Professional Documents
Culture Documents
DOI 10.1007/s11242-005-1398-x
Springer 2006
Nomenclature
decline constant 1/[T ]
matrix surface area [L2 ]
total matrix compressibility [LT 2 /M]
time domain function
laplace domain function
matrix block thickness [L]
porous media permeability [L2 ]
modied Bessel function of rst kind
a
A
cm
f
F
hm
k
I0
52
pf
pm
pm
pD
Pm
Pm
Qm
r
Rm
s
s
t
tD
Vm
x
Greek
letters
dimensionless decline constant
uid viscosity [M/LT ]
porosity
euler constant 0.5772156
uid density [M/L3 ]
shape factor constant [1/L2 ]
matrix hydraulic diffusivity, L2 /T
I1
Subscripts
f
fracture
m
matrix
t
total
1. Introduction
Based on Barenblatt et al. (1960) work on uid ow in fractured media, Warren
and Root (1963) introduced the double porosity concept into petroleum engineering. In their model, a fractured medium comprises two overlapping media:
matrix and fracture. Matrix has a low permeability and a high uid storage. In
contrast, fracture has a high permeability and a low storage.
In one form of double porosity models, matrix blocks act as a source or
sink for a fracture system, where the uid transfer between a matrix block
and fracture system is proportional to the difference between fracture pressure and average matrix block pressure as given by the following equation:
Qm =
km
(pm pf ),
(1)
53
where Qm is the rate of the uid transfer between the matrix and fracture, km the matrix permeability, the uid viscosity, the uid density,
pm represents the average matrix block pressure, pf is the fracture pressure,
and is called matrix-fracture transfer shape factor and has dimensions of
L2 . The model based on this assumption is called a pseudo-steady state
(PSS) transfer model. The matrix-fracture transfer term, Qm is related to
the matrix pressure by the following relationship:
pm
Qm = m cm
,
(2)
t
where m is the matrix porosity and cm is the matrix compressibility. The
traditional double porosity model of Warren and Root uses Equation (1) to
model the uid transfer between the matrix and the fracture in a naturally
fractured reservoir.
This type of model does not account for the pressure transient within
the matrix. There is another model for uid exchange between the matrix
and fracture, which accounts for the pressure transient in the matrix block
by solving the following equation
1 pm
2
pm =
,
(3)
m t
where m = km /cm is the matrix hydraulic diffusivity. This type of model
is called a transient transfer model and does not use the matrix-fracture transfer shape factor. In the transient model, the uid transfer rate
between the matrix and fracture is proportional to the pressure gradient at
the matrix block surface as given by Nanba (1991)
Qm =
Akm
pm |matrix
Vm
face ,
(4)
where A and Vm are the matrix block surface area and matrix block volume, respectively. Petroleum engineering literature shows that the shape
factor remains a controversial topic. A large body of research in the
area of naturally fractured reservoirs simulation is devoted to representing
an accurate matrix fracture exchange term (Kazemi et al., 1976; Thomas
et al., 1983; Kazemi and Gilman, 1993; Lim and Aziz, 1995; Quintard and
Whittaker, 1996; Noetinger and Estebenet, 1998; Bourbiaux et al., 1999;
Coats, 1999; Noetinger et al., 2000; Penuela et al., 2002a,b; Sadra et al.,
2002). In spite of extensive research on fractured reservoir modeling, no
critical improvements have been made during the last two decades. Even for
the single-phase ow problem, various investigators derive different shape
factors. The shape factor is usually derived from a simple mechanism of
pressure diffusion with constant fracture pressure as a boundary condition,
54
55
Investigator(s)
Approach
Fluid ow
1D
2D
3D
Transient/
PSS
Geometrical
Single phase
12
32
60
PSS
Numerical
Single phase
12
PSS
Numerical
Two phase
25
Transient
Analytical
Numerical
Analytical
Single phase
Two phase
Single phase
12
8
28.45
16
49.58
24
29.61
PSS
PSS
Transient
Analytical
Single phase
19.74
29.61
Transient
Averaging
Single phase
12
28.4
49.6
Numerical
Single phase
20
Random walk
Single phase
11.5
27.1
Numerical
Two phase
9.87
Transient
Numerical
Single phase
24
48
Transient
9.87
PSS
porosity simulations. They suggested that the shape factor depends on the
geometry and physics of pressure diffusion in the matrix. Here, we will
show that the shape factor also depends on the way the pressure changes
in the fracture.
Quintard and Whitaker (1996) used the volume-averaging technique to
derive the shape factors for single-phase ow of slightly compressible uids.
Their values are exactly the values obtained by Coats using the analytical
solution of the diffusivity equation under the PSS assumption and constant
pressure at the matrix boundaries. Bourbiaux et al. (1999) derived a shape
factor for two-dimensional matrix-fracture transfer based on a single-phase
ne grid simulation and PSS assumption. To evaluate the shape factor, they
performed a ne grid simulation on a square matrix block under constant
fracture pressure at the matrix boundaries. Using the results of the ne grid
simulation, the shape factor is back calculated. They derived a shape factor
56
57
Determination of p m in Equation (5) requires solution of the matrix pressure subject to the appropriate initial and boundary conditions. The governing partial differential equation that describes the single-phase pressure
diffusion in a matrix block is given by Equation (3). This equation can be
solved by the Laplace transform method with an arbitrary boundary condition. The Laplace domain solution for matrix pressure Pm (x, s ), can then
be integrated to obtain the average matrix block pressure in the Laplace
domain P m (s ), as given below:
1
Pm =
Pm dVm ,
(6)
Vm
Vm
= s P m pm (0).
(7)
t
Substituting all terms in Equation (5) gives the matrix-fracture transfer
shape factor.
1 s P m p m (0)
=
(8)
.
m
1 P m pf
Equation (8) suggests that calculation of shape factor can be performed by
nding an expression for average matrix pressure in Laplace space, P m (s ).
In the following section, Equation (8) is used to obtain the shape factor
for different geometries, subject to different boundary conditions in the
fracture.
4. Solutions
The solution of the diffusivity equation that leads to the derivation of the
shape factor for various boundary conditions and different matrix block
geometries is presented in this part. This is demonstrated here for a particular case. The solutions for other cases are presented later.
Consider a slab shape matrix block of thickness hm , with initial pressure
pi sandwiched by two parallel planes of fractures with pressure pf where
In this paper, capital letters are used for variables in Laplace domain, and the
sign is used to denote average values.
58
(9)
tD = 0, 0 xD 1,
pm (xD , tD ) = 0,
pm (xD , tD ) = pf , tD > 0, xD = 1,
pm (xD , tD )
= 0,
tD > 0, xD = 0,
xD
where
p = p(xD , tD ) pi .
The pressure pi is the initial matrix block pressure. The parameters xD and
tD are dimensionless length and time respectively and are dened as follows:
xD = x/ lc
and
tD = m t/ lc2 ,
(10)
where lc is the matrix block characteristic length. For a slab shaped matrix
block we consider half of the matrix-block thickness as the characteristic
length. The characteristic lengths of cylindrical and spherical matrix blocks
are considered to be equal to the block radius.
To investigate the effect of fracture pressure on shape factor, pm is
obtained for different boundary conditions using the Laplace transform
method. In this study, we are considering step change, exponential, and linear pressure depletion as well as constant ux boundary cases. In addition to
the slab geometry, the diffusivity equation is also solved for cylindrical and
spherical geometry subject to the mentioned boundary condition. Functions
describing the pressure regime in the fracture are given in Table II.
Equation (9) with a constant boundary condition pf can be solved to
nd the pressure distribution in the matrix block. The Laplace base solution is given by the following equation (Ozisik, 1980)
1 hm 2 pf cosh(xD s)
Pm =
,
(11)
m 2
s cosh s
where s is the Laplace variable with respect to tD . Equation (11) can be
integrated over the block volume to nd the average block pressure given
by:
1 hm 2 pf tanh s
P m =
.
(12)
m 2
s s
59
Table II. Matrix pressure and its average in slab shape for different boundary conditions
Matrix and
fracture BC
Pm
Constant
1
m
pressure pf
1
m
Linear
decline, pf =
pi (1 t) 1/t
1
m
Exponential,
pf = pi et
1
m
Constant
Pm
hm 2
pi
2
hm 2
pi
2
hm 2
pi
2
hm 2
pi
2
D s)
+ cosh(x
s cosh s
sp2i
pi
1
m
cosh(xD s)
cosh( s)
1
s
1
s+
2
m hm
Q4k
m
1
m
cosh(xD s)
cosh( s)
cosh(xD D s)
s s sinh s
1
m
1
m
hm 2
pi
2
hm 2
pi
2
hm 2
pi
2
hm 2
pi
2
) tanh
+ (pf psi
s
s 2pi s tanh
s
spi s 1 s+
tanh
2
s
hm
Qm
4k
s2
m
ux, Qm
4
1 {sP m }
1 {P m } pf
.
1 tanh
s
2
hm =
.
s
m
1 1 tanh
4
h2
s s
(13)
(14)
The product group h2m is dimensionless and will be called shape factor.
For cylindrical and spherical matrix blocks, hm will be replaced by R. The
transfer-shape factor for a slab matrix with step change in fracture pressure
is obtained using Equation (14) and by inverting its Laplace functions into
the real-time domain using an appropriate Laplace inversion algorithm. In
this work we have used the Stehfest algorithm (Stehfest, 1979).
Similarly, we have obtained the shape factor when the fracture pressure
is a continuous function of time. The solutions are given in Table II. Exponential and linear functions have been considered. The shape factors for
other matrix geometries are calculated using a similar approach. Tables III
and IV show the results for cylindrical and spherical blocks, respectively
(Hassanzadeh, 2002). The analogy between cylindrical (spherical) blocks,
60
Table III. Matrix pressure and its average in cylindrical shape for different boundary
conditions
Matrix and fracture
BC
Pm
2
Rm
m
2
Rm
m
Constant pressure pf
Linear decline, pf =
pi (1 t) 1/t
2
Rm
m
Exponential,
pf = pi et
2
Rm
m
Constant ux, Qm
Pm
pi
s
pi
s
pi
s
pi
s
i ) I0 (r s)
+ (pf p
s
I1 (Ro s)
2
Rm
m
2
Rm
m
pi
1
s
1
s+
2 Q
Rm
m
2km
I0 (rD s)
2
Rm
m
I1 ( s)
I0 (rD s)
s sI1 ( s)
2
Rm
m
pi
s
pi
s
s2p
2 s I ( s)
pi
s
2
s
psi 1 s+
s
pi
s
2
Rm
2km
Qm
s2
I0 ( s)
I1 ( s)
Table IV. Matrix pressure and its average in spherical matrix block for different boundary
conditions
Matrix and
fracture BC
Constant
Pm
2
Rm
m
pressure pf
Linear
2
Rm
m
decline, pf =
pi (1 t) 1/t
Exponential,
2
Rm
m
pf = pi et
Constant
2
Rm
m
Pm
pi
s
i) 1
+ (pf p
s
rD
pi
s
sp2i
pi
s
pis
pi
s
sinh(rD s)
sinh s
2
Rm
m
1 sinh(rD s)
s rD sinh s
1
s
1
s+
2
Rm
m
1 sinh(rD s)
rD sinh s
2 Q
Rm
3 sinh(rD s)
m
3km s cosh s 1 sinh s
R
m
2
Rm
m
2
Rm
m
pi
s
3
i)
+ (pf p
coth s 1s
s
s
pi
s
sp2i
coth
s 1s
i 1 s
s3p
s+
s
coth s 1s
2
Rm
pi
Qm
3k
s
s2
m
pi
s
ux, Qm
61
Figure 1. Shape factor constant for slab shape matrix block subject to different
boundary conditions.
62
63
Figure 2. Shape factor constant for cylindrical shape matrix block subject to different boundary conditions.
Figure 3. Shape factor constant for spherical shape matrix block subject to different
boundary conditions.
64
Table V. Shape factor constant for different geometry matrix block subject to different
boundary conditions
Type of boundary condition
Cylindrical
Spherical
9.87
9.87
12
12
12
18.2
18.2
25.13
25.5
25.13
25.65
25.65
39
39
39
the shape factor converges to a constant value of 39. The linear pressure
decline boundary condition gives a similar shape factor as the exponential
case with a small value of a. Applying the constant ux boundary condition results in a stabilized value of 39 similar to the linear decline boundary condition.
6. Signicance of the Results
In summary, it is found that depending on the boundary condition two
stabilized shape factor constant can be obtained for each geometry, which
could be 2040% apart. Results also reveal that the time to stabilization
depends on the boundary condition imposed on the matrix block. Furthermore, it was shown in Figures 13 that the different stabilized values can
be obtained by applying an exponential boundary condition with various
a exponents, where large values of the exponent give the smaller stabilized
value of the shape-factor constant. These stabilized shape factor constants
for different cases are presented in Table V.
The stabilized value of the shape-factor constant is usually used in the
traditional double porosity model. This can cause the following two types
of errors: (1) the matrix-fracture transfer would be underestimated at early
times, because Figures 13 show that at early times, the actual value of
shape-factor constant is larger than the stabilized value. (2) Depending on
the stabilized value of the shape factor chosen, the calculated value could
be signicantly different even at late times.
The essential question is what is the magnitude of the error if one uses the
stabilized value of shape factor instead of its transient value? We investigated
the error introduced in the single-phase matrix fracture mass exchange term
if one uses the stabilized value instead of the appropriate transient value.
Here, we dene the relative errors as the difference between the ow rates
as calculated using the transient shape-factor and the stabilized-shape factor,
65
Figure 4. Relative error for a slab shape matrix block subject to different boundary
conditions.
divided by the rate using the transient-shape factor. A positive value would
indicate an underestimation of the transfer term by using the stabilized values. The results in Figures 46 illustrate that using the stabilized value of the
shape-factor constant at early time results a signicant underestimation of
the rate of uid transfer from matrix into fracture. For example, for a slab
shape matrix block if one uses the stabilized value of shape factor, the calculated rate of mass transfer from matrix block into fracture at dimensionless
time of 0.04 would be 3050% lower than the actual transient case. The relative error dies down for cases where the larger stabilized shape factor was
used. On the other hand, using the lower value of the stabilized shape-factor
leads to an overestimation of the matrix-fracture transfer term late times. The
difference, which is of the order of 2060% depending on the matrix geometry, does not approach zero. While, depending on the boundary condition,
the magnitude of the transfer rate may be small, nevertheless the error would
remain if one uses the lower values of the stabilized shape factor.
7. Discussion
Lim and Aziz (1995) approximated the pressure diffusion in a matrix block
surrounded by two and three sets of perpendicular fractures by solving
66
Figure 5. Relative error for a cylindrical shape matrix block subject to different
boundary conditions.
Figure 6. Relative error for a spherical shape matrix block subject to different
boundary conditions.
67
(15)
where = km /m cm
(16)
We solve the above problem using Duhamels theorem and its stepwise
approximation and compare the solutions. Solution for this ODE for constant and linear (pf = pi (1 tD ), 1/tD ) fracture pressure are given by
the following equations, respectively.
p m (tD ) = pf + pi pf e
2
4
tD
2
4
4 tD
p m (tD ) = pi (1 tD ) + 2 1 e
(17)
68
As mentioned the average matrix pressure and uid efux for a linear
fracture pressure can be obtained by superposition of the constant fracture pressure solutions. However, the accuracy of the superposition solution
depends on the number of steps used. To demonstrate the effect of number
of steps on the average matrix pressure and uid efux, result of the linear
pressure decline in the fracture and the superposition solutions of the constant fracture pressure with different number of steps are compared in Figure 7. The dimensionless efux in this gure (7a) represents the uid efux
divided by its stabilized value under the linear fracture decline. Results
show that using superposition of large pressure steps underestimates the
uid efux signicantly. The average matrix pressure and the uid efux for
a linear fracture pressure can be obtained by superposition of the constant
pressure solution only if small pressure steps are used. Therefore, in practice using large time-steps in numerical simulations can potentially introduce large errors in simulation results.
We found a shape factor constant of 9.87 and 12 for a slab shape matrix
block. Warren and Root (1963) and Lim and Aziz (1995) reported values
of 12 and 9.87, respectively. For the two-dimensional case, we found values of 18.2 and 25.13. A shape factor in this range was reported before
by different authors such as Kazemi and Gillman (1993), Lim and Aziz
(1995), Bourbiaux et al. (1999), and Sadra et al. (2002). For the threedimensional cases, we derived values of roughly 25 and 39. Thomas et al.
(1983), Kazemi and Gillman (1993), and Lim and Aziz (1995) reported
shape factor values of 25, 29.61, and 29.61, respectively, for three-dimensional transfer cases.
(a)
(b)
Figure 7. Dimensionless uid efux (a) and dimensionless matrix pressure (b) for a
linear pressure decline in fracture as a function of dimensionless time compared
with the superposition solutions of the constant fracture pressure steps. Solutions
are compared for different numbers of steps.
69
Acknowledgements
We would like to thank Jacques Hagoort (Hagoort & Assoc.) for his
comments and discussion that greatly improved the earlier version of the
70
71
Stehfest, H.: 1979, Numerical inversion of Laplace transforms algorithm 368 communications of the ACM. 13(1), 47.
Thomas, L. K., Dixon, T. N. and Pierson, R. G.: 1983, Fractured reservoir simulation,
SPEJ, February, 4254.
Warren, J. E. and Root, P. J.: 1963, The behavior of naturally fractured reservoirs, SPEJ,
September, 245255.