Model-based diagnosis
Description
This dissertation presents an approach to the goal of automated diagnosis, for which diagnosis is defined to be the task of detecting and isolating component faults within a system. The solution is applicable to systems for which a behavioral model is available. Importantly, no information regarding the failure modes of the system, their likelihood, nor their symptoms, is required. Engineered systems are within this domain. The theoretical framework for this solution is presented along with the procedures, an analysis of their complexity, and sample problems The solution is a graph-theoretic extension to the existing consistency-based approach. Using constraint propagation over a constraint graph as the basis results in a simpler formulation of the solution than those presented to date. The use of appropriate constraint solution techniques enables the required focusing to achieve an efficient consistency-based solution. Intuitions regarding adding sensors, swap testing, and bench testing have been formally incorporated in the consistency-based approach. The formal diagnosis has been reconciled with the presentation requirements. The interval algebra has been extended to have open and closed intervals. The computational costs of the steps in the diagnosis have been analyzed