Gianfranco Lamperti, Marina Zanella
In diagnosis, the notion of observation varies according to the class of considered systems. In discrete-event systems, an observation usually consists of a sequence, or a set of sequences, of totally ordered observable events. This paper extends the concept of discrete-event observation in several ways. First, observable events (messages) may be uncertain in nature, both in behavioral models and in system observations. Uncertain messages are specified by variables ranging on finite sets of observable labels. Second, messages relevant to a system observation are accommodated within a DAG, the observation graph, whose edges define a partial temporal ordering among (uncertain) messages. This way, an observation graph implicitly defines a finite set of system observations in the traditional sense. Consequently, solving a diagnostic problem amounts to solving at one time several traditional diagnostic problems. Finally, the (possibly distributed) reconstruction of the system behavior is further complicated by the fact that homonymous observable labels can be shared by different components. This raises the need of dealing with null messages. The method is appropriate for several real systems, where messages may get lost, are noisy, or attached timestamps are generated by different clocks.
Keywords: Diagnosis, Model-Based Reasoning, Knowledge Representation
Citation: Gianfranco Lamperti, Marina Zanella: Uncertain Temporal Observations in Diagnosis. In W.Horn (ed.): ECAI2000, Proceedings of the 14th European Conference on Artificial Intelligence, IOS Press, Amsterdam, 2000, pp.151-155.