A signal segmentation and feature fusion based RUL prediction method for railway point system

Abstract

Railway point system (RPS) is an important connecting element in railway industry. Faults in RPS increase railway costs of operation and maintenance, and cause delays and even train accidents. Prognostic and health management (PHM) technique can reduce the risks of equipment due to its powerful ability of remaining useful life (RUL) prediction. However, RUL prediction for RPS is a challenging task due to the difficulty of identifying the state for each operating phase related with specific failure modes (FMs) and constructing a comprehensive degradation state from the selected features. To cope with this challenge, this study presents a novel framework for the RUL prediction of RPS. Specifically, a segmentation approach of the power signals is first proposed by constructing an optimization model, in order to identify the FM-related operating phases. Then, a construction method of the health index (HI) is put forward to obtain the comprehensive degradation information by fusing the selected features, which indicate the partial information of the degradation state, and are extracted with popular approaches of feature extraction. Next, the RUL prediction model is developed based on the constructed HIs. Final, this framework is evaluated using the monitoring data of RPS collected from China Railway Guangzhou Group. The experimental results demonstrate that the proposed RUL prediction framework is very effective for RPS.