Research On Fault Diagnosis Of Railway Wagon Air Brake System Based On Bayesian Network

With the continuous increase of the carrying capacity and running speed of China's railway wagons,the running safety of wagons has become increasingly prominent.As an important part of the railway w.agon,air brake system plays an important role in maintaining the running safety and the order of railway transportation.China's railway wagons are mainly equipped with air brake system with 120 type control valve,the complex mechanical structure and harsh environment caused brake system faults occurrence frequently.Accurate diagnosis and location after faults occurrence have become a research hotspot of railway wagons.This dissertation carried out the research on fault diagnosis of railway wagons air brake system.By analyzing the common failure mechanism of the brake system and the existing inspection mode,a fault diagnosis method based on Bayesian network for the air brake system was proposed.Based on the statistics of braking faults and probabilistic reasoning in Bayesian networks,this method obtained the results of common brake fault diagnosis and fault location,and solved the difficulties of fault diagnosis caused by the fuzzy of brake system faults to a certain extent.Following research works were completed in this dissertation:(1)The structure and working principle of the air brake system equipped with 120 type control valve were analyzed.Based on single car test data of brake system,four kinds of common failures and their failure mechanisms were analyzed in terms of brake sensitivity,brake stability,improper release and natural release.(2)By analyzing the diagnosis and inspection methods of the brake faults in actual operation of railway wagons in China,a fault diagnosis method of air brake system based on Bayesian network was proposed.(3)Fault classification method based on support vector machine(SVM)for four kinds of common brake failures was proposed.The training set and test set data were obtained from the air pressure data of train tube and brake cylinder in single car test.In MATLAB,based on the libsvm-3.22 toolbox,SVM training and verification experiment of classification results were completed.(4)The Bayesian network structures for four kinds of common brake failures diagnosis were established by expert knowledge preliminary.The training set data for Bayesian network learning were obtained from the running fault statistics.Based on the training set data and K2 algorithm in GeNIe2.2 software,the Bayesian network structures obtained through expert knowledge were modified.Finally,the parameter learning of the diagnostic Bayesian networks were completed based on EM algorithm.(5)The results of fault classification of SVM were used as evidence inputs for diagnostic Bayesian networks.In GeNIe2.2,reasoning of Bayesian networks for four kinds of common braking fault diagnosis were completed based on joint tree algorithm.According to the reasoning results,the brake fault diagnosis and location were completed.