Spectral Clustering Analysis of CAD Model Based on Multi-Feature Fusion

Geometry and topology are the key descriptive factors of CAD models. In order to improve the description ability of models, a multi-feature fusion method is adopted to express models. Aiming at the defects of traditional model clustering algorithm, such as sensitivity to irregular error data and poor effect, this paper use STEP file to represent CAD model and spectral clustering algorithm to carry out model clustering analysis. First, the STEP (product data interchange) standard is used to obtain the geometric and topological data of the model. The CAD model is represented as adjacency graph. Then, according to the graph theory, models are represented as attribute adjacency graph. The multidimensional features of models are weighted to measure the classification ability, so as to calculate the similarity between models. The new similarity calculation method replaces the traditional similarity measurement matrix. The new similarity measurement method is applied to the spectral clustering algorithm of models, which makes the processing of arbitrary shape data of CAD model more simple and feasible. Experimental results show that this algorithm can better deal with the problem of CAD model clustering, and greatly improve the effect of CAD model clustering.