In order to enhance mechanical fault detection of high voltage circuit breaker and the generalization of diagnosis methods

a diagnosis method for mechanical faults based on identity map convolutional neural network（CNN） is proposed. The vibration information of the circuit breaker is obtained roundly by using multiple acceleration sensors. The fault diagnosis model based on CNN with the identity mapping module is designed

and the signal is input through the preprocessing means such as down sampling and data splicing. The model synthetically analyzes the signal characteristics of each sensor and identifies faults. In this method

a direct connection channel is added to the network structure. Compared with the CNN used in previous papers

this method can be adopted to alleviate the problem of gradient disappearance and improve the failure recognition rate of verification set from about 96% to 100%. Compared with the method of artificial feature extraction

this method can be adopted to extract fault feature information independently. The comparative results reveal that

compared with traditional model which extracts the sum of amplitude of the difference interval and then identifies using the weighted support vector machine

and compared with the convolutional neural network model which does not contain the identity mapping module

the proposed method can still reach 100% diagnosis accuracy under different severity of faults

and has better generalization.