Research On Remaining Useful Life Prediction Based On The Adaptive Wiener Process

Remaining useful life (RUL) prediction is the core of the condition monitoring, prognostics and health management. In view of the difficulty of mechanism modelling, data-driven approaches to establishing models for RUL prediction enjoy a wide popularity. Once the accurate and infor-mative RUL prediction is obtained, an effective support for the subsequent management can be launched, so the losses introduced by the sudden failures of devices and the maintenance cost-s can be brought down. To enhance the accuracy of the RUL prediction, this thesis focuses on how to establish accurate and reasonable degradation models to characterize practical degradation trajectories.We first review the recent related work of RUL prediction. And then, we introduce detailedly statistic based data-driven approaches to RUL prediction, i.e., RUL prediction based on random variable regression models, Wiener process models, and gamma process models. The advantages and disadvantages of these models are thoroughly analyzed.Based on the sufficient survey, we propose three different degradation models, i.e., the lat-ten Wiener degradation model with an adaptive drift, the non-linear and heterogeneous Wiener degradation model with an adaptive drift, and the adaptive skew-Wiener degradation model. Each model is proposed to relax some assumptions of the previous model to approximate the practical degradation trajectories. According to the characteristics of each model, we propose correspond-ing online learning methodologies, i.e., with more measurements coming out, the latten states and the unknown parameters are updated at each sampling instant. Furthermore, the explicit RUL distributions are derived, so the RULs can be predicted in real time.Finally, we summarize the thesis and point out the future research.