摘要

Assembly retrieval is a significant technology to reuse the abundant design knowledge embedded in the products. However, the existing assembly retrieval methods are very few and the research on this topic is still needed. In this paper, an assembly retrieval approach based on shape distributions and Earth Mover's Distance (EMD) is proposed. This approach concerns the shape information rather than the high-level information (such as kinematical information) of assembly model, which gives the user a simpler way to retrieve assemblies. First, a histogram is generated to capture the shape information of a part model based on shape distributions, and a point which represents the part model is constructed by taking the heights of the bins in the histogram as the coordinates. Through this way, an assembly model is simply and quantitatively described as a point set, and the comparison of assembly models is transformed into the comparison of point sets. Afterward, the corresponding EMD-based matching method is given to comprehensively evaluate the differences between assembly models by calculating the dissimilarities between the generated point sets. This matching method also enables the fuzzy retrieval, which allows the user to input several part models rather than the exact assembly model to query. Finally, an assembly retrieval prototype system is developed, and the experiments on the system have shown the effectiveness of the proposed assembly retrieval approach.