Neural networks for multiple fault diagnosis in analog circuits

Abstract

Fault diagnosis of analog circuits is a complex problem. The authors discuss how the features of neural networks of learning from examples and of generalizing may be used to solve this problem. In a detailed applicative example, it is shown how, given the voltages values in a set of test points, a network may be trained to recognize catastrophic single faults on a circuit part of a direct current motor drive. The network is then used to diagnose multiple faults on two and three components. In this case the network is generally able to detect at least one of the malfunctioning components, although less sharply than in the case of single faults.