In this paper

a vacuum circuit breaker test platform was built and four kinds of mechanical failure state

such as principle axles jam fault

were realized. The intrinsic mode function（IMF）needed was gotten by empirical mode decomposition（EMD）of vibration signal measured under normal and fault condition. The energy method was used to get the total energy of all intrinsic mode function weights containing main fault feature information. The total energy of all intrinsic mode function weights was taken as a feature vector

and put into support vector machine（SVM）. The different classification strategy as well as the classification time and accuracy of kernel function were analyzed and compared. By the experiment

OAOT classification strategy and RBF kernel function were chosen for their better classification effect. The study provides the theoretical basis and actual data base for the R & D of the circuit breaker fault diagnosis system.