Predicting permeability via statistical learning on higher-order microstructural information

Dataset and code used in M. Röding, et al, "Predicting permeability via statistical learning on higher-order microstructural information", published in Scientific Reports, 2020. In this work, we study permeability prediction in a large data set of 30,000 virtual, porous microstructures of different types, including both granular and continuous solid phases. The permeabilities are computed using the lattice Boltzmann method. The pore space geometries are charaterized using the following descriptors: one-point correlation functions (porosity, specific surface), two-point surface-surface, surface-void, and void-void correlation functions, and geodesic tortuosity. Linear regression with linear and quadratic terms as well articifical neural networks are used for prediction. As a reference, Kozeny-Carman regression with only lowest-order descriptors (porosity and specific surface) is also studied. Herein, the descriptors, the permeabilities, and the Matlab and Python/Tensorflow code used for prediction are supplied.