Qualitative Reasoning Group

This paper reports on research in progress, investigating the use of process-centered quali_ tative models in the diagnosis of continuous variable systems. I view diagnosis as a problem of minimally perturbing a faulty model of a system until it matches observed behavior. I claim that qualitative models can account for observed symptoms without relying on explicit fault models, thereby in(--easing robustness for novel faults. Both the mechanism of failure acid the resulting state of the failed system can be modeled in terms of their underlying active processes. By focusing on the process structure rather than the physical structure, the search space is both smaller and more focused, since only those processes which could influence the system in the observed directions need to be considered. Several diagnostic strategies are being explored within this framework. An implementation is in progress, building on a modified ATNIS and an incremental qualitative envisioner. The kinds of systems I am examining are taken from the domain of thermodynamics, and include piston and gas-turbine engines, a steam boiler and a thermal control system.