Case-based reasoning, learning and refinement for industrial diagnosis

A description is given of a case-based reasoning system, REFINER, and how it can be applied to an industrial diagnosis problem. The system is capable of identifying inconsistencies among cases and helping a user resolve them. The use of such a system co-operating with a rule based system on the same industrial problem is discussed. The REFINER program is a knowledge refinement tool, for differential diagnosis, which helps an expert elicit and refine his knowledge in a more `natural´ way by talking about individual cases rather than general rules. Further, the system has the ability to point out that two classifications are not distinct, and can then suggest ways in which the inconsistency may be resolved. REFINER, which uses an inductive learning algorithm and background knowledge if it is available, explores a variety of possibilities if the learning algorithm is initially unable to find a complete and consistent concept description. The system assumes that inconsistencies arise due to inaccurately described cases, misclassified cases or indeed, lack of domain knowledge. The system was designed specifically to analyse cases, learn the classification criteria used, and resolve inconsistent uses