A rolling bearing fault diagnosis method

Abstract

The present invention relates to a rolling bearing fault diagnosis method, and belongs to the technical field of mechanical equipment fault diagnosis. First, the vibration signal of the rolling bearing is decomposed into a series of modal components by using variational modal decomposition, and the sample entropy of each modal component is calculated and input as a feature vector. Then, the extreme learning machine weights and thresholds are optimized by using a cuckoo search algorithm, and a CS-ELM model is established. Finally, the sample entropy eigenvalue is input into the model, and the bearing fault types under different working conditions are classified and identified. The present invention can effectively overcome signal modal aliasing and improve the accuracy of fault identification.

Description

Technical Field

[0001] The invention relates to a rolling bearing fault diagnosis method, belonging to the category of mechanical fault diagnosis. Background Art

[0002] Rolling bearings are the most widely used components in heavy rotating machinery. Problems in their operation can cause the entire machinery to shut down or even cause casualties, leading to serious economic losses. Therefore, real-time monitoring and fault diagnosis of the health status of rolling bearings are extremely important.

[0003] The vibration signal of rolling bearing is usually characterized by non-stationary and nonlinear characteristics, which restricts the effective extraction of fault features. Traditional feature extraction methods have modal aliasing, boundary effects, over-decomposition and other problems. Extreme learning machine is a single-layer feedforward neural network. It does not need to repeatedly adjust the hidden layer parameters, but the random selection of the connection weights between the input layer and the hidden layer and the bias value of the hidden layer will affect its stability and classification accuracy. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide a rolling bearing fault diagnosis method to obtain the highest possible fault diagnosis accuracy in a shorter time.

[0005] The technical solution adopted by the present invention is: firstly, the vibration signal of the rolling bearing is decomposed into a series of intrinsic modal components by using VMD, and the sample entropy of each modal component is calculated and input as a feature vector. Then, the extreme learning machine weight and threshold are optimized by using the cuckoo search algorithm, and the CS-ELM model is established. Finally, the sample entropy feature value is input into the model to classify and identify the fault types of the bearing under different working conditions.

[0006] The details are as follows:

[0007] A rolling bearing fault diagnosis method, the specific steps are:

[0008] 1) Collect vibration signals of normal bearings, inner ring fault bearings, outer ring fault bearings, and rolling element fault bearings under a certain load state;

[0009] 2) Decompose the bearing fault vibration signal f(t) under different working conditions into k eigenmodes u k , while satisfying the requirement that the sum of the estimated bandwidth of each mode is minimum and the sum of each modal function is equal to the input bearing fault vibration signal f(t);

[0010] 3) For the k eigenmodes of the bearing vibration signal obtained in step 2), the sample entropy of each eigenmode is calculated, and a high-dimensional feature vector is constructed, and the vector is used as the fault feature vector and input into the model;

[0011] 4) Optimizing ELM through cuckoo search algorithm to determine the optimal connection weight and threshold, and obtain the CS-ELM model;

[0012] 5) The feature vector is divided into a training set and a test set and input into the CS-ELM classifier for classification training. The fault type and working status of the bearing are obtained through the output results of the classifier to realize the fault diagnosis of the rolling bearing.

[0013] Specifically, the step 2) uses a variational mode decomposition method to decompose the vibration signal data of the bearing in four states, and specifically includes the following steps:

[0014] Assume that the bearing fault vibration signal f(t) is decomposed into k eigenmodes u k , and then find the optimal solution from the variational model to determine the center frequency and bandwidth of each eigenmode, and satisfy the minimum sum of the estimated bandwidth of each mode and the sum of all modes equal to f(t). Therefore, the mathematical expression for constructing the constrained variational model is as follows:

[0015]

[0016] Where δ(t) is the impulse function, t is time, ω k represents the center frequency of the kth modal component; {u k} represents the decomposed intrinsic mode component; {ω k} represents the center frequency corresponding to the eigenmode component; Indicates that the IMF component u is obtained by Hilbert transform k (t) One-sided spectrum.

[0017] In order to obtain the optimal solution to the above constrained variational problem, the Granger multiplication operator λ and the quadratic penalty factor α are introduced to obtain the augmented function, that is,

[0018]

[0019] It means that the single-sided spectrum of the intrinsic mode component uk(t) is obtained by Hilbert transform;

[0020] Iterate the update by alternating directions using the multiplication operator The optimal solution of the variational problem is sought to decompose the bearing fault vibration signal f(t) into k eigenmodes.

[0021] Specifically, the step 3) of calculating the sample entropy of the complexity characteristics of each modal component specifically includes the following steps: the time series {x(n)}=x(1), x(2), x(3)...x(N) is represented as an m-dimensional vector X m (1), X m (2), Xm (3)…X m (N-m+1), solve for vector X m (i) With X m (j) The maximum difference between the corresponding elements d[X m (i),X m (j)]:

[0022] d m (X(i),X(j))=max 0≤k≤m-1 |x(i+k)-x(j+k)|

[0023] d[X m (i), X m The number of (j)] < r is recorded as B i , r is the similarity tolerance valve, and its ratio to the total number of vectors Nm is recorded as B m (r), the dimension increases to m+1, and we get B m+1 (r), so the sample entropy is defined as:

[0024]

[0025] Specifically, the step 4) sets specific parameters of the cuckoo search algorithm: setting the bird's nest size N=30 and the maximum number of iterations M=200.

[0026] The beneficial effects of the present invention are:

[0027] (1) VMD can adaptively decompose fault signals of different states and effectively express fault information.

[0028] (2) By calculating the sample entropy value to obtain the feature vector, the fault signal is quantitatively described, which is conducive to fault identification.

[0029] (3) The CS algorithm optimizes the ELM parameters and improves the accuracy of fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 This is a schematic diagram of the diagnostic model of the present invention;

[0031] Figure 2 Optimize the ELM flow chart of the cuckoo search algorithm of the present invention;

[0032] Figure 3 It is the vibration signal diagram of different faults of the present invention;

[0033] Figure 4 The present invention is a variational modal decomposition result diagram of a rolling bearing fault signal;

[0034] Figure 5Comparison chart of ELM prediction and actual classification of the present invention;

[0035] Figure 6 Comparison chart of CS-ELM prediction and actual classification of the present invention. DETAILED DESCRIPTION

[0036] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments.

[0037] Example 1: Figure 1 As shown in the figure, a rolling bearing fault diagnosis method is first used to decompose the vibration signal of the rolling bearing into a series of modal components using VMD, and the sample entropy of each modal component is calculated and input as a feature vector. Then, the cuckoo search algorithm is used to optimize the extreme learning machine weights and thresholds, and the CS-ELM model is established. Finally, the sample entropy feature value is input into the model to classify and identify the fault types of the bearing under different working conditions.

[0038] The specific steps are:

[0039] 1) Collect vibration signals of normal bearings, inner ring fault bearings, outer ring fault bearings, and rolling element fault bearings under a certain load state;

[0040] 2) Decompose the bearing fault vibration signal f(t) under different working conditions into k eigenmodes u k , while satisfying the requirement that the sum of the estimated bandwidth of each mode is minimum and the sum of each modal function is equal to the input bearing fault vibration signal f(t);

[0041] 3) For the k eigenmodes of the bearing vibration signal obtained in step 2), the sample entropy of each eigenmode is calculated, and a high-dimensional feature vector is constructed, and the vector is used as the fault feature vector and input into the model;

[0042] 4) Optimizing ELM through cuckoo search algorithm to determine the optimal connection weight and threshold, and obtain the CS-ELM model;

[0043] 5) The feature vector is divided into a training set and a test set and input into the CS-ELM classifier for classification training. The fault type and working status of the bearing are obtained through the output results of the classifier to realize the fault diagnosis of the rolling bearing.

[0044] Furthermore, the step 2) uses a variational mode decomposition method to decompose the vibration signal data of the bearing in four states, specifically comprising the following steps:

[0045] Decompose the rolling bearing fault vibration signal f(t) into k eigenmodes u k, and then find the optimal solution from the variational model to determine the center frequency and bandwidth of each eigenmode, and satisfy the minimum sum of the estimated bandwidth of each mode and the sum of all modes equal to f(t). Therefore, the mathematical expression for constructing the constrained variational model is as follows:

[0046]

[0047] In the formula, δ(t) is the impulse function, t is time, ω k represents the center frequency of the kth modal component; {u k} represents the decomposed intrinsic mode component; {ω k} represents the center frequency corresponding to the eigenmode component; Indicates that the intrinsic mode component u is obtained by Hilbert transform k (t) One-sided spectrum.

[0048] To find the optimal solution to the constrained variational problem, it is necessary to introduce the Lagrange multiplication operator λ and the quadratic penalty factor α to transform the above constrained problem into an unconstrained problem. The function expression is:

[0049]

[0050] In order to find the optimal solution to the variational problem, it is necessary to use the multiplication operator alternating direction algorithm to iteratively update The "saddle point" of the augmented Lagrangian function. The corresponding variable update expression in the process of finding the "saddle point" is as follows:

[0051]

[0052]

[0053]

[0054] In the formula is the corresponding time domain function f(t),λ n+1 The Fourier isometric transform of . Represents the center of gravity of the power spectrum of the current eigenmode component.

[0055] Furthermore, the step 3) of calculating the sample entropy of each modal component specifically includes:

[0056] The step 3) of calculating the sample entropy of the complexity characteristics of each modal component specifically includes the following steps: expressing the time series {x(n)}=x(1), x(2), x(3)...x(N) as an m-dimensional vector X m (1), X m (2), X m (3)…X m(N-m+1), solve for vector X m (i) With X m (j) The maximum difference between the corresponding elements d[X m (i),X m (j)] absolute value: d m (X(i),X(j))=max 0≤k≤m-1 |x(i+k)-x(j+k)|

[0057] X m (i) With X m The number of (j) whose distance is less than or equal to r is recorded as B i , calculation formula r is the similarity tolerance threshold.

[0058] When the dimension increases to m+1, we get

[0059] The sample entropy is defined as:

[0060] When N is a finite value, the sample entropy estimate is:

[0061] Furthermore, the step 4) optimizes the connection weights and thresholds of the (CS) extreme learning machine by using a cuckoo search algorithm.

[0062] Specifically, there are N populations, namely bird nests, in the CS algorithm. Each bird nest position represents a rational number. The rounded bird nest position is used as the number of hidden layer nodes M of the ELM. The bird nest scale N=30 and the maximum number of iterations M=200 are set. Input sample data to generate a set of input weight matrix w, hidden layer threshold b and input weight matrix β of the ELM network, and calculate the sample data prediction error as the fitness value of the current bird nest position. At the end of each iterative search, the optimal bird nest position w, b, β is stored. When the iteration termination condition is met, the fitness value corresponding to the bird nest is compared, and the optimal bird nest position and its corresponding w, b, β are output.

[0063] Embodiment 2: In this embodiment, the method shown in Embodiment 1 is used to diagnose bearing faults. The specific implementation steps are as follows:

[0064] The acceleration sensor is now installed on the driving end of the rolling bearing. The bearing damage diameter is 0.1778mm, the speed is set to 1797r / min, the sampling frequency is 12kHz, and the different vibration signals of the rolling bearing under inner ring fault, outer ring fault, rolling element fault and normal state are measured respectively. Figure 3 The waveform shown.

[0065] Take 200 sets of data for each bearing state signal, and randomly select 170 sets of data from each state sample data as training samples, and use the remaining sets of data of bearing vibration signals in the four states of normal, inner ring fault, outer ring fault, and rolling element fault as test samples.

[0066] The number of decompositions is determined based on the principle of avoiding modal aliasing and retaining all characteristic information of the signal. The group signal is decomposed to obtain the modal components of each bearing state. The number of decompositions for the rolling element fault signal is 4, and the decomposition results are as follows: Figure 4 shown.

[0067] The sample entropy values ​​of the decomposed modal components are calculated, and some eigenvectors are shown in Table 1.

[0068] Table 1 Sample entropy feature value extraction

[0069]

[0070] Set the bird's nest scale N = 30, the maximum number of iterations M = 200, and use the CS algorithm to optimize the weights and thresholds of the ELM network to form the best parameter combination and form a CS-ELM model for fault identification. The specific steps are as follows: Figure 2 shown.

[0071] In order to prove the superiority of the CS-ELM model, a comparative experiment is set up to identify the fault type when the training data and test data of the two models are the same. The experimental results confirm that the CS-ELM diagnostic model has higher accuracy. Figure 5 and Figure 6 The following are the classification results of the prediction test set of the two fault diagnosis models. Table 2 more intuitively describes the diagnostic accuracy of the two fault diagnosis models:

[0072] Table 2 Algorithm classification comparison

[0073]

[0074] Figure 5 , Figure 6 As can be seen from Table 2, when the CS-ELM model is used for fault diagnosis, its diagnostic accuracy is higher and it can better identify and classify various types of bearing faults.

[0075] The specific implementation modes of the present invention are described in detail above in conjunction with the accompanying drawings, but the present invention is not limited to the above implementation modes, and various changes can be made within the knowledge scope of ordinary technicians in this field without departing from the purpose of the present invention.

Claims

1. A rolling bearing fault diagnosis method, characterized in that : The specific steps are: 1) Collect vibration signals of normal rolling bearings, inner ring fault bearings, outer ring fault bearings, and rolling element fault bearings under a certain load state; 2) Decompose the bearing fault vibration signal f(t) under different working conditions into k eigenmodes u k , while satisfying the requirement that the sum of the estimated bandwidth of each mode is minimum and the sum of each modal function is equal to the input bearing fault vibration signal f(t); 3) for the k eigenmodes of the bearing fault vibration signal f(t) obtained in step 2), calculate the sample entropy of each eigenmode, and construct a high-dimensional feature vector, which is input into the model as a fault feature vector; 4) Optimizing ELM through cuckoo search algorithm to determine the optimal connection weight and threshold, and obtain the CS-ELM model; 5) Divide the feature vector into a training set and a test set and input them into the CS-ELM classifier for classification training. The fault type and working status of the bearing are obtained through the output of the classifier to realize the fault diagnosis of the rolling bearing; The step 2) uses a variational mode decomposition method to decompose the vibration signal data of the bearing under four states, which specifically includes the following steps: assuming that the bearing fault vibration signal f(t) is decomposed into k eigenmodes u k , and then find the optimal solution from the variational model to determine the center frequency and bandwidth of each eigenmode, and satisfy the minimum sum of the estimated bandwidth of each mode and the sum of all modes equal to f(t). Therefore, the mathematical expression for constructing the constrained variational model is as follows: Where δ(t) is the impulse function, t is time, and w k represents the center frequency of the kth modal component; In order to obtain the optimal solution of the above constrained variational problem, the Granger multiplication operator λ and the quadratic penalty factor α are introduced to obtain the augmented function, that is, Indicates that the intrinsic mode component u is obtained by Hilbert transform k (t) single-sided spectrum; Iterative update using alternating directions of multiplication operators Seek the optimal solution of the variational problem, so as to decompose the bearing fault vibration signal f(t) into k eigenmodes; Step 3) Calculate the sample entropy of the complexity characteristics of each modal component. Specifically, the steps include: Represent the time series {x(n)}=x(1), x(2), x(3)...x(N) as an m-dimensional vector X m (1), X m (2), X m (3)...X m (N-m+1), solve for vector X m (i) With X m (j) The maximum difference between the corresponding elements d[X m (i), X m (j)]: d m (X(i),X(j))=max 0≤k≤m-1 |x(i+k)-x(j+k)| d[X m (i), X m The number of (j)] < r is recorded as B i , r is the similarity tolerance valve, and its ratio to the total number of vectors Nm is recorded as B m (r), the dimension increases to m+1, and we get B m+1 (r), so the sample entropy is defined as: Step 4) Specific parameter settings of the cuckoo search algorithm: set the bird's nest size N = 30, and the maximum number of iterations M = 200; Specifically, there are N populations, namely bird nests, in the CS algorithm, and each bird nest position represents a rational number, and the rounded bird nest position is used as the number of hidden layer nodes M of the ELM; the bird nest scale N = 30, and the maximum number of iterations M = 200; Input sample data, generate a set of ELM network input weight matrix w, hidden layer threshold b and input weight matrix β, and calculate the sample data prediction error as the fitness value of the current nest position; At the end of each iterative search, the optimal nest position w, b, β is stored; when the iteration termination condition is met, the fitness values ​​corresponding to the nests are compared, and the optimal nest position and its corresponding w, b, β are output; The steps for bearing fault diagnosis are as follows: The acceleration sensor is now installed on the driving end of the rolling bearing; the bearing damage diameter is 0.1778mm, the speed is set to 1797r / min, the sampling frequency is 12kHz, and the different vibration signals of the rolling bearing under inner ring fault, outer ring fault, rolling element fault and normal state are measured respectively; 200 groups of data are taken for each bearing state signal, 170 groups of data are randomly selected from each state sample data, that is, as training samples, and the remaining groups of data of the bearing vibration signals in the four states of normal, inner ring fault, outer ring fault and rolling element fault are used as test samples; the number of decompositions is determined according to the principle of avoiding modal aliasing and retaining all characteristic information of the signal, and the group signal is decomposed to obtain the modal components of each bearing state; the number of decompositions for the rolling element fault signal is 4; the sample entropy value of the decomposed modal components is calculated, and some eigenvectors are as follows: Normal: IMF1: 0.270606, 0.275523, IMF2: 0.538289, 0.547741, IMF3: 0.266344, 0.261999, IMF4: 0.756758, 0.741383; Inner circle: IMF1: 0.583038, 0.592259, IMF2: 0.507441, 0.510691, IMF3: 0.245161, 0.239027, IMF4: 0.27667, 0.287463; Rolling element: IMF1: 0.478097, 0.490156, IMF2: 0.502078, 0.512955, IMF3: 0.259033, 0.240769, IMF4: 0.162716, 0.177347; Outer ring: IMF1: 0.427306, 0.398368, IMF2: 0.609396, 0.595303, IMF3: 0.267466, 0.228638, IMF4: 0.155322, 0.157244; The bird's nest scale N is set to 30, the maximum number of iterations M is set to 200, and the weights and thresholds of the ELM network are optimized by the CS algorithm to form the best parameter combination and the CS-ELM model for fault identification. In order to prove the superiority of the CS-ELM model, a comparative experiment is set to identify the fault type when the training data and test data of the two models are the same. The experimental results confirm that the CS-ELM diagnosis model has higher accuracy. The diagnostic accuracy of the two fault diagnosis models is: ELM model: training sample accuracy: 91.61%, test sample accuracy: 92.5%; CS-ELM model: training sample accuracy: 99.26%, test sample accuracy: 99.15%; When the CS-ELM model is used for fault diagnosis, the diagnostic accuracy is high and various bearing faults can be better identified and classified.

Images