Abstract
Automatic 3D part search is advantageous to reducing duplicate designs and facilitating component outsourcing in new product development. Previous studies converted a CAD model into a shape signature and compared the signatures of different models according to a pre-defined similarity function. Various definitions of the shape signature have been proposed and each ha its own limitations in discriminating 3D shapes. Any comparison scheme only based on a single signature may not offer satisfying discernability. Therefore, this research develops a novel search scheme that adopts multiple shape signatures for part comparison. It transforms a solid model into a form-feature adjacency graph with node as a single feature or the shape of feature intersections, and edge indicating the connectivity between nodes. Each node corresponds to a topological graph of the B-Rep volume it represents. Such a hybrid approach can effectively solve the feature intersection problem, and capture the user's intent more accurately during the search. In addition, we propose a set of heuristic algorithms for accelerating the graphs comparison in the similarity assessment. A shape distribution algorithm is applied at the last stage for discriminating the geometric information that the feature or topology-based methods fail to handle. The implementation results validate the practicality of this work in 3D part search.