Fault Diagnosis Of Three-Level Inverter In High-Speed Train With Compound Faults

This paper focuses on three kinds of compound faults in three-level inverter.There are(1)IGBT compound faults in the same arm,(2)IGBT compound faults in different arms,(3)clamping diode and dc-link capacitor faults.The compound fault diagnosis methods are based on data driven.The work of this paper is as follows:Firstly,the research background,origin and signif icance of this paper are introduced.Based on the main circuit structure and physical mechanism of three-level inverter,the electrical simulation model is built using Simulink and Sim Power Systems.According to the operating environment and operating principle of inverter,three kinds of compound faults which are prone to occur are proposed,there are IGBT compound fault in the same arm,IGBT compound fault in different arms,clamping diode and dc-link capacitor faults.These three kinds of faults are injected into the simulation model,and all the patterns of these faults are analyzed.The qualitative criterion which can judge the two kinds of IGBT faults is obtained.The simulation verif ies the accuracy of the model and the rationality of the compound faults.For IGBT compound fault in the same arm,a diagnosis method using FFT and decision tree is proposed.The feature of fault currents are extracted using FFT,then the decision tree mode is trained using the fault features.The simulation results show that the proposed decision tree model has good classification ability.For IGBT compound fault in different arms,a diagnosis method using wavelet and support vector machine(SVM)is proposed.The adaptive mult iresolution analysis of three-phase current signals is made by using wavelet transform.In order to train and test the SVM classifier,the wavelet coefficients energy value in the main frequency band of the fault signal is extrac ted as the fault feature vector.During the training of SVM,an improved grid search method(IGSM)is proposed to optimize the SVM parameters,so that the training time of classifier is greatly reduced.The simulation results show that the proposed method not only has the fault detection function,but also has a good ability of faults classification.Finally,a method for diagnosing clamping diode and dc-link capacitor faults is proposed.The three-phase output currents with faults are transformed into two-phase dq currents using d-q transformation.An improved empirical mode decomposition algor ithm based on wavelet denoising is used to detect the d-q currents.The results show that the fault features extracted through correlation analys is are more accurate.SVM is employed as the classifying method.During the training of SVM,genetic algorithm(GA)and IGSM are applied separately to design SVM classifiers.The fault classification is performed online using the better SVM classifier.A comparison of the results of the simulation indicates the effectiveness of the method.