AI That Enables Real-Time Analysis of Steel Microstructure Using X-Ray Diffraction

This work presents a convolutional neural network capable of directly predicting phase fractions (ferrite and austenite) in steels from experimental X-ray diffraction patterns. Trained on over 40,000 synchrotron datasets spanning diverse compositions and conditions, the model achieves high accuracy (~2% error), comparable to conventional Rietveld refinement.

Importantly, predictions are obtained in milliseconds, enabling real-time analysis during high-throughput experiments and paving the way for adaptive and autonomous materials research. 

Title: Deep Learning for Real-Time Phase Quantification from X-ray Diffraction: Toward High-Throughput Steel Microstructure Mapping
Authors: Imed-Eddine Benrabah, Guillaume Geandier, Olha Nakonechna, Benoît Denand, Hugo Van Landeghem, Alexis Deschamps, Sébastien Y. P. Allain 
References: Advanced Engineering Materials, 2026 
DOI: https://doi.org/10.1002/adem.202503172