| About this Abstract |
| Meeting |
2024 TMS Annual Meeting & Exhibition
|
| Symposium
|
Accelerated Discovery and Insertion of Next Generation Structural Materials
|
| Presentation Title |
Interoperable Batch Bayesian Optimization Techniques for Efficient Property Discovery of Metals |
| Author(s) |
Trevor Hastings, James Paramore, Brady Butler, Raymundo Arroyave, Danial Khatamsaz, Douglas Allaire |
| On-Site Speaker (Planned) |
Trevor Hastings |
| Abstract Scope |
When optimizing a material property space for a set of objectives, experimental techniques alone lack the rapidity required to produce new materials on necessary timescales. By employing choice functions within Bayesian statistics, one can optimize an arbitrary material space for any number of desired objectives based on a phase space of possible input data, using tested material specimens as prior knowledge. Herein, high entropy alloy datasets are used as quintessential examples, their atomic fractions as inputs with properties inquirable via calphad models. Using Bayesian optimization in batches, pareto fronts of known alloy datasets are found in short order, proving this method utilizable for “black box” engineering problems. The effect on various project parameters are illuminated—phase space size, number of objectives, elements / crystal systems explored, experimental uncertainty, and optional syntax complexities—illustrating how this type of framework can be used for mechanically relevant industry property targets. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Computational Materials Science & Engineering, High-Entropy Alloys, Machine Learning |