Abstract:In this study, we introduce a method designed to eliminate parallax artefacts present in X-ray powder diffraction computed tomography data acquired from large samples. These parallax artefacts manifest as artificial peak shifting, broadening and splitting, leading to inaccurate physicochemical information, such as lattice parameters and crystallite sizes. Our approach integrates a 3D artificial neural network architecture with a forward projector that accounts for the experimental geometry and sample thickness. It is a self-supervised tomographic volume reconstruction approach designed to be chemistry-agnostic, eliminating the need for prior knowledge of the sample's chemical composition. We showcase the efficacy of this method through its application on both simulated and experimental X-ray powder diffraction tomography data, acquired from a phantom sample and an NMC532 cylindrical Lithium-ion battery.

What problem does this paper attempt to address?

This paper aims to address the issue of parallax artifacts in X-ray powder diffraction computed tomography (XRD-CT) when dealing with large samples. Parallax artifacts can cause shifts, broadening, and even splitting of diffraction peaks, thereby affecting the accuracy of physicochemical information extracted from XRD-CT data, such as lattice parameters and grain sizes. To solve this problem, the authors propose a method based on a 3D artificial neural network, which incorporates a forward projector that considers experimental geometry and sample thickness. This method is a self-supervised XRD-CT volume reconstruction approach designed to be chemically agnostic, meaning it does not require prior knowledge of the sample's chemical composition. Specifically, the core of this method is a generator network that can create a set of realistic XRD-CT images (i.e., an image stack). The input to this generator is a random non-zero constant. The generated images are then converted into sinograms containing parallax artifacts through a differentiable forward operator. This forward operator consists of two parts: the first part adds artificial parallax artifacts based on the experimental setup; the second part is the Radon transform that converts the images into sinograms. The generated XRD-CT sinogram dataset is compared with experimentally obtained sinograms, and a specified loss function is used to evaluate the differences. Based on this comparison, the weights of the generator network are updated to produce sinograms that more closely match the experimental data. The authors validated the effectiveness of this method through both simulated and experimental data. On simulated XRD-CT data, the method accurately reconstructed the signal positions and effectively addressed the parallax artifact issue. Additionally, on experimental data, including custom phantom samples and commercial NMC532 cylindrical lithium-ion battery data, the method also demonstrated good performance. It not only resolved the issues of diffraction peak shifts and broadening caused by parallax but also achieved significant results in the map reconstruction of lattice parameters and grain sizes. In summary, the neural network method proposed in this study provides an effective and chemically agnostic solution for eliminating parallax artifacts in large sample XRD-CT data.

A neural network reconstruction approach for obtaining parallax-free X-ray powder diffraction computed tomography data from large samples

A neural network reconstruction approach for obtaining parallax-free X-ray powder diffraction computed tomography data from large samples

Neural networks for rapid phase quantification of cultural heritage X‐ray powder diffraction data

A Self-Supervised Approach to Reconstruction in Sparse X-Ray Computed Tomography

Accelerating Materials Discovery: Automated Identification of Prospects from X‐Ray Diffraction Data in Fast Screening Experiments

AI-enabled high-resolution scanning coherent diffraction imaging

Parallax in angular sensitive powder diffraction tomography

Towards end-to-end structure determination from x-ray diffraction data using deep learning

Texture tomography with high angular resolution utilizing sparsity

Artifact Identification in X-ray Diffraction Data using Machine Learning Methods

Reconstruction of nanoscale particles from single-shot wide-angle FEL diffractions patterns with physics-informed neural networks

AI-enhanced X-ray diffraction analysis: towards real-time mineral phase identification and quantification

Advantage of Machine Learning over Maximum Likelihood in Limited-Angle Low-Photon X-Ray Tomography

Enhancing deep-learning training for phase identification in powder X-ray diffractograms

Artifact identification in X‐ray diffraction data using machine learning methods

Autonomous Electron Tomography Reconstruction with Machine Learning

Autonomous Polycrystalline Material Decomposition for Hyperspectral Neutron Tomography

Neural rendering enables dynamic tomography

Physics-informed neural networks for diffraction tomography

Phase quantification using deep neural network processing of XRD patterns

Deep Learning of Structural Morphology Imaged by Scanning X-ray Diffraction Microscopy