Fault Diagnosis on Train Brake System Based on Multi-dimensional Feature Fusion and GBDT Enhanced Classification

Abstract

The condition of train brake system directly affects the performance and safety of the train. In view of the low accuracy and inefficiency of train brake system fault diagnosis, a new fault diagnosis method for train brake system based on multi-dimension feature fusion and GBDT enhanced classification is proposed. Firstly, the initial features of the brake system data are extracted from four dimensions, including time domain, frequency domain, wavelet packet decomposition and correlation. Secondly, features, which have great contributions to the fault diagnosis model, are screened out by the ReliefF algorithm, and the uncorrelated components are eliminated with the KPCA algorithm. So the core feature vectors can be obtained with the help of feature selection and reduction. Finally, a gradient boosting decision tree (GBDT) model for fault diagnosis is trained by the core feature vectors. And the model will be used to identify and diagnose the faults of the brake system. Experimental results show that, the fault diagnosis model proposed in this paper can identify the common fault types of brake system, and has high model train efficiency and excellent fault recognition performance.