About this Abstract |
| Meeting |
2027 TMS Annual Meeting & Exhibition
|
| Symposium
|
AI-Enabled Materials Processing: Integrating Accelerated Experimental Workflows and Processing-Aware Machine Learning
|
| Presentation Title |
Automating Rietveld Refinement like a Human Expert in GSAS-II |
| Author(s) |
Chidozie Ezeakunne, Reeju Pokharel |
| On-Site Speaker (Planned) |
Chidozie Ezeakunne |
| Abstract Scope |
High-throughput synchrotron X-ray diffraction generates data in volumes faster than traditional manual analysis can process it, creating a critical bottleneck in the rapid characterization of materials. Rietveld refinement is extensively used for determining crystal structures, extracting precise structural and microstructural parameters, yet automating it correctly without naively treating it as a black box can be challenging as it demands complex expert judgments, phase identification and parameter tuning. To address this, we introduce a mostly autonomous framework with the ability to perform detector calibration, single peak fitting, and Rietveld refinement, combining LLM-based reasoning agents and a deterministic interface to GSAS-II. Refinement parameters are selected through a guided search utilizing Bayesian optimization, with physicality screening to ensure the process remains interpretable.
We compare the results we obtain to both entry-level and expert human users on in-situ datasets and find it capable of facilitating quick feedback at synchrotron beamlines, helping accelerate materials discovery. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Computational Materials Science & Engineering, Machine Learning, Powder Materials |