Multi-stage exponential model based on subspace clustering distribution for bearing remaining useful life prediction

Abstract As one of the key components in rotating machinery, the rolling element bearing has been widely used in actual production, such as wind turbines, vehicles and machine tools. A bearing’s remaining useful life (RUL) is an important indicator for its performance assessment, which is related to maintenance and production safety. To overcome the insensitivity of the conventional health indicator (HI) on bearing degradation assessment, this study proposes a subspace clustering method based on manifold learning to evaluate the evolution of health status, which describes the degenerate distribution via a two-class model and realizes the identification of the degradation of each stage. Motivated by the inconsistent degradation process in the application of actual bearing, this study proposes a multi-stage degradation identification criterion in an adaptive way, which can effectively identify different degradation rates of bearing. Based on the different degradation states, a multi-stage degradation exponential model is established to accurately predict the RUL. The effectiveness of the proposed method is validated through open datasets. The experimental results prove that the proposed method can effectively identify different degradation rates and accurately give the boundary time of the multi-stage degradation. The RUL prediction accuracy is significantly improved compared with traditional HI.