Combined rapid bearing state fault diagnosis method

Abstract

The invention discloses a combined rapid bearing state fault diagnosis method, which comprises the following steps of: (1) acquiring a plurality of bearing fault data sets from different sources, preprocessing the bearing fault data sets, and dividing the bearing fault data sets into a training set and a test set; (2) decomposing the bearing vibration signal by using fast ensemble empirical mode decomposition FEEMD, and extracting a fault feature vector; (3) optimizing a kernel function parameter and a penalty coefficient of the LIBSVM model by using an IEGWO algorithm, constructing a bearing fault diagnosis model, and training the bearing fault diagnosis model; (4) processing the test data set by using the optimized LIBSVM model in combination with FEEMD to obtain a fault diagnosis result of the experiment; according to the invention, the operation stability and safety of the equipment can be improved.

Description

Technical Field

[0001] The invention relates to the technical field of bearing state monitoring, and in particular to a combined rapid bearing state fault diagnosis method. Background Art

[0002] With the advancement of industrialization, rotating machinery is increasingly used in various fields. The working conditions of these mechanical equipment are becoming increasingly harsh, which inevitably leads to various faults during operation. As a key component of rotating machinery, the development of fault diagnosis technology for bearings is particularly important. At present, fault diagnosis methods based on the collection of bearing vibration signals have been widely used, but traditional fault feature extraction methods, such as wavelet transform, Hilbert-Huang transform and set empirical mode decomposition, have problems such as large amount of calculation, poor noise resistance and slow operation speed, which makes it difficult to meet the needs of fast and accurate diagnosis in practical applications. At the same time, traditional fault diagnosis methods are also easily affected by personal experience and proficiency, which is not conducive to fast and accurate diagnosis. Summary of the invention

[0003] Purpose of the invention: The purpose of the present invention is to provide a joint rapid bearing state fault diagnosis method, which combines fast ensemble empirical mode decomposition (FEEMD) and LIBSVM bearing fault diagnosis optimized by improved enhancement grey wolf optimizer (IEGWO) to improve the speed and accuracy of bearing state fault diagnosis and achieve effective diagnosis of different bearing fault states.

[0004] Technical solution: A combined rapid bearing status fault diagnosis method according to the present invention comprises the following steps: (1) Obtain bearing fault data sets from multiple different sources, preprocess them, and divide them into training sets and test sets; (2) Use fast ensemble empirical mode decomposition (FEEMD) to decompose the bearing vibration signal and extract the fault feature vector; (3) The kernel function parameters and penalty coefficients of the LIBSVM model were optimized using the IEGWO algorithm, and the bearing fault diagnosis model was constructed and trained; (4) The optimized LIBSVM model is combined with FEEMD to process the test data set to obtain the experimental fault diagnosis results.

[0005] Furthermore, step (1) is specifically as follows: downsampling is performed using a linear interpolation method, and the sampling is uniformly downsampled to 40960 Hz; label processing is performed on the data set, and the normal is marked as 1, the inner race fault is marked as 2, the rolling element fault is marked as 3, and the outer race fault is marked as 4.

[0006] Furthermore, step (2) is as follows: the number of decomposition layers of FEEMD is set to seven; FEEMD is decomposed on the training set samples to obtain the corresponding IMF components and calculate their square averages to construct feature vectors.

[0007] Furthermore, step (3) includes the following steps: (31) Initialize the IEGWO algorithm and LIBSVM parameters: including the number of wolf packs, the maximum number of iterations, and the range of LIBSVM parameters; (32) Based on the preset parameters, the LIBSVM model starts the learning process of the training data; (33) The LIBSVM model was validated using the dataset and the classification accuracy of the obtained model was evaluated as an indicator of the adaptation of the gray wolf population to the environment; (34) Based on the fitness score of each gray wolf, a hierarchical structure is implemented within the group to determine the levels of α, β, δ and ω wolves; (35) Enter the gray wolf position update process; (36) Check whether the current iteration round count has reached the preset upper limit. If so, the optimization process is terminated and the position of the leader wolf in the group, i.e., wolf α, is published as the optimal solution. If the upper limit has not been reached, return to step (34) and continue the iterative optimization process.

[0008] Furthermore, the ranges of the LIBSVM parameters in step (31) include: the range of the kernel function parameter g is [1e-5, 1000], and the range of the penalty parameter c is [1e-5, 1000].

[0009] Furthermore, step (32) is specifically as follows: according to the relevant parameters set in step (31), the gray wolf population is initialized using the Tent chaotic map, and the LIBSVM model begins to be trained based on the initial parameters generated by the Tent chaotic map.

[0010] Furthermore, step (35) is as follows: in the process of updating the gray wolf's position, a random selection control strategy is adopted to randomly select another individual's information to guide the search of candidate individuals; at the same time, an adaptive control convergence factor strategy is adopted to coordinate the global and local development capabilities of the IEGWO algorithm, and finally obtain a better position update.

[0011] Furthermore, the random selection control strategy is as follows: ; Among them, X' is an individual randomly selected from the total population, r3 and r4 are uniform random numbers in [0,1], b1∈(0,1) and b2∈(0,1) are constant coefficients used to adjust the global exploration and development capabilities of the equation.

[0012] Furthermore, the strategy of adaptive control convergence factor is as follows: nonlinear control parameters are used: ; Among them, α ini and α Fin are the initial and final values ​​of the control parameters respectively; t is the current number of iterations; T max is the maximum number of iterations.

[0013] An electronic device described in the present invention comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the computer program is loaded into the processor, any one of the combined rapid bearing status fault diagnosis methods is implemented.

[0014] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages: a LIBSVM bearing fault diagnosis method optimized by combining fast ensemble empirical mode decomposition (FEEMD) and grey wolf algorithm (IEGWO). First, the bearing vibration signal is decomposed by FEEMD to extract the fault feature vector; then, the kernel function parameter g and penalty coefficient c of the LIBSVM model are optimized by the IEGWO algorithm to construct a bearing fault diagnosis model. The best classification effect is obtained by optimizing the stop condition, setting the threshold, improving the classification algorithm and other measures; the proposed method provides an effective technical means for bearing condition monitoring, which helps to improve the operating stability and safety of the equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is the overall flow chart of the present invention; Figure 2 It is a model flow chart of the present invention. DETAILED DESCRIPTION

[0016] The technical solution of the present invention is further described below in conjunction with the accompanying drawings.

[0017] like Figure 1 As shown, an embodiment of the present invention provides a combined rapid bearing status fault diagnosis method, comprising the following steps: (1) Obtain bearing fault data sets from multiple different sources and preprocess them to divide them into training sets and test sets; the data sets include Case Western Reserve University data sets and Jiangnan University data sets; specifically: use linear interpolation to downsample, and uniformly downsample to 40960 Hz; label the data sets, label normal as 1, inner race fault as 2, rolling element fault as 3, and outer race fault as 4. After collecting bearing vibration signals in four states from the data sets, randomly mix them according to the states, and then randomly collect 30 groups of sample data for each state, a total of 120 groups of sample data for the four states, each group of sample data points is 2048, which is equivalent to a sampling point of 0.05s for each group of data. The 120 groups of samples after label processing are randomly divided into 7:3 ratios, including 84 groups of training set samples and 36 groups of test set samples.

[0018] (2) Use fast ensemble empirical mode decomposition (FEEMD) to decompose the bearing vibration signal and extract the fault feature vector; specifically: One feature of FEEMD is that it can set the number of IMFs required for modal decomposition. When FEEMD performs modal decomposition, it is necessary to set an appropriate number of IMFs to be extracted. This is because if the number of decomposition layers is too small, the inherent trend of the original signal cannot be reflected. On the contrary, if the number of decomposition layers is too large, it will lead to over-decomposition, resulting in additional errors, and ultimately the inherent characteristics of the signal cannot be reflected. In order to obtain the optimal number of FEEMD modal decomposition layers, different numbers of decomposition layers are set for prediction, and the prediction results are shown in Table 1.

[0019] Table 1 Comparison of different decomposition levels of FEEMD

[0020] It can be seen from Table 1 that although the accuracy fluctuates when the modal decomposition layer number is 9, in general, when the modal decomposition layer number is less than 7, the accuracy gradually decreases, and when the modal decomposition layer number is greater than 7, the accuracy also gradually decreases, so the optimal choice of FEEMD modal decomposition layer number is 7. FEEMD is performed on the training set samples to obtain the corresponding IMF components and calculate their square averages to construct the eigenvectors.

[0021] (3) The kernel function parameters and penalty coefficients of the LIBSVM model are optimized using the IEGWO algorithm, and the bearing fault diagnosis model is constructed and trained; Figure 2 As shown, the following steps are included: (31) Initialize the IEGWO algorithm and LIBSVM parameters: including the number of wolf packs, the maximum number of iterations, and the range of LIBSVM parameters; Step (31) The range of LIBSVM parameters includes: the range of kernel function parameter g is [1e-5, 1000], and the range of penalty parameter c is [1e-5, 1000].

[0022] (32) Based on the preset parameters, the LIBSVM model starts the learning process of the training data; specifically, according to the relevant parameters set in step (31), the gray wolf population is initialized using the Tent chaotic map, and based on the initial parameters generated by the Tent chaotic map, the LIBSVM model begins training.

[0023] Among them, the formula of Tent chaos mapping is as follows: ; The specific steps of Tent chaotic mapping to generate sequence values ​​are as follows: (a) Randomly generate an initial value x0 (between [0, 1] and avoid x0 falling within (0.2, 0.4, 0.6, 0.8)), and record it in the marking group, that is, y (1) = x0, i = 1, j = 1; (b) Generate the x sequence according to the formula of Tent chaotic mapping, i=i+1; (c) If the maximum number of iterations is reached, go to step (d), otherwise x i ={0, 0.25, 0.5, 0.75} or x i =x i-k , k={0, 1, 2, 3, 4}, then change the initial value of iteration according to the formula x(i)=y(j+1)=y(j)+c, c is a random number, j=j+1; otherwise return to (b); (d) The operation stops and the x sequence is retained.

[0024] (33) The LIBSVM model was validated using the data set, and the classification accuracy of the obtained model was evaluated, which was used as an indicator to measure the adaptation of the gray wolf population to the environment: first, the model was trained using the training set, and the accuracy of the model on the training set was obtained. Then, the trained model was applied to the test set, and the accuracy of the model on the test set was obtained. The combination of the two can be used to obtain the cross-validation accuracy as the fitness of the gray wolf; the LIBSVM model formula is as follows: ; In the formula, n is the number of training samples, c is the penalty coefficient, is the relaxation coefficient. When the sample set data is not completely linearly separable, the introduction of penalty coefficient and relaxation coefficient can reduce the impact of abnormal data points on the model. represents the classification hyperplane, is the i-th category flag.

[0025] The kernel function formula is as follows: ; Among them, x and is the data point in the original feature space, represents the square of the Euclidean distance between two points, is the kernel parameter.

[0026] (34) Based on the fitness score of each gray wolf, a hierarchical structure is implemented within the group to determine the levels of α, β, δ and ω wolves; The mathematical model of the gray wolf tracking the location of prey is described as follows: , , Represent the positions of α, β, and δ wolves respectively: ; The mathematical model of the length and direction of the progress of the ω wolf in the wolf pack towards the α, β, and δ wolves is as follows: ; (35) Enter the gray wolf position update process; the details are as follows: in the process of updating the gray wolf position, a random selection control strategy is adopted to randomly select another individual's information to guide the search of candidate individuals; at the same time, the strategy of adaptive control convergence factor is adopted to coordinate the global and local development capabilities of the IEGWO algorithm, and finally obtain a better position update. Among them, the formula of the random selection control strategy is as follows: ; Among them, X' is an individual randomly selected from the total population, r3 and r4 are uniform random numbers in [0,1], b1∈(0,1) and b2∈(0,1) are constant coefficients used to adjust the global exploration and development capabilities of the equation.

[0027] The strategy of adaptive control convergence factor is as follows: The IEGWO algorithm is mainly composed of two steps: prey positioning and individual movement behavior of gray wolves; the formula is as follows: ; Parameter A plays a very critical role in coordinating the global and local development capabilities of the IEGWO algorithm. When |A|>1, the group will expand the search range to find better candidate solutions, which is the global exploration capability of the IEGWO algorithm; when |A|<1, the group will narrow the search range and conduct a fine search in the local area, which is the local development capability of the IEGWO algorithm. At the same time, during the evolution process, the value of A changes with the change of the control parameter convergence factor a. That is, the global and local development capabilities of the IEGWO algorithm depend on the change of the control parameter a. However, the optimization process of the IEGWO algorithm is very complicated, and the linear change of parameter a cannot reflect the actual optimization search process of the algorithm. It is better to use a nonlinear change strategy than a linear strategy for optimization. Using nonlinear control parameters: ; Among them, α ini and α Fin are the initial and final values ​​of the control parameters respectively; t is the current number of iterations; T max is the maximum number of iterations.

[0028] (36) Check whether the current iteration round count has reached the preset upper limit. If so, the optimization process is terminated and the position of the leader wolf in the group, i.e., wolf α, is published as the optimal solution. If the upper limit has not been reached, return to step (34) and continue the iterative optimization process.

[0029] For the IEGWO-LIBSVM model, the number of wolf packs is set to 20 and the maximum number of iterations is set to 100. At the same time, the optimization range of the SVM parameter penalty factor c is set to [1e-5, 1000] and the optimization range of the kernel function parameter g is set to [1e-5, 1000].

[0030] (4) The optimized LIBSVM model is combined with FEEMD to process the test data set to obtain the experimental fault diagnosis results.

[0031] In order to better demonstrate the advantages of the bearing condition fault diagnosis algorithm based on LIBSVM optimized by FEEMD and IEGWO, 6 groups of comparative experiments were designed again to compare with EMD-SVM, FEEMD-SVM, FEEMD-GWO-SVM, EEMD-DE-SVM, EEMD-PSO-SVM, and VMD-GWO-SVM. The comparison results are shown in Table 2.

[0032] Table 2 Performance comparison of different diagnostic models

[0033] It can be seen that although the EMD-SVM diagnostic model requires the shortest time, its accuracy is the lowest. Even if it is also an unoptimized SVM, the use of FEEMD for modal decomposition can still improve the accuracy by 20%, which fully demonstrates that the FEEMD algorithm can better extract the characteristics of bearing vibration signals. By comparing with optimization algorithms such as DE, PSO, and GWO, it can also be seen that the IEGWO algorithm is effective in optimizing machine learning algorithms such as SVM, and can reduce the dependence on manually set parameters. By comparing LIBSVM and SVM under the same conditions, it can be seen that as an integrated and optimized support vector machine algorithm, it can further improve the accuracy while shortening the recognition time. Therefore, it can be concluded that compared with the other 6 groups of diagnostic models, the diagnostic model proposed in the present invention is superior in accuracy and time required, so it can be seen that the FEEMD-IEGWO-LIBSVM algorithm has the best classification effect and has certain advantages in the field of bearing state fault diagnosis.

[0034] An embodiment of the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the computer program is loaded into the processor, any one of the combined rapid bearing status fault diagnosis methods is implemented.

Claims

1. A combined rapid bearing status fault diagnosis method, characterized in that: The steps include: (1) Obtain bearing fault data sets from multiple different sources, preprocess them, and divide them into training sets and test sets; (2) Use fast ensemble empirical mode decomposition (FEEMD) to decompose the bearing vibration signal and extract the fault feature vector; (3) The kernel function parameters and penalty coefficients of the LIBSVM model were optimized using the IEGWO algorithm, and the bearing fault diagnosis model was constructed and trained; (4) The optimized LIBSVM model is combined with FEEMD to process the test data set to obtain the experimental fault diagnosis results.

2. A combined rapid bearing status fault diagnosis method according to claim 1, characterized in that: Step (1) is as follows: downsampling is performed using a linear interpolation method, and the sampling is uniformly downsampled to 40960 Hz; label processing is performed on the data set, and normal is marked as 1, inner race fault is marked as 2, rolling element fault is marked as 3, and outer race fault is marked as 4.

3. A combined rapid bearing status fault diagnosis method according to claim 1, characterized in that: Step (2) is as follows: set the number of decomposition layers of FEEMD to seven; perform FEEMD decomposition on the training set samples, obtain the corresponding IMF components and calculate their square averages to construct the feature vector.

4. A combined rapid bearing status fault diagnosis method according to claim 1, characterized in that: Step (3) includes the following steps: (31) Initialize the IEGWO algorithm and LIBSVM parameters: including the number of wolf packs, the maximum number of iterations, and the range of LIBSVM parameters; (32) Based on the preset parameters, the LIBSVM model starts the learning process of the training data; (33) The LIBSVM model was validated using the dataset and the classification accuracy of the obtained model was evaluated as an indicator of the adaptation of the gray wolf population to the environment; (34) Based on the fitness score of each gray wolf, a hierarchical structure is implemented within the group to determine the levels of α, β, δ and ω wolves; (35) Enter the gray wolf position update process; (36) Check whether the current iteration round count has reached the preset upper limit. If so, the optimization process is terminated and the position of the leader wolf in the group, i.e., wolf α, is published as the optimal solution. If the upper limit has not been reached, return to step (34) and continue the iterative optimization process.

5. A combined rapid bearing status fault diagnosis method according to claim 4, characterized in that: Step (31) The range of LIBSVM parameters includes: kernel function parameter g ranges from [1e-5, 1000], penalty parameter c ranges from [1e-5, 1000].

6. A combined rapid bearing status fault diagnosis method according to claim 4, characterized in that: Step (32) is as follows: According to the relevant parameters set in step (31), the gray wolf population is initialized using the Tent chaotic map, and the LIBSVM model begins training based on the initial parameters generated by the Tent chaotic map.

7. A combined rapid bearing status fault diagnosis method according to claim 4, characterized in that: Step (35) is as follows: In the process of updating the gray wolf's position, a random selection control strategy is adopted to randomly select another individual's information to guide the search of candidate individuals; at the same time, an adaptive control convergence factor strategy is adopted to coordinate the global and local development capabilities of the IEGWO algorithm, and finally obtain a better position update.

8. A combined rapid bearing status fault diagnosis method according to claim 7, characterized in that: The random selection control strategy is as follows: ； Among them, X' is an individual randomly selected from the total population, r3 and r4 are uniform random numbers in [0,1], b1∈(0,1) and b2∈(0,1) are constant coefficients used to adjust the global exploration and development capabilities of the equation.

9. A combined rapid bearing status fault diagnosis method according to claim 7, characterized in that: The strategy of adaptive control convergence factor is as follows: nonlinear control parameters are used: ； Among them, α ini and α Fin are the initial and final values ​​of the control parameters respectively; t is the current number of iterations; T max is the maximum number of iterations.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the computer program is loaded into a processor, a combined rapid bearing status fault diagnosis method according to any one of claims 1 to 9 is implemented.

Images