Research On Fault Diagnosis Of Metro Traction Inverter

Urban rail transit plays an important role in the development of the city with high population density,and the normal and stable operation of the subway provides the guarantee for the social production and social life.As an important part of the locomotive,traction inverters keep changing states between operation on stop.There are so many kinds of complex operating conditions and the faults are inevitable.Traction inverter fault will not only disrupt the normal operation of the subway order,but also inconvenience passengers.The research on the fault detection methods of traction inverter for metro vehicle can improve the efficiency of solving the problem,shorten the repair time of locomotive,and ensure the safe operation of the locomotive.On the basis of understanding the causes and types of the failure,the stimulation model of subway locomotive based on the simulation model of topological structure and working principle is built in this paper.After setting all types of single and double IGBT failure modes,current waveforms under various modes are collected.Because the wavelet theory is suitable for the application of time-frequency analysis,the db3 wavelet analysis is used to extract the fault characteristic vectors.Because the support vector machine(SVM)has the advantages of effective generalization performance,avoiding the curse of dimensionality and achieving the global optimum,the support vector machine(SVM)is used to classify the fault characteristic vectors.Firstly,the theoretical basis of SVM classification is introduced in this paper.Then in order to improve the classification accuracy,how to choose the kernel function and optimize the kernel parameters is studied.Two different types of fault characteristic vectors are used to be the sample data.The relationship between the selection of kernel function and parameter and classification results is researched with binary-class SVMs.Considering the operation time and classification accuracy,results show that the radial basis kernel function and cross validation based on the grid parameter optimization method are suitable for classifying fault characteristic vectors of inverters.Because the traction inverter fault classification is classed as multi category,this paper studies one-to-one combination classification algorithm and one-to-many combination classification algorithm,and the two algorithms are based on binary-class SVMs.The results show that one-to-many combination classification algorithm is more suitable for this research problem.The last chapter includes the summary of the paper,improvement in the research progress,and the future research works of the subject.