Extracting geometric attributes directly from scanned data sets for feature recognition

Abstract

Reverse engineering (RE) is the process of digitally capturing physical entities of an existing part from its scanned data. A novel methodology is proposed for reconstruction of CAD models by recognizing prismatic features from a data set of three-dimensional scanned points, which utilizes the concepts of artificial neural networks (ANNs). Four geometric attributes, such as chain code, convex/concave, circular/rectangular, and open/closed attribute, are extracted from a scanned point set first, and then they are presented to the ANN module for feature recognition. Each generic feature in the feature library is uniquely described by those geometric attributes. Identifying each feature requires the determination of these attributes beforehand. Once these attributes are determined, a segmented point set may be uniquely identified as a valid feature through the ANN module. Since feature recognition is carried out based on these attributes, this paper focuses on algorithms for determining these attributes directly from a scanned point set. The system validation and sample results are also discussed.