An Analogical Theory of Creativity in Design

Creative design often involves large, complex modifications to the design topology. Making these modifications typically requires transfer of design configurations from different designs to the new problem. We describe an analogical theory of creative design called modelbased analogy (MBA). In this theory, case-specific structure-behaviorfunction (SBF) models of past designs enable abstraction of generic telelogical mechanisms (GTMs) at storage time. The goal of adapting a specific design to address a new design problem leads to the retrieval of a relevant GTM and its instantiation in the context of the case-specific SBF model at transfer time. Thus GTMs learned from past designs mediate the transfer of abstract design configurations to a new problem. 1 Background, Motivat ions and Goals Traditional case-based reasoning provides a process account of adaptive design in which the design modifications are small, simple and local. The process is characterized by reminding of a past design case based on its similarity to a given design problem, and direct transfer of the structure of the design case to the current problem. In earlier work (Goel 1991a), we showed that a StructureBehavior-Function (SBF) ontology of physical devices provides a vocabulary for indexing known design cases and measures for determining the functional similarity of a past design with a given f u n c t i o n ~ s t ruc ture design problem. We also demonstrated that case-specific, hierarchically-organized SBF models give rise to an array of modification strategies that enable design modifications of two kinds: modifications to the parameters of design elements (i.e., subdesigns that are recursively decomposable into the primitive components and substances) in the structure of the known design, and replacement of design elements by functionally-similar elements. In addition, we showed that the SBF models enable evaluation of modified designs, and, if the evaluation succeeds, their storage in the case memory organized in function-oriented discrimination networks. We call this computational theory adaptive modeling. In this paper, we describe recent work on creative design characterized by large, complex modifications to the topology of a past design. In particular, we describe a computational theory of extracting, abstracting, and composing design configurations from different designs and composing them with a candidate design. In this theory, case-specific SBF models of past designs enable abstraction of generic behavior-function (BF) models. The goal of adapting a specific past design to address a new problem leads to the retrieval of a relevant generic