A case study of maintenance of a commercially fielded case-based reasoningsystem

This article is a case study of the maintenance required to the case base o f a commercially fielded case-based reasoning (CBR) system that provides su pport for HVAC engineers enabling them to better specify HVAC installations . The article briefly describes the system and details how the case base gr ew rapidly, causing a problem of case redundancy. A simple algorithm to ide ntify and remove redundant cases is described, along with the results of ap plying it to the case base. Case obsolescence also was encountered and part ially remedied using DBMS techniques. The article analyzes the case-base ma intenance (CBM) required by the system in terms of Richter's knowledge cont ainers and Leake and Wilson's CBM framework and contrasts this case study w ith 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 bee n unnecessary. The article concludes by identifying lessons learned and hig hlighting the relationship between the sophistication of the case-represent ation and similarity metrics and the ease with which CBM can be undertaken by nontechnical staff. This relationship does not always work in favor of t he maintainer.