Prediction of the Remaining Useful Life of a Switch Machine, Based on Multi-Source Data

Abstract

Aimed at the shortcomings of a single feature to characterize the health status and accurately predict the remaining life span of the equipment, a prediction method for a switch machine, based on the weighted Mahalanobis distance (WDMD), is proposed. The method consists of two parts: the construction of a health indicator, based on the weighted Markov distance and the prediction of the remaining useful life, based on the hidden Markov model (HMM). Firstly, a kernel principal component analysis (KPCA) is used to extract the characteristics of the power curve data of the switch machine, and the characteristics with a high correlation with the degradation process are screened, according to the trend indicators. Secondly, the resulting features are combined with multi-source information, as the input, and a comprehensive health indicator (HI) is constructed by the weighted fusion of the WDMD algorithm, to characterize the degradation process of the switch machine. The degradation model of this HI is established and trained by the HMM, so as to predict the remaining life span of the equipment. Finally, the actual operation data of the railway field is selected to verify the prediction method proposed in the paper. The results show that the state recognition and the life prediction accuracy of the proposed method is higher, which can provide effective opinions for the predictive maintenance of the switch machine equipment.