A dual adaptive generalized remaining life prediction method based on GP-Wiener

Abstract Remaining useful life prediction (RUL), as an important part of equipment health management (PHM), how to improve its prediction accuracy becomes the key of the problem. In this paper, a digital-model fusion method for RUL prediction with dual adaptive nature is proposed for complex systems with multiple sensor monitoring signals. Aiming at multiple ideal properties of health indicators (HIs), a nonlinear fusion method based on genetic programming (GP) is used to construct composite virtual health indicators (VHIs) to realize adaptive selection and fusion of sensors; a degradation model is established using a nonlinear Wiener process with implicit scale transformations, and the a priori parameters of the model are estimated based on the historical data using the method of great likelihood (MLE) and updated a posteriori through the Kalman filter (KF) to update the model parameters a posteriori based on the field measurement data; At the same time, an adaptive RUL correction method is used to reduce the huge error of RUL prediction due to the difference between the filtered and measured values, and the corresponding corrected RUL probability density function is obtained. Finally, the proposed method is validated in the simulated degradation data and C-MPASS dataset, and compared with various methods horizontally and vertically, the proposed method has the highest prediction accuracy and the highest prediction stability, which proves the validity and generalizability of the proposed method.