New stochastic pore-scale simulation and machine learning approach to predicting permeability and tortuosity of heterogeneous porous media

Abstract

A new 3D stochastic pore-scale simulation approach was introduced in this study to investigate how stochastic pore connectivity impacts the permeability and hydraulic tortuosity of heterogeneous porous media. Multiple three-dimensional pore microstructures with the same porosity, pore size distribution, and number of pores were created to examine the role that pore connectivity plays in permeability and hydraulic tortuosity of rocks. According to the findings of this study, stochastic pore connectivity has a sizeable influence on permeability but a comparatively insignificant influence on hydraulic tortuosity. In addition, there is a negative correlation between the amount of heterogeneity and the predictability of permeability based on hydraulic tortuosity. The study used four carbonate and five siliciclastic rock cores to validate the findings, and the predicted permeability was generally closer to the measured permeability than five popular empirical model equations. In addition, machine learning was utilized to optimize the workflow, which resulted in a reduction in the necessary number of pore-scale simulations by a factor of 157 and the reproduction of pore-scale permeability estimates was with a mean absolute percentage error of 10%. The conventional use of SEM and micro-CT technologies in generating pore microstructures was augmented through the addition of MICP data to capture high-resolution information in rocks at a representative scale. Pore microstructures generated through the approach were validated via a comparable permeability of the pore microstructure with the measured permeability of the rock sample modelled. The findings of this study has a broad range of geoscience applications including, petroleum exploration and production, carbon storage, environmental protection, and groundwater exploration.