| About this Abstract |
| Meeting |
2025 TMS Annual Meeting & Exhibition
|
| Symposium
|
Chemistry and Physics of Interfaces
|
| Presentation Title |
Describe, Transform, Machine Learning: Feature Engineering for Grain Boundaries and Other Variable-Sized Atom Clusters |
| Author(s) |
Eric R. Homer, Braxton Owens, Gus LW Hart, Tyce W Olaveson, Jacob P Tavenner, Edward M Kober, Garritt J Tucker, Nithin Mathew |
| On-Site Speaker (Planned) |
Eric R. Homer |
| Abstract Scope |
Machine learning approaches have emerged as a useful tool to obtain structure-property relationships in complex, high-dimensional scenarios, like those of the atomic structures of grain boundaries. However, grain boundaries and other variable-sized atom clusters rarely have feature sizes that are consistent across a given dataset, thereby requiring a transform to fix the feature sizes. We examine a combinations of atomic structure descriptors, transforms, and machine learning algorithms for their impact on the machine learning predictions. Specifically, we predict interface energy in a dataset of 7000 aluminum grain boundaries. The discussion of the results focuses on accuracy of the predictions as well as interpretability of the different combinations in order to learn the physics of the structure-property relationships. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Machine Learning, Computational Materials Science & Engineering, Aluminum |