Two-stage prediction technique for rolling bearings based on adaptive prediction model

Abstract

Due to the discrepancy of the components themselves, operating conditions and loads, the bearings show different degradation processes, and a single fixed model cannot describe different degradation processes accurately. In order to solve the adaptive prediction model for component degradation, a hybrid prediction technique is proposed in this paper, firstly, the Theil-sen estimator (TSE) is proposed as the health indicator of bearing operation, and it has been validated that the TSE has good trendability, monotonicity, and robustness. Based on the TSE, in order to identify the time to start prediction (TSP) accurately, the idea of deviation degree is proposed to reflect the difference between local and global degradation trend. Then, the RUL prediction is performed by matching the Wiener model suitable for the prediction task through the model adaptive algorithm. Finally, the effectiveness of the proposed method is verified by XJTU-SY bearing dataset and bearing accelerated degradation experiments.