Robust Operative Diagnosis as Problem Solving in a Hypothesis Space

Kathy H. Abbott

The lack of robustness in current diagnostic systems is an important research issue because it has two major consequences: inability to diagnose novel faults and inability to diagnose more than one type of fault. This paper describes an approach that formulates diagnosis of physical systems in operation (operative diagnosis) as problem solving in a hypothesis space. Such a formulation increases robustness by: (1) incremental hypotheses construction via dynamic inputs, since the fault propagation results in changes in symptoms over time; (2) reasoning at a higher level of abstraction to construct hypotheses, albeit less specific ones, when specific knowledge is not available; and (3) partitioning the space by grouping fault hypotheses according to the type of physical system representation and problem solving techniques used in their construction. The approach was implemented for aircraft subsystems and evaluated on eight actual aircraft accident cases involving engine faults, with very promising results.


This page is copyrighted by AAAI. All rights reserved. Your use of this site constitutes acceptance of all of AAAI's terms and conditions and privacy policy.