A novel Multi-Channel Weighted Convolutional Neural Network for fault identification of high-voltage circuit breakers

To solve the problem that single-channel vibration signal input cannot fully express fault feature information and early diagnosis of high-voltage circuit breaker (HVCB) faults, a multi-channel weighted convolutional neural network (MCW-CNN) is proposed and applied to HVCB vibration signal characteristic learning and fault diagnosis. Firstly, wavelet transform is used to extract the time-frequency features of the vibration signal. The vibration signal is converted into multi-channel image input, so as to give full play to the excellent performance of CNN in image feature extraction. Secondly, according to the difference in image frequency and bandwidth of each channel, the dynamic receptive field is used in the convolution layer to extract image features. Aiming at the difference in the intensity of shock features carried by each channel image, a multi-channel fusion method based on kurtosis weighting is proposed to enhance the channel fault features with strong shock features. Finally, the effectiveness of the proposed method is verified by fault diagnosis test of HVCB. Through the fault simulation experiment, the results verify that the MCW-CNN can effectively extract the fault features of the vibration signal, and the recognition accuracy is significantly higher than that of the traditional method.