A Novel Fault Detection Method Based on Multiple Features for Analog Circuits
Tianyu Gao,Jingli Yang,Jianfeng Wang,Shouda Jiang
DOI: https://doi.org/10.1109/phm-nanjing52125.2021.9612952
2021-01-01
Abstract:The reliability and security of analog circuits are becoming increasingly significant. Fault diagnosis methods can identify fault classes of analog circuits, thus locating the fault components. However, the fault diagnosis methods based on multi-classification learning framework suffer from the problem of desirable classification effect in the case of lack of fault samples. To address these issues, a fault detection method based on multiple features for analog circuits is proposed in this paper. By learning only normal samples to obtain control limits, the proposed fault detection method can effectively determine the health states of analog circuits. First, features of the output signals of the circuit under test (CUT) in the time domain, frequency domain, and time-frequency domain are calculated to comprehensively reflect its states. In addition, the construction method of related similarity (RS) features is introduced to achieve feature enhancement, which further explores the essential information in the features. Then, to remove redundant features, the feature selection is adaptively performed by using the quantum particle swarm optimization (QPSO) algorithm, where the fitness is the improved Wilks statistic (IWS). Finally, the feature vectors are transmitted to the fault detection model based on kernel principal component analysis (KPCA) to identify the health states of CUT. The experimental results indicate that the proposed method exhibits excellent detection performance for analog circuits in the case of lack of fault samples.