Accelerating in situ X-ray tomography using sparse projections and deep learning

Published

May 30, 2026

Author(s)

Nathan Johnson, Orion Kafka, Newell Moser

Abstract

In situ X-ray tomography experiments provide powerful insight into material deformation and damage evolution under mechanical load. However, laboratory-based tomography instruments often suffer from long acquisition times that limit temporal resolution. In this work, we evaluate a deep-learning-based reconstruction method capable of producing high-quality volume reconstructions from sparse (10x fewer) projection datasets. We apply this approach to in situ tensile experiments on additively manufactured Inconel 718 dog-bone specimens, increasing sampling density from a few tomography scans per experiment to over 20 scans at different load steps. We compare conventional Feldkamp–Davis–Kress (FDK) reconstructions using 1001 projections with both (a) FDK reconstructions using 101 projections and (b) deep-learning reconstructions using 101 projections. Qualitative and quantitative comparisons between reconstructions show that the deep-learning approach reduces noise and preserves major structural features, enabling reliable tracking of pore evolution during deformation. However, differences in intensity distributions and boundary smoothing introduced by the deep learning reconstruction influence segmentation results, particularly for small and intermediate pore sizes. These effects lead to differences in pore size distributions compared to traditional reconstructions. The deep-learning workflow substantially increases temporal resolution, enabling observation of deformation mechanisms such as pore growth, coalescence, and shear band formation that would otherwise be difficult to capture. These results demonstrate that deep-learning-enabled sparse acquisition provides a practical approach for accelerating laboratory in situ XCT experiments, particularly for studies focused on feature tracking and qualitative defect evolution.

Citation

Materials Characterization

Volume

238

Pub Type

Journals

Download Paper

https://doi.org/10.1016/j.matchar.2026.116524

Local Download

Keywords

Materials science, Deep learning, In situ, X-ray microscopy, X-ray tomography, Electron microscopy, Fractography, Mechanical properties, High throughput, Additive manufacturing, Inconel

Citation

Johnson, N. , Kafka, O. and Moser, N. (2026), Accelerating in situ X-ray tomography using sparse projections and deep learning, Materials Characterization, [online], https://doi.org/10.1016/j.matchar.2026.116524, https://tsapps.nist.gov/publication/get_pdf.cfm?pub_id=960470 (Accessed June 4, 2026)

Additional citation formats