A rotating machinery fault diagnosis method based on AHO-VMD-CNN-BiGRU

Abstract

This invention provides an AHO-VMD-CNN-BiGRU method for diagnosing rotating machinery faults, belonging to the field of intelligent fault diagnosis technology. The method includes: acquiring vibration signals of rotating machinery under normal conditions and different fault types; preprocessing the vibration signals using variational mode decomposition, decomposing and removing noise to reconstruct new signals; converting the new vibration signals into a two-dimensional time-frequency image using short-time Fourier transform; optimizing the hyperparameters of a convolutional neural network (CNN)-bidirectional gated recurrent unit neural network (BiGRU) using the African Hyena Swarm Optimization (AHO) algorithm; and inputting the two-dimensional time-frequency image into the optimized CNN-BiGRU model to achieve rotating machinery fault diagnosis. This invention has significant practical implications for improving the accuracy of rotating machinery fault diagnosis and guiding equipment maintenance.

Description

Technical Field

[0001] This invention relates to the field of intelligent fault diagnosis technology, and more specifically, to a rotating machinery fault diagnosis method based on AHO-VMD-CNN-BiGRU. Background Technology

[0002] Rotating machinery, as a key component of industrial equipment systems, directly determines the continuous operation capability and overall reliability of production lines. Equipment failure not only leads to unplanned downtime and increased maintenance costs but can also trigger cascading production interruptions and even endanger personnel safety. Therefore, developing accurate and efficient intelligent fault diagnosis methods has significant engineering value.

[0003] In the context of Industry 4.0 and intelligent operation and maintenance, data-driven fault diagnosis methods have gradually become the mainstream research approach. Deep learning models, in particular, are widely used for fault identification and classification of rotating machinery due to their ability to automatically extract sensitive features from complex vibration signals. A common technical approach is to convert one-dimensional time-series vibration signals into two-dimensional time-frequency maps using short-time Fourier transform or continuous wavelet transform, and then employ convolutional neural networks and their improved architectures for feature extraction and fault classification. However, the diagnostic performance of such methods is largely constrained by the network structure and hyperparameter selection. Traditional parameter tuning methods relying on human experience are not only time-consuming but also fail to guarantee that the model is in its optimal state. To address these issues, systematically exploring the optimal hyperparameter combination by constructing a parameter optimization objective function and combining it with intelligent optimization algorithms has become a research hotspot. However, existing intelligent optimization algorithms still have room for improvement in terms of optimization accuracy and computational efficiency. Summary of the Invention

[0004] The main objective of this invention is to address the shortcomings of existing technologies and provide an AHO-VMD-CNN-BiGRU method for fault diagnosis of rotating machinery. First, Variational Model Decomposition (VMD) is used for signal denoising preprocessing. Then, a Convolutional Neural Network (CNN)-Bidirectional Gated Recurrent Unit (BiGRU) model optimized using the African Hyena Optimizer (AHO) algorithm is established. This invention innovatively proposes a novel African Hyena Optimizer (AHO) algorithm and uses it to optimize the hyperparameters of the CNN-BiGRU, resulting in a lower iterative loss value and higher diagnostic accuracy, providing reliable technical support for improving the accuracy of fault diagnosis in rotating machinery.

[0005] In nature, African hyena groups have developed highly structured social systems, with their hunting behaviors and hierarchical mechanisms jointly ensuring the survival and efficiency of the population. Individuals within a hyena group differ in size, strength, and agility: smaller, more agile young or female hyenas can quickly locate and harass prey, but due to limited strength, they often struggle to kill alone; while larger, stronger adult hyenas take on the task of subduing the prey, effectively preventing injured prey from escaping. This division of labor and cooperation fully utilizes the differences among individuals within the group, increasing the overall hunting success rate. Furthermore, the hyena group has a clear hierarchy, a social structure that significantly reduces internal conflict and enhances cooperative efficiency. During periods of severe resource scarcity, the worst-performing and least cooperative teams within the hyena population are naturally eliminated or disbanded, with individuals integrating into other groups or forming new ones. This mechanism not only effectively reduces resource consumption within the population but also promotes continuous optimization of the group structure, thereby enhancing the overall adaptability and survival ability of the population in harsh environments. The African hyena pack optimization algorithm proposed in this paper is a metaheuristic swarm intelligence optimization algorithm inspired by the competition mechanism, social hierarchy, division of labor and cooperation, and team elimination and reorganization strategies within the aforementioned group.

[0006] This invention is implemented as follows:

[0007] The technical solution to achieve the purpose of this invention is: a rotating machinery fault diagnosis method based on AHO-VMD-CNN-BiGRU, comprising the following steps:

[0008] Step 1: Obtain the original vibration signals of the rotating machinery under normal operating conditions and different fault types as training samples, and label the corresponding fault categories;

[0009] Step 2: The original vibration signal is preprocessed for noise reduction using variational mode decomposition. The denoising reconstructed signal is obtained by decomposition, filtering and reconstruction.

[0010] Step 3: Convert the reconstructed signal into a two-dimensional time-frequency image using a short-time Fourier transform;

[0011] Step 4: Set the basic parameters of the African hyena pack optimization algorithm;

[0012] Step 5: Use the African hyena pack optimization algorithm to automatically optimize the hyperparameters of the convolutional neural network-bidirectional gated recurrent unit network model;

[0013] Step 6: Based on the optimized hyperparameters, combined with the reconstructed signal and its corresponding fault label, train a rotating machinery fault diagnosis model based on a convolutional neural network-bidirectional gated recurrent unit network model;

[0014] Step 7: The vibration signal to be diagnosed is denoised using the variational mode decomposition method described in Step 2. The denoised signal is then converted into a time-frequency image using the short-time Fourier transform described in Step 3. The time-frequency image is then input into the trained convolutional neural network-bidirectional gated recurrent unit network model to realize the identification of the fault type of rotating machinery.

[0015] Furthermore, in step two, the noise reduction preprocessing includes the following steps: First, the key parameters required for variational mode decomposition noise reduction are set, including the number of intrinsic mode functions and the penalty factor; then, the variational mode decomposition algorithm is applied to decompose the original vibration signal to obtain several intrinsic mode function components; next, each intrinsic mode function component is arranged from low to high according to its center frequency, and its correlation coefficient with the original vibration signal is calculated, with the mode component with the lowest correlation being marked as a noise component; finally, all marked noise components are removed, and the remaining components are superimposed and reconstructed to obtain the denoised signal.

[0016] Furthermore, in step four, the African hyena pack optimization algorithm consists of five steps executed sequentially:

[0017] Step S41: During the initialization phase, randomly generate... Y A hyena, its expression is as follows:

[0018] X k = lb+(ub-lb)×rand (0,1)

[0019] in, X k Representing the k The location of a single hyena. k =1,2,… Y , lb This is the lower bound of the search domain. ub This is the upper limit of the search domain. X k , lb , ub Both are N-dimensional vectors, where N represents the dimension of the problem being solved. rand (0,1) represents an N-dimensional vector composed of random numbers between 0 and 1;

[0020] Step S42: In the team building and leader election phase, hyena individuals randomly form hunting teams, and the fitness value of each individual in each team is calculated. The hyena individual with the best fitness value is selected as the leader of the current hunting team. The number of members in each team is... Ts It is dynamically adjusted, and its formula is as follows:

[0021] Ts =max( Tsmin ,min( Ts max ,round(3+2sin(2π× t / T ))))

[0022] like Y / Ts If the number of elements is not divisible, the remaining individuals form a team; the `round()` function rounds the value within the parentheses to the nearest integer. t This represents the current iteration number. T The maximum number of iterations, Ts max and Ts min Representing the maximum and minimum team size; Step S43: During the team encirclement phase, the non-fittest individual in each team will move towards the location of the team leader; each team leader will conduct a more detailed local search around itself; the update formula for this phase is as follows:

[0023] X i,m t+1 = X i,m t +C(0,1)× α t ×( X i,l t - X i,m t ) F i,m t ≠ F i,l t

[0024] X i,l t+1 = X i,l t +C(0,1)×( X i,l t -h×X i,m t ) F i,m t = F i,l t

[0025] in, Xi,m t+1 and X i,m t They represent the first t+1 Second and third t In the nth iteration i The first team m The position of each member X i,l t+1 and X i,l t The first t +1 and t In the nth iteration i The position of leader in a team F i,m t and F i,l t They represent the first t In the nth iteration i The first team m The fitness value of each member and the team leader's position. C (0,1) is the N-dimensional Cauchy perturbation vector. α t and h Let represent the step size when the non-optimal individual and the leader move, respectively, with the following expressions:

[0026] α t = α max -( α max - α min )× t / T

[0027] h =2cos( r )

[0028] in, α max For the maximum step size, α min For the minimum step size, r A random number uniformly distributed in (0,1);

[0029] Step S44: During the team transition phase, after each D In the next iteration, the team with the worst average fitness value is removed from all teams, and individuals are regenerated within the search range according to the following formula:

[0030]

[0031]

[0032] in, X w,m t+1 Indicates the first t+1 The worst team in the next iteration m The position of each member X w,l t+1 and X w,,l t They represent the first t+1 Second and third t The position of the leader in the worst team in the next iteration rand ( X i,l t ) indicates from the first t The position of a leader randomly selected from among the team leaders in the next iteration; For random Gaussian perturbation terms, ;

[0033] Step S45: Determine whether the current iteration count has reached the maximum iteration count. If so, complete the iteration and end the search, outputting the optimal position. Otherwise, return to step S42 and repeat steps S42 to S45.

[0034] Furthermore, in step four, the parameters required for the African hyena pack optimization algorithm include population size. Y Maximum number of iterations T Maximum step size α max Minimum step size α min Largest scale Ts max and minimum size Ts min Team turnover and iteration count D wait.

[0035] Furthermore, in step five, the hyperparameters of the convolutional neural network-bidirectional gated recurrent unit model are automatically optimized using the African hyena swarm optimization algorithm. The optimized hyperparameters include the number of neurons in the convolutional layer, the size of the convolutional kernel, and the number of neurons in the bidirectional gated recurrent unit layer.

[0036] Furthermore, in step six, the convolutional neural network-bidirectional gated recurrent unit network model includes four convolutional layers, three pooling layers, one flattening layer, one bidirectional gated recurrent unit layer, and one output layer. The activation function of the convolutional layers is the ReLU function, the stride is (2, 2), and zero padding is used. The pooling layer size is 2×2. The output layer outputs 4, and the activation function is the softmax function.

[0037] The beneficial effects of this invention are as follows: It proposes a novel metaheuristic swarm intelligence optimization algorithm—the African Hyena Group Optimization Algorithm (AHO). Inspired by the highly structured social system of hyena groups, this algorithm improves optimization performance by simulating their division of labor, cooperation, and hierarchical mechanisms. Unlike existing algorithms, AHO innovatively introduces a team elimination and reorganization mechanism based on group hierarchy. This mechanism simulates the natural elimination process of the worst-performing and least efficient teams in a hyena population during periods of resource scarcity. The algorithm periodically removes teams with the worst average fitness and reorganizes them into new teams, effectively enhancing population diversity and significantly improving the algorithm's ability to escape local optima. Furthermore, this invention applies the AHO algorithm to the hyperparameter optimization of the CNN-BiGRU model. The optimized CNN-BiGRU can more effectively extract fault features and accurately identify fault types with lower iteration loss, thereby significantly improving the accuracy of rotating machinery fault diagnosis. Attached Figure Description

[0038] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained from these drawings without creative effort.

[0039] Figure 1 This is a flowchart of an intelligent fault diagnosis method for rotating machinery provided by an embodiment of the present invention;

[0040] Figure 2 This is a detailed flowchart of the AHO optimization algorithm provided in the embodiments of the present invention;

[0041] Figure 3 This is a topology diagram of the CNN-BiGRU network model;

[0042] Figure 4 This is a confusion matrix diagram of the fault diagnosis results of the AHO-VMD-CNN-BiGRU method;

[0043] Figure 5 This is a comparison chart of iterative losses for different methods;

[0044] Figure 6This is a comparison chart of the accuracy of different iterative methods. Detailed Implementation

[0045] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the implementation examples and accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to represent selected embodiments of the invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0046] The method of the present invention will be illustrated below using a specific case of rotating machinery fault diagnosis as an example.

[0047] like Figure 1 A rotating machinery fault diagnosis method using AHO-VMD-CNN-BiGRU includes the following steps:

[0048] Step 1: Obtain the original vibration signals of the rotating machinery under normal operating conditions and different fault types as training samples, and label the corresponding fault categories;

[0049] Step 2: The original vibration signal is preprocessed for noise reduction using variational mode decomposition. The denoising reconstructed signal is obtained by decomposition, filtering and reconstruction.

[0050] Step 3: Convert the reconstructed signal into a two-dimensional time-frequency image using a short-time Fourier transform;

[0051] Step 4: Set the basic parameters of the African hyena pack optimization algorithm;

[0052] Step 5: Use the African hyena pack optimization algorithm to automatically optimize the hyperparameters of the convolutional neural network-bidirectional gated recurrent unit network model;

[0053] Step 6: Based on the optimized hyperparameters, combined with the reconstructed signal and its corresponding fault label, train a rotating machinery fault diagnosis model based on a convolutional neural network-bidirectional gated recurrent unit network model;

[0054] Step 7: The vibration signal to be diagnosed is denoised using the variational mode decomposition method described in Step 2. The denoised signal is then converted into a time-frequency image using the short-time Fourier transform described in Step 3. The time-frequency image is then input into the trained convolutional neural network-bidirectional gated recurrent unit network model to realize the identification of the fault type of rotating machinery.

[0055] Furthermore, in step two, the noise reduction preprocessing includes the following steps: First, the key parameters required for variational mode decomposition noise reduction are set, including the number of intrinsic mode functions and the penalty factor; then, the variational mode decomposition algorithm is applied to decompose the original vibration signal to obtain several intrinsic mode function components; next, each intrinsic mode function component is arranged from low to high according to its center frequency, and its correlation coefficient with the original vibration signal is calculated, with the mode component with the lowest correlation being marked as a noise component; finally, all marked noise components are removed, and the remaining components are superimposed and reconstructed to obtain the denoised signal.

[0056] Furthermore, in step four, the African hyena pack optimization algorithm consists of five steps executed sequentially:

[0057] Step S41: During the initialization phase, randomly generate... Y A hyena, its expression is as follows:

[0058] X k = lb+(ub-lb)×rand (0,1)

[0059] in, X k Representing the k The location of a single hyena. k =1,2,… Y , lb This is the lower bound of the search domain. ub This is the upper limit of the search domain. X k , lb , ub Both are N-dimensional vectors, where N represents the dimension of the problem being solved. rand (0,1) represents an N-dimensional vector composed of random numbers between 0 and 1;

[0060] Step S42: In the team building and leader election phase, hyena individuals randomly form hunting teams, and the fitness value of each individual in each team is calculated. The hyena individual with the best fitness value is selected as the leader of the current hunting team. The number of members in each team is... Ts It is dynamically adjusted, and its formula is as follows:

[0061] Ts =max( Tsmin ,min( Ts max ,round(3+2sin(2π× t / T ))))

[0062] like Y / Ts If the number of elements is not divisible, the remaining individuals form a team; the `round()` function rounds the value within the parentheses to the nearest integer. t This represents the current iteration number. T The maximum number of iterations, Ts max and Ts min Representing the maximum and minimum team size; Step S43: During the team encirclement phase, the non-fittest individual in each team will move towards the location of the team leader; each team leader will conduct a more detailed local search around itself; the update formula for this phase is as follows:

[0063] X i,m t+1 = X i,m t +C(0,1)× α t ×( X i,l t - X i,m t ) F i,m t ≠ F i,l t

[0064] X i,l t+1 = X i,l t +C(0,1)×( X i,l t -h×X i,m t ) F i,m t = F i,l t

[0065] in, Xi,m t+1 and X i,m t They represent the first t+1 Second and third t In the nth iteration i The first team m The position of each member X i,l t+1 and X i,l t The first t +1 and t In the nth iteration i The position of leader in a team F i,m t and F i,l t They represent the first t In the nth iteration i The first team m The fitness value of each member and the team leader's position. C (0,1) is the N-dimensional Cauchy perturbation vector. α t and h Let represent the step size when the non-optimal individual and the leader move, respectively, with the following expressions:

[0066] α t = α max -( α max - α min )× t / T

[0067] h =2cos( r )

[0068] in, α max For the maximum step size, α min For the minimum step size, r A random number uniformly distributed in (0,1);

[0069] Step S44: During the team transition phase, after each D In the next iteration, the team with the worst average fitness value is removed from all teams, and individuals are regenerated within the search range according to the following formula:

[0070]

[0071]

[0072] in, X w,m t+1 Indicates the first t+1 The worst team in the next iteration m The position of each member X w,l t+1 and X w,,l t They represent the first t+1 Second and third t The position of the leader in the worst team in the next iteration rand ( X i,l t ) indicates from the first t The position of a leader randomly selected from among the team leaders in the next iteration; For random Gaussian perturbation terms, ;

[0073] Step S45: Determine whether the current iteration count has reached the maximum iteration count. If so, complete the iteration and end the search, outputting the optimal position. Otherwise, return to step S42 and repeat steps S42 to S45.

[0074] Furthermore, in step four, the parameters required for the African hyena pack optimization algorithm include population size. Y Maximum number of iterations T Maximum step size α max Minimum step size α min Largest scale Ts max and minimum size Ts min Team turnover and iteration count D wait.

[0075] Furthermore, in step five, the hyperparameters of the convolutional neural network-bidirectional gated recurrent unit model are automatically optimized using the African hyena swarm optimization algorithm. The optimized hyperparameters include the number of neurons in the convolutional layer, the size of the convolutional kernel, and the number of neurons in the bidirectional gated recurrent unit layer.

[0076] Furthermore, in step six, the convolutional neural network-bidirectional gated recurrent unit network model includes four convolutional layers, three pooling layers, one flattening layer, one bidirectional gated recurrent unit layer, and one output layer. The activation function of the convolutional layers is the ReLU function, the stride is (2, 2), and zero padding is used. The pooling layer size is 2×2. The output layer outputs 4, and the activation function is the softmax function.

[0077] The method of the present invention will be illustrated below using a specific case of rotating machinery failure as an example.

[0078] The experimental data comes from the public dataset of the CWRU Bearing Data Center, and the experimental data information is shown in Table 1.

[0079] Table 1 Experimental Data Information Table

[0080]

[0081] The specific implementation process is as follows:

[0082] like Figure 1 A rotating machinery fault diagnosis method using AHO-VMD-CNN-BiGRU includes the following steps:

[0083] Step 1: Obtain the original vibration signals of the rotating machinery under normal conditions and different fault types as training samples, and label the corresponding fault categories. In this embodiment, 70% of the total number of samples are selected as training samples, 10% as verification samples, and 20% as test samples.

[0084] Step 2: The original vibration signal is preprocessed for noise reduction using variational mode decomposition. The denoising reconstructed signal is obtained by decomposition, filtering and reconstruction.

[0085] Step 3: Convert the reconstructed signal into a two-dimensional time-frequency image using a short-time Fourier transform;

[0086] Step 4: Set the basic parameters of the African hyena pack optimization algorithm, including population size. Y =30, Maximum number of iterations T =350, maximum step size α max =2, minimum step size α min =0.1, maximum scale Ts max =5 and minimum size Ts min =3 Team turnover iterations D =2, the flowchart of the optimization algorithm for African hyena packs is as follows: Figure 2 As shown;

[0087] Step 5: The hyperparameters of the convolutional neural network-bidirectional gated recurrent unit model are automatically optimized using the African hyena pack optimization algorithm. The optimized hyperparameters include the number of neurons in the convolutional layer, the size of the convolutional kernel, and the number of neurons in the bidirectional gated recurrent unit layer. The evaluation results after optimization are shown in [link to optimization]. Figure 4 ;

[0088] Step Six: Based on the optimized hyperparameters, and combined with the reconstructed signals and their corresponding fault labels, train a rotating machinery fault diagnosis model based on a convolutional neural network-bidirectional gated recurrent unit network model; the topology diagram of the convolutional neural network-bidirectional gated recurrent unit network model is shown below. Figure 3 As shown;

[0089] Table 2 Comparison of Model Accuracy and Loss Value

[0090]

[0091] Step 7: The vibration signal to be diagnosed is denoised using the variational mode decomposition method described in Step 2. The denoised signal is then converted into a time-frequency image using the short-time Fourier transform described in Step 3. The time-frequency image is then input into the trained convolutional neural network-bidirectional gated recurrent unit network model to realize the identification of the fault type of rotating machinery.

[0092] To evaluate the advantages of the model, multiple models, including CNN, CNN-BiGRU, and VMD-CNN-BiGRU, were compared with the method of this invention. Accuracy and F1 score were used to evaluate the diagnostic results, and the results are shown in Table 2. Figure 5 , Figure 6 As shown in Table 2, the method of the present invention significantly outperforms the comparative model in both accuracy and F1 score, demonstrating the advantages of the method of the present invention in the diagnosis of rotating machinery faults.

[0093] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A rotating machinery fault diagnosis method based on AHO-VMD-CNN-BiGRU, characterized in that, Includes the following steps: Step 1: Obtain the original vibration signals of the rotating machinery under normal operating conditions and different fault types as training samples, and label the corresponding fault categories; Step 2: The original vibration signal is preprocessed for noise reduction using variational mode decomposition. The denoising reconstructed signal is obtained by decomposition, filtering and reconstruction. Step 3: Convert the reconstructed signal into a two-dimensional time-frequency image using a short-time Fourier transform; Step 4: Set the basic parameters of the African hyena pack optimization algorithm; Step 5: Use the African hyena pack optimization algorithm to automatically optimize the hyperparameters of the convolutional neural network-bidirectional gated recurrent unit network model; Step 6: Based on the optimized hyperparameters, combined with the reconstructed signal and its corresponding fault label, train a rotating machinery fault diagnosis model based on a convolutional neural network-bidirectional gated recurrent unit network model; Step 7: The vibration signal to be diagnosed is denoised using the variational mode decomposition method described in Step 2. The denoised signal is then converted into a time-frequency image using the short-time Fourier transform described in Step 3. The time-frequency image is then input into the trained convolutional neural network-bidirectional gated recurrent unit network model to realize the identification of the fault type of rotating machinery. In step four, the African hyena pack optimization algorithm consists of five steps executed sequentially: Step S41: During the initialization phase, randomly generate... Y A hyena, its expression is as follows: X k = lb+(ub-lb)×rand (0,1) in, X k Representing the k The location of a single hyena. k =1,2,… Y , lb This is the lower bound of the search domain. ub This is the upper limit of the search domain. X k , lb , ub All N A dimensional vector, where N Represents the dimension of the problem to be solved. rand (0,1) represents a random number between 0 and 1. N dimensional vector; Step S42: In the team building and leader election phase, hyena individuals randomly form hunting teams, and the fitness value of each individual in each team is calculated. The hyena individual with the best fitness value is selected as the leader of the current hunting team. The number of members in each team is... Ts It is dynamically adjusted, and its formula is as follows: Ts =max( Ts min ,min( Ts max ,round(3+2sin(2π× t / T )))) like Y / Ts If the number of elements is not divisible, the remaining individuals form a team; the `round()` function rounds the value within the parentheses to the nearest integer. t This represents the current iteration number. T The maximum number of iterations, Ts max and Ts min Represents the largest and smallest team sizes; Step S43: During the team encirclement phase, the non-fittest individual in each team will move towards the location of the team leader; each team leader will then conduct a more detailed local search around itself; the update formula for this phase is as follows: X i,m t+1 = X i,m t +C(0,1)× α t ×( X i,l t - X i,m t ) F i,m t ≠ F i,l t X i,l t+1 = X i,l t +C(0,1)×( X i,l t -h×X i,m t ) F i,m t = F i,l t in, X i,m t+1 and X i,m t They represent the first t+1 Second and third t In the nth iteration i The first team m The position of each member X i,l t+1 and X i,l t The first t +1 and t In the nth iteration i The position of leader in a team F i,m t and F i,l t They represent the first t In the nth iteration i The first team m The fitness value of each member and the team leader's position. C (0,1) is N Vicosy perturbation vector α t and h Let represent the step size when the non-optimal individual and the leader move, respectively, with the following expressions: α t = α max -( α max - α min )× t / T h =2cos( r ) in, α max For the maximum step size, α min For the minimum step size, r A random number uniformly distributed in (0,1); Step S44: During the team transition phase, after each D In the next iteration, the team with the worst average fitness value is removed from all teams, and individuals are regenerated within the search range according to the following formula: in, X w,m t+1 Indicates the first t+1 The worst team in the next iteration m The position of each member X w,l t+1 and X w,l t They represent the first t+1 Second and third t The position of the leader in the worst team in the next iteration rand ( X i,l t ) indicates from the first t The position of a leader randomly selected from among the team leaders in the next iteration; For random Gaussian perturbation terms, ; Step S45: Determine whether the current iteration count has reached the maximum iteration count. If so, complete the iteration and end the search, outputting the optimal position. Otherwise, return to step S42 and repeat steps S42 to S45.

2. The rotating machinery fault diagnosis method based on AHO-VMD-CNN-BiGRU as described in claim 1, characterized in that, In step two, the noise reduction preprocessing includes the following steps: First, the key parameters required for variational mode decomposition denoising are set, including the number of intrinsic mode functions and the penalty factor; Subsequently, the variational mode decomposition algorithm is applied to decompose the original vibration signal to obtain several intrinsic mode function components. Then, each intrinsic mode function component is arranged in order of its center frequency from low to high, and its correlation coefficient with the original vibration signal is calculated. The mode component with the lowest correlation is marked as a noise component. Finally, all marked noise components are removed, and the remaining components are superimposed and reconstructed to obtain the denoised signal.

3. The rotating machinery fault diagnosis method based on AHO-VMD-CNN-BiGRU as described in claim 1, characterized in that, In step four, the parameters required for the African hyena pack optimization algorithm include the total number of individuals. Y The largest team Ts max Minimum size Ts min Maximum step size α max minimum step size α min Maximum number of iterations T Team turnover and iteration count D .

4. The rotating machinery fault diagnosis method based on AHO-VMD-CNN-BiGRU as described in claim 1, characterized in that, In step five, the hyperparameters of the convolutional neural network-bidirectional gated recurrent unit model are automatically optimized using the African hyena swarm optimization algorithm. The optimized hyperparameters include the number of neurons in the convolutional layer, the size of the convolutional kernel, and the number of neurons in the bidirectional gated recurrent unit layer.

5. The rotating machinery fault diagnosis method based on AHO-VMD-CNN-BiGRU as described in claim 1, characterized in that, In step six, the convolutional neural network-bidirectional gated recurrent unit network model includes four convolutional layers, three pooling layers, one flattening layer, one bidirectional gated recurrent unit layer, and one output layer. The activation function of the convolutional layers is the ReLU function, the stride is (2, 2), and zero padding is used. The pooling layer size is 2×2. The output layer outputs 4, and the activation function is the softmax function.

Images