Bearing fault diagnosis model construction method and bearing fault diagnosis method and system

Abstract

The invention discloses a method for constructing a bearing fault diagnosis model, and the method comprises the steps: collecting vibration signals of a rolling bearing under different working conditions, and constructing a sample set; constructing a CNN-BiLSTM neural network for rolling bearing fault diagnosis, and training, verifying and testing the neural network by using the sample set to obtain a classification diagnosis result; constructing an objective function according to the cross entropy loss of the neural network in the verification set, and optimizing the objective function by using a variant grey wolf algorithm to obtain an optimal neural network weight and hyper-parameter combination; and updating the neural network model by using the optimal neural network weight and hyper-parameter combination to serve as a bearing fault diagnosis model. According to the method, the objective function is constructed through the cross entropy loss of the neural network in the verification set, the hyperparameters of the neural network are optimized by using the variant grey wolf algorithm, and the constructed fault diagnosis model is insensitive to an initial value, is not liable to fall into a local optimal solution, and has better efficiency and precision.

Description

Technical Field

[0001] The present invention relates to the technical field of fault diagnosis, and in particular to a bearing fault diagnosis model construction method, a bearing fault diagnosis method and a bearing fault diagnosis system. Background Art

[0002] Rolling bearings are key components of mechanical equipment, and their health directly affects the operating efficiency and safety of the machinery. Due to their excellent radial load-bearing capacity, rolling bearings are widely used in complex working conditions such as high speed, heavy load and variable load. Compared with ordinary rotor bearing systems, rolling bearings are more susceptible to damage and troubleshooting is more challenging. In complex and harsh mechanical environments, bearings are affected by various factors such as vibration, shock and poor lubrication, which lead to wear, fatigue and damage. This not only increases maintenance costs, but also brings huge safety risks. Therefore, timely and accurate diagnosis and identification of bearing faults are crucial to ensure the safe and stable operation of mechanical equipment.

[0003] With the rise of neural network algorithms, there have been some studies on establishing fault diagnosis models based on neural networks in the diagnosis and identification of bearing faults. In neural networks, hyperparameters are crucial to model performance. However, adjusting these parameters is a time-consuming and labor-intensive task. The main reason is that bearings may fail after months or years, so data collection usually takes a long time and the number of fault samples is scarce. Therefore, in the field of fault diagnosis models, the adjustment of hyperparameters mainly relies on empirical methods, which lack theoretical support, thus limiting the transparency and interpretability of the optimization process. In addition, the current bearing fault diagnosis model has the disadvantage of being easily trapped in the local optimal solution, resulting in poor recognition effect. Therefore, based on the above problems, it is necessary to further improve the neural network model for bearing fault diagnosis to meet current needs. Summary of the invention

[0004] The present invention aims to solve one of the technical problems existing in the related art at least to a certain extent.

[0005] One purpose of the present invention is to provide a method for constructing a bearing fault diagnosis model based on a variant gray wolf optimization algorithm, which uses the variant gray wolf algorithm to optimize the neural network hyperparameters, effectively solves the model training and fault diagnosis challenges caused by data scarcity in common fault classification, and improves the generalization ability and classification accuracy of the model.

[0006] Another object of the present invention is to provide a bearing fault diagnosis method.

[0007] Another object of the present invention is to provide a bearing fault diagnosis system.

[0008] In order to achieve the above-mentioned purpose, the present invention first provides a method for constructing a bearing fault diagnosis model, comprising:

[0009] Collect vibration signals of rolling bearings under different working conditions and construct a sample set;

[0010] Construct a CNN-BiLSTM neural network for rolling bearing fault diagnosis, use the sample set to train, verify and test the neural network, and obtain classification diagnosis results;

[0011] The objective function is constructed according to the cross entropy loss of the neural network in the validation set, and the objective function is optimized using the mutant gray wolf algorithm to obtain the optimal combination of neural network weights and hyperparameters; and the neural network model is updated with the optimal combination of neural network weights and hyperparameters as the bearing fault diagnosis model.

[0012] A further preferred technical solution of the present invention is that the objective function is optimized by using the variant gray wolf algorithm to obtain the optimal combination of neural network weights and hyperparameters; comprising:

[0013] Construct a real number vector based on the hyperparameter range and threshold range in the initial neural network model;

[0014] Substitute the real number vector into the objective function to obtain the fitness value; compare the fitness value with the upper and lower thresholds of the updated neural network model to obtain the optimal combination of neural network weights and hyperparameters.

[0015] Preferably, the objective function is expressed as:

[0016] F(a)=(L val (a),A val (a))

[0017]

[0018] a = {convolution kernel size, number of filters, number of BiLSTM hidden layer units, learning rate}

[0019] Among them, N val is the number of samples; C is the number of categories; y ic is the true label of the i-th sample in category c; is the predicted probability of the i-th sample in category c; L val (a) represents the loss of model a on the validation set; A val (a) represents the accuracy of model a on the validation set; The index of the category with the highest probability among the category probabilities predicted by the model; max c y ic refers to the index of the correct category in the true label.

[0020] Preferably, the specific process of optimizing the objective function by using the variant grey wolf algorithm is as follows:

[0021] (1) Initialize the population and then optimize individual selection through genetic operations and mutation operations;

[0022] (2) Evaluate the algorithm fitness and select the one with the highest fitness as the decision layer;

[0023] (3) Perform arithmetic crossover operations on the decision layer;

[0024] (4) Introducing randomness through mutation operations to increase the diversity of the population;

[0025] (5) Select the three individuals with the highest fitness from the current population and the newly generated population as new leaders;

[0026] (6) The decision-making level updates its position and retains new leadership individuals;

[0027] (7) Update the proxy location;

[0028] (8) Define an allowable small threshold between the last two consecutive results. The maximum number of iterations is considered as the stopping criterion. Save the current optimal solution and determine whether it meets the stopping criterion. If it meets the stopping criterion, output the optimal solution. Otherwise, repeat steps (1) to (7).

[0029] Preferably, the CNN-BiLSTM neural network includes a convolutional network layer and a bidirectional long short-term memory network layer.

[0030] Preferably, the hyperparameters of the neural network include learning rate, convolution kernel size, number of filters and number of hidden layer units of BiLSTM.

[0031] Another aspect of the present invention provides a bearing fault diagnosis method, comprising:

[0032] Using the above-mentioned bearing fault diagnosis model construction method, a bearing fault diagnosis model is obtained;

[0033] Perform transfer learning training on the bearing fault diagnosis model to obtain a trained bearing fault diagnosis model;

[0034] The running vibration signal of the bearing to be tested is input into the trained bearing fault diagnosis model to obtain the fault diagnosis result of the bearing to be tested.

[0035] Preferably, the transfer learning of the model includes freezing the weights of the pre-trained model and only fine-tuning the weights of the output layer.

[0036] Another aspect of the present invention provides a bearing fault diagnosis system, comprising:

[0037] A model building module, used to apply the above-mentioned bearing fault diagnosis model building method to build a bearing fault diagnosis model;

[0038] A model training module is used to perform transfer learning training on the bearing fault diagnosis model to obtain a trained bearing fault diagnosis model;

[0039] The fault diagnosis module is used to obtain the operating vibration signal of the bearing to be detected, and input it into the trained bearing fault diagnosis model to obtain the fault diagnosis result of the bearing to be detected.

[0040] The present invention also provides a non-transitory computer-readable storage medium on which computer instructions are stored. The computer instructions enable a computer to execute the above-mentioned bearing fault diagnosis method.

[0041] The present invention also provides an electronic device, comprising: a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus, and the processor calls the logic instructions in the memory to execute the above-mentioned bearing fault diagnosis method.

[0042] The present invention finally provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer executes the above-mentioned bearing fault diagnosis method.

[0043] Beneficial effects: The present invention constructs the objective function through the cross entropy loss of the neural network in the validation set, and uses the variant gray wolf algorithm to optimize the neural network hyperparameters. The constructed fault diagnosis model is insensitive to the initial value and is not easy to fall into the local optimal solution, and has better efficiency and accuracy.

[0044] The present invention uses the optimized CNN-BiLSTM network as a pre-training model and transplants it to new working conditions for fault diagnosis. Even when there are few training samples, the present invention can achieve high fault diagnosis accuracy, effectively improve the performance and application effect of fault diagnosis, and is particularly suitable for scenarios where training samples are scarce under multiple working conditions. Multiple comprehensive experiments have verified the excellent performance of the scheme, effectively solving the problem of model training and fault diagnosis under conditions of scarce samples. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 is a flow chart of a method for constructing a bearing fault diagnosis model in an embodiment of the present invention;

[0046] Figure 2 It is a multi-operating-condition high-precision bearing fault diagnosis framework in an embodiment of the present invention. DETAILED DESCRIPTION

[0047] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings in the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments, and they should not be understood as limitations on the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. In the description of the present invention, it should be understood that the terms used are only for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0048] Combine the following Figure 1-Figure 2 The present invention describes a bearing fault diagnosis model construction method, a bearing fault diagnosis method and a system.

[0049] Embodiment 1: This embodiment provides a method for constructing a bearing fault diagnosis model, such as Figure 1 Shown include:

[0050] Collect vibration signals of rolling bearings under different working conditions and construct a sample set;

[0051] Construct a CNN-BiLSTM neural network for rolling bearing fault diagnosis, use the sample set to train, verify and test the neural network, and obtain classification diagnosis results;

[0052] The objective function is constructed according to the cross entropy loss of the neural network in the validation set, and the objective function is optimized using the mutant gray wolf algorithm to obtain the optimal combination of neural network weights and hyperparameters; and the neural network model is updated with the optimal combination of neural network weights and hyperparameters as the bearing fault diagnosis model.

[0053] The focus of this embodiment is to use the mutant gray wolf algorithm to optimize the objective function and obtain the optimal combination of neural network weights and hyperparameters.

[0054] The Gray Wolf Algorithm (GWO) is an intelligent optimization algorithm that simulates wolf hunting behavior. Although GWO performs well in solving many complex optimization problems, in the final stage of the GWO algorithm, a large number of gray wolf individuals approaching the decision layer will reduce the diversity of the population. If a local optimal solution appears in the decision layer at this time, GWO will converge prematurely. In order to overcome these problems, a hybrid algorithm based on the Gray Wolf Algorithm and the Genetic Algorithm, the HGWA, is proposed in this embodiment.

[0055] The purpose of the variant gray wolf algorithm is to obtain the optimal combination of neural network weights and hyperparameters. The overall idea is:

[0056] Construct a real number vector based on the hyperparameter range and threshold range in the initial neural network model;

[0057] Substitute the real number vector into the objective function to obtain the fitness value; compare the fitness value with the upper and lower thresholds of the updated neural network model to obtain the optimal combination of neural network weights and hyperparameters.

[0058] The objective function here is expressed as:

[0059] F(a)=(L val (a),A val (a))

[0060]

[0061] a = {convolution kernel size, number of filters, number of BiLSTM hidden layer units, learning rate}

[0062] Among them, N val is the number of samples; C is the number of categories; y ic is the true label of the i-th sample in category c; is the predicted probability of the i-th sample in category c; L val (a) represents the loss of model a on the validation set; A val (a) represents the accuracy of model a on the validation set; Refers to the index of the category with the highest probability among the category probabilities predicted by the model; max c y ic refers to the index of the correct category in the true label.

[0063] The specific process of optimizing the objective function using the mutant gray wolf algorithm is as follows:

[0064] Step 1: After the population is initialized, genetic operations and mutation operations are applied to optimize individual selection. First, individuals in the current population are recombined and crossovered through genetic operations to generate new individuals.

[0065] 1) Cross

[0066] New individuals are generated by exchanging some genes of two parent individuals. The crossover strategy formula is as follows:

[0067]

[0068] Among them, k is the crossover point, N is the individual length. X1, X2 are the parents, X' c1 , X′ c2 For offspring.

[0069] 2) Mutation

[0070] Randomly change some genes of new individuals to introduce diversity. The mutation strategy formula is:

[0071]

[0072] Among them, Random(bounds[j]) means randomly generating a new value within the gene value range j; Represents the proxy original location.

[0073] Step 2: Evaluate the algorithm fitness, put the value of the decision variable (solution vector) into the user-defined fitness function, calculate the position fitness of each agent, and select the one with the highest fitness as the decision layer. The position update formula is as follows:

[0074]

[0075] Where t represents the current iteration number, A and C are coefficient vectors representing the location of the prey, and X(t) represents the location of the agent. is the distance between the agent and the target.

[0076] The coefficient vectors A and C are calculated as follows:

[0077]

[0078] In the formula, and are random vectors in [0, 1], and as t increases, the coefficient a gradually decreases from 2 to 0. Decide when Represents the search agent switching from the exploration phase to the hunting phase. The calculation is as follows:

[0079]

[0080] Step 3: Perform arithmetic crossover on the decision layer. For three individuals randomly selected from the population and and individuals in decision-making and The new position update formula is:

[0081]

[0082] Where a, θ are random numbers between (0, 1) and are updated in each iteration.

[0083] Step 4: To further improve the local search capability, refer to the crossover operation in step 1. and Perform crossover operation, the formula is as follows:

[0084]

[0085] Step 5: The mutation operation introduces randomness to increase the diversity of the population and escape from the local optimal solution. The mutation formula is:

[0086]

[0087] Here, ifrand() is a random number in the range [0,1].

[0088] Step 6: Select the three individuals with the highest fitness from the current population and the newly generated population as new α, β and δ individuals:

[0089]

[0090] Step 7: The decision layer updates the position, are the positions of all agents in the parent population, is the position of all agents in the new population, and the three most adaptable agents are retained as the leaders of the new generation. The final position update formula is:

[0091]

[0092] Step 8: The update formula for the final proxy position is:

[0093]

[0094] Step 9: Define an allowable small threshold between the last two consecutive results. The maximum number of iterations is considered as the stopping criterion. Save the current optimal solution and determine whether it meets the stopping criterion. If it meets the stopping criterion, output the optimal solution. Otherwise, repeat steps 2 to 8.

[0095] Embodiment 2: This embodiment provides a bearing fault diagnosis method, comprising:

[0096] Using the method for constructing a bearing fault diagnosis model in Example 1, a bearing fault diagnosis model is obtained;

[0097] The bearing fault diagnosis model is trained by transfer learning to obtain a trained bearing fault diagnosis model; Figure 2 As shown in Figure 1, the transfer learning of the model involves freezing the weights of the pre-trained model and only fine-tuning the weights of the output layer.

[0098] The running vibration signal of the bearing to be tested is input into the trained bearing fault diagnosis model to obtain the fault diagnosis result of the bearing to be tested.

[0099] Embodiment 3: This embodiment provides a bearing fault diagnosis system, including:

[0100] A model building module, used to apply the bearing fault diagnosis model building method of Example 1 to build a bearing fault diagnosis model;

[0101] A model training module is used to perform transfer learning training on the bearing fault diagnosis model to obtain a trained bearing fault diagnosis model;

[0102] The fault diagnosis module is used to obtain the operating vibration signal of the bearing to be detected, and input it into the trained bearing fault diagnosis model to obtain the fault diagnosis result of the bearing to be detected.

[0103] Embodiment 4: This embodiment provides a non-transitory computer-readable storage medium on which computer instructions are stored. The computer instructions enable a computer to execute the bearing fault diagnosis method of embodiment 2.

[0104] Embodiment 5: This embodiment provides an electronic device, which may include: a processor, a communications interface, a memory, and a communication bus, wherein the processor, the communications interface, and the memory communicate with each other via the communication bus. The processor may call the logic instructions in the memory to execute the bearing fault diagnosis method of embodiment 2.

[0105] In addition, the logic instructions in the above-mentioned memory can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0106] Embodiment 6: This embodiment provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the bearing fault diagnosis method of Embodiment 2.

[0107] In order to verify the technical effect that the present invention can produce, an experimental verification was conducted. The present invention experiment is based on a 64-bit Windows 11 operating system, the system processor is an Intel Core i5-12490KF@3.00GHz, the memory is 32GB; the development tool used is the PyCharm editor, and the deep learning framework is Keras 2.31.

[0108] (1) Prepare experimental data

[0109] The experiment was applied to a vibration dataset of Case Western Reserve University (CWRU) Bearing Center data. The CWRU dataset includes four different fault states: no fault (NF), ball fault (BF), inner race fault (IF), and outer race fault (OF). A discharge device was used to introduce faults on the surface of the bearing element, with fault diameters of 0.007 inches, 0.014 inches, and 0.021 inches. The bearing vibration signal was collected by an accelerometer mounted on the motor drive with a sampling frequency of 12kHz. As shown in Table 1, the motor was subjected to different load powers: 0HP, 1HP, 2HP, and 3HP. Under these operating conditions, four bearing fault datasets were generated and labeled A, B, C, and D respectively. Each operating condition has 10 fault states (including no fault as the only category) based on different damage levels at different locations on the bearing. Therefore, a set of 10 class labels from 0 to 9 were assigned to classify the different fault states.

[0110] Table 1 CWRU dataset

[0111]

[0112]

[0113] (2) Data preprocessing

[0114] In order to expand the data set and improve the classification accuracy of the model, overlapping sampling is used to sample the sequence signal. Each data segment contains 1024 data points with a step size of 150. 250 samples are collected for each fault state, so each data set has a total of 2500 data samples.

[0115] (3) Model training parameter setting

[0116] The search range of hyperparameters is set as follows: learning rate is 0.0001-0.01; convolution kernel size is 3-32; the number of filters is 3-128; the number of hidden units of BiLSTM is 3-128.

[0117] (4) Experimental results

[0118] The results of the CNN-BiLSTM network model used in the present invention are compared with those of the CNN and CNN-LSTM models, as shown in Table 2.

[0119] Table 2 Model comparison

[0120]

[0121]

[0122] Four algorithms are used to optimize the hyperparameters of the CNN-BiLST network. The four algorithms are HGWA, PSO, GS and RS. The results of optimizing the hyperparameters of the CNN-BiLST network are shown in Table 3.

[0123] Table 3 Network hyperparameter optimization results

[0124]

[0125]

[0126] Since the working environment of rolling bearings is complex and changeable, it is challenging to collect sufficient training data under different conditions. Therefore, in this section, we will study how to use model-based transfer learning techniques to diagnose bearing faults under multiple working conditions with limited training samples.

[0127] In the experiment, the data of one working condition in the CWUR dataset was used as the source domain, while the data of the remaining three conditions were used as the target domain. In order to verify the effectiveness of the proposed cross-working condition bearing fault diagnosis method, multiple comparative tests were carried out using different numbers of target domain training samples. For each target domain, 10, 20, 30 and 40 random samples were selected as training sets, while 500 samples were retained as test sets. The CNN-BiLSTM model optimized by HGWA was used as the pre-trained model for the target domain. The weights of all layers of the pre-trained model were frozen except for the input layer and the output layer. The weights of the output layer were then fine-tuned using the training set data. Finally, the performance of the fine-tuned model was evaluated using the test set. The training parameters of the experiment were as follows: 10 iterations, a batch size of 32, and a learning rate determined by the optimized pre-trained model and the Adam optimizer. In order to reduce the influence of randomness, all experimental results are the average of 10 independent trials.

[0128] The experimental results show that in the cross-condition fault diagnosis experiment with dataset A as the source domain and dataset D as the target domain, the average accuracy of 10 training samples is only 86.8%. The main reason for the low accuracy is that there are significant differences in data features. The source domain data represents no-load conditions, while the target domain data includes load conditions. In addition, the limited amount of training data is also one of the reasons for the low classification accuracy. As the number of training samples increases from 10 to 40, the diagnosis accuracy of the model increases from 86.8% to 99.76%. This trend is also confirmed in other experimental results. Among the 12 cross-condition fault diagnosis experiments, only one experiment did not exceed 99% classification accuracy when the number of training samples reached 30. However, when the number of training samples increased to 40, the classification accuracy of all experiments exceeded 99%. As the number of training samples increased, the average accuracy of the 12 experiments increased from 95.26% to 99.65%. These results further show that even with a limited number of training samples, model-based transfer learning can achieve high classification accuracy.

[0129] To evaluate the performance of the HGWA-optimized CNN-BiLSTM model on the JNU dataset, we conducted a preliminary experiment. The data preprocessing steps and hyperparameter search ranges were consistent with those used in the CWRU experiment. Given the increased complexity of the JNU training set, the number of training epochs was adjusted to 20. It is worth noting that the previous experiments on this model were conducted under a single working condition. To verify the performance of the HGWA-optimized CNN-BiLSTM model under multiple working conditions, this study trained and optimized the model using samples from different working conditions. This approach aims to verify the performance of the optimized model under different conditions. It is important to emphasize that the transition from single-condition training to multi-condition training is critical because it shows that the model has the ability to handle more complex and varied data, thereby demonstrating its robustness and adaptability.

[0130] The present invention designs an optimization algorithm comparison test and a multi-operating condition diagnosis test under limited samples. The optimization algorithm comparison test uses HPSO, GS, RS and HGWA to optimize the CNN-BiLSTM network hyperparameters respectively. The experimental results show that the performance of HGWA and the ability to find the global optimal solution are better than the other three optimization algorithms, proving the rationality of the optimization algorithm of the present invention; in the multi-operating condition diagnosis test under limited samples, the problem of few bearing fault samples is solved by using the transfer learning of the model under limited data samples, proving the generalization ability and effectiveness of the optimized network.

[0131] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0132] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0133] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for constructing a bearing fault diagnosis model, characterized in that: include: Collect vibration signals of rolling bearings under different working conditions and construct a sample set; Construct a CNN-BiLSTM neural network for rolling bearing fault diagnosis, use the sample set to train, verify and test the neural network, and obtain classification diagnosis results; The objective function is constructed according to the cross entropy loss of the neural network in the validation set, and the objective function is optimized using the mutant gray wolf algorithm to obtain the optimal combination of neural network weights and hyperparameters; and the neural network model is updated with the optimal combination of neural network weights and hyperparameters as the bearing fault diagnosis model.

2. The method for constructing a bearing fault diagnosis model according to claim 1, characterized in that: The objective function is optimized by using the mutant gray wolf algorithm to obtain the optimal combination of neural network weights and hyperparameters; include: Construct a real number vector based on the hyperparameter range and threshold range in the initial neural network model; Substitute the real number vector into the objective function to obtain the fitness value; compare the fitness value with the upper and lower thresholds of the updated neural network model to obtain the optimal combination of neural network weights and hyperparameters.

3. The method for constructing a bearing fault diagnosis model according to claim 2, characterized in that: The objective function is expressed as: F(a)=(L val (a),A val (a)) a = {convolution kernel size, number of filters, number of BiLSTM hidden layer units, learning rate} Among them, N val is the number of samples; C is the number of categories; y ic is the true label of the i-th sample in category c; is the predicted probability of the i-th sample in category c; L val (a) represents the loss of model a on the validation set; A val (a) represents the accuracy of model a on the validation set; The index of the category with the highest probability among the category probabilities predicted by the model; refers to the index of the correct category in the true label.

4. The method for constructing a bearing fault diagnosis model according to claim 1, characterized in that: The specific process of optimizing the objective function using the mutant gray wolf algorithm is as follows: (1) Initialize the population and then optimize individual selection through genetic operations and mutation operations; (2) Evaluate the algorithm fitness and select the one with the highest fitness as the decision layer; (3) Perform arithmetic crossover operations on the decision layer; (4) Introducing randomness through mutation operations to increase the diversity of the population; (5) Select the three individuals with the highest fitness from the current population and the newly generated population as new leaders; (6) The decision-making level updates its position and retains new leadership individuals; (7) Update the proxy location; (8) Define an allowable small threshold between the last two consecutive results. The maximum number of iterations is considered as the stopping criterion. Save the current optimal solution and determine whether it meets the stopping criterion. If it meets the stopping criterion, output the optimal solution. Otherwise, repeat steps (1) to (7).

5. The method for constructing a bearing fault diagnosis model according to claim 1, characterized in that: The CNN-BiLSTM neural network includes a convolutional network layer and a bidirectional long short-term memory network layer.

6. The method for constructing a bearing fault diagnosis model according to claim 1, characterized in that: The hyperparameters of the neural network include learning rate, convolution kernel size, number of filters and number of hidden units of BiLSTM.

7. A bearing fault diagnosis method, characterized in that: include: Using the method for constructing a bearing fault diagnosis model according to any one of claims 1 to 6, a bearing fault diagnosis model is obtained; Perform transfer learning training on the bearing fault diagnosis model to obtain a trained bearing fault diagnosis model; The running vibration signal of the bearing to be tested is input into the trained bearing fault diagnosis model to obtain the fault diagnosis result of the bearing to be tested.

8. The bearing fault diagnosis method according to claim 7, characterized in that: Transfer learning of the model involves freezing the weights of the pre-trained model and only fine-tuning the weights of the output layer.

9. A bearing fault diagnosis system, characterized in that: include: A model building module, used to apply the method for building a bearing fault diagnosis model according to any one of claims 1 to 6 to build a bearing fault diagnosis model; A model training module is used to perform transfer learning training on the bearing fault diagnosis model to obtain a trained bearing fault diagnosis model; The fault diagnosis module is used to obtain the operating vibration signal of the bearing to be detected, and input it into the trained bearing fault diagnosis model to obtain the fault diagnosis result of the bearing to be detected.

Images