A deep convolutional neural network for real-time full profile analysis of big powder diffraction data

This paper introduces Parameter Quantification Network (PQ-Net), a regression deep convolutional neural network designed for quantitative analysis of powder X-ray diffraction patterns from multi-phase systems. The network's accuracy is validated using simulated and experimental datasets of increasing complexity, culminating in an X-ray diffraction computed tomography (XRD-CT) dataset of a Ni-Pd/CeO₂-ZrO₂/Al₂O₃ catalytic material. PQ-Net accurately predicts scale factors, lattice parameters, and crystallite sizes, comparable to Rietveld method results, and provides reliable uncertainty measures. Its significantly faster processing speed highlights its potential for real-time diffraction data analysis in in situ/operando experiments.

Hongyang Dong, Keith T. Butler, Dorota Matras, Stephen W. T. Price, Yaroslav Odarchenko, Rahul Khatry, Andrew Thompson, Vesna Middelkoop, Simon D. M. Jacques, Andrew M. Beale, Antonis Vamvakeros