Effects of Domain Size on Transverse Permeability through Random Arrays of Cylinders

Abstract

Researchers using Darcy's law to model flow in porous media must satisfy the requirement for sufficient scale separation between the pore scale and the model scale. This requirement is analogous to that for any continuum model, where application is restricted to scales larger than the underlying discrete structure. In the case of Darcy's law when the model scale becomes too small, the measurement of the permeability - the material property required to close the relationship - becomes polluted by the boundary conditions, either physical or numerical. The requirements for adequate scale separation to obtain permeability measurements (also known as satisfying the conditions for a representative elementary volume, or REV, for permeability) have not been previously reported. Likewise, the behavior of Darcy models when applied at sub-REV length scales has not been reported. Here, the results of Stokes simulations of transverse flow in 90,000 sequential random packings of monodisperse cylinders at a variety of liquid fractions and averaging-volume sizes show that approximately 200 cylinders must be present in an averaging volume before the effects of periodic boundary conditions on the Stokes simulations (the conventional choice for permeability measurements using Stokes flow) are no longer evident in the measured permeability. Direct comparisons between flow predictions from a two-dimensional, tensor-based Darcy model and a Stokes model for additional 10,000 domains show that the Darcy model is an unbiased predictor of the flow distribution in the system, even when the permeability is expected to contain boundary-condition artifacts. Though unbiased, the Darcy models do show considerable reduction in accuracy as the model scale shrinks toward the pore scale, with significant declines observed after the side length of a square averaging volume reaches 10 times the cylinder diameter. Finally, a novel approach for visualizing flows using the linear properties of the Stokes equations shows how the periodic boundary conditions affect the flow, and motivates the development of a generalized approach for obtaining permeability that does not require periodic boundary conditions. Modest improvements in the Darcy model relative to the actual Stokes flow result when the new approach is used to obtain permeability at small averaging volumes.