Automatic Abstraction of Time-Varying System Models for Model Based Diagnosis

This paper addresses the problem of automatic abstraction of component variables in the context of the MBD of Time-Varying Systems (i.e. systems where the behavioral modes of components can evolve over time); the main goal is to produce abstract models capable of deriving fewer and more general diagnoses when the current observability of the system is reduced and/or the system operates under specific operating conditions. The notion of indiscriminability among instantiations of a subset of components is introduced and constitutes the basis for a formal definition of abstractions which preserve all the distinctions that are relevant for diagnosis given the current observability and operating conditions of the system. The automatic synthesis of abstract models further restricts abstractions so that the temporal behavior of abstract components can be expressed in terms of a simple combination of the temporal behavior of their subcomponents. As a validation of our proposal, we present the results obtained with the automatic abstraction of a non-trivial model adapted from the spacecraft domain.