PCA-Based Analog Fault Detection by Combining Features of Time Domain and Spectrum

Abstract

In view of the difficulties caused by component tolerance, a method based on principal component analysis (PCA) is proposed for fault detection of analog circuits. The basic model of the proposed method and the general rule for analog fault detection are described in detail. At first, the principal component model of fault-free circuit is constructed. Then the circuits-under-test are compared with the principal component model to calculate the statistic for fault detection. The features in both time and frequency domain are combined by this method to detect faults of analog circuits with tolerances. This method was applied to detect faults of the Sallen-Key filter. The results show that it can detect both catastrophic and parametric faults of analog circuits effectively, avoid the shortcomings of single variable statistics and overcome the difficulty to determine threshold by empirical knowledge.