Analog circuit fault diagnosis based on feature attention

Abstract

Aiming at the troubles of effective extraction of fault features, large model calculation, low-accuracy diagnosis and poor stability, this paper proposes an analog circuit fault diagnosis method that is based on an improved CNN-Transformer model. To achieve comprehensive and effective extraction of fault features, one-dimensional convolution is implemented to obtain the local features in the data, and multi-head attention is employed to catch the global features. A Sallen-Key band-pass filter, a fourth-order state-variable filter and a Butterworth low-pass filter circuits are applied as the experimental subjects for comparison to verify the effectiveness and advancement of the proposed CNN-Transformer method. The results indicate that of the suggested CNN-Transformer model is able to effectively enhance diagnostic accuracy and stability, achieve accurate diagnosis and localization of circuit fault components, which could be a helpful reference for engineering practice in analog circuit fault diagnosis.