Test and diagnosis of analog circuits: When fuzziness can lead to accuracy

Abstract

Testing and diagnosing analog circuits is a very challenging problem. The inaccuracy of measurement and the infinite domain of possible values are the principal difficulties. AI approaches were the base of many systems which tried to overcome these problems. The first part of this paper is a state of the art of this research area. We present two fundamental approaches, model-based reasoning and qualitative reasoning, and the systems implementing them; a discussion and an evaluation of these systems are given. In the second part, we present and propose a novel approach based on fuzzy logic in order to go further in dealing with analog circuits testing and diagnosis. Tolerance is treated by means of fuzzy intervals which are more general, more efficient and of higher fidelity to represent the imprecision in its different forms than other approaches. Fuzzy intervals are also able to be semi-qualitative which is more suitable to the simulation of analog systems. We use this idea to develop a best test point finding strategy based on fuzzy probabilities and fuzzy decision-making methodology. Finally, a complete expert system which implements this approach is presented.