A Non Uniform Algebraic Dynamic Multilevel for Unstructured Meshes for the numerical Simulation of Fluid Flow in Heterogeneous Porous Media

Autores

  • João Paulo Rodrigues de Andrade
  • José Cícero Araujo dos Santos
  • Artur Castiel Reis de Souza
  • Paulo Roberto Maciel Lyra
  • Darlan Karlo Elisiário de Carvalho

Palavras-chave:

Reservoir Simulation, Finite Volume Method,, Multiscale-Multilevel Method, Unstructured Mesh

Resumo

In the present work, we introduce, for the first time in the literature, the Non-Uniform Algebraic Dynamic Multilevel with Unstructured Meshes (NU-ADM-UM) method, which extends the original Non-Uniform Algebraic Dynamic Multilevel method to the simulation of single-phase and two-phase flows on unstructured meshes. In this new method, the employed multiscale transfer operators are the Algebraic Multiscale Solver for General Unstructured Grids (AMS-U) and the Multiscale Restriction Smoothed Basis (MsRSB). To construct the coarse meshes (primal and dual), we adopted the procedure proposed by AMS-U method. The dual mesh obtained via this approach was then used to define the support and boundary regions required by the MsRSB method. The mathematical model results in a system solved using an IMPES (Implicit Pressure Explicit Saturation) strategy, in which pressure is solved implicitly and saturation explicitly. For the pressure field, we adopted a finite volume formulation of the MPFA-D (Multi-Point Flux Approximation with a Diamond scheme) type, with weights computed using the Global Least Squares (GLS) method. This formulation can handle non-K-orthogonal meshes and yields highly accurate solutions. The saturation field is solved using a first-order upwind method. The results show that the mesh refinement parameters of NU-ADM-UM successfully preserved, at the fine scale, the regions that would otherwise produce spurious pressure oscillations. Moreover, the method was able to follow the saturation front in two-phase flow problems, delivering results very close to those obtained by solving the problem on the full fine scale, but with a reduced number of active control volumes throughout the simulation and, consequently, a reduction in total simulation time.

Downloads

Publicado

2026-03-02