How to optimize structure parameters of support vector machine (SVM) classifier is reseached based on particle swarm optimization (PSO).The information entropy band of the fault signal is used as SVM training sample. Singular spectrum entropy, power spectrum entropy, wavelet energy spectrum entropy and wavelet space feature entropy values are obtained after computing the entropy band value of fault signal in the time, frequency and time-frequency domains. A fault identification method for the rotor system ｉs established based on SVM, whose structure parameters are optimized by PSO training features of SVM are the four entropy band values processed before. Experimental results show that the designed classifier has many advantages of high training speed, short test time and high classification accuracy on fault diagnosis for rotor systems online.