The porous structure of hydrated cement has a significant effect on its governing properties, such as strength, density, ion-diffusion, acoustic performance etc. However, direct observation of the 3D pore structure is limited by high stochasticity and multiscale features. We constructed a statistically equivalent 3D micropore structure using a digital concrete model that used a probabilistic map derived from 2D cross-sections and parameter sets calibrated via a deep learning neural network. Based on the replica of cement pore structure, image analyses and virtual mercury intrusion porosimetry (MIP) tests were compared with experimental outcomes. In particular, the pore size distribution curves matched well for both MIP and image analysis approaches. However, although the fractal dimension derived from each 2D image linearly correlated with the corresponding porosity, it tended to underestimate the complexity of the pore structure. Correspondingly, the 3D fractal dimension showed a periodic pattern to voxel size in log-scale and was slightly higher than the MIP results. The tortuosity factor also had a linear correlation to voxel size, resulting in an overestimation of the pore structure's tortuosity. We demonstrated that the digital concrete model could serve as an alternative approach to studying the microstructural pore analysis of cementitious material through such a comparison. This study's outcome helps to understand the structure–property link in cement with the potential to transform traditional analytical processes of cementitious materials.
- BSE image
- Image analysis
- Microstructure characterisation
- Microstructure reconstruction
- Pore structure