A parameter adaptive data-driven approach for remaining useful life prediction of solenoid valves

As crucial parts of various engineering systems, solenoid valves (SVs) are of great importance and their failure may cause unexpected casualties. Accurately predicting the remaining useful life (RUL) of SVs helps making maintenance decision before they break down. It is hard to establish accurate physical model of SVs as they are characterized by complicated structure, multi-physics coupled working mechanism and complex degradation mechanisms. Different individuals may experience distincted degradation processes in various working environment. In this paper, a data-driven prognostic method is proposed for SVs. Firstly, a health index based on the dynamic driven current of SVs is constructed and an exponential model is established to characterize the degradation path. Then, the particle filter (PF) is introduced to reduce the noise of online measurement. Based on the denoised measurement, the parameters of the exponential model are adaptively updated with Bayesian estimation dynamically. Finally, the effectiveness and practicability of proposed method is validated by the designed experiments on SVs.