Early results from ART2-based clustering for CAD-R like triangular mesh models

The increasing variety and complexity of engineering designs entails taking an automated approach for clustering analysis. Because of data sparseness and nearest neighbor property of high dimensional space, traditional clustering algorithms are not applicable to CAD model clustering. A modified type of ART (Adaptive Resonance Theory) network (ART2) was chosen as a solution to clustering of engineering designs. To describe a triangular mesh CAD-like model, the method of Maximum Normal Distribution was improved, and then Moment Fourier Descriptor (MFD) was extended to Principal Sectional Drawing Moment Fourier Descriptor (PSD-MFD). Experiments are presented that show model clustering result based on the approach is consistent with human visual perception. © 2008 IEEE.