Experience learning in model-based diagnostic systems. (English) Zbl 0718.68083
IJCAI 89, Proc. Int. Conf., Detroit, MI/USA 1989, 1356-1362 (1989).
[For the entire collection see Zbl 0707.68001
The paper introduces a model-based diagnostic system architecture which has test generation capability, and is able to learn from its experience in order to incrementally improve its performances. The system learns heuristic knowledge from experience in terms of symptom-failure association rules and component failure models. It also caches general rules for model-based diagnosis, and test pattern generation for efficient computations. The learning capabilities are supported by the following three constraints on the system:
(1) single fault assumption (i.e., there is only one malfunctioning component in a failure device);
(2) non-intermittency fault assumption (i.e., the behaviour of a failing device does not change during the diagnosis process);
(3) fault locality (i.e., a failed subcomponent tends to fail again in a similar manner in the future).
Implemented in Prolog in a concise form, the experimental results indicate promising data.
|68T20||AI problem solving (heuristics, search strategies, etc.)|
|68T05||Learning and adaptive systems|