Prediction of both diffusive and hydraulic conductance in the pore network model extracted from 3D images using deep learning

Abstract Pore network modeling (PNM), with network extracted from 3D images, is widely used to study mass transport in porous media. Its accuracy heavily depends on the precise determination of hydraulic and diffusive conductance values of local pore pairs. Deep learning (DL), leveraging true pore geometry, offers a promising approach for conductance estimation. However, due to the arbitrary complexity of 3D pore pair geometries, most studies have relied exclusively on 2D cross sectional images for hydraulic conductance prediction. Therefore, in this work, a convolutional neural network was trained to estimate diffusive and hydraulic conductance values, with reference ground truth values obtained by applying the lattice Boltzmann method (LBM) on 3D pore pair geometries. The main contributions include: (1) 3D image preprocessing to align pore pair geometries with LBM fluid flow, (2) high prediction accuracy with a 3D residual neural network-18 model ( R 2 = 0.95 for diffusive and R 2 = 0.94 for hydraulic conductance), and (3) insights on the importance of full 3D geometry for accurate hydraulic conductance evaluation. To evaluate the generality of our method, we calculated the permeability and diffusivity of 3D porous media across a wide range of porosities using PNM, with DL predicted local conductance values. Results show accurate diffusivity predictions, but permeability was underestimated due to over-segmentation of pore spaces, which reduced flow rate estimates.