This paper was focused on the faults of main converter with the Simulink toolbox of Matlab, the output voltages of different faults had been obtained. Making use of wavelet analysis to process the output voltage, the fault feature had been extracted and the feature vector had been constructed. The samples had been accessed, which was containing the fault information with the SVM toolbox, the multi-fault classifiers had been established and the samples had been diagnosed. However, in the infinite space, SVM easily lead to wrong classification. For this problem, an improved method had been proposed, which cought make SVM to classify in a limited space. Finally, the simulation samples had been trained and tested. Experimental results show that the classification result is fine and achieves the desired objective.