High energy geothermal simulations in faulted reservoirs with code ComPASS

• Team: this project started in 2014 in collaboration between


• BRGM: Simon Lopez and Farid Smai
• Joint INRIA-LJAD team Coffee: Feng Xing, Laurence Beaude, Konstantin Brenner and Roland Masson


• The ANR project CHARMS started in october 2016 for 4 years strenghens the project and enlarges it to the additional partners Storengy (Delphine Patriarche, Vincent Seignole, Jean-Frédéric Thébault), La Maison de la Simulation (Michel Kern, Julien Bigot, Pascal Tremblin, Thomas Dufaut) and the Laboratoire Jacques Louis Lions (Cindy Guichard, Robert Eymard).



• Objectives: our objectives are to develop efficient parallel numerical methods to simulate gas liquid compositional thermal flow in high energy geothermal reservoirs. The spatial discretization is adapted to unstructured polyhedral meshes and takes into account discrete fracture or fault networks represented as 2D surfaces connected to the surrounding 3D heterogeneous anisotropic porous medium (the matrix). The difficulties result from the highly contrasted spatial and temporal scales and petrophysical properties between the 3D matrix and the fault network. The second difficulty is to account for the strong couplings and high nonlinearities induced by the large range of pressure and temperature and by the phase appearance and disapearance.
  
  These numerical methods and physical models are implemented in code ComPASS (Computing Architecture to Speed up Simulation).


• Brief history of the code ComPASS


• First version developed during the Cemracs summer school in 2012 by C. Guichard, C. Yang, R. Masson, P. Havé. Compass: a tool for distributed parallel finite volume discretizations on general unstructured polyhedral meshes
• The second version implementing an immiscible two phase Darcy flow model and developed by C. Guichard, R. Eymard and R. Masson has been deposited at APP in 2014. High Performance Computing linear algorithms for two-phase flow in porous media
• The third version implementing a non-isothermal compositional multiphase Darcy flow model taking into account fault or fracture networks was developed during the inter Carnot postdoctoral fellowship of F. Xing in collaboration between BRGM and LJAD-INRIA. Parallel Numerical Modeling of hybrid-dimensional compositional non-isothermal Darcy flows in fractured porous media
• A new version of the code is currently developed in the ANR project CHARMS


• Current version of ComPASS: it has the following features


• Computer science


• Version management of the code using Inria GitLab
• SPMD Paradigm
• Fortran 2003 + C/C++ + MPI
• Parallelism based on a partitioning of the mesh with Metis
• One layer of ghost cells
• Connected to the solver libraries Petsc + Hypre + Trilinos
• Visualization ouputs using parallel vtk format
• Checkpointing using HDF5
• High level python API (in development)


• Numerical methods


• 3D conforming polyhedral meshes
• Parallel Vertex Approximate Gradient (VAG) scheme accounting for the flow in the 3D matrix and the 2D fault network. See Parallel Vertex Approximate Gradient discretization of hybrid dimensional Darcy flow and transport in discrete fracture networks
• Fully implicit Coat's type formulation of multiphase compositional thermal models. See Vertex Centred Discretization of Multiphase Compositional Darcy Flows on General Meshes
• Newton Raphson algorithm with phase appearance and disappearance
• Preelimination of the secondary unknowns before assembly of the Jacobian matrix
• GMRES or BiCGStab linear solver with CPR-AMG preconditioner


• Physics


• Multiphase compositional thermal Darcy flow model
• 2D discrete fracture or fault network coupled with the surrounding 3D matrix
• Multi-branch well model (in development)

• Gas liquid thermal model with a single component H2O
• Reservoir of size 3kmx3kmx3km initially at hydrostatic pressure and 300 K (liquid phase)
• Temperature fixed to 600 K at the bottom boundary (liquid phase) and to 300 K at the top boundary
• Homogeneous porous medium
• 3D Cartesian mesh of size 100x100x100
• 15000 days of simulation



• Temperature and gas saturation function of time

• Gas liquid thermal model with a single component H2O
• Reservoir of size 3kmx3kmx3km initially at hydrostatic pressure and 300 K (liquid phase)
• Temperature fixed to 600 K at the bottom boundary (liquid phase) and to 300 K at the top boundary
• Homogeneous matrix of permeability 1 mDarcy with four fractures of width 1 m and permeability 1 Darcy
• 3D Cartesian mesh of size 240x240x240
• 220000 days of simulation



• Temperature and gas saturation in the matrix and in the fractures function of time

• Gas liquid thermal model with a single component H2O
• Reservoir of size 6kmx3kmx250m with roughly 1000 fractures of width 1 m
• Permeability ratio between the fracture network and the matrix of 1000
• 3D Mesh with 1.000.000 prismatic cells
• Reservoir initially at 450 K and 1 MPa (liquid phase)
• Left output boundary at 450 K and 1 MPa (liquid phase) and right input boundary at 550K and 2 Mpa (gas phase)



• Gas saturation and Temperature in the matrix and fracture network function of time





• Speed up

• Reservoir of size 6kmx3kmx250m with roughly 1000 fractures of width 1 m
• Ratio of permeability between the fractures and the matrix of 10000
• 3D Mesh with 10.000.0000 prismatic cells
• Flow and tracer simulation
• One injection well and two producers (see below)
• 8 years of simulation



• Prismatic mesh of the matrix and triangular mesh of the fracture network




• Tracer in the matrix and in the fracture network at the end of the simulation




• Pressure field and tracer production at both wells




• Speed up

• Two phases water and oil and two components H2O and HC
• Dissolution of the HC component in the water phase
• Reservoir of size 100 mx 100 mx 100m with fractures of width 1 cm
• Permeability ratio between the fracture network and the matrix of 10000
• 3D Mesh with 6.000.000 tetrahedra
• Reservoir initially saturated with pure water and injection of oil from the bottom boundary with a gravity dominant flow
• 10000 days of simulation



• Oil saturation and molar fraction of HC in the water phase both in the matrix and the fracture network function of time





• Speed up