Generalized network modeling: Network extraction as a coarse-scale discretization of the void space of porous media

TLDR

The workflow aims to reduce uncertainties in conventional network modeling by better capturing complex pore geometries in natural porous media. It extracts a generalized network by medial‑axis transformation of 3‑D images, discretizing void space into pores, half‑throat connections, and corner segments, and computes conductivities from single‑phase flow simulations to match the image permeability. Validation on synthetic angular pores shows accurate corner‑angle reproduction, while application to micro‑CT rock images preserves permeability, formation factor, and image statistics.

Abstract

A generalized network extraction workflow is developed for parameterizing three-dimensional (3D) images of porous media. The aim of this workflow is to reduce the uncertainties in conventional network modeling predictions introduced due to the oversimplification of complex pore geometries encountered in natural porous media. The generalized network serves as a coarse discretization of the surface generated from a medial-axis transformation of the 3D image. This discretization divides the void space into individual pores and then subdivides each pore into sub-elements called half-throat connections. Each half-throat connection is further segmented into corners by analyzing the medial axis curves of its axial plane. The parameters approximating each corner---corner angle, volume, and conductivity---are extracted at different discretization levels, corresponding to different wetting layer thickness and local capillary pressures during multiphase flow simulations. Conductivities are calculated using direct single-phase flow simulation so that the network can reproduce the single-phase flow permeability of the underlying image exactly. We first validate the algorithm by using it to discretize synthetic angular pore geometries and show that the network model reproduces the corner angles accurately. We then extract network models from micro-CT images of porous rocks and show that the network extraction preserves macroscopic properties, the permeability and formation factor, and the statistics of the micro-CT images.