An Auxiliary Classifier Generative Adversarial Network based Fault Diagnosis for Analog Circuit

To solve the analog fault diagnosis problem with fewer samples, a transformer based auxiliary classifier generative adversarial network (ACGAN) is investigated for circuit by constructing both generator and discriminator in ACGAN pure components. The has high model generality due to its weak inductive bias, but it also increases risk of overfitting on small datasets. Therefore, we use generate sample data enrich dataset mitigate overfitting. However, severely unstable during training period, this reason, confidence mechanism new layer normalization method studied avoid loss conditional information. Take sallen-key filter biquad high-pass as experiment objects. results indicate that can effectively improve accuracy, reach 96.22% 95.35%, respectively.