Clustering Techniques for Databases of Cad Models

Computer-Aided Design data and Solid Models are of immense importance across the full spectrum of engineering disciplines. Current industrial estimates point to the existence of over 30 billion CAD models in the world, with increasing numbers of them including 3D solid models. While the database and data mining communities have made great strides in management of image, audio and video media, little work exists on how to perform engineering, content-based mining of CAD data and solid models. CAD data, and the engineering consumers of CAD data, have greatly different needs and requirements from traditional database multimedia: large, individual database elements; multidisciplinary components (mechanical, electrical, etc.) with unique access critera; and the lack of an accepted set of readily identifiable features that can be used for model indexing and clustering. This paper presents our approach to perform similarity-based clustering on large databases of solid models of mechanical CAD designs. A publicly available source of solid models is the National Design Repository (http://www.designrepository.org), which contains over 55,000 publicly available solid models from many different engineering disciplines. It is this repository which we test our techniques against. We create a mapping of a solid model's boundary representation and engineering attributes into Model Signature Graphs (MSG). From the MSG's, we create an Invariant Topology Vector (ITV) which captures shape and engineering properties of the CAD models. We use a-clustering algorithm to group the models and provide some of our empirical results to illustrate the approach. We believe that our methodology can form the basis for adapting data mining technquies to large engineering databases and building tools for handling 3D solid models as database media.