Abnormal State Detection for Rolling Bearings Using DBSE Approach

Abstract

Rolling bearings play an important role in mechanical equipment. Abnormal state detection of rolling bearings is significant to avoid industrial accident. In this paper, we use Distribution-based States Extraction (DBSE) approach to detect and analyze the abnormal state of rolling bearings. We apply DBSE approach on three different datasets. Each dataset contains time series of horizon acceleration and vertical acceleration right before a specific fault of a rolling bearing. The results show that the DBSE is able to extract normal states and abnormal states before all three faults and effective for fault prediction. By analyzing statistical properties of two states for all three faults, we find distributions of acceleration data on two states are quite different. Further more, we find that standard deviations are more effective to describe features of different states than means of acceleration data. Our finding are great useful for abnormal detection and fault prediction.