RUL prediction for two-phase degrading systems considering physical damage observations

Abstract

This paper focuses on a specific type of two-phase degrading system commonly encountered in industrial practice. The first phase is moderate with a low degradation rate while the second is rapid with a high rate. Current studies usually rely solely on sensor measurements to divide phases and predict the remaining useful life (RUL), ignoring the utilization of actual physical damage observations, such as wear depth and crack length. These observations, available during system shutdown periods, directly reflect system states and phase changes. To this end, we propose a novel RUL prediction framework consisting of offline training and online prediction processes. In the offline training process, the physical damage observations and sensor measurements are utilized to estimate the parameters of a two-phase Wiener process and a bijective function matrix. In the online prediction process, real-time sensor measurements are transformed into virtual damage observations for RUL prediction. To enhance the accuracy of phase change point detection, a change point detection algorithm is proposed for both processes. The effectiveness is demonstrated using a simulation and a real case study.