Diagnosis of Discrete Event Systems by Independent Windows

Xingyu Su

ARTIFICIAL INTELLIGENCE SEMINAR PhD Monitoring

DATE: 2013-05-29
TIME: 11:30:00 - 12:00:00
LOCATION: NICTA CRL Boardroom (level 2)
CONTACT: JavaScript must be enabled to display this email address.

ABSTRACT:
In the literature, diagnosis of discrete event systems (DES) is performed by computing the paths on the complete model that generate the observations received on the system, or equivalently the belief state. This belief state can be computed on-line, by iteratively computing the set of states can be reached from the current set through transitions that would produce exactly the next observation. If this is done explicitly, the number of these states makes the approach inapplicable for many real-world problems. Symbolic approaches have been proposed when the states are represented in propositional logic, but this representation is also subject to exponential blow-up. Finally, the potential belief states can be pre-computed, but their number is double exponential.

We propose new algorithms to slice the observations into windows, each diagnosed independently. Doing so allows to cope with intermittent observations and to ignore the overhead of maintaining a precise estimate of the system state. We show how diagnosability can be verified using this approach a" and any diagnosis algorithm in general a" through the notion of a simulator, which is a modified model that simulates how the diagnosis algorithm computes the diagnosis.