|
Abstract : |
As technology advances, more and more complex devices and systems are designed and built. Consider electronic circuits, nuclear-power plants, petrochemical refineries, and space shuttle launch processing systems. Because of their complexity, these devices are difficult to diagnose when they malfunction. Research in automated diagnosis addresses this problem by constructing computerized diagnostics that can pinpoint faults based on a device's observed misbehavior. Traditional expert-system technology can be applied to automated diagnosis, as illustrated by successful medical diagnostic programs such as Mycin and Internist. This approach uses production rules that directly relate symptoms to the possible underlying causes. An alternative approach to diagnosis is based on an underlying model of a device 's structure and behavior [1, 2]. Device diagnosis from first principles uses a model of a physical device to explain discrepancies between its faulty and normal behavior. This model-based approach has some advantages over the associational, heuristic-rule approach of conventional expert systems. It is compositional in that it, |
