Generating digital twins of mesoporous silica by graph-based stochastic microstructure modeling

Silica monoliths are hierarchically structured, versatile materials that are widely used in analytical separation science, e.g., liquid chromatography. Their functionality strongly depends on the 3D morphology of their macropore and mesopore spaces. In the present paper, we consider three differently manufactured silica monolith samples, where the process conditions of their hydrothermal treatment (affecting, e.g., mesopore size) have been varied, and present a parametric stochastic 3D microstructure model that is able to generate digital twins of the resulting mesopore spaces. The model, which is based on random point processes and relative neighborhood graphs, theoretically guarantees the complete connectivity of both, the silica phase and the mesopores. The parametric model is fitted to electron tomographic image data. For this purpose, we optimize a cost function that is based on empirically derived relationships between model parameters and volume fraction, mean geodesic tortuosity and constrictivity. Validation is performed regarding further microstructure characteristics, which are not used for model fitting, and regarding effective diffusivity, which is numerically simulated by a particle-tracking algorithm based on random walks.