ABSTRACT
In design, past experience is often used to guide and inspire new design solutions. Thus, CBR is a natural framework for an automated design system, as it allows prior designs to be stored in the case base to help solve new problems. The complexities of engineering design, however, rarely permit an existing solution to entirely solve a new design situation. Therefore, existing designs typically are used as a basis for the new design, but must be further adapted in order to properly address the characteristics and issues associated with the new problem situation. This adaptation process is difficult because it must consider many existing designs in conjunction to solve a new problem, each of which may be represented differently and also may not readily lead to a valid solution if not combined and adapted in a systematic way. In order to alleviate some of these difficulties, we investigate a methodology for adaptation that uses constraint satisfaction techniques. Each existing design case is formulated and stored in the case base as a CSP, and a repair-based CSP algorithm (Minton, Johnston, Phillips, and Laird, 1992) is then used to adapt these existing cases to solve new problems. In this way, we provide a general formalism for adaptation that can systematically find solutions to new problems and, because of its formalization as a CSP, it can be applied across a general class of design problems. We have implemented our methodology in COMPOSER, a case-based adaptation system, and tested it in two design domains: assembly sequence generation and configuration design.
