About this Abstract |
| Meeting |
2026 TMS Annual Meeting & Exhibition
|
| Symposium
|
Artificial Intelligence Applications in Integrated Computational Materials Engineering (AI-ICME)
|
| Presentation Title |
Graph Neural Network-Based Forecasting of Atomic Dynamics at Grain Boundaries |
| Author(s) |
Sohrab Salimi Bani, Matt Uffenheimer, Jun Wang, Panthea Sepehrband |
| On-Site Speaker (Planned) |
Sohrab Salimi Bani |
| Abstract Scope |
A graph-based machine learning framework is developed to efficiently predict long-term atomic trajectories based on short-time molecular dynamics (MD) simulation data. The model is designed to capture atomic diffusion behavior along grain boundary (GB) planes and is implemented on simulations of twist GBs in FCC aluminum (110). A Graph Neural Network (GNN) is trained using atomic positions from early MD time steps. Atoms are represented as nodes in a graph, with their connectivity defined through a K-Nearest Neighbors (KNN) Ball-Tree algorithm. The trained model predicts future atomic positions across sequential time steps, enabling trajectory extrapolation far beyond the original simulation window. This approach provides a promising path toward reducing the computational cost of long MD simulations and enabling extended-timescale analyses of GB diffusion behavior. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Modeling and Simulation, Machine Learning, Electronic Materials |