A novel approach for analog circuit fault diagnosis based on Deep Belief Network

Traditional fault diagnosis of analog circuits relies heavily on feature extraction and selection, which is ad-hoc and often needs complex signal processing and domain knowledge. This has severely limited the applications of fault diagnosis. To address this issue, this paper proposes an analog circuit fault diagnosis method based on Deep Belief Network (DBN). Our contributions include development of an intelligent diagnosis solution that does not rely on manual feature extraction and selection, and providing comprehensive comparison studies on two representative experimental circuits with different levels of complexities under soft fault modes. One significant advantage of the proposed method is that it extracts features adaptively from the raw time series signals and automatically classifies the fault mode, which significantly simplifies the design of diagnosis and increases the flexibility so that it can be applied to different diagnosis problems. The experimental comparison studies show that the proposed method has higher performance, lower requirements on data (small number of sampling points in learning instance), and more reliable performance (consistent diagnosis accuracy for different fault modes) than existing methods. Performance regarding the number of instances and the number of sampling points in instances are studied. The results demonstrate the effectiveness of the proposed method in analog circuit diagnosis.