About this Abstract |
| Meeting |
2026 TMS Annual Meeting & Exhibition
|
| Symposium
|
Vacancy Engineering in Metals and Alloys
|
| Presentation Title |
A combined physics-based and data-driven prediction of vacancy formation energies in refractory non-dilute random alloys |
| Author(s) |
Mahshad Fani, Oluwatimilehin Akinloye, Anvesh Nathani, Subah Mubassira, Iman Ghamarian, Shuozhi Xu |
| On-Site Speaker (Planned) |
Mahshad Fani |
| Abstract Scope |
Vacancy formation energy (VFE) is a fundamental descriptor of defect-mediated phenomena, such as diffusion, creep and radiation tolerance, in metallic alloys. In this work, density-functional theory (DFT) calculations are first employed to generate a comprehensive dataset of VFEs across pure metals, binaries, ternaries, quaternaries and a single quinary composition in the Mo–Nb–Ta–V–W system. We structure this dataset so that each input provides (i) the VFEs of the constituent pure metals, (ii) the alloy’s chemical composition, and (iii) the specific element for which vacancy energy is to be predicted. These inputs feed into four machine learning surrogates, including random forest, XGBoost, graph neural network, and graph attention network (GAT), whose outputs are the VFEs of elements in multicomponent alloys. The resulting models achieve accuracy comparable to DFT while reducing the computational effort by orders of magnitude, making it practical to screen defect properties across many alloy compositions. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Modeling and Simulation, Machine Learning, High-Entropy Alloys |