Phase segmentation in X-ray CT images of concrete with implications for mesoscale modeling

X-ray computed tomography (XRCT) is a valuable tool for characterizing the microstructure of concrete and for developing 3D mesoscale numerical models directly from experimental data. However, the results of imaging and subsequent modeling are reliable only if individual phases can be identified and segmented accurately. Siliceous aggregates and cement paste are difficult to separate in XRCT images because of their similar X-ray attenuation coefficients. This work examines the quality of aggregate phase segmentation in XRCT images using (1) a standard deviation thresholding approach and (2) a random forest classification. Both approaches were validated with ground truth data for concrete samples with different aggregate volume fractions. Our findings show that either approach may successfully be used to segment aggregate phases if appropriate post processing is performed. However, our results emphasize the critical need to preserve both aggregate size and shape during post-processing as illustrated through mesoscale modeling.