Degradation-trend-dependent Remaining Useful Life Prediction for Bearing with BiLSTM and Attention Mechanism

Abstract

Rolling bearing is an important basic component widely used in mechanical equipment, and its remaining useful life (RUL) prediction is one of the important technologies to realize the health management and predictive maintenance. It can be found in this paper that the degradation law of bearings can be divided into fast degradation and slow degradation after the degenerate point. Based on this important fact, the main exploration is listed into the following parts: Firstly, the root mean square (RMS) is selected as the monitoring indicator from the original signal, and it can be used to judge the degradation mode and determine the incipient degradation time (IDT) according to the 3σ criterion; Secondly, relative transformation is chosen to transform the selected degenerate feature, and input the proposed bidirectional long short-term memory (BiLSTM) with attention mechanism, then the RUL is predicted accurately. Finally, experimental verification is carried out on the validation dataset in the PRONOSTIA, and the comparison with other approaches demonstrate that the proposed method achieves better performance.