Automatic systems diagnosis without behavioral models

Abstract

Recent feedback obtained while applying Model-based diagnosis (MBD) in industry suggests that the costs involved in behavioral modeling (both expertise and labor) can outweigh the benefits of MBD as a high-performance diagnosis approach. In this paper, we propose an automatic approach, called ANTARES, that completely avoids behavioral modeling. Decreasing modeling sacrifices diagnostic accuracy, as the size of the ambiguity group (i.e., components which cannot be discriminated because of the lack of information) increases, which in turn increases misdiagnosis penalty. ANTARES further breaks the ambiguity group size by considering the component's false negative rate (FNR), which is estimated using an analytical expression. Furthermore, we study the performance of ANTARES for a number of logic circuits taken from the 74XXX/ISCAS benchmark suite. Our results clearly indicate that sacrificing modeling information degrades the diagnosis quality. However, considering FNR information improves the quality, attaining the diagnostic performance of an MBD approach.