A case study of maintenance of a commercially fielded case-based reasoning system

This article is a case study of the maintenance required to the case base of a commercially fielded case-based reasoning (CBR) system that provides support for HVAC engineers enabling them to better specify HVAC installations. The article briefly describes the system and details how the case base grew rapidly, causing a problem of case redundancy. A simple algorithm to identify and remove redundant cases is described, along with the results of applying it to the case base. Case obsolescence also was encountered and partially remedied using DBMS techniques. The article analyzes the case-base maintenance (CBM) required by the system in terms of Richter's knowledge containers and Leake and Wilson's CBM framework and contrasts this case study with experience from NEC and DaimlerChrysler. The article observes that had maintenance of the case base been considered more explicitly during system design and implementation, some of the resulting maintenance would have been unnecessary. The article concludes by identifying lessons learned and highlighting the relationship between the sophistication of the case-representation and similarity metrics and the ease with which CBM can be undertaken by nontechnical staff. This relationship does not always work in favor of the maintainer.