A New Kind of Dual-level Retrieval Approach for CAD Models

As the first step of a common case-based design, retrieving a candidate CAD model meeting the new requirements quickly and effectively has a profound effect on the subsequent product design. Accordingly, a new kind of dual-level retrieval approach for CAD models is proposed in this study. It combines geometric information with topological information to carry out model retrieval in a coarse-to-fine manner. In detail, to accurately describe the global and local (geometric) shapes of the query model, a new geometric descriptor is designed based on D2 and Point Feature Histogram. Using the geometric descriptor to make a coarse retrieval, the CAD models that have similar global and local shapes will be found as preliminary candidate models. Then, to finely find the one(s) that has the most similar topology with the query model from them, as well as see their primarily global and local shapes through their geometric details, a new topological descriptor is designed after determining the key faces of each CAD model. The experiments show that the accuracy of our proposed approach is better than that of MVCNN++ based on deep learning on average. Compared with the approaches developed based on D2 and attributed graph, the precision of our proposed approach is improved by about 10% on average; and the recall is improved by 5% on average as well. Moreover, our proposed approach improves the efficiency of graph-based retrieval. © 2023 CAD Solutions, LLC.