Research On Bayesian Network Based Fault Diagnosis Of Train Control Systems

With (he rapid development of high-speed railway and the innovation of high-speed rail technology, the train transport has played more and more important role in people’s daily lives. However, the train control system with long-time running may lead to inevitable failures, which do great harm to the RAMS (Reliability. Availability, Maintainability and Safety) of train control systems. Therefore efficient and accurate fault diagnosis of the train control system would be a key problem in the assurance cf safe and efficient operation of high-speed trains.Currently, many researchers have invested a lot of efforts to research fault diagnosis in high speed railway by all kinds of methods, such as the Neural Networks, Fault Tree analysis and so on. Although the Neural Network method is an intelligent diagnostic method with self-learning ability, it is not a good solution to the problem of learning incomplete data. The Fault Tree analysis method is a simple method, but it’s difficult to adapt to uncertain characteristics of fault decision in train control systems. Considering the fact that faults generated from the running train has typical features, such as incomplete data and uncertain decision-making, this thesis proposed the novel ideas and conesponding methods to develop intelligent fault diagnosis and maintenance strategies for train control systems by using Bayesian network, which has been proved to be good enough in learning ability and the perfect mechanism for the probability calculation. It can significantly improve the accuracy of fault diagnosis, the RAMS of the train control system,and optimize the maintenance strategy.This thesis’s main works are listed as follows:(1) Based on the current situation of Fault diagnosis in train control system, a novel fault diagnosis method based on Bayesian Network is proposed, in which fault samples are generated through the general description of the train control system failure and the data mining of the fault tables on train tracks. In addition, the analysis and designs method in fault diagnosis system based on Bayesian Network are proposed in this thesis.(2) Constructing Bayesian network is the most important works in completing the fault diagnosis of the train control system. In this thesis, the models of fault diagnosis are built by the fault trace table as a data source. Through research on the relevance between all of fault by combing data mining technologies and the knowledge of expert. the Bayesian network slructure of fault diagnosis of train control system is conslructcd. In addition, the priori probability is obtained by the fault samples mined from the fault trace table.(3) By integration of Matlab, featuring with an unparalleled advantage in dealing with the mathematical problems, and Microsoft Visual Studio2010, an irreplaceable position on the design of forms and interface displays, we developed the fault diagnosis demo system of the train control system.In this thesis, a fault diagnosis method based on Bayesian network was proposed. Fault samples and Bayesian network model of fault diagnosis was made, moreover, fault diagnosis system based on Bayesian network was developed.In summary, the proposed fault diagnosis system will achieve the conversion from the theory ofBayesian network analysis to the practical application in fault diagnosis of high-speed railway. It can effectively improve the efficiency of fault diagnosis, reduce labor intensity for maintenance personnel, provide technical support for the system maintenance, moreover, provide strong support for safe and efficient operation of train.