Online fault prognostics based on degradation-oriented slow feature analysis and temporal smoothness analysis

Fault prognostic, which estimates the time to failure and risk for existing and future failure modes has recognized as a key component in maintenance strategies currently. The fault degradation process is in general slowly varying autocorrelated and shows typical nonstationary nature. However, most of existing prognostic methods don’t consider the nonstationary characteristics of the fault degradation process. Besides, stationary fault information may bury the fault degradation trend. In this work, a fault degradation modeling and online fault prognosticating strategy is proposed based on degradation-oriented slow feature analysis and temporal smoothness analysis. First, a fault degradation-oriented slow feature analysis (FDSFA) algorithm is proposed to extract critical fault degradation features. Second, a stability weighting factor (SW) is defined to evaluate the stability of each feature and a key fault degradation factor (KFDF) is calculated by combining critical features and SW. Considering the nonstationary characteristics of KFDF, a time-varying regression model with temporal smoothness regularization is developed for online fault prognostics. Case study on a real industrial process illustrates the validity of the proposed method.