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 |
Scalable Thermal Surrogate Modeling for LPBF Using STEAM |
Author(s) |
Berkay Bostan, Praveen Vulimiri, Dhruba Aryal, Albert To |
On-Site Speaker (Planned) |
Berkay Bostan |
Abstract Scope |
This work presents STEAM (Scanwise Thermal Emulator for Additive Manufacturing), a scalable surrogate modeling framework for scanwise thermal simulations in LPBF. STEAM is trained on a compact yet geometrically diverse dataset and captures representative thermal behavior through engineered features that account for both node–heat source interactions and the influence of local geometry. Its memory-efficient design supports training and inference on datasets exceeding 40 billion points. On average, test parts are 21 times larger than the training blocks in the spatio-temporal scale and include thermal predictions for up to three melting events per node. STEAM attains a mean absolute percentage error of 6.5%, a mean absolute error of 30.2 °C, and an R² score of 0.97, while achieving up to 1421× speedup using four GPUs. With linear scalability across multiple GPUs and robust generalization to complex geometries and diverse toolpath strategies, STEAM offers an effective solution for high-fidelity, real-scale LPBF simulation and process optimization. |
Proceedings Inclusion? |
Planned: Post-meeting proceedings |