A Rolling Bearing Fault Early Warning and Diagnosis Technology Based on Spectrum Analysis and Improved MSET

Abstract

Timely fault early warning and accurate fault location diagnosis in rolling bearing are significant to improve the reliable operation of rotating machinery. In this paper, a method for monitoring rolling bearing condition is proposed based on spectrum analysis and improved Multivariate State Estimation Technology (MSET). First, the envelope spectrum of original signal is obtained through Fast Kurtogram (FK). The fixed rotation frequency and fault characteristic frequency of bearing are obtained according to the empirical equation, and the corresponding amplitude of these frequencies in the envelope spectrum is used as the monitoring parameter. Secondly, a nonparametric model of the bearing under normal operating conditions is established via improved MSET, and similarity is introduced to quantitatively measure the similarity degree between the observed state and the normal state. Thirdly, the fault contribution rates of different fault frequencies are calculated for diagnosis of bearing fault types. Finally, the actual operating data of a certain bearing is used as an example for verification. The result shows that the proposed method can provide early warning of bearing faults and accurately identify fault types in the early stage.