Research on wear rate of train brake pads driven by small sample data

Abstract

Increased wear leads to the failure of brake pads. Both the International Union of Railways (UIC) and the China Railway Production Certification (CRCC) take wear rate as an extremely important parameter that characterizes the degree of wear, and its value can provide reference for the braking performance of the brake pads. Since the braking performance test of brake pads is time-consuming and costly, the sample data of wear rate is small. In this study, we explore the coupling relationship between the wear rate of train brake pads and its features such as initial braking speed (IBS), braking pressure (FB), braking temperature and average coefficient of friction (ACOF), and propose a Butterfly Optimization Algorithm-Back Propagation (BOA-BP) method suitable for small sample data for the wear rate prediction of brake pads. The grey relational analysis (GRA) and Pearson correlation analysis (PCA) are used to arrive at preferred features for the prediction of wear rate of brake pads, such as IBS, FB, braking distance and ACOF. The dynamic GM(1, N) and the least square method are used for sample expansion. Furthermore, the wear rate is designated as the output feature parameter and the preferred features are designated as the input feature parameters, a BOA-BP model is established for predicting the wear rate of brake pads. The results show that the wear rate obtained by dynamic GM(1, N) is consistent with the actual braking test, the maximum temperature and the braking pressure are important factors affecting wear rate. Compared to the comparison methods such as PSO-BP, GA-BP, and BP, the BOA-BP exhibits better advantages in prediction with small samples, and the average prediction accuracy with 33 sets of data and 99 sets of data are 95.70 % and 97.21 % respectively.