About this Abstract |
| Meeting |
2026 TMS Annual Meeting & Exhibition
|
| Symposium
|
Computational Thermodynamics and Kinetics
|
| Presentation Title |
Bayesian Design of Experiments for Calphad Modeling |
| Author(s) |
Brandon Bocklund, Isabel R. Crystal, Elizabeth Sobalvarro Converse |
| On-Site Speaker (Planned) |
Brandon Bocklund |
| Abstract Scope |
Rapid exploration of materials design spaces relies on computational models and advanced experiments. Calphad models are often used to predict phase stability and thermophysical properties, however only about 25% of binary systems have thermodynamic Calphad assessments and fewer include phase-based thermophysical properties, primarily due to limited experimental data. Although high-throughput experimental techniques are advancing, there is still a need to efficiently design experiments for fundamental measurements of phase diagrams and properties. While optimal experimental design (OED) techniques are emerging in materials science, they often rely on physically inconsistent surrogate models. Here, we adopt a consistent Bayesian approach for OED that operates directly on Calphad-type models with quantified uncertainty. By propagating uncertainty in Gibbs energy parameters, we demonstrate that the expected information gain for experiments in the temperature-composition space of a phase diagram and correlates with actual information gain when new phase diagram data are incorporated via ESPEI. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Computational Materials Science & Engineering, ICME, Other |