Analog Circuit Fault Diagnosis Based on Machine Learning Using Frequency Domain Features

This paper is devoted to the analog circuit fault diagnosis via frequency measurements. The presented fault diagnosis technique deals with the frequency response of the circuit to create the fault dictionary and employs machine learning classifier for locating hard faults. The selected classifier features are peak values of the real and imaginary part of the voltages at test nodes. This study gives a comparative analysis of a set of supervised machine learning classification algorithms for detecting and localizing faults based on the features extracted from frequency domain. Diagnostic results confirm effectiveness of the method in terms of classification accuracy and diagnosis time.