In view of the influence of the accuracy of the empirical mode decomposition（EMD）on mechanical fault diagnosis of high-voltage circuit breaker

a new method combining the improved EMD with boundary extension and the RBF neural network is proposed to improve the mechanical fault diagnosis for high-voltage circuit breaker.The signal boundary is extended with the least squares method to overcome the end-effect in EMD decomposition and to reduce the fitting error of signal.The mechanical vibration signal is decomposed into a finite number of independent IMF functions by the improved EMD method.Then

the energy entropy of each IMF envelope with different frequency is calculated

and the energy entropy vector is taken as the input of RBF neural network.Finally

the RBF neural network model is trained with the alternating gradient method to achieve accurate diagnosis of mechanical fault of circuit breaker.Experimental results show that the proposed method can effectively diagnose mechanical fault type of circuit breakers.