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 |
STEAM: A Scalable Surrogate Modeling Framework for Scanwise Thermal Simulation in Laser Powder Bed Fusion |
| Author(s) |
Berkay Bostan, Praveen Vulimiri, Dhruba Aryal, Albert To |
| On-Site Speaker (Planned) |
Berkay Bostan |
| Abstract Scope |
This work introduces STEAM (Scanwise Thermal Emulator for Additive Manufacturing), a scalable surrogate model for scanwise thermal simulations in LPBF. STEAM is trained on a compact, geometrically diverse dataset. It captures representative thermal patterns using carefully engineered features that encode node–heat source interactions and local geometric effects on the thermal field. STEAM enables memory-efficient training and inference on datasets exceeding 40 billion data points. On average, test parts are 21× larger than training blocks in the spatio-temporal scale and include predictions for up to three melting events per node. STEAM achieves a 6.5% mean absolute percentage error, 30.2 °C mean absolute error, and an R² score of 0.97, while delivering up to 1421× speedup on four GPUs. Its linear multi-GPU scalability and strong generalization across geometrically complex parts and alternative toolpath strategies make it well suited for real-scale LPBF simulation and process optimization. |
| Proceedings Inclusion? |
Planned: Post-meeting proceedings |