Time: Wednesday 7.6.2006, 13:30
Place: Room A5-006, Fr. Bajersvej 7
Problems of observation such as monitoring, fault-diagnosis, etc, generally suppose that the observable part of the plant is fixed. This is modeled by a set of observable events, or an observable part of the plant's state. What if observables are not fixed in advance? In this talk we examine some problems that arise, in the context of fault-diagnosis (observe a plant and detect if a fault occurred). In particular, we show that minimizing the set of observable events so that the plant is diagnosable (i.e., faults can be detected) is an NP-complete problem. We also study dynamic observers, which are allowed to change their observation set during execution.