Automatic Abstraction in Component-Based Diagnosis Driven by System Observability

The paper addresses the problem of automatic abstraction of component variables in the context of Model Based Diagnosis, in order to produce models capable of deriving fewer and more general diagnoses when the current observability of the system is reduced. The notion of indiscriminability among faults of a set of components is introduced and constitutes the basis for a formal definition of admissible abstractions which preserve all the distinctions that are relevant for diagnosis given the current observability of the system. The automatic synthesis of abstract models further restricts abstractions such that the behavior of abstract components is expressed in terms of a simple and intuitive combination of the behavior of their subcomponents. As a validation of our proposal, we present experimental results which show the reduction in the number of diagnoses returned by a diagnostic agent for a space robotic arm.