Fully automated groupings of mechanical designs using CADseek’s Shape Analytics Engine.

Figure 1: Fully automated groupings of mechanical designs using CADseek’s Shape Analytics Engine.

Ames, Iowa – October 7, 2022 – The research lab at the iSEEK Corporation is sharing a sample CADseek Shape Analytics Report for the 3D dataset compiled by the Convergence Design Lab at Purdue University, MCB: Mechanical Components Benchmark (mechanical-components.herokuapp.com). The CADseek Shape Analytics Report is presented on 3DShapeIndex.com free of charge to both research institutions and industry as part of iSEEK Corporation’s effort to promote and advance 3D shape classification, search, and analytics research.

The dataset consisted of 76,727 3D CAD models of mechanical parts and assemblies. The CADseek Shape Analytics Engine generated the shape groupings in 100 seconds on a 4-core i7 workstation using a minimum CADseek Shape Similarity Score of 54%. The report displays groupings with more than two members. Depending on the dataset type, grouping can get larger in size if the similarity score is further reduced. The CADseek Shape Analytics Engine is not a neural networks-based classification engine and, therefore, doesn’t require the manual tagging and training of a neural network. The CADseek Shape Analytics Engine classification algorithm is fully automated and is based on the minimum distance classification of the CADseek shape signature. A dataset can be analyzed at multiple CADseek Shape Similarity Score cutoffs ranging from duplicates, to highly similar, to including similar dominant features, and even fit relationship.

As a reference point, a 75% CADseek Shape Similarity cutoff has optimal correlation between the shape of the design and the manufacturing process and, hence, the cost of manufacturing the design in contract manufacturing and part supplier qualification applications.

According to the Convergence Design Lab website, 3D data was aggregated from TraceParts, 3DWarehouse, and GrabCAD. Their research paper, titled “A Large-scale Annotated Mechanical Components Benchmark for Classification and Retrieval Tasks with Deep Neural Networks”, by Kim, Sangpil and Chi, Hyung-gun and Hu, Xiao and Huang, Qixing and Ramani, Karthik, used the dataset to explore the effectiveness of deep learning shape classifiers on mechanical components. Their paper was published in the Proceedings of the 16th European Conference on Computer Vision (ECCV) in 2020.

The grouping presented in this sample CADseek Shape Analytics Report was generated without any consideration for dimensions, units, or size of the design. However, the CADseek Shape Analytics Engine can generate shape analytics reports where dimensions and units are considered in the CADseek Shape Similarity Score. In the next sample analysis, an automated grouping report will be generated for the Smithsonian 3D dataset on 3DShapeIndex.com as an example to the applicability of CADseek’s fully automated shape analytics algorithms on non-mechanical shapes.