Abstract:Wavelet Packet method is used to analyse fault signals from the Ball Screw of CNC machine. By combing with the time-domain analysis, feature values under different fault conditions were drawn so as to achieving the feature vector which was put into the Support Vector Machine (SVM). The binary tree algorithm and RBF kernel function were studied and chosen, then the genetic algorithm (GA) for the optimization of SVM parameters was used, lastly the SVM multi-fault classifier was established to realize fault diagnose and classify of Ball Screw. Finally through experimental results, that shows the feasibility and validity of this multi-fault classifier.