Fault Diagnosis Method of Analog Circuit Based on Enhanced Boundary Equilibrium Generative Adversarial Networks

Abstract

In the actual working process of the analog circuit, the probability of multiple component failures at the same time is lower than the probability of a single component failure, which makes the single fault data samples and multiple fault data samples tend to show imbalanced characteristics. However, most of the existing data-driven analog circuit diagnosis methods focus on the balance data sample set. Therefore, it is hard to satisfy the needs of fault diagnosis during the actual working of analog circuits. In response to the problems raised above, an analog circuit fault diagnosis method based on enhanced boundary equilibrium generative adversarial network (EBEGAN) is proposed. The generator of boundary equilibrium generative adversarial networks (BEGAN) uses conditional variational auto encoder (CVAE), which can enhance the generated sample quality while ensuring sample diversity. In addition, by introducing the classified loss factor into the loss function, the discriminator has the ability to distinguish the true and false and the type of samples. The experimental results indicate that this study proposes the new method in the situation of imbalanced data, the type of fault in the analog circuit can be accurately identified. compared with the existing analog circuit fault diagnosis methods.