A cross-modal bearing fault diagnosis method based on self-supervised spatial learning

Abstract

This invention discloses a cross-modal bearing fault diagnosis method based on self-supervised spatial learning. The main method involves constructing a unified optimization model for class labels and cross-modal spatial projection directions, thereby enabling self-supervised learning of class labels in unsupervised fault diagnosis and effectively improving the accuracy of fault diagnosis. The specific implementation process is as follows: (1) Constructing a cross-modal self-supervised spatial learning model using cross-modal class label information and related spatial theories; (2) Theoretically deriving the indirect representation of cross-modal class centers, and then obtaining the analytical solution of class labels in the learned cross-modal consistent space; (3) Obtaining cross-modal fault features of fault sample data based on the cross-modal self-supervised spatial learning projection directions, and inputting the low-dimensional unified fault features into a classifier to obtain the final bearing fault diagnosis result. Compared with existing technologies, the fault diagnosis method of this invention is more effective and robust.

Description

Technical Field

[0001] This invention relates to a cross-modal bearing fault diagnosis method based on self-supervised spatial learning, which belongs to the field of pattern recognition and fault diagnosis. Background Technology

[0002] In the field of fault diagnosis, bearings, as key components of mechanical equipment, ensure stable operation of mechanical components by reducing friction and improving rotational smoothness, significantly improving equipment operating efficiency and reducing energy consumption. However, equipment failure can cause huge economic losses or even safety accidents. Therefore, effective fault diagnosis of key components of mechanical equipment, extracting valuable discriminative features from high-dimensional fault data, and improving diagnostic performance are currently key issues. Among the methods for solving such problems, spatial learning is an effective approach. However, unsupervised cross-modal spatial learning lacks class labels for rolling bearing faults, making it difficult to extract richer feature information and effectively improve the class separability of the learning space. To address this, this paper proposes a unified optimization model for class labels and cross-modal spatial projection directions using relevant spatial theory and cross-modal class label functions. This model can solve the problem of lacking class label discriminative information in unsupervised learning. Through theoretical derivation of the optimization model, an analytical solution for class labels in the cross-modal self-supervised spatial learning model is obtained, realizing self-supervised learning of class labels in unsupervised situations. Furthermore, this invention effectively solves the problem of difficulty in fault separation caused by the lack of adaptability between cross-modal fusion and downstream task clustering. Summary of the Invention

[0003] To effectively utilize the rich discriminative information of self-supervised pseudo-labels, this invention constructs a cross-modal self-supervised spatial learning model based on spatial learning theory, and theoretically derives the analytical solution of this model, thereby solving the problem of lacking class label discriminative information in unsupervised learning. The specific implementation steps of this invention are as follows:

[0004] 1. By collecting operational data on bearing failures in mechanical equipment through sensors, relevant statistical features of multiple bearing failure data are extracted from the time domain, frequency domain, and time-frequency domain perspectives to construct a cross-modal failure data sample set. Where d i The feature dimension of the i-th dataset is represented by N, the number of samples is represented by m, and the number of modes of the fault data is represented by m. The fault data is divided into training sample sets trainX1 and trainX2 and test sample sets testX1 and testX2 according to the proportion.

[0005] 2. Construct a cross-modal self-supervised spatial learning model to obtain a low-dimensional representation of high-dimensional fault features.

[0006] The specific steps for constructing a cross-modal self-supervised spatial learning model are as follows:

[0007] (2a) Construction of the class label optimization function:

[0008] Assuming each modality has K class centers, after each modality is mapped... The class center is In the shared subspace, the class center learning function for the m-th modality is expressed as:

[0009]

[0010] in, This represents the number of samples belonging to class v in modality m. And further define... For class label functions:

[0011]

[0012] Here, the class label is represented as F = [f1, f2, ..., f n ]∈R (K×N) ,f n =[f n1 ,f n2 ,…,f nK ]∈R (K×N) , n=1,2,…,N. The meaning of m = 1, 2 is that f = 1, 2 if and only if f nv When = 1, It belongs to class v, otherwise f nv =0.

[0013] (2b) Cross-modal neighborhood consistent class center objective function:

[0014] Cross-modal data is typically high-dimensional data containing noise and redundant information. Neighborhood relationships based on high-dimensional data often deviate from inherent neighborhood relationships, resulting in weak class separability when fusing related features. To address this problem, this invention projects high-dimensional data into a related fusion subspace to obtain inter-modal correlation features, where the correlation features between different modalities have the same dimension. Cross-modal class centers are constructed in the fused subspace, and an indirect representation of these centers is theoretically derived.

[0015]

[0016] Where F represents the class label matrix, c v This represents the cross-modal class center, which can be viewed as a set of discrete points within the class.

[0017] The cross-modal neighborhood consensus class center objective function can effectively integrate the discriminative information between modalities, improve intermodal complementarity, and weaken the impact of label distortion. The cross-modal neighborhood consensus class center objective function is defined as follows:

[0018]

[0019] in, S represents the characteristic matrix.

[0020] (2c) Construction of cross-modal self-supervised spatial learning models:

[0021] Based on the solutions to the aforementioned class label optimization function and cross-modal neighborhood consistent class center objective optimization function, as well as related spatial learning theories, a cross-modal self-supervised spatial learning model is constructed to address the problem of lacking class label discrimination information in unsupervised learning:

[0022]

[0023] in, Cross-modal self-supervised spatial learning models seek shared low-dimensional representations among different modalities, minimize the intra-class sparseness of each modality, effectively enhance the class separability of related features, and maximize the correlation between modalities, thus preserving complementary information between cross-modal data to a greater extent.

[0024] 3. An iterative optimization method is used to solve for the class center matrix C and the projection dimension reduction matrix H of the cross-modal self-supervised spatial learning model. m And the class label matrix F.

[0025] (3a) Fix the class label matrix F and optimize the projection matrix H m And the central matrix C:

[0026] The analytical representation of C is obtained using the Lagrange multiplier method, and the Lagrange multiplier function L(c) of the cross-modal self-supervised spatial learning model is constructed. v ):

[0027]

[0028] Where λ is a Lagrange multiplier, and... Substitute into the above equation and apply c v Find the partial derivative:

[0029]

[0030] make To get zero:

[0031]

[0032] Therefore, c vThe representation of is:

[0033]

[0034] Extending the above equation, we obtain the cross-modal class center matrix C:

[0035]

[0036] When using the Lagrange multiplier method, the solution is the PR corresponding to the first d largest eigenvalues. -1 The generalized eigenvectors. For block matrices, The projection vector can be obtained from S.

[0037] (3b) Fixed projection matrix H m And the center matrix C, optimize the class label matrix F:

[0038] The optimization objective function for solving the cross-modal self-supervised spatial learning model is restated as follows:

[0039]

[0040] When H m and c v When c is fixed, the denominator in the formula is a constant, and only the numerator in the formula is minimized. Since c v It is fixed, F = [f1, f2, ..., f n The optimization problem of f is simplified to the nearest neighbor problem of the class center. n =[f n1 ,f n2 ,…,f nK ]∈R (K×N) The optimal solution for (n = 1, 2, ..., N) is:

[0041]

[0042] 4. Obtain low-dimensional training sample feature sets (trainY1, trainY2) and low-dimensional test sample feature sets (testY1, testY2) of bearing fault data through spatial projection; use a parallel fusion strategy to fuse low-dimensional training sample sets (trainY1, trainY2) or low-dimensional test sample feature sets (testY1, testY2) of different modalities, thereby obtaining fused low-dimensional training sample sets and fused low-dimensional test sample sets; finally, use a classifier for classification to obtain fault diagnosis results.

[0043] The method of the present invention has the following advantages:

[0044] (1) This invention can effectively utilize the rich identification information of self-supervised pseudo-labels and learn the cross-modal self-supervised space based on this, effectively improving the class separation of cross-modal bearing fault features;

[0045] (2) This invention, by leveraging relevant spatial learning theory and cross-modal class label function, further constrains the learning of spatial projection matrix, effectively solving the problem of lack of class label identification information in unsupervised learning, realizing self-supervised learning in an unsupervised environment, and significantly improving the accuracy of fault diagnosis.

[0046] (3) This invention, through theoretical derivation, obtains the analytical solution of the cross-modal self-supervised spatial learning model, effectively utilizing class label information and capturing similar structures between different modalities. By minimizing the intra-class dispersion of each modality, the class separability of related features is effectively enhanced; by maximizing the correlation between modalities, complementary information between cross-modal data is preserved as much as possible. Thus, it is possible to quickly obtain the cross-modal fault test features of fault test samples and achieve more accurate fault type classification. Attached Figure Description

[0047] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0048] Figure 1 This is a flowchart of the present invention, where n is the number of fault categories.

[0049] Figure 2 It is the classification accuracy of a randomized experiment. Detailed Implementation

[0050] The specific implementation steps of this invention are as follows:

[0051] 1. By collecting operational data on bearing failures in mechanical equipment through sensors, relevant statistical features of multiple bearing failure data are extracted from the time domain, frequency domain, and time-frequency domain perspectives to construct a cross-modal failure data sample set. Where d i The feature dimension of the i-th dataset is represented by N, the number of samples is represented by m, and the number of modes of the fault data is represented by m. The fault data is divided into training sample sets trainX1, trainX2 and test sample sets testX1, testX2 according to the proportion.

[0052] 2. Based on the cross-modal self-supervised spatial learning model, an iterative solution method is used to continuously update and optimize the class center matrix C and the projection dimension reduction matrix H. m And the class label matrix F, to obtain the low-dimensional projection direction. By mitigating the incompatibility between dimensionality reduction and clustering of high-dimensional data samples, clustering performance is optimized, enabling class labels to indicate class centers in an unsupervised environment.

[0053] 3. Obtain low-dimensional training sample feature sets and low-dimensional test sample feature sets of bearing fault data through spatial projection; use a parallel fusion strategy to fuse low-dimensional training sample sets or low-dimensional test sample feature sets of different modalities to obtain fused low-dimensional training sample sets and low-dimensional test sample sets; finally, use a classifier to classify and obtain fault diagnosis results.

[0054] The effectiveness of this invention was further verified through the following experiments:

[0055] Two data modes were selected from the Paderborn bearing dataset for experimental verification: motor current signal and vibration signal. In this experiment, the selected bearing fault data consisted of 1000 samples at a sampling frequency of 64 kHz. Both the current and vibration signals were divided into 250 samples with a sampling length of 1024. The experiment selected two sets of man-made damage fault data, one type of accelerated life damage data, and one type of fault-free data. Inner race faults and outer race faults were man-made damage faults, denoted as F1 and F2; mixed inner and outer race faults were accelerated life damage, denoted as F3; and fault-free data was denoted as F4. Figure 2 The classification accuracy of bearing fault diagnosis in each random experiment is visually demonstrated. From Figure 2 As can be seen, the accuracy of the method of this invention increases with the increase of the number of training samples, and the average diagnostic accuracy of the optimal randomized experiment reaches 99.79%, with good stability. Experimental results show that the method disclosed in this invention is an effective bearing fault diagnosis method.

Claims

1. A cross-modal bearing fault diagnosis method based on self-supervised spatial learning, characterized in that... The method includes the following steps: (1) By collecting operating data of bearing failures in mechanical equipment through sensors, relevant statistical features of multiple bearing failure data are extracted from the time domain, frequency domain, and time-frequency domain to construct a cross-modal failure data sample set. Where d i The feature dimension of the i-th dataset is represented by N, the number of samples is represented by m, and the number of modes of the fault data is represented by m. The fault data is divided into training sample sets trainX1, trainX2 and test sample sets testX1, testX2 according to the proportion. (2) Construct a self-supervised spatial learning model for cross-modal fault data to obtain a low-dimensional representation of high-dimensional fault features. The steps are as follows: (2a) Assume that each modality has K class centers, and each modality is mapped after... The class center is definition The class label optimization function is derived theoretically, and the cross-modal neighborhood consistent class center objective optimization function is obtained. (2b) Combining the solution of the class label optimization function and the cross-modal neighborhood consistent class center objective optimization function with relevant spatial learning theories, a cross-modal self-supervised spatial learning model is constructed to address the problem of lack of class label discrimination information in unsupervised learning: Among them, H m Let S be the projection vector we want to find, S represent the feature matrix, F represent the class label matrix, and c represent the projection vector. v Indicates the cross-modal class center, Taking class tag functions as an example, constructing class tag optimization functions in, F represents the number of samples belonging to class v in modality m, where F = [f1, f2, ..., f n ]∈R (K×N) ,f n =[f n1 ,f n2 ,…,f nK ]∈R (K×N) n = 1, 2, ..., N; It means if and only if f nv When = 1, It belongs to class v, otherwise f nv =0; The cross-modal neighborhood consistent class center objective optimization function can effectively integrate the discriminative information between modalities, improve the complementarity between modalities, and weaken the influence of label distortion. The cross-modal neighborhood consistent class center objective optimization function is defined as follows: Cross-modal self-supervised spatial learning models seek shared low-dimensional representations among different modalities, minimize the intra-class sparseness of each modality, effectively enhance the class separability of related features, and maximize the correlation between modalities, thus preserving complementary information between cross-modal data to a greater extent. (3) The class center matrix C and the projection dimension reduction matrix H of the cross-modal self-supervised spatial learning model are solved by iterative optimization. m And the class label matrix F; the steps are as follows: (3a) Fix the class label matrix F and optimize the projection matrix H m And class center matrix C: The analytical representation of C is obtained using the Lagrange multiplier method, and the Lagrange multiplier function L(c) of the cross-modal self-supervised spatial learning model is constructed. v ): Where λ is a Lagrange multiplier, and... Substitute into the above equation and apply c v Find the partial derivative: make To get zero: Therefore, c v The representation of is: Extending the above equation, we obtain the cross-modal class center matrix C: When using the Lagrange multiplier method, the solution is the PR corresponding to the first d largest eigenvalues. -1 The generalized eigenvectors, where For block matrices, The projection vector can be obtained from S. (3b) Fixed projection matrix H m And the center matrix C, optimize the class label matrix F: The optimization objective function for solving the cross-modal self-supervised spatial learning model is restated as follows: When H m and c v When fixed, the denominator in the formula is a constant, and only the numerator in the formula is minimized; since c v It is fixed, F = [f1, f2, ..., f n The optimization problem of f is simplified to the nearest neighbor problem of the class center. n =[f n1 ,f n2 ,…,f nK ]∈R (K×N) The optimal solution for n = 1, 2, ..., N is: By fusing cross-modal information using canonical correlation analysis, the complementarity of fault samples is enhanced, resulting in richer discrimination information. Furthermore, cross-modal class centers are considered to mitigate the impact of class label distortion, ultimately yielding the projection matrix. The class center matrix C and class label matrix F enable self-supervised learning and class center sharing in the unsupervised case. (4) Obtain low-dimensional training sample feature set and low-dimensional test sample feature set of bearing fault data through spatial projection; use parallel fusion strategy to fuse low-dimensional training sample set or low-dimensional test sample feature set of different modalities, and then obtain fused low-dimensional training sample set and fused low-dimensional test sample set; finally, use classifier to classify and obtain fault diagnosis results.

Images