Analog circuit fault diagnosis based on across space-channel attention and double 1D-convolution network

Abstract Circuit fault diagnosis based on deep learning is an important means to mitigate risks and reduce losses. To extract circuit fault features and explore their correlation, a circuit fault diagnosis method based on Across Space-Channel Attention and Double 1D-Convolution (ASCDC) is proposed in this paper. Firstly, a Double 1D-Convolutional Neural Network (DoCNN) is designed to extract local and global features of the electrical voltage signal. Secondly, an Improved Stacked Residual Structure (ISRS) is constructed to enhance channel feature and eliminate redundant information. Thirdly, to explore deeply the key features and their correlation, an Across Spatial-and-Channel Attention Network (ASCAN) is proposed. It can assign higher weights to critical channel feature and position since ASCAN can perform fine-grained weight processing for each channel and different positions within those channels. Finally, the circuit fault is diagnosed using the full connection layer and the Softmax classifier. The two-stage differential amplifier circuit is widely used in the aerospace field, so it is employed for data acquisition in this paper. Experimental results show the proposed ASCDC model can effectively diagnose the circuit fault with an overall accuracy 98.66%, 4.57% higher than the RNDCAM method.