A new method for change-point identification and RUL prediction of rolling bearings using SIC and incremental Kalman filtering

Remaining useful life (RUL) prediction for rolling bearings is a key aspect in equipment prognosis and health management. To predict the RUL of rolling bearings, a two-stage degradation model that simultaneously considers environmental noise was first constructed to characterize the evolution of the health indicator (HI). A change-point identification method based on the Schwarz Information Criterion (SIC) is proposed to achieve adaptive switching between the two-stage degradation processes. Then, to overcome the issue of the Kalman filter (KF)-based method ignoring fluctuations in adjacent states, an incremental Kalman filtering (IKF) algorithm is proposed for RUL prediction using online observed HI data. Meanwhile, the expectation maximization (EM) algorithm is used in the absence of prior information to estimate the initial parameters. Finally, the effectiveness of this approach is verified using 16,004 rolling bearing test data points. The results show that the proposed method enhances RUL prediction accuracy by at least 57.12% over traditional KF-based methods and by 31.53% compared to methods that ignore environmental noise.