Rolling bearing fault diagnosis method and device based on composite multi-scale attention entropy and optimized SVM and medium

Abstract

The invention relates to a rolling bearing fault diagnosis method and device based on a composite multi-scale attention entropy and an optimized SVM, and a medium, and the method comprises the steps: collecting vibration signals of a rolling bearing in different states, obtaining the damage size grade in each state according to the vibration signals, and obtaining a fault data set; performing coarse graining processing on the vibration signal to obtain a coarse grain sequence, and calculating an entropy sequence of the coarse grain sequence under different scale factors by adopting a composite multi-scale attention entropy algorithm improved by a fractional order algorithm; entropy values of a plurality of first scale factors in the entropy sequence are selected as bearing fault feature vectors; optimizing hyper-parameters of the support vector machine by using a collaborative group optimization algorithm; and constructing a support vector machine classification model based on the optimal hyper-parameter, and performing fault prediction and judgment on the test sample. Compared with the prior art, the method has the advantages of high noise robustness, high global convergence, high adaptability and the like.

Description

Technical Field

[0001] The present invention relates to the field of fault diagnosis, and in particular to a rolling bearing fault diagnosis method, equipment and medium based on composite multi-scale attention entropy and optimized SVM. Background Art

[0002] Rolling bearings are core components of rotating machinery, and their operating status directly affects the performance, reliability, and service life of the entire machine. Therefore, rolling bearing fault diagnosis is of vital importance. Analysis methods based on vibration signals have become a common means of bearing fault diagnosis due to their effectiveness and practicality. However, vibration signals collected in actual industrial environments are often interfered with by strong noise, making it difficult to extract effective fault features. This is one of the difficulties in current bearing fault diagnosis. At the same time, during bearing operation, factors such as friction, clearance, and nonlinear stiffness can cause the vibration signal to exhibit significant nonlinear and non-steady-state characteristics. Traditional linear system analysis methods, such as frequency domain analysis based on Fourier transform, have difficulty in fully characterizing the dynamic characteristics of such complex signals. They have obvious limitations in feature extraction and cannot accurately reflect the true operating status of the bearing.

[0003] To address these issues, nonlinear dynamic analysis methods, such as approximate entropy, sample entropy, and multiscale entropy, have been introduced into the field of bearing fault diagnosis. Approximate entropy provides a basis for measuring time series complexity. Sample entropy reduces its dependence on sequence length and optimizes self-matching defects. Fuzzy entropy improves consistency by introducing fuzzy set theory. However, these methods all require manual parameter setting before feature extraction, and parameter selection directly affects the diagnostic results. Note that entropy, as a new time series complexity measurement method, does not require hyperparameter setting and is robust to data length. However, it is sensitive to noise and has difficulty effectively capturing the transient impact characteristics of non-stationary signals in low signal-to-noise ratio environments, resulting in a decrease in the quality of fault feature extraction. In the machine learning classification stage, support vector machines (SVMs) are widely used due to their advantages in processing nonlinear and small sample data. However, their classification performance is highly dependent on hyperparameter settings, such as penalty parameters and kernel function parameters. Traditional optimization algorithms, such as particle swarm optimization and genetic algorithms, are prone to falling into local optimality when solving high-dimensional and nonlinear problems. In addition, the parameter tuning process relies on experience and has high computational costs, which affects the accuracy and efficiency of fault diagnosis.

[0004] Chinese patent application CN115406657A discloses a rolling bearing fault diagnosis method that effectively eliminates noise in vibration signals through wavelet threshold denoising; decomposes the denoised signal based on the empirical wavelet transform, and proposes the attention entropy of the IMF component as a feature vector. The attention entropy can effectively distinguish fault categories. However, this application relies on wavelet threshold denoising, which is not effective in suppressing noise interference in strong noise environments such as low signal-to-noise ratio environments. In addition, the marine predator algorithm used may converge slowly during parameter optimization and easily fall into local optimality, affecting model training efficiency and diagnostic accuracy. Therefore, how to improve the feature extraction capability of nonlinear and non-steady-state signals in a strong noise background, and how to efficiently optimize the hyperparameters of the machine learning model to accurately diagnose rolling bearing faults are technical problems that need to be solved. Summary of the Invention

[0005] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and to provide a rolling bearing fault diagnosis method, equipment and medium based on composite multi-scale attention entropy and optimized SVM. The entropy value calculation method of the composite multi-scale attention entropy algorithm is improved by using a fractional order algorithm, and the hyperparameters of the support vector machine are optimized by using a collaborative group optimization algorithm to obtain accurate rolling bearing fault diagnosis results under a noisy background.

[0006] The purpose of the present invention can be achieved by the following technical solutions:

[0007] According to one aspect of the present invention, a rolling bearing fault diagnosis method based on composite multi-scale attention entropy and optimized SVM is provided, and the specific steps include:

[0008] S1. Collect vibration signals of the rolling bearing in different states, including normal state, inner ring fault, outer ring fault, and rolling element fault, and obtain the damage size level in each state based on the vibration signals to obtain a fault data set;

[0009] S2. Coarse-graining the vibration signal to obtain a coarse-grained sequence, and calculating the entropy sequence of the coarse-grained sequence under different scale factors using a composite multi-scale attention entropy algorithm improved by a fractional order algorithm; the composite multi-scale attention entropy algorithm improved by a fractional order algorithm introduces a fractional order into the calculation of Shannon entropy and uses a composite multi-scale analysis method to generate the entropy sequence;

[0010] S3, selecting the entropy values of the first several scale factors in the entropy sequence as the bearing fault feature vectors, and dividing them into training samples and test samples;

[0011] S4. Optimize the hyperparameters of the support vector machine using a collaborative swarm optimization algorithm. Use the support vector machine to classify the training set, use the training set accuracy as the fitness function, and update the particle position and velocity through the dynamic attraction equation until the optimal hyperparameters are obtained.

[0012] S5. Construct a support vector machine classification model based on the optimal hyperparameters, perform fault prediction on the test samples, and identify the fault type and working status of the rolling bearing according to the prediction results.

[0013] Furthermore, the specific steps of the composite multi-scale attention entropy algorithm in S2 include: coarse-graining the vibration signal to obtain coarse-grained sequences under different scale factors, extracting the local peak points of the coarse-grained sequences in the form of time series under each scale factor as key points, and arranging them in combinations of minimum-minimum, minimum-maximum, maximum-minimum and maximum-maximum to generate a new time series; calculating the entropy value of the new time series by the fractional-order Shannon entropy formula, taking the average to obtain the fractional-order attention entropy, and traversing all scale factors to obtain a composite multi-scale fractional-order attention entropy sequence.

[0014] Furthermore, the fractional-order attention entropy is calculated by the fractional-order improved entropy value, and the expression is:

[0015] FrAE(X)=(H α 1+H α 2+H α 3+H α 4) / 4,

[0016] Among them, FrAE(X) is the fractional attention entropy, H α 1. H α 2. H α 3 and H α 4 are the entropy values of the four new time series, that is, the entropy value of the fractional order improvement, and the calculation method is:

[0017]

[0018] Among them, H α (x) is the fractional-order improved entropy value, p(x) is the probability of x occurring in the new time series, α is the fractional order, and its value is [0, 1]. When the fractional order is 0, the fractional Shannon entropy is the classical Shannon entropy, Γ(·) and ψ(·) are the derivative functions of the gamma function and the logarithm of the gamma function, respectively.

[0019] Furthermore, the fractional-order attention entropy values of different sequences under different scale factors are calculated and averaged to obtain the composite multi-scale fractional-order attention entropy sequence CMFrAE(X,τ) of the original coarse-grained sequence under the corresponding scale factor. The expression is:

[0020]

[0021] Where τ is the scale factor, X is the time series, For a coarse-grained sequence, for a time series X with a length of N, the expression for the coarse-grained processing is:

[0022]

[0023] Furthermore, the collaborative swarm optimization algorithm in S4 optimizes the hyperparameters in the support vector machine algorithm. The specific steps include: initializing the particle swarm position and velocity, setting the individual and global optimal positions and fitness values; for each particle in the swarm, classifying the training set through the support vector machine algorithm, and using the accuracy of the training set as the fitness function; updating the particle position and velocity through the velocity update equation and the dynamic attraction equation; according to the preset number of iterations and stopping conditions, iteratively updating until the stopping conditions are met to obtain the optimal hyperparameters.

[0024] Furthermore, the initialization expression is:

[0025] T=rand(N,Dim)*(UB-LB)+LB,

[0026] Where T is the particle population, rand(N,Dim) is an N×Dim random matrix, N is the number of particles, Dim is the dimension of the parameter to be optimized, and UB and LB are the upper and lower bounds respectively.

[0027] Furthermore, the dynamic attraction equation adaptively guides particles according to the local and global attraction of the position, including inertia weight, personal optimal coefficient, global optimal coefficient, dynamic attraction coefficient, adaptive neighborhood interaction coefficient and diversity maintenance coefficient.

[0028] Furthermore, the inertia weight IWV is obtained according to the fitness, which makes the group concentrate the search space, and the expression is:

[0029] IWV=w(t)*v(i,j),

[0030] Among them, v(i,j) is the current velocity of the particle, w(t) is the inertia weight, and the specific expression is:

[0031] w(t+1)=w(t)*(1-exp(-k*t)),

[0032] Where k is the decay rate constant and t is the current iteration;

[0033] The expression of the personal best coefficient PBC is:

[0034] PBC=r1*(eps*rand(pbest)-T i ),

[0035] Among them, r1 is a random value, rand(pbest) is a random solution of the current position and velocity candidate solution, and pbest is the particle T i Best historical position

[0036] The expression of the global optimal coefficient GBC is:

[0037] GBC=r2*gbes t -T i ,

[0038] Among them, r2 is a random value, gbes t is the current global optimal solution,

[0039] The expression of the dynamic attraction coefficient DAC is:

[0040]

[0041] Among them, r3 is a random value, attract i is the position with the highest local attraction value in the neighborhood of the i-th particle, c1 is the scaling factor,

[0042] The expression of the adaptive neighborhood interaction coefficient ANIC is:

[0043] ANIC=r4*rand(bestf)-bestf i ,

[0044] Among them, r4 is a random value, rand(bestf) is the random fitness value from the current fitness solution, bestf i is the fitness value of the i-th position and velocity solution

[0045] The expression of the diversity maintenance coefficient DMC is:

[0046]

[0047] Among them, r5 is a random value, diversity i is the position of the diversity in the neighborhood of the i-th particle in the particle swarm, and c2 is the scaling factor.

[0048] According to a second aspect of the present invention, an electronic device is provided, comprising a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the method when executing the program.

[0049] According to a third aspect of the present invention, a computer-readable storage medium is provided, on which a computer program is stored, and when the program is executed by a processor, the method described above is implemented.

[0050] Compared with the prior art, the present invention has the following beneficial effects:

[0051] (1) Improved noise robustness: By introducing a fractional-order improved composite multi-scale attention entropy algorithm and incorporating fractional-order parameters into the Shannon entropy calculation, the algorithm is given adaptive capabilities and can effectively capture transient impact features in non-stationary signals. Compared with traditional attention entropy, the fractional-order improved composite multi-scale attention entropy can still maintain high-precision feature extraction in a strong noise environment without relying on complex pre-noise reduction processing, thus solving the problem of noise sensitivity of existing technologies.

[0052] (2) Enhanced parameter optimization efficiency and global convergence: The collaborative swarm optimization algorithm is used to optimize the support vector machine hyperparameters. The dynamic attraction equation and the adaptive neighborhood interaction mechanism are used to balance global exploration and local development. The collaborative swarm optimization algorithm combines particle collaboration and diversity maintenance strategies to avoid falling into local optimality, significantly improving the convergence speed and accuracy of parameter optimization, thereby optimizing the generalization ability of the classification model.

[0053] (3) Balancing diagnostic accuracy and real-time performance: Combining the composite multi-scale fractional-order attention entropy feature extraction with the support vector machine classification model optimized by the collaborative group optimization algorithm, a comprehensive characterization and efficient classification of multi-level and multi-scale fault features is achieved. This method reduces computational complexity through a lightweight model structure while ensuring high diagnostic accuracy, and is suitable for real-time fault diagnosis needs in industrial scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 Flowchart of the rolling bearing fault diagnosis method based on composite multi-scale attention entropy and optimized SVM;

[0055] Figure 2 The fractional attention entropy graphs of ten different bearing faults at twenty different scales in the embodiment;

[0056] Figure 3 This is the recognition result diagram of SVM optimized based on the composite multi-scale fractional-order attention entropy algorithm and collaborative group optimization algorithm;

[0057] Figure 4 This is the recognition result diagram of SVM optimized based on the original algorithm and the collaborative group optimization algorithm;

[0058] Figure 5 This is the recognition result diagram of SVM based on the improved algorithm and particle swarm optimization;

[0059] Figure 6This is the recognition result diagram of SVM optimized based on the improved algorithm and the gray wolf algorithm;

[0060] Figure 7 This is the recognition result diagram of SVM based on improved algorithm and genetic algorithm optimization;

[0061] Figure 8 The results of using different optimization algorithms combined with different feature extraction methods under different signal-to-noise ratios are shown. DETAILED DESCRIPTION

[0062] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0063] In the attention entropy algorithm, a fractional-order algorithm is used to improve the entropy calculation method of the composite multi-scale attention entropy algorithm. This improved attention entropy is calculated at different scales. The introduction of fractional-order calculus improves the algorithm's performance, maintaining excellent feature extraction capabilities even in noisy environments, addressing the noise sensitivity of traditional attention entropy. Furthermore, in the support vector machine algorithm, hyperparameters significantly influence classification results. Therefore, a swarm optimization algorithm is used to optimize these hyperparameters in the support vector machine algorithm, improving accuracy and enabling more accurate assessment of the rolling bearing condition.

[0064] like Figure 1 As shown in FIG, a rolling bearing fault diagnosis method based on composite multi-scale attention entropy and optimized SVM is shown, and the specific steps include:

[0065] S1. Collect vibration signals of rolling bearings in different states, including normal state, inner ring fault, outer ring fault, and rolling element fault. Determine the damage size level in each state based on the vibration signals to obtain a fault data set.

[0066] S2. Coarse-graining the vibration signal to obtain a coarse-grained sequence, and using a composite multi-scale attention entropy algorithm improved by a fractional order algorithm to calculate the entropy sequence of the coarse-grained sequence under different scale factors; the composite multi-scale attention entropy algorithm improved by a fractional order algorithm introduces fractional orders in the calculation of Shannon entropy, and uses a composite multi-scale analysis method to generate entropy sequences;

[0067] S3, selecting the entropy values of the first several scale factors in the entropy sequence as the bearing fault feature vectors, and dividing them into training samples and test samples;

[0068] S4. Optimize the hyperparameters of the support vector machine using a collaborative swarm optimization algorithm. Use the support vector machine to classify the training set, use the training set accuracy as the fitness function, and update the particle position and velocity through the dynamic attraction equation until the optimal hyperparameters are obtained.

[0069] S5. Build a support vector machine classification model based on the optimal hyperparameters, perform fault prediction on the test samples, and identify the fault type and working status of the rolling bearing based on the prediction results.

[0070] In S1, the time domain signals of the rolling bearing are obtained under normal conditions, inner ring fault, outer ring fault, and rolling element fault conditions. The damage size of the bearing fault is divided into three levels: 0.007 inches, 0.014 inches, and 0.028 inches, respectively. A total of 10 types of vibration signals are obtained.

[0071] For an N-point time series, each point in the time series is considered a system, and changes in its state can be seen as adjustments made by the system to its environment. Peak points can effectively characterize changes in the upper and lower bounds of local states, so local peak points are defined as key points. The specific steps of the composite multi-scale attention entropy algorithm in S2 include coarse-graining the vibration signal to obtain coarse-grained sequences at different scale factors. The local peak points of the coarse-grained time series at each scale factor are extracted as key points, and then arranged in combinations of minimum-minimum, minimum-maximum, maximum-minimum, and maximum-maximum to generate a new time series. The entropy of the new time series is calculated using the fractional Shannon entropy formula, and the fractional attention entropy is obtained by taking the average. The composite multi-scale fractional attention entropy sequence is then generated by traversing all scale factors.

[0072] The fractional-order attention entropy is calculated from the fractional-order improved entropy value, and the expression is:

[0073] FrAE(X)=(H α 1+H α 2+H α 3+H α 4) / 4,

[0074] Among them, FrAE(X) is the fractional attention entropy, H α 1. H α 2. H α 3 and H α 4 are the entropy values of the four new time series, that is, the entropy value of the fractional order improvement, and the calculation method is:

[0075]

[0076] Among them, H α(x) is the fractional-order improved entropy value, p(x) is the probability of x occurring in the new time series, α is the fractional order, and its value is [0, 1]. When the fractional order is 0, the fractional Shannon entropy is the classical Shannon entropy, Γ(·) and ψ(·) are the derivative functions of the gamma function and the logarithm of the gamma function, respectively.

[0077] Calculate the fractional attention entropy values of different sequences under different scale factors and take the average value to obtain the composite multi-scale fractional attention entropy sequence CMFrAE(X,τ) of the original coarse-grained sequence under the corresponding scale factor. The expression is:

[0078]

[0079] Where τ is the scale factor, X is the time series, For a coarse-grained sequence, for a time series X with a length of N, the expression for coarse-grained processing is:

[0080]

[0081] The composite multi-scale fractional attention entropy on the first twenty scales is selected as the bearing fault feature vector.

[0082] In S4, a collaborative swarm optimization algorithm is used to optimize the hyperparameters in the support vector machine algorithm. The specific steps include initializing the particle swarm position and velocity, setting the individual and global optimal positions and fitness values; for each particle in the swarm, the training set is classified using the support vector machine algorithm, and the accuracy of the training set is used as the fitness function; the particle position and velocity are updated using the velocity update equation and the dynamic attraction equation; and according to the preset number of iterations and stopping conditions, the iterative update is performed until the stopping conditions are met to obtain the optimal hyperparameters.

[0083] The initialization expression is:

[0084] T=rand(N,Dim)*(UB-LB)+LB,

[0085] Where T is the particle population, rand(N,Dim) is an N×Dim random matrix, N is the number of particles, Dim is the dimension of the parameter to be optimized, and UB and LB are the upper and lower bounds respectively.

[0086] The expression for updating the candidate solution Tnew(i,j) is:

[0087] Tnew(i,j)=T(i,j)+v(i,j),

[0088] Among them, T(i,j) is the current position of the particle, and v(i,j) is the current velocity of the particle.

[0089] Dynamic attraction equation v new(i, j) adaptively guides particles according to the local and global attraction of the position, including inertia weight, personal best coefficient, global best coefficient, dynamic attraction coefficient, adaptive neighborhood interaction coefficient and diversity maintenance coefficient, which are expressed as:

[0090] v new (i,j)=IWV+PBC+GBC+DAC+ANIC+MDC.

[0091] The inertia weight IWV is obtained according to the fitness, which makes the group concentrate the search space. The expression is:

[0092] IWV=w(t)*v(i,j),

[0093] Among them, ν(i,j) is the current velocity of the particle, and w(t) is the inertia weight, which is used to dynamically control the balance between exploration and exploitation. The specific expression is:

[0094] w(t+1)=w(t)*(1-exp(-k*t)),

[0095] Where k is the decay rate constant and t is the current iteration;

[0096] The expression of personal best coefficient PBC is:

[0097] PBC=r1*(eps*rand(pbest)-T i ),

[0098] Among them, r1 is a random value, rand(pbest) is a random solution of the current position and velocity candidate solution, and pbest is the particle T i Best historical position

[0099] The expression of the global optimal coefficient GBC is:

[0100] GBC=r2*gbes t -T i ,

[0101] Among them, r2 is a random value, gbes t is the current global optimal solution,

[0102] The expression of dynamic attraction coefficient DAC is:

[0103]

[0104] Among them, r3 is a random value, attract i is the position with the highest local attraction value in the neighborhood of the i-th particle, c1 is the scaling factor,

[0105] The expression of the adaptive neighborhood interaction coefficient ANIC is:

[0106] ANIC=r4*rand(bestf)-bestf i ,

[0107] Among them, r4 is a random value, rand(bestf) is the random fitness value from the current fitness solution, bestf i is the fitness value of the i-th position and velocity solution

[0108] The expression of diversity maintenance coefficient DMC is:

[0109]

[0110] Among them, r5 is a random value, diversity i is the position of the diversity in the neighborhood of the i-th particle in the particle swarm, and c2 is the scaling factor.

[0111] This example uses deep groove ball bearings for testing, employing electrospark machining (EDM) to create single-point faults of varying sizes. The bearing speed is 1797 rpm, and the sampling frequency is 12 kHz. Ten fault types, including normal state, inner race fault, outer race fault, and rolling element fault, are analyzed, along with data of varying sizes. Each type contains 100 data points, each with a length of 8000 points, for a total of 1000 data points. The specific test data set is shown in Table 1.

[0112] Table 1 Experimental dataset

[0113]

[0114] In order to verify the feature extraction ability of the algorithm under strong noise background, noise with different signal-to-noise ratios was added to the data set, so that the signal-to-noise ratios of the signals after adding noise were 0dB, 3dB, and 5dB respectively, generating three data sets. The following experimental results are based on the analysis of the data set with SNR=3dB.

[0115] Calculate the composite multi-scale fractional-order attention entropy of the bearing vibration signal as the input fault feature matrix, the composite multi-scale fractional-order attention entropy sequence CMFrAE(X,τ):

[0116]

[0117] Among them, τ=20, fractional order fo=0.1, and N=8000.

[0118] The vibration data of the ten different rolling bearing states mentioned above are analyzed using composite multi-scale fractional attention entropy. The results are as follows: Figure 2As shown in the figure, although the signals under different fault states have similar change trends at different scales, the entropy values of different fault states at different scales are different, indicating that the complexity of the signals under different fault states is different. Therefore, the composite multi-scale fractional-order attention entropy is an effective method to reflect and distinguish the characteristics of rolling bearing faults. By introducing fractional-order parameters, the algorithm is given adaptive capabilities, effectively capturing fault impact signals and maintaining excellent feature extraction capabilities in a noisy background, solving the defect of traditional attention entropy being sensitive to noise.

[0119] The composite multi-scale fractional attention entropy at the first twenty scales was selected as the bearing fault feature vector. The obtained bearing fault features were divided into training samples and test samples. 100 sets of data were collected for each of the 10 bearing states, of which 70 sets were used for training and the remaining 30 sets were used for testing.

[0120] To reduce the impact of human factors on fault identification and verify the superiority of composite multi-scale fractional attention entropy over the original composite multi-scale attention entropy algorithm in extracting bearing condition information, a support vector machine was used to automatically diagnose rolling bearing faults. A radial basis kernel function was used as the kernel function, and a collaborative group optimization algorithm was used to find the optimal penalty and kernel parameters.

[0121] In this example, the training samples are input into the support vector machine for training, and the trained model is used to predict the test data. All the training samples are correctly identified, and the recognition results of the test samples are as follows: Figure 3 As shown in the figure, 6 of the 300 test samples were misclassified, and the prediction accuracy was 98%.

[0122] In order to highlight the superiority of the composite multi-scale fractional attention entropy algorithm, the original composite multi-scale entropy of the above data was calculated, and the same steps were used to train and establish the support vector machine prediction model, and then the test data was classified and recognized. All training samples were correctly recognized, and the recognition results of the test samples are shown in Figure 4 , a total of 20 out of 300 test samples were misclassified, with a classification accuracy of 93.33%. This comparison result also shows that the composite multi-scale fractional-order attention entropy algorithm can more effectively extract the state information implicit in the bearing vibration signal under noisy background than the original algorithm, and thus can more accurately identify different bearing fault states under noisy background.

[0123] In order to verify the superiority of the optimization algorithm used, different optimization algorithms are used for comparative verification, including particle swarm optimization algorithm, gray wolf optimization algorithm and genetic algorithm. The results are as follows: Figure 5 、 Figure 6 、 Figure 7As shown in the figure, the result of the particle swarm algorithm is that 7 of the 300 test samples were misclassified, and the prediction accuracy was 97.67%. The result of the gray wolf optimization algorithm is that 12 of the 300 test samples were misclassified, and the prediction accuracy was 96%. The result of the genetic optimization algorithm is that 16 of the 300 test samples were misclassified, and the prediction accuracy was 94.67%. This comparison result also shows that the optimization algorithm used can better perform parameter optimization, and thus can more accurately identify different bearing fault states. The results of using different optimization algorithms combined with different feature extraction methods under different signal-to-noise ratios are shown in the figure. Figure 8 And as shown in Table 2:

[0124] Table 2 Results of using different optimization algorithms combined with different feature extraction methods under different signal-to-noise ratios

[0125]

[0126]

[0127] As can be seen from the results in the figure and table, the method proposed in the present invention can effectively diagnose bearing data sets under different signal-to-noise ratios. The fault diagnosis model based on the composite multi-scale fractional-order attention entropy algorithm shows significant anti-interference ability under all experimental conditions, and its diagnostic accuracy is always better than the traditional composite multi-scale attention entropy algorithm. This comparison result shows that the composite multi-scale fractional-order attention entropy algorithm can more effectively extract the state information implicit in the bearing vibration signal under the noise background than the original algorithm, and thus can more accurately identify different bearing fault states under the noise background. The classification results of the optimization algorithm used in the present invention are also significantly higher than other optimization algorithms, verifying that the optimization algorithm used in the present invention can better perform parameter optimization, thereby improving accuracy and more accurately evaluating the operating status of the bearing.

[0128] This embodiment is based on a rolling bearing fault diagnosis method based on composite multi-scale attention entropy and optimized SVM, which improves the traditional attention entropy algorithm, better handles the non-stationarity and complexity of data, and maintains high-precision feature extraction in a noisy environment. The improved algorithm gives the algorithm adaptive capabilities by introducing fractional-order parameters, effectively capturing fault impact signals. On this basis, the algorithm adopts a composite multi-scale analysis method to calculate entropy values from different scales, which can better extract fault features and improve the reliability of extracted entropy value features, thereby improving the feature extraction capability of the algorithm under strong noise backgrounds. In the model training set, the collaborative group optimization algorithm is used to optimize the parameters of the support vector machine. The collaborative group optimization algorithm combines the principles of group intelligence and collaborative cooperation, adopts a collaborative mechanism, and particles exchange information with each other, learn from each other, and improve search behavior, so that the algorithm avoids falling into local optimality, improves the convergence speed, effectively finds the optimal solution, obtains the optimal parameters, and has a higher recognition rate in the fault pattern recognition process.

[0129] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the described module can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0130] The electronic device of the present invention includes a central processing unit (CPU), which can perform various appropriate actions and processes according to computer program instructions stored in a read-only memory (ROM) or loaded from a storage unit into a random access memory (RAM). In the RAM, various programs and data required for device operation can also be stored. The CPU, ROM, and RAM are connected to each other via a bus. An input / output (I / O) interface is also connected to the bus.

[0131] Multiple components in the device are connected to the I / O interface, including: input units, such as a keyboard, mouse, etc.; output units, such as various types of displays, speakers, etc.; storage units, such as magnetic disks, optical disks, etc.; and communication units, such as network cards, modems, wireless communication transceivers, etc. The communication unit allows the device to exchange information / data with other devices via computer networks such as the Internet and / or various telecommunications networks. The processing unit performs the various methods and processes described above, such as the method of the present invention. For example, in some embodiments, the method of the present invention can be implemented as a computer software program that is tangibly contained in a machine-readable medium, such as a storage unit. In some embodiments, part or all of the computer program can be loaded and / or installed onto the device via ROM and / or the communication unit. When the computer program is loaded into RAM and executed by the CPU, one or more steps of the method of the present invention described above can be performed. Alternatively, in other embodiments, the CPU can be configured to perform the method of the present invention by any other suitable means (e.g., by means of firmware).

[0132] The functions described above herein may be performed, at least in part, by one or more hardware logic components. For example, and without limitation, exemplary types of hardware logic components that may be used include: field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chip (SOCs), complex programmable logic devices (CPLDs), and the like.

[0133] The program code for implementing the method of the present invention can be written in any combination of one or more programming languages. Such program code can be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device so that when the program code is executed by the processor or controller, the functions / operations specified in the flow chart and / or block diagram are implemented. The program code can be executed entirely on the machine, partially on the machine, as a stand-alone software package, partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0134] In the context of the present invention, machine-readable medium can be a tangible medium that can contain or store a program for use with an instruction execution system, device or equipment or used in combination with an instruction execution system, device or equipment. Machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. Machine-readable medium can include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. More specific examples of machine-readable storage media can include electrical connections based on one or more lines, portable computer disks, hard disks, random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memory), optical fibers, portable compact disk read-only memories (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0135] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and such modifications or substitutions are intended to be within the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be subject to the scope of protection of the claims.

Claims

1. A rolling bearing fault diagnosis method based on composite multi-scale attention entropy and optimized SVM, characterized in that: The specific steps include: S1. Collect vibration signals of the rolling bearing in different states, including normal state, inner ring fault, outer ring fault, and rolling element fault, and obtain the damage size level in each state based on the vibration signals to obtain a fault data set; S2. Coarse-graining the vibration signal to obtain a coarse-grained sequence, and calculating the entropy sequence of the coarse-grained sequence under different scale factors using a composite multi-scale attention entropy algorithm improved by a fractional order algorithm; the composite multi-scale attention entropy algorithm improved by a fractional order algorithm introduces a fractional order into the calculation of Shannon entropy and uses a composite multi-scale analysis method to generate the entropy sequence; S3. Select the entropy values of the first several scale factors in the entropy sequence as the bearing fault feature vectors, and divide them into training samples and test samples; S4. Optimize the hyperparameters of the support vector machine using a collaborative swarm optimization algorithm. Use the support vector machine to classify the training set, use the training set accuracy as the fitness function, and update the particle position and velocity through the dynamic attraction equation until the optimal hyperparameters are obtained. S5. Construct a support vector machine classification model based on the optimal hyperparameters, perform fault prediction on the test samples, and identify the fault type and working status of the rolling bearing according to the prediction results.

2. The rolling bearing fault diagnosis method based on composite multi-scale attention entropy and optimized SVM according to claim 1 is characterized in that: The specific steps of the composite multi-scale attention entropy algorithm in S2 include: coarse-graining the vibration signal to obtain coarse-grained sequences under different scale factors, extracting the local peak points of the coarse-grained sequences in the form of time series under each scale factor as key points, and arranging them in combinations of minimum-minimum, minimum-maximum, maximum-minimum and maximum-maximum to generate a new time series; calculating the entropy value of the new time series by the fractional-order Shannon entropy formula, taking the average to obtain the fractional-order attention entropy, and traversing all scale factors to obtain a composite multi-scale fractional-order attention entropy sequence.

3. The rolling bearing fault diagnosis method based on composite multi-scale attention entropy and optimized SVM according to claim 2 is characterized in that: The fractional-order attention entropy is calculated by the fractional-order improved entropy value, and the expression is: FrAE(X)=(H α 1+H α 2+H α 3+H α 4) / 4, Among them, FrAE(X) is the fractional attention entropy, H α 1. H α 2. H α 3 and H α 4 are the entropy values of the four new time series, that is, the entropy value of the fractional order improvement, and the calculation method is: Among them, H α (x) is the fractional-order improved entropy value, p(x) is the probability of x occurring in the new time series, α is the fractional order, and its value is [0, 1]. When the fractional order is 0, the fractional Shannon entropy is the classical Shannon entropy, Γ(·) and ψ(·) are the derivative functions of the gamma function and the logarithm of the gamma function, respectively.

4. The rolling bearing fault diagnosis method based on composite multi-scale attention entropy and optimized SVM according to claim 3 is characterized in that: Calculate the fractional attention entropy values of different sequences under different scale factors and take the average value to obtain the composite multi-scale fractional attention entropy sequence CMFrAE(X,τ) of the original coarse-grained sequence under the corresponding scale factor. The expression is: Where τ is the scale factor, X is the time series, For a coarse-grained sequence, for a time series X with a length of N, the expression for the coarse-grained processing is:

5. The rolling bearing fault diagnosis method based on composite multi-scale attention entropy and optimized SVM according to claim 1 is characterized in that: The collaborative swarm optimization algorithm in S4 optimizes the hyperparameters in the support vector machine algorithm. The specific steps include: initializing the particle swarm position and velocity, setting the individual and global optimal positions and fitness values; for each particle in the swarm, classifying the training set through the support vector machine algorithm, and using the accuracy of the training set as the fitness function; updating the particle position and velocity through the velocity update equation and the dynamic attraction equation; and iterating and updating according to the preset number of iterations and stopping conditions until the stopping conditions are met to obtain the optimal hyperparameters.

6. The rolling bearing fault diagnosis method based on composite multi-scale attention entropy and optimized SVM according to claim 5 is characterized in that: The initialization expression is: T=rand(N,Dim)*(UB-LB)+LB, Where T is the particle population, rand(N,Dim) is an N×Dim random matrix, N is the number of particles, Dim is the dimension of the parameter to be optimized, and UB and LB are the upper and lower bounds respectively.

7. The rolling bearing fault diagnosis method based on composite multi-scale attention entropy and optimized SVM according to claim 1 is characterized in that: The dynamic attraction equation adaptively guides particles according to the local and global attraction of the position, including inertia weight, personal optimal coefficient, global optimal coefficient, dynamic attraction coefficient, adaptive neighborhood interaction coefficient and diversity maintenance coefficient.

8. The rolling bearing fault diagnosis method based on composite multi-scale attention entropy and optimized SVM according to claim 7 is characterized in that: The inertia weight IWV is obtained according to the fitness, which makes the group concentrate the search space. The expression is: IWV=w(t)*v(i,j), Among them, v(i,j) is the current velocity of the particle, w(t) is the inertia weight, and the specific expression is: w(t+1)=w(t)*(1-exp(-k*t)), Where k is the decay rate constant and t is the current iteration; The expression of the personal best coefficient PBC is: PBC=r1*(eps*rand(pbest)-T i ), Among them, r1 is a random value, rand(pbest) is a random solution of the current position and velocity candidate solution, and pbest is the particle T i Best historical position The expression of the global optimal coefficient GBC is: GBC=r2*gbes t -T i , Among them, r2 is a random value, gbes t is the current global optimal solution, The expression of the dynamic attraction coefficient DAC is: Among them, r3 is a random value, attract i is the position with the highest local attraction value in the neighborhood of the i-th particle, c1 is the scaling factor, The expression of the adaptive neighborhood interaction coefficient ANIC is: ANIC=r4*rand(bestf)-bestfi i , Among them, r4 is a random value, rand(bestf) is the random fitness value from the current fitness solution, bestfi i is the fitness value of the i-th position and velocity solution The expression of the diversity maintenance coefficient DMC is: Among them, r5 is a random value, diversity i is the position of the diversity in the neighborhood of the i-th particle in the particle swarm, and c2 is the scaling factor.

9. An electronic device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: When the processor executes the program, the method according to any one of claims 1 to 8 is implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method according to any one of claims 1 to 8 is implemented.

Images