Toward automated topology optimization : identification of non-design features of CAD models using graph neural networks

Volume
Part F3256
Start Page
267
End Page
279
Citation
Proceedings of the International Conference on Additive Manufacturing in Products and Applications (AMPA 2023)
Contribution to Conference
International Conference on Additive Manufacturing in Products and Applications, AMPA 2023  
Publisher DOI
10.1007/978-3-031-42983-5_19
Scopus ID
2-s2.0-85201294151
Publisher
Springer
ISSN
27309576
ISBN
978-3-031-42982-8
978-3-031-42983-5
This paper presents an automated identification of non-design features of CAD models for topology optimization using learning-based segmentation. The CAD files are taken from a large database of industry-relevant components. Based on the geometry and topology of the components, a graph structure is created and processed by a deep neural network. The results show good match with real cases and can be continuously improved by training with additional data.
Subjects
Design Automation
Graph Neural Networks
Topology Optimization
DDC Class
620: Engineering
005: Computer Programming, Programs, Data and Security