Circuit Fault Diagnosis Using Simulation and Bayesian Inference

Abstract

Electronics is an integral part of almost all prime industrial equipment, whether it is the computer that drives the device or the sensors that monitor and control equipment. Electronics circuit designs are based on nominal values, tolerance stack ups and physical properties of the components have a major effect on the reliability of electronics. Test upon discovery of a problem will not be able to identify the exact component problem because most failures are not hard failures but only out-of-specification failures (soft failures), which are usually difficult to locate. Instead of replacing the whole circuit when soft failures happen, significant costs can be saved by having a specific, faster and less costly solutions for fault isolation. With the help of simulation-based methods for solving electronic system manufacturing problems, we propose to apply adaptive Markov Chain Monte Carlo (AMCMC) to track and update estimates of the system condition, and to diagnose malfunctions in electronic systems in an efficient and accurate way. Our methods have great advantages when there are no enough data and the failure modes of the system are unknown.