A Health Indicator Construction Method based on Deep Belief Network for Remaining Useful Life Prediction

Abstract

Remaining useful life (RUL) prediction is of great importance in a successful prognostics and health management system. The performance of RUL prediction is mainly decided by the development of an appropriate health indicator (HI), which can accurately indicate the degree of degradation of the equipment. Therefore, we proposed an unsupervised method for HI construction based on deep belief network (DBN) by using multisensory historical data. Firstly, DBN is employed to describe the hidden representation corresponding to the healthy state. With the running of the system, its performance will decrease over time and the corresponding potential characteristics tend to be different. The deviation degree of degraded state can be used to establish HI so as to estimate the RUL. Finally, a case study is conducted to validate the effectiveness of the new method, where it can be seen that the new approach achieves better performance compared to traditional methods.