Numerical Methods for Flow in Fractured Porous Media

Formaggia, Luca and Scotti, Anna and Fumagalli, Alessio

arXiv:2003.14279v1 [physics.comp-ph] 31 Mar 2020

Definition iii) Fractures usually form networks, often highly

connected. An example of fracture network is de-
• porous medium: A material containing voids picted in Fig 4.
(pores), whose size is small compared to the size With the previous definitions one can consider
of the sample. The pores are typically filled with as fractures, depending on the scale of interest, dif-
a fluid. A porous medium is characterized by ferent objects such as: tectonic faults at the scale
macroscopic properties which are obtained with of sedimentary basins, cracks in glaciers, as well as
an averaging procedure. fractures in concrete or in rocks.
• fracture: a break in a material with a small thick- When they are empty or filled with highly perme-
ness compared to its global extension. able materials, fractures may provide a preferential
• fracture network: a network composed by several path to fluid flow, but in some cases the deposits in-
intersecting fractures. side fractures can become nearly impermeable. We
refer to the latter situation as blocking fracture.
The presence of fractures greatly alters the macro-
scopic properties of the material in a complex way,
1 Introduction in particular its mechanical and flow characteristics.
We will be here concerned with the second aspect,
Fractures are ubiquitous in porous media. Here, with and specifically on the mathematical modelling and
the term fracture we denote a void in the porous computing techniques that may be adopted in the
material that has the following characteristics. presence of fractured porous media. For a more gen-
i) One of its dimensions, the aperture, is orders of eral treatment of fractures in porous media the reader
magnitude smaller than the other dimensions and the may refer to the Book of P.M. Adler et al. [1].
size of the domain of interest, but still large compared The irregular spatial distribution of fractures and
to pore size. We will indicate with extension the the presence of multiple scales makes it difficult and
size of a fracture in the directions orthogonal to the often impossible to account for fractures by deriv-
aperture. The extension of fractures in a network has ing effective upscaled parameters, permeability for
a distribution that is usually assumed to be governed instance, by volume averaging or homogeneization
by a power law, which implies the presence of a large techniques. Indeed, such procedures assume a strong
variation of space scales. See Fig 1. separation of scales. Therefore, methods have been
ii) Fractures may be either open or infilled by a developed that model fractures explicitly, at least
porous medium, whose physical characteristics may those crucial for the flow. These models are based
be strongly different from those of the surrounding on the assumption that in the porous medium flow is
material. governed by Darcy’s law, and often a similar model

1
2 Formaggia, L. and Scotti, A. and Fumagalli, A.

is adopted for the flow taking place in the fractures

as well.
We recall the main characteristic of the Darcy’s
model, considering, for simplicity, just the case of
single-phase flow. In this mathematical framework
the two main variables are the pressure p and the
macroscopic velocity field u, also called Darcy’s ve-
locity. The two quantities are related by Darcy’s law,

µK −1 u + ∇p = 0 in Ω, (1a)
where Ω ⊂ Rd represents the domain occupied by
the porous material and ∇ indicates the gradient.
In the case where gravity effect are relevant, equa-
tion (1) may be modified by replacing the pressure
term with p − ρgz, where ρ is the fluid density, g
the magnitude of the gravity acceleration and z is the
vertical coordinate pointing upwards from the Earth
surface. The main hypotheses behind the model are
that fluid velocity is small, so we can neglect inertial
effects, and the main model parameters are: µ, the
fluid viscosity, and K, the permeability tensor of the Fig. 1: Choice of the numerical model for domains
porous medium, which is a symmetric and positive at different space scales.
definite tensor. Permeability may be heterogeneous
in space and often with high variations.
The second equation expresses continuity of mass 2.1 Continuum Fracture Models
by the following differential equation,

cφ∂t p + ∇ · u = q in Ω, (1b) CFMs are early models introduced in the 1960’s [8],
later justified mathematically in [2], and currently
where ∇· is the divergence operator, q a source/sink implemented in many industrial software. They as-
term, c accounts for the medium and fluid compress- sume a highly permeable and interconnected fracture
ibility and φ is the porosity. Sometimes one is inter- network so that it can be modeled as a continuum su-
ested in the steady state solution or the compressibil- perimposed to that of the porous medium.
ity can be neglected, in which case cφ∂t p = 0. A commonly adopted method is the dual-poros-
Equations (1) form a system of partial differen- ity/dual-permeability scheme. It assumes that at each
tial equations which, complemented by appropriate point of the domain Ω we may use the Darcy’s equa-
boundary and initial conditions, allows to describe tions for flow in the fractures and in the porous
the evolution of (u, p) in the porous medium. medium, respectively, with a term representing the
interchange of mass between them. A basic model of
this type may be written as:

2 Numerical models for fractured cm φm ∂t pm + ∇ · um − α(pm − p f ) = qm

in Ω,
porous media c f φ f ∂t p f + ∇ · u f + α(pm − p f ) = q f
(2a)
Numerical models for fractured porous media may
be subdivided into two main categories: Continuum where suffixes m and f refer to quantities related to
Fracture Models (CFMs) and Discrete Fracture Mod- the porous medium and fractures, respectively. The
els (DFMs). term α(pm − p f ) represents the mass exchange be-
tween the two components, with α a rate parameter.
The Darcy’s velocities um and u f are given by

µKm
−1
um + ∇pm = 0
in Ω. (2b)
µK −1
f u f + ∇p f = 0
Title Suppressed Due to Excessive Length 3

A simpler model, called dual-porosity/single- µK −1

f ,t u f +  ∇τ p f = 0
permeability, may be obtained formally by setting in Γ, (3a)
∇τ · u f − [[um · n f ]] = q f
Km = 0. It assumes that the porous medium acts as
storage volume for flow occurring only along frac- where ∇τ · and ∇τ are the divergence and gradient
tures. A further simplification, appropriate only for operating on the tangent plane of the fractures, re-
highly connected networks of fracture of small ex- spectively. The jump of normal velocity across the
tension relative to the domain size, consists in the fracture acts as a source term in the mass conser-
use of a single equivalent continuum with upscaled vation equation and represent the net flux entering
properties that account for the combined effect of (or leaving) the fracture. This model must be com-
porous medium and fractures. plemented by appropriate boundary conditions. In
the portion of the boundary of Γ that touches the
boundary of Ω, the conditions are determined by the
specific problem at hand: either pressure or mass
2.2 Discrete Fracture Models
flux is prescribed. In the case of a fracture tip that
ends inside Ω, a zero mass flux condition is generally
DFMs represent fractures explicitly, modeled as a adopted.
network of (typically planar) surfaces Γ, immersed A network normally exhibits intersection between
in the porous medium. In the fractures we typically fractures. A common approach is to assume that the
use a Darcy-type model where some special source pressure is continuous at the intersection and the net
terms are added to account for the fluid exchange sum of fluxes is zero. More sophisticated models
with the porous medium. The Darcy equations in are available, to account for fractures with different
the latter are also modified, with terms that act as hydraulic properties and flow along fracture inter-
interface conditions. sections.
DFMs are computationally more demanding that To close the problem we need also to specify in-
CFMs, but also more accurate, particularly when terface conditions to couple Γ to the porous medium.
fracture of relative large extension are present. For In [7] a family of models is presented, including the
this reason the choice between CFM and DFM may following, often adopted in practice
depend on the spatial scale of interest, as illustrated
+ +
in Fig. 1. µ K f−1
,n um · n f + 2(p f − pm ) = 0
On each fracture we may identify a unit normal on Γ. (3b)
µ K f−1
,n um · n f + 2(pm − p f ) = 0
− −
vector n f , and thus a positive and a negative side of
Γ , see Fig. 2. We indicate with [[ f ]] = f + − f − the These conditions can be interpreted as the applica-
jump of a quantity f across the fracture. tion of a discrete Darcy law across the fracture, in-
deed they link the flux across the fracture to pressure
differences.
nf Remark 1 If the fractures are highly permeable in the
Γ normal direction, continuity of pressure across Γ is
+ often assumed, i.e. p+m = p−m . This induces a certain
- simplification in the model, since it does not account
for a net mass flow across the fracture, but only for
porous medium-fracture exchanges.
Fig. 2: Positive and negative side of Γ.
Remark 2 For nearly impermeable porous media, a
different simplification of these models, called Dis-
A commonly used model is derived in [7], in crete Fracture Networks (DFNs), consists in neglect-
which the fracture permeability is split into a normal, ing the effect of the porous material and simulate
K f ,n , and tangential, K f ,t , components to account flow just in the fracture network. They can be used
for the fact that those quantities may be different and in the presence of highly connected and permeable
scale differently with the fracture aperture . In the fractures.
porous medium Ω \ Γ we consider Eqs. (1), while in
the fractures we have, neglecting the compressibility
term,
4 Formaggia, L. and Scotti, A. and Fumagalli, A.

able operators to transfer the discrete solution in the

fractures to the porous medium grid and viceversa.
The so-called mortar technique, which is based on
the set up of additional variables at the interface, is
sometimes used to simplify the construction of the
transfer operators. The process of grid generation is
eased and it is simpler to avoid the generation of
highly distorted elements, however it is still rather
demanding in complex situations.
iii) Non-matching methods. The mesh for the
porous medium and the fracture network are com-
pletely independent and the fracture grids can cut
the porous medium mesh elements in an arbitrary
way. This simplifies mesh generation greatly, since
porous medium and fracture grids can be generated
independently. It is still necessary to find the inter-
Fig. 3: Porous medium and fracture grids. sections, but this is simpler than generating a con-
forming mesh and can be done with standard geo-
metric search tools. However, the implementation of
3 Discretization schemes for DFM conditions (3b) is more complex.
For every category, many different numerical
Numerical schemes for the approximation of the schemes are at disposal. For the cases i) and ii),
equations presented in the previous section are based it is beneficial to use techniques able to operate on
on partitioning the domain Ω into a mesh of polye- arbitrary polyhedral grids, like Finite Volumes or
dral elements Th . The unknowns are then discretized Mimetic Finite Differences, to mention the more es-
by assuming a given variation inside each element, tablished one. The research in this field is, however,
for instance constant or linear. The continuous so- very active, and we mention also Gradient Schemes,
lution is then replaced by the discrete values at the the Hybrid High Order (HHO) method and the Vir-
mesh nodes. tual Elements Method (VEM). A reference contain-
In the case of DFM we need to mesh both the ing examples of numerical schemes applied in a
porous medium and fracture domains and construct DFM context is [5].
suitable ways to couple the two via (3b). The many In the case iii), as already stated, the main diffi-
discretization techniques available in the literature culty is how to impose the interface conditions. In
may be roughly subdivided into three categories, de- that respect we have two class of procedures. The
pending on the relation between porous medium and first is to represent the possible jumps in the solu-
fracture grids, see Fig. 3. tion across the elements cut by a fracture explicitly.
i) Conforming methods. The mesh for the porous This is what is done in eXtended Finite Elements
medium conforms to that adopted for the fracture net- (XFEM), where the finite element basis functions
work. It means that a fracture mesh element coincide are locally enriched to allow discontinuous solutions
(geometrically) with the faces of two porous medium that can satisfy the coupling conditions (3b). Con-
mesh elements, on the positive and negative side of trarily, if one accepts a less accurate representation
the fracture. Consequently, no porous medium mesh of the solution, some manipulations of the interface
elements are cut by the fracture. This requirement conditions are possible to transform them into source
poses strong constraints on the mesh generation pro- terms acting both on the fracture elements and on the
cess, which indeed can be the most time-consuming elements in the porous medium that are cut by the
part of the simulation, particularly in the presence fracture. The resulting source terms have some sim-
of complex networks. On the other hand, the imple- ilarity with those present in CFM techniques. The
mentation of (3b) is rather straightforward. Embedded Discrete Fracture Model (EDFM), usu-
ii) Non-conforming methods. Still no elements ally coupled with a simple Finite Volumes approxi-
in the porous medium grid are cut by the fractures, mation, falls into this second category of methods.
however the grid at the two sides of the fracture, An reference on EDFM techniques is in [6].
as well as the fracture grid, are independent. The The performances of many classes of methods
implementation of (3b) involves the set up of suit- are presented and discussed in [4] for bi-dimensional
problems and [3] for three-dimensional problems.
Title Suppressed Due to Excessive Length 5

4 An example of computational solutions are obtained with the library PorePy, see
workflow https://github.com/pmgbergen/porepy.

Numerical simulation of flow in fractured porous

media is challenging due to the intrinsic geometrical 5 Conclusion
complexity of the fractures, as well as the measure-
ment of real fractures and their properties buried
Modeling flow in porous media in presence of frac-
deep in the underground. These data are difficult
tures and fracture networks is a challenging task.
to obtain and usually affected by large uncertainty
Several approaches are available in literature, and a
which compromises the reliability of the numerical
few of them implemented in specialised software.
solutions.
The choice depends on the scales at which the phe-
Several approaches can be considered to detect
nomenon has to be considered, on the connectivity
fractures in the underground, from seismic inversion
of the fracture network and on the level of accu-
to outcrop interpretation, the former effective to de-
racy desired. Simpler continuum fracture models are
tect big fractures few kilometres below the surface
suitable for highly connected and dense networks of
while the latter normally used as an analogue of the
fracture, while in presence of fracture with a larger
underground. Once the fractures are collected and
extension and more sparse, discrete fracture models
digitalized by means of one of these methods, a suit-
provide more accurate results. Clearly, it is also pos-
able mathematical model can be adopted to perform
sible to use a combination of the two approaches.
the simulations.
In this chapter we gave a rapid review of the dif-
We focus our attention on the case of outcrop in-
ferent strategies and, for the sake of simplicity, we
terpretation. From highly detailed photographs of the
focused the attention on single-phase flow. The gen-
interested region, all the fractures are collected and
eral conclusions can however be extended to the more
interpreted up to a minimum sampling scale deter-
complex situation of multi-phase flows.
mined by the quality of the data or by computational
constraints. Smaller fractures can be accounted for by
a suitable change of the porous medium properties Acknowledgements The authors acknowledge Davide Losa-
pio for his help in the research underlying this manuscript.
and seen as upscaled or homogenized. What quan-
tity defines a fracture “small” and thus not explicitly
represented is still an open question, many authors
consider the fracture extension a good proxy for its Cross-References
importance. After this operation, the digitalized ver-
sion of the outcrop is thus available, see Fig. 4 on the
top as an example where natural and human factors • Lithosphere, Mechanical Properties
limit its exposure. The outcrop is a portion of Sotra • Numerical Methods, Finite Element
• Sedimentary Basins
island, near Bergen in Norway.
Another challenging aspect is the collection of
the physical data, again affected by uncertainty, that
are needed to run the simulation. If some data are References
not available, it is possible to fill the gap considering
models from the literature to, for example, relate the
1. Adler, P.M., Thovert, J.F., Mourzenko, V.V.: Fractured
permeability with the aperture and the latter with Porous Media. Oxford University Press (2012)
the fracture extension. The model can, at this point, 2. Arbogast, T., Douglas Jr., J., Hornung, U.: Derivation
be discretized numerically by means of one of the of the double porosity model of single phase flow via
methods mentioned in the previous sections. homogenization theory. SIAM J. Math. Anal. 21(4),
823–836 (1990). DOI 10.1137/0521046. URL https:
Fig. 4 reports numerical results of a simulation, in
//doi.org/10.1137/0521046
particular the pressure and concentration of a tracer 3. Berre, I., Boon, W.M., Flemisch, B., Fumagalli, A.,
(for instance a contaminant) in the interpreted out- Gläser, D., Keilegavlen, E., Scotti, A., Stefansson, I.,
crop. Two extreme cases are considered, if the frac- Tatomir, A., Brenner, K., Burbulla, S., Devloo, P., Du-
tures are higher or lower permeable than the sur- ran, O., Favino, M., Hennicker, J., Lee, I.H., Lipnikov,
K., Masson, R., Mosthaf, K., Nestola, M.G.C., Ni, C.F.,
rounding porous medium. Data are homogeneous
Nikitin, K., Schädle, P., Svyatskiy, D., Yanbarisov, R., Zu-
and a left to right pressure gradient is imposed. The lian, P.: Verification benchmarks for single-phase flow in
three-dimensional fractured porous media. Tech. rep.,
6 Formaggia, L. and Scotti, A. and Fumagalli, A.

(a) Interpreted outcrop. (b) Digitalized fractures.

(c) p and u with high permeable fractures. (d) Scalar with high permeable fractures.

(e) p and u with low permeable fractures. (f) Scalar with low permeable fractures.

Fig. 4: On the top the interpreted outcrop and digitalized fractures. The blue fractures are geometrically
simplified due to software constraint. Images are taken from [4]. The others represent pressure, velocity
field, and a scalar tracer at a specific time. The colour scheme spans from the lowest in blue to the highest
in red. Images are taken from Fumagalli, A., Keilegavlen, E.: Dual virtual element methods for discrete
fracture matrix models. Oil & Gas Science and Technology - Revue d’IFP Energies nouvelles 74(41), 1–17
(2019). under CC BY 4.0 licence.
Title Suppressed Due to Excessive Length 7

arXiv:2002.07005 [math.NA] (2020). URL https://

arxiv.org/abs/2002.07005
4. Flemisch, B., Berre, I., Boon, W., Fumagalli, A.,
Schwenck, N., Scotti, A., Stefansson, I., Tatomir,
A.: Benchmarks for single-phase flow in fractured
porous media. Advances in Water Resources 111,
239–258 (2018). DOI 10.1016/j.advwatres.2017.
10.036. URL https://www.sciencedirect.com/
science/article/pii/S0309170817300143
5. Fumagalli, A., Berre, I., Formaggia, L., Keilegavlen, E.,
Scotti, A. (eds.): Numerical methods for processes in frac-
tured porous media. Lecture Notes in Geosystems Mathe-
matics and Computing. Birkhäuser, Basel (2019)
6. Lee, S., Jensen, C., Lough, M.: Efficient finite-difference
model for flow in a reservoir through discrete fracture
networks and homogenized media. SPE Reservoir Eval.
Eng. 3, 268–275 (2000)
7. Martin, V., Jaffré, J., Roberts, J.E.: Modeling fractures and
barriers as interfaces for flow in porous media. SIAM
Journal on Scientific Computing 26(5), 1667–1691 (2005)
8. Warren J., E., Root, P.J.: The behaviour of naturally frac-
tured reservoir. Soc. Petr. Eng. 3, 245 (1963)