Sensor Fault Classification And Tracing Method Based On MDBN For Traction Drive System Of High-Speed Train

As one of the core equipment systems of the high-speed railway transportation network,high-speed train(HST)is a bright business card in China.As the main power source of HST,the stable operation of traction drive system is very important to the safe operation of the HST.However,due to the long-term operation in the harsh environment such as large temperature difference,humidity and heat and frequent signal acquisition,the sensor is a high fault prone component of the traction drive system.Once the sensor fails,it will feed back the error signal to the control unit,which will affect the operation performance of the system,and even lead to the collapse of the whole system and serious accidents.Therefore,the research on fault diagnosis of sensor in the traction drive system is of great significance to ensure the safe operation of HST.This thesis focuses on sensor fault classification and tracing methods.The main innovative work is as follows:1.For the problem that it is difficult to classify the fault caused by the coupling of sensor signals in the traction drive system and the neglect of correlation information of a single sensor signal,a fault classification method of traction drive system sensor based on modular Bayesian network is proposed.This method divides different system modules according to the system topology,and constructs the incidence matrix between each module;A modular Bayesian network model is established to solve the problem of slow model training caused by too many nodes.The effectiveness of the proposed method is verified by the hardware in the loop experimental platform.2.For the problem that the sensor fault of traction drive system is easy to change with the working conditions and is easily disturbed by the correlation between the front and back time data,a fault classification method of traction drive system sensor based on real-time modular dynamic Bayesian network is proposed.This method can effectively extract relevant data similar to the current working conditions by learning and tracking the normal changes of working conditions in real time;A local modular dynamic Bayesian network model is constructed to describe the relationship between the time variables before and after,and the sensor multi condition fault classification is realized.The experimental results verify the effectiveness of the proposed method in the multi condition fault classification of traction drive system sensors.3.For the problem that the cascading of traction drive system components makes the sensor fault propagate rapidly,a fault tracing method combining small step real-time modular dynamic Bayesian network and Granger causality is proposed.According to the classification results,the fault zone is found and the initial fault time is roughly located;In terms of time,a method based on small step real-time modular dynamic Bayesian network is proposed to determine the sensor with the earliest fault as the fault source;From the perspective of space,the Granger causality test method is used to analyze the causality of each sensor signal,and the most fundamental Granger cause is determined as the root cause of the fault;The final traceability result is determined by combining the time and space fault information.The effectiveness of the proposed fault tracing method is verified by experiments.