IEEE Industrial Electronics Society, Vol. 64, Issue. 1, pp. 605-614, 2017, DOI: https://doi.org/10.1016/j.microrel.2016.12.015

Signal Model-Based Fault Coding for Diagnostics and Prognostics of Analog Electronic Circuits


Zhenbao Liu 1, Taimin Liu1, Junwei Han1, Shuhui Bu1, Xiaojun Tang1, Michael G. Pecht2
1Northwestern Polytechnical University, Xi'an, China
2CALCE, Center for Advanced Life Cycle Engineering, Department of Mechanical Engineering, University of Maryland, College Park, Maryland 20740, USA

Abstract:

Analog circuits have been extensively used in industrial systems, and their failure may make the systems work abnormally and even cause accidents. In order to monitor their status, detect faults, and predict their failure early, this study proposes a signal model-based fault coding to monitor the circuit response after being stimulated to perform a fault diagnosis without training a large amount of sample data and fault classifiers. Manifold features extracted from circuit responses are associated with a fault-indicating curve in the feature space, in which a group of fault bases are uniformly and continuously distributed along with gradual deviation from the nominal value of one critical component. These bases can be deployed in a factory setting but used during field operation. Fault coding is converted to a novel optimization problem, and the optimized solution forms a fault code representing fault class, suitable for realizing fault detection, and isolation for different components. A fault indicator based on comparison between fault codes can describe performance degradation trends. To improve the prediction accuracy, historical degradation data are collected and considered as a priori exemplars, and a novel exemplar-based conditional particle filter is proposed to track a degradation process for the prediction of remaining useful performance. Case studies on two analog filter circuits demonstrate that the proposed method achieves relatively high fault diagnosis and prognosis accuracy. The main advantages of our study are two-fold: first, the high diagnostic accuracy can still be obtained even if there is no large amount of training data; second, the prognostic effect remains relatively stable whenever triggering prognosis module.

This article is available online here and to CALCE Consortium Members for personal review.

[Home Page] [Articles Page]
Copyright © 2018 by CALCE and the University of Maryland, All Rights Reserved