An anomaly detection algorithm of rolling bearing combining random matrix eigenvalues and principal eigenvectors

Abstract

An anomaly detection algorithm combining eigenvalues and eigenvectors in random matrix is proposed, which can solve the problem that present algorithms based on random matrix theory overly rely on eigenvalues and ignore useful information contained in eigenvectors. Firtsly, the time window method is used to select the original data of rolling bearing, and the sampling feature matrix is constructed by extracting bearing feature. Secondly, the eigenvalues and principal eigenvector of sampling feature matrix are investigated and combined to construct comprehensive feature index and its corresponding threshold. Finally, an anomaly detection based on comprehensive feature index is proposed to detect the early anomaly of rolling bearing. The application research is carried out by using bearing datasets of intelligent maintenance center and Xi’an Jiaotong University; the result shows that compared with the single eigenvalue index and kurtosis index, the algorithm based on the comprehensive feature index can detect the abnormal condition of bearing earlier. And the accuracy and effectiveness of the anomaly detection algorithm are proved through spectrum analysis, which provides guidance and basis for fault warning and equipment maintenance.