ECAI-2002 Conference Paper

[full paper]

Diagnosis of Discrete-Event Systems with Model-Based Prospection Knowledge

Roberto Garatti, Gianfranco Lamperti, Marina Zanella

Diagnosis of discrete-event systems is a complex and challenging task. Within the class of active systems, such a complexity is exacerbated by the possibility of queuing (possibly uncertain) events within connection links, thereby making essential the simulation of link behavior during the reconstruction of the system reaction. However, reconstructing such a reaction without any prospection in the search space is generally bound to detrimental backtracking. To cope with this short-sightedness, we present a technique which allows the automatic generation of prospection knowledge relevant to the mode in which events are produced and consumed in links. Such a far-sighted diagnosis requires that a collection of prospection graphs be generated off-line, based on the system model, which are then exploited on-line to guide the search process. As a result, both time and space can be considerably reduced on-line. The approach is worthwhile whenever time constraints are far more severe on-line (when the diagnostic engine is running) than off-line (when no diagnostic process is ongoing), which is commonplace in a large variety of real systems.

Keywords: Diagnosis, Model-Based Reasoning, Discrete-Event Systems, Knowledge Compilation

Citation: Roberto Garatti, Gianfranco Lamperti, Marina Zanella: Diagnosis of Discrete-Event Systems with Model-Based Prospection Knowledge. In F. van Harmelen (ed.): ECAI2002, Proceedings of the 15th European Conference on Artificial Intelligence, IOS Press, Amsterdam, 2002, pp.427-431.