Researchers at Russia's Skolkovo Institute of Science and Technology and Belgium's Katholieke Universiteit Leuven have reconstructed three-dimensional (3D) micro-computed tomography (micro-CT) images of fibrous materials using machine learning.
To compensate for shaded, missing, or damaged areas in images, the researchers turned to inpainting, which is used to restore damaged paintings while preserving their overall integrity.
To fill the gap in available inpainting tools for 3D micro-CT images, the researchers used 3D encoder-decoder generative adversarial networks (GANs).
They tested three GAN architectures on micro-CT scans of short glass fiber composite and selected the architecture that achieved high inpainting quality and performance and relatively low GPU memory usage.
Researcher Radmir Karamov said, "With the inpainting algorithm, we can eliminate all defects in micro-CT scans for a more precise simulation of material behavior and analyze how material performance will increase if all inside pores and voids are removed during the manufacturing process."
From Skolkovo Institute of Science and Technology (Russia)
View Full Article
Abstracts Copyright © 2021 SmithBucklin, Washington, DC, USA