Multiscale model reduction technique for fluid flows with heterogeneous porous inclusions

TL;DR

This paper introduces a multiscale model reduction technique based on GMsFEM for simulating viscous fluid flow in heterogeneous porous media, reducing computational complexity while maintaining accuracy across various flow regimes.

Contribution

The paper develops a multiscale basis function construction method for fluid flow in porous media using GMsFEM, incorporating oversampling strategies to improve efficiency and accuracy.

Findings

The multiscale method shows good convergence as the number of basis functions increases.

The approach effectively handles different flow regimes characterized by Reynolds, Forchheimer, and Darcy numbers.

Numerical results demonstrate the method's robustness and computational efficiency.

Abstract

Numerical treatment of the problem of two-dimensional viscous fluid flow in and around circular porous inclusions is considered. The mathematical model is described by Navier-Stokes equation in the free flow domain $\Omega_f$ and nonlinear convective Darcy-Brinkman-Forchheimer equations in porous subdomains $\Omega_p$. It is well-known that numerical solutions of the problems in such heterogeneous domains require a very fine computational mesh that resolve inclusions on the grid level. The size alteration of the relevant system requires model reduction techniques. Here, we present a multiscale model reduction technique based on the Generalized Multiscale Finite Element Method (GMsFEM). We discuss construction of the multiscale basis functions for the velocity fields based on the solution of the local problems with and without oversampling strategy. Three test cases are considered for a given choice of the three key model parameters, namely, the Reynolds number ($Re$), the Forchheimer coefficient ($C$) and the Darcy number ($Da$). For the test runs, the Reynolds number values are taken to be $Re = 1, 10, 100$ while the Forchheimer coefficient and Darcy number are chosen as $C= 1, 10$ and $Da = 10^{-5}, 10^{-4}, 10^{-3}$, respectively. We numerically study the convergence of the method as we increase the number of multiscale basis functions in each domain, and observe good performance of the multiscale method.