Intelligent bearing fault diagnosis method of VMD-SCO-MCNN

Abstract

The invention provides an intelligent bearing fault diagnosis method based on VMD-SCO-MCNN, and belongs to the technical field of fault diagnosis, and the method comprises the steps: obtaining vibration signals of a bearing under normal and different fault types; performing preprocessing, decomposition and noise removal on the vibration signal by adopting a variational mode decomposition method VMD, and performing reconstruction to generate a new signal; the new signal is converted into a two-dimensional time-frequency image by adopting short-time Fourier transform STFT; adopting a shark cooperative optimization algorithm SCO to optimize hyper-parameters of the multi-scale convolutional neural network MCNN; inputting the two-dimensional time-frequency image into the MCNN model after the hyper-parameter optimization for training; and after noise reduction and two-dimensional time-frequency image conversion are carried out on a fault diagnosis signal of a bearing to be diagnosed, the signal is input into the VMD-SCO-MCNN model to realize bearing fault diagnosis. The method has very important practical significance for improving the bearing fault diagnosis accuracy and guiding equipment maintenance and repair.

Description

Technical Field

[0001] The present invention relates to the technical field of bearing fault diagnosis, and in particular to a VMD-SCO-MCNN intelligent bearing fault diagnosis method. Background Art

[0002] Bearings are one of the most prone to failure in various rotating machinery and equipment. If a bearing fails, it will lead to a decline in product quality or even production suspension, causing economic losses to the company, or even cause casualties and bring adverse social impacts. Therefore, it is of great practical significance to effectively diagnose bearing faults in rotating machinery and guide equipment maintenance and repair.

[0003] At present, with the development of machine learning method research, artificial intelligence technology has been widely used in various fields. In the field of bearing fault diagnosis, applying machine learning methods, especially various neural network methods, to bearing fault diagnosis has become one of the mainstream research directions. However, the performance of the neural network model will be affected by hyperparameters. Inappropriate hyperparameters may directly lead to diagnostic errors and affect the diagnostic accuracy. Therefore, optimizing the neural network based on the optimization algorithm and finding the optimal match of hyperparameters has also become a hot topic in current research. However, the existing intelligent optimization algorithms still have room for further improvement in optimization accuracy and computational efficiency. Summary of the invention

[0004] The present invention provides an intelligent bearing fault diagnosis method of VMD-SCO-MCNN. VMD-SCO-MCNN refers to first using variational mode decomposition (VMD) to pre-process the signal for noise reduction, and then establishing a multi-scale convolutional neural network (MCNN) model optimized by a shark collaboration optimization algorithm (SCO). In the present invention, a new shark collaboration optimization algorithm (Shark collaboration optimization, SCO) is originally proposed, and the MCNN hyperparameters are optimized by the SCO algorithm, so that the network has a lower iteration loss value and a higher diagnostic accuracy, providing reliable technical support for improving the accuracy of bearing equipment fault diagnosis.

[0005] In nature, white tip reef sharks and grey reef sharks are two different shark species, but they often cooperate in hunting together to improve the efficiency of hunting. Among them, white tip reef sharks are small and flexible, so they can shuttle through the corals to hunt, while grey reef sharks are larger and are powerless against prey hiding in the corals. Therefore, grey reef sharks often gather near the corals where white tip reef sharks are located, looking for prey that escapes from the corals chased by white tip reef sharks, thereby improving the efficiency of hunting. The white tip reef sharks also improve their hunting efficiency because the grey reef sharks drive the prey into the corals. The shark cooperation optimization algorithm proposed in the present invention is a meta-heuristic swarm intelligence optimization algorithm inspired by the above natural phenomenon.

[0006] The technical solution to achieve the purpose of the present invention is: a VMD-SCO-MCNN intelligent bearing fault diagnosis method, comprising the following steps: Step 1: Obtain the vibration signals to be trained of the bearing under normal conditions and different fault types; Step 2: Use variational mode decomposition (VMD) to pre-process the vibration signal to reduce noise, decompose, denoise and reconstruct the signal to generate a new vibration signal; Step 3: Convert the new vibration signal into a two-dimensional time-frequency image through short-time Fourier transform (STFT); Step 4: Set the parameters required for the Shark Cooperation Optimization Algorithm (SCO) optimization algorithm; Step 5: Optimize the hyperparameters of the multi-scale convolutional neural network (MCNN) through the shark cooperative optimization algorithm (SCO); Step 6: Input the two-dimensional time-frequency image into the MCNN model after optimizing the hyperparameters, and train the bearing fault diagnosis model based on VMD-SCO-MCNN; Step 7: The fault diagnosis signal to be diagnosed is preprocessed for noise reduction, converted into a two-dimensional time-frequency image through STFT, and the image is input into the trained VMD-SCO-MCNN bearing fault diagnosis model to realize bearing fault diagnosis.

[0007] Furthermore, the noise reduction preprocessing includes the following processes: (1) setting the number of intrinsic mode functions (IMFs) and penalty factors required for VMD noise reduction; (2) decomposing the original signal using the VMD algorithm to obtain multiple IMF components; (3) arranging the IMFs from high to low according to the center frequency, and calculating the correlation of each IMF. The IMF with the lowest correlation is marked as a noise IMF; (4) removing all IMFs marked as noise, and adding the remaining IMFs to obtain the denoised reconstructed signal.

[0008] Furthermore, the parameters required for the Shark Cooperative Optimization (SCO) algorithm include the number of whitetip reef shark individuals M, the proportion of shark leaders Mp (0 < Mp < 1), the maximum number of grey reef sharks SH, the team search range Tp (0 < Tp < 0.1), the maximum movement step ratio St (0 < St < 0.1), the number of iterations to abandon pursuit D, the proportion of abandoning pursuit DP (0 < DP < 0.3), and the maximum number of iterations tmax.

[0009] Furthermore, in step 4, the Shark Cooperative Optimization algorithm is a brand-new meta-heuristic swarm intelligence search optimization algorithm. This algorithm involves the cooperation of two different types of individuals, whitetip reef sharks and grey reef sharks. After the initialization phase, its search steps need to repeatedly calculate six key stages: (1) leader election; (2) generating grey reef sharks; (3) constructing a hunting team; (4) team encirclement; (5) pursuit; (6) abandoning pursuit; In the initialization phase, M whitetip reef sharks are randomly generated, and its expression is: B i 0 =lb+ ( ub-lb ) × rand (0, 1) where, B i represents the position of the i th whitetip reef shark individual. The superscript 0 represents the initialization phase. Then B i 0 represents the position of the i th whitetip reef shark in the initialization phase; i i = 1, 2, … M, lb is the lower limit of the search domain, ub is the upper limit of the search domain, B i 0 , lb , ub are all N-dimensional vectors, where N represents the dimension of the problem to be solved, rand (0, 1) represents an N-dimensional vector composed of random numbers between 0 and 1; (1) In the leader election phase, calculate the fitness values of M whitetip reef sharks, and select the P whitetip reef sharks with the best fitness values as leaders: P = round(M × Mp) where, P is the number of leader whitetip reef sharks, Mp is the proportion of shark leaders, and the round() function represents rounding the value in the parentheses to an integer; (2) In the gray reef shark generation stage, for the P elected leader whitetip reef sharks, different numbers of gray reef sharks are randomly generated within the team search range Tp around each leader. The number of gray reef sharks is related to the fitness value of the whitetip reef shark leader. The number of gray reef sharks generated around each leader can be expressed as: S a t =SH×(1- ( F a t -F min t ) / ( F max t - F min t )) Among them, the maximum number of gray reef sharks in SH, superscript t Represents the number of current loops. S a t For the a The number of Grey Reef Sharks that spawn around a Whitetip Reef Shark leader. a =1,2,…P, F a yes No. a The fitness value of the leader of the white tip reef shark, F max t and F min t yes a The maximum and minimum fitness values ​​among white tip reef shark leaders; The positions of each grey reef shark generated are: H aj t = BA a t + ( ub-lb )× rand (0,1)×Tp Among them, Tp is the team search range, H aj t Representative a The first generation of white tip reef sharks around the leader j Grey reef shark location, BA a t is the position of the leader of the white tip reef shark,H aj t and BA a t are all N-dimensional vectors; (3) In the hunting team formation stage, for whitetip reef shark leaders, the leader and the gray reef sharks generated by the leader form a hunting team. For non-leader whitetip reef sharks, each whitetip reef shark individual forms a hunting team. (4) In the team hunting phase, first calculate the fitness value of each shark in each team, and then calculate the average fitness value of each team FT i t and the average position BS i t Finally, by comparing the average fitness values, we can get the average fitness value of the team with the best fitness value. FT min t and the average position BS min t , and the average fitness value of the worst team FT max t ； (5) In the pursuit phase, first, for each team, the white-tip reef sharks in that team move to the position with the best fitness value of the whole team; second, for the team with non-optimal fitness value, the white-tip reef sharks in that team will move towards the average position of the optimal team, and the step size of the movement is D i Related to the team's average fitness value: D i t = St×( ub-lb )×( FT i t -FT min t ) / ( FT max t -FT min t ) Among them, St is the maximum moving step ratio, and the position of the white-tip reef shark of the non-optimal team will be updated as follows: B i t+1 = B it + D i t ×( BS min t - B i t ) in, B i t+1 For the t+1 Second cycle i Location of individual white tip reef sharks; Finally, remove all grey reef sharks; (6) In the abandonment stage, after every D rounds of iterations, the white-tip reef sharks in the M×DP team with the worst average fitness value are deleted, and M×DP white-tip reef sharks are regenerated in the search domain according to the method of the initialization stage, where D is the number of abandonment iterations and DP is the abandonment ratio; After the calculation of the six key stages is completed, it is determined whether the iteration is completed. If the iteration is completed, the search ends and the optimal position is output. If the iteration is not completed, the next round of iteration is entered.

[0010] Furthermore, the multi-scale convolutional neural network (MCNN) includes convolution kernels of three different scales, and the hyperparameters include the number of neurons in the convolution layer, the size of the pooling layer, the number of neurons in the fully connected layer, and the batch size.

[0011] The beneficial effects of the present invention are as follows: the present invention proposes a novel meta-heuristic swarm intelligence optimization algorithm - Shark Cooperation Optimization Algorithm (SCO), which focuses the optimization attention on the individuals with the best fitness value through two different types of shark individuals, thereby improving the optimization efficiency; at the same time, unlike other algorithms, the present invention adopts team optimality instead of individual optimality as the potential object to be pursued, which helps to avoid falling into local optimality; the present invention optimizes the hyperparameters of the multi-scale convolutional neural network MCNN based on the SCO algorithm, so that the optimized MCNN can more effectively extract fault features and improve the accuracy of bearing fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0013] Figure 1It is a flow chart of the VMD-SCO-MCNN intelligent bearing fault diagnosis method; Figure 2 It is a detailed flow chart of SCO optimization algorithm; Figure 3 It is the confusion matrix diagram of the fault diagnosis results of the SCO-MCNN method; Figure 4 This is a comparison chart of iterative losses of different methods; Figure 5 This is a comparison chart of the iterative accuracy of different methods. DETAILED DESCRIPTION

[0014] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in combination with the implementation cases and drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not 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 belong to the scope of protection of the present invention. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the invention claimed for protection, but merely represents the selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work belong to the scope of protection of the present invention.

[0015] The method of the present invention is described below by taking a specific bearing fault diagnosis case as an example.

[0016] Example: Figure 1 , a VMD-SCO-MCNN intelligent bearing fault diagnosis method, comprising the following steps: Step 1: Obtain the vibration signals to be trained of the bearing under normal conditions and different fault types; Step 2: Use variational mode decomposition (VMD) to pre-process the vibration signal to reduce noise, decompose, denoise and reconstruct the signal to generate a new vibration signal; Step 3: Convert the new vibration signal into a two-dimensional time-frequency image through short-time Fourier transform (STFT); Step 5: Using the two-dimensional time-frequency image as input, optimize the hyperparameters of the multi-scale convolutional neural network through the shark cooperative optimization algorithm; Step 6: Based on the optimized hyperparameters, a bearing fault diagnosis model of an intelligent bearing fault diagnosis method based on VMD-SCO-MCNN is trained; Step 6: Input the two-dimensional time-frequency image into the MCNN model with optimized hyperparameters, and train to obtain a bearing fault diagnosis model based on VMD-SCO-MCNN; Step 7: Perform noise reduction preprocessing on the fault diagnosis signal to be diagnosed, convert it into a two-dimensional time-frequency image through STFT, and input the image into the trained bearing fault diagnosis model of VMD-SCO-MCNN to achieve bearing fault diagnosis.

[0017] Further, the noise reduction preprocessing includes the following processes: (1) Set the number of intrinsic mode functions (IMFs) and penalty factors required for VMD noise reduction; (2) Decompose the original signal using the VMD algorithm to obtain multiple IMF components; (3) Arrange the IMFs in descending order of central frequency, and calculate the correlation of each IMF. The IMF with the lowest correlation is marked as the noise IMF; (4) Remove all the marked noise IMFs, and add the remaining IMFs to obtain the denoised reconstructed signal.

[0018] Further, the parameters required for the shark cooperation optimization algorithm (SCO) include the number of whitetip reef sharks M, the proportion of shark leaders Mp (0 < Mp < 1), the maximum number of grey reef sharks SH, the team search range Tp (0 < Tp < 0.1), the maximum movement step ratio St (0 < St < 0.1), the number of iterations to give up chasing D, and the proportion of giving up chasing DP (0 < DP < 0.3), and the maximum number of iterations tmax.

[0019] Further, the shark cooperation optimization algorithm (SCO) is a brand-new meta-heuristic swarm intelligence search optimization algorithm. This algorithm includes the cooperation of two different types of individuals, whitetip reef sharks and grey reef sharks. After the initialization stage, its search steps need to repeatedly calculate six key stages: (1) Leader election; (2) Generate grey reef sharks; (3) Build a hunting team; (4) Team encirclement: (5) Chase; (6) Give up chasing; In the initialization stage, randomly generate M whitetip reef sharks, and its expression is: B i 0 =lb+ ( ub-lb ) × rand (0, 1) where B i represents the position of the i th whitetip reef shark individual. The superscript 0 represents the initialization stage, i = 1, 2, … M, lb is the lower limit of the search domain, ub is the upper limit of the search domain, Bi 0 , lb , ub They are all 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; (1) In the leader election stage, the fitness values ​​of M white-tip reef sharks are calculated, and the P white-tip reef sharks with the best fitness values ​​are selected as leaders: P = round(M×Mp) Among them, the round() function represents rounding the value in the brackets to an integer; (2) In the gray reef shark generation stage, for the P white-tip reef shark leaders elected, different numbers of gray reef sharks are randomly generated within the team search range Tp around each leader. The number of gray reef sharks is related to the fitness value of the white-tip reef shark leader. The number of gray reef sharks generated around each leader can be expressed as: S a t =SH×(1- ( F a t -F min t ) / ( F max t - F min t )) Among them, the superscript t Represents the number of current loops. S a For the a The number of Grey Reef Sharks that spawn around a Whitetip Reef Shark leader. a =1,2,…P, F a yes No. a The fitness value of the leader of the white tip reef shark, F max and F min yes a The maximum and minimum fitness values ​​among white tip reef shark leaders; The positions of each grey reef shark generated are: H aj t = BA a t + ( ub-lb )× rand (0,1)×Tp in, H aj Representative a The first generation of white tip reef sharks around the leader j Grey reef shark location, BA a is the position of the leader of the white tip reef shark, H aj and BA a are all N-dimensional vectors; (3) In the hunting team formation stage, for whitetip reef shark leaders, the leader and the gray reef sharks generated by the leader form a hunting team. For non-leader whitetip reef sharks, each whitetip reef shark individual forms a hunting team. (4) In the team hunting phase, first calculate the fitness value of each shark in each team, and then calculate the average fitness value of each team FT i and the average position BS i Finally, by comparing the average fitness values, we can get the average fitness value of the team with the best fitness value. FT min and the average position BS min , and the average fitness value of the worst team FT max ； (5) In the pursuit phase, first, for each team, the white-tip reef sharks in that team move to the position with the best fitness value of the whole team; second, for the team with non-optimal fitness value, the white-tip reef sharks in that team will move towards the average position of the optimal team, and the step size of the movement is D i Related to the team's average fitness value: D i t = St×( ub-lb )×( FT i t -FT min t ) / ( FT max t -FT min t ) Then the position of the white tip reef shark of the non-optimal team will be updated to: B i t+1 = B i t +D i t ×( BS min- B i t ) in, B i t+1 For the t+1 Second cycle i Location of individual white tip reef sharks; Finally, remove all grey reef sharks; (6) In the abandonment stage, after every D rounds of iterations, the white-tip reef sharks in the M × DP teams with the worst average fitness values ​​are deleted, and M × DP white-tip reef sharks are regenerated within the search domain according to the method of the initialization stage; After the calculation of the six key stages is completed, it is determined whether the iteration is completed. If the iteration is completed, the search ends and the optimal position is output. If the iteration is not completed, the next round of iteration is entered.

[0020] Furthermore, the multi-scale convolutional neural network (MCNN) includes convolution kernels of three different scales, and the hyperparameters include the number of neurons in the convolution layer, the size of the pooling layer, the number of neurons in the fully connected layer, and the batch size.

[0021] The test data comes from the XJTU-SY rolling bearing accelerated life test, and the experimental data information is shown in Table 1.

[0022] Table 1 Experimental data information table

[0023] The specific implementation process is as follows: A VMD-SCO-MCNN intelligent bearing fault diagnosis method comprises the following steps: Step 1: Obtain the vibration signals of the bearing under normal conditions and different fault types to be trained. In this embodiment, 70% of the total number of samples are selected as training samples, 10% as verification samples, and 20% as test samples; Step 2: Use variational mode decomposition (VMD) to pre-process the vibration signal to reduce noise, decompose, denoise and reconstruct the signal to generate a new vibration signal; Step 3: Convert the new vibration signal into a two-dimensional time-frequency image through short-time Fourier transform (STFT); Step 4: Set the parameters required for the Shark Cooperative Optimization (SCO) algorithm. The number of individual whitetip reef sharks is M = 20, the proportion of shark leaders is Mp = 0.1, the maximum number of grey reef sharks is SH = 5, the team search range is Tp = 0.05, the maximum movement step ratio is St = 0.05, the number of iterations to abandon pursuit is D = 5, and the proportion of abandoning pursuit is DP = 0.15 (0 < DP < 0.3), and the maximum number of iterations is tmax = 50; Step 5: Use the two-dimensional time-frequency image as the input, and optimize the hyperparameters of the multi-scale convolutional neural network through the Shark Cooperative Optimization algorithm. The process of the Shark Cooperative Optimization algorithm is as Figure 2 shown; Step 6: Based on the optimized hyperparameters, train a bearing fault diagnosis model for the intelligent bearing fault diagnosis method based on VMD-SCO-MCNN; Step 7: Select the test sample as the signal to be diagnosed, perform noise reduction preprocessing on the fault diagnosis signal to be diagnosed, convert it into a two-dimensional time-frequency image through STFT, and input the image into the trained bearing fault diagnosis model of VMD-SCO-MCNN to achieve bearing fault diagnosis.

[0024] To verify the advantages of the method of the present invention, after VMD noise reduction preprocessing and STFT conversion of the two-dimensional time-frequency image, calculations were respectively performed using the ordinary convolutional neural network CNN, multi-scale convolutional neural network MCNN, MCNN optimized by the Sparrow Search Algorithm (SSA), and MCNN optimized by the Sand Cat Swarm Optimization (SCSO) algorithm, and comparisons were made with the method of the present invention. The present invention evaluated the classification results of each method using accuracy and F1 value, and the evaluation results are shown in Table 2, Figure 3 、 Figure 4 、 Figure 5 . It can be seen that the SCO-MCNN method proposed by the present invention has better performance than the comparative algorithms in terms of accuracy, F1 value, and iterative loss. Among them, Figure 4 is the comparison chart of iterative losses of different methods. The iterative loss represents the degree of difference between the model prediction and the true value. Figure 4 The figure in the figure in[] is the iterative loss from the 46th to the 50th generation during the model training iteration process.

[0025] Table 2 Comparison of accuracy and F1 value of each model

[0026] The above is only the preferred embodiment of the present invention and is not used to limit the present invention. For those skilled in the art, the present invention can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A VMD-SCO-MCNN intelligent bearing fault diagnosis method, characterized in that: Including the following steps: Step 1: Obtain the vibration signals to be trained for the bearing under normal conditions and different fault types; Step 2: Use variational mode decomposition to perform noise reduction preprocessing on the vibration signals, decompose, denoise, and reconstruct the signals to generate new vibration signals; Step 3: Convert the new vibration signals into two-dimensional time-frequency images through short-time Fourier transform; Step 4: Set the parameters required for the shark cooperation optimization algorithm; Step 5: Use the two-dimensional time-frequency images as input, and optimize the hyperparameters of the multi-scale convolutional neural network through the shark cooperation optimization algorithm; Step 6: Based on the optimized hyperparameters, train a bearing fault diagnosis model for the intelligent bearing fault diagnosis method based on VMD-SCO-MCNN; Step 7: Perform noise reduction preprocessing on the fault diagnosis signals to be diagnosed, convert them into two-dimensional time-frequency images through short-time Fourier transform, and input the images into the trained bearing fault diagnosis model of variational mode decomposition-shark cooperation optimization algorithm-multi-scale convolutional neural network to achieve bearing fault diagnosis.

2. The VMD-SCO-MCNN intelligent bearing fault diagnosis method according to claim 1, characterized in that: In Step 2, the noise reduction preprocessing includes the following process: Set the number of intrinsic mode functions and the penalty factor required for variational mode decomposition noise reduction; Decompose the original signal using the variational mode decomposition algorithm to obtain multiple intrinsic mode function components; Arrange the intrinsic mode functions in descending order of central frequency, and calculate the correlation of each intrinsic mode function. The intrinsic mode function with the lowest correlation is marked as the noise intrinsic mode function; Remove all the marked noise intrinsic mode functions, and add the remaining intrinsic mode functions to obtain the denoised reconstructed signal.

3. The VMD-SCO-MCNN intelligent bearing fault diagnosis method according to claim 1, characterized in that: In Step 4, the parameters required for the shark cooperation optimization algorithm include the number of whitetip reef sharks M, the proportion of shark leaders Mp, where 0 < Mp < 1, the maximum number of grey reef sharks SH, the team search range Tp, where 0 < Tp < 0.1, the maximum moving step ratio St, where 0 < St < 0.1, the number of iterations to give up chasing D, and the proportion of giving up chasing DP, where 0 < DP < 0.3, and the maximum number of iterations tmax.

4. The VMD-SCO-MCNN intelligent bearing fault diagnosis method according to claim 1, characterized in that: In Step 4, the shark cooperation optimization algorithm is a new metaheuristic swarm intelligence search optimization algorithm. This algorithm involves the cooperation of two different types of individuals, whitetip reef sharks and grey reef sharks. After the initialization stage, its search steps need to repeatedly calculate six key stages: (1) Leader election; (2) Generating grey reef sharks; (3) Constructing hunting teams; (4) Team encirclement: (5) Chasing; (6) Giving up chasing; In the initialization stage, randomly generate M whitetip reef sharks, and its expression is: B i 0 =lb+ ( ub - lb )× rand (0,1) in, B i Representative i The position of the individual white tip reef shark, the superscript 0 represents the initialization stage, then B i 0 Representative i Position of white tip reef sharks during initialization phase; i =1,2,…M, lb is the lower limit of the search domain, ub is the upper limit of the search domain, B i 0 , lb , ub They are all 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; (1) In the leader election stage, calculate the fitness values of M whitetip reef sharks, and select the P whitetip reef sharks with the best fitness values as leaders: P = round(M × Mp) where P is the number of leader whitetip reef sharks, Mp is the proportion of shark leaders, and the round() function represents rounding the value in the parentheses to an integer; (2) In the gray reef shark generation stage, for the P elected leader whitetip reef sharks, different numbers of gray reef sharks are randomly generated within the team search range Tp around each leader. The number of gray reef sharks is related to the fitness value of the whitetip reef shark leader. The number of gray reef sharks generated around each leader can be expressed as: S a t =SH×(1- ( F a t -F min t ) / ( F max t - F min t )) Among them, the maximum number of gray reef sharks in SH, superscript t Represents the number of current loops. S a t For the a The number of Grey Reef Sharks that spawn around a Whitetip Reef Shark leader. a =1,2,…P, F a yes No. a The fitness value of the leader of the white tip reef shark, F max t and F min t yes a The maximum and minimum fitness values ​​among white tip reef shark leaders; The positions of each grey reef shark generated are: H aj t = BA a t + ( ub-lb )× rand (0,1)×Tp in, H aj t Representative a The first generation of white tip reef sharks around the leader j Grey reef shark location, BA a t is the position of the leader of the white tip reef shark, H aj t and BA a t are all N-dimensional vectors; (3) In the hunting team formation stage, for whitetip reef shark leaders, the leader and the gray reef sharks generated by the leader form a hunting team. For non-leader whitetip reef sharks, each whitetip reef shark individual forms a hunting team. (4) In the team hunting phase, first calculate the fitness value of each shark in each team, and then calculate the average fitness value of each team FT i t and the average position BS i t Finally, by comparing the average fitness values, we can get the average fitness value of the team with the best fitness value. FT min t and the average position BS min t , and the average fitness value of the worst team FT max t ； (5) In the pursuit phase, first, for each team, the white-tip reef sharks in that team move to the position with the best fitness value of the whole team; second, for the team with non-optimal fitness value, the white-tip reef sharks in that team will move towards the average position of the optimal team, and the step size of the movement is D i Related to the team's average fitness value: D i t = St×( ub-lb )×( FT i t -FT min t ) / ( FT max t -FT min t ) Among them, St is the maximum moving step ratio, and the position of the white-tip reef shark of the non-optimal team will be updated as follows: B i t+1 = B i t + D i t ×( BS min t - B i t ) in, B i t+1 For the t+1 Second cycle i Location of individual white tip reef sharks; Finally, remove all grey reef sharks; (6) In the abandonment stage, after every D rounds of iterations, the white-tip reef sharks in the M×DP team with the worst average fitness value are deleted, and M×DP white-tip reef sharks are regenerated in the search domain according to the method of the initialization stage, where D is the number of abandonment iterations and DP is the abandonment ratio; After the calculation of the six key stages is completed, it is determined whether the iteration is completed. If the iteration is completed, the search ends and the optimal position is output. If the iteration is not completed, the next round of iteration is entered.

5. The VMD-SCO-MCNN intelligent bearing fault diagnosis method according to claim 1, characterized in that: In step five, the multi-scale convolutional neural network includes convolution kernels of three different scales, and the hyperparameters include the number of neurons in the convolution layer, the size of the pooling layer, the number of neurons in the fully connected layer, and the batch size.

Images