| About this Abstract |
| Meeting |
MS&T24: Materials Science & Technology
|
| Symposium
|
Materials Informatics for Images and Multi-dimensional Datasets
|
| Presentation Title |
Enhancing Rietveld Refinement Analyses with Machine Learning Techniques |
| Author(s) |
Finley Holt, Redad Mehdi, Weiqi Yue, PawanT Tripathi, Daniel Savage, Matthew Willard, Frank Ernst, Roger French |
| On-Site Speaker (Planned) |
Finley Holt |
| Abstract Scope |
Neutron and X-ray powder diffraction techniques generate patterns characterized by reflections in accordance with Bragg’s Law. The Rietveld refinement method is instrumental in analyzing crystalline materials. Currently, Rietveld refinement software necessitates substantial expertise and multiple analysis iterations to achieve convergence. However, by leveraging adaptive neural networks specifically tailored to the experimental setup and materials, the number of iterations required for satisfactory convergence is significantly reduced. This advancement enhances the usability and efficiency of Rietveld refinement software.
The specialized neural networks are trained using simulated diffractograms generated by the diffraction simulation capabilities of GSASII. These neural networks are designed to accurately predict phase fractions, lattice parameters, and the phases present in the sample. Our approach has demonstrated an average deviation of 0.39% from the true phase fractions, with a maximum deviation of 4.45%. Similarly, lattice parameters have been predicted with an average error of 0.46% and a maximum error of 4.06%. |