A deep belief network based health indicator construction and remaining useful life prediction using improved particle filter

Abstract The prediction of remaining useful life plays a significant role in prognostics and health management, which can give helpful reference for maintenance. The construction of health indicator is an important part of prediction, and a suitable health indicator can reflect the degradation degree of the system and provide useful information for remaining useful life estimation. This paper presents an unsupervised health indicator construction method based on deep belief network and combines it with particle filter for remaining useful life prediction. Firstly, deep belief network is trained to extract the hidden characteristics corresponding to fault state of a system, and the distance between degraded state and failed state is used to construct health indicator. Afterwards, the remaining useful life before failure is predicted by particle filter which is improved by introducing a fuzzy inference system. At last, a case study using aircraft engine dataset is performed to verify the effectiveness of the proposed method, and the results show that the proposed method gives better prediction accuracy compared to traditional methods.