You are on page 1of 15

International Journal of Engineering Science 42 (2004) 1289–1303

www.elsevier.com/locate/ijengsci

Buckley–Leverett mathematical and numerical


models describing vertical equilibrium process in porous media
M.C.C. Cunha a, M.M. Santos a,*
, J.E. Bonet b

a
Department of Mathematics, IMECC–UNICAMP, CP 6065, 13083-970 Campinas, SP, Brazil
b
FEM–UNICAMP, CP 6065, 13083-970 Campinas, SP, Brazil
Received 15 April 2003; received in revised form 16 September 2003; accepted 6 October 2003
Available online 24 June 2004

(Communicated by I. STAKGOLD)

Abstract
A laboratory experiment is performed to analyze the motion of two-phase flow in a porous medium
under the condition of vertical equilibrium. The mathematical model is described. The analytical solution is
exhibited for short time. Numerical schemes of Godunov type are implemented, with and without viscosity.
Analytical and numerical solutions that take into account hysteresis effects are also considered.
 2004 Elsevier Ltd. All rights reserved.

AMS: 35L65; 76S05; 76Txx


Keywords: Porous media; Conservation laws; Two-phase flow; Godunov method; Riemann problem; Vertical
equilibrium; Hysteresis; Stationary discontinuity

1. Introduction

The assumption of vertical equilibrium is useful in some two-phase processes taking place in
oil recovery, ground water contamination or other practical situations. Formally, the vertical
equilibrium in a porous medium occurs when, at each point of this medium, the sum of the fluid

*
Corresponding author. Tel.: +55-19-3788-6008; fax: +55-19-3289-5766.
E-mail addresses: cunha@ime.unicamp.br (M.C.C. Cunha), msantos@ime.unicamp.br (M.M. Santos), bonet@
fem.unicamp.br (J.E. Bonet).

0020-7225/$ - see front matter  2004 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ijengsci.2003.10.005
1290 M.C.C. Cunha et al. / International Journal of Engineering Science 42 (2004) 1289–1303

velocities is zero (see e.g. Lake [12, p. 205]). This means that the velocities of two immiscible
fluids are redistributed in accordance with capillary and gravitational force balance: the denser
fluid flows downward while the less dense fluid flows upward, with velocities equal in magni-
tude.
Several authors have analyzed physical processes under vertical equilibrium assumption. For
instance, Templeton et al. [19] to study counter-flow processes, Briggs and Katz [3] to study gas
storage reservoir in aquifer sands, Coats and Henderson [4] in a two-dimensional simulation of
three-dimensional reservoir performance. Vertical equilibrium models were also used by Siddiqui
and Lake [16] to analyze secondary hydrocarbon migration, van Dijke and van der Zee [20] and
Bear and Ryzhik [1] to study aquifer contamination.
In this paper we describe a laboratory experiment designed to follow the displacement of a
gas phase through a liquid phase and to measure the volume of gas retained in a porous
medium. Assuming the hypothesis of vertical equilibrium, the mathematical model is described.
We present the mathematical and numerical aspects from the point of view of the mathe-
matical theory of conservation laws (see e.g. [6,10,13,15,17]). This approach, we believe, is
much more used in mathematical literature than in engineering, but we think it has the
advantage of providing more general solutions and treating the equation in a conservative
form, differently from the characteristic method (similarity solutions) more usual in engineering
when dealing with hyperbolic equations. A summary of the paper is described in the following
paragraphs.
In Section 2, Darcy law and mass conservation are used to obtain the equation that models
vertical equilibrium displacements, Eq. (2.1). The liquid relative permeability is zero when the
liquid saturation, s, vanishes. Similarly, the gas relative permeability is zero when s ¼ 1. Therefore
the flux function, Eq. (2.2), vanishes at these values of saturation. Furthermore, applying the
usual expressions for relative permeabilities, this flux function is a bell shaped function, changing
concavity twice. This fact adds more difficulties to the mathematical and numerical analysis than
does the best known two phase porous medium equation: the Buckley–Leverett equation, in
which the flux function is an S shaped function. The third term in Eq. (2.1) is due to capillary
effects. It approaches zero for small capillary number N, that is, for long porous media. Therefore,
capillary effects are negligible in several practical situations and in this case the mathematical
equation is a conservation law.
In Section 3, this conservation law is solved. Riemann solution to the problem with initial and
boundary condition is obtained by compositions of shock and rarefaction waves that satisfy the
entropy condition. The ideas are also extended to cases in which the hysteresis affects relative
permeabilities.
In Section 4, the numerical approach is presented. It is a Godunov type method, which includes
the transonic rarefaction waves near the static saturation (maximum of flux function). In fact, the
usual upwind methods are unable to capture the rarefaction near these points. Also in Section 4,
we show in figures that the numerical results agree fairly well with the analytical solution pre-
sented in Section 3. The numerical method is extended to simulate the model including capillary
effects. In this case the shocks are smeared or spread around the shock front, as was expected.
In Section 5, details of the experiment done in laboratory are explained. The comparison
between experimental and numerical results shows that the gas phase trapped at the lower part
of the porous medium is due mainly to the hysteretic effects.
M.C.C. Cunha et al. / International Journal of Engineering Science 42 (2004) 1289–1303 1291

2. Mathematical model

Darcy law extended to two-phase flow can be written for each fluid as
 
kkrw opw
vw ¼   qw g ;
lw ox

and
 
kkrg opg
vg ¼   qg g ;
lg ox

where vw ðvg Þ is the Darcy flux velocity for the liquid (gas) phase. The other parameters in Darcy
law are: k is the absolute permeability, krw ðkrg Þ is the liquid (gas) relative permeability, lw ðlg Þ is the
liquid (gas) viscosity, pw ðpg Þ is the liquid (gas) pressure and qw ðqg Þ is the liquid (gas) density and g
is the acceleration due to gravity. We assume incompressibility of the gas phase, which is a
realistic assumption for the experiment done in laboratory.
To eliminate the unknown pressure gradients, following the fractional flow theory, we use the
total flow velocity, v ¼ vw þ vg , and the capillary pressure, as a function of the liquid saturation, s:
pc ðsÞ ¼ pg  pw :
Under the hypothesis of vertical equilibrium, the total flow velocity vanishes, v ¼ 0, and we have
vg ¼ vw . Then, some mathematical manipulations give
 
1 kkrw krg opc
vw ¼ þ Dqg ;
lg krw þ Mkrg ox

where M ¼ llwg is the viscosity ratio and Dq ¼ qw  qg .


Using this expression in the liquid conservation equation,
os ovw
/ þ ¼ 0;
ot ox
where sðx; tÞ is the liquid saturation and / is the medium porosity, we have
   
os k o krw krg k o krw krg opc
/ þ Dqg þ ¼ 0:
ot lg ox krw þ Mkrg lg ox krw þ Mkrg ox

Changing the variables to xD ¼ Lx , tD ¼ lkDqg


w /L
t, we obtain the equation in dimensionless variables,
ðx; tÞ hereafter, 0 < x < 1, t > 0:
 
os o o oPc
þ ðF ðsÞÞ þ N F ðsÞ ¼ 0: ð2:1Þ
ot ox ox ox

In (2.1), Pc ðsÞ ¼ pcPeðsÞ, Pe is a constant reference pressure, N ¼ LDqg


Pe
is the capillary number and the
flux function is:
1292 M.C.C. Cunha et al. / International Journal of Engineering Science 42 (2004) 1289–1303

0.14

0.12
A

0.1
B
Flux Function

0.08

0.06
C

0.04

0.02

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Saturation
S R
Fig. 1. Flux function and two illustrations of wave compositions: (i) the composition II in Proposition 1, ð1; 0Þ!A!B,
S;R
(ii) the composition V in Proposition 2, C!ð1; 0Þ.

Mkrw krg
F ðsÞ ¼ : ð2:2Þ
krw þ Mkrg

In general, if we need analytical expressions we use Corey’s expressions for relative permeabilities:
krw ðsÞ ¼ sew and krg ðsÞ ¼ ð1  sÞeg , with positive exponents ew and eg. In this case the function
F ðsÞ has a bell shaped graphic, as illustrated in Fig. 1 and in other figures in this paper.

3. Analytical solution

3.1. The model as a conservation law

Laboratory experiments were performed in L  90 cm long core, which means N  0:03 in our
experiments. However, in reservoir conditions L is much bigger, which makes the capillary
pressure effects usually negligible. In this case Eq. (2.1) is a conservation law. We want to find the
liquid saturation sðx; tÞ, 0 < x < 1, t > 0, such that

os o
þ ðF ðsÞÞ ¼ 0: ð3:1Þ
ot ox

Initially the gas is concentrated near the base ðx ¼ 1Þ and the rest of the sample is filled with the
liquid. Thus the initial condition is the step function:
M.C.C. Cunha et al. / International Journal of Engineering Science 42 (2004) 1289–1303 1293

1; 0 < x 6 x0 ;
sðx; 0Þ ¼ ð3:2Þ
b; x0 < x < 1;

for given values of x0 and b such that 0 < x0 < 1 and 0 < b < 1.
Eqs. (3.1) and (3.2) define a Riemann problem centered at x ¼ x0 , with left state
ðs ; F  Þ ¼ ð1; 0Þ and right state ðsþ ; F þ Þ ¼ ðb; F ðbÞÞ. To obtain the physically correct solution,
following the ideas used by Isaacson [10], Johansen and Winther [11], we use the ‘‘convex envelope
construction’’ that satisfies the Oleinick shock entropy condition (see e.g. Dafermos [6], p. 184).
The basic points of this methodology used to solve non-convex conservation laws are the fol-
lowing, where ui ’s are given states:
S
A. Shock waves will be denoted by u1 !u2 . The speed r at which a discontinuity travels, shock
speed, can be computed using the Rankine–Hugoniot jump condition

F ðu1 Þ  F ðu2 Þ
r¼ :
u1  u2
R
B. Rarefaction waves will be denoted by u1 !u2 . In this case the characteristic speed is F 0 ðsÞ.
a b
C. Two waves, u1 !u2 and u2 !u3 , where a, b can be either shock or rarefaction waves, can be
composed satisfying the entropy criterion if the final speed of the first wave is less than or
equal to the initial speed of the second wave. In this case we say that the waves are compatible,
i.e. they can be composed to solve the Riemann problem with left and right states, u1 and u3 ,
respectively.

Taking into account the above concepts, entropic compositions of shock and rarefactions
waves can be obtained using geometrical constructions in the ðs; F Þ-plane. This approach is usual
in solving the Buckley–Leverett equation in which the flux function is S shaped (Appendix 2 in
Isaacson [10]). It seems to us that the ðs; F Þ-constructions approach is more intuitive and probably
more understandable in petroleum engineering. For this reason we will use this methodology in
our arguments.
In the ðs; F Þ-constructions the secant and tangent lines represent shock waves; parts of F ðsÞ
graphic represent rarefaction waves. These waves are connected to find the convex hull of F ðsÞ in
the interval defined by the left ðs Þ and right ðsþ Þ states. We can see directly that the convex hull
will be upward if s < sþ ; otherwise the convex hull will be downward. Examples of the ðs; F Þ-
constructions are presented in Figs. 1 and 2, illustrations of the next proposition.
In Proposition 1 we present the compositions used to solve the Riemann problem (3.1) and
(3.2). The statement of this proposition depends on the following three points:

• ðs1 ; F ðs1 ÞÞ is such that the tangent line to the graph of F at ðs1 ; F ðs1 ÞÞ with slope F 0 ðs1 Þ < 0 inter-
sects this graph at ð1; 0Þ;
• Si1 is the smallest of the two inflection points of F ;
• ðs2 ; F ðs2 ÞÞ is such that the tangent line to the graph of F at ðs2 ; F ðs2 ÞÞ with F 0 ðs2 Þ > 0 intersects
this graph at ðb; F ðbÞÞ, b < Si1 .
1294 M.C.C. Cunha et al. / International Journal of Engineering Science 42 (2004) 1289–1303

0.14

0.12
A

0.1
B
Flux Function

0.08

0.06

0.04

0.02

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Saturation
S R S
Fig. 2. Illustration of the composition III, Proposition 1, with b ¼ 0:1, ð1; 0Þ!A!B!C.

Proposition 1. The solution of the Riemann problem (3.1) and (3.2) is given by one of the following
compositions:
S
I. If b P s1 : ð1; 0Þ!ðb; F ðbÞÞ.
S R
II. If s1 > b P Si1 : ð0; 1Þ!ðs1 ; F ðs1 ÞÞ!ðb; F ðbÞÞ.
S R R
III. If b < Si1 : ð1; 0Þ!ðs1 ; F ðs1 ÞÞ!ðs2 ; F ðs2 ÞÞ!ðb; F ðbÞÞ.

Proof. We can show directly that all these compositions are compatible. As an illustration, the
ðs; F Þ-constructions in the cases II and III are presented in Figs. 1 and 2, respectively. In these
cases it was necessary to introduce the intermediary states A ¼ ðs1 ; F ðs1 ÞÞ and B ¼ ðs2 ; F ðs2 ÞÞ
defined above. The construction II is compatible, since the speed of the first wave, F 0 ðs1 Þ, is equal
to the initial speed of the rarefaction wave plotted in Fig. 1. h

Besides the initial condition (3.2), we also need the boundary conditions for the differential
equation. To simulate the vertical equilibrium in laboratory both ends of the core were sealed.
This means zero flux at x ¼ 1, the lower end, and at x ¼ 0, the upper end:

F ðsÞ ¼ 0 at x ¼ 0 and x ¼ 1: ð3:3Þ

Since F ðsÞ ¼ 0 if s ¼ 0 or s ¼ 1, (3.3) means that sð0; tÞ ¼ 0 or sð0; tÞ ¼ 1. We discard sð0; tÞ ¼ 0


because the wave connecting, s ¼ 0 from the left to sþ ¼ 1 (sample saturated with liquid for
small time), has zero speed. On the other hand, sð1; tÞ ¼ 0 is not allowed because the wave
M.C.C. Cunha et al. / International Journal of Engineering Science 42 (2004) 1289–1303 1295

connecting s ¼ b, the left state, to sþ ¼ 1, the right state, has positive speed, a contradiction with
the zero flux assumption at x ¼ 1. Therefore,

sð0; tÞ ¼ 1 and sð1; tÞ ¼ 1 ð3:4Þ

are the boundary conditions, corresponding to (3.3) for all time before the interactions of char-
acteristics.
Thus the boundary conditions generate a new Riemann problem centered at x ¼ 1, with
s ð1; tÞ ¼ b and sþ ð1; tÞ ¼ 1. The ðs; F Þ-constructions can be used to visualize the entropic solution
for this initial and boundary problem for (3.1). In Proposition 2 we present the compositions used
to solve the boundary problem. The statement of this proposition depends on the following two
points:

• ðs3 ; F ðs3 ÞÞ is such that the tangent line to the graph of F , at ðs3 ; F ðs3 ÞÞ with slope F 0 ðs3 Þ < 0
intersects the graph of F at ðb; F ðbÞÞ;
• Si2 is the biggest inflection point of F .

Proposition 2. The solution of Eq. (3.1) with s ð1; tÞ ¼ b and sþ ð1; tÞ ¼ 1 is given by one of the
following compositions:
R
IV. If b P si2 : ðb; F ðbÞÞ!ð1; 0Þ.
S R
IV. If b < si2 : ðb; F ðbÞÞ!ðs3 ; F ðs3 ÞÞ!ð1; 0Þ.

Proof. We can prove the proposition directly, using the ðs; F Þ-constructions of the compositions
presented in IV and V. As an illustration, the case V is presented in Fig. 1. h

The Propositions 1 and 2 are used to obtain the final solution of the initial and boundary
problem for Eq. (3.1). As an illustration we present a saturation distribution
2 ð1sÞ2
calculated with this
methodology in Fig. 3. The calculations were done with F ðsÞ ¼ ss2 þð1sÞ 2 , b ¼ 0:3 and x0 ¼ 0:6, in

(3.2). With this data, the solution can be obtained coupling III, in Proposition 1, and V, in
Proposition 2. The saturation profile in Fig. 3 is plotted at a time before the interactions between
characteristics. To plot this ‘‘analytical’’ solution we use the characteristics in ðx; tÞ-plane, com-
pute the shock speed and use interpolation to find values of rarefactions using ðF 0 ðuÞÞ1 in some
characteristics, LeVeque [13]. The usual name of semi-analytic for solutions computed with the
method of characteristics is because they need these complementary numerical calculations. We
make use of Fig. 3 to compare this semi-analytical solution with a numerical approximation
computed with a Godunov finite difference discretization, described in the next section.

3.2. The model as a conservation law with hysteresis

We recall that in the laboratory experiments the gas had initially occupied part of the sample––
the interval ðx0 ; 1Þ in (3.2). For t ¼ 0, when the core is turned upside down, this gas slug starts
flowing upward, with the liquid imbibing the part previously with gas. Hysteretic behavior on
cyclic processes like this has long been recognized and studied. Among the vast literature we
1296 M.C.C. Cunha et al. / International Journal of Engineering Science 42 (2004) 1289–1303

0.9

0.8

0.7
Liquid Saturation

0.6

0.5

0.4

0.3

0.2
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

x = Distance from the upper face

Fig. 3. Analytical and numerical solutions plotted at t ¼ 0:5, a time before the interactions between characteristics,
using model without hysteresis. In the numerical simulations we used and h ¼ 103 and Dt=Dx ¼ 0:1.

mention Colonna et al. [5], Gladfelter and Gupta [9], Braun and Holland [2], Furati [8] and the
references therein.
Following the literature, for the same value of liquid saturation, the relative permeability in the
imbibing process, that occur in ðx0 ; 1Þ, is smaller than the values in the drainage process.
Therefore, in this case drainage and imbibing flux functions are bell-shaped and the imbibing flux
function is below the drainage flux functions as shown in Fig. 4. Similar processes occur in other
applications of the vertical equilibrium model, Briggs and Katz [3], Siddiqui and Lake [16], for
example.
Analytical solutions taking in account hysteresis in vertical equilibrium can also be obtained
using the ideas presented in Section 3.1. However, in hysteretic models we must deal with two
different flux functions. In this case, besides shock and rarefaction waves used before to obtain the
entropic solution, we also use the immobile shock for conservation laws, as used by Furati [7].
Saturation profiles with immobile shocks also have been found by Philip [14] in solving equations
with capillary hysteresis.
There are differences in flux functions because, as observed before, for t > 0 we must use the
imbibing flux function, named F I ðsÞ, in x0 6 x < 1, and a draining flux function, named F D ðsÞ,
elsewhere. Therefore, the hysteretic model is a conservation law,

os o
þ ðF ðsÞÞ ¼ 0; ð3:5Þ
ot ox
M.C.C. Cunha et al. / International Journal of Engineering Science 42 (2004) 1289–1303 1297

0.35

Drainage and Imbibition Functions 0.3

0.25

B
0.2
A

0.15

0.1

I
C
0.05

E
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Saturation

Fig. 4. Drainage (the upper curve) and Imbibing flux curves. Also in this figure we plot an illustration of the com-
S I
position (3.9), Proposition 3, ð1; 0Þ!A!B, coupled with the boundary Riemann problem in the imbibing curve
S R
B!C!E.

where

F D ðsÞ; 0 < x < x0 ;
F ðsÞ ¼ ð3:6Þ
F I ðsÞ; x0 < x < 1:

The initial condition, which defines the Riemann problem centered in x0 is the same used before,

1; 0 < x 6 x0 ;
sðx; 0Þ ¼ ð3:7Þ
b; x0 < x < 1;

but now the left state is at the drainage curve, ðs ¼ 1; F ¼ F D ¼ 0Þ, while the right state is a point
on the imbibing curve, ðs ¼ b; F ¼ F I ðbÞÞ.
In the following proposition we use shock, rarefactions and immobile shocks waves to describe
the entropic constructions for the hysteretic case. First we introduce the notation.
I
The immobile shocks, the discontinuities with zero speed, will be denoted by u1 !u2 . By
ðSMI ; F I ðSMI ÞÞ we denote the maximal point in the imbibing flux function, and ðSMD ; F D ðSMD ÞÞ is the
point, at the drainage curve, such that F I ðSMI Þ ¼ FD ðSMD Þ. The intermediary states ðs1 ; F D ðs1 ÞÞ and
ðs3 ; F I ðs3 ÞÞ were used in Propositions 1 and 2.

Proposition 3. The solution of the Riemann problem 3.3, 3.4, 3.5 is given by one of the following
compositions
1298 M.C.C. Cunha et al. / International Journal of Engineering Science 42 (2004) 1289–1303

• If b P SMI :
S R I
ð1; 0Þ!ðs1 ; F D ðs1 ÞÞ!ðSMD ; F D ðSMD ÞÞ!ðb; F I ðbÞÞ: ð3:8Þ

• If b < SMI :
S R I R S
ð1; 0Þ!ðs1 ; F D ðs1 ÞÞ!ðSMD ; F D ðSMD ÞÞ!ðSMI ; F I ðSMI ÞÞ!ðs3 ; F I ðs3 ÞÞ!ðb; F I ðbÞÞ: ð3:9Þ

Proof. Direct calculations show that the connection, used in both cases (3.8) and (3.9), are
compatible, i.e., on each connection the final speed of the first wave is less than or equal to the
initial speed of the second wave. For instance, we present the construction (3.9) in Fig. 4. We
observe that rarefaction in construction (3.8) is absent if F D ðSMD Þ < F D ðs1 Þ. Also, one or both
rarefactions in (3.9) can be absent, subject to the geometric behavior of the drainage or imbibing
flux functions. h

To obtain the final solution for the initial and boundary problem we can use the same pro-
cedure described before, in the case without hysteresis. In the hysteretic case we use the final point
of the imbibing curve as the zero flux boundary condition.

4. Numerical methods

The ideas presented before are addressed to Riemann problems for conservation laws. Thus the
methodology can not be used in two practical situations: general initial conditions for vertical
equilibrium equation, because this is not a Riemann problem, and the model including capillary
effects, because the differential equation is not a conservation law.
In the conservation law case, the Godunov method can be written as LeVeque [13]:

k 
sjþ1
i ¼ sji  Hðsji ; sjiþ1 Þ  Hðsji1 ; sji Þ ; ð4:1Þ
h

where H is the numerical flux and sji ¼ sðxi ; tj Þ, h is the space mesh width and k is the time step. In
our particular case, we can use the shape of the flux function to write the numerical flux in a
simplified form, more convenient for computational implementations. Let sM be the value such
that F 0 ðsM Þ ¼ 0. There are four possibilities of si ; siþ1 and to each one we have the following
adapted expressions of the numerical fluxes:

1: si ; siþ1 < sM ) H ðsi ; siþ1 Þ ¼ F ðsi Þ;


2: si ; siþ1 > sM ) H ðsi ; siþ1 Þ ¼ F ðsiþ1 Þ;
ð4:2Þ
3: si < sM < siþ1 ) Hðsi ; siþ1 Þ ¼ min f F ðsi Þ; F ðsiþ1 Þg;
4: si > sM > siþ1 ) Hðsi ; siþ1 Þ ¼ F ðsM Þ:
M.C.C. Cunha et al. / International Journal of Engineering Science 42 (2004) 1289–1303 1299

The case 4 is crucial because without it the numerical method does not capture the rarefaction
wave near the static saturation, the maximum of the flux function.
If the capillary pressure is included in the model, Eq. (2.1), its term can be treated with the usual
discretization for second order derivatives. The final expression for the explicit method used in our
calculations is:

k  Nk h j i
sjþ1
i ¼ sji  H ðsji ; sjiþ1 Þ  H ðsji1 ; sji Þ  2 Fiþ j j j j j
1 ðPc ðsiþ1 Þ  Pc ðsi ÞÞ  Fi1 ðPc ðsi Þ  Pc ðsi1 ÞÞ ;
h h 2 2

j 1 j j
where Fiþ 1 ¼ 2 ½F ðsiþ1 Þ þ F ðsi Þ.
2
We observe that this formulation is very convenient in introducing the zero flux boundary
condition without the necessity of establishing values for the saturation at the ends of the core. To
show that, we write both equations in the form
os o
þ ðvÞ ¼ 0 ð4:3Þ
ot ox
where v ¼ F , in the conservation law case, Eq. (3.1), and
 
opc
v¼F 1N
ox

in the general equation (2.1), which includes the capillary pressure term. In the following, we
describe the conservation law case but the same arguments can be used in the general equation if
we incorporate the capillary pressure in the flux term.
First we integrate (4.3) in ½xih=2 ; xiþh=2   ½tj ; tjþ1 . In the Godunov method the solution of the
Riemann problem centered in xi is used to construct the numerical flux given in (4.2). Therefore
we can use the zero flux boundary condition at both ends of the core by correcting the term
corresponding to the numerical fluxes on the first and last equations of the discretization, the
equations for the first and the last cell. In other words, besides the discretization (4.1), used for
i ¼ 2 : n  1, we use the boundary condition in the first and last equation, i ¼ 1 and i ¼ n to
obtain:

k
sjþ1
1 ¼ sj1 þ H ðsj1 ; sj2 Þ;
h
k
sjþ1
n ¼ sjn  H ðsjn1 ; sjn Þ:
h
In this way, the values of saturations are calculated naturally in each time step in a marching
process, starting from the initial condition. We emphasize the advantages of this procedure be-
cause we do not need to predefine values of the saturation at x ¼ 0 and 1 as we did before in (3.4),
conditions valid only for times before characteristic interactions.
The first illustration of the numerical simulation is presented in Fig. 3 in which, as explained
before, the numerical approximation is compared with the semi-analytical solution in the
2 ð1sÞ2
case where F ðsÞ ¼ ss2 þð1sÞ 2 and b ¼ 0:3. The profile is plotted at a time before characteristics
1300 M.C.C. Cunha et al. / International Journal of Engineering Science 42 (2004) 1289–1303

0.9

0.8

0.7
Liquid Saturation

0.6

0.5

0.4

0.3

0.2
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x = Distance from the upper face

Fig. 5. The numerical solution for the model with hysteresis and capillary pressure: at t ¼ 0, the full line, t ¼ 0:3,
dashed line, and t ¼ 0:5, the dotted line. This shows that the modified Godunov scheme reproduces the immobile shock
at x ¼ 0:6.

interactions. These results show us that the numerical approximation and semi-analytical solution
agree quite well, one validating the other.
We use the hysteretic model with capillary pressure as the second illustration of the numerical
simulations. In Fig. 5 we plot some saturation profiles at three different times: the initial condi-
tion, t ¼ 0:3 and 5 (about 20 h). As we can see in Fig. 5, the numerical method shows high
resolution of the immobile shock, following the methodology used by Furati [7,8].

5. About the experiment

The physical model used in the laboratory is simple but includes the physical processes dis-
cussed in this paper: flow in vertical equilibrium and hysteresis effects. The experimental work
consisted of the measurement of the liquid saturation. The liquid phase contained a tracer and its
saturation was determined by measuring its radioactivity. We used either a composition of sodium
chloride with a concentration of 30 kppm or sodium iodine with a concentration of 65 kppm. The
gaseous phase was the atmosphere air. Here we present only a brief resume on the apparatus and
used techniques, since in fact we repeated the experiments done by Briggs and Katz [3]. The flow
took place along sealed samples collected from a sandstone formation (porosity 20%) in the re-
gion of Botucatu/Brazil. They were sealed in a standard encapsulating process. The saturation
profiles were measured using an X-ray emitter, X-ray detector, multi-channels analyzer (MCA).
M.C.C. Cunha et al. / International Journal of Engineering Science 42 (2004) 1289–1303 1301

We used a XZ positioning system with step motor (0.01 mm) to control the measurement posi-
tions. All the data were acquired and controlled using a computer program system. The sample
dimensions were about 90 cm lengthwise and 2 in. of diameter.
During each run, radiation measurements were made in 43 locations along the length of the
column. These readings were converted to saturation values. The procedure for each run, in a
total of eight, was the following. Initially we saturated the vertical column with a measured
amount of the radioactive liquid phase. Then we allowed the liquid to flow through the bottom of
the column and the gas to occupy the free space entering the top of the column. After a while, we
closed the sample at both ends, the saturation along the sample was measured, the sample was
turned upside down and the subsequent measurements were done until the two phases reached the
equilibrium distribution.
The laboratory experiments were our inspiration during this study. For instance, the presence
of gas near the base of the sample even when all flow ceased, observed in all eight experiments
done in the laboratory, motivated the analysis of hysteretic effects. In fact, the models without
hysteresis do not predict the trapped gas observed in the laboratory.
However, even when we considered the hysteresis effect with only two different curves for flux
functions––the drainage and imbibing curves––the comparison between experiments and math-
ematical results did not work with our presumed efficacy. Here we discuss some aspects to justify
these differences.
If we consider (3.2) as initial condition, i.e., a Riemann problem, the solution including hys-
teresis effects can be obtained using only two flux functions, as described in Section 3.2.

0.35

0.3
Saturation Measured and Simulated

0.25

0.2

0.15

0.1

0.05

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

x = Distance from the upper face

Fig. 6. Laboratory data, the dotted plot, compared with the numerical solution, using four scanning curves, the full
line. The initial condition is plotted in dashed line.
1302 M.C.C. Cunha et al. / International Journal of Engineering Science 42 (2004) 1289–1303

However, the initial conditions of our laboratory experiments are not like those of the
step functions. As an illustration, in Fig. 6 we plot the initial condition for one experiment.
This plot shows that the liquid saturation varies continuously. Several attempts were made to
obtain a step function as initial condition in the laboratory experiments, but the results were
frustrating.
Several researchers have studied the behavior of relative permeabilities when saturation
changes in different directions, Braun and Holland [2] and their references provide an overview of
the work in this area.
It is necessary to have more information about scanning curves in cases where the direction of
saturation change is reversed, and the direction changes at a number of intermediate saturations
between the end points. Scanning curves are the portions of relative permeability curves that link
drainage and imbibing curves when the direction of saturation change is reversed at an inter-
mediary saturation. In our case, the drainage process––gas entering through the upper face––was
interrupted to give the initial condition for the imbibing process occurring in part of the sample.
As a consequence, there are intermediary values for the initial saturations, as Fig. 6 shows.
The study of scanning curves and their effects in analytical and numerical solutions will be
studied in future works. However, we tried to fit the laboratory data introducing four scanning
curves in our numerical simulation as an exercise. To obtain these scanning curves we changed the
exponents of Corey’s expression presented at the end of the Section 2. The results are compared
with laboratory data in Fig. 6. The four steps near x ¼ 1 are a consequence of these scanning
curves. Using more scanning curves we can obtain better approximations for the data. The time
corresponding to the laboratory data and numerical simulation is about six days from the initial
condition ðt ¼ 0Þ. We can observe that the X-ray apparatus did not capture the thin layer of gas at
the top of the sample, x  0 in Fig. 6.

References

[1] J. Bear, V. Ryzhik, On the displacement of NAPL lenses and plumes in a phreatic aquifer, Transp. Porous Media
33 (1998) 227–255.
[2] E.M. Braun, R.F. Holland, Relative permeability hysteresis: laboratory measurements and a conceptual model,
SPE Reservoir Eng. (1995) 222–228.
[3] J.E. Brggs, D.L. Katz, Drainage of water from sand in developing aquifer storage, Tech. Rep. SPE, Soc. Pet. Eng.
1502 (1996).
[4] K.H. Coats, J.H. Henderson, The use of vertical equilibrium in two-dimensional simulation of three-dimensional
reservoir performance, Soc. Pet. Eng. J. (1971) 63–71.
[5] J. Colonna, F. Brissaud, J.L. Millet, Evolution of capillary and relative permeability hysteresis, Soc. Pet. Eng. J.
(1972) 28–37.
[6] C.M. Dafermos, Hyperbolic Conservation Laws in Continuous Physics, Springer-Verlag, 2000.
[7] K. Furati, Effects of relative permeability hysteresis dependence on two-phase flow in porous media, Trans. Porous
Media 28 (1997) 181–203.
[8] K. Furati, The solution of the Riemann problem for a hyperbolic system modeling polymer flooding with
hysteresis, J. Math. Anal. Appl. 206 (1997) 205–233.
[9] R.E. Gladfelter, S.P. Gupta, Effect of fractional flow hysteresis on recovery of tertiary oil, Soc. Pet. Eng. J. (1980)
508–520.
[10] E. Isaacson, Global solution of a Riemann problem for a non-strictly hyperbolic system of conservation laws
arising in enhanced oil recovery, Rockefeller University, NY, (1984) preprint.
M.C.C. Cunha et al. / International Journal of Engineering Science 42 (2004) 1289–1303 1303

[11] T. Johansen, R. Winther, The solution of the Riemann problem for a hyperbolic system of conservation laws
modeling polymer flooding, SIAM J. Math. Anal. 19 (3) (1988) 541–566.
[12] L.W. Lake, Enhanced Oil Recovery, Prentice Hall, 1989.
[13] R. LeVeque, Numerical Methods for Conservation laws, Birkhauser-Verlag, 1990.
[14] J. Philip, Horizontal redistribution with capillary hysteresis, Water Resour. Res. (1991) 1459–1469.
[15] D. Serre, Systems of conservation laws, vol. 1, Cambridge University Press, Cambridge, 1996.
[16] F. Siddiqui, L. Lake, A comprehensive dynamical theory of hydrocarbon migration and trapping, Tech. Rep. SPE
38682 Soc. Pet. Eng., 1997.
[17] J. Smoller, Shock waves and reaction–diffusion equations, second ed., Grundlehren der Mathematischen
Wissenschaften (Fundamental Principles of Mathematical Sciences), vol. 258, Springer-Verlag, New York, 1994.
[19] E.E. Templeton, R.F. Nielsen, C.D. Stahl, A study of gravity counterflow segregation, Soc. Pet. Eng. J. (1962) 185–
193.
[20] M. Van Dijke, S.V.D. Zee, A similarity solution for oil lens redistribution including capillary forces and oil
entrapment, Transp. Porous Media 29 (1) (1997) 99–125.

You might also like