Analog circuit fault diagnosis method based on CEEMDAN and SA-DCGAN

Abstract Fault diagnosis of analog circuits is crucial for enhancing the safety and reliability of power systems and reducing operation and maintenance losses. However, under noisy environments and limited fault samples, the accuracy of existing diagnostic methods often declines significantly. To overcome these limitations, this paper suggests an analog circuit fault diagnosis method that combines complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) and an improved deep convolutional generative adversarial network (DCGAN). First, CEEMDAN is employed to decompose the original circuit signals into several intrinsic mode functions (IMFs). The denoised signals are then obtained by reconstructing the IMFs based on correlation coefficients. Subsequently, short-time Fourier transform is applied to the denoised signals to generate time–frequency images for extracting more comprehensive fault information. The SA-DCGAN model is created by incorporating a self-attention mechanism (SA) and spectral normalization into DCGAN to enhance model stability and feature extraction. The time–frequency images are fed into SA-DCGAN for data augmentation, effectively alleviating the issue of limited samples. Finally, a 2D convolutional neural network is used for classification. Experimental results demonstrate that the proposed method achieves diagnostic accuracies of 99.36%, 97.12%, and 95.08% on three benchmark circuits, respectively, with only 20 training samples per class. Furthermore, across a signal-to-noise ratio range of 0–100 dB, it consistently maintains the highest diagnostic accuracy.