Talk at the 2015 Cargese summer school on
Flow and Transport in Porous and Fractured media: Development, Protection, Management and Sequestration of Subsurface Fluids
4. Upscaling solute transport processes (heterogeneous media)
▪ Purposes
▪ Effect of heterogeneity on inert and reactive solute processes
▪ Enhancement of dispersion and mixing induced by permeability heterogeneity
▪ Determine effective, upscale laws: Multiple scales in the same simulations
▪ Conceptual model (Assumptions)
▪ Stochasticly well-defined heterogeneity fields with evolving levels of complexity
▪ Simplification of boundary and initial conditions to focus on the processes
▪ Stochastic simulations
▪ Mathematical model
▪ Advection-diffusion-dispersion equations
7. Transport model
v
vv
dvD
cDuc
t
c
ji
TLijTij
0
Periodic boundary
conditions
extendedinjection
Reflecting
boundary
conditions
c=0
Adsorbing
boundary
conditions
point-source
Initial conditions
c(x,t=0)=0
8. Numerical methods
▪ Multi-scale stochastic simulations
▪ requires parallel computation
▪ Flow equation
▪ finite volume discretization
▪ algebraic multigrid linear solver
▪ Transport equation
▪ Lagrangian method: random walks
▪ Numerical strategy
▪ Macrodispersion: stochastic simulations with limited number of particles
▪ Mixing: few simulations with large number of particles
Beaudoin, A., J. R. de Dreuzy, and J. Erhel (2007), An efficient parallel tracker for advection-diffusion simulations in
heterogeneous porous media, paper presented at Europar, Rennes, France, 28-31 August 2007, Lecture Notes in
Computer Science 4641 705-714 Springer-Verlag, Berlin, Heidelberg
9. Some examples of software for porous and fractured media
▪ Classical hydrogeological models
▪ MODFLOW
▪ FEEFLOW
▪ HYDROGEOSPHERE
▪ Specialized modelling plateforms
▪ Tough, Berkeley, reactive transport
▪ DUMUX, DUNE, Stutgart, Multiphase flow, Multiphysics
▪ GEOSYS, UFZ, THMC
▪ PROOST, Barcelona
▪ H20lab, Rennes, heterogeneity (porous,fracture) and transport
▪ Multiphysics models
▪ COMSOL
▪ ABACUS
▪ Fluid mechanics models
▪ Open foam
12. Permeability variance 0.25 < y
2 < 9
Domain size Nx = 16384, Ny Nz = 128
500 Monte Carlo simulations
10 000 particles
Extensive parameter study :
Cluster = 64 nodes of 2 processors Intel Quad Core
x5472. Each processor is composed of 4 cores
(Harpertown 3GHz) and 4GB of memory per core.
permeability generation = 20 s
time for flow = 213s
time for transport = 1605s
Example of CPU times :
Performances
13. Temporal evolution of the dimensionless longitudinal effective
dispersivity L(t) for various values of y²
Validation against analytical predictions Y
2<1
14. Predictions
2D and 3D longitudinal macro dispersivities
LA as function of y²
3D transverse macro dispersivity TA for
various values of y²
A. Beaudoin and J.R. de Dreuzy, Numerical assessment of 3D macro dispersion in heterogeneous porous media, Water Resources Research, Vol. 43, 2013
15. A. Beaudoin and J.R. de Dreuzy, Numerical assessment of 3D macro dispersion in heterogeneous porous media, Water Resources Research, Vol. 43, 2013
Presentation of results
Low heterogeneity
Y
2=1
High heterogeneity
Y
2=6.25
17. Simulation and analysis of concentration distributions
Probability distribution of concentrations
macrodispersion model
Simulations at different times
18. Definition and validation of a new effective mixing model
Lamella representation
Villermaux, Cargèse summer school 2010
19. Quantification of fluid deformation processes
Map of fluid deformation Distribution of elongations
Le Borgne et al., JFM 2015
20. Definition and validation of a new effective mixing model
t1
t2
𝒑 𝒄, 𝒕
t2 t3
Fluid deformation Concentrations
𝒑(𝒄|𝝆)
Lamella representation Concentration PDF
Le Borgne et al. PRL 2013
macrodispersion model
23. Scope: develop simulation projects in interaction with lecturers
(on a voluntary basis)
▪ Projects linked to practical courses
▪ Simulation of saltwater/freshwater interface
▪ Simulation of heat transport and potential fiber optic signal
▪ Direct modelling of geophysical signals (Resistivity, Spontaneaous Potential…)
▪ Projects linked to lectures
▪ Transport in heterogeneous media
▪ Reactive transport, colloid transport
▪ Multiphase or Non-Newtonian flows..
▪ Hydro-mechanics
▪ Projects linked to students PhD topics
24. Tool: COMSOL multiphysics
▪ Advantages
▪ Easy to learn in a week (friendly interface)
▪ Handles a large spectrum of coupled flow and transport processes
▪ Disadvantages
▪ Commercial licence
▪ Limited in terms of simulation size
▪ Free alternatives
▪ OpenFoam
▪ FreeFem
▪ …
In-silico experiments
Numerical experiments
Global objective
Build up understanding: Model is not understanding, it is leading to it
Establish Predictions
Interest of numerical experiments
Some are easy to set up
Less expensive than data, experiments
Difficulties/Limitations
Not reality, just a model, an approximation: complementary to experiments,
Poorly constrained: everything is possible but is it good, understandable…
Using numerics to get access to processes, understanding
Raise some general issues: to which point should I develop or understand numerical methods to run them
Same for experimental or field work
Not necessary to understand the computer to run a simulation, to understand the car to drive it….
It is just when it breaks that I need somehow what to do, when it does not fulfill my needs
Underline some basic logic when running simulations
The software is rarely wrong, I am rarely wrong, The interface can be
A global prospective on numerical development
Algorithms, Benchmark, Validation, Coupling
Deploymeent, Interface
Documentation, Community
Models
Develop at least as possible
Understanding and assembling (limits and pitfalls to assemble)
Differences between
Assumptions
Model: ensemble of equations, boundary and initial conditions that fully determine a problem that can be mathematically treated
Numerical Method
Software (Comsol)
Simulation
Purposes
Influence at all levels:
Mathematical model
Implementation Choice: software
Numerical scheme
Parameter choice:
Tradeoffs: the best numerical method, the worst interface, Worth? Necessary?
Choice of software
Free solutions can be very expensive: time
Approximations
Are approximations, numerical errors small enough.
Decompose simulation and interpretation
Segmentation of the process
Organize simulations
Less simulations, better analyzed
Reference cases
Model is not understanding
Understanding comes from the deep understanding of analytical solutions, high understanding of the equations and their mathematical properties
Understanding comes from the simulations
Question:
Use numerical models but not more
Models
Develop at least as possible
Understanding and assembling (limits and pitfalls to assemble)
Differences between
Assumptions
Model: ensemble of equations, boundary and initial conditions that fully determine a problem that can be mathematically treated
Numerical Method
Software (Comsol)
Simulation
Purposes
Influence at all levels:
Mathematical model
Implementation Choice: software
Numerical scheme
Parameter choice:
Tradeoffs: the best numerical method, the worst interface, Worth? Necessary?
Choice of software
Free solutions can be very expensive: time
Approximations
Are approximations, numerical errors small enough.
Decompose simulation and interpretation
Segmentation of the process
Organize simulations
Less simulations, better analyzed
Reference cases
Model is not understanding
Understanding comes from the deep understanding of analytical solutions, high understanding of the equations and their mathematical properties
Understanding comes from the simulations