Fourier-Based Non-Rigid Slice-to-Volume Registration of Segmented Petrographic LM and CT Scans of Concrete Specimens

Cyclic freezing and thawing of concrete specimens are one of the main causes of cracking in concrete. Current procedures for assessing frost resistance of concrete in Germany rely mainly on the CIF and CDF tests that utilise qualitative estimation of cracks. Although these standard tests provide a general overview of the condition of concrete damage through the estimation of water saturation through capillary suction, mass of surface delamination, qualitative open surface damage, and relative dynamic modulus of elasticity, they do not take quantitative analysis of cracks directly into account. To facilitate this quantitative approach, cracks are studied in concrete samples exposed to specific standard cycles of freezing and thawing, then scanned using micro computed tomographic (µCT) imaging, and consequently cut for petrographic thin section analysis. The thin sections are scanned using light microscopy (LM). Deep learning frameworks were deployed to train semantic segmentation models to identify cracks, air pores, aggregate, and cement matrix. Both scanned modalities were co-registered using three experimental variations of varying processing complexity that rely on matching of Fourier-based shape descriptors and underlying features thereof to verify and validate the quality of segmentation inferred for various phases of concrete.