UBC Theses and Dissertations
Consistency-based diagnosis using dynamic models Watkins, Andrew.
As society grows ever more reliant on increasingly complex technology, so does the importance of being able to quickly detect and diagnose the inevitable failures. Thus, the relatively new field of computer-assisted diagnosis has emerged to help fill this need. To date, expert-systems form the most common paradigm for computer-assisted diagnosis but another approach, that of modelbased diagnosis (MBD), is superior in many respects and offers several advantages over expertsystems. These advantages stem from the fact that MBD uses system models that capture underlying information about both the behaviour and the structure of an artifact. This representation of the artifact allows unanticipated faults to be diagnosed, the system to be simulated, reactions to unusual inputs predicted and makes possible reactive, real-time system control. However, the most important and difficult aspect of M B D is creating useable models; this thesis presents a new approach to modeling for diagnosis that allows simple and declarative models to be written that accurately reflect the behaviour of the artifact of interest. This is done by applying the technique of consistency-based diagnosis (one form of MBD) to the logic-based languages cc and tec. The cc language allows models to be written using concurrent logic agents and constraints. Tec extends this by allowing temporal models and default logic to be expressed. The application of consistency-based diagnosis to cc represents an important contribution to the field of modelbased diagnosis because the concurrent nature of these languages correspond well to the multiple, concurrently operating components found in most artifacts and their use of constraints allow flexible, qualitative models to be written. These benefits are extended by tec, which allows temporal models exhibiting default behaviour to be realized. Together, cc and tec form a new approach to modelbased diagnosis that is declarative, expressive, flexible and able to diagnose temporal models with timeouts.
Item Citations and Data