Two-stage Remaining Useful Life Prediction Based on the Wiener Process With Multi-feature Fusion and Stage Division

Abstract

Remaining life prediction (RUL) is a critical link of maintenance decision-making, the accurate RUL prediction is an important means to monitor the operating status and achieve the safe operation of equipment. However, existing studies rarely considered the multi-stage characteristics of indicator fusion in the degradation process, and directly used the Wiener process to establish degradation model, which results in significant errors in RUL prediction results. Therefore, to solve above issues, a two-stage RUL prediction method of bearing based on the Wiener process model with data fusion and stage division is proposed in the paper. Firstly, the concept of multi-feature fusion is introduced to construct a comprehensive health indicator (CHI) that considers indicator performance. After that, a two-stage RUL prediction model based on the CHI is developed, and a method for detecting change points and dividing stages is proposed. Finally, the effectiveness and predictability of the proposed method and CHI are demonstrated based on the bearing test datasets.