An Evaluation Method for Rolling Bearing Performance Degradation Combining Random Matrix Theory and Principal Component Analysis

An evaluation method for rolling bearing performance degradation (RMT-PCA) combining random matrix theory (RMT) and principal component analysis (PCA) is proposed to solve the problems that the traditional feature extraction method will cause the loss of some useful information when processing the high-dimensional monitoring data of a bearing and the existing bearing performance degradation state indicators are difficult to accurately represent the actual operating state. Firstly, through the translation-time window, the rolling bearing monitoring data are locked and a random matrix model is constructed. Secondly, the single-ring theorem and M-P law in the random matrix theory are used to decompose and extract matrix features, and 14 feature indicators are constructed. Finally, the PCA algorithm is used to fuse multiple feature indicators, and the principal components with a large contribution rate are extracted to construct the fused feature indicators for bearing performance degradation assessment. An application research is carried out using the bearing test data in the Experimental Center of University of Cincinnati. Experimental results and a comparison with the abnormal detection algorithm based on the ratio of the maximum and minimum eigenvalues show that: the RMT-PCA method can detect the early abnormality of the bearing 12.5 h earlier; Compared with the hierarchical Dirichlet process-continuous hidden Markov model, the results of RMT-PCA method in detection of early abnormal points and serious fault points are basically the same as that of the former, but its fusion indicator can more clearly reflect the occurrence of the "healing phenomenon" of the bearing in the middle and severe degradation stages. © 2021, Editorial Office of Journal of Xi'an Jiaotong University. All right reserved.