Bearing RUL Prediction Using Composite Health Index-Based Gaussian Process Regression Model With a Dynamic Change Point

Predicting the remaining useful life (RUL) of bearings is crucial for system reliability and preventing unexpected shutdowns. Aiming at two critical issues in RUL prediction, the first change point (FCP) detection and health index (HI) construction, a novel bearing RUL prediction method is proposed. In the proposed method, a composite HI-based Gaussian process regression (GPR) model is integrated with a dynamic change point detection mechanism. First, an exponentially weighted moving average (EWMA)-based dynamic FCP detection technique is developed, which uses the shift in HI to detect the beginning of degradation. Second, a two-step feature selection technique is proposed to identify the optimal subset of features from candidate features. A novel virtual HI (VHI), termed the revised principal component (RPC), is constructed, which can effectively capture the degradation process and is robust to individual bearing differences. Finally, the GPR model is applied for RUL prediction with a confidence interval. Comparative analyses demonstrate that the proposed EWMA-based FCP detection technique exhibits greater sensitivity to incipient faults compared to traditional methods. Furthermore, the constructed VHI effectively represents the bearing degradation process, and the HI-based GPR model outperforms the existing approaches in RUL prediction accuracy and reliability.