Interpretation of Design Drawings by Analogy

Visual analogical mapping and transfer can be used to derive a structural model of a drawing by analogy, and, moreover, the problem of analogical mapping can be guided by using functional knowledge. We view the interpretation drawings of designs as deriving a structural model of the components and connections of the depicted device. This problem is not deductive in nature but, rather, it is abductive, as there is no a priori reason a shape must represent one object and not another; only with significant help from the design context can a model be derived, and in particular we propose to do it by deriving the model by analogy to a similar drawing with a known structural and teleological model. This requires (1) an analogical mapping from the source (known) drawing to the target drawing that is derived on the basis of shapes and spatial relations, and (2) a transfer and adaptation process by which the old model is transferred to the new drawing and adapted to it. In this paper we are focusing, in particular, on the first task, that of analogical mapping. Analogical mapping is known to be hard. Computing an analogical mapping from a source to a target or query with no further information to guide the search, to determine what ought to correspond to what, is very difficult. We propose to focus the search for a mapping by employing the functional knowledge represented in the teleological model. Let us consider, for example, the task of mapping the source drawing illustrated in figure 1(a) to the target drawing illustrated in figure 1(b), both drawings of a piston and crankshaft. If an agent has a low-level geometric reasoner to recognize the shapes and spatial relations between them, then the agent can treat this shape representation as a set of (mostly binary) relations—A contains B, C is parallel to D, and so on. This is, then, a labelled graph. Analogical mapping must then find a projection of the one graph onto the other that preserves some but not necessarily all of the relations. This is the well-known NP-Hard problem of maximum common subgraph. We begin with a teleological model of the source case, based on structure-behavior-function (SBF) model of physical devices. SBF models treat functional models as consisting of three pieces: (1) the structural model of the components and connections and their functionally relevant proper-