Coupling EDS hypermaps and X-ray microtomography for advanced 3D microstructure characterization of cement paste: A step forward in multiscale modeling

Investigating the microstructure of ordinary Portland cement (OPC) paste using X-ray micro-computed tomography (μ-CT) requires optimized acquisition and precise image segmentation to reliably differentiate phases. μ-CT image segmentation is challenged by the heterogeneous microstructure and limited...

Full description

Saved in:
Bibliographic Details
Published in:Cement and concrete research Vol. 199; p. 108047
Main Authors: Krameche, Mohammed, Wilson, William, Tagnit-Hamou, Arezki
Format: Journal Article
Language:English
Published: Elsevier Ltd 01.01.2026
Subjects:
Cement paste
Data analysis
Image analysis
micro-CT
Microcube
Microstructure
Multiscale modeling
SEM-EDS
Image analysis
Data analysis
SEM-EDS
Microcube
Cement paste
Microstructure
micro-CT
Multiscale modeling
ISSN:0008-8846
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Investigating the microstructure of ordinary Portland cement (OPC) paste using X-ray micro-computed tomography (μ-CT) requires optimized acquisition and precise image segmentation to reliably differentiate phases. μ-CT image segmentation is challenged by the heterogeneous microstructure and limited contrast between microstructure phases in the X-ray linear attenuation coefficient. Conventional gray scale value thresholding often misclassifies phases, while previous machine learning (ML) approaches have relied on manually labeled training leading to subjectivity and replicability issues. This study proposes an innovative μ-CT image segmentation method for OPC paste, leveraging chemical information from quantitative energy dispersive spectroscopy (QEDS) mapping. The method workflow involves five steps: (1) μ-CT imaging to capture the 3D microstructure, (2) scanning electron microscopy backscattered electron (SEM-BSE) imaging and QEDS mapping to generate 2D phase maps, (3) image registration to align QEDS phase maps with μ-CT images, (4) phase separability optimization using denoising and sharpening of the μ-CT images, and (5) ML-based segmentation using the random forest approach and training labels from QEDS-derived phase maps. The proposed method effectively differentiates portlandite from hydrates matrix, as well as ferrite from clinker phases. This paper further details two main applications of the method: (1) quantification of phase assemblage, hydration degree, and particle size distribution (PSD) of residual anhydrous phases, with results compared to thermodynamic modeling, and (2) the first-ever comprehensive quantitative microstructural characterization of a grid of 32 microcube samples, revealing spatial heterogeneity in phase fractions, porosity, particle counts, and microcube volumes. These results provide critical input for calibrating and validating micromechanical upscaling models. •A novel μ-CT image segmentation method integrating QEDS mapping for cement paste 3D microstructural analysis and modeling.•Enhanced phase separation using optimized BM4D image filtering and similarity calculations.•First-ever 3D microstructural analysis of 32 cement paste microcubes.•Quantified spatial heterogeneity in phase fractions, porosity, and particle counts across microcubes.•Microstructure segmented into 5 groups: ferrite, silicates & aluminate, portlandite & calcite, hydrate matrix, and pores.
ISSN:0008-8846
DOI:10.1016/j.cemconres.2025.108047