A Geometric Similarity-based CAD Assembly Model Retrieval for Digital Twin.

The digital twin model can be constructed in cyberspace that is fully equivalent to the physical entity, and it can enable designers to explore and optimize new designs based on existing CAD models during the design phase. This requires the ability to accurately retrieve similar assembly models from the existing library. However, in designing single and small-batch products like tools or fixtures, it's challenging to find assembly models with similar structures. A solution is to use digital scanning devices to capture point clouds of the physical product and reverse-engineer a digital model in cyberspace. By comparing the point cloud data with the existing assembly models, designers can find parts or components with similar shapes and improve their design efficiency. This paper proposes a method for retrieving CAD assembly models based on geometric similarity by comparing the shape descriptor derived from the point cloud of the product with the existing models. An experimental study was conducted based on a mixture of public model datasets and local designs. The experimental results showed high accuracy in retrieving partial or whole assembly models based on geometric similarity.