Content area
Full Text
(ProQuest: ... denotes non-US-ASCII text omitted.)
Reprint requests to: Patrick W. Yaner, LogicBlox, Inc., Two Midtown Plaza, Suite 1880, 1349 West Peachtree Street, NE, Atlanta, GA 30309, USA. E-mail:
1.
MOTIVATION AND GOALS
Drawings, that is, external two-dimensional (2-D) graphical representations, are a central component of the design process (e.g., Ferguson, 1992). Larkin and Simon (1987) describe some of the advantages of using drawings in problem solving in general: drawings focus search, afford perceptual inferences, and enable easy recognition of elements such as shapes (e.g., circle) and spatial relations (e.g., between). According to them, these cognitive advantages accrue because drawings use location to group information about a single element, avoiding the need to match symbolic labels; drawings group all information that is used together, which helps avoid large amounts of search; and drawings automatically support perceptual inferences that are easy for humans. Ullman et al. (1990) analyze the importance of drawings in engineering design.
Recognition of shapes and spatial relations in a design representation enables classification and indexing of the representations, and retrieval of appropriate design knowledge. In CAD, shape similarity among three-dimensional (3-D) design representations is a major research issue in indexing and retrieval of design and manufacturing knowledge. Cardone et al. (2003) and Iyer et al. (2005) survey the state-of-the-art in computational techniques for 3-D shape search. In contrast, we are interested in computational techniques for deeper semantic analysis of 2-D CAD drawings generated with vector-graphics tools. In particular, in addition to recognition of shapes and spatial relations in an unlabeled 2-D design drawing, we are interested in the recognition and labeling of structural components and connections depicted in the drawing. Labeling of the structural components and connections should enable deeper classification and indexing of design drawings and indexing and retrieval of the functions and behaviors of the components and connections.
This theory is implemented in a program called Archytas, which analogically infers shape and structure in a target drawing (the input) from that given in a source (or base) case. This program reads in a 2-D unlabeled drawing and, given the source drawing and associated teleological model, attempts to infer by analogy a representation of the shapes and spatial relations in the target drawing and a representation of the structural...