTION

[0054] To make the above-mentioned purpose, technical solution and advantages of the present invention more clear, the present invention will be further described in detail below with reference to the accompanying drawings. Obviously, the embodiments described in the present invention are only used to explain the present invention, not to limit the present invention.

[0055] Example 1

[0056] Figure 1 It is a flow chart of a preferred embodiment of a rolling bearing fault diagnosis method based on WSET_CNN_LBKA_LSSVM provided by the present invention.

[0057] A rolling bearing fault diagnosis method based on WSET_CNN_LBKA_LSSVM, comprising:

[0058] S1. Using the high time-frequency resolution of wavelet synchronous extraction transform (WSET), the rolling bearing vibration signal collected by the sensor is modally decomposed to obtain the decomposed time-frequency image, and the time-frequency image is losslessly compressed;

[0059] S2, dividing the time-frequency image data after lossless compression processing into a training set and a test set according to a set ratio, wherein the training samples are used for training the neural network model, and the test samples are used for evaluating the performance of the neural network model;

[0060] S3, input the generated time-frequency image into a two-dimensional convolutional neural network (CNN) for adaptive fault feature extraction;

[0061] S4. The output result of the fully connected layer of the two-dimensional convolutional neural network is used as the input of the least squares support vector machine (LSSVM), and the improved black kite optimization algorithm (LBKA) is used to optimize the hyperparameters of LSSVM to further improve the classification accuracy. The fault types are then classified through LSSVM.

[0062] In step one, the vibration signal data of the rolling bearing at the same sampling frequency, load and speed are obtained through the public data set, and the vibration signal data is classified according to the fault state of the rolling bearing, including roller fault, inner ring fault, outer ring fault, compound fault and normal state. Multiple data points of the same category are taken as a sample data, and a sample data set is composed of multiple sample data. Finally, the sample data set is divided into training data and test data set as shown in Table 1.

[0063] In step 2, the acquired vibration signal data is processed using WSET to improve the time-frequency resolution. The vibration signal is modally decomposed by the WSET method to extract the frequency characteristics of different time periods, thereby generating a high-resolution time-frequency image.

[0064] The generated time-frequency images are compressed to improve the efficiency of subsequent model training and reduce processing time.

[0065] The compressed data is divided into training set and test set according to a certain ratio. Then, 15% of the data of each fault type is randomly selected as the training set, and the remaining 85% is used as the test set, as shown in Table 1.

[0066] Table 1 is the data set partition table

[0067] Fault type Category Label Number of training sets Number of test sets Sample length Roller 1 30 170 2048 complex 2 30 170 2048 normal 3 30 170 2048 Inner Circle 4 30 170 2048 Outer ring 5 30 170 2048

[0068] Example 2

[0069] A rolling bearing fault diagnosis method based on WSET_CNN_LBKA_LSSVM, comprising:

[0070] S1. Using the high time-frequency resolution of wavelet synchronous extraction transform (WSET), the rolling bearing vibration signal collected by the sensor is modally decomposed to obtain the decomposed time-frequency image, and the time-frequency image is losslessly compressed;

[0071] S2, dividing the time-frequency image data after lossless compression processing into a training set and a test set according to a set ratio, wherein the training samples are used for training the neural network model, and the test samples are used for evaluating the performance of the neural network model;

[0072] S3, input the generated time-frequency image into a two-dimensional convolutional neural network (CNN) for adaptive fault feature extraction;

[0073] S4. The output result of the fully connected layer of the two-dimensional convolutional neural network is used as the input of the least squares support vector machine (LSSVM), and the improved black kite optimization algorithm (LBKA) is used to optimize the hyperparameters of LSSVM to further improve the classification accuracy. The fault types are then classified through LSSVM.

[0074] In step three, the generated time-frequency image is input into a two-dimensional convolutional neural network (CNN) to achieve adaptive fault feature extraction. The specific steps are as follows:

[0075] S31: The data first passes through a convolution kernel of size 3×3, scanning the local area of ​​the image to extract preliminary features, including edges and basic textures.

[0076] The convolution kernel generates 16 feature maps, each of which contains the response information of the image under different convolution kernels.

[0077] To ensure information integrity, the convolution operation uses the "same" padding strategy to make the output feature map size consistent with the original input image. Batch normalization is performed after convolution to standardize the values ​​of the feature map in order to accelerate training and improve model stability. The ReLU activation function is then used to convert the linear combination features into nonlinearities, allowing the model to learn more complex feature patterns.

[0078] S32: After preliminary feature extraction, the data flows into the first pooling layer, and the redundant information in the image feature map is compressed and filtered through a 2×2 maximum pooling operation.

[0079] The pooling layer retains the maximum value of each 2×2 region, thereby reducing the resolution of the feature map while reducing redundant information and retaining key features. In addition, the pooling layer uses a sliding window with a stride of 2 for downsampling, thereby reducing the amount of calculation and preventing overfitting. At the same time, the downsampling operation of the pooling layer produces translation invariance on the input image of the model, thereby enhancing the robustness of the model.

[0080] S33: The image feature map is downsampled by the first pooling layer and then flows into the second set of convolutional layers and the second pooling layer operation.

[0081] The second set of convolutional layers uses a 5×5 convolution kernel to generate 32 feature maps to capture more complex and deep feature structures. Similar to the first convolutional layer, batch normalization is performed; the output of the convolutional layer is nonlinearly mapped using the ReLU activation function, thereby enhancing the model's ability to express complex patterns, avoiding the linear limitations of the model, and providing richer input for subsequent feature learning;

[0082] The dropout operation is then added to reduce the risk of overfitting by randomly removing the outputs of some neurons; the second pooling layer further compresses the feature map to retain the most representative abstract information, ensuring that the model has good generalization capabilities.

[0083] S34: After two rounds of convolution and pooling, the feature maps have been highly compressed and integrated into key patterns. These feature maps are flattened into one-dimensional vectors and further processed through multiple fully connected layers.

[0084] First, a 64-unit fully connected layer is used to map the pooled features to a smaller feature space, retaining the core information of the image, and then a dropout operation is performed again to obtain a fully connected feature vector.

[0085] Subsequently, a 32-unit fully connected layer is used to integrate the fully connected feature vector in a more detailed manner to further optimize the feature expression of fault classification. Finally, the output layer maps the features to the dimension consistent with the fault category, so that the output of the CNN module meets the classification requirements. During the CNN model training process, the Adam optimizer is used to adjust the parameters, and the number of training is set to 150 times to speed up the convergence speed and improve the model accuracy.

[0086] Example 3

[0087] A rolling bearing fault diagnosis method based on WSET_CNN_LBKA_LSSVM, comprising:

[0088] S1. Using the high time-frequency resolution of wavelet synchronous extraction transform (WSET), the rolling bearing vibration signal collected by the sensor is modally decomposed to obtain the decomposed time-frequency image, and the time-frequency image is losslessly compressed;

[0089] S2, dividing the time-frequency image data after lossless compression processing into a training set and a test set according to a set ratio, wherein the training samples are used for training the neural network model, and the test samples are used for evaluating the performance of the neural network model;

[0090] S3, input the generated time-frequency image into a two-dimensional convolutional neural network (CNN) for adaptive fault feature extraction;

[0091] S4. The output result of the fully connected layer of the two-dimensional convolutional neural network is used as the input of the least squares support vector machine (LSSVM), and the improved black kite optimization algorithm (LBKA) is used to optimize the hyperparameters of LSSVM to further improve the classification accuracy. The fault types are then classified through LSSVM.

[0092] In step 4, the output of the fully connected layer is used as the input of the least squares support vector machine (LSSVM), and the improved black kite optimization algorithm (LBKA) is used to optimize the hyperparameters of the LSSVM to further improve the classification accuracy, and then the fault type is classified by LSSVM. The specific steps are as follows:

[0093] S41: The specific process of the LBKA algorithm is as follows:

[0094] Initialize the population randomly and assume that the positions of black kites are as follows:

[0095]

[0096] in, and are the lower and upper bounds of the black-winged kite search, respectively, and rand is a random value between [0,1];

[0097] The Levy flight calculation formula is as follows:

[0098]

[0099]

[0100]

[0101] in , S is the step length of the Levy flight, , , Levy Flight is a random walk model consisting of a series of flight steps with random directions and random lengths. The distribution of flight steps follows a probability distribution with heavy-tail characteristics, namely the Levy distribution. , is a random number generated by normal distribution, t is the current iteration number, represents the gamma function, , is to generate random variables The standard deviation of is to generate random variables The standard deviation of

[0102] Introduce Levi's flying attack behavior, the specific formula is as follows:

[0103]

[0104] in is the scaling factor used to control the scale of the step size, is the probability parameter of attack behavior, , , is the number of current iterations, is the total number of iterations, , Indicates the new position of the i-th individual in the j-th dimension at the t+1-th iteration. Due to the large jump step size generated by the Levy distribution, the individual can jump directly from the current solution position to an area far away from the current position, thereby enhancing the global search capability. By adding this long-distance random search to the attack behavior, it can avoid falling into the local optimum and promote the algorithm to search in a larger range, so that it can detect more areas and improve the global search effect.

[0105] Levy flight is also added to the migration behavior to improve the diversity of solutions and global search performance. The formula is as follows:

[0106]

[0107] in, represents the Cauchy mutation, represents that as of the tth iteration, the black-winged kite is the leading scorer, In each iteration, Levy flight and migration behavior are combined. In this way, the global search capability can be effectively enhanced, the diversity of solutions can be improved, exploration and development can be balanced, convergence can be accelerated, and the risk of falling into local optimality can be reduced, which significantly improves the overall performance and robustness of the algorithm.

[0108] S42: The process of using LBKA algorithm to optimize the hyperparameters gam and sig2 of LSSVM is as follows Figure 2 As shown, find the optimal value and input it into the LSSVM model;

[0109] S43: The feature vector output by the last fully connected layer of the CNN network is used as the input of the LSSVM model, and the LSSVM model finally completes the classification task.

[0110] The model performance was analyzed by using mean square error (MSE), root mean square error (RMSE), coefficient of determination (R 2 ) Three performance indicators are used to evaluate the model as shown in Table 2.

[0111] Table 2 is a performance comparison table of cnn_bka_lssvm and cnn_lbka_lssvm on the test data set

[0112] Model Performance Indicators MSE RMSE <![CDATA[R 2 ]]> CNN_BKA_LSSVM 0.22353 0.14951 0.98882 CNN_LBKA_LSSVM 0.00236 0.04851 0.99882

[0113] The model optimized by LBKA outperforms the unoptimized model, which verifies the effectiveness of the LBKA_LSSVM method.

[0114] Figure 3 This is the framework diagram of the WSET_CNN_LBKA_LSSVM model.

[0115] Figure 4This is the confusion matrix diagram of the WSET_CNN_LBKA_LSSVM test set. It can be seen that one of the labels 3 is incorrectly assigned to label 2, and one of the labels 4 is incorrectly assigned to label 5, but the other three fault types are exactly the same as their true categories, and the accuracy rate reaches 99.76%. It proves that the rolling bearing fault diagnosis method based on WSET_CNN_LBKA_LSSVM proposed in the present invention has a strong fault diagnosis ability.

[0116] A computer device / equipment / system comprises a memory, a processor and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the above method.

[0117] A computer-readable storage medium stores a computer program / instruction, which implements the steps of the above method when executed by a processor.

[0118] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks 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 produce 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.

[0119] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0120] 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.

[0121] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A rolling bearing fault diagnosis method based on WSET_CNN_LBKA_LSSVM, characterized in that: include: S1. Using the high time-frequency resolution characteristics of wavelet synchronous extraction transformation, the rolling bearing vibration signal collected by the sensor is modally decomposed to obtain the decomposed time-frequency image, and the time-frequency image is losslessly compressed; S2, dividing the time-frequency image data after lossless compression processing into a training set and a test set according to a set ratio, wherein the training samples are used for training the neural network model, and the test samples are used for evaluating the performance of the neural network model; S3, inputting the generated time-frequency image data in the training set into the two-dimensional convolutional neural network to perform adaptive fault feature extraction; S4. The output result of the fully connected layer of the two-dimensional convolutional neural network is used as the input of LSSVM, and the improved black kite optimization algorithm is used to optimize the hyperparameters of LSSVM, and then the fault types are classified through LSSVM.

2. The rolling bearing fault diagnosis method based on WSET_CNN_LBKA_LSSVM according to claim 1 is characterized in that: In step one, the vibration signal data of the rolling bearing at the same sampling frequency, load and speed are obtained through a public data set, and the vibration signal data is classified according to the fault state of the rolling bearing, including roller fault, inner ring fault, outer ring fault, compound fault, and normal state. Multiple data points of the same category are taken as a sample data, and a sample data set is formed by multiple sample data. Finally, the sample data set is divided into training data and test data set.

3. The rolling bearing fault diagnosis method based on WSET_CNN_LBKA_LSSVM according to claim 1 is characterized in that: In step 2, the acquired vibration signal data is modally decomposed by the WSET method to extract the frequency characteristics of different time periods, thereby generating a high-resolution time-frequency image; Performing data compression on the generated time-frequency image; The obtained compressed data is divided into training set and test set according to the set ratio.

4. A rolling bearing fault diagnosis method based on WSET_CNN_LBKA_LSSVM according to claim 1, characterized in that: In step three, the specific steps are as follows: S31: The data first passes through a convolution kernel of size 3×3 to scan the local area of ​​the image to extract preliminary features, including edges and basic textures; S32: After preliminary feature extraction, the data flows into the first pooling layer, and the redundant information in the image feature map is compressed and filtered through a 2×2 maximum pooling operation; S33: The image feature map flows into the second group of convolutional layers and the second pooling layer after being downsampled by the first pooling layer; S34: After two rounds of convolution and pooling, the feature map is highly compressed and integrated into the key pattern. After the feature map is flattened into a one-dimensional vector, it is further processed through multiple fully connected layers in sequence.

5. A rolling bearing fault diagnosis method based on WSET_CNN_LBKA_LSSVM according to claim 4, characterized in that: In step S31, the convolution kernel generates 16 feature maps, each of which contains the response information of the image under different convolution kernels; The convolution operation uses the "same" padding strategy to make the output feature map size consistent with the original input image. Batch normalization is performed after convolution to standardize the value of the feature map. Then the ReLU activation function is used to transform the linear combination features into nonlinear ones.

6. A rolling bearing fault diagnosis method based on WSET_CNN_LBKA_LSSVM according to claim 4, characterized in that: In step S33, the second set of convolutional layers uses a 5×5 convolution kernel to generate 32 feature maps, including the following operations: Perform batch normalization; Use the ReLU activation function to perform nonlinear mapping on the output of the convolutional layer; Then the dropout operation is added to randomly remove the outputs of some neurons.

7. A rolling bearing fault diagnosis method based on WSET_CNN_LBKA_LSSVM according to claim 4, characterized in that: In step S34, a fully connected layer of 64 units is first used to map the pooled features to a smaller feature space, retaining the core information of the image, and then a dropout operation is performed again to obtain a fully connected feature vector; Subsequently, a 32-unit fully connected layer is used to integrate the fully connected feature vectors to further optimize the feature expression of fault classification. Finally, the output layer maps the features to the dimension consistent with the fault category, so that the output of the CNN module meets the classification requirements.

8. The rolling bearing fault diagnosis method based on WSET_CNN_LBKA_LSSVM according to claim 1 is characterized in that: In step 4, the specific steps are as follows: S41: The process of the LBKA algorithm is as follows: Initialize the population randomly and assume that the positions of black kites are as follows: ； in, and are the lower and upper bounds of the black-winged kite search, respectively, and rand is a random value between [0,1]; The Levy flight calculation formula is as follows: ； ； ； in , S is the step length of the Levy flight, , , , represents the Levy distribution, is a random number generated by normal distribution, t is the current iteration number, represents the gamma function, , is to generate random variables The standard deviation of is to generate random variables The standard deviation of The attack behavior formula introduced by Levi Flight is as follows: ； in is the scaling factor used to control the scale of the step size, is the probability parameter of attack behavior, , is the number of current iterations, is the total number of iterations, , represents the new position of the i-th individual in the j-th dimension at the t+1-th iteration; Adding Levi flight to the migration behavior, the formula is as follows: ； in, represents the Cauchy mutation, represents that as of the tth iteration, the black-winged kite is the leading scorer, ; S42: Use the LBKA algorithm to optimize the hyperparameters gam and sig2 of LSSVM, find the optimal values ​​and input them into the LSSVM model; S43: The feature vector output by the last fully connected layer of the CNN network is used as the input of the LSSVM model, and the LSSVM model finally completes the classification.

9. A computer device / equipment / system comprising a memory, a processor and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program / instruction stored thereon, characterized in that: When the computer program / instructions are executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

Images