You are on page 1of 12

Comput Geosci (2010) 14:527538

DOI 10.1007/s10596-009-9168-4
ORIGINAL PAPER
Simulation of multiphase ow in fractured reservoirs
using a fracture-only model with transfer functions
Evren Unsal Stephan K. Matthi Martin J. Blunt
Received: 17 April 2008 / Accepted: 5 October 2009 / Published online: 13 November 2009
Springer Science + Business Media B.V. 2009
Abstract We present a fracture-only reservoir simula-
tor for multiphase ow: the fracture geometry is mod-
eled explicitly, while uid movement between fracture
and matrix is accommodated using empirical transfer
functions. This is a hybrid between discrete fracture dis-
crete matrix modeling where both the fracture and ma-
trix are gridded and dual-porosity or dual-permeability
simulation where both fracture and matrix continua
are upscaled. The advantage of this approach is that
the complex fracture geometry that controls the main
ow paths is retained. The use of transfer functions,
however, simplies meshing and makes the simula-
tion method considerably more efcient than discrete
fracture discrete matrix models. The transfer functions
accommodate capillary- and gravity-mediated ow be-
tween fracture and matrix and have been shown to
be accurate for simple fracture geometries, capturing
both the early- and late-time average behavior. We
verify our simulator by comparing its predictions with
simulation results where the fracture and matrix are
explicitly modeled. We then show the utility of the ap-
proach by simulating multiphase ow in a geologically
E. Unsal S. K. Matthi M. J. Blunt (B)
Department of Earth Science and Engineering,
Imperial College London, London, UK
e-mail: m.blunt@imperial.ac.uk
Present Address:
E. Unsal
Schlumberger Cambridge Research,
Cambridge, CB3 0EL, UK
Present Address:
S. K. Matthi
Department of Mineral Resources and Petroleum Eng.,
Montan University of Leoben, Leoben, Austria
realistic fracture network. Waterooding runs reveal
the fraction of the fracturematrix interface area that is
inltrated by water so that matrix imbibition can occur.
The evolving fraction of the fracturematrix interface
area turns out to be an important characteristic of
any particular fracture system to be used as a scaling
parameter for capillary driven fracturematrix transfer.
Keywords Fractured reservoirs
Countercurrent imbibition Dual porosity
Finite element/nite volume method
1 Introduction
In a fractured reservoir, uids ow primarily through
the interconnected fracture network, while the matrix
stores the vast majority of the oil and water in the
system [17]. The hydraulic properties of a reservoir
are determined by the nature of the fracture network,
while the matrix determines the ultimate recovery of
oil. One of the principal recovery mechanisms is im-
bibition, controlled by capillary forces, where water
enters the matrix from the fractures, displacing oil [16].
This process can either be co- or countercurrent. In
cocurrent imbibition, both phases ow in the same
direction, while in countercurrent imbibition, they ow
in opposite directions. In a fractured system where the
fractures are the main owpath, it is possible to observe
both imbibition modes; the water ows cocurrently with
oil through the fractures, while there is countercurrent
imbibition between the fracture and the matrix, as
shown schematically in Fig. 1. Countercurrent imbibi-
tion is a much slower process than cocurrent imbibition
528 Comput Geosci (2010) 14:527538
Fig. 1 Countercurrent imbibition between the rock matrix and
fracture. Due to capillary forces, as soon as injected water invades
a fracture, water (shown hatched) begins to imbibe the oil satu-
rated matrix displacing oil into the fractures
[21], hence is possible for water to displace oil in the
matrix cocurrently with ow parallel to the direction of
the fracture.
The most physically realistic and computationally
accurate way to model ow and transport in fractured
media is by using a discrete fracture/discrete matrix
(DFDM) approach where both fracture and matrix are
gridded explicitly [7, 8, 10, 13, 14, 19]. However, cap-
turing displacement in a geologically realistic fracture
network requires a nely resolved grid and intricate
indirect discretization approaches. Simulations on such
meshes require very large amounts of time.
A common alternative is to consider the reservoir
as containing two interacting media: the fractures that
carry the ow connected to a relatively stagnant matrix.
The geometry of fracture and matrix is not represented
in any detail; instead, they are replaced by averaged
properties in a regularized (grid-block) representation
of the eld. This is the dual-porosity approach that is
universally used for eld-scale ow modeling of frac-
tured reservoirs [1, 6, 8, 9, 24]. The uid exchange
between the two media is accommodated using a trans-
fer function. This method works well if the fractures
are interconnected and the permeability contrast be-
tween the matrix and the fractures is very high. This
method is computationally much less expensive. How-
ever, it makes three main approximations. The rst is
that there is no ow in the matrix. This can be over-
come using a dual-permeability model [4, 5, 23] that
allows ow between matrix blocks. However, in many
cases, it is appropriate to consider that the majority
of the ow is conned to the fractures. The second
approximation is to replace the complex interaction
between fracture and matrix mediated by capillary and
gravitational forces by an empirical transfer function
that acts as a source/sink in the governing transport
equations (see the next section). While the original
formulation for the transfer function was based on uid
expansion and assumed a pseudosteady-state between
fracture and matrix [9], recent work has led to the
development of transfer functions that do accurately
capture the dynamics of the displacement, albeit for
simple fracture geometries [12]. The thirdand most
seriousapproximation is to use averaged properties
for the fracture network. It is not possible to capture
the full complexity of the connectivity, topology, and
permeability of a network by a simple grid-block aver-
age, even if tensor permeabilities are used.
We propose a hybrid model called a fracture-only
model that captures the advantages of both dual-
porosity and DFDM models. It can also be considered
as a dual-porosity discrete fracture model. The matrix
is treated using a dual-porosity approach, where all
the matrix properties are averaged and the transfer
between fracture and matrix is captured by a transfer
function. Since most of the reservoir volume is ma-
trix, we save signicant computational time by making
this simplication. Fractures, on the other hand, are
modeled using a discrete fracture approach, where all
the fracture heterogeneities are considered. We will as-
sume, however, that all the fractures are interconnected
and there is no viscous ow in the matrix.
In the following section, we introduce the CSMP++
reservoir simulator which we used for our numerical
simulations. A detailed description of the formula-
tion is provided. Next, we present the implementation
of transfer functions into CSMP++, and nally, we
present results in one, two, and three dimensions. The
code is veried against analytical solutions (in one
dimension) and explicit fracturematrix simulations in
two dimensions.
2 Simulation method
2.1 CSMP++
The fracture-only model is implemented using the
CSMP++ reservoir simulator [14]. It uses a combina-
tion of nite element (FEM) and nite volume methods
(FVM) using operator splitting [3, 18]. The geometric
exibility of the nite element method is retained, but
the nite volumes are better suited for the solution
of the hyperbolic part of the equation. Therefore, this
nite elementnite volume method (FEFVM) gives
better results. The FEFVM used in this article involves
an implicit pressure explicit saturation discretization of
time to simulate two-phase fracture ow.
Comput Geosci (2010) 14:527538 529
The new transfer calculations are performed after
each implicit pressure explicit saturation solution cycle
and represent a simple saturation update. We have
augmented the standard CSMP fracture-only code to
include matrix/fracture transfer. In the next sections,
the conceptual model and governing equations used in
the numerical work are presented.
2.2 Conceptual model
We assume that all the ow occurs in an interconnected
fracture network. Only the fractures are gridded and
a lower 2D representation is used; consequently, the
number of grid blocks and computation time is re-
duced by one to two orders of magnitude over DFDM
methods while the complex fracture geometry is still
retained. The fracture geometry is modeled explicitly
using the FEFVM mesh, and each fracture element is
associated with a virtual matrix element via a trans-
fer function (Fig. 2). The number and orientation of
the elements vary depending on the dimensions of
the system and the meshing method used. A virtual
matrix network is also formed; each matrix connects
to a fracture element via the transfer function. In the
governing transport equationssee the next section
we need to assign an effective porosity to fracture and
matrix. Essentially, we need to know the ratio of the
volume of the fracture element to the pore volume of
the matrix block. While it may appear that we have
to divide up the matrix and assign every region in
space to a specic fracture element, in this work, we
chose a simpler approach. We compute the total pore
volume of the matrix and the total pore volume of the
fracture: this is the ratio
m
/
f
, where
f
is the fracture
porosity and
m
is the matrix porosity. This xed value
is then assigned to each fracturematrix element. In
a heterogeneous medium, we could assign different
volume ratios to each element dependent on the local
geometry.
In the fracture network, the pressure and the velocity
distributions are calculated by CSMP++. In the matrix
network, the viscous owvelocity is assumed to be zero,
meaning there is no ow between the matrix elements.
2.3 Governing equations
For a uid phase a in a fracture network (the fracture
system uses no-ow boundary conditions on the faces
of the fractures), the mass balance equation for incom-
pressible ow is given by the following equation [3],

f
S
af
t
= u
a
+ q
a
, a {w, n} (1)
where S is saturation, t is time, u is uid velocity, and
q are uid sources or sink terms including those origi-
nating due to ow from the rock matrix. The subscript
a will refer to either the wetting (water) phase w or
the nonwetting (oil) phase n. Subscript f refers to the
fracture network. The saturations S satisfy,
S
w
+ S
n
= 1. (2)
The uid velocity of phase a is given empirically by
Darcys law
u
a
=
a
k(p
a

a
g) , (3)
where
a
is the mobility and is equal to the ratio of the
relative permeability k
ra
and to the viscosity
a
of phase
a (
a
= k
ra
/
a
). Tensor k represents the permeability
of the porous medium, p
a
is the uid pressure, and
a
is the uid phase density, while g is the gravitational
acceleration. The uid pressure for the wetting and
nonwetting phases are related through the capillary
pressure P
c
,
P
n
= P
w
+ P
c
. (4)
Fig. 2 Fracture-only virtual
matrix discretization of the
method. The matrix is
represented by a series of
virtual blocks
530 Comput Geosci (2010) 14:527538
Capillary effects in the fractures are neglected such that
P
w
= P
n
. Then, Darcys law can be expressed for both
phases as
u
t
=
t
kP +kg (
w

w
+
n

n
) (5)
where u
t
is the total velocity and
t
is the total mobility.
If we assume that the uids and rock are incompress-
ible, the divergence of the ow eld is equal to the total
uid source or sink q
t
= q
w
+q
n
u
t
= q
t
. (6)
Equation 6 can be used to derive an elliptic equation for
pressure. Once the total velocity is known, Eq. 1 can be
solved for saturation. An operator-splitting technique
is used to solve for pressure with the FEM and for
saturation with the FVM.
For this work, a transfer term T is added to Eq. 1:

f
S
af
t
= u
a
+q
a
T
a
. (7)
Since ow is treated as incompressible T
w
+ T
n
= 0,
transfer has no direct effect on the solution for pressure.
The formulation of transfer function will be given in
the next section. We also present the implementation
of relevant transfer functions into the CSMP++ code.
3 Fracture/matrix transfer functions
In this section, we will briey review the dual-porosity
approach, including alternative ways of formulating the
matrix/fracture transfer function.
3.1 Dual-porosity models
In dual-porosity models, the fractured porous medium
is divided into two main regions: (1) fractures, where
the vast majority of ows occur due to their high
permeability, and (2) matrix, which has much lower
permeability but stores the vast majority of the uid.
It is assumed that there is no viscous ow in the matrix.
Fluid movement between the two regions is modeled
using a transfer function. The transfer term is included
in the relevant mass conservation equation for the frac-
tures, Eq. 7, while for the matrix:

m
S
am
t
= T
a
, (8)
where the subscript m represents the matrix.
3.2 Transfer function
In this numerical study, we use two types of transfer
function. The rst one is a relatively simple linear
transfer function from Di Donato and Blunt [2]:
T =
m

_
S

S
wm
_
, (9)
where is a rate constant, S* is the maximum water
saturation reached in the displacement, and S
wm
is the
water saturation in the matrix. We have dropped the
subscript w on T in Eq. 9 for convenience: T
n
= T
w

T. This function represents the late-time behavior of


capillary-controlled displacements accurately, but does
not model the early-time behavior properly [20]. An
improved formulation is given below; however, we will
use the linear form since it allows us to compare with
analytical solutions in one dimension (see Section 3.4).
The generic behavior of recovery for constant frac-
ture conditions is well known: initially, it scales at

t before the advancing front reaches any boundaries


and, approximately, as (1 e
t
) at late time with
some geometry and uid-dependent rate [22, 25].
Lu et al. [12] extended the linear transfer function
further to capture both early and late time behaviors.
They included a Vermuelen-type correction factor B
e
in the transfer function:
T = B
e

m
_
S

S
wm
_
, (10)
where
=
3
wm

om

t
_

wo
J

L
2
_
K
m

m
+
K
m
|
wo
| g

m
L
_
, (11)
B
e
=

wm
S
wim

+|S
wm
S
wim
|
2 max ((S
wm
) , |S
wm
S
wim
|)
, (12)
where
wm
,
om
, and
t
are water, oil, and total mo-
bilities in the matrix, K
m
is the matrix permeability,
J* is the dimensionless capillary pressure,
wo
is the
oil/water interfacial tension, L is the effective distance
through which oil moves to reach the fracture (it is
related to the size of the matrix block), S
wim
is the initial
water saturation in the matrix, and is a dimensionless
convergence factor. We will refer to Eq. 10 as the
nonlinear transfer function.
3.3 Implementation of the transfer functions
into CSMP++
The transfer term, T, in Eq. 7 is implemented using
the appropriate transfer function, Eq. 9 or 10. At each
time step, the oil and water saturations at each node
in the fractures are calculated using Eqs. 16, ignor-
ing transfer. Then, the saturations in the matrix and
Comput Geosci (2010) 14:527538 531
Fig. 3 Flowchart summarizing how the fracture/matrix satura-
tions are calculated
fracture nite volumes are updated using the transfer
function, ignoring advection. If there is no transfer,
the matrix saturation stays unchanged. To determine
whether there should be a transfer or not, the following
condition is tested:
S
of
S
o
, (13)
where S
of
is the current oil saturation in each nite vol-
ume and S
o
is the initial fracture oil saturation. If the
water has reached a certain nite volume via advection,
then this decreases its oil saturation from the initial
value. If this condition is satised, then transfer occurs.
The saturations in the fracture nodes are updated using
the following equations for oil and water, respectively.
S
n
of
= S
n1
of
+
Tt

f
, (14)
S
n
wf
= S
n1
wf

Tt

f
, (15)
where the superscript n refers to the updated satura-
tion at the current time level and n 1 refers to the
saturation updated to accommodate advection only. In
CSMP++, there is a CourantFriedrichsLewy (CFL)
criterion which limits the time steps for the explicit
simulations [15]. However, in the fracture-only model,
the time steps are also limited by the transfer terms,
which are sources/sinks. Especially at early times, when
the transfer rates are high, the time step is chosen to be
sufciently small that during the update, the saturations
never drop lower than 0 or become greater than 1. Once
the transfer terms are smaller, longer time steps are
taken. Finally, the matrix saturations at the correspond-
ing node are updated for oil and water, respectively.
S
n
om
= S
n1
om

Tt

m
, (16)
S
n
wm
= S
n1
wm
+
Tt

m
. (17)
This procedure is repeated for fracture and then matrix
nite volume at each time step. A owchart which
summarizes this process is given in Fig. 3.
3.4 Analytical solution
To verify our numerical scheme with linear transfer
function, we use the time-dependent analytical solution
from Di Donato and Blunt [2]. They constructed 1D
analytical solutions for the fracture water saturation
that is valid at early and late times for the front location
using the linear transfer function. To develop a trans-
port equation in 1Dat early times, they dened the time
of ight (s) as the time taken for a particle to move a
distance x in the fracture.
(t) =

f
u
t
x (t) . (18)
Assuming S
wf
(,t = 0) = 0, S
wf
( = 0,t) = 1 and linear
relative permeabilities such that u
w
= S
w
u
t
. the leading
edge of water front moves at unit speed, meaning that
imbibition has not had sufcient time to reduce fracture
saturation to zero anywhere. Hence, water rst enters
the fracture at time = t, and at that moment, water
transfer from fracture to matrix starts with an exponen-
tially decaying function:
S
wf
(, t) =
_
1 e
(t)
_
H(t ) , (19)
where
=

f
(1 S
omr
S
wmi
) . (20)
and S
omr
is the residual oil saturation in the matrix,
H(t ) is the Heaviside step function: H(t ) = 0 if
t > and H(t ) = 1 if t .
At late times, when the amount of imbibition into
the matrix exceeds the pore volume of the fracture,
the advancing front slows down, and its advance is
given by:
(t) =
1
( +)
2
_
( +) t +
_
1 e
( +)t)
__
. (21)
532 Comput Geosci (2010) 14:527538
Table 1 Parameters and
values for the 1D CSMP++
solution
K
f
is the fracture
permeability
Parameter CSMP++
parameters

f
0.01

m
1
q
i
10
5
m s
1

w
0.001 Pa s

o
0.001 Pa s

w
1,000 kg m
3

o
1,000 kg m
3
k
rwf
S
wf
k
rof
1 S
wf
5.15 10
7
K
f
10
13
m
2
S
wmi
0.03
S
omr
0.03
S* 0.7
4 Results
The transfer term is evaluated by comparing simulation
results to those obtained using equivalent 1D analytical
solutions and 2D discrete fracture simulations in simple
geometries.
4.1 1D simulations
In this section, we compare the 1D numerical solution
with the corresponding analytical solutions for the lin-
ear transfer function. We used a 1D grid with 200 nodes
with an internode distance of 0.005 m (see Table 1 for
parameters used in CSMP++; Table 2 for the analyt-
ical model). The rst nite volume contains the uid
source and the last one the sink. Water is injected at a
constant rate (Fig. 4) while transfer rate constant is
kept constant. The fracture water saturation is plotted
against distance at different times. A time step of 10 s
was used.
Table 2 Parameters and values for the 1D analytical solutions
Parameter Figures 5 and 6 Figure 7

f
0.01 0.01

m
0.2 0.2
q
i
10
5
m s
1
1.8 10
5
m s
1

w
0.0003 Pa s 0.0003 Pa s

o
0.0003 Pa s 0.0003 Pa s

w
1,000 kg m
3
1,000 kg m
3

o
1,000 kg m
3
1,000 kg m
3
k
rwf
S
wf
S
wf
k
rof
1 S
wf
1 S
wf
0 day
1
(Fig. 5); 1.86 10
6
day
1
5.15 10
7
day
1
K
f
10
13
m
2
10
13
m
2
S
wmi
0.03 0.03
S
omr
0.03 0.03
S* 0.7 0.7
Fig. 4 One dimensional fracture geometry
0
0.2
0.4
0.6
0.8
1
0 0.2 0.4 0.6 0.8 1
f
r
a
c
t
u
r
e

w
a
t
e
r

s
a
t
u
r
a
t
i
o
n
distance [m]
10000s 20000s 50000s
numerical
analytical
10000s 20000s 50000s
numerical
analytical
10000s 20000s 50000s
numerical
analytical
Fig. 5 The fracture water saturationdistance prole when there
is no transfer. The analytical solution is piston-like advance
0
0.2
0.4
0.6
0.8
1
0 0.2 0.4 0.6 0.8 1
f
r
a
c
t
u
r
e

w
a
t
e
r

s
a
t
u
r
a
t
i
o
n
distance [m]
10000s 20000s 50000s
numerical
analytical
10000s 20000s 50000s
numerical
analytical
10000s 20000s 50000s
numerical
analytical
10000s 20000s 50000s
numerical
analytical
10000s 20000s 50000s
numerical
analytical
Fig. 6 Comparison of numerical and analytical results at differ-
ent times for 1D ow when there is transfer
Comput Geosci (2010) 14:527538 533
0
0.2
0.4
0.6
0.8
0 20000 40000 60000 80000 100000
f
r
o
n
t

l
o
c
a
t
i
o
n

[
m
]
time [s]
analytical
numerical
Fig. 7 Comparison between analytical and numerical 1D solu-
tions for early and late times
Our rst test case examines ow with no transfer
where the analytical solution is simply a step function
moving at unit dimensionless speed. Figure 5 compares
the analytical and numerical solutions; the smearing of
the front in the CSMP++ implementation is due to the
numerical dispersion associated with the second-order
accurate high-resolution method using the MINMOD
spatial limiter [11].
Figure 6 shows a comparison of numerical and an-
alytical solution, Eq. 19, at early time when there is
transfer. Here, the water saturation in the fracture
decreases with time and distance. The CSMP++results
are in good agreement with the analytical solution.
Figure 7 shows example results at late time long after
the front passed through the right-hand boundary of
the model. In CSMP++, the location of the front is
where the water saturation is rst nonzero. There is a
transition from fast ow of water in the fracture at early
times to a slower ow associated with increasing water
imbibition into the matrix at later timesthe leading
front is slowed down due to complete removal of the
water into the matrix by imbibition.
4.2 2D model
For a 2D single fracture model, solutions for two-phase
ow are compared to a commercial nite difference
reservoir simulator, ECLIPSE (Eclipse 100, Black Oil
Model). In the CSMP++ model, the fracture is repre-
sented as a plane with dimensions of 10 m length and
Fig. 9 Regular grid of 2D fracture for comparison with commer-
cial nite difference simulator and associated virtual matrix. The
fracture plane is at the top; the matrix block is at the bottom. Each
fracture nite volume surrounding each nite element node is
connected a matrix nite volume via a transfer function T. For
simplicity, the connections are shown only for the rst row of
elements
1 m width (Fig. 8). Each fracture element is assigned a
thickness, W
f
, of 1 cm. The grid has 200 nite elements
in the X direction and ve elements in the Y direction,
1,000 elements in total. Initially, the fracture is fully
oil saturated, and water is injected from the left-side
boundary.
Each fracture element is associated with a virtual
matrix element via the transfer function as illustrated
in Fig. 9. Each virtual matrix element has a cross-
section equal to the fracture element cross-section it
is connected to and also has a thickness, W
m
, which is
used in the denition of Eq. 11. The matrix thickness
W
m
is 0.42 m. The relative permeability of the fracture
is assumed to be a linear function of saturation, and the
capillary pressure in the fracture is zero.
For the nite difference model, a total of 9,800 grid
blocks were used in the simulation, with smaller blocks
near the inlet to capture accurately the initial advance
of water into the matrix (Fig. 10). The fracture has 200
grid blocks, and the matrix has 48 grid blocks along the
fracture (Table 3). All other faces in contact with the
fracture were closed off (no ow boundaries). Again,
the relative permeability of the fractures was treated
as a linear function of saturation, and the capillary
pressure in the fracture was set to zero; the matrix
capillary pressure is given in Table 4. The fracture was
dened explicitly as high permeability region with a
Fig. 8 Regular grid used to
represent a 2D fracture
(dimensions 10 1 m)
534 Comput Geosci (2010) 14:527538
Fig. 10 Finite difference grid
system for 2D simulations of
countercurrent imbibition. A
total of 9,800 grid blocks were
used
porosity of 1. Quadratic relative permeabilities were
dened in the matrix (Table 4).
Water was injected at a constant rate via an injection
well into the rst grid block of the fracture, and oil
was produced at a constant rate from a producing
well which was situated in the last fracture grid block
(Fig. 10). The uid, matrix, and fracture medium prop-
erties are listed in Table 4. The matrix permeability was
set to 1 mD (10
15
m
2
) to ensure a huge disparity in
permeability between fracture and matrix. A time step
of 10 s was used.
4.2.1 Analysis of results
2D CSMP++ results are now compared with nite
difference simulations. In the CSMP++ model, non-
linear transfer functions were used (Eqs. 1012). The
following equations were used to dene fracture and
matrix porosities in the dual-porosity model:

f
=
W
f
W
f
+ W
m
(22)

m
=

mm
W
m
W
f
+ W
m
(23)
where W is the thickness and
mm
is the matrix porosity
in the nite difference model.
In the rst test, we considered just imbibition with
no ow in the fractures. This was a test to determine
if the nonlinear transfer function accurately captured
Table 3 Dimensions of the 2D nite difference model
Dimension Unit Fracture Matrix
X m 10 10
Y m 0.01 0.42
Z m 1 1
the transfer of water into the matrix. For this case, the
fracture was represented as one big water saturated
block in the nite difference simulationsthe fracture
acted as a water reservoir. There was no additional
water injection. Water imbibed in the matrix only due
to spontaneous imbibition. In CSMP++, similar condi-
tions were created. The fracture was kept fully water
saturated. Figure 11 shows a comparison between the
nite difference code and CSMP++ using a best-t
value of 9 10
7
s
1
. Both the early- and late-time
imbibition behavior is accurately captured.
Now that the value was determined, the simu-
lations were run with water owing in the fracture.
Figures 12 and 13 compare the matrix and fracture
saturation proles in CSMP++and the nite difference
code at different times. The agreement was close for
both matrix and fracture.
4.3 3D model
Having validated the fracture-only model for simple
one- and 2D examples, we apply it to a geologically rep-
resentative 3D system. An example of a 3D fracture-
only geometry is shown in Fig. 14. It consists of about 30
interconnected fractures; the fractures are represented
as 2Dsurfaces; however, all assigned a thickness of 1 cm
in the simulations. The thickness is the same for each
nite volume. Only the fractures are gridded; there is
no matrix. The fractures have an initial oil saturation
of 1 (S
of
). There is a water source in one of the frac-
tures (green circle) at the top corner and a producer
at the bottom corner. It can be seen how as the time
progresses the water imbibes through the fractures.
The injection rate is 1.2 10
7
m
3
s
1
. The matrix
and fracture porosities are 0.2 and 1, respectively. The
transfer rate constant, , is set to 9 10
7
s
1
. The time
step was 10 s.
Comput Geosci (2010) 14:527538 535
Table 4 Data used in 2D
simulations
Parameter Unit Value
Fracture porosity,
f
Fraction 0.023
Matrix porosity,
m
Fraction 0.2
Fracture permeability, K
f
m
2
10
10
Matrix permeability, K
m
m
2
10
15
Initial matrix water saturation, S
wmi
Fraction 0
Water density,
w
kg m
3
1,000
Oil density,
o
kg m
3
1,000
Water viscosity,
w
Pa s 1 10
3
Oil viscosity,
o
Pa s 1 10
3
Fracture water relative permeability K
fw
= S
w
Fracture oil relative permeability K
fo
= (1 S
w
)
Fracture capillary pressure P
c
= 0
Matrix water relative permeability K
mw
= S
2
w
Matrix oil relative permeability K
mo
= (1 S
w
)
2
Matrix capillary pressure P
c
=
_
m
Km

om
_
S
n
wm
S
n
_
n = 0.17
Figure 15 shows the average matrix saturation across
the whole system. As water contacts more of the frac-
tures, there is a greater surface area for transfer, and
the imbibition rate increases rapidly. After around
1,000 days, however, the fraction of the fracture net-
work contacted by water increases more slowly, as wa-
ter is simply being cycled through the more conductive
portion of the network. The recovery rate decreases, as
imbibition reaches its late-time limit in those regions of
the matrix contacted by water-saturated fractures. In a
eld-scale model, we would need to nd an empirical
formulation to capture this average behavior; the whole
system is approximately the size of a single grid block in
a eld-scale simulation: note that the overall recovery
is dissimilar to that observed locally for a single matrix
block (Fig. 11).
0
0.2
0.4
0.6
0.8
1
0 2e+006 4e+006 6e+006 8e+006
m
a
t
r
i
x

w
a
t
e
r

s
a
t
u
r
a
t
i
o
n
time [s]
Finite difference
CSMP++
Fig. 11 Transfer rate comparison between nite difference simu-
lator and CSMP++ model using transfer functions. Matrix water
saturation when there is no fracture ow, is determined as 9
10
7
s
1
More realistic fracture geometries are also tested.
Figure 16 shows a fracture system (6 6 3 km) from
the Clair Field which is located in the North Sea, 75 km
west of Shetland Islands. The reservoir is made up of
fractured sandstones; current interpretations suggest as
total volume of oil in place in excess of 410 million
metric tonnes of oil. Water is injected from the source
which is located in the center of the geometry. The
initial fracture water saturation (S
wf
) is 0. Water moves
through the fractures toward the sink, choosing a path
through the better connected, higher permeability por-
tions of the network.
It takes less time to run this simulation compared to
when the matrix is gridded; in these examples, it was
not possible to run a simulation including the matrix,
but similar runs have taken around 3 months of CPU
time [14] compared to 1 week for the Clair model.
0
0.03
0.06
0.09
0.12
0.15
0 2 4 6 8 10
m
a
t
r
i
x

w
a
t
e
r

s
a
t
u
r
a
t
i
o
n
distance [m]
0.6 day-CSMP++
1 day-CSMP++
0.6 day-Finite difference
1 day-Finite difference
Fig. 12 Comparison of matrix water saturation proles from
CSMP++ and nite difference simulations
536 Comput Geosci (2010) 14:527538
0
0.2
0.4
0.6
0.8
1
0 2 4 6 8 10
f
r
a
c
t
u
r
e

w
a
t
e
r

s
a
t
u
r
a
t
i
o
n
distance [m]
0.6 day-CSMP++
1 day-CSMP++
0.6 day-Finite difference
1 day-Finite difference
Fig. 13 Comparison of fracture water saturation proles in
CSMP++ and nite difference simulation
5 Conclusions and future work
We have presented a new extended DFN model of
fracture transport combined with capillary uid ex-
change with a virtual matrix. It is a combination of
discrete fracture and a dual-porosity model retaining
the realistic ow eld of DFDM. Its domain of ap-
plicability is naturally fractured reservoirs with a strong
0
0.2
0.4
0.6
0.8
1
0 1000 2000 3000 4000 5000 6000
a
v
e
r
a
g
e

m
a
t
r
i
x

w
a
t
e
r

s
a
t
u
r
a
t
i
o
n
time [day]
Fig. 15 Average water matrix saturation across the whole
fracturematrix system
inuence of the fractures on the ow and matrix with
low permeability. The rock matrix is treated using
an effective medium representation where all matrix
properties are averaged and acts as the uid reservoir.
Exchange of uid between fracture and the matrix is
captured by a transfer function with a parameterization
that can be customized for each fracture cell-virtual
matrix block couple. This ability is used to set transfer
Fig. 14 3D 30-fracture model
at different time steps, a
0.1 days, b 20 days, c 200 days,
and d 5,475 days. Green
indicates the movement of
injected water along the oil
saturated fractures. The size
of the system is 50 50
1 m. The fractures have a
height of 1 m and aperture of
1 mm; is 9 10
7
s
1
Comput Geosci (2010) 14:527538 537
Fig. 16 An example of a 3D geometry of a fracture system from
Clair Field. The system dimensions are 6 km 6 km 3 km
initiation times to the time of the water front arrival
in each fracturematrix cell pair. We used both linear
and nonlinear transfer functions in this approach which
circumvents the need to grid the matrix via use of
transfer functions and thus represents an attractive,
more efcient simulation technique as compared with
DFDMs.
We tested the model for simple one- and 2D geome-
tries. 1D results for fracture water saturation versus
distance at different times were successfully compared
against analytical solutions from Di Donato and Blunt
[2], validating the simulator. 2D numerical model re-
sults were compared with those from a nite difference
reservoir simulator. The transfer function accurately
captures the dynamics of the matrix saturation changes.
Proof of concept, 3D results indicate that bulk re-
covery from the rock matrix of a naturally fractured
reservoir is controlled by both the local transfer rate
and the fraction of the fracture surface area contacted
by injected water. Future work will study this behavior
for a suite of geologically representative models to
develop eld-scale dual-porosity models with an up-
scaled transfer function that accommodates the effects
of fracture contact area in the transfer function.
Acknowledgements The authors would like to thank the DTI,
EPSRC, BP, Petro-Canada, Total, ExxonMobil, and Chevron for
nancial support, BP for generously providing the Clair Field
data, Virginie Mucha for her contributions in building the Clair
Model, and Hamid Maghami Nick for his help during the simula-
tions. This work was carried out as part of the project Improved
simulation of ow in fractured reservoirs.
References
1. Barenblatt, G.I., Zheltov, I.P., Kochina, I.N.: Basic concepts
in the theory of seepage of homogeneous liquids in ssured
rocks. J. Appl. Math. Mech., Eng. Trans. 24, 12861303 (1960)
2. Di Donato, G., Blunt, M.J.: Streamline-based dual-porosity
simulation of reactive transport and ow in fractured reser-
voirs. Water Resour. Res. 40, W04203 (2004)
3. Geiger, S., Roberts, S., Matthai, S.K., Zoppou, C., Burri, A.:
Combining nite element and nite volume methods for ef-
cient multiphase ow simulations in highly heterogeneous
and structurally complex geologic media. Geouids 4, 284
299 (2004)
4. Gerke, H.H., Van Genuchten, M.T.: A dual-porosity model
for simulating the preferential movement of water and
solutes in structured porous media. Water Resour. J. 29, 305
319 (1993)
5. Gerke, H.H., Van Genuchten, M.T.: Evaluation of a rst-
order water transfer term for variably saturated dual porosity
models. Water Resour. Res. 29, 12251238 (1993)
6. Gilman, J.R., Kazemi, H.: Improvement in simulation of nat-
urally fractured reservoirs. SPE J. 23, 695707 (1983)
7. Hotetot, H., Firoozabadi, A.: An efcient model for incom-
pressible two-phase ow in fractured media. Adv. Water Re-
sour. 31, 891905 (2008)
8. Karimi-Fard, M., Durlofsky, L.J., Aziz, K.: An efcient
discrete-fracture model applicable for general purpose reser-
voir simulators. SPE J. 9, 227236 (2004)
9. Kazemi, H., Merrill, L.S., Portereld, K.L., Zeman, P.R.: Nu-
merical simulation of wateroil ow in naturally fractured
reservoirs. SPE J. 16, 318326 (1976)
10. Kim, J.-G., Deo, M.D.: Finite-element, discrete-fracture
model for multiphase ow in porous media. Alche J. 46(6),
11201130 (2000)
11. LeVeque, R. J.: Finite Volume Methods for Hyperbolic
Problems. Cambridge Texts in Applied Mathematics, p. 558.
Cambridge University Press, Cambridge
12. Lu, H., Di Donato, G., Blunt, M.J.: General transfer functions
for multiphase ow in fractured reservoirs. SPE J. 13(3), 289
297 (2008)
13. Matthi, S.K., Mezentsev, A., and Belayneh, M.: Finite-
element node-centered nite-volume experiments with
fractured rock represented by unstructured hybrid ele-
ment meshes. SPE Reserv. Evalu. Eng. 10(6), 740756
(2007)
14. Matthi, S.K., Geiger, S., Roberts, S.G., Paluszny, A.,
Belayneh, M., Burri, A., Mezentsez, A., Lu, H., Coumou, D.,
Driesner, T., Heinrich, C.A.: Numerical Simulation of Multi-
phase Fluid Flow in Structurally Complex Reservoirs. Special
Publications, 292, pp. 405429. Geological Society, London
(2007)
15. Matthi, S.K., Nick, H.M., Pain, C., Neuweiler, I.: Simulation
of solute transport through fractured rock: a higher-order ac-
curate nite-element nite-volume method permitting large
time steps. Trans. Porous Media (2009). doi:10.1007/s11242-
009-9440-z
16. Morrow, M.R., Mason, G.: Recovery of oil by spontaneous
imbibition. Curr. Opin. Colloid Interface Sci. 6, 321337
(2001)
17. Nelson, R.A.: Geologic Analysis of Naturally Fractured
Reservoirs, p. 332. Gulf Professional, Boston (1987/2001)
18. Paluszny, A., Matthai, S.K., Hohmeyer, M.: Hybrid nite
element-nite volume discretization of complex geologic
structures and a new simulation workow demonstrated on
fractured rocks. Geouids 7(2), 186208 (2007)
538 Comput Geosci (2010) 14:527538
19. Reichenberger, V., Jakobs, H., Bastian, P., Helmig, R.: A
mixed-dimensional nite volume method for multiphase ow
in fractured porous media. Adv. Water Res. 29, 10301036
(2006)
20. Tavassoli, Z., Zimmerman, R.W., Blunt, M.J.: Analytic analy-
sis for oil recovery during counter-current imbibition in
strongly water-wet systems. Trans. Porous Media 58, 173189
(2005)
21. Unsal, E., Mason, G., Morrow, N.R., Ruth, D.W.: Co-current
and counter-current imbibition in independent tubes of non-
axisymmetric geometry. J. Colloid Interface Sci. 306, 105117
(2007)
22. Vermuelen, T.: Theory of irreversible and constant-pattern
solid diffusion. Ind. Chem. Eng. 45, 16641670 (1953)
23. Vogel, T., Gerke, H.H., Zhang, R., Van Genuchten, M.T.:
Modeling ow and transport in a two-dimensional dual-
permeability system with spatially variable hydraulic proper-
ties. J. Hydrol. 238, 7889 (2000)
24. Warren, J.E., Root, P.J.: The behavior of naturally fractured
reservoirs. SPE J. 3, 245255 (1963)
25. Zimmerman, R.W., Chen, G., Hadgu, T., Bodvarsson, G.S.:
A numerical dual-porosity model with semianalytical treat-
ment of fracture/matrix ow. Water Resour. Res. 29(7),
21272137 (1993)

You might also like