Bearing fault diagnosis method based on PCNN-BILSTM

Abstract

The invention discloses a bearing fault diagnosis method based on PCNN-BILSTM, and the method specifically comprises the following steps: 1, carrying out the fault diagnosis of a bearing; acquiring a sample data set, performing data preprocessing on the acquired vibration signals, and dividing the preprocessed sample data set into a training data set and a test data set; a PCNN-BILSTM bearing fault diagnosis model is constructed, the training data set is input into the PCNN-BILSTM bearing fault diagnosis model for training, and the PCNN-BILSTM bearing fault diagnosis model comprises two parallel convolutional neural networks, a feature fusion module, a bidirectional long and short term memory network, a full connection layer and a Softmax output layer; and finally inputting a test data set to evaluate the performance of the whole model. Through combination of a parallel convolutional neural network and a bidirectional long short-term memory network, adaptive feature extraction and rolling bearing fault diagnosis under high-noise and complex working conditions are realized.

Description

Technical Field

[0001] The present invention relates to the field of bearing fault diagnosis, and in particular to a bearing fault diagnosis method based on PCNN-BILSTM. Background Art

[0002] Rolling bearings play a key role in wind turbines. Their performance and reliability are crucial to ensuring the stable operation of the entire wind turbine. Rolling bearings operate in harsh working environments and are subject to complex load changes, making them prone to various faults such as wear, cracks, and peeling. If the fault is not discovered and handled in a timely manner, the equipment may shut down, production efficiency may be reduced, and even serious safety accidents may occur. Therefore, in order to ensure the safe operation of wind turbines, it is necessary to conduct research on intelligent monitoring and diagnosis of rolling bearings.

[0003] With the advancement of monitoring technology and the increasing complexity of mechanical equipment structures, deep learning technology has been valued for its ability to automatically extract deep fault features. It abandons the need to rely on traditional expert systems. Its multi-level structure gives it powerful data fitting and processing capabilities, and can efficiently process and analyze complex data sets. Wang Lihua et al. developed a fault diagnosis technology that combines short-time Fourier transform and convolutional neural network. This technology effectively utilizes the time-frequency characteristics of the signal and enhances the accuracy and efficiency of fault diagnosis through the efficient feature extraction ability of the deep learning model. Zhao Xiaoqiang et al. (Zhao Xiaoqiang, Zhang Long, Zhao Lijuan, et al. Rolling bearing fault diagnosis using recursive graph coding technology and residual network [J]. Journal of Xi'an Jiaotong University, 2023, 57(2): 110-120.) proposed an improved CNN model for rolling bearing fault diagnosis under varying working conditions. This method can effectively capture key features in sample data even in a high-noise environment by distributing feature extraction tasks in convolutional layers of different scales. Sun et al. (Sun GD, Wang YR, Sun CF, et al. Intelligent Detection of a Planetary Gearbox Composite Fault Based on Adaptive Separation and Deep Learning [J]. Sensors, 2019, 19 (23): 5222.) proposed an improved particle swarm optimization variational mode decomposition (IPVMD) and an improved convolutional neural network (I-CNN) to solve the problem of composite fault diagnosis of planetary gearboxes. Hu Bingbing et al. (Hu Bingbing, Tang Jiahui, Wu Jimei, et al. Research on rolling bearing fault diagnosis method based on Triple GAN [J]. Noise and Vibration Control, 2021.) Based on the rolling bearing fault diagnosis method of printing equipment, the research explored the rolling bearing fault diagnosis method based on target detection theory from the two aspects of time domain and time-frequency domain, and verified the performance of the method through multiple experiments. Rolling bearings, due to their complex operating structure, drastically changing loads and extremely harsh working environment, make the collected signals very complex and show nonlinear characteristics. This increases the complexity of manually selecting appropriate characteristic indicators for signal processing. Simple fault diagnosis models cannot meet the changing working conditions, while overly complex algorithm models may affect the diagnosis efficiency and reduce the versatility. Summary of the invention

[0004] In order to address the deficiencies in the prior art, the present invention provides a rolling bearing fault diagnosis method based on a parallel convolutional neural network (PCNN)-bidirectional long short-term memory network (BILSTM or BiLSTM). By integrating the parallel convolutional neural network with the bidirectional long short-term memory network, the fault characteristics of the rolling bearing under complex working conditions are automatically extracted and identified, thereby realizing adaptive fault feature extraction under multi-dimensional signals and rolling bearing fault diagnosis under high noise and complex working conditions.

[0005] In order to achieve the above object, the specific scheme adopted by the present invention is: A bearing fault diagnosis method based on PCNN-BILSTM mainly includes the following steps: Step S1, obtaining vibration signals of rolling bearings under multiple working conditions through public data sets, labeling and segmenting the collected vibration signals, and obtaining a sample data set of vibration signals; Step S2, preprocessing the sample data set, and dividing the preprocessed sample data set into a training data set and a test data set; Step S3, constructing a PCNN-BILSTM bearing fault diagnosis model, and inputting the training data set into the PCNN-BILSTM bearing fault diagnosis model for training; Step S4: input the test data set into the PCNN-BILSTM bearing fault diagnosis model trained in step S3, and calculate the diagnostic accuracy of the test data set.

[0006] Furthermore, in step S3, the PCNN-BILSTM bearing fault diagnosis model includes two parallel convolutional neural networks, a feature fusion module, a bidirectional long short-term memory network, a fully connected layer and a softmax output layer. Furthermore, the bidirectional long short-term memory network includes an input layer, a forward propagation layer, a back propagation layer and an output layer, wherein the calculation formula of the output layer is as follows: In the formula, y t represents the output result of the final output layer, which is calculated by superimposing the output values ​​of the two hidden layers at each time step. g represents the activation function, U represents the weight matrix, which is used to linearly transform the forward and backward hidden state vectors, ht represents the hidden state vector at time step t, and c is the bias term.

[0007] Furthermore, the function used in the softmax output layer is the softmax activation function, and its formula is as follows: Where: q(x i ) represents the output probability corresponding to the i-th element xi in the input vector, and x is the logical input of the function layer, which produces a C-dimensional probability distribution. The function converts the output of the last layer of the network into a probability value that sums to 1 through a specific mathematical formula. The category of the sample is determined by the label with the highest probability value.

[0008] Further, in step S3, the training data set data is firstly subjected to two parallel convolutional neural networks to extract deep features of sensor data and extract local features in the spatial domain from the vibration signal; The extracted local features are weighted fused through the feature fusion module; The weighted fused features are passed through a bidirectional long short-term memory network, which extracts temporal features based on the fused spatial features. Finally, the extracted time series features are passed through the fully connected layer and the softmax output layer in sequence.

[0009] Furthermore, the two parallel convolutional neural networks are composed of three convolutional layers, each of which contains a ReLU activation function, and the convolution kernel sizes are 3, 5, and 7 respectively.

[0010] 7. A bearing fault diagnosis method based on PCNN-BILSTM according to claim 5, characterized in that when the extracted local features are weighted fused through the feature fusion module, the weights of the local features output by each convolutional neural network are all 0.5.

[0011] Furthermore, step S1 specifically includes the following steps: obtaining vibration signals of rolling bearings at the same sampling frequency, different loads, and different speeds through the Southeast University wind power transmission system simulation test bench, taking 500 data points as a sample data, and assigning sample data labels according to the rolling bearing state, combining multiple sample data to obtain a sample data set, 80% of the sample data set is divided out as a training data set, and the remaining 20% ​​is used as a test data set.

[0012] Furthermore, the sampling frequency is 1365HZ, and the rolling bearing includes three states: inner ring 0.5mm crack, inner ring 0.8mm crack and normal state.

[0013] Furthermore, step S2 specifically includes the following steps: The Kalman filter is used to denoise the signal. It calculates the Kalman gain by continuously updating the covariance of the system state estimate and combining the filter error covariance at the previous moment and the prediction error covariance at the current moment. In the continuous filtering process of the system state, new measurement data is introduced at each step to provide additional information. The discrete form of the Kalman filter linear state equation is as follows: xk+1 =Ax k +Bu k +w k y k =Hx k +v k Where: x k and x k+1 Represent the state vectors at the kth moment and the k+1th moment respectively; w k is the system noise vector; u k is the system input; y k is the system output, that is, the observed value; v k is the measurement noise vector; A is the state transfer matrix; B is the input matrix; H is the output matrix; The Kalman filter updates the state estimate in real time, and the update process follows the following formula: P k =AP k-1 A T +Q Where: It represents the state estimate without considering the observed value at the kth moment, that is, the prior state estimate at the kth moment; It uses the k-1th moment observation value y k-1 The state estimate obtained after is the posterior state estimate at the k-1th moment; P k is the prior estimation error covariance matrix; P k-1 is the posterior estimation error covariance matrix, and Q represents the covariance matrix of the process noise.

[0014] Beneficial effects:

[0015] (1) The rolling bearing fault diagnosis method disclosed in the present invention automatically extracts and identifies the fault features of rolling bearings under complex working conditions by integrating a parallel convolutional neural network with a bidirectional long short-term memory network, thereby realizing adaptive fault feature extraction under multi-dimensional signals and rolling bearing fault diagnosis under high noise and complex working conditions.

[0016] (2) The rolling bearing fault diagnosis method disclosed in the present invention uses a Kalman filter, which can separate useful information from noise in the signal and achieve signal noise reduction.

[0017] (3) The rolling bearing fault diagnosis method disclosed in the present invention can effectively and adaptively extract characteristic signals because of its parallel convolutional neural network.

[0018] (4) In the rolling bearing fault diagnosis method disclosed in the present invention, the bidirectional long short-term memory network captures the time-dependent characteristics of the signal with its advantages in time series analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is a flow chart of the present invention.

[0020] Figure 2 It is a structural diagram of the fault diagnosis model of the present invention.

[0021] Figure 3 It is a comparative experimental result diagram of the present invention.

[0022] Figure 4 It is the confusion matrix result diagram of the present invention.

[0023] Figure 5 It is a diagram of the ablation test results of the present invention. DETAILED DESCRIPTION

[0024] The technical solution of the present invention will be clearly and completely described below in conjunction with specific embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0025] The present invention discloses a rolling bearing fault diagnosis method based on PCNN-BILSTM, such as Figure 1 As shown, the specific steps include: obtaining a sample data set: obtaining the vibration signal of the rolling bearing under different working conditions through a public data set, obtaining a vibration signal sample data set, performing data preprocessing on the collected vibration signal, and dividing the preprocessed sample data set into a training data set and a test data set; then constructing a PCNN-BILSTM bearing fault diagnosis model, and inputting the training data set into the PCNN-BILSTM bearing fault diagnosis model for training; finally inputting the test data set to evaluate the performance of the entire model; the present invention realizes adaptive feature extraction and rolling bearing fault diagnosis under high noise and complex working conditions by combining a parallel convolutional neural network and a bidirectional long short-term memory network. Each step is described in detail below.

[0026] Step S1, obtaining a sample data set: obtaining vibration signals of rolling bearings under multiple working conditions through public data sets, labeling and segmenting the collected vibration signals, and obtaining a sample data set of vibration signals, wherein the sample data set includes a training data set and a test data set.

[0027] The vibration signals of rolling bearings at the same sampling frequency, different loads and different speeds are obtained through the wind power transmission system simulation test bench of Southeast University. The sampling frequency is 1365HZ. The rolling bearings have three states: 0.5mm crack on the inner ring, 0.8mm crack on the inner ring and normal state. 500 data points are taken as a sample data, and the sample data labels are assigned according to the rolling bearing state. Multiple sample data are combined to obtain a sample data set.

[0028] Step S2: preprocess the sample data set, and divide the preprocessed sample data set into a training data set and a test data set.

[0029] The sample data set is preprocessed, that is, outliers are removed, duplicate values ​​are integrated, missing values ​​are filled, and signal noise reduction is performed on the data set. The signal noise reduction process is: the Kalman filter is used to denoise the signal, which continuously updates the covariance of the system state estimate, combines the filter error covariance at the previous moment and the prediction error covariance at the current moment, to calculate the Kalman gain. In the continuous filtering process of the system state, new measurement data is introduced at each step to provide additional information. The discrete form of the Kalman filter linear state equation is as follows: x k+1 =Ax k +Bu k +w k y k =Hx k +v k Where: x k and x k+1 Represent the state vectors at the kth moment and the k+1th moment respectively; w k is the system noise vector; u k is the system input; y k is the system output, that is, the observed value; v k is the measurement noise vector; A is the state transfer matrix; B is the input matrix; H is the output matrix.

[0030] The Kalman filter updates the state estimate in real time, and the update process follows the following formula: P k =AP k-1 A T +Q Where: It represents the state estimate without considering the observed value at the kth moment, that is, the prior state estimate at the kth moment; It uses the k-1th moment observation value y k-1 The state estimate obtained after is the posterior state estimate at the k-1th moment; is the prior estimation error covariance matrix; P k-1 is the posterior estimation error covariance matrix, and Q represents the covariance matrix of the process noise.

[0031] 80% of the preprocessed sample data set is divided into training data sets, and the remaining 20% ​​is used as a test data set.

[0032] Step S3: construct a PCNN-BILSTM bearing fault diagnosis model, and input the training data set into the PCNN-BILSTM bearing fault diagnosis model for training.

[0033] Step 31: First, establish the PCNN-BILSTM fault diagnosis model. The model structure diagram is as follows: Figure 2 As shown, the PCNN-BILSTM fault diagnosis model includes a parallel convolutional neural network (PCNN) and a bidirectional long short-term memory network (BILSTM).

[0034] Step 32: Input the preprocessed training data set into two PCNN branches with different convolution kernels respectively. CNN uses its feature extraction capability to extract local features in the spatial domain from the vibration signal. Each convolution neural network consists of three convolution layers, each of which contains a ReLU activation function, and the convolution kernel sizes are 3, 5, and 7 respectively.

[0035] Step S33: Perform weighted fusion on the local features extracted in step S32, and the weight of each feature output by the convolutional neural network is 0.5.

[0036] Step S34, input the weighted fused features into the bidirectional long short-term memory network, and the bidirectional long short-term memory network uses its advantages in time series analysis to capture the time-dependent characteristics of the signal. The bidirectional long short-term memory network contains two complete sets of long short-term memory network parameters, one long short-term memory network processes the forward information of the sequence information (from the past to the future), and the other long short-term memory network processes the reverse information of the sequence information (from the future to the past). The forward and backward mathematical expressions of the bidirectional long short-term memory network are as follows: Where h represents the calculation result of the hidden layer of the long short-term memory network; x t represents the input data at time step t; w represents the input weight of the long short-term memory network; v represents the weight at time step t-1; h t-1 represents the output of the previous moment, b represents the bias term of the long short-term memory network layer; f represents the activation function.

[0037] The bidirectional long short-term memory network includes an input layer, a forward propagation layer, a back propagation layer, and an output layer, where the calculation formula of the output layer is as follows: In the formula, y t represents the output result of the final output layer, which is calculated by superimposing the output values ​​of the two hidden layers at each time step. g represents the activation function, U represents the weight matrix, which is used to linearly transform the forward and backward hidden state vectors, and h t represents the hidden state vector at time step t, and c is the bias term.

[0038] Step S35: Input the extracted final features into the fully connected layer and the softmax output layer to obtain the predicted bearing state. The function used in the softmax output layer is the softmax activation function, which converts the outputs of multiple neurons into the (0,1) interval for processing multi-classification problems. The formula is as follows: Where: q(x i ) represents the i-th element x in the input vector i The corresponding output probability, x is the logical input of the function layer, which produces a C-dimensional probability distribution. The function converts the output of the last layer of the network into a probability value that sums to 1 through a specific mathematical formula. The category of the sample is determined by the label with the highest probability value.

[0039] The mean of the cross entropy function between the predicted bearing state and the sample data label of the test data set is calculated as the loss function of the fault diagnosis model. The model parameters are optimized by back propagation using the Adam stochastic gradient descent algorithm until the maximum training round is reached. The number of iterations is 50 and the batch size is 32. The back propagation algorithm is used to iteratively update the parameters of the parallel convolutional neural network and the bidirectional long short-term memory network until the number of iterations is reached, indicating that the PCNN-BILSTM fault diagnosis model training is completed and the process proceeds to step S4.

[0040] Step S4: input the test data set into the trained model, and calculate the diagnostic accuracy of the test data set.

[0041] Step S41: After the parameters of the PCNN-BILSTM stop updating, the trained PCNN-BILSTM is obtained, and the test data set in S2 is input into the trained parallel convolutional neural network to obtain a feature parameter matrix;

[0042] Step S42, input the feature parameter matrix in step S41 into the trained bidirectional long short-term memory network. The bidirectional long short-term memory network uses its advantage in time series analysis to capture the time-dependent characteristics of the signal. Then the final features are input into the fully connected layer and the softmax output layer, and the bearing state of the identified test data set is output. The bearing state of the identified test data set is compared with the sample data label of the test data set to calculate the diagnostic recognition accuracy of the rolling bearing fault.

[0043] In order to specifically evaluate the performance of the proposed method, the present invention compares the algorithm diagnosis rate with the traditional machine learning methods artificial neural network (ANN), temporal convolutional neural network (TCN) and multi-scale convolutional neural network (MSCNN). To ensure fairness, the above methods are all trained using the same settings. The evaluation criteria are diagnostic accuracy, F1 score, precision, and recall. The diagnostic accuracy is the ratio of correctly diagnosed samples to the total number of samples, reflecting the overall performance of the evaluation method; the F1 score reflects the performance of the evaluation method in each category; the precision reflects the "accuracy" of the model when predicting positive class labels; the recall rate reflects the model's ability to identify all positive samples, that is, the proportion of all actual positive samples found by the model. In order to reduce the impact of randomness, each experiment was repeated three times. The diagnostic accuracy of each model is obtained as follows Figure 3 The results of F1 score, precision and recall are shown in Table 1.

[0044] Table 1 Diagnostic results of each model

[0045] From Table 1 and Figure 3 It can be seen that the method proposed in the present invention significantly surpasses the other three models in terms of accuracy, F1 score, precision and recall. The performances of the three models of ANN, TCN and MSCNN are all below 40%, and only the average accuracy and average recall of MSCNN are just over 40%. In contrast, the diagnostic accuracy of the method of the present invention has increased significantly, with increases of 61.96%, 65.19% and 58.30% respectively compared with the comparison models, which effectively verifies its excellent performance improvement effect. In particular, the model of the present invention exhibits an extremely low standard deviation of 0.18, which is not only lower than all the comparison models, but also proves the high stability and excellent robustness of the proposed model on the data set.

[0046] In order to more intuitively show the accuracy of the proposed model for each classification, the confusion matrix of the fault classification results is plotted. Figure 4As shown, the horizontal axis represents the diagnostic state and the vertical axis represents the actual fault state. In the figure, 0-2 correspond to 3 state categories respectively, the values ​​on the diagonal are the recognition accuracy of each state, and the remaining values ​​are the recognition error rates. The results in the figure show that the classification accuracy of category 1 and category 2 are both 100%, and 4.76% of category 0 are identified as category 2. It is verified that the method proposed in the present invention has demonstrated extremely high classification accuracy and stability in the field of rolling bearing fault identification, and effectively reduces confusion between categories.

[0047] In order to further verify the superiority of PCNN-BILSTM in rolling bearing fault diagnosis and classification under high noise and complex working conditions, ablation experiments were performed on it, mainly including the following: a. PCNN-BILSTM model of the present invention; b. Only PCNN, no BILSTM; c. Only BILSTM, no PCNN; d. Replace the parallel CNN with a single-scale CNN; e. Replace BILSTM with LSTM.

[0048] During the model training process, the learning rate, mini-batch, and training batch settings of each model are consistent. The optimizer uses the Adam optimizer, the learning rate is 0.0001, the number of training rounds is 50, and the mini-batch size is set to 32. In order to reduce the impact of randomness, each experiment is performed three times. The evaluation criteria are still diagnostic accuracy, F1 score, precision, and recall. The results are as follows Figure 5 And as shown in Table 2.

[0049] Table 2 Diagnostic results of each model

[0050] Figure 5 The experimental results in Table 2 strongly confirm the significant advantages and effectiveness of the model proposed in the present invention in rolling bearing fault diagnosis. The comparative analysis results show that each module of the model plays an important role in the overall performance. Specifically, when the BILSTM module is removed, the diagnostic accuracy of the model suffers a significant decline of 49.06%, and the stability is also greatly weakened; similarly, if the PCNN module is removed, the accuracy drops sharply by 44.76%, highlighting its importance at the feature extraction level. Simplifying PCNN into a single CNN architecture results in a 53.19% drop in accuracy. Replacing BILSTM with traditional LSTM drops the accuracy by 56.68%, highlighting the importance of BILSTM in capturing complex dependencies of time series.

[0051] In summary, the PCNN-BILSTM fault diagnosis model proposed in this paper not only demonstrates the powerful ability of CNN in feature extraction, but also highlights the outstanding advantages of BILSTM in processing time series data. The combination of the two significantly improves the effectiveness and stability of rolling bearing fault diagnosis.

[0052] The above description is only a preferred embodiment of the present invention and does not limit the present invention in any form. Any equivalent changes or modifications made according to the essence of the present invention should be included in the protection scope of the present invention.

Claims

1. A bearing fault diagnosis method based on PCNN-BILSTM, characterized in that: The main steps are as follows: Step S1, obtaining vibration signals of rolling bearings under multiple working conditions through public data sets, labeling and segmenting the collected vibration signals, and obtaining a sample data set of vibration signals; Step S2, preprocessing the sample data set, and dividing the preprocessed sample data set into a training data set and a test data set; Step S3, constructing a PCNN-BILSTM bearing fault diagnosis model, and inputting the training data set into the PCNN-BILSTM bearing fault diagnosis model for training; Step S4: input the test data set into the PCNN-BILSTM bearing fault diagnosis model trained in step S3, and calculate the diagnostic accuracy of the test data set.

2. A bearing fault diagnosis method based on PCNN-BILSTM according to claim 1, characterized in that: In step S3, the PCNN-BILSTM bearing fault diagnosis model includes two parallel convolutional neural networks, a feature fusion module, a bidirectional long short-term memory network, a fully connected layer and a Softmax output layer.

3. A bearing fault diagnosis method based on PCNN-BILSTM according to claim 2, characterized in that: The bidirectional long short-term memory network includes an input layer, a forward propagation layer, a back propagation layer, and an output layer, where the calculation formula of the output layer is as follows: In the formula, y t represents the output result of the final output layer, which is calculated by superimposing the output values ​​of the two hidden layers at each time step. g represents the activation function, U represents the weight matrix, which is used to linearly transform the forward and backward hidden state vectors, and h t represents the hidden state vector at time step t, and c is the bias term.

4. A bearing fault diagnosis method based on PCNN-BILSTM according to claim 2, characterized in that: The function used in the softmax output layer is the softmax activation function, and its formula is as follows: Where: q(x i ) represents the i-th element x in the input vector i The corresponding output probability, x is the logical input of the function layer, which produces a C-dimensional probability distribution. The function converts the output of the last layer of the network into a probability value that sums to 1 through a specific mathematical formula. The category of the sample is determined by the label with the highest probability value.

5. A bearing fault diagnosis method based on PCNN-BILSTM according to claim 2, characterized in that: In step S3, the training data set data first passes through two parallel convolutional neural networks to extract deep features of sensor data and extract local features in the spatial domain from the vibration signal; The extracted local features are weighted fused through the feature fusion module; The weighted fused features are passed through a bidirectional long short-term memory network, which extracts temporal features based on the fused spatial features. Finally, the extracted time series features are passed through the fully connected layer and the softmax output layer in sequence.

6. A bearing fault diagnosis method based on PCNN-BILSTM according to claim 5, characterized in that: The two parallel convolutional neural networks are composed of three convolutional layers, each of which contains a ReLU activation function, and the convolution kernel sizes are 3, 5, and 7 respectively.

7. The bearing fault diagnosis method based on PCNN-BILSTM according to claim 5 is characterized in that: When the extracted local features are weighted fused through the feature fusion module, the weights of the local features output by each convolutional neural network are all 0.

5.

8. The bearing fault diagnosis method based on PCNN-BILSTM according to claim 1 is characterized in that: Step S1 specifically includes the following steps: obtaining the vibration signals of the rolling bearing at the same sampling frequency, different loads, and different speeds through the Southeast University wind power transmission system simulation test bench, taking 500 data points as a sample data, and assigning sample data labels according to the rolling bearing state, combining multiple sample data to obtain a sample data set, 80% of the sample data set is divided out as a training data set, and the remaining 20% ​​is used as a test data set.

9. The bearing fault diagnosis method based on PCNN-BILSTM according to claim 8 is characterized in that: The sampling frequency is 1365HZ, and the rolling bearing has three states: inner ring crack of 0.5mm, inner ring crack of 0.8mm and normal state.

10. The bearing fault diagnosis method based on PCNN-BILSTM according to claim 1, characterized in that: Step S2 specifically includes the following steps: The Kalman filter is used to denoise the signal. It calculates the Kalman gain by continuously updating the covariance of the system state estimate and combining the filter error covariance at the previous moment and the prediction error covariance at the current moment. In the continuous filtering process of the system state, new measurement data is introduced at each step to provide additional information. The discrete form of the Kalman filter linear state equation is as follows: x k+1 =Ax k +Bu k +w k y k =Hx k +v k Where: x k and x k+1 Represent the state vectors at the kth moment and the k+1th moment respectively; w k is the system noise vector; u k is the system input; y k is the system output, that is, the observed value; v k is the measurement noise vector; A is the state transfer matrix; B is the input matrix; H is the output matrix; The Kalman filter updates the state estimate in real time, and the update process follows the following formula: Where: It represents the state estimate without considering the observed value at the kth moment, that is, the prior state estimate at the kth moment; It uses the k-1th moment observation value y k-1 The state estimate obtained after is the posterior state estimate at the k-1th moment; P k is the prior estimation error covariance matrix; P k-1 is the posterior estimation error covariance matrix, and Q represents the covariance matrix of the process noise.

Images