Content-Based Classification of CAD Models with Supervised Learning

Abstract

This paper describes how to apply machine learning to adjust shape matching techniques to suit different classification of CAD models. Existing research based on either group technology or fixed modeling matching algorithms, impose a priori categorization schemes on engineering data or require significant human labeling of design data. This paper describes a general technique for “teaching” existing model comparison algorithms to adapt to different classifications that are relevant in many engineering applications. In this way, the core shape matching algorithms can be learned to adapt to wide variety of model classifications based on user input and training data. This allows for great flexibility in search and data mining of engineering data. Results are presented that show how to optimize different parameters to fit different categorization schema. The paper presents a comprehensive example and provides empirical results using a nearest neighbor classification on mechanical CAD models from the National Design Repository. © 2005 by CAD Solutions Company Limited.