About this Abstract |
| Meeting |
TMS Specialty Congress 2026
|
| Symposium
|
4th World Congress on Artificial Intelligence in Materials & Manufacturing (AIM 2026)
|
| Presentation Title |
3D PRIMME for Learning Grain Growth Behaviors from Simulated and Experimental Datasets |
| Author(s) |
Zhihui Tian, Hailey Hall, Michael Tonks, Amanda Krause, joel B harley |
| On-Site Speaker (Planned) |
Zhihui Tian |
| Abstract Scope |
Data-driven grain-growth modeling has advanced in recent years, yet most methods focus on 2D domains, which conflicts with the inherently 3D nature of experimental grain evolution. This work learns 3D grain-growth dynamics from simulation data and extends the model to experimental datasets.
We extend our PRIMME machine learning based grain-growth framework from 2D to 3D and demonstrate its spatiotemporal extrapolation capability. In a representative case, 3D-PRIMME is trained on 96^3 Monte Carlo Potts (MCP) microstructure datasets with a few time steps, yet it performs inference on 1024^3 microstructure with statistically consistent long-term behavior.
Leveraging this robustness, we further apply 3D-PRIMME to Lab-DCT experimental dataset. Results show that our model generalizes well to real microstructures, with validation performed at both the whole microstructure statistical scale and the individual grain scale. Overall, 3D-PRIMME provides an efficient and physically grounded surrogate for 3D grain growth across both synthetic and experimental domains. |
| Proceedings Inclusion? |
Undecided |