Most model-based approaches to diagnosis require a procedure which checks for the consistency of the observations, given a model of the diagnosed system. When dealing with a dynamically changing system such a procedure must take into account time-varying data. Additional difficulties arise when delays are involved in interactions between variables. The worst case occurs when some of the delays are completely unspecified.
This paper presents an approach to consistency-checking which handles qualitative models of dynamic systems exhibiting time lags. A component-centered ontology is adopted to model the structure of the physical system and an episode-based approach is adopted for representing its behavior over time. An example consisting of a physical process exhibiting transportation lags is used to illustrate the power of the approach. Algorithms are presented and illustrated by an output from an implementation in Prolog called C-CAT (Consistency-Checking Along Time).
The solution proposed represents an extension to Brian Williams' Temporal Constraint Propagation methodology. It also extends the applicability range of existing approaches to model-based diagnosis, permitting its use in tasks such as on-line diagnosis of dynamic systems.