Prognosis of wheel tread degradation status based on PCA-NARX neural network

Abstract

The degradation status of the wheels directly affects the operational reliability of the railway wagons, so the prediction of the degradation status is very important for the health management of the wheels. In this paper, wheel tread wear is taken as explanatory variables, and relevant historical wheel profile measurement and wheel temperature rise are taken as external related variables. The huge amount of wheel temperature rise data is fused by principal component analysis, and the feature combination is selected by wrapper method. Nonlinear autoregressive with external input (NARX) is used to establish the relationship between wheel tread wear and external related variables, and predict the wheel tread wear value. This paper uses the actual measurement data to verify the model, and the results show that the proposed model considering the historical degradation state of wheel treads and related variables could improve the prediction accuracy of the model.