Research On Fault Diagnosis Method Of Election Unit Traction Motor

Since October,2009, our country’s first high-speed railway, Wu-Guang High-speed Rail,has been ran, election unit, which marks that our country’s railway has been into the moderntechnologies, is widely applied to various railway lines of our country. Traction motor is thecore equipment of election unit, which is closely related to the safe operation of the train.With the continuous increasing of the new type of Election Unit and the continuousimprovement of the operational speed, more and more requirements on safety performance ofthe moto also are put forward. Therefore, the research of fault diagnosis to traction motor hasvery important practical significance.Based on the analysis of the various existing fault diagnosis methods of traction motor,the fault problem of traction motor is studied in this thesis by using wavelet analysis andsupport vector machine(SVM). The main research works are as follows:(1) Traction motor’s structure, working principle, common faults and fault mechanism ofbullet train are analyzed, common fault types of traction motor and internal connectionbetween the various motor fault and characteristic frequency are summarized and the faultcharacteristics are obtained by analyzing.(2) In view of the traditional signal processing method is only suitable for treatment ofstable, non time varying signal, and having no local analysis ability, using the superiority ofwavelet packet analysis to signal denoising and mutation detection, the thesis realizes noisereduction and energy feature extraction to the fault signal with wavelet packet technology andfrequency band energy analysis technology, which is validated by simulation.(3) The classification principles of SVM and least square support vector machines arestudied, the fault diagnosis method of traction motor based on wavelet packet and least squaresupport vector Machines is used, the Wavelet Packet Transformation is used as featureextraction method, the energy value of characteristic frequency band is selected as the trainingand testing samples of least square support vector machines classifier, which is compared withthe fault diagnosis method based on neural network by simulation.(4) Because of the least square support vector machines model parameters have greatinfluence on the classification accuracy and generalization ability, using attractive andrepulsive particle swarm optimizer (ARPSO) and quantum particle swarm optimization(QPSO) to select optimal parameters. After optimization, by using the fault diagnosis methodand traditional least square support vector machines method and the particle swarmoptimization method of parameter optimization, the simulation experiments are carried out,and the results were compared and analyzed.Theoretical analysis and simulation results show that the diagnosis method of least square support vector machines based on wavelet packet analysis technology and QPSO caneffectively detect and diagnose the traction motor fault.