Lattice Boltzmann methods in porous media simulations: From laminar to turbulent flow

The lattice Boltzmann method has become a popular tool for determining the correlations for drag force and permeability in porous media for a wide range of Reynolds numbers and solid volume fractions. In order to achieve accurate and relevant results, it is important not only to implement very efficient code, but also to choose the most appropriate simulation setup. Moreover, it is essential to accurately evaluate the boundary conditions and collision models that are effective from the Stokes regime to the inertial and turbulent flow regimes. In this paper, we compare various no-slip boundary schemes and collision operators to assess their efficiency and accuracy. Instead of assuming a constant volume force driving the flow, a periodic pressure drop boundary condition is employed to mimic the pressure-driven flow through the simple sphere pack in a periodic domain. We first consider the convergence rates of various boundary conditions with different collision operators in the Stokes flow regime. Additionally, we choose different boundary conditions that are representatives of first- to third-order schemes at curved boundaries in order to evaluate their convergence rates numerically for both inertial and turbulent flow. We find that the multi-reflection boundary condition yields second order for inertial flow while it converges with third order in the Stokes regime. Taking into account the both computational cost and accuracy requirements, we choose the central linear interpolation bounce-back scheme in combination with the two-relaxation-time collision model. This combination is characterized by providing viscosity independent results and second order spatial convergence. This method is applied to perform simulations of touching spheres arranged in a simple cubic array. Full- and reduced-stencil lattice models, i.e., the D-3(Q(27) and D-3(Q(19), respectively, are compared and the drag force and friction factor results are presented for Reynolds numbers in the range of 0.001 to 2,477. The drag forces computed using these two different lattice models have a relative difference below 3% for the highest Reynolds number considered in this study. Using the evaluation results, we demonstrate the flexibility of the models and software in two large scale computations, first a flow through an unstructured packing of spherical particles, and second for the turbulent flow over a permeable bed. (C) 2016 Elsevier Ltd. All rights reserved.