Infrared image-enhanced convolutional neural network for solid-state power controllers fault diagnosis

Aiming at the problem that the temperature information of the solid-state power controllers (SSPCs) contains both fault features and a large amount of redundant data, a convolutional neural network (CNN) fault diagnosis algorithm based on information fusion is proposed. The quality of infrared image is optimised by image pre-processing algorithm, the fault features in multiple information domains are extracted by combining the electrical signal features, and the high-dimensional feature data are down-scaled by principal component analysis (PCA) technique. On this basis, a SSPCs fault diagnosis model is constructed with the outstanding feature learning capability of CNN, and its effectiveness is verified through experiments. The proposed method is superior in both diagnostic accuracy and efficiency, and is of great significance to ensure the safe and reliable operation of SSPCs.