About this Abstract |
| Meeting |
2025 Annual International Solid Freeform Fabrication Symposium (SFF Symp 2025)
|
| Symposium
|
2025 Annual International Solid Freeform Fabrication Symposium (SFF Symp 2025)
|
| Presentation Title |
Generative Adversarial Networks for Topology Optimization With Orientation |
| Author(s) |
Nathan Hertlein, Joseph Kubalak |
| On-Site Speaker (Planned) |
Nathan Hertlein |
| Abstract Scope |
Leveraging the unique design space of additive manufacturing (AM), many AM-specific topology optimization (TO) techniques help designers find high-performing, readily manufacturable parts. Topology optimization with orientation (TOO) adds considerations for the anisotropic mechanical properties inherent to fused filament fabrication by enabling simultaneous optimization of a deposition orientation field. However, these techniques tend to entail higher computational expense and potential for increased local optimality. Generative models have demonstrated the ability to perform well on conventional TO, creating opportunities for computational savings after an initial training period. In this work, we study how a generative adversarial network (GAN) could accelerate AM-specific optimization such as TOO. We compare GAN configurations and loss functions, and quantify their impact on structural performance. Orientation predictions typically show larger error than topology predictions, but overall performance is promising for our problem class. These techniques could fit into ongoing efforts to increase generality in data-driven approaches for TOO. |
| Proceedings Inclusion? |
Planned: Post-meeting proceedings |