Fault diagnosis and learning in discrete-event systems with incomplete information

Most state-based approaches to fault diagnosis require a complete and accurate model of the system to be diagnosed. In this thesis, the state-based framework for fault diagnosis in discrete-event systems (DES) first proposed in the Ph.D. thesis of Shahin Hashtrudi Zad [HZ99] is extended to solve the diagnosis problem for models which may be missing a small amount of information. An attempt is also made to recover the missing information using discrepancies between the system's expected behaviour and its actual output. Learning is achieved by using the set-covering model of abductive inference described in the parsimonious covering theory (PCT) of Reggia and Peng [PR90]. The conditions under which failures may be diagnosed and learning is possible are analyzed, and the computational complexity of the diagnosis and learning process is examined. Some methods for reducing the complexity are also discussed.