DeepComOR predicts crystallographic orientation relationships between parent and product phases from chemical composition using a composition-aware transformer encoder and a hypernetwork-based rotation predictor, achieving significant advancements in materials screening applications.
Key findings
DeepComOR predicts crystallographic orientation relationships solely from chemical composition.
The framework integrates a composition-aware transformer encoder with a hypernetwork-based rotation predictor.
Achieves mean angular errors of 2.3°, 3.1°, and 4.2° for KS, NW, and Burgers ORs respectively.
Chemical composition encodes sufficient information to constrain crystallography of phase transformations.
Limitations & open questions
The framework's applicability is currently limited to metallic alloys.
Further research is needed to expand the model's capabilities to other material systems.