Novel intelligent rotating machine fault diagnosis method

Abstract

The invention provides a novel intelligent rotating machine fault diagnosis method, and belongs to the technical field of rotating machine fault diagnosis, and the method comprises the steps: obtaining vibration signals of a rotating machine under normal and different fault types; a variation mode decomposition method is adopted to pre-process and decompose the vibration signal of the rotating machine, noise is removed, and then a new signal is reconstructed and generated; the new vibration signal is converted into a two-dimensional time-frequency image through short-time Fourier transform; optimizing hyper-parameters of a convolutional neural network (CNN)-long and short term memory neural network (BiLSTM) by adopting a Sharla silver ant optimization algorithm (SSAO), wherein the hyper-parameters comprise the number of neurons of a convolutional layer, the size of a convolution kernel and the number of neurons of a BiLSTM layer; and inputting the two-dimensional time-frequency image into the CNN-BiLSTM model after the hyper-parameter optimization to realize fault diagnosis of the rotating machine. The method has very important practical significance for improving the rotating machine fault diagnosis accuracy and guiding equipment maintenance and repair.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of rotating machinery fault diagnosis, in particular to a novel intelligent rotating machinery fault diagnosis method. BACKGROUND

[0002] As the core equipment in industrial production, the running state of rotating machinery is directly related to the stability of the production system. Equipment failure may cause production process interruption and product quality decline, and even cause major safety accidents and huge economic losses. Therefore, it is of great engineering value to build an efficient fault diagnosis system.

[0003] In recent years, deep learning technology has made significant progress in rotating machinery fault diagnosis due to its strong feature learning ability. The mainstream method converts the vibration signal into a time-frequency image through time-frequency analysis technology, and then uses convolutional neural network and its improved architecture for feature extraction and fault classification. However, the diagnostic performance of existing methods is highly dependent on the optimized configuration of network hyperparameters, and traditional manual parameter tuning methods have bottlenecks. To solve these problems, it has become a research hotspot to systematically explore the optimal hyperparameter combination by constructing a parameter optimization objective function and combining intelligent optimization algorithms. However, there is still room for improvement in the optimization accuracy and computational efficiency of existing intelligent optimization algorithms. SUMMARY

[0004] The main purpose of the present application is to solve the problems existing in the prior art, and to provide a novel intelligent rotating machinery fault diagnosis method. First, the signal is preprocessed by variational model decomposition (VMD) for noise reduction, and then a convolutional neural network-bidirectional long short-term memory network model (CNN-BiLSTM) optimized by Saharan Silver Ant Optimization (SSAO) is established. The present application originally proposes a novel Saharan Silver Ant Optimization (SSAO) algorithm, and optimizes the CNN-BiLSTM hyperparameters through the SSAO 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 rotating machinery equipment fault diagnosis.

[0005] In nature, Saharan silver ants have evolved an efficient foraging strategy in the extreme high-temperature desert environment. When the temperature exceeds a critical value, the ant colony adopts a dynamic team formation mechanism to form cooperative teams, and individuals continuously share information to optimize the foraging path. In team behavior, experienced individuals gradually evolve into leaders, responsible for decision-making whether to continue exploring near the area where food resources are concentrated or to follow the current optimal path steadily. The behavior of team members is influenced by the leader: when an individual is close to the food source, it usually follows the trail left by the leader of its team and performs local exploration on the path; when the distance is far, it tends to quickly approach the leader's position and search cooperatively. In addition, when the local information is limited or the surrounding environment changes dramatically, some individuals may abandon the guidance of the original leader and follow a random leader from other teams to explore. Through continuous information exchange and path adjustment, the ant colony maintains efficient cooperation and foraging ability in high-temperature environments. The Saharan silver ant optimization algorithm proposed in the invention is inspired by the above natural phenomenon and is a meta-heuristic swarm intelligence optimization algorithm.

[0006] The present application is implemented as follows: The technical solution for achieving the object of the present application is a novel intelligent rotating machinery fault diagnosis method, comprising the following steps: Step one: obtaining the vibration signals to be trained and fault classification of rotating machinery under normal conditions and different fault types; Step two: using variational mode decomposition to pre-process the noise reduction of the vibration signals to be trained, and obtaining the reconstructed signals after pre-processing; Step three: converting the reconstructed signals into two-dimensional time-frequency images through short-time Fourier transform; Step four: setting the parameters required by the Saharan silver ant optimization algorithm; Step five: optimizing the hyperparameters of the convolutional neural network-bidirectional long short-term memory network based on the Saharan silver ant optimization algorithm; Step six: training the rotating machinery fault diagnosis model based on SSAO-VMD-CNN-BiLSTM based on the optimized hyperparameters, reconstructed signals, and fault classification; Step seven: pre-processing the fault diagnosis signals to be diagnosed, converting them into two-dimensional time-frequency images through short-time Fourier transform, and inputting the images into the trained rotating machinery fault diagnosis model based on SSAO-VMD-CNN-BiLSTM to realize rotating machinery fault diagnosis.

[0007] Furthermore, in step 2, the noise reduction preprocessing includes the following process: setting the number of intrinsic mode functions and penalty factors required for variational mode decomposition denoising; using the variational mode decomposition algorithm to decompose the original signal to obtain multiple intrinsic mode function components; arranging the intrinsic mode functions from high to low according to the center frequency, and calculating the correlation of each intrinsic mode function, and marking the intrinsic mode function with the lowest correlation as a noise intrinsic mode function; removing all intrinsic mode functions marked as noise, and adding the remaining intrinsic mode functions to obtain the denoised reconstructed signal.

[0008] Furthermore, in step 4, the Sahara Silver Ant optimization algorithm is divided into five steps that are performed sequentially: Step S41: In the initialization phase, Latin hypercube sampling is used to generate M The initial distribution of silver ants in the thermal domain is expressed mathematically as follows: X k = lb +1 / M ×( Y + μ- 1)( ub - lb ) in, X k Indicates the k The location of individual silver ants; lb and ub are the lower and upper bounds of the search space, respectively. Y is {1,2,…, M}, μ Represents a random number from the interval (0,1); Step S42: During the heat adaptation formation phase, the silver ants will randomly form teams to update their positions. The number of members in each team is Ts It is dynamically adjusted as follows: Ts =max(3,min(5,round(3+2sin(2π× t / T )))) Among them, the round() function represents rounding the value in the brackets to an integer. t For the current iteration, T is the total number of iterations; Step S43: In the leader election phase, by evaluating the Ts The fitness values ​​of the silver ants are calculated, and the individual with the best fitness is selected as the leader of the team; Step S44: In the ant path optimization phase, for each team leader, according to its fitnessF i,b t the ratio between the current global optimum fitness and the best fitness of the population determines the search strategy: if F g t F i,b t β F g t path development is performed in the direction towards the global optimum, where β is the behavior control parameter; otherwise, local exploration is performed directly in the neighborhood of the global optimum, by the following method: β =1+1 / 2× U (0,1) X i,b t+1 X i,b t +| N (0,1)|×( X g t X i,b t + N (0,1)× τ ( n )× P F i,b t β × F g t X i,b t+1 X g t ×( U (0,1) +τ ( n ) ×P Other where, U (0,1) denotes a uniformly distributed random variable in the interval [0,1], F i,b t represents the fitness value of the leader of the i-th team in the j-th iteration, t i F g t represents the fitness value of the leader of the i-th team in the j-th iteration,​​​​​​​​​t The global optimal solution in the iteration; X i,b t+1 and X i,b t Respectively represent t+ 1st iteration and t In the iteration i A team leader position, X g t Indicates the t The global optimal position of the iteration, F i,b t and F g t are the fitness values ​​of the corresponding team leader and the global optimal position, respectively. t Represents the current iteration number, N (0,1) is a standard normal distribution random variable, τ ( n ) is subject to the degree of freedom n T-distributed random variable, where n =5+round(10×(1–( t / T ))), the round() function rounds the value in the brackets to an integer; P = (X k t -lb) / (ub-lb) ,in X k t Indicates the k Silver ant individuals t The position in the iteration; Step S45: In the unknown area exploration phase, for each team member, according to their fitness F i,m t Compatibility with team leaders F i,b t Choose the exploration strategy based on the proportional relationship between F i,m t <β×F i,b t , a cross-team learning approach is adopted to introduce the information of the best individuals from other teams for search; otherwise, local exploration is performed directly near the current team leader. The update formula for this stage is as follows: X i,mt+1 =rand ( X i,b t ) + | N (0,1)| × ( rand ( X i,b t ) -X i,m t ) F i,m t <β×F i,b t X i,m t+1 = X i,b t ×( U (0,1) +τ ( n ) ×P ) other in, X i,m t+1 and X i,m t Respectively represent t +1 iteration and t In the first iteration, the team i Middle m The location of each member, rand ( X i,b t ) indicates that t The position of a leader randomly selected from the team leaders of the iteration; After the five-stage calculation 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, it returns to step S42 and repeats S42 to S45.

[0009] Furthermore, in step 4, the parameters required for the Sahara silver ant optimization algorithm include the total number of individuals M , the largest team Ts max , minimum scale Ts min , maximum number of iterations T .

[0010] Further, in step five, the Sahara silver ant optimization algorithm optimizes the hyperparameters of the CNN-BiLSTM, including the number of convolutional layer neurons, the convolution kernel size, and the number of long short-term memory layer neurons.

[0011] Further, in step six, the CNN-BiLSTM model includes four convolutional layers, three pooling layers, one flattening layer, one long short-term memory layer, and one output layer, the convolutional layer activation function is the relu function, the stride is (2, 2), and the zero padding; the pooling layer size is 2x2; the output layer output is 4, and the activation function is the softmax function; wherein the number of convolutional layer neurons, the convolution kernel size, and the number of long short-term memory layer neurons are obtained from step five.

[0012] The beneficial effects of the present application are: the present application proposes a new meta-heuristic swarm intelligence optimization algorithm, Sahara silver ant cooperative optimization algorithm SSAO, through dynamic team formation and multi-source information sharing mechanism, the individuals continuously transmit environmental and position information, so that the search behavior focuses on the individual around the optimal fitness, thereby improving the optimization efficiency; at the same time, unlike other algorithms, the present application introduces a cross-team leader switching strategy under certain conditions, individuals can abandon the original team leader and follow the random leader of other teams for exploration, effectively enhancing the population diversity and helping to avoid falling into local optimum; the present application optimizes the hyperparameters of CNN-BiLSTM based on SSAO algorithm, which can make the optimized CNN-BiLSTM more effectively extract fault features and improve the fault diagnosis accuracy. BRIEF DESCRIPTION OF DRAWINGS

[0013] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments. It should be understood that the following drawings only show some embodiments of the present application, and therefore should not be regarded as a limitation on the scope, and for those skilled in the art, other related drawings can also be obtained without creative labor.

[0014] Figure 1 is a flow chart of a new intelligent rotating machinery fault diagnosis method provided by the embodiments of the present application; Figure 2 is a detailed flow chart of the SSAO optimization algorithm provided by the embodiments of the present application; Figure 3 is a CNN-BiLSTM network model structure topology diagram; Figure 4 is a fault diagnosis result confusion matrix diagram of the SSAO-VMD-CNN-BiLSTM method;

[0015] Figure 5 is a comparison diagram of iteration loss of different methods; Figure 6 is a comparison chart of iteration accuracy of different methods. DETAILED DESCRIPTION

[0016] In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the embodiments in the present application and the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative labor fall within the protection scope of the present application. Therefore, the following detailed description of the embodiments of the present application provided in the accompanying drawings is not intended to limit the scope of the claimed present application, but only represents selected embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative labor fall within the protection scope of the present application.

[0017] The method of the present application will be described below with reference to a specific rotating machine fault diagnosis case.

[0018] As Figure 1 A novel intelligent rotating machine fault diagnosis method, comprising the following steps: Step one: obtaining the vibration signals to be trained and fault classification of the rotating machine under normal conditions and different fault types; Step two: adopting variational mode decomposition to carry out noise reduction pretreatment on the vibration signals to be trained, to obtain a reconstructed signal after pretreatment; Step three: converting the reconstructed signal into a two-dimensional time-frequency image through short-time Fourier transform; Step four: setting the parameters required by the Sahara silver ant optimization algorithm; Step five: optimizing the hyperparameters of the convolutional neural network-bidirectional long short-term memory network based on the Sahara silver ant optimization algorithm; Step six: training a rotating machine fault diagnosis model based on SSAO-VMD-CNN-BiLSTM based on the optimized hyperparameters, the reconstructed signal and the fault classification; Step seven: carrying out noise reduction pretreatment on the fault diagnosis signal to be diagnosed, converting it into a two-dimensional time-frequency image through short-time Fourier transform, and inputting the image into the trained rotating machine fault diagnosis model based on SSAO-VMD-CNN-BiLSTM, to realize rotating machine fault diagnosis.

[0019] Furthermore, the noise reduction preprocessing includes the following processes: setting the number of intrinsic mode functions and penalty factors required for variational mode decomposition noise reduction; using the variational mode decomposition algorithm to decompose the original signal to obtain multiple intrinsic mode function components; arranging the intrinsic mode functions from high to low according to the center frequency, and calculating the correlation of each intrinsic mode function, and marking the intrinsic mode function with the lowest correlation as a noise intrinsic mode function; removing all intrinsic mode functions marked as noise, and adding the remaining intrinsic mode functions to obtain the denoised reconstructed signal.

[0020] Furthermore, the Sahara Silver Ant optimization algorithm is divided into five steps that are performed sequentially: Step S41: In the initialization phase, Latin hypercube sampling is used to generate M The initial distribution of silver ants in the thermal domain is expressed mathematically as follows: X k = lb +1 / M ×( Y + μ- 1)( ub - lb ) in, X k Indicates the k The location of individual silver ants; lb and ub are the lower and upper bounds of the search space, respectively. Y is {1,2,…, M}, μ Represents a random number from the interval (0,1); Step S42: During the heat adaptation formation phase, the silver ants will randomly form teams to update their positions. The number of members in each team is Ts It is dynamically adjusted as follows: Ts =max(3,min(5,round(3+2sin(2π× t / T )))) Among them, the round() function represents rounding the value in the brackets to an integer. t For the current iteration, T is the total number of iterations; Step S43: In the leader election phase, by evaluating the Ts The fitness values ​​of the silver ants are calculated, and the individual with the best fitness is selected as the leader of the team; Step S44: In the ant path optimization phase, for each team leader, according to its fitness Fi,b t The ratio between the current global optimal solution fitness F g t determines the search strategy: if F i,b t , β x F g t , path development is performed in the direction towards the global optimum, where β is the behavior control parameter; otherwise, local exploration is performed directly in the vicinity of the global optimal position, as follows: β = 1 + 1 / 2 x U (0,1) X i,b t+1 = X i,b t + N (0,1) x ( X g t - X i,b t + N (0,1) x τ ( n ) x P F i,b t β x F g t X i,b t+1 = X g t x ( U (0,1) +τ ( n ) ×P ) otherwise where, U (0,1) denotes a uniformly distributed random variable in the interval [0,1], F i,b t represents the fitness value of the t th team leader in the i th iteration, F g t represents the fitness value of the t ​the global optimal solution in the next iteration; X i,b t+1 and X i,b t denote the position of the i-th team leader in the 1st t+ iteration and the i-th iteration, respectively, t i X g t denote the global optimal position in the i-th iteration, t F i,b t and F g t are the fitness values of the corresponding team leader and the global optimal position, respectively, and the superscript t represents the current iteration number, N (0, 1) is a standard normal distributed random variable, τ (0, 1) is a standard normal distributed random variable, n is a T-distributed random variable with degrees of freedom n , where n = 5 + round(10 x (1 - ( t / T )))),and the round() function represents rounding the value inside the parentheses to the nearest integer; P = (X k t -lb) / (ub-lb) , where X k t denotes the position of the i-th silver ant individual in the i-th iteration; k t Step S45: In the exploration phase of the unknown region, for each team member, the exploration strategy is selected according to the proportional relationship between its fitness value i,m t and the fitness value of the team leader F i,b t : if F i,m t F i,b t , the cross-team learning method is adopted to introduce the information of the optimal individual of other teams for search; otherwise, local exploration is directly performed near the current team leader, and the update formula in this phase is as follows: <β×F X i,m t+1 ​​​​=rand ( X i,b t ) + | N (0,1)| × ( rand ( X i,b t ) -X i,m t ) F i,m t <β×F i,b t X i,m t+1 = X i,b t ×( U (0,1) +τ ( n ) ×P ) other in, X i,m t+1 and X i,m t Respectively represent t +1 iteration and t In the first iteration, the team i Middle m The location of each member, rand ( X i,b t ) indicates that t The position of a leader randomly selected from the team leaders of the iteration; After the five-stage calculation 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, it returns to step S42 and repeats S42 to S45.

[0021] Furthermore, in step 4, the parameters required for the Sahara silver ant optimization algorithm include the total number of individuals M , the largest team Ts max , minimum scale Ts min , maximum number of iterations T .

[0022] Further, in step five, the Sahara silver ant optimization algorithm optimizes the hyperparameters of the CNN-BiLSTM, including the number of convolutional layer neurons, the convolution kernel size, and the number of long short-term memory layer neurons.

[0023] Further, in step six, the CNN-BiLSTM model includes four convolutional layers, three pooling layers, one flattening layer, one long short-term memory layer, and one output layer. The convolutional layer activation function is the relu function, the stride is (2, 2), and the zero padding is used. The pooling layer size is 2x2. The output layer output is 4, and the activation function is the softmax function. The number of convolutional layer neurons, the convolution kernel size, and the number of long short-term memory layer neurons are obtained from step five. The method of the present application is described below with specific rotating machinery fault cases.

[0024] The experimental data comes from the public data set of the CWRU bearing data center, and the experimental data information is shown in Table 1.

[0025] Table 1 Experimental data information table

[0026] The specific implementation process is as follows: As Figure 1 , a new intelligent rotating machinery fault diagnosis method, comprising the following steps: Step one: obtain the vibration signals of rotating machinery under normal conditions and different fault types for training. In this embodiment, 70% of the total number of samples is selected as the training sample, 10% as the validation sample, and 20% as the test sample. Step two: use VMD (Variational Modal Decomposition) to preprocess the vibration signal, decompose, denoise, and reconstruct the signal to generate a new vibration signal. Step three: convert the new vibration signal into a two-dimensional time-frequency image through STFT (Short-Time Fourier Transform). Step four: set the parameters required by the Sahara silver ant optimization algorithm, wherein the total number of individuals M = 30, the maximum team size Ts max = 5, the minimum size Ts min = 3, and the maximum number of iterations T = 350. Step five: optimize the hyperparameters of the CNN-BiLSTM based on the Sahara silver ant optimization algorithm. The flowchart of the Sahara silver ant optimization algorithm is shown in Figure 2 , and the evaluation results after optimization are shown in Figure 4 . Step six: input the two-dimensional time-frequency image into the CNN-BiLSTM model with optimized hyperparameters to realize rotating machinery fault diagnosis; the CNN-BiLSTM network model structure topological graph is shown in Figure 3 , and the evaluation results are shown in Table 2, Figure 5 , Figure 6 .

[0027] Step seven: the fault diagnosis signal to be diagnosed is preprocessed by noise reduction, and is converted into a two-dimensional time-frequency image by short-time Fourier transform, and the image is input into the trained rotating machinery fault diagnosis model based on SSAO-VMD-CNN-BiLSTM to realize rotating machinery fault diagnosis.

[0028] Table 2: Comparison table of model accuracy and loss value

Claims

1. A novel intelligent rotating machinery fault diagnosis method, characterized in that: The following steps are involved: Step 1: Obtain the vibration signals and fault classifications to be trained for rotating machinery under normal conditions and different fault types; Step 2: Use variational mode decomposition to perform noise reduction preprocessing on the training vibration signal to obtain the preprocessed reconstructed signal; Step 3: Convert the reconstructed signal into a two-dimensional time-frequency image through short-time Fourier transform; Step 4: Set the parameters required for the Sahara Silver Ant optimization algorithm; Step 5: Optimize the hyperparameters of the convolutional neural network-bidirectional long short-term memory network based on the Sahara Silver Ant optimization algorithm; Step 6: Based on the optimized hyperparameters, reconstructed signals, and fault classification, a rotating machinery fault diagnosis model based on SSAO-VMD-CNN-BiLSTM is trained. Step 7: The fault diagnosis signal to be diagnosed is preprocessed for noise reduction and converted into a two-dimensional time-frequency image through short-time Fourier transform. The image is then input into the trained rotating machinery fault diagnosis model based on SSAO-VMD-CNN-BiLSTM to realize rotating machinery fault diagnosis.

2. A novel intelligent rotating machinery fault diagnosis method according to claim 1, characterized in that: In step 2, the noise reduction preprocessing includes the following process: setting the number of intrinsic mode functions and penalty factors required for variational mode decomposition noise reduction; The variational mode decomposition algorithm is used to decompose the signal to be preprocessed to obtain multiple intrinsic mode function components; the intrinsic mode functions are arranged from high to low according to the center frequency, and the correlation of each intrinsic mode function is calculated, and the intrinsic mode function with the lowest correlation is marked as the noise intrinsic mode function; all the intrinsic mode functions marked as noise are removed, and the remaining intrinsic mode functions are added to obtain the denoised reconstructed signal.

3. A novel intelligent rotating machinery fault diagnosis method according to claim 1, characterized in that: In step 4, the Sahara Silver Ant optimization algorithm is divided into five steps that are performed in sequence: Step S41: In the initialization phase, Latin hypercube sampling is used to generate M The initial distribution of silver ants in the thermal domain is expressed mathematically as follows: X k = lb +1 / M ×( Y + μ- 1)( ub - lb ) in, X k Indicates the k The location of individual silver ants; lb and ub are the lower and upper bounds of the search space, respectively. Y is {1,2,…, M }, μ Represents a random number from the interval (0,1); Step S42: During the heat adaptation formation phase, silver ants randomly form teams to update their positions. The number of members in each team is Ts It is dynamically adjusted as follows: Ts =max(3,min(5,round(3+2sin(2π× t / T )))) Among them, the round() function represents rounding the value in the brackets to an integer. t For the current iteration, T is the total number of iterations; Step S43: In the leader election phase, by evaluating the Ts The fitness values ​​of the silver ants are calculated, and the individual with the best fitness is selected as the leader of the team; Step S44: In the ant path optimization phase, for each team leader, according to its fitness F i,b t Fitness with the current global optimal solution F g t The proportional relationship between determines the search strategy: if F i,b t < β × F g t , develop paths along the direction towards the global optimum, where β is the behavior control parameter; otherwise, local exploration is performed directly near the global optimal position as follows: β =1+1 / 2× U (0,1) X i,b t+1 = X i,b t +| N (0,1)|×( X g t - X i,b t )+ N (0,1)× τ ( n )× PF i,b t < β × F g t X i,b t+1 = X g t ×( U (0,1) +τ ( n ) ×P ) other in, U (0,1) represents a uniformly distributed random variable in the interval [0,1], F i,b t Representative t In the iteration i The fitness value of a team leader, F g t Representative t The global optimal solution in the iteration; X i,b t+1 and X i,b t Respectively represent t+ 1st iteration and t In the iteration i A team leader position, X g t Indicates the t The global optimal position of the iteration, F i,b t and F g t Respectively t The fitness value of the leader of the i-th team in the iteration and the global optimal position, N (0,1) is a standard normal distribution random variable, τ ( n ) is subject to the degree of freedom n T-distributed random variable, where n =5+round(10×(1–( t / T ))), the round() function rounds the value in the brackets to an integer; P=(X k t -lb) / (ub-lb) ,in X k t Indicates the k Silver ant individuals t The position in the iteration; Step S45: In the unknown area exploration phase, for each team member, according to their fitness F i,m t Compatibility with team leaders F i,b t Choose the exploration strategy based on the proportional relationship between F i,m t <β×F i,b t , a cross-team learning approach is adopted to introduce the information of the best individuals from other teams for search; otherwise, local exploration is performed directly near the current team leader. The update formula for this stage is as follows: X i,m t+1 =rand ( X i,b t ) + | N (0,1)| × ( rand ( X i,b t ) -X i,m t ) F i,m t <β×F i,b t X i,m t+1 = X i,b t ×( U (0,1) +τ ( n ) ×P ) other in, X i,m t+1 and X i,m t Respectively represent t +1 iteration and t In the first iteration, the team i Middle m The location of each member, rand ( X i,b t ) indicates that t The position of a leader randomly selected from the team leaders of the iteration; After the five-stage calculation 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, it returns to step S42 and repeats S42 to S45.

4. A novel intelligent rotating machinery fault diagnosis method according to claim 1, characterized in that: In step 4, the parameters required for the Sahara silver ant optimization algorithm include the total number of individuals M , the largest team Ts max , minimum scale Ts min , maximum number of iterations T .

5. A novel intelligent rotating machinery fault diagnosis method according to claim 1, characterized in that: In step 5, the Sahara Silver Ant optimization algorithm optimizes the hyperparameters of CNN-BiLSTM, including the number of neurons in the convolutional layer, the size of the convolution kernel, and the number of neurons in the long short-term memory layer.

6. A novel intelligent rotating machinery fault diagnosis method according to claim 1, characterized in that: In step 6, the CNN-BiLSTM model includes 4 convolutional layers, 3 pooling layers, a flattening layer, a long short-term memory layer, and an output layer. The activation function of the convolutional layer is the relu function with a stride of (2, 2) and zero padding; the pooling layer size is 2×2; the output of the output layer is 4, and the activation function is the softmax function; wherein, the number of neurons in the convolutional layer, the convolution kernel size, and the number of neurons in the long short-term memory layer are obtained from step 5.

Images