Aiming at the problem that support vector machine（SVM） parameters are given arbitrarily or by means of test experience

and the difficulty in selecting

a SVM which optimize parameter by particle swarm optimization（PSO） algorithm is proposed.Data are equally obtained from three types of faults which are normal

bump and wear are used as research training sample set.PSO-SVM classification sample is trained.Some data records from the train data are used to test the model.The research shows that classification accuracy of PSO-SVM model is 93.8 %.Compared with SVM whose parameters are not optimized

the proposed algorithm can find the optimal solution better and improve the classification accuracy of model.