Intelligent fault diagnosis method for rotating machinery based on minimum average composite entropy

Abstract

The invention provides a rotating machine fault intelligent diagnosis method based on minimum average composite entropy, and belongs to the technical field of rotating machine fault diagnosis. The method comprises the following steps: inputting a rotary mechanical vibration signal; a minimum average composite entropy is used as a fitness function, and an improved dung beetle optimization algorithm is used to optimize two key parameter modal quantity k and a penalty factor alpha of variational modal decomposition (VMD). The composite entropy divergence is used as a characteristic parameter to extract an effective mode of variational mode decomposition, and a new vibration signal is reconstructed. And converting the new vibration signal into a two-dimensional time-frequency image through short-time Fourier transform, and inputting the two-dimensional time-frequency image into a CNN model integrated with a channel attention mechanism to realize fault diagnosis of the rotating machinery.

Description

Technical Field

[0001] The present invention relates to the technical field of rotating machinery fault diagnosis, and in particular to a rotating machinery fault intelligent diagnosis method based on minimum mean composite entropy. Background Art

[0002] The machinery industry is a strategic industry that provides technical equipment for the national economy and national defense construction. In recent years, mechanical equipment has gradually moved towards high speed, large-scale, integrated, and automated. Mechanical parts often need to work continuously for a long time under extremely harsh working conditions. Once a part of a mechanical equipment fails, it will cause a chain reaction, causing the entire equipment and even all systems to fail and be damaged, causing huge economic losses and even casualties.

[0003] Intelligent diagnosis methods such as deep convolutional neural networks have the advantages of being able to mine deep connections in data, prevent overfitting, and have high diagnostic accuracy. They are widely used in the field of rotating machinery fault diagnosis. The diagnostic accuracy of intelligent fault diagnosis methods is closely related to the quality of the input signal. Variational Mode Decomposition (VMD) can effectively solve the modal aliasing phenomenon in empirical mode decomposition (EMD) and improve the quality of the input signal. It has attracted widespread attention from a large number of scientific researchers and engineers. The two key parameters of VMD penalty parameter α and decomposition layer number k have a great influence on the fault signal decomposition and extraction effect of VMD. Using optimization algorithms to select key parameters of VMD can effectively reduce the selection of human factors, thereby improving the quality of input signals. However, most of the current methods that use single entropy as a VMD parameter optimization index and a screening method for effective modal components have limitations. For example, sample entropy measures the complexity of time series by measuring the probability of generating new modes in the signal. Information entropy is usually used to measure the information content of the system. The different emphases of different entropies may lead to different results of VMD parameter optimization, thus affecting the quality of the input signal and causing huge fluctuations in the fault diagnosis ability of the deep learning model.

[0004] The optimization ability of the optimization algorithm is crucial for the selection of key parameters of VMD. The dung beetle optimizer (DBO) is an optimization algorithm inspired by the behaviors of dung beetles such as rolling, dancing, foraging, stealing and reproduction. It has attracted widespread attention from the academic community due to its strong local development ability. However, DBO also has problems such as premature maturity and easy to fall into local optimal solutions, which greatly limits its optimization ability. Summary of the invention

[0005] The present invention provides a method for intelligent diagnosis of rotating machinery faults based on minimum mean composite entropy. The method aims to solve the problem that the current intelligent diagnosis method has poor diagnostic accuracy due to poor input signal quality. The present invention selects the minimum mean composite entropy that can better balance noise robustness and fault signal sensitivity as the fitness function, optimizes the two key parameters of VMD by improving the dung beetle optimization algorithm, and selects the composite entropy divergence as the characteristic parameter for extracting the effective mode of VMD, extracts the effective mode of VMD that can reflect the fault characteristics, and reconstructs the new vibration signal, thereby improving the quality of the input signal of the deep neural network. The newly reconstructed signal is then converted into a two-dimensional graph through a short-time Fourier transform and a CNN network that incorporates a channel attention mechanism is used to realize the fault diagnosis of rotating machinery.

[0006] The technical solution to achieve the purpose of the present invention is: a rotating machinery fault intelligent diagnosis method based on minimum average composite entropy, comprising the following steps: S100 step 1: input the vibration signal of the rotating machinery; S200 Step 2: Taking the minimum average composite entropy as the objective function, the improved dung beetle optimization algorithm is used to optimize the hyperparameters of VMD, namely the number of modes K and the penalty factor α. S300 Step 3: Using the composite Raney divergence as the characteristic parameter, extract the effective modes of VMD and reconstruct the new vibration signal; S400 Step 4: Convert the new vibration signal into a two-dimensional time-frequency image through short-time Fourier transform STFT; S500 Step 5: Input the two-dimensional time-frequency image into the CNN model with channel attention mechanism to realize rotating machinery fault diagnosis;

[0007] Furthermore, in step 2, the minimum mean composite entropy is used as the fitness function of the optimization algorithm to solve the problem that the currently used fitness function is difficult to balance the noise robustness and fault signal sensitivity. The composite entropy is composed of Rainey entropy and envelope entropy. The k-order modal Rainey entropy R after VMD decomposition is e (k) The formula is: where u k (i) is the time series u of its k-order mode after VMD decomposition k The i-th value of . The k-order modal envelope entropy R after VMD decomposition b (k) The formula is: The formula for the k-order modal composite entropy C(k) is: in The normalized Rainey entropy is defined as: Same reason The normalized envelope entropy is defined as:

[0008] Furthermore, in step 2, the improved dung beetle optimization algorithm mainly includes initialization behavior, fitness behavior, rolling behavior, dancing behavior, breeding behavior, foraging behavior, and stealing behavior. In the initialization behavior stage of S210, the initial population of the standard dung beetle optimization algorithm is randomly generated, which has the disadvantages of poor population diversity and uneven distribution. The improved dung beetle optimization initialization behavior uses tent chaotic mapping to initialize the dung beetle population. Its mathematical expression is: Where a is the chaos coefficient of the tent chaotic map; Z k is a random sequence, Z k ∈(0,1); Z k+1 According to Z k Generated chaotic sequence. S220 uses compound entropy as the fitness function in the fitness behavior stage to calculate the initial fitness of the dung beetle S230 In the ball rolling behavior stage, the ball rolling behavior of the dung beetle depends on celestial clues to ensure that the dung ball moves in a straight line, and considering the influence of light intensity on its ball rolling path, the ball rolling behavior position update expression of the dung beetle is: x(t+1)=x i (t)+α×c×x i (t-1)+b×Δx Δx=|x i (t)-X w | Among them, x i (t) is the position information of the ith agent at the tth iteration; α is a natural number, indicating the deflection direction; the value is -1 or 1. b represents the deflection coefficient constant, b∈(0, 0.02]; and c∈(0, 1). w represents the global worst position. In the dancing behavior stage S240, when the dung beetle encounters an obstacle, it can change direction through dancing behavior to obtain a new route. The position update formula of the dancing behavior of the standard dung beetle optimization algorithm is expressed as: x i (t+1)=x i (t)+tan(β)|x i (t)-x i (t-1)| Among them, β∈[0,π], when β=0,π / 2, orπ, the dancing behavior does not work and the position of the dung beetle is not updated. The standard dancing behavior position update is prone to fall into the local optimum, resulting in optimization failure. Levy flight is a random behavior strategy that simulates the step size and direction during the search process. Its expression is: Among them, λ follows the standard normal distribution, u follows the mean of 0 and the variance is σ u Gaussian distribution, improved dung beetle optimization algorithm to introduce Levy flight into dancing behavior Among them, x′ i (t+1) is the position function of the t+1th dung beetle after the introduction of Levy flight, α is the step size control, and Levy(u, λ) is the random step size; is the dot multiplication operation. S250 In the reproductive behavior stage, in order to provide a safe environment for offspring, dung beetles will look for a suitable environment to lay eggs. The boundary selection expression of dung beetles when laying eggs is: Lb * =max(X * ×(1-R), Lb) Ub * =max(X * ×(1+R),Ub) Where, R = 1-t / T, Lb * and Ub * are the upper and lower bounds of the reproductive behavior of dung beetles, Lb and Ub are the upper and lower bounds of the solution problem, R = 1-t / T, T is the maximum number of iterations. After the egg-laying area is determined, the dung beetle lays eggs in the egg-laying area. The updated expression of the egg-laying position is B i (t+1)=X*+b1×(B i (t)-Lb*)+b2×(B i (t)-Ub*) Among them, B i (t+1) is the position of the t+1th ball in the i-th iteration; b1 and b2 are 1×d random vectors. In the foraging behavior stage of S260, the foraging behavior of the long and small dung beetles of the standard dung beetle optimization algorithm sets the optimal foraging area, which is expressed as: Lb b =max(X b ×(1-R), Lb) Ub b =max(X b×(1+R),Ub) Among them, X b is the global optimal position, Lb b and Ub b are the upper and lower bounds of the foraging dung beetle's activities, and the position update of the foraging dung beetle is expressed as: x i (t+1)=x i (t)+C1×(x i (t)-Lb b )+C2×(x i (t)-Ub b ) Among them, C1 is a random number that follows a normal distribution, and C2 is a 1×d random vector. The standard foraging behavior of dung beetles has low position diversity and is prone to fall into local optimality. The improved dung beetle optimization algorithm uses adaptive t-distribution perturbation to perturb the foraging behavior of small dung beetles, improving the problem that the dung beetle algorithm is prone to fall into local optimal solutions. The small dung beetle position update formula is transformed into: Among them, t(iter) represents the adaptive t distribution function, x new represents the new position after the adaptive t-distribution perturbation. S270 In the stealing behavior stage, some dung beetles obtain better food by observing the behavior of other dung beetles, which is similar to stealing the dung balls of other dung beetles. The position update expression simulating this behavior is: x i (t+1)=X b +S×g×(|x i (t)-X * |-|x i (t)-X b |) Where g is a 1×d random vector that follows a normal distribution. The optimized VMD hyperparameter expression is: Where k is the number of modes of key parameters for optimizing variational mode decomposition (VMD), and α is the penalty factor for key parameters for optimizing variational mode decomposition (VMD).

[0009] Furthermore, in step 3, the effective modes of VMD are extracted using the composite entropy divergence as the characteristic parameter. The composite entropy divergence is composed of the correlation coefficient and the Raney entropy. The composite entropy divergence expression is: Among them, α is the parameter of the two probability distributions, C c (k) is the person correlation coefficient: Arrange J(k) in descending order, i.e. p=[p(1) p(2)…p(l)], where p(i)>p(i+1) The modes corresponding to the first K items are selected as valid modes.

[0010] Furthermore, the rotating machinery fault diagnosis method based on the minimum mean composite entropy network described in step 4 is characterized in that the CNN model containing the channel attention mechanism described in step 5 includes 4 convolutional layers, 3 pooling layers, a flattening layer, a channel attention mechanism layer, and an output layer. The activation function of the convolutional layer is the relu function, the stride is (3, 3), and zero padding; the pooling layer size is 3×3; the output layer output is 4, and the activation function is the softmax function; Channel Attention MechanismA CAM It can be expressed as: A CAM =S[W d (W mp (X CAM ))]×X CAM Where X CAM is the input of the channel attention mechanism, S(·) is the Softmax function, and W d is the convolution operation, W mp is the average pooling operation.

[0011] Compared with the prior art, the technical progress achieved by the present invention is that: the present invention proposes the minimum average composite entropy that can better balance the noise robustness and fault signal sensitivity as the fitness function of the optimization algorithm, thereby improving the anti-noise and optimization capabilities of the optimization algorithm. The present invention uses tent chaotic mapping, Levy flight and adaptive t distribution as improvement strategies to improve the dung beetle optimization algorithm, avoid falling into local optimal solutions, improve the global search capability of the dung beetle optimization algorithm, and proposes composite entropy divergence as a characteristic parameter for VMD effective mode extraction. The improved algorithm proposed by the present invention comprehensively improves the quality of the input signal of the intelligent fault diagnosis algorithm, thereby improving the accuracy of fault diagnosis, especially the diagnosis accuracy under noisy conditions. 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 1 is a flow chart of the present invention; Figure 2 It is the flow chart of improving the dung beetle optimization algorithm; Figure 3 It is the topological diagram of the CNN-BiLSTM network model structure; Figure 4 This is a comparison of the sensitivity of the Rainey entropy to the kurtosis and divergence to the fault signal. Figure 5 This is a comparison of the robustness of the Rainey entropy, kurtosis and divergence to occasional noise. Figure 6 Comparison of convergence curves of VMD hyperparameters optimized by improved dung beetle algorithm and standard dung beetle algorithm Figure 7 This is a comparison of the composite entropy divergence, Rainey entropy and correlation coefficient for VMD modal analysis. Figure 8 The frequency domain and time domain curves of the reconstructed signal using the method of the present invention are Fig. 9 The diagnostic results of the example using the method of the present invention are Fig.10 The method of the present invention is used to visualize the results of the example fault diagnosis DETAILED DESCRIPTION

[0014] The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without making creative work are within the scope of protection of the present application. In addition, it should be understood that the specific embodiments described herein are only used to illustrate and explain the present application, and are not used to limit the present application.

[0015] The present application provides a rotating machinery fault diagnosis method based on minimum mean composite entropy, which selects the minimum mean composite entropy as the fitness function to realize the optimization of two key parameters of VMD by the improved dung beetle optimization algorithm, selects the composite entropy divergence to realize the extraction and reconstruction of the effective mode of VMD, and simultaneously uses a convolutional neural network with a channel attention mechanism to extract features of the fault signal after significant information extraction, so as to realize the fault diagnosis and classification of rotating machinery parts.

[0016] See also Figure 1 The rotating machinery fault diagnosis method based on minimum mean composite entropy specifically includes the following steps:

[0017] Step 1 S100: Using a vibration sensor to extract the vibration signal of the rotating machinery

[0018] Step 2 S200 uses the minimum average composite entropy as the objective function and uses the improved dung beetle optimization algorithm to optimize the hyperparameters of VMD, namely the number of modes k and the penalty factor α. The average composite entropy with good noise robustness and fault signal sensitivity is obtained. The expression of the average composite entropy is: The composite entropy consists of the Rainey entropy and the envelope entropy. The k-order modal Rainey entropy R after VMD decomposition is e (k) The formula is: where u k (i) is the time series u of its k-order mode after VMD decomposition k The i-th value of . The k-order modal envelope entropy R after VMD decomposition b (k) The formula is: The formula for the k-order modal composite entropy C(k) is: in The normalized Rainey entropy is defined as: Same reason The normalized envelope entropy is defined as

[0019] See also Figure 2 , the improved dung beetle optimization algorithm includes the following steps S210 uses the improved algorithm of tent chaos mapping to initialize the dung beetle population. The improved algorithm of tent chaos mapping is expressed as: Among them, a is the tent chaos coefficient, Z k is a random sequence. k ∈(0,1),Z k+1 According to Z k Generated chaotic sequence S220, using compound entropy as fitness function, calculate the initialization fitness of dung beetle S230, updating the position of the rolling dung beetle, and the position update expression is: x(t+1)=x i (t)+α×c×x i (t-1)+b×Δx Δx=|x i (t)-X w | Among them, x i(t) is the position information of the ith agent at the tth iteration; α is a natural number, indicating the deflection direction; the value is -1 or 1. b represents the deflection coefficient constant, b∈(0, 0.02]; and c∈(0, 1). w represents the global worst position. S240, using Levy flight to improve the position update of dancing dung beetles to solve the problem of being easily trapped in local optimality. The position expression of the improved algorithm is: Among them, x′ i (t+1) is the position function of the t+1th dung beetle after the introduction of Levy flight, α is the step size control, and Levy(u, λ) is the random step size; is the dot multiplication operation. S250 Updated the breeding location of dung beetles The boundary selection formula for breeding dung beetles during the birth of dung beetles is: Lb*=max(X*×(1-R), Lb) Ub*=max(X*×(1+R),Ub) Where, R = 1-t / T, Lb * and Ub * are the upper and lower bounds of the reproductive behavior of dung beetles, Lb and Ub are the upper and lower bounds of the solution problem, R = 1-t / T, T is the maximum number of iterations. After the egg-laying area is determined, the dung beetle lays eggs in the egg-laying area. The updated expression of the egg-laying position is B i (t+1)=X * +b1×(B i (t)-Lb * )+b2×(B i (t)-Ub * ) Among them, B i (t+1) is the position of the t+1th ball in the i-th iteration; b1 and b2 are 1×d random vectors. S260 uses adaptive t distribution to update the foraging dung beetle position and sets the optimal foraging area for the foraging behavior of the dung beetle, which is expressed as Lb b =max(X b ×(1-R), Lb) Ub b =max(X b ×(1+R),Ub) The improved algorithm uses adaptive t-distribution perturbation to perturb the foraging behavior of dung beetles, improving the problem that the dung beetle algorithm is prone to fall into the local optimal solution. The position update expression of the foraging dung beetle in the improved algorithm is: Among them, t(iter) represents the adaptive t distribution function, x new represents the new position after the adaptive t-distribution perturbation. S270 updates the position of the stealing dung beetle, and its expression is: x i (t+1)=X b +S×g×(|x i (t)-X * |-|x i (t)-X b |) g is a 1×d random vector that follows a normal distribution. S280 determines whether the number of iterations reaches the maximum number of iterations. If it reaches the maximum number of iterations, the optimal fitness value is output and the process ends. If it does not reach the maximum number of iterations, the process continues to perform iterative calculation optimization.

[0020] S300 uses compound entropy divergence as a characteristic parameter to extract the effective mode of VMD. The compound entropy divergence is composed of correlation coefficient and Rainey entropy. The compound entropy divergence formula is: Among them, α is the parameter of the two probability distributions, C c (k) is the person correlation coefficient: Arrange J(k) in descending order, i.e. p=[p(1) p(2) …p(l)], where p(i)>p(i+1) The modes corresponding to the first K items are selected as valid modes.

[0021] S400 converts the new vibration signal into a two-dimensional time-frequency image through short-time Fourier transform (STFT)

[0022] S500 inputs the two-dimensional video image into the CNN model with channel attention mechanism to realize the fault diagnosis of rotating machinery. The CNN model with channel attention mechanism includes 4 convolution layers, 3 pooling layers, a flattening layer, a channel attention mechanism layer, and an output layer. The activation function of the convolution layer is the relu function, the stride is (3, 3), and zero padding; the size of the pooling layer is 3×3; the output of the output layer is 4, and the activation function is the softmax function; Channel Attention MechanismA CAM It can be expressed as: A CAM =S[W d (W mp (X CAM ))]×X CAM Among them, X CAM is the input of the channel attention mechanism, S(·) is the Softmax function, and W d is the convolution operation, W mp is the average pooling operation.

[0023] Specific example analysis Take the bearing fault signal data from Case Western Reserve University as an example. The data signal is taken from the vibration signal of the drive end of the experimental system, and the sampling frequency is 12000Hz. This experiment uses discharge machining technology to determine the single point damage of the rolling bearing. The bearing model is 6205-2RS JEM SKF ball bearing, and the structural parameters are shown in Table 1. Taking the outer ring fault signal as an example, Figure 7 is the time domain waveform of the outer ring fault. The minimum average composite entropy is used as the fitness function of the improved dung beetle algorithm. Figure 7 is the fitness function change curve. Table 1 Experimental parameters

[0024] The Raney entropy R proposed in this invention e and the currently popular kurtosis S k and skewness R e By comparison, we can find that the skewness S r It is not sensitive to the defect size, R e With S k has a similar trend. e With S k It is highly sensitive to changes in fault defects and can indicate bearing faults more accurately. The composite entropy, which can better balance noise robustness and sensitivity, is used as the fitness function of the optimization algorithm. Figure 4 Schematic diagram of the change of kurtosis, divergence and Raney entropy proposed by the present invention with the inner ring fault defect of the rolling bearing. Figure 4 As shown in the figure, skewness is not sensitive to defect size, and kurtosis and Reiney entropy have similar change trends. It can be seen that kurtosis and Reiney entropy are highly sensitive to changes in fault defects and can accurately indicate bearing faults. Figure 5 The kurtosis, divergence and Rainey entropy proposed by the present invention are the changes in the response to occasional impulses. Figure 5 The data show that the Raney entropy has excellent robustness to occasional pulse responses, and the kurtosis is highly sensitive to occasional pulse responses. When occasional interference occurs, the kurtosis value increases by more than 3 times, which shows that when occasional pulses occur in rotating machinery, the kurtosis is greatly affected. Although the Raney entropy has a certain sensitivity to occasional pulses, its sensitivity to occasional noise is only 18.4% of the kurtosis. Figure 5This shows that the Rainey entropy has a good ability to balance the sensitivity and occasional robustness of bearing defects.

[0025] The improved dung beetle optimization algorithm of the present invention is compared with the standard dung beetle algorithm. Figure 6 This is a comparison chart of the two algorithms. Figure 6 As shown in the figure, the minimum average composite entropy of the two algorithms gradually decreases with the increase of iterations, and the optimization results of the two algorithms are basically the same. The dung beetle algorithm can reach the minimum value in the 5th iteration, while the fitness function of the dung beetle algorithm needs to reach the minimum value in the 10th iteration. The faster convergence speed can greatly reduce the training time and improve the efficiency. The best optimization parameter of the improved dung beetle optimization algorithm is [α k] = [1360 8]. VMD is decomposed according to the best optimization parameter, and 8 modal components can be obtained.

[0026] The composite entropy divergence proposed in this invention is applied to VMD effective mode analysis, such as Figure 7 It can be found that the Reiney entropy and correlation coefficient of each modal component do not show the same trend of change. For example, IMF1 has a very high Reiney entropy, but a relatively small correlation coefficient. Therefore, it is not comprehensive to evaluate the effective mode based solely on the correlation coefficient or the Reiney entropy.

[0027] According to the effective mode selection index proposed in the present invention, IMF2 and IMF4 are selected as effective IMF components, and the signal is reconstructed. The time domain waveform and envelope spectrum of the reconstructed signal are as follows: Figure 8 As shown, Figure 8 As shown in (a), the reconstructed signal can clearly identify the 4th-order bearing inner race fault frequency information. Figure 8 (b) The repeated impact phenomenon shown is also quite obvious. Figure (9) fully demonstrates the effectiveness of the proposed dung beetle optimized signal feature extraction method based on minimum mean composite entropy. Table 2 Dataset classification

[0028] The reconstructed signal is converted into a two-dimensional image through SETFT, and trained and verified through a CNN network with a channel attention mechanism. The results are as follows Fig. 9 As shown, the fault diagnosis accuracy reaches 99.3%, and the visualization results Fig.10 The same results were also shown.

Claims

1. A rotating machinery fault diagnosis method based on minimum average composite entropy comprises the following steps: S100 step 1: input the vibration signal of the rotating machinery; S200 Step 2: Taking the minimum average composite entropy as the objective function, the improved dung beetle optimization algorithm is used to optimize the hyperparameters of VMD, namely the number of modes k and the penalty factor α; S300 Step 3: Using the composite entropy divergence as a characteristic parameter, extract the effective modes of VMD and reconstruct a new vibration signal; S400 Step 4: Convert the new vibration signal into a two-dimensional time-frequency image through short-time Fourier transform STFT; S500 Step 5: Input the two-dimensional time-frequency image into the CNN model with channel attention mechanism to realize rotating machinery fault diagnosis.

2. The rotating machinery fault diagnosis method based on minimum mean composite entropy according to claim 1, characterized in that: In step 2, the minimum mean composite entropy is used as the fitness function of the optimization algorithm to solve the problem that the current fitness function is difficult to balance the noise robustness and fault signal sensitivity. The composite entropy consists of Rainey entropy and envelope entropy. The k-order modal Rainey entropy R after VMD decomposition is e (k) The formula is: Among them, u k (i) is the time series u of its k-order mode after VMD decomposition k The i-th value of . The k-order modal envelope entropy R after VMD decomposition b (k) The formula is: The formula for the k-order modal composite entropy C(k) is: in, The normalized Rainey entropy is defined as: Same reason The normalized envelope entropy is defined as 3. The rotating machinery fault diagnosis method based on minimum mean composite entropy network according to claim 1, characterized in that: In step 2, the improved dung beetle optimization algorithm mainly includes initialization behavior, fitness behavior, rolling behavior, dancing behavior, breeding behavior, foraging behavior, and stealing behavior. In the initialization behavior stage of S210, the initial population of the standard dung beetle optimization algorithm is randomly generated, which has the disadvantages of poor population diversity and uneven distribution. The improved dung beetle optimization initialization behavior uses tent chaotic mapping to initialize the dung beetle population. Its mathematical expression is: Where a is the chaos coefficient of the tent chaotic map; Z k is a random sequence, Z k ∈(0,1); Z k+1 According to Z k Generated chaotic sequence. S220 uses compound entropy as the fitness function in the fitness behavior stage to calculate the initial fitness of the dung beetle S230 In the ball rolling behavior stage, the ball rolling behavior of the dung beetle depends on celestial clues to ensure that the dung ball moves in a straight line, and considering the influence of light intensity on its ball rolling path, the ball rolling behavior position update expression of the dung beetle is: x(t+1)=x i (t)+α×c×x i (t-1)+b×Δx Δx=|x i (t)-X w | Among them, x i (t) is the position information of the ith agent at the tth iteration; α is a natural number, indicating the deflection direction; the value is -1 or 1. b represents the deflection coefficient constant, b∈(0, 0.02]; and c∈(0, 1). w represents the global worst position. In the dancing behavior stage S240, when the dung beetle encounters an obstacle, it can change direction through dancing behavior to obtain a new route. The position update formula of the dancing behavior of the standard dung beetle optimization algorithm is expressed as: x i (t+1)=x i (t)+tan(β)|x i (t)-x i (t-1)| Among them, β∈[0,π], when β=0,π / 2, orπ, the dancing behavior does not work and the position of the dung beetle is not updated. The standard dancing behavior position update is prone to fall into the local optimum, resulting in optimization failure. Levy flight is a random behavior strategy that simulates the step size and direction during the search process. Its expression is: Among them, λ follows the standard normal distribution, u follows the mean of 0 and the variance is σ u Gaussian distribution, improved dung beetle optimization algorithm to introduce Levy flight into dancing behavior Among them, x' i (t+1) is the position function of the t+1th dung beetle after the introduction of Levy flight, α is the step size control, and Levy(u,λ) is the random step size; is the dot multiplication operation. In the S250 reproductive behavior stage, in order to provide a safe environment for their offspring, dung beetles will look for a suitable environment to lay eggs. When laying eggs, dung beetles select the egg-laying site, and the boundary selection expression is: Lb * =max(X * ×(1-R),Lb) Ub * =max(X * ×(1+R),Ub) Where, R = 1-t / T, Lb * and Ub * are the upper and lower bounds of the reproductive behavior of dung beetles, Lb and Ub are the upper and lower bounds of the problem, The lower bound is T, and T is the maximum number of iterations. After the egg-laying area is determined, the dung beetle lays eggs in the egg-laying area. The updated expression of the egg-laying position is B i (t+1)=X * +b1×B i (t-Lb * )+b2×(B i (t)-Ub * ) Among them, B i (t+1) is the position of the t+1th egg in the i-th iteration; b1 and b2 are 1×d random vectors. In the foraging behavior stage of S260, the standard dung beetle optimization algorithm sets the optimal foraging area for the foraging behavior of the grown-up dung beetle, which is expressed as: Lb b =max(X b ×(1-R),Lb) Ub b =max(X b ×(1+R),Ub) Among them, X b is the global optimal position, Lb b and Ub b are the upper and lower bounds of the foraging dung beetle's activities, and the position update of the foraging dung beetle is expressed as: x i (t+1)=x i (t)+C1×(x i (t)-Lb b )+C2×(x i (t)-Ub b ) Among them, C1 is a random number that follows a normal distribution, and C2 is a 1×d random vector. The standard foraging behavior of dung beetles has low position diversity and is prone to fall into local optimality. The improved dung beetle optimization algorithm uses adaptive t-distribution perturbation to perturb the foraging behavior of small dung beetles, improving the problem that the dung beetle algorithm is prone to fall into local optimal solutions. The small dung beetle position update formula is transformed into: Among them, t(iter) represents the adaptive t distribution function, x new represents the new position after the adaptive t-distribution perturbation. S270 In the stealing behavior stage, some dung beetles obtain better food by observing the behavior of other dung beetles, which is similar to stealing the dung balls of other dung beetles. The position update expression simulating this behavior is: x i (t+1)=X b +S×g×(|x i (t)-X * |-(|x i (t)-X b |) Where g is a 1×d random vector that follows a normal distribution. The optimized VMD hyperparameter expression is: Where k is the number of modes of key parameters for optimizing variational mode decomposition (VMD), and α is the penalty factor for key parameters for optimizing variational mode decomposition (VMD).

4. The rotating machinery fault diagnosis method based on minimum mean composite entropy network according to claim 1, characterized in that: In step 3, the effective modes of VMD are extracted using the composite entropy divergence as the characteristic parameter. The composite entropy divergence consists of the correlation coefficient and the Rainey entropy. The composite entropy divergence expression is: Among them, α is the parameter of the two probability distributions, C c (k) is the person correlation coefficient: Arrange J(k) in descending order, that is, p=[p(1) p(2) …p(l), where p(i)>p(i+1) The modes corresponding to the first K items are selected as valid modes.

Images