关键词: 相似性评估, 自组织特征映射神经网络, 聚类, 检索

Abstract: To overcome the shortcomings of traditional attribute-based retrieval method and to realize geometrical-content-based retrieval for mechanical three-dimensional (3D) CAD models, a new approach of clustering and retrieval of mechanical 3D CAD models was presented. Firstly, The STandard for the Exchange of Product model data (STEP) AP203 Part21 files of CAD model were transformed into attributed-graph files by searching and matching keywords. Secondly, feature invariants were extracted and feature invariant vector of CAD model was formed by calculating graph related attributes such as the total number of nodes and edges. Finally, a Self-Organization feature Mapping (SOM) neural network model was employed to cluster and retrieve CAD models by using the extracted invariant vector as its input to train the neural network. The proposed approach was verified to be valid and feasible based on 60 real industry 3D CAD models, and the experimental results showed that it could meet general requirements of engineering retrieval.

Key words: similarity assessment, self-organization feature mapping neural network, clustering