Tensor domain health state assessment method for fusing multi-channel signals

Abstract

The invention provides a tensor domain health state assessment method for fusing multi-channel signals. According to the method, a multi-channel sensitive feature optimization method is provided based on rank mutual information and improved spectral clustering. The multi-channel adjacent matrix constructed by the improved spectral clustering can realize the overall evaluation of the multi-channel characteristics, thereby realizing the clustering of the similar degradation curve characteristics in each channel. Rank mutual information is used for measuring nonlinear correlation between features and time, and the optimal feature is selected from each cluster. In a health state assessment stage, adaptive kernel spectrum clustering is used as basic clustering to realize a preliminary assessment stage of a multi-channel data health state, and then health state information among multiple channels is fused based on a tensor clustering integration method of low-rank tensor approximation to realize final health state assessment of multi-channel data.

Description

Technical Field

[0001] The present invention relates to the field of gear technology assessment, and in particular to a tensor domain health status assessment method for fusing multi-channel signals Background Art

[0002] Rolling bearings play a vital role in mechanical systems. As key components for bearing and transmitting force, they ensure the smooth operation and efficient work of equipment. Whether in industrial manufacturing, transportation or aerospace, the performance of rolling bearings directly affects the overall efficiency and life of mechanical equipment. By regularly monitoring and evaluating the working status of bearings, potential faults can be discovered in time and targeted maintenance strategies can be implemented, thereby effectively reducing the risks caused by faults and reducing equipment downtime. This not only improves the reliability of mechanical systems, but also saves maintenance costs and resources for enterprises, ensuring the safety and smoothness of production operations. Therefore, the health status assessment of rolling bearings is the core link to ensure the efficient operation of mechanical systems, and has important economic and safety significance.

[0003] Most of the existing rolling bearing health status assessment methods are based on single-channel implementation. The patents related to the above methods are: (1) Invention patent CN202010557799 discloses a ball bearing health status online assessment method and system. The invention better reflects the structural information of the signal based on local mean decomposition and construction of a graphical model, thereby realizing the health status assessment of the bearing; (2) Invention patent CN202110192810 discloses a spindle bearing health status assessment and remaining life prediction method and device. The invention combines the optimal generalized high-order moment coefficient and the improved Paris crack growth theory to construct a remaining life prediction model, which can stably predict the remaining life of the bearing; (3) Invention patent CN202210627736 discloses an explosion-proof motor bearing health status assessment method and system. The invention can evaluate the slight changes in the health status of the motor bearing in real time and accurately through real-time data collection, multi-scale sparse measurement fusion indicators and confidence conversion, thereby providing timely early fault warning.

[0004] However, most of the existing bearing health status assessment methods are based on a single channel and do not fully utilize the degradation information between multiple channels. With the popularization of multi-channel / multi-sensor in the era of Industry 4.0, multi-channel signals containing richer equipment status information show greater potential in the accurate assessment of health status. Patents related to the above method include: Invention patent CN202311706693 discloses a bearing health status assessment method, device and related equipment based on digital twins. The invention integrates multi-dimensional physical information data through digital twins, and combines time series prediction with status assessment, thereby improving the accuracy of bearing health status assessment. Although the above patent incorporates multi-channel signals into the health status assessment of rolling bearings, there are still shortcomings. Its method cannot explore the correlation characteristics between multiple channels and integrate multi-channel health status information, making it difficult to achieve accurate assessment of the health status of rolling bearings. Summary of the invention

[0005] The technical problem to be solved by the present invention is to provide a tensor domain health status assessment method for fusing multi-channel signals in view of the problems existing in the background technology.

[0006] The technical solution adopted by the present invention to solve the above technical problems is:

[0007] A tensor domain health status assessment method for fusing multi-channel signals is proposed. The method proposes a multi-channel sensitive feature optimization method based on rank mutual information and improved spectral clustering. The multi-channel adjacency matrix constructed by improved spectral clustering can realize the overall evaluation of multi-channel features, and then realize the clustering of similar degradation curve features in each channel. Rank mutual information is used to measure the nonlinear correlation between features and time, and select the optimal features from each clustering cluster. In the health status assessment stage, adaptive kernel spectral clustering is used as the basic clustering to realize the preliminary evaluation stage of the health status of multi-channel data. Then, the tensor clustering integration method based on low-rank tensor approximation is used to fuse the health status information between multiple channels to realize the final health status assessment of multi-channel data.

[0008] The method includes the following features:

[0009] Step S1, using a sensor to collect multi-channel vibration signal data of a rolling bearing;

[0010] Step S2, extracting the time domain, frequency domain and time-frequency domain degradation features of each channel 41 from the collected data;

[0011] Step S3, based on rank mutual information and improved spectral clustering, multi-channel sensitive feature optimization is realized, similar feature clustering is realized, and feature set redundancy is reduced;

[0012] Step S4, using adaptive kernel spectral clustering to achieve a preliminary assessment of the health status of multiple channels. The assessment of the three stages (initialization stage, calibration stage and test stage) in the adaptive kernel spectral clustering is achieved based on the full life root mean square characteristics passed by each;

[0013] Step S5, inputting the basic status assessment information obtained by adaptive kernel spectrum clustering into tensor clustering integration to realize synchronous assessment of multi-channel health status.

[0014] Furthermore, the features of step S3 specifically include:

[0015] Step S3-1, extracting the full life feature set of each channel;

[0016] Step S3-2, calculate each feature f in each channel feature set m Three indicators: Spearman coefficient ρ m 、f m Corresponding RMI 1 、-f m Corresponding RMI 2 , RMI refers to Rank Mutual Information;

[0017] Step S3-3, trend uniform correction and RMI calculation: When ρ m ≥0, the final required RMI=RMI 1 , the required feature F m =f m When ρ m <0, the final required RMI = RMI 2 , the required feature F m =-f m ;

[0018] Step S3-4, calculate the improved similarity matrix of each channel, and use the variable step size grid search algorithm to improve the scale factor a in the similarity matrix ij and offset b ij Parameter estimates of ;

[0019] Step S3-5, taking the means of the corresponding elements of the improved similarity matrices of multiple channels to construct a multi-channel adjacency matrix;

[0020] Step S3-6, perform feature clustering based on improved spectral clustering, extract the feature with the largest RMI in each cluster of each channel as the sensitive feature, and realize the optimization of the sensitive feature set for multi-channel performance degradation assessment.

[0021] Furthermore, the features of step S5 specifically include:

[0022] Step S5-1, using a basic clustering algorithm to achieve a preliminary assessment of the health status of the rolling bearing;

[0023] Step S5-2, forming a basic cluster Π of the preliminary evaluation results of the health status of multiple channels to construct a connection matrix;

[0024] Step S5-3, constructing each connection matrix into a joint matrix and a coherent link matrix, and concatenating the two into a third-order tensor;

[0025] Step S5-4, processing the third-order tensor based on a low-rank tensor approximation method to obtain a redefined joint matrix;

[0026] Step S5-5, the redefined joint matrix is ​​used to construct the adjacency matrix in spectral clustering, and finally the health status evaluation of multi-channel data is realized. On the one hand, the present invention proposes a method for optimizing multi-channel sensitive features based on rank mutual information and improved spectral clustering, which can simplify the feature set used for health status evaluation and reduce feature set redundancy. On the other hand, a health status evaluation method based on tensor clustering integration is proposed. Through the low-rank tensor approximation method, the correlation characteristics between multiple channels can be mined, and the multi-channel health status information can be integrated to realize the health status evaluation of rolling bearings.

[0027] The beneficial effects of the present invention are:

[0028] The present invention proposes a tensor domain health status assessment method for fusing multi-channel signals, which solves some difficulties and shortcomings in the accurate assessment of the health status of rolling bearings of mechanical equipment under multi-source signal data at the current stage. For example: (1) Most of the existing bearing health status assessment methods are based on single-channel implementation, and do not make full use of the degradation information between multiple channels; (2) The existing feature selection method ignores the detailed change information in the degradation process and is more suitable for remaining service life prediction rather than health status assessment. Therefore, it is particularly important to extract the key feature set suitable for health status assessment. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 It is a flowchart of a tensor domain health status assessment method for fusing multi-channel signals provided in an embodiment of the present invention.

[0030] Figure 2 These are three collected rolling bearing multi-channel vibration signals in an embodiment of the present invention, wherein the left figure represents channel 1: horizontal vibration signal, and the right figure represents channel 2: vertical vibration signal.

[0031] Figure 3 This is the feature clustering result diagram of channel 1 in bearing 1-1 based on rank mutual information and improved spectral clustering method.

[0032] Figure 4 is a graph of RMI values ​​for preferred features.

[0033] Figure 5 This is a preliminary health status assessment diagram based on adaptive kernel spectral clustering according to an embodiment of the present invention, wherein the upper left is bearing 1-1, the upper right is bearing 1-2, the lower left is bearing 1-3, and the lower right is bearing 2-3.

[0034] Figure 6 It is a health status assessment diagram based on tensor clustering integration of an embodiment of the present invention, wherein the upper left is bearing 1-1, the upper right is bearing 1-2, the lower left is bearing 1-3, and the lower right is bearing 2-3.

[0035] Numbers in the figure: DETAILED DESCRIPTION

[0036] In order to enable those skilled in the art to better understand and implement the technical solution of the present invention, the present invention is specifically, clearly and completely described in the following embodiments in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Moreover, based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work, any modifications, equivalent substitutions, improvements, etc., should be included in the protection scope of the present invention.

[0037] Figure 1 It is a flowchart of a tensor domain health status assessment method for fusing multi-channel signals provided in an embodiment of the present invention.

[0038] Step S1, using a sensor to collect multi-channel vibration signal data of a rolling bearing;

[0039] Figure 2 1-1 is a multi-channel vibration signal of three rolling bearings collected in the embodiment of the present invention. Bearing 1-1 is used to select the optimal degradation characteristics. All three bearings are further used to verify the effectiveness of the proposed performance degradation assessment method.

[0040] Step S2, extracting 41-dimensional time domain, frequency domain and time-frequency domain degradation features of each channel from the collected data, including 11 time domain features, 12 frequency domain features, 16 time-frequency domain features and 2 information entropy features;

[0041] Step S3, based on rank mutual information and improved spectral clustering, multi-channel sensitive feature optimization is realized, similar feature clustering is realized, and feature set redundancy is reduced;

[0042] The specific implementation steps of step S3 are as follows:

[0043] Step S3-1, extracting the full life feature set of each channel;

[0044] Step S3-2, calculate each feature f in each channel feature set m Three indicators: Spearman coefficient ρ m 、f m Corresponding RMI 1 、-f m Corresponding RMI 2 , RMI refers to Rank Mutual Information;

[0045] Step S3-3, trend uniform correction and RMI calculation: When ρ m ≥0, the final required RMI=RMI 1 , the required feature F m =f m When ρ m <0, the final required RMI = RMI 2 , the required feature F m =-f m ;

[0046] Step S3-4, calculate the improved similarity matrix of each channel, and use the variable step size grid search algorithm to improve the scale factor a in the similarity matrix ij and offset b ij Parameter estimation; the improved similarity matrix and single-channel adjacency matrix are defined as follows:

[0047]

[0048] where s ij is the similarity matrix, w ij is the adjacency matrix.

[0049] Step S3-5, taking the means of the corresponding elements of the improved similarity matrices of multiple channels to construct a multi-channel adjacency matrix;

[0050] Step S3-6, perform feature clustering based on improved spectral clustering, extract the feature with the largest RMI in each cluster of each channel as the sensitive feature, and realize the optimization of the sensitive feature set for multi-channel performance degradation assessment.

[0051] Figure 3 This is the feature clustering result diagram of channel 1 in bearing 1-1 based on rank mutual information and improved spectral clustering method. It can be seen that the feature curves in the same cluster are similar or even overlapped, which verifies the effectiveness of spectral clustering in feature clustering. The multi-channel feature optimization method based on rank mutual information and improved spectral clustering can not only reduce redundancy, but also effectively avoid the excessive influence of features by a single type of features. In addition, "bad" features may also contain key degenerate information (such as cluster 7), so feature clusters with usable information in multiple channels should be retained.

[0052] Figure 4 is the RMI value of the preferred feature. It can be seen that the RMI value of feature F1 from cluster 8 is much smaller than that of other features and does not contain useful degradation information, so it is excluded. Feature 16 from cluster 7 is retained because it contains stage information. The preferred 8 features will be used in the subsequent multi-channel health status assessment work.

[0053] Step S4, using adaptive kernel spectral clustering to achieve a preliminary assessment of the health status of multiple channels. The assessment of the three stages (initialization stage, calibration stage and test stage) in the adaptive kernel spectral clustering is achieved based on the full life root mean square characteristics passed by each;

[0054] Figure 5 This is a preliminary health status assessment diagram based on adaptive kernel spectral clustering of an embodiment of the present invention. As can be seen from the figure, adaptive kernel spectral clustering has a certain effect on the health status assessment of four different bearings. Among them, bearing 1-1 has basically achieved health status assessment. However, for the other three bearings, the membership of the clusters fluctuates back and forth within a certain period of time. In the performance degradation assessment, the health status of the bearing is generally assessed as 2-4, but the highest values ​​of the clusters of bearings 1-3 and bearing 2-3 have exceeded 10, which is not favorable in the health status assessment, so it is necessary to perform secondary processing of the multi-channel clustering data to achieve accurate health status assessment.

[0055] Step S5, inputting the basic status assessment information obtained by adaptive kernel spectrum clustering into tensor clustering integration to realize synchronous assessment of multi-channel health status.

[0056] The specific implementation steps of step S5 are as follows:

[0057] Step S5-1, using a basic clustering algorithm to achieve a preliminary assessment of the health status of the rolling bearing;

[0058] Step S5-2, forming a basic cluster Π of the preliminary evaluation results of the health status of multiple channels to construct a connection matrix;

[0059] Step S5-3, constructing each connection matrix into a joint matrix and a coherent link matrix, and concatenating the two into a third-order tensor;

[0060] Step S5-4, processing the third-order tensor based on a low-rank tensor approximation method to obtain a redefined joint matrix;

[0061] Step S5-5, using the redefined joint matrix to construct the adjacency matrix in spectral clustering, and finally realizing the evaluation of the health status of multi-channel data.

[0062] Figure 6It is a health status assessment diagram based on tensor clustering integration according to an embodiment of the present invention. Figure 5 and Figure 6 The distribution of the number of clusters changes significantly. In the preliminary health status assessment, the membership of a large number of clusters fluctuates significantly. However, in the final health status assessment, a large number of adjacent time period data are assessed as the same stage, such as the data from 78 to 80 minutes in bearing 1-1 and the data from 45 to 65 minutes in bearing 1-2. It can be seen that the health status assessment method based on tensor clustering integration can extract the public health information hidden in multi-channel data, thereby efficiently realizing health status assessment.

[0063] Although the present invention has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments, or to make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A tensor domain health status assessment method for fusing multi-channel signals, characterized in that: The method includes the following features: Step S1, using a sensor to collect multi-channel vibration signal data of a rolling bearing; Step S2, extracting the time domain, frequency domain and time-frequency domain degradation features of each channel 41 from the collected data; Step S3, based on rank mutual information and improved spectral clustering, multi-channel sensitive feature optimization is realized, similar feature clustering is realized, and feature set redundancy is reduced; Step S4, using adaptive kernel spectral clustering to achieve a preliminary assessment of the health status of multiple channels. The assessment of the three stages (initialization stage, calibration stage and test stage) in the adaptive kernel spectral clustering is achieved based on the full life root mean square characteristics passed by each; Step S5, inputting the basic status assessment information obtained by adaptive kernel spectrum clustering into tensor clustering integration to realize synchronous assessment of multi-channel health status.

2. The tensor domain health status assessment method for fusing multi-channel signals according to claim 1, characterized in that: The features of step S3 specifically include: Step S3-1, extracting the full life feature set of each channel; Step S3-2, calculate each feature f in each channel feature set m Three indicators: Spearman coefficient ρ m 、f m Corresponding RMI1, -f m Corresponding to RMI2, RMI refers to Rank Mutual Information; Step S3-3, trend uniform correction and RMI calculation: When ρ m ≥0, the final required RMI=RMI1, the required feature F m =f m When ρ m <0, the final required RMI = RMI2, the required feature F m =-f m ; Step S3-4, calculate the improved similarity matrix of each channel, and use the variable step size grid search algorithm to improve the scale factor a in the similarity matrix ij and offset b ij Parameter estimates of ; Step S3-5, taking the means of the corresponding elements of the improved similarity matrices of multiple channels to construct a multi-channel adjacency matrix; Step S3-6, perform feature clustering based on improved spectral clustering, extract the feature with the largest RMI in each cluster of each channel as the sensitive feature, and realize the optimization of the sensitive feature set for multi-channel performance degradation assessment.

3. The tensor domain health status assessment method for fusing multi-channel signals according to claim 1, characterized in that: The features of step S5 specifically include: Step S5-1, using a basic clustering algorithm to achieve a preliminary assessment of the health status of the rolling bearing; Step S5-2, forming a basic cluster Π of the preliminary evaluation results of the health status of multiple channels to construct a connection matrix; Step S5-3, constructing each connection matrix into a joint matrix and a coherent link matrix, and concatenating the two into a third-order tensor; Step S5-4, processing the third-order tensor based on a low-rank tensor approximation method to obtain a redefined joint matrix; Step S5-5, using the redefined joint matrix to construct the adjacency matrix in spectral clustering, and finally realizing the evaluation of the health status of multi-channel data.

Images