TY - JOUR
T1 - Multiscale computation of pore-scale fluid dynamics
T2 - Single-phase flow
AU - Mehmani, Yashar
AU - Tchelepi, Hamdi A.
N1 - Funding Information:
Funding for this work was provided by the Stanford University Petroleum Research Institute (SUPRI-B affiliates). We also acknowledge the Office of Basic Energy Sciences Energy Frontier Research Center under Contract number DE-AC02-05CH11231 for financial support. We are grateful to the Stanford Research Computing Facility (SRCF) for access to computational resources. Mehrdad Yousefzadeh is thanked for discussions on the SIMPLE-C Navier–Stokes solver.
Publisher Copyright:
© 2018 Elsevier Inc.
PY - 2018/12/15
Y1 - 2018/12/15
N2 - Direct numerical simulation (DNS) of interstitial fluid dynamics in porous media is hindered by the sheer size and complexity of the discretized equations. While reduced-complexity methods such as pore-network models (PNM) can occasionally yield satisfactory solutions at a much lower cost, they can neither estimate nor control their error. We focus on the single-phase Navier–Stokes equations and develop a computationally efficient multiscale method that produces a sequence of increasingly accurate approximations to DNS. The pore-level multiscale method (PLMM) decomposes the pore space into several subdomains and constructs a set of local basis functions on them. The bases are coupled with a global interface problem to obtain an initial approximation to DNS. The approximation is excellent because subdomains coincide with physical pores in the void space and physics-informed boundary conditions are used to construct the bases. Errors in the initial approximation can be arbitrarily reduced with an iterative strategy presented. The method is parallelizable, memory efficient, and allows for different physics, models, and meshes to be incorporated within each subdomain.
AB - Direct numerical simulation (DNS) of interstitial fluid dynamics in porous media is hindered by the sheer size and complexity of the discretized equations. While reduced-complexity methods such as pore-network models (PNM) can occasionally yield satisfactory solutions at a much lower cost, they can neither estimate nor control their error. We focus on the single-phase Navier–Stokes equations and develop a computationally efficient multiscale method that produces a sequence of increasingly accurate approximations to DNS. The pore-level multiscale method (PLMM) decomposes the pore space into several subdomains and constructs a set of local basis functions on them. The bases are coupled with a global interface problem to obtain an initial approximation to DNS. The approximation is excellent because subdomains coincide with physical pores in the void space and physics-informed boundary conditions are used to construct the bases. Errors in the initial approximation can be arbitrarily reduced with an iterative strategy presented. The method is parallelizable, memory efficient, and allows for different physics, models, and meshes to be incorporated within each subdomain.
UR - http://www.scopus.com/inward/record.url?scp=85056218335&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85056218335&partnerID=8YFLogxK
U2 - 10.1016/j.jcp.2018.08.045
DO - 10.1016/j.jcp.2018.08.045
M3 - Article
AN - SCOPUS:85056218335
SN - 0021-9991
VL - 375
SP - 1469
EP - 1487
JO - Journal of Computational Physics
JF - Journal of Computational Physics
ER -