Fault Diagnosis Of IGBT Based On Signal Processing Methods In Traction Inverter System

Railway is an important transportation infrastructure,which plays a vital role in the economic development.With the rapid development of high-speed railway,the reliability of the traction drive system,as a driving source of electric multiple units and high-power electric locomotives,has a large impact on the safe and stable operation of trains.According to the statistics of the faulty devices,IGBT is one of the most fagile components in the traction drive system.The fault of IGBT can be classified into two types: open-switch fault and short-switch fault.Since the time of short-switch fault is so short that it is mainly eliminated by the hardware.If an open-switch fault occurs,it is not only difficult to find its specific location,but it would also cause other faults to damage the whole system seriously.Therefore,it is of great significance to detect and locate the open switch fault of IGBT timely.First of all,this paper takes the IGBT in the traction inverter as the research object,and it briefly introduces the working principle,control mode of the inverter and the simulation model of the traction inverter system on MATLAB/Simulink software platform.The failure classification of IGBT is briefly described.Then,four kinds of typical fault conditions are selected,and the electric quantities(stator currents,speed and torque of asynchronous motor)are analyzed and discussed,which provides a theoretical basis for the fault diagnosis method of IGBT.Then,according to the characteristics of electrical quantities,three-phase currents are selected as the fault electrical quantities,wavelet packet energy Shannon entropy,wavelet packet energy Tsallis entropy,empirical mode decomposition energy Shannon entropy,Shannon empirical mode decomposition energy Tsallis entropy are applied for fault detection respectively.In order to avoid false alarms,the wavelet packet energy Shannon entropy is selected.And it is combined with DC component method to diagnose the open-switch fault of IGBT.However,the entropy method is easy interfered by system noise and is unable to diagnose all the two-IGBT faults.In order to solve that,spectral kurtosis based on Choi–Williams distribution is added to original method,and the performance of the proposed method is analyzed.In addition,the improved method is verified on the RT-LAB platform.Finally,a new method based on the combination of fault feature and classifier is proposed to diagnose the fault of IGBT.For the IGBT failure,wavelet transform and EMD are applied to calculate the three-phase current energy probability and construct the fault features that combined with typical BP and Elman neural network to train and diagnose samples.At last,the fault features and classifiers are compared by the accuracy of diagnosis and the results of training and diagnosis time.