Dynamic Diagnosis of Active Systems with Fragmented Observations

Diagnosis of discrete-event systems (DESs) is a complex and challenging task. Typical application domains include telecommunication networks, power networks, and digital-hardware networks. Recent blackouts in northern America and southern Europe offer evidence for the claim that automated diagnosis of large-scale DESs is a major requirement for the reliability of this sort of critical systems. The paper is meant as a little step toward this direction. A technique for the dynamic diagnosis of active systems with uncertain observations is presented. The essential contribution of the method lies in its ability to cope with uncertainty conditions while monitoring the systems, by generating diagnostic information at the occurrence of each newly-received fragment of observation. Uncertainty stems, on the one hand, from the complexity and distribution of the systems, where noise may affect the communication channels between the system and the control rooms, on the other, from the multiplicity of such channels, which is bound to relax the absolute temporal ordering of the observable events generated by the system during operation. The solution of these diagnostic problems requires nonmonotonic reasoning, where estimates of the system state and the relevant candidate diagnoses may not survive the occurrence of new observation fragments.