Transp Porous Med (2009) 76:153–178

DOI 10.1007/s11242-008-9241-9

Effective Permeability of Media with a Dense Network

of Long and Micro Fractures

J.-F. Barthélémy

Received: 13 December 2006 / Accepted: 25 April 2008 / Published online: 4 June 2008
© Springer Science+Business Media B.V. 2008

Abstract This article presents a new methodology to estimate the effective permeability
of random fractured media of any anisotropy containing both microfractures and a large
number of long fractures crosscutting the representative volume element. The fractures are
replaced by fictitious permeable materials for which the tangential permeability is deduced
from a Poiseuille flow. A self-consistent scheme is proposed to derive the macroscopic
permeability. On the one hand, the contribution of long fractures to the effective permeability
writes by simple superposition of the fracture tangential permeabilities. On the other hand, the
contribution of microfractures needs to resort to auxiliary problems requiring the computation
of second-order Hill (or Eshelby) tensors related to ellipsoids embedded in an anisotropic
matrix, for which a complete procedure is detailed. The effect of the microfracture normal
permeability is put in evidence in the upscaling scheme and analyzed. In particular, it is shown
that it must be chosen large enough to allow the connections between families. Examples are
finally developed and compared to numerical simulations in the 2D case.

Keywords Anisotropic effective permeability · Self-consistent scheme · Fracture

permeability · Double scale fracture network

Nomenclature
X Vector notation
7 Second-order tensor notation
p Pressure field
∇P Macroscopic pressure gradient
q Microscopic filtration vector
Q Macroscopic filtration vector
µ Fluid viscosity

J.-F. Barthélémy (B)

Structural Geology Department, IFP, 1-4 av. Bois Préau, 92852 Rueil-Malmaison, France
e-mail: [email protected]

123
154 J.-F. Barthélémy

? Second-order identity tensor

?n Orthogonal projector on the plane of normal n(?n = ? − n ⊗ n)
*s Permeability of the solid phase
*i Permeability of a fracture of the ith family
Kt Tangential permeability of the fracture
Kn Normal permeability of the fracture
*hom Macroscopic permeability
*o Permeability of the reference medium
Go Green function for the reference permeability *o
/Io Hill polarization tensor for an ellipsoidal inclusion I in the reference medium
hessx Hessian operator with respect to the position x(hessx ≡ gradx gradx )

1 Introduction

A lot of industrial applications, e.g., hydrocarbon production, water supply or subcritical CO2
storage, are concerned by fluid flows in rocks. The issue of modeling such flows, for example,
at the scale of a reservoir for the assessment of field productivity and reserves, requires to
focus on the porous domain and more specifically on the fracture network. Indeed the latter
can be composed of several geological entities of different orders of length magnitude which
can all play a very important role in the overall hydraulic behavior of the field. In addition
to the pores and fractures at very small scale (decimetric or less) accounting for the intrinsic
porosity and permeability of the rock matrix, several fracturation scales are put in evidence
in Bourbiaux et al. (2002). On the one hand, large scale faults observable by seismic surveys,
sub-seismic faults and fracture swarms consisting in clusters of fractures can all be considered
as large scale fractures (hectometric or more) taken explicitly into account in flow simulation
models. On the other hand, diffuse fractures (metric to decametric) are so numerous that
they can not reasonably individually appear in those models; however, their presence might
significantly affect the fluid flows.
A workflow incorporating the hydraulic effects of fractures at all scales and their possible
interactions is detailed in Bourbiaux et al. (2002) so as to build a complete flow simulation
model. An example of model built and calibrated from available static and dynamic data and
containing fractures at different scales can be found in de Galard et al. (2005). Many real case
studies show that, as expected, the presence of large scale fractures of high transmissivity has
a strong impact on the field behavior and particularly on well production data, especially when
they cross the well (Cosentino et al. 2001; Jenni et al. 2007). Nevertheless, as emphasized in
Cosentino et al. (2001), the requested accuracy of the simulation used for prediction imposes
to build a dual medium model accounting for the interactions between the large draining faults
and their neighborhood, namely media of lower permeability but with a high fluid storage
capacity as diffuse fracture networks or even the matrix itself. For this purpose, the intrinsic
porosity and permeability of the matrix can sometimes be deduced from fracture observations
at the core scale (centimetric) as in Howard and Nolen-Hoeksema (1990) or from inversion
methods based on flowmeter data as in Cosentino et al. (2001). For some reservoirs, dynamic
data can be successfully numerically reproduced by a model involving only a permeable
matrix and networks of diffuse (metric to decametric) fractures but no large fault (Verga
et al. 2001). The combined influence of diffuse fractures and a permeable matrix suggests
resorting to techniques allowing to derive effective properties of dual porosity/permeability
media (Bourbiaux et al. 1998). In many articles proposing workflows for flow simulation
(Bourbiaux et al. 1998; Sabathier et al. 1998; Cacas et al. 2001; Garcia et al. 2007) or

123
Effective Permeability of Media 155

presenting real cases (Cosentino et al. 2001; de Galard et al. 2005), an important step consists
in representing the hydraulic effect of diffuse fractures by an effective permeability. In some
cases, the proposed upscaling method relies on a numerical resolution on an explicit local
discrete fracture network (Sabathier et al. 1998; Pouya 2005). Besides, the issue of upscaling
hydraulic properties of fracture networks by analytical methods has also been raised by several
authors: Oda (1986) presents a method based on a linear superposition of the contributions
of different fracture families, Zimmerman (1996) implements the differential scheme and
Maxwell’s method for 2D isotropic fracture networks, Fokker (2001) and Dormieux and
Kondo (2004) take advantage of the self-consistent scheme for, respectively, 2D anisotropic
networks and 3D isotropic networks.
This article deals with the estimate of the permeability of a fractured medium with a 3D
anisotropic fracture network. As the existence of a r.v.e. (representative volume element) is
assumed in the sequel, it becomes natural to define the macroscopic scale as that considering
the r.v.e. as an elementary particle with effective properties and the microscopic scale as that
refering to the different phases composing the r.v.e. (fracture families and solid phase). The
diffuse fracture network considered here contains a high density of families of microfrac-
tures and long fractures with small spacing. The adjective “long” characterizes fractures
crosscutting the r.v.e. The fracture centers are randomly distributed and the orientations of
the families are arbitrary, leading then to a possible anisotropic effective permeability. The
notion of r.v.e. implies the separation scale between its characteristic length and those of the
fracture families, namely the size of microfractures or the average spacing between long frac-
tures crosscutting the r.v.e. It should be recalled that the prefix “micro” only refers to a small
size compared to that of the r.v.e., whatever the order of magnitude of the latter (decimetric to
hectometric according to real case studies Sabathier et al. 1998). The framework of random
media upscaling, which is also commonly used to estimate mechanical effective properties
(Zaoui 2002), is considered here so as to provide a methodology based on the self-consistent
scheme to estimate the effective permeability *hom in the general case of a 3D anisotropic
distribution of fractures. The contribution of microfractures, requiring the determination of
anisotropic Hill polarization tensors, extends to 3D the results obtained in 2D in Fokker
(2001). The self-consistent scheme allows to highlight percolation thresholds in the case of
a strong contrast between the phases (fracture space and matrix). Considering fractures as
fictitious Darcean materials, i.e., characterized by a permeability tensor, this article focuses
on the relevance of the different components of this tensor. In particular, the influence of the
normal component on the macroscopic permeability is analyzed in details for both fracture
types and illustrated by (3D or 2D) isotropic distributions of fractures allowing to derive
closed-form solutions of *hom . In the general case of an anisotropic distribution, numerical
solutions of the self-consistent scheme can be obtained and compared to calculations made
on explicit discrete fracture networks.

2 Description of the Phases

2.1 Description of the Fracture Network

In natural diffuse fractured media as reservoirs, the issue of the existence of a r.v.e. on
which upscaling techniques can provide macroscopic properties such as permeability and
percolation thresholds has sometimes to be reconsidered. This is for example the case for
fracture networks with a fractal correlation (Bonnet et al. 2001; Darcel et al. 2003; Darcel
2002). Nevertheless, aiming at applying upscaling methods, the media which are studied

123
156 J.-F. Barthélémy

in the present article are based on the assumption of a random fracture distribution. In this
case, some articles (de Dreuzy et al. 2001a,b) have shown the influence of the geometrical
parameters of the fractures, in particular the length distribution, on the connectivity and the
permeability of the medium. A domain of validity of the effective medium theory based
on the self-consistent scheme that will be developed hereafter is provided in de Dreuzy et
al. (2001a). According to the latter, this technique can only be applied for a connectivity
beyond the percolation threshold whereas other theories or numerical devices depending on
the length distribution should be applied before and at the threshold. Nevertheless, the latter
as well as the permeability tensor can be rather well estimated by the self-consistent scheme
(Fokker 2001; Pozdniakov and Tsang 2004) except in the very neighborhood of the threshold
(Sahimi 1995) which appears as a transition at which the permeability progressively changes
of order of magnitude. It is worth mentioning that a percolation threshold can also be obtained
by means of Maxwell’s method for a 2D isotropic network in Zimmerman (1996). Although
thresholds derived by such effective media theories do not exactly coincide with numerical
simulations, those theories can anyway provide rough but quick estimates which can be
refined by numerical computations on effective fracture networks or which can also be used
to assess the sensitivity to the fracture network characteristics.
The present study relies on the assumption of existence of a r.v.e. with N fracture fami-
lies. The ith family (1 ≤ i ≤ N ) contains a large number of fractures all characterized by the
same length, aspect ratio, orientation and hydraulic properties. It must be emphasized here
that the following developments are based on a finite number of families. This means that the
characteristics (length, shape, orientation or conductivity) must take discrete values. Never-
theless, continuous distributions of the latter (defined for example by means of probability
density functions) could be accounted for by building an equivalent discrete set of families.
For example, in the framework of mechanics, Zhu (2006) presents methods either based on
numerical integration by gaussian quadrature or on the definition of a representative set of
families allowing to replace a continuous orientation distribution by a discrete one. The set
of N fracture families is partitioned in two subsets: families of long fractures crosscutting
the r.v.e. (i ∈ C ) and families of microfractures (i ∈ M).

• Families of fractures crosscutting the r.v.e. (i ∈ C )

A fracture crosscutting the r.v.e. is geometrically described here as a plane, assuming then
that its curvature can be neglected at the scale of the r.v.e.
The average spacing between fractures of the ith family is denoted by di . This distance
can be considered as the characteristic dimension of the family and must be very small
compared to the size of the fractures and to that of the r.v.e. (Fig. 1). The orientation is
entirely defined by a normal vector, i.e., two Euler angles. Adopting a 3D description of the
fracture, the latter can be seen as a volume delimited by two parallel planes. Introducing the
average geometrical aperture h i , possibly different from the hydraulical aperture defined
later, the infinitesimal volume fraction of the family is:
hi
fi = (1)
di
• Families of microfractures (i ∈ M)
Considering fractures as 2D plane objects, the microfractures are modeled as ellipses and
the characteristic length of each family (largest radius ai ) is assumed to be very small
compared to the characteristic dimension of the r.v.e. The aspect ratio characterizing the
shape in the fracture plane is denoted by ηi = bi /ai where bi is the small radius. If Ni
denotes the number of fractures per volume unit and i = Ni ai3 the classical density

123
Effective Permeability of Media 157

Fig. 1 Fractures crosscutting the

r.v.e.

Fig. 2 Network of
microfractures

introduced by Budiansky (Budiansky and O’Connell 1976), the average spacing between
two elements of a family is then the inverse of the average number of fractures intersecting
a segment of unit length orthogonal to the fracture plane:

1 ai
di = = (2)
Ni π ai ηi
2 π i ηi

The orientation of all the microfractures of a given family is defined by a 3D frame (three
orthonormal vectors) i.e., three Euler angles as described later in Fig. 7. A network
composed of three families of microfractures is shown on Fig. 2.

123
158 J.-F. Barthélémy

Fig. 3 Ellipsoidal fracture

For sake of convenience in the self-consistent scheme based on Eshelby’s problem and
developed in the following, we also introduce a 3D model of microfractures considered
as ellipsoids of largest radii ai and bi . The third dimension is then defined by the small
radius ci or the ratio ωi = ci /ai  1 (see Fig. 3). Hence, the volume fraction of the
microfracture family writes:
4
fi = π i ηi ωi (3)
3
It is assumed that all the characteristic lengths of the families (spacing in the case of fractures
crosscutting the r.v.e. and size in the case of microfractures) are of the same order of mag-
nitude, i.e., infinitesimal with respect to the dimension of the r.v.e., in order to incorporate
all the families to the same upscaling process. Otherwise several successive steps have to be
performed.

2.2 Hydraulic Properties of the Phases

Ene and Sanchez-Palencia have shown, in the framework of periodic homogenization, that
the steady-state, laminar, single-phase and incompressible flow of a newtonian viscous fluid
in a porous medium obeys to the Darcy law at the upper scale, which means that the filtration
velocity is linearly related to the pressure gradient (Ene and Sanchez-Palencia 1975):
*hom
Q=− · ∇P (4)
µ
where the second-order tensor *hom has the dimension of a surface (m 2 ) and depends on
the sole morphology of the porous space (i.e., not on the fluid viscosity µ). The aim of
the upscaling procedure is to determine or at least estimate *hom . It is commonly assumed
that the relationship (4) also applies to random media but estimating *hom from Stokes
equations at the lower scale proves to be a hard task. That is why some authors (Fokker
2001; Dormieux and Kondo 2004; Pozdniakov and Tsang 2004) have proposed to replace
the different constituents of the microscopic scale (fractures and solid phase) by fictitious
Darcy media. The advantage of such a method is that the same type of equation, namely the
Darcy law, applies at both the microscopic and macroscopic scale.
The permeability of the solid phase is assumed to be isotropic *s = K s ?. This perme-
ability results from the presence of micropores at a lower scale and is possibly obtained by
a previous upscaling procedure. The permeability of the fractures writes with tangential and
normal permeabilities which are a priori different:

123
Effective Permeability of Media 159


*i = *it + K ni ni ⊗ ni with *it = K ti ?ni = K ti ? − ni ⊗ ni (5)

Following a classical result of rock hydraulics (Guéguen and Palciauskas 1994), K ti accounts
for the overall tangential permeability of the fracture and is identified with the ratio between
the average velocity of a Poiseuille flow between two idealized planes and the pressure
gradient in the direction of the planes. The space ei between those fictitious planes is called
the hydraulic aperture. It may differ from its geometrical counterpart because the latter may
not be uniform along the fracture due to the face rugosity or simply due to the specific 3D
geometry, for instance in the case of an ellipsoid. The equivalent Poiseuille flow then gives:

ei2
K ti = (6)
12

It is assumed in the following that the tangential permeability K ti is much larger than K s
implying the impermeability of the fracture faces. As the equivalent Poiseuille problem is
based on a flow parallel to the fracture plane, it does not give any indication on how to
estimate K ni . Recalling that a permeability is the macroscopic evidence of charge loss due
to friction at the fluid-solid interface at the microscopic scale, the intuition could lead us to
choose a value of K ni much larger than K ti because the latter should be much more impacted
by the narrowness of the fracture aperture (6) than the former. Nevertheless, some authors
have proposed different other choices: for example, the microfracture permeability is taken
isotropic (K ni = K ti ) in Fokker (2001) or K ni = K s  K ti in Dormieux and Kondo (2004).
Thus, replacing the real fluid flow in the fracture space by an effective flow in a fictitious
Darcy medium somehow makes K ni in (5) a mathematical parameter which does not seem
so obvious to choose. It is then necessary to raise the issue of the relevance of any choice of
K ni by analyzing its impact on the overall permeability, which is the purpose of Sect. 4.

3 Determination of the Macroscopic Permeability

The upscaling problem to solve on the r.v.e.  is similar to that allowing to exhibit effective
thermal or electrical properties, i.e., is the following elliptic problem:

⎧
⎪ div q = 0 () (a)
⎪
⎪
⎪
⎪ Ks
⎪
⎨q = − grad p (s ) (b)
µ (7)
⎪
⎪ * i
⎪
⎪ q=− · grad p (i ) (c)
⎪
⎪ µ
⎩
p = ∇P · x (∂) (d)

where s is the space occupied by the matrix, i the space occupied by the ith fracture
family and ∇ P is a given uniform vector identified as the macroscopic pressure gradient.
The linearity of the equations at stake implies the existence of a concentration tensor field
such that the solution of (7) writes:

∀x ∈ , grad p(x) = (x) · ∇ P (8)

123
160 J.-F. Barthélémy

The macroscopic permeability is then given by:

Q =< q > = − *µ · ∇ P
hom

 (9)

N 
N
with *hom =< * · > = 1 − f i *s · < >s + f i *i · < >i
i=1 i=1

where < · >ω denotes the average operator over the domain ω. The upscaling methods
aim at providing estimates of the averages of the concentration tensor or equivalently the
relationships between the averages of q in each phase and ∇ P.

3.1 Contribution of the Fractures Crosscutting the r.v.e.

As they crosscut the r.v.e., those fractures are subjected to a pressure difference directly
related to the boundary condition (7 d). In fact, as pointed out in Oda (1986), the fluid flow in
those fractures sees an effective pressure gradient obtained by projection of the macroscopic
pressure gradient onto the fracture plane. By construction of K ti and recalling that the fracture
faces are impermeable, the flow in such a family is purely tangential and writes:

K ti *i
< q >i = − ?ni · ∇ P = − t · ∇ P (∀i ∈ C ) (10)
µ µ

The contribution of this family to *hom (9) can then be identified in (10) by:

*i · < >i = *it = K ti ?ni (∀i ∈ C ) (11)

which appears as a superposed term in (9) without any interaction with other fracture fami-
lies. K ni is obviously not involved in the explicit contribution of the corresponding family to
*hom (11).
Focusing on the estimate of the pressure gradient average, i.e., < >i , it can first be
noted that (11) leads to

?ni · < >i = ?ni and K ni ni ⊗ ni · < >i  K ti (∀i ∈ C ) (12)

The first relationship of (12) allows to determine without ambiguity the projection < >i
onto the fracture plane whereas the projection onto the normal depends on the choice of K ni .
If K ni is chosen at least of the same order of magnitude as K ti , then (12) implies that

< >i = ?ni (∀i ∈ C ) (13)

The choice K ni  K ti still satisfies the consistency with (11) but leads to three undetermined
components of < >i . In fact, the following developments in Sects. 2.2 and 2.4 will show
that those components of < >i do not actually play a role in the upscaling process
provided that they keep bounded values. (13) Can therefore still be used.

3.2 Contribution of the Microfractures

Unlike the previous case of fractures crosscutting the r.v.e., the interactions of smaller frac-
tures with their neighborhood must be strong. Their influence on the overall permeability
comes from the connections between them, which means that each phase of the heterogeneous
medium is in contact with elements of the other phases. According to Dormieux and Kondo

123
Effective Permeability of Media 161

(2004), the self-consistent scheme is adapted for its capacity to account for a disordered
microstructure and to exhibit percolation thresholds.
In the framework of the self-consistent scheme, the average of the concentration tensor
in each phase is estimated by means of an auxiliary problem of an ellipsoid I embedded
in the infinite effective medium itself and subjected to a remote pressure gradient ∇ P o .
The use of ∇ P o different from ∇ P corresponds to a classical way to take into account
interactions in upscaling schemes (see for example Zaoui 2002 in the mechanical framework).
The relationship between ∇ P o and ∇ P is obtained by means of the consistency rule:

< grad p > = ∇ P (14)

The auxiliary problem, known as the Eshelby problem (Eshelby 1957), shows an homoge-
neous pressure gradient within the ellipsoid, which is used as an estimate of the pressure
gradient average in the fractures of the ith family in the r.v.e. problem:
−1
< grad p >i = *i · ∇ P o with *i = ? + /ihom · *i − *hom (∀i ∈ M)
(15)

where the second-order Hill tensor /ihom depends on *hom and on the shape and orientation
of the ellipsoid. The ellipsoid representing the ith family is defined thanks to a second-order
tensor 'i by the equation:

−1
x · t 'i · 'i ·x ≤1 (16)

For any ellipsoid, /ihom results from *hom in the general 3D anisotropic case according to
the following procedure (see Appendix for a detailed proof):

1. Diagonalize *hom and introduce the tensor 1 such that

3
t
0 · 0 = ? and t
0 · *hom · 0 = K hom
j ej ⊗ ej (17)
j=1

3 
1= K hom
j 0 · ej ⊗ ej (18)
j=1

t
2. Compute the tensor 'i = 'i · 1−1 and diagonalize t 'i · 'i

2i · 'i · 'i · 2i = Ai2 e1 ⊗ e1 + Bi2 e2 ⊗ e2 + Ci2 e3 ⊗ e3 (19)

t
t
2i · 2i = ? and t

with Ai ≥ Bi ≥ Ci 1
3. Compute the tensor

#i = I Ai e1 ⊗ e1 + I Bi e2 ⊗ e2 + ICi e3 ⊗ e3 (20)

with

1 A permutation between the eigenvectors is possibly used so as to respect this order.

123
162 J.-F. Barthélémy

• if Ai > Bi > Ci

Ai Bi Ci
I Ai =
 2 (Fi − Ei ) (21)
Ai2 − Bi2 Ai − Ci2
⎛  ⎞
Ai Bi Ci Bi Ai2 − Ci2
⎝ − Ei ⎠
IC =
 2
i
(22)
B2 − C 2 A − C2 Ai C i
i i i i

I Bi = 1− I Ai − ICi (23)
where Fi = F (θi , ki ) and Ei = E (θi , ki ) are respectively the elliptic integrals of the
first and second kinds of amplitude and parameter:
 
Ci2 Ai2 − Bi2
θi = arcsin 1 − 2 ; ki = (24)
Ai Ai2 − Ci2
• if Ai > Bi = Ci


1 Ai Ai2 − Ci2 − Ci2 argch (Ai /Ci )
I Bi = ICi = Ai
2 3/2 (25)
2 Ai − Ci2
I Ai = 1 − 2 I Bi (26)
• if Ai = Bi > Ci


A 2 arccos (C /A ) − C Ai2 − Ci2
1 i i i i
I Ai = I Bi = Ci
2 3/2 (27)
2 Ai − Ci2
ICi = 1 − 2 I Ai (28)
• if Ai = Bi = Ci

1
I Ai = I Bi = ICi = (29)
3
4. Finally compute
t
/ihom = 1−1 · 2i · #i · t 2i · 1−1 (30)
The determination of the average of the concentration tensor < >i will be achieved when
the relationship between ∇ P o and ∇ P is found. However, as this relationship comes from
the consistency rule (14), the average of the pressure gradient within the matrix must also be
estimated through an auxiliary problem even if the explicit contribution of the matrix in (9)
could be neglected because of the low permeability of the matrix.

3.3 Contribution of the Matrix

In the self-consistent scheme, the matrix phase also requires the resolution of an auxiliary
problem to estimate the average of the pressure gradient. Nevertheless, unlike polycrystals

123
Effective Permeability of Media 163

Fig. 4 Matrix ellipsoid shape

in which each phase plays an equivalent role, the matrix here plays a specific role from
the geometrical point of view. The latter, which occupies almost all the r.v.e. space, is also
represented by an ellipsoid in its auxiliary problem but the shape of the ellipsoid may not be as
obvious as for the polycrystals. This ellipsoid is supposed to account for the spatial distribution
of the matrix with respect to the fractures crossing it. More precisely, it seems judicious to
quantify the anisotropy of the inter-fracture space by means of the family orientations ni
and spacings di as shown on Fig. 4 and build the 's tensor related to the matrix ellipsoid
according to the definition:
 N −1
1
' =s
ni ⊗ ni (31)
di
i=1

Knowing the shape of the ellipsoid, the average of the pressure gradient within the matrix
can then be estimated by means of the same procedure as for the microfractures, i.e.,:


−1
< grad p >s = *s · ∇ P o with *s = ? + /shom · *s − *hom (32)

where /shom depends on *hom and 's (31).

3.4 Macroscopic Permeability

Using the estimates (15) and (32) in the consistency rule (14) provides the relationship
between the auxiliary pressure gradient ∇ P o and the actual one ∇ P:
 N
 −1  
∇ Po = 1− f i *s + f i *i · ?− fi < >i · ∇P (33)
i=1 i∈M i∈C

Since the fracture volume fractions are infinitesimal, the average pressure gradients, i.e.,
< >i of the families of long fractures play a role in (33) only if it presents singular
values. If < >i is bounded as in (13), (33) gives:
 −1
∇ P o = *s + f i *i · ∇P (34)
i∈M

A singular component of < >i , along ni ⊗ ni to remain consistent with (12), would have
implied a reduction of the level of ∇ P o and consequently of *hom , which would have made
the corresponding long fracture family play the unexpected role of a barrier against the fluid

123
164 J.-F. Barthélémy

flow possibly coming from microfracture families. Finally, the average of the concentration
tensor in the microfracture families (resp. in the matrix) can be written with the help of (15)
(resp. 32) and (34):
 −1
< >i = *i · *s + f i *i (∀i ∈ M) (35)
i∈M

 −1
< >s = *s · *s + f i *i (36)
i∈M

The macroscopic permeability *hom can now be estimated by taking advantage of (10), (35)
and (36) in (9):
 N
   −1
*hom = 1− f i *s · *s + f i *i · *i · *s + f i *i + f i *it
i=1 i∈M i∈M i∈C
(37)
It is recalled that, in the framework of the self-consistent scheme, the tensors *i (15) and
*s (32) nonlinearly depend on *hom , which makes (37) an implicit equation characterizing
*hom :


*hom = G *hom (38)
Although closed-form solutions can be obtained in simple cases (see Sect. 5), it is necessary
to use a numerical method to find a solution of (38). A recursive fixed-point method will be
implemented in the following examples:


*0 = *s and * j = G * j−1 until * j − * j−1 < ε * j−1 (39)
However, the relevance of the upscaling scheme still relies on the choice of the normal
component of the microfracture permeability, which is the purpose of the next paragraph.

4 Influence of the Normal Component of the Fracture Permeability

It has been shown earlier that the macroscopic permeability does not depend on K ni of
the long fractures, which contributions are obtained by simple superpositions. The case of
microfractures is different since they are confined inside the r.v.e. and the flow and the pressure
gradient established in them strongly depend on their neighborhood. In the self-consistent
scheme, the solution in the microfractures has been estimated by that of an ellipsoid sur-
rounded by the effective material. Moreover, according to Voigt bound, *hom as a quadratic
form is always lower than < * > and is reasonably of the same order of magnitude of
the latter when the fractures are connected or in the presence of long fractures crosscutting
the r.v.e. (37). This implies that *hom is necessarily much lower than the tangential perme-
ability K ti of any fracture contributing to the overall permeability since the fracture volume
fraction is infinitesimal. Therefore, the fluid flow is almost parallel to the fracture plane in
the auxiliary problem without any condition on K ni as the neighboring material is much less
permeable than the ellipsoidal fracture. The quantitative influence of K ni on *hom in the
case of microfractures has then to be examined more in details. To this end, the behavior of
*i (15) and /ihom for small values of the aspect ratio ωi is first studied. The Taylor expansion

123
Effective Permeability of Media 165

of /ihom with respect to ωi can be obtained from another expression of the Hill tensor than
the one using the Green function exploited in Sect. 3.2 (Dormieux et al. 2006):

det 'i ξ ⊗ξ
/ihom =

t 3/2 dSξ (40)
4π ξ · *hom · ξ ξ · 'i · 'i · ξ
ξ =1

Calling (l i , mi , ni ) the eigenvectors of the ellipsoid (see Fig. 3), 'i writes:

'i = ai (l i ⊗ l i + ηi mi ⊗ mi + ωi ni ⊗ ni ) (41)
√
Using in (40) the parametrization ξ = 1 − z 2 (cos φ l i + sin φ mi ) + z ni and dSξ = dz dφ
as well as an analogy with a result obtained in the framework of elasticity in Huang and Liu
(1998), /ihom writes:
 
ni ⊗ ni ωi2
/ihom = + ωi /ihom + O (42)
ni · *hom · ni *hom

with:

1 2 π 
ηi z 1 ∂ ξ ⊗ξ
/ihom = − √
 dφ dz
4π 1 − z 2 cos2 φ + ηi2 sin2 φ 3/2 ∂z ξ · *hom · ξ
z=−1 φ=0
(43)


It is worth noting on (43) that /ihom = O 1/ *hom .
In the case of an oblate spheroid (ηi = 1) in an isotropic effective medium *hom = K hom ?,
(42) becomes:
 
ni ⊗ ni ωi2 π
/ihom = + ωi /ihom + O with /ihom = ?ni − 2 ni ⊗ ni
K hom K hom 4 K hom
(44)

Substituting (42) in (15), it comes out that:

ni ⊗ *hom · ni K ni ni ⊗ ni
*i −1 = ? − + + ωi /ihom · *i − *hom
K nhom
i ni
K nhom
i ni
 
*i
+O ωi2
(45)
*hom

where K nhomi ni
= ni · *hom · ni . As already previously mentioned, it seems reasonable to as-
sume that, beyond the percolation threshold, *hom is of the same order of magnitude as f i *it
if the ith family significantly contributes to *hom . Keeping the terms of highest order, *i −1
can be written by blocks:

123
166 J.-F. Barthélémy

*i −1 = ?ni + ωi ?ni · /ihom · *it +ωi ?ni · −/ihom · *hom · ni ⊗ ni

     
@i Ai
 
*hom
+ ni ⊗ ni · ωi K ti /ihom − hom · ?ni
K ni ni
   (46)
 Bi 
K ni
+ − ωi ni · /ihom · *hom · ni ni ⊗ ni
K nhom
n
i i
  
δi

The tensors @i , Ai and Bi introduced in (46) do not explicitely depend on K ni and are all of the
same order of magnitude, namely O (1). As an endomorphism of R3 , @i is singular whereas,
as an endomorphism of the fracture plane, @i can be inverted using the following notation:
−1 −1
@i −1 = ?ni · ? + ωi ?ni · /ihom · *it = ?ni · ? + ωi K ti ?ni · /ihom · ?ni (47)

In addition, (46) shows that K ni explicitely appears only in δi (bottom right term, i.e., along
ni ⊗ ni ), which order of magnitude controls that of *i . As regards the order of magnitude
of δi and its dependence on K ni , two cases can be examined:

• K ni ωi *hom ⇒ δi = K ni /K nhom
i ni
For this condition, *i writes:

K nhom

*i = @i −1 + i ni
−Bi · @i −1 + ni ⊗ ni (48)
K ni

On the one hand, recalling that f i is given by (3), it comes that f i *i can be neglected in
the interaction term (34). On the other hand, (48) also allows to compute the term *i · *i
which will be crucial in (37). Indeed, as ?ni · Bi =  (46), it comes:


*i · *i = *it · @i −1 + O *hom
−1

= *it · ? + ωi ?ni · /ihom · *it +O *hom (49)

and, in the case of an oblate spheroid in an isotropic effective medium:

*it

*i · *i = + O K hom (50)
π ωi K t
i
1+ 4 K hom

Taking (49) and (32) into account as well as the impermeability of the matrix, the implicit
expression of *hom (37) can be simplified:
 
−1

* hom
= f i *t · ? + ωi ?ni · /hom · *t
i i i
· ? − /shom · *hom
i∈M

+ f i *it (51)
i∈C

123
Effective Permeability of Media 167


• K ni = O ωi *hom
In this case, δi can be written:
K ni
δi = ωi δi with δi = − ni · /ihom · *hom · ni = O (1) (52)
ωi K nhom
i ni

To the first non-zero order, *i writes:

  −1 
Bi −1 Ai · Bi · @i −1 Ai ni ⊗ ni
*i = ?ni −  · @i · ?− · ?ni −  + (53)
ωi δi δi δi ωi δi
Unlike the previous case, f i *i can not be neglected when ωi takes small values. Using (3),
f i *i writes to the first order:
  −1 

4 π i ηi −1 Ai · Bi · @i −1 Ai
f i *i = −Bi · @i · ? − · ?ni −  + ni ⊗ ni (54)
3 δi δi δi

leading, in the case of an oblate spheroid in an isotropic effective medium for which /ihom
is given by (44), to:
4 ni ⊗ ni
f i *i = π i (55)
3 K ni π
ωi K hom
+ 2

*i · *i can also be obtained:

 −1 
Ai · Bi · @i −1 Ai

*i · *i = *it · @i −1 · ? − · ?ni −  +O *hom (56)
δi δi
leading to (50) in the case of an oblate spheroid in an isotropic effective medium.
The non-zero term (54) involved in (34) and (37) highlights the negative influence of a
too small value of K ni on *hom . Indeed, this choice induces an unexpected role of barrier
of the fracture in its transverse direction ni contributing to decrease at least the value of
K nhom
i ni
. Nevertheless, keeping such a small value of K ni could be used to somehow account
for cementing on the fracture faces impeding the fluid flow from other fractures.
From the results obtained hereabove, one can now conclude that, in the modeling consisting
in replacing a fracture by a fictitious Darcy medium, the most relevant choice of the transverse
permeability is specific to the fracture type. Whereas the choice of K ni of the long fractures
has no consequence on the macroscopic permeability, in the case of a microfracture, K ni
must be large enough (K ni ωi2 K ti ) to avoid creating an artificial obstacle in the transverse
direction except if it corresponds to the aim of the model.

5 Implementation of the Self-Consistent Scheme on Examples

5.1 Isotropic Distributions of Microfractures

• 3D case
In this paragraph, the fracture orientation distribution is assumed to be isotropic and all the
fractures are modeled by identical oblate spheroids of aspect ratio ω with the same perme-
ability. The macroscopic permeability is isotropic, i.e., *hom = K hom ? and the ellipsoid
related to the matrix is a sphere according to (31). It is worth noting that, given an overall

123
168 J.-F. Barthélémy

Fig. 5 3D isotropic distribution of spheroidal fractures (ω = 10−3 )

fracture density , the same permeability tensor *hom can be obtained with a network of
three orthogonal families of density /3 each. Recalling that /ihom is written in (44) for
−1
oblate spheroids and /shom = (3K hom ) ? for a sphere, the macroscopic permeability is
obtained by solving (37), which leads to a relationship of the type:

K hom K s Kn
=L , , , ω (57)
ω Kt Kt Kt

Figure 5 shows numerical computations of K hom /K t with respect to the density  for
ω = 10−3 and different values of K s /K t and K n /K t , thus illustrating the role of the order
of magnitude of the latter which has been highlighted in Sect. 4. When K s tends toward
0, closed-form solutions of K hom can be found.

– If K n ω2 K t , a percolation threshold is detected for  = 27/64 and, as already found

in Fokker (2001), the permeability beyond the threshold writes:

K hom π 27
= (64  − 27) + O (ω) ,  ≥ (58)
ω Kt 108 64
It is recalled that this percolation threshold  = 27/64 ≈ 0.42 has been compared in
Fokker (2001) to an estimate  ≈ 0.30 (i.e., a fracture volume fraction f ≈ 1.27ω)
coming from numerical computations on networks of randomly placed ellipsoids pro-
vided by Garboczi et al. (1995), showing that the self-consistent scheme overestimates
the threshold in 3D.
– If K n  ω2 K t , the percolation threshold is reached for a larger value namely  = 9/16
highlighting the negative effect of a too small value of K n compared to the previous
case. As already found in Dormieux and Kondo (2004), the permeability beyond the
threshold is inferior to that of (58) and writes:

K hom 3 π 16  − 9 9
= + O (ω) ,  ≥ (59)
ω Kt 4 16  + 27 16

123
Effective Permeability of Media 169

The irrelevance of the model K n  ω2 K t can also be deduced from the expression (59).
Indeed, let the number of fractures per volume unit N and the aperture c be given and
let the radius a increase so as to examine its influence on K hom . Recalling that  = N a 3
and K t = c2 /3, it comes from (59) that K hom varies as 1/a when a is large, meaning that
the fracture length would have a negative effect on the permeability, which is unrealistic.
This, once again, shows that a small value of K n makes the fractures play the unexpected
role of obstacles in their transverse direction.

• 2D case
The same computation can be done on a 2D fractured medium. The geometry is invariant
along e3 and isotropic in the (e1 , e2 ) plane, i.e., K hom is transversely isotropic:
*hom = K /hom
/ (e1 ⊗ e1 + e2 ⊗ e2 ) + K ⊥
hom
e3 ⊗ e3 (60)
The fractures are modeled as flat cylinders of elliptic section. As regards the permeability,
one can equivalently consider a random distribution of orientation or two orthogonal
orientations. Following the procedure described in Sect. 3.2 in the limit case of a largest
axis tending toward infinity, the Hill tensor related to a cylinder of elliptic section (same
aspect ratio ω for all the families) embedded in a transversely isotropic medium with the
same axis is:

1 ω 1
/ihom = hom mi ⊗ mi + ni ⊗ ni (61)
K // 1+ω 1+ω
where ni denotes the normal vector, i.e., the small axis direction and mi the direction
orthogonal to ni in the (e1 , e2 ) plane, i.e., the large axis direction. It is worth noting that
the expression (61) is also found in Fokker (2001) in the framework of a complete 2D
resolution based on the 2D Green function. The isotropy in the (e1 , e2 ) plane leads to the
following expression of /shom
1
/shom = (e1 ⊗ e1 + e2 ⊗ e2 ) (62)
2 K /hom
/

If f denotes the total fracture volume fraction, the fracture density  in the 2D framework
is such that f = π  ω. By projection of (37) on e3 , K ⊥ hom is found, as expected, as an

arithmetic average:
hom
K⊥ = (1 − π  ω) K s + π  ω K t (63)
where the K s term disappears for an impermeable matrix. The solution in the (e1 , e2 )
plane is shown on Fig. 6 for ω = 10−3 for different values of K s /K t and K n /K t .
Moreover, a closed-form solution is available in the case of an impermeable matrix and
Kn ω2 K t :
K /hom
/ π 4
=  − 1 + O (ω) ,  ≥ (64)
ω Kt 4 π
This percolation threshold  = 4/π ≈ 1.27 can be compared to the value  ≈ 1.4 obtained
from numerical computations by several authors for the same problem of draining fractures
in an impermeable matrix (Robinson 1983; Berkowitz 1995; Bour and Davy 1997). Unlike
the 3D case, the self-consistent scheme underestimates the percolation threshold but gives
a better estimate. It can be mentioned that, in the inverted problem of impermeable needle-
like slits in a permeable matrix, the self-consistent scheme as well as Maxwell’s method

123
170 J.-F. Barthélémy

Fig. 6 2D isotropic distribution of elliptic fractures (ω = 10−3 )

give the same percolation threshold  = 4/π (above which K /hom / = 0) and even the same
function K // () which is rather successfully compared to numerical simulations except
hom

in the very neighborhood of the threshold (Zimmerman 1996).

Whereas a percolation threshold is found here for a large enough value of K n , the case
K n  ω2 K t does not allow the macroscopic permeability to reach the order of magnitude
of ω K t i.e., does not exhibit any percolation threshold for a finite value of  (see Fig. 6),
once more illustrating the effect of K n .

5.2 Anisotropic Distributions of Fractures

• 3D case
The algorithm leading to the solution to (37) is now implemented in the case of an
anisotropic 3D network. As shown on Fig. 7, the orientation of a given family with re-
spect to a fixed basis (e1 , e2 , e3 ) of R3 is determined by three Euler angles: two angles
(spherical coordinates θ and φ) are needed to define the normal orientation n and a third
angle ψ (between the major axis and eθ on Fig. 7) allows to define the rotation around n
of the fracture plane. Three families are considered in this example. The aspect ratio ωi
(see Fig. 3) and the tangential permeability are the same for all the families ω = 10−3 .
The other characteristics are the following:

– family 1: θ1 = 75◦ ; φ1 = 0◦ ; ψ1 = 0◦ ; η1 = 0.5

– family 2: θ2 = 90◦ ; φ2 = 90◦ ; ψ2 = 0◦ ; η2 = 1
– family 3: θ3 = 90◦ ; φ3 = 45◦ ; ψ3 = 45◦ ; η3 = 0.5

The three family orientations and shapes are represented on Fig. 8.

In addition, the three fracture volume fractions are assumed to be equal, i.e., one-third
of the total microfracture volume fraction. The effective permeability of such a fractured
medium is computed for different values of the latter (from 0 to 60/00). Moreover, the same
computations are carried out in the presence of a family crosscutting the r.v.e. of normal

123
Effective Permeability of Media 171

Fig. 7 Angular coordinates

of a fracture

Fig. 8 Three families

of microfractures

e3 (horizontal plane) and of fixed volume fraction 10/00. According to the conclusions of
Sect. 4, the normal permeability of the crosscutting family is zero whereas the tangential
permeability is chosen of the same order of magnitude as that of the microfractures. In
both cases (with or without the family of long fractures), the eigenvalues of *hom are
plotted against the total fracture volume fraction on Fig. 9.
As regards the case without the family of long fractures, a percolation threshold is detected
as expected. Beyond a fracture volume fraction of about 3.10/00, the three eigenvalues of
*hom reach the order of magnitude of ω K t and the principal directions remain constant.
It is worth precising that the direction related to the highest permeability is characterized
by the angles θ = 9◦ and φ = 174◦ , which is close to the vertical axis e3 .

In the presence of the crosscutting family, the percolation threshold disappears and three
phases can be observed on Fig. 9:

– f < 30/00
The permeability is essentially controlled by the long fractures, i.e., the directions related
to the two highest permeabilities are close to the direction of the plane (e1 , e2 ) but
the microfractures play nevertheless a role. Indeed, in the orthogonal direction, the
permeability starts to increase as soon as the microfracture volume fraction increases.

123
172 J.-F. Barthélémy

Fig. 9 Eigenvalues of the permeability of a 3D network with and without a family of long fractures crosscutting
the r.v.e.

This result can be somehow justified by the fact that even few microfractures create a
connection between the crosscutting planes because the spacing between these ones is
of the same order of magnitude as the length of the microfractures.
– 30/00 ≤ f ≤ 40/00
This phase is a transition during which the effects of the crossing fractures and those of
the microfractures are of the same order of magnitude. The computations also show a
progressive change of orientation of the principal directions. In particular, the direction
related to the highest permeability initially in the horizontal plane rotates and tends
toward the major direction of the case without crosscutting fractures.
– f > 40/00
The permeabilities linearly increase with the fracture volume fraction. Due to the den-
sity of the microfractures, the effect of the crosscutting family gets slightly attenuated.
Nevertheless, it has still to be considered within the abscisse range of Fig. 9 for the two
lowest permeabilites. Indeed, the associated directions are close to the horizontal plane
in which the crosscutting family still plays a role for f < 70/00.

• 2D case: comparisons with computations on explicit discrete fracture networks

The same methodology can be used in the 2D case for which we consider for example two
families of microfractures (modeled as flat cylinders of elliptic section). The angle between
the orientations is 45◦ and the lengths, the aspect ratio and the densities are identical. The
permeability is computed for several values of the total density and compared to numer-
ical simulations on explicit discrete fracture networks by means of the software FRACA
developed by IFP. Among many functionalities aiming at analyzing the characteristics
of fractured reservoirs, this software allows to randomly generate fractures according to
length, orientation and conductivity laws and compute an effective permeability accord-
ing to the methodology presented in Bourbiaux et al. (1998), Sarda et al. (2001), Lange
et al. (2002) and Bourbiaux et al. (2002). In the present case, the length, orientation and
conductivity laws are deterministic. The number of generated fractures is computed with
respect to each required density while the length remains constant and the fracture centers
are randomly chosen on a square (see Fig. 10). In FRACA, the fractures are considered as
an infinitesimal space between rectangles, the aspect ratio being therefore defined as the
ratio between the aperture and the length. In order to ensure a relevant comparison between
the upscaling model (elliptic profile of the fractures) and the numerical simulation (rec-
tangular profile), the length of the rectangular profile, expected to have a straightforward

123
Effective Permeability of Media 173

Fig. 10 2D discrete fracture

network

Fig. 11 Permeability of a 2D fracture network of two families—angle between the two directions = 45◦

influence on the percolation threshold, is equal to the major diameter of the corresponding
ellipse. Besides, the aperture is such that both areas are identical, which means that the
aspect ratio of the rectangle profile over the aspect ratio of the corresponding ellipse is
equal to π/4.
The two principal permeabilites provided by the effective medium theory (the com-
putations are quasi-instantaneous on a 2.8 GHz PC) and the numerical simulations (each
cross requires tens of seconds) are plotted on Fig. 11. It appears that both methods give
approximately the same value of percolation threshold (about  = 1.85). Nevertheless, the
effective medium theory provides a linear evolution of *hom beyond the threshold, which
is not expected from the percolation theory (Sahimi 1995) and the numerical computations
here. For higher densities (about twice the threshold value), the theory underestimates the
minimal principal permeability computed from the explicit network but both methods give
concordant estimates of the maximal permeabilities in the chosen density range (not much

123
174 J.-F. Barthélémy

than three times the threshold). However, for much higher densities ( 1), the spacing
between fractures (2) becomes very small compared to the length if the latter remains con-
stant, i.e., the increase of density is only due to an increase of the fracture number. In that
case, the fractures should not be considered as microfractures but long ones with respect to
a lower scale r.v.e. Moreover, one can show on the numerical computations of Fig. 11, that
the slopes of the permeabilities tend toward those corresponding to a network of crosscut-
ting fractures recalling that *hom is then analytically obtained by the second term of (37).

6 Conclusion

The computation of the macroscopic permeability of a fractured medium is not an easy task.
Indeed, natural networks may show spatial correlations preventing the definition of a r.v.e.
on which effective properties could be determined. Nevertheless, in the case of a random
fractured medium such that a r.v.e. exists, an estimate of the macroscopic permeability is
derived in this article. Two kinds of fractures have been considered: fractures crosscutting
the r.v.e. and microfractures. The determination of the macroscopic permeability of such a
medium with several families has been proposed by means of the self-consistent scheme. The
application of the latter relies first on the substitution of the fractures by equivalent Darcy
media with a tangential permeability deduced from the solution of a Poiseuille flow between
two planes and an undefined normal permeability. An important step of this scheme consists
in computing the second-order Hill tensor. The algorithm leading to the expression of this
tensor has been provided for any anisotropic reference medium and any ellipsoidal inclusion.
Implementing then the self-consistent scheme, it has been shown that the relevance of the
fictitious normal permeability depends on the fracture type: for fractures crosscutting the
r.v.e., any value can be taken whereas for microfractures, the same value as the tangential
permeability may be chosen not to artificially prevent connections between fractures. Results
have finally been provided in the 2D and 3D isotropic and anisotropic cases. The effective
permeability of a 2D anisotropic network has been compared to numerical simulations with
a good agreement. The need to take carefully into account the characteristic length of each
fracture family has also been put in evidence, being understood that all the phases taking
part to a given upscling process have all their characteristic length of the same order of
magnitude. If the density is of the order of 1 or less, the characteristic length is the fracture
length and the family is considered as a microfracture family whereas if the density is large,
the characteristic length is the spacing and the family is crosscutting the r.v.e.

Acknowledgements The author gratefully acknowledges fruitful discussions with Jean-Marc Daniel, Marie-
Christine Cacas and Martin Guiton (IFP).

Appendix: Hill Tensor in the 3D Anisotropic Case

This paragraph aims at providing the expression of the Hill tensor /Io related to an ellipsoid
I of any shape characterised by the second-order tensor ' (16) embedded in an infinite
medium of any anisotropic permeability *o .
Let us first introduce the Green function G o for the reference permeability *o , i.e., the
function satisfying:


divx *o · gradx G o = δ0 and lim G o (x) = 0 (65)
x →∞

123
Effective Permeability of Media 175

where δ0 denotes the Dirac distribution at point 0 of R3 . It can be shown that the Hill tensor
/Io can write by means of G o (Dormieux et al. 2006):
⎛ ⎞
 

 ∂ 2

/Io (x) = hessx G o x − x  dx  = ⎝ G o x − x  dx  ⎠ ei ⊗ e j (66)
∂ xi ∂ x j
I I

where the integration is performed with respect to the variable x  whereas the differentiations
concern the variable x. It will be shown in the following that /Io (x) is uniform over the
ellipsoid I , which constitutes the transposition of Eshelby’s fundamental result (Eshelby
1957) to the second-order tensor problem class.
The determination of /Io (66) requires to solve (65). To this end, it is first convenient to
simplify (65) by introducing the orthogonal tensor 0 diagonalizing *o
3
t
0 · 0 = ? and t
0 · *o · 0 = # = K oj e j ⊗ e j (67)
j=1

the following tensors

3 
#1/2 = K oj e j ⊗ e j and 1 = 0 · #1/2 (68)
j=1

as well as the new variable of R3

⎛ ⎞
3
1
y = 1−1 · x = #−1/2 · t 0 · x = ⎝  ej ⊗ ej⎠ · t0 · x (69)
j=1 K oj

−1
where the obvious notation #−1/2 = #1/2 has been used.
Using the change of variable (69), the left hand side of the first equation of (65) writes:


divx *o · gradx G o = *o : hessx G o


= #−1/2 · t 0 · *o · 0 · #−1/2 : hessy G o
= ? : hessy G o
= yG
o
(70)

In addition, a classical result of the theory of distributions allows the same change of variable
on the right hand side:
1 1
δ0 (x) = δ0 (1 · y) = δ0 (y) = √ δ0 (y) (71)
det 1 det *o
Finally, the first equation of (65) write:
1
yG
o
= √ δ0 (y) (72)
det *o
The solution of (72) satisfying the remote condition (65) is known:
1 1
G o (x = 1 · y) = − √ =− √  (73)
4π det *o y 4 π det * x · *o−1 · x
o

123
176 J.-F. Barthélémy

Introducing (73) in (66), it comes:


1 1
/Io (x) =− √ hessx  dx  (74)
4 π det *o (x − x  ) · *o−1 · (x − x  )
I

or, with x = 1 · y and x  = 1 · y  :

⎛ ⎞

1 t −1 1
/Io (x = 1 · y) = − 1 · ⎝hessy d y  ⎠ · 1−1 (75)
4π y − y
J

where the integration is performed over the ellipsoid J of equation:

−1 t
' · ' · y ≤ 1 with ' = ' · 1−1 = ' · 0 · #−1/2
t
y∈J ⇔y· (76)

The classical newtonian potential is recognized in (75):


1 1 t
 (y) = 
d y  ⇒ /Io (x = 1 · y) = − 1−1 · hessy  (y) · 1−1
4π y−y
J
(77)

The potential  can simply be expressed by means of the eigenvectors and eigenvalues of
the characteristic tensor t ' · ' of J . Let us then introduce the orthogonal tensor 2 such
that:

2 · ' · ' · 2 = A2 e1 ⊗ e1 + B 2 e2 ⊗ e2 + C 2 e3 ⊗ e3 (A ≥ B ≥ C)
t
t
2 · 2 = ? and t

(78)

Using these notations, it can be shown that the potential (77) is a quadratic form of y for
y ∈ J , i.e., x ∈ I (Kellogg 1929; Eshelby 1957):
1
2  1
∀y ∈ J ,  (y) = A I A + B 2 I B + C 2 IC − y · 2 · # I · t 2 · y (79)
2 2
where # I , I A , I B and IC are defined with respect to A, B and C in (20)–(29).
Finally, the homogeneity of /Io within I is proved by introducing (79) in (77):
t
/Io = 1−1 · 2 · # I · t 2 · 1−1 (80)

References

Berkowitz, B.: Analysis of fracture network connectivity using percolation theory. Math. Geol. 27(4), 467–
483 (1995)
Bonnet, E., Bour, O., Odling, N., Main, I., Berkowitz, B., Davy, P., Cowie, P.: Scaling of fracture systems in
geological media. Rev. Geophys. 39(3), 347–383 (2001)
Bour, O., Davy, P.: Connectivity of random fault networks following a power law fault length distribution. Water
Resour. Res. 33(7), 1567–1583 (1997)
Bourbiaux, B., Cacas, M.-C., Sarda, S., Sabathier, J.-C.: A rapid and efficient methodology to convert fractured
reservoir images into a dual-porosity model. Oil Gas Sci. Technol. Rev. IFP 53(06), 785–799 (1998)
Bourbiaux, B., Basquet, R., Cacas, M.-C., Daniel, J.-M.: An integrated workflow to account for multi-scale
fractures in reservoir simulation models: implementation and benefits. In: 10th International Petroleum
Exhibition and Conference, Abu Dhabi. SPE, Reservoir simulation (2002)
Budiansky, B., O’Connell, R.: Elastic moduli of a cracked solid. Int. J. Solids Struct. 12, 81–97 (1976)

123
Effective Permeability of Media 177

Cacas, M.-C., Daniel, J.-M., Letouzey, J.: Nested geological modelling of naturally fractured reservoirs. Petrol.
Geosci. 7, S43–S52 (2001)
Cosentino, L., Coury, Y., Daniel, J.-M., Manceau, E., Ravenne, C., Van Lingen, P., Cole, J., Sengul, M.:
Integrated study of a fractured middle east reservoir with stratiform super-K intervals—part2: Upscaling
and dual media simulation. In: SPE Middle East Oil Show, Bahrain. SPE (2001)
Darcel, C.: Corrélations dans les réseaux de fractures: caractérisation et conséquences sur les propriétés
hydrauliques. PhD thesis, Université Rennes (2002)
Darcel, C., Bour, O., Davy, P., de Dreuzy, J.-R.: Connectivity properties of two-dimensional fracture networks
with stochastic fractal correlation. Water Resour. Res. 39(10), SBH1.1–SBH1.13 (2003)
de Dreuzy, J.-R., Davy, P., Bour, O.: Hydraulic properties of two-dimensional random fracture net-
works following a power law length distribution: 1-Effective connectivity. Water Resour. Res. 37(8),
2065–2078 (2001)
de Dreuzy, J.-R., Davy, P., Bour, O.: Hydraulic properties of two-dimensional random fracture networks
following a power law length distribution: 2-Permeability. Water Resour. Res. 37(8), 2079–2095 (2001)
de Galard, J.-H., Zoormand, G., Ghanizadeh, M., Daltaban, S., Camus, D.: A case study on redevelopment
of a giant highly fractured carbonate reservoir in iran based on integrated reservoir characterization and
3D modeling studies. In: SPE Middle East Oil & Gas Show and Conference, Bahrain. SPE (2005)
Dormieux, L., Kondo, D.: Approche micromécanique du couplage perméabilité-endommagement. C. R.
Mécanique 332, 135–140 (2004)
Dormieux, L., Kondo, D., Ulm, F.: Microporomechanics. Wiley (2006)
Ene, H., Sanchez-Palencia, E.: Équations et phénoménes de surface pour l’écoulement dans un modéle de
milieu poreux. Journal de Mécanique 14(1), 73–108 (1975)
Eshelby, J.: The determination of the elastic field of an ellipsoidal inclusion, and related problems. Proc. Roy.
Soc. Lond. 241, 376–396 (1957)
Fokker, P.: General anisotropic effective medium theory for the effective permeability of heterogeneous reser-
voirs. Transp. Porous Media 44, 205–218 (2001)
Garboczi, E., Snyder, K., Douglas, J., Thorpe, M.: Geometrical percolation threshold of overlapping ellip-
soids. Phys. Rev. E 52(1), 819–828 (1995)
Garcia, M., Gouth, F., Gosselin, O.: Fast and efficient modeling and conditioning of naturally fractured reservoir
models using static and dynamic data. In: SPE Europec/EAGE Conference and Exhibition, Paper 107525,
London (2007)
Guéguen, Y., Palciauskas, V.: Introduction to Physics of Rocks. Princeton Univ. Press (1994)
Howard, J., Nolen-Hoeksema, R.: Description of natural fracture systems for quantitative use in petroleum
geology. AAPG Bull. 74(2), 151–162 (1990)
Huang, J., Liu, H.: On a flat ellipsoidal inclusion or crack in three-dimensional anisotropic media. Int. J. Eng.
Sci. 36(2), 143–155 (1998)
Jenni, S., Hu, L., Basquet, R., de Marsily, G., Bourbiaux, B.: History matching of a stochastic model of
field-scale fractures: methodology and case study. Oil Gas Sci. Technol. Rev. IFP 62(2), 265–276 (2007)
Kellogg, O.: Potential Theory. Springer-Verlag, Berlin (1929)
Lange, A., Basquet, R., Bourbiaux, B.: Hydraulic characterization of faults and fractures using a dual medium
discrete fracture network simulator. In: 10th International Petroleum Exhibition and Conference, Abu
Dhabi. SPE (2002)
Oda, M.: An equivalent continuum model for coupled stress and fluid flow analysis in jointed rock masses. Wa-
ter Resour. Res. 22(13), 1845–1856 (1986)
Pouya, A.: Tenseurs de perméabilité équivalente d’un domaine hétérogène fini. C. R. Géoscience 337, 581–
588 (2005)
Pozdniakov, S., Tsang, C.-F.: A Self-Consistent Approach for Calculating the Effective Hydraulic Conductivity
of a Bimodal, Heterogeneous Medium. Lawrence Berkeley National Laboratory (2004)
Robinson, P.: Connectivity of fracture systems—a percolation theory approach. J. Phys. A Math.
Gen. 16(3), 605–614 (1983)
Sabathier, J.-C., Bourbiaux, B., Cacas, M.-C., Sarda, S.: A new approach of fractured reservoirs. In: Interna-
tional Petroleum Conference and Exhibition, Paper 39825, Mexico. SPE (1998)
Sahimi, M.: Flow and Transport in Porous Media and Fractured Rock. VCH (1995)
Sarda, S., Jeannin, L., Bourbiaux, B.: Hydraulic characterization of fractured reservoirs: simulation on dis-
crete fracture models. In: Symposium of the Society of Petroleum Engineers, Houston. SPE, Reservoir
simulation (2001)
Verga, F., Giglio, G., Masserano, F., Ruvo L.: Calibration of fractured reservoirs with dynamic data. In: SPE
Reservoir Simulation Symposium, Paper 66395, Houston (2001)
Zaoui, A.: Continuum micromechanics: survey. JEM 128(8), 808–816 (2002)

123
178 J.-F. Barthélémy

Zhu, Q.: Applications des approches d’homogénéisation à la modélisation tridimensionnelle de

l’endommagement des matériaux quasi-fragiles: formulations, validations et implémentations
numériques. Thèse de doctorat, UniversitT des Sciences et Technologies de Lille (2006)
Zimmerman, R.: Effective conductivity of a two-dimensional medium containing elliptical inhomo-
geneities. Proc. Roy. Soc. Ser. A 452(1950), 1713–1727 (1996)

123