Fault Prediction Based On Bayesian Network For Traction Converter

Electric locomotive is quite a complex structure which has a high requirement of reliability so that it can keep working all the time. Monitoring the real-time state of locomotive and establishing a scheme of failure prediction are very important to prevent failures and to ensure safe operation. Traction converter is installed in electric locomotive and other locomotives which have electric drive systems. It is set in the main circuit of the traction converter and its function is to convert the electric energy between DC system and AC system. Traction converter can control and update a variety of traction motors to make sure the locomotives run regularly. So the reliability of traction converter is an important guarantee for the safe operation of locomotives. The main contributions of this thesis are organized as follows:1. Proposed a method which can apply Bayesian networks to fault prediction of traction converter.This thesis expounded the purpose, significance and the status of the development of fault prediction. It presented the urgency of applying the intelligence fault prediction method to locomotive equipment. An advanced tool named Bayesian network was introduced to fault prediction in this thesis and the superiority was illustrated comparing with the other diagnostic methods. A new technology which applied Bayesian network to fault prediction of traction converter was proposed..2. Clarified the basic theory and method of modeling approach of Bayesian network.Studied the basic theories and methods of Bayesian networks, Bayesian network provides a causal relationship of the data representation method. It can be used to dig the potential relationship of the data and it has plenty of advancements in the uncertainty reasoning and data modeling. Bayesian conditional independence can express relationship of faults significantly.3. Bayesian theory was applied to the fault prediction of traction converter.The thesis selected electric locomotive traction converter as the object of the study. It described the Shoshanna eight electric locomotive converter circuit structures and its working principle. Through the analysis of the converter mechanism and common faults like over-current fault, over-voltage fault, voltage fault, overheating fault, overload, it dealt with the signals which were collected from the signal processing system and determined using fuzzy theory. The thesis used simulation software MATLAB toolbox in FullBNT-1.0.7to build Bayesian network, strengthen the model by Bayesian network parameter learning. It diagnosed the faults and determined what to do next relying on the reasoning ability.The results of the studies had shown that the Bayesian network used for fault prediction was a real-time method with high accuracy. It provided a new, more effective technical support for the reliability study of locomotive traction converter.