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

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

doi:10.1038/s41524-021-00542-4

Abstract

We present Parameter Quantification Network (PQ-Net), a regression deep convolutional neural network providing quantitative analysis of powder X-ray diffraction patterns from multi-phase systems. The network is tested against simulated and experimental datasets of increasing complexity with the last one being an X-ray diffraction computed tomography dataset of a multi-phase Ni-Pd/CeO2-ZrO2/Al2O3 catalytic material system consisting of ca. 20,000 diffraction patterns. It is shown that the network predicts accurate scale factor, lattice parameter and crystallite size maps for all phases, which are comparable to those obtained through full profile analysis using the Rietveld method, also providing a reliable uncertainty measure on the results. The main advantage of PQ-Net is its ability to yield these results orders of magnitude faster showing its potential as a tool for real-time diffraction data analysis during in situ/operando experiments.

Highlights

Over the past decade, advancements in X-ray sources, optics and detector technologies have led to a dramatic increase in the volume and data quality of experimental powder diffraction patterns[1,2,3,4,5,6,7]
It should be pointed out that the mean absolute error (MAE) values lated 1D X-ray diffraction (XRD) patterns and it is shown that it can yield accurate provide an indication of accuracy regarding the predictions of the predictions for scale factors, crystallite sizes and lattice parameters network but a Parameter Quantification Network (PQ-Net) model trained with relatively high MAE can for both simulated and experimental XRD data
The results presented in this work demonstrate that the PQ-Net model is able to extract accurate physico-chemical information from XRD patterns

Summary

INTRODUCTION

Advancements in X-ray sources, optics and detector technologies have led to a dramatic increase in the volume and data quality of experimental powder diffraction patterns[1,2,3,4,5,6,7]. It should be pointed out that the MAE values lated 1D XRD patterns and it is shown that it can yield accurate provide an indication of accuracy regarding the predictions of the predictions for scale factors, crystallite sizes and lattice parameters network but a PQ-Net model trained with relatively high MAE can for both simulated and experimental XRD data. The parameter-block part of the architecture contained the dataset majority of the trained weights so our initial work focused on The logical step was to test the PQ-Net against simulated minimizing both the depth and width of the dense layers It was multi-phase XRD patterns which is a closer approximation to most found that the width of the second dense layer has to be doubled experimental data and closer to real world applications.

Dong et al 3

DISCUSSION

Findings

CODE AVAILABILITY