An Improved Constraint-Propagation Algorithm for Diagnosis

Diagnosing a system requires the identification of a set of components whose abnormal behavior could explain the faulty sys­ tem behavior. Previously, model-based diagnosis schemes have pro­ ceeded through a cycle of assumptions -* predictions observations assumptions-adjustment, where the basic assumptions entail the proper functioning of those components whose failure is not esta­ blished. Here we propose a scheme in which every component's status is treated as a variable; therefore, predictions covering all pos­ sible behavior of the system can be generated. Remarkably, the algo­ rithm exhibits a drastic reduction in complexity for a large family of system-models. Additionally, the intermediate computations pro­ vide useful guidance for selecting new tests. The proposed scheme may be considered as either an enhancement of the scheme proposed in [de Kleer, 1986] or an adap­ tation of the probabilistic propagation scheme proposed in [Pearl, 1986] for the diagnosis of deterministic systems.