Evolution of the microporous structure in cement hydration: A deep learning-based image translation method

Abstract

Investigating the evolution of the microporous structure is essential for gaining insights into the hydration process of the cementitious materials. However, conventional methods for describing cement hydration rely on complex formulas and idealized assumptions, limiting their accuracy and efficiency in representing hydration kinetics. Here, we present a novel deep learning-based image translation method for modelling the in-situ evolution of the microporous structure during cement hydration. The translated images are of high spatial resolution, which can identify the pores smaller than around 100 nm. The performance of our model was evaluated by comparing various characteristics of the translated microporous structures with the experimental data, including porosity, pore size distribution, pore shape and spatial properties. The results of this comparison indicate our approach exhibits an excellent performance in accurately modelling the evolution of the microporous structures. Furthermore, due to the data-driven nature, our model presented a high efficiency, which is capable of completing the image translation within a few minutes. The findings in this study offer a novel approach to trace or predict the structural development of cement during its hydration, which is crucial for material design, durability assessment, and performance prediction of cementitious composites.