Bayesian Network Based Fault Diagnosis Method Of On-board Equipment For Train Control System

On-board equipment is an important part of train control system, which is the key to ensure the safety of train operation. However, in the process of the train operation, on-board equipment faults have occurred from time to time. The existing fault diagnosis methods of on-board equipment have been unable to meet the needs of high-speed operation of modern railway. Hence, it is of great significance to achieve the fault diagnosis of on-board equipment with the method of intelligent fault diagnosis.Currently, domestic fault diagnosis of on-board equipment is dependent on the expert knowledge, and in the process of fault diagnosis, the utilization of data is very low. On the basis of deep research on the structure and function of on-board equipment, the thesis proposed a fault diagnosis method based on Bayesian network for on-board equipment. This method is aimed at the diversity and uncertainty of the fault in train control system on-board equipment and combined the expertise knowledge with historical fault data, which realized the intelligent analysis in fault diagnosis of on-board equipment.The main works of this thesis include:(1) Based on the historical fault data of on-board equipment, fault dictionary is established with the method of data mining. Besides, the weight of fault feature words in fault text entries is calculated and the fault text attribute weight table is discretized to design the on-board equipment fault diagnosis decision table. After that, attribute reduction is conducted for the fault decision table with the method of rough set and attributes reduction based on genetic algorithm. The reduction results are considered as fault symptom nodes for the diagnosis network to reduce the complexity of Bayesian network model.(2) Based on the training data set and expert experience, Bayesian network diagnosis model is established to improve the accuracy of the diagnosis model. The training data set and the expert experience are based on the fault data of on-board equipment. Besides, parameter learning of Bayesian network is applied with the method of maximum likelihood estimation to perfect the fault diagnosis model based on Bayesian network and to create the condition for the accurate inference of Bayesian network.(3) Based on the historical fault data of on-board equipment, Text mining for fault symptoms is studied to figure out the relationship between the potential fault symptoms and to update the diagnostic Bayesian network model. According to the fault data, parameter learning is carried out for the fault symptom association Bayesian network to further complete the parameters of the diagnosis network model. Verified with the fault texts, it is proved that Bayesian network fault diagnosis with symptom association have better accuracy and higher reliability.(4) According to the requirements of the fault diagnosis system, the visual fault diagnosis system of on-board equipment is developed based on mixed programming of C# and Matlab, which provides the decision support for the on-board equipment fault diagnosis.This thesis proposed a fault diagnosis method based on Bayesian network for train control system on-board equipment, realized the intelligent fault diagnosis of on-board equipment, improved the precision and efficiency of diagnosis, and provided a new effective method for fault diagnosis of train control system.