Commutation failure diagnosis of superconducting magnet power supply based on dual-channel CNN

Aiming at the difficultly of identification and low diagnostic accuracy, a superconducting magnet power supply fault diagnosis model based on a dual-channel convolutional neural network is established. The feature expression ability of data from one dimension to two dimensions is enhanced by constructing the first-order derivative Gramian angular difference field, and the squeeze and excitation dilated residual block is introduced to improve the deep feature extraction performance. The fusion-gated bidirectional long short-term memory is designed to weight the time series dependency. Finally, the convolutional block attention module is introduced to optimize the key feature expression effect. According to the experimental results, the accuracy of the established fault diagnosis model is as high as 99.31%, while the accuracy is still 93.52% under the noise interference of −4 dB. The established commutation failure diagnosis model exhibits strong anti-noise performance and high accuracy, which can ensure the reliable operation of the power supply and safe operation of the fusion device.