Research On Fault Diagnosis Of High-speed Train Bo1gies Based On Hierarchical Classification And Evidence Reasoning Rules

With the rapid development of the high-speed railway,train traffic safety problems can not be ignored.Bogie is an important part of high-speed trains,the state of its key components directly affect the smooth running of the train.At present,the research on the fault diagnosis of key components of the bogie is mainly focused on the fault feature extraction and analysis,the sudy from the system level has not been performed.The research object is the bogie two-line suspension system.This paper studies the fault diagnosis method of high-speed train bogies from two aspects:multi-classifier design and multi-channel fusion decision.According to the geometry of the body symmetrical structure,7 single fault conditions are chosen referring different locations of the air spring,lateral shock absorber and anti-snake shock absorber.Firstly,the dynamic monitoring signals of 6 channels on the vehicle body are decomposed by EEMD(Ensemble Empirical Mode Decomposition)method.Then,three IMF(Intrinsic Mode Function)components are selected by the correlation coefficient method.Finally,the approximate entropy and fuzzy entropy are extracted according to the selected IMF component,and each sample has a 6-dimensional information entropy feature.In order to diagnose 7 kinds of faults in single-channel way,the research of multi-classifier is carried out.Based on the standard K-means algorithm,this paper proposes a K-means clustering method based on kernel method to construct hierarchical multi-classification model,with binary decision tree structure to store and top-down decision-making,the SVM with the kernel function as RBF(Radial Basis Function)is selected as base classifier.Then the invasive weed algorithm is introduced to optimize the SVM parameters and compare it with the improved particle swarm optimization algorithm.The experimental results show that the K-means algorithm based on kernel method is more balanced than the standard K-means,and the fault diagnosis recognition rate is higher.The multi-classifier composed of SVM optimized by invasive weed algorithm improves the recognition rate by about 2%compared with the improved particle swarm optimization algorithm.Aiming at the problem that the single channel can not carry on the comprehensive information expression and the fault diagnosis precision is not high,the multi-channel fusion decision-making fault diagnosis method is carried out.Firstly,the evidence,reliability factor and importance weight of each channel are obtained by using the fault diagnosis result of the hierarchical multi-classifier.Then,the method of evidence reasoning is used to fuse the diagnosis results of multiple channels at the decision level.The experimental results show that the fault diagnosis recognition rate of multi-channel fusion decision-making method with evidence reasoning rule is improved by about 3%on the basis of single channel fault diagnosis.