Causal interaction: from a high-level representation to an operational event-based representation

We propose to extend the temporal causal graph formalisms used in model-based diagnosis in order to deal with non trivial interactions like (partial) cancellation of fault effects. A high-level causal language is defined in which properties such as the persistence of effects and the triggering or sustaining properties of causes can be expressed. Various interaction phenomena are associated with these features. Instead of proposing an ad hoc reasoning mechanism to process this extended formalism, the specifications in this language are automatically translated into an event calculus based language having well-established semantics. Our approach improves the way fault interaction and intermittent faults are coped with in temporal abductive diagnosis.