You are on page 1of 13

SPE 113343

Statistical Model of Dispersion in a 2-D Glass Micromodel


Mohammad Hossein Ghazanfari, Sharif University of Technology, Tehran, Iran; Riyaz Kharrat, Petroleum
University of Technology,Tehran, Iran; Dawood Rashtchian, Sharif University of Technology,Tehran, Iran;
Shapour Vossoughi, University of Kansas, Lawrence, KS, USA
Copyright 2008, Society of Petroleum Engineers

This paper was prepared for presentation at the 2008 SPE/DOE Improved Oil Recovery Symposium held in Tulsa, Oklahoma, U.S.A., 1923April2008.

This paper was selected for presentation by an SPE program committee following review of information contained in an abstract submitted by the author(s). Contents of the paper have not been
reviewed by the Society of Petroleum Engineers and are subject to correction by the author(s). The material does not necessarily reflect any position of the Society of Petroleum Engineers, its
officers, or members. Electronic reproduction, distribution, or storage of any part of this paper without the written consent of the Society of Petroleum Engineers is prohibited. Permission to
reproduce in print is restricted to an abstract of not more than 300 words; illustrations may not be copied. The abstract must contain conspicuous acknowledgment of SPE copyright.


Abstract
Microscopic Visualization of the porous media can provide valuable information to enhance understanding of pore-scale
transport phenomena. Here, a micromodel which its grains and pores are non-uniform in size, shape and distribution is
considered as porous medium. The pore size distribution as well as pore length distribution was extracted by applying an
image analysis technique. A two-dimensional random network model of the micromodel has been constructed which the non-
uniformity is considered by assigning measured distribution functions. The random particle method was applied for
correlating and predicting dispersion coefficients based on probabilistic approaches. Statistical derivations result in a new
functional dependence for the longitudinal and transverse dispersion coefficients in terms of pore velocity and ensemble
averages. Prediction from derived model for both longitude and transverse dispersion is in close agreement with the
experimental data. Despite the simplicity of the proposed network model, its accurate prediction provides some confidence
that it can be considered as a reasonable approximation of the complex nature of porous media with reliable predictive
capability.

Introduction
Dispersion in porous media is one of the most extensively studied fundamental transport properties in physics with
applications in improved oil recovery, hydrology and chemical engineering. The geometry and topology of the microscopic
pores control transport properties of porous medium, e.g., dispersion and permeability (Sahimi, 1993). On the other hand, the
exact solutions of fluid flow and convection-diffusion equations at the pore scale are extremely difficult to obtain, due to the
complexity of the boundary conditions at the irregular pore/grain interface (Man and Jing, 2000). Therefore, it is important to
have a reliable tool that can provide plausible estimates of the properties. Instead of searching for exact solution, research
efforts have focused on ways to simplify the irregular pore systems. Such simplified versions of porous medium are called
'network model', which can be the only possible means of understanding the flow through porous media from a microscopic
point of view (Dullien, 1992).
Network models, which are exemplified by the work of Fatt (1956), can be applied either as stochastic model or as a
fixed model, depending on whether or not the pores which constitute the connections in the network are considered having a
probability distribution of sizes. In a sense, the randomly oriented network models are simply more general network models
in which the connecting links in the networks are permitted a distribution of orientations, radii and lengths rather than being
fixed as, for example, the edges of regular polyhedral.
In comparison with the deterministic model of Taylor (1953 and 1054) and Aris (1956), there has been much work to
represent dispersion through statistical models. The idea behind a statistical approach is that to attribute probabilities to
predict the distribution of many tracer particles at a certain time, which were initially in close proximity to each other at
initial time. In short, over large enough time periods, the time-averaged velocity of a single particle can be used in place of
the averages taken over the whole group of particles; thus the problem reduces to that of the random motion of a single
particle (Bear, 1072).
Scheidegger (1954) employed the movement of a random walk particle, under laminar flow conditions, through a
homogeneous isotropic porous medium comprising system of capillaries with identical macroscopic characteristics. His
predicted values for dispersion coefficient make no distinction between longitudinal and transverse spreading of a tracer. De
Josselin de Jong (1958) and Saffman (1959) employed a statistical approach for obtaining estimates of both longitudinal and
transversal dispersion coefficients as separate (albeit coupled) processes. De Josselin de Jong (1958) visualizes a porous
2 SPE 113343
medium as randomly interconnected straight channels of equal sizes, orientated at random, uniformly distributed in all
directions, in which average uniform flow takes place. Saffman (1959) studied a similar geometric porous structure, in which
capillaries are orientated randomly. He accounts not only for the probability of a particle choosing a capillary of random
orientation but also for the velocity distribution within a capillary, i.e., the probability that a particle chooses a streamline in
that capillary. Greenkorn and Kessler (1970) extended the model to the case of non-uniform media by the use of a two
parametric density function, beta distribution, for both the radius and for the length distributions. However, in this work a
four parametric probability density function has been used to express the pore size distribution of the micromodel which is
measured using image analysis technique.
Micromodel is a small scale artificial model of porous medium which is known as a novel approach for simulating flow
and transport in porous media. This provides insight to the pore scale interplay of various aspects of transport phenomena. It
has been proved that microscopic visualization of the micromodel provides the opportunity to discover yet unrecognized
process and to enhance the understanding of existing theories and assumptions (Wilson, 1994). Micromodel studies reported
in the literature vary widely in their methods and applications (Buckley, 1991). Mattax and Kyte (1961) made the first etched
glass network, but this approach was significantly improved by application of photo etching technique (McKellar and
Wardlaw, 1982). Almost all researchers used micromodel to obtain greater understanding of transport mechanism visually,
and to the best of our knowledge, no attention or only little attention has been paid to quantify the static observation of pore
size characteristics for estimating of dispersion in micromodel.
Standard numerical techniques such as the finite difference and finite element methods, when applied to solve solute
transport problems, impose some restrictive spatial discretization requirements to avoid numerical dispersion which
artificially smears concentration fronts in simulation results (Neuman, 1984; Chiang, et. al., 1989; Wolfsberg and Freyberg,
1994). Varieties of approaches have been developed to overcome numerical dispersion. One effective approach is the random
particle method (Uffink, 1990; Prickett, et. al., 1981; Kinzelbach, 1988; LaBolle, et. al., 1996; Bijeljic and Blunt, 2006;
Hassan and Mohamed, 2003). A random walk is generally defined as the movement of a particle (walker) that will undergo a
displacement with a magnitude that depends on the chance. The irregularity and randomness of the grain skeleton of a porous
medium make it impossible to fully describe the solute displacement in a deterministic fashion. As long as the movement of
solute particles is unpredictable, it is useful to consider all possible displacements and the probabilities they are realized
(Uffink, 1990). Here, random particle method was applied to a simple statistical non-uniform network model, and the
longitude and transverse dispersion of micromodel was estimated.
In this work a glass type micromodel is considered as porous medium which its grains and pores are non-uniform in size,
shape and distribution. The network of pores in the porous model is regarded as a network of random capillaries with
diameters and lengths governed by probability distribution function. The model can be made anisotropic by distributing the
orientation angle. The macroscopic properties, e.g. dispersion coefficients and permeability, of flow are related to the pore
structure of micromodel. In order to model and then to estimate the dispersion in flow through porous medium, the pore size
and pore length distributions of micromodel are measured using image analysis technique. A parameteric probability
distribution function was used to express the measured distributions. The random particle method was applied for estimation
of dispersion coefficients.

Statistical Distributions
In general, pore size distribution can be described in terms of cumulative distribution function, which is equal to
probability that pore diameter is equal to or less than D, or in terms of probability density function, which is equal to the
derivative of cumulative distribution with respect to pore size. In practice, one can measure pore size distribution function
giving the fraction of pore space which has a pore diameter greater than a given value. The basic empirical statistical
model, ) (D f , used in this study is based on a parametric probability density function (Ghazanfari, et. al., 2007) as follows,

n
D D
n
n
e
D D n
D f

min 1
min
) (
) (
max min
D D D
(1)
Where , and n are adjustable parameters, and might be related to physical properties of porous medium. More details
about the physical significance of model parameters and the behavior of proposed model by changing the parameters are
given elsewhere (Ghazanfari, et. al., 2007).
min
D and
max
D are minimum and maximum pore throat size, respectively. If
max
D goes to infinity, parameter becomes equal to unity. Therefore, 1 = corresponds to the threshold capillary
pressure being equal to zero. It is assumed that pore length, l , distribution also obeys probability distribution function of
equation (1) as presented by equation 2.



SPE 113343 3
n
l l
n
n
e
l l n
l f

min 1
min
) (
) (
max min
l l l
(2)

where,
min
l and
max
l are minimum and maximum pore length, respectively. The parameters of equations (1) and (2) are
calculated by fitting the statistical model to the measured data of pore throat, pore body and pore length distributions of
porous medium. Most of the probability distribution function models are two-parametric model. This causes lack of
flexibility to simulate natural petrography phenomena and particularly they suffer from the shortcoming of infinite range for
the pore size and pore length. Contrary, equations (1) and (2) allow finite size of pore diameter and pore length, respectively.

Micromodel Pattern
The glass micromodels are mostly fabricated by etching the desired pore network pattern on two plates of mirror glass
which are then fused together. Using this method, highly intricate and detailed patterns can be etched with the dimensions of
pores and throats as low as a few microns. Details of the model production procedure are given elsewhere (McKellar and
Wardlaw, 1982; Ghazanfari, 2007). The micromodel used in this study included grains and pores which are non-uniform in
size, shape and distribution. Figure 1 shows the micromodel which is saturated with the blue colored water. In the black and
white prints the color of water and grains are gray and white, respectively.



Figure 1. Micromodel pattern saturated with blue colored water


Micromodel Properties
Details about the micromodel setup and method of measurements of micomodel properties are given elsewhere
(Ghazanfari, 2007; Ghazanfari, et. al., 2006). Because of the relevance, a brief description is presented here. The permeability
and pore volume of micromodel were determined using conventional methods. The porosity was measured using image
analysis technique. The average etched depth was calculated using measured values of pore volume and porosity. The
physical and hydraulic properties of micromodel are shown in Table 1.


Table 1. Physical and hydraulic properties of micromodel
Length (cm) 7.5

Width (cm) 1.2
Absolute permeability (D) 13.3 Average areal porosity 0.485
Average depth (micron) 32 Pore volume (cm
3
)

0.014


Micromodel Pore Size Characteristics
Details of the method of measurement applied for characterization of pore throat, pore body and pore length of the
micromodel presented in Figure 1 are given elsewhere (Ghazanfari, et. al., 2007; Ghazanfari, 2007). Because of the
importance and its relevance to the materials of this paper, it is preferred to repeat the summary of the procedure here as well.
The grains and pores of micromodel shown in Figure 1 are non-uniform in size, shape and distribution. This non-
uniformity of geometry and topology of the microscopic pores control the fluid content and transport properties of the porous
medium, e.g., dispersion and permeability. Therefore, it is important to measure the pore size characteristics of porous model.
A pore can be described in terms of its diameter, length and orientation in a flow field. In systems of interconnected pores,
this implies a partitioning of pore space into pores body, pores throat and pores length. In fact, pore structure of porous
medium is described by the pore size distribution. A pore size distribution curve represents the cumulative fraction of total
pore volume within porous media sample made up by particular ranges of pore sizes.
Dimensions of pore throat and pore body as well as pore length of porous model are computed from data obtained by
two dimensional image analysis. The area of pore body and the width of pore throat within a representative region of the
micromodel were measured by the image analysis of porous model shown in Figure 1. An important condition for successful
application of image analysis in the statistical analysis of the porous model is a sufficient contrast between pores and grains.
Employing the micromodel average etched depth, the total volume of the pore body was calculated by multiplying the planar
pore area by the average edged depth. An average diameter was determined using an equivalent sphere the volume of which
4 SPE 113343
was equal to the volume of the pore body. To determine the pore throat size, the width of the pore throat was measured and
the cross-sectional area was obtained by multiplying this by the average edged depth. An average pore throat diameter was
determined from the area of the circle equal to the cross sectional area of a pore throat. The pore length is the distance
between two connected pore bodies along the flow field. To present the exact values of the measured pore size
characteristics, the frequency distribution diagrams of pore throat diameter in the form of ) (D f versus
min
D D and pore
lengths in the form of ) (l f

versus
min
l l are plotted as given in Figures 3a,b. These figures also provide the minimum,
maximum and average values of pore throat, pore diameter, and pore length.



Table 2. Fitted parameters derived from pore throat diameter and pore length distribution models,
and the measured minimum, maximum and average values.

n

,
m


) .( m Min

) .( m Max

) .( m Ave

Pore throat 5.1 55 1.07 4 80 41.0
Pore length 1.6 50 1.11 5 165 49.4


The parameters , and n for the pore throat size distribution model were estimated by fitting the statistical
model ) (D f of equation (1) to the measured data of pore throat size distribution as given in Figure 2a. The same parameters
for the pore length distribution model were estimated by fitting the statistical model ) (l f of equation (2) to the measured data
of pore length distribution as presented in Figure 2b. The measured minimum, maximum and average values of pore throat
diameters and pore lengths and the fitted parameters of distribution models are given in Table 2. The plotted solid line in
Figures 2a and 2b is the behavior of the statistical distribution functions of equations (1) and (2) with the models parameters
of the Table 2.
The frequency distribution diagram of pores body of the micromodel is not shown in Figure 2. The measured pore body
distribution as well as pore throat and pore length distributions were used for generating a physically-based network model of
the micromodel which have been used as input for pore-scale simulation of the micromodel. The developed pore-scale
simulation model was used for prediction of longitudinal dispersion in micromodel as well as prediction of relative
permeability of the drainage process in the micromodel. Details of this development are given elsewhere (Ghazanfari, 2007).


(a) (b)
Figure 2. Frequency distribution diagrams (a) pore throat size (b) pore length -- solid lines are the fitted
statistical model behavior.

Network Model
One of the difficulties in calculating the transport properties of flow through porous media is that the detailed geometry
of the media is not usually known and the flow field can not be computed in detail. Therefore, theoretical method is at the
present confined to the investigation of models which can be handled mathematically. A pore can be described in terms of its
diameter, length and orientation in a flow field. The pore space model is approximated by a random, statistically anisotropic,
network of straight cylindrical capillaries of length l and diameter D, with several capillaries starting and finishing at each
junction. The porous model considered here is two dimensional. The pore element in this network which is presented in
Figure 3 is oriented by an angle from the y axis. It is clear that 0 for the case of flow in the xdirection.
SPE 113343 5


Figure 3. Pore element description

Assuming that the size, length and orientation of pore elements are independent, then the distinct probability distribution
functions of pore size, ) (D f , and pore lengths, ) (l f , are as equations (1) and (2), respectively. The model is conceived as
a large number of randomly intersecting pore elements. The probability of a given pore to exist with the size and orientation
in the range of l to dl l + , Dto dD D + , and to d + is given by the product of the independent probabilities such
as,

d dD D f dl l f dP sin . ) ( . ) (
2
1
= (3)

Where the coefficient of 1/2 is normalization factor, and is equal to 2 / 1 for the case of 3D pore element. The probability
function shown by equation (3) is later used for calculating longitudinal and transverse dispersion coefficients and
permeability of the micromodel.

Permeability
The permeability of the model is found by relating the average velocity in a pore element to the average velocity in the
ensemble. Consider the bulk flow through the model is only in the x direction, and consider the Reynolds number of flow
through capillaries is small, then according to the Hagen-Poiseuille equation,

sin
32
2
x
p D
v

=

(4)

The components of this velocity are,

cos
sin
v v
v v
y
x
=
=

(5)

The average velocities in each direction for the ensemble of pores are found by integrating the velocity components given by
equation (5) over the entire range of pore size and orientation such as,

=
=
P
y
P
x
dP v v
dP v v

cos
sin

(6)

where, dP is given by equation (3) and integration limits are the range of l D, and . For the case of the bulk flow through
the model being only in the x direction, integration of the velocity in the y direction yields 0 =
y
v . The average velocity in
the xdirection is,

x
p D
v
x


=
48
2
(7)
6 SPE 113343

Where
2
D refers to the second moment of pore throat diameter. The average pore velocity for a fluid flowing in porous
medium is given by Darcy law as,

x
p k
V


(8)

where, p V k , , , , are permeability, viscosity, porosity, Darcy velocity and pressure, respectively. Since
x
v V = ,
combination of equation (7) and equation (8) yields,

48
2

=
D k

(9)

It should be mentioned that without considering the statistical concepts, the factor 1/48 on the right hand side of equation (9)
becomes 1/32 which is a result of direct combination of Darcys law and Hagen-Poiseuille equation (Dullien, 1992).
The permeability-porosity ratio is a function of the average pore diameter squared. Therefore, it depends on the pore size
distribution function. Thus, the permeability-porosity ratio causes dissipation due to entrance-exit effect. Since the flow in
porous media is controlled by pore throat size, substituting the average pore throat diameter of porous model, 41 m , which
is measured experimentally, into equation (9) results in the absolute permeability-porosity ratio equal to 35.02
2
m . It is in
a good agreement with the experimentally measured value of 27.79
2
m (27.42 D), and the error is 26.0%. Absolute
permeability calculations without considering the statistical concepts gives the absolute permeability-porosity ratio equal to
52.53
2
m producing an error of 89.0% in comparison with the measured data. The foregoing discussion leads to an
interesting possibility to determine the average pore throat size of porous model using permeability and porosity
measurements with equation (9).

Dispersion
The macroscopic transport of a non-reactive solute in a porous medium can be described by three main transport
processes: the advective motion with the mean velocity field of the flow, the molecular diffusion and the hydrodynamic
dispersion of the solute. On the pore scale, the dispersive flux is due to sub-scale variations in velocity caused by the varying
thickness of pores, the bending of streamlines around the grains and the variation of the velocity profiles within the pores. On
the macroscopic scale, dispersion is caused by in-homogeneities of the media. Dispersion is always anisotropic even in the
case of an isotropic medium and at least one order of magnitude larger in flow direction than orthogonal to the flow direction
(Bear and Bachmat, 1990). The convection-diffusion equation, mass conservation law, has been commonly used (Bear, 1972)
to describe solute dispersion in porous media,

t
C
y
C
v
x
C
v
y
C
D
x
C
D Q
y x T L

+
2
2
2
2

(10)

where C is the solute concentration, t is time, Q is a source term representing the quantity of solute injected in the system,
x
v and
y
v are the components of flow velocity,
L
D and
T
D are the

hydrodynamic longitude and transverse dispersion
coefficients, respectively. It has been proven that the y-component of flow velocity for the points that are not much far from
the injection source would be negligible, so the convective term associated with the
y
v in equation (10) can be omitted. In
the other words, it would be reasonable if it is assumed that only dispersion process happens due to convection of x-
component of flow velocity (Ghazanfari, 2007).
The solution of the convection-diffusion equation (Equation 10) is difficult due to the co-existence of spatial first and
second order terms. The first order term, which describes the advective motion, introduces a hyperbolic character, whereas
the second-order term, which describes the diffusive/dispersive motion, introduces a parabolic character. Up to date, there
exists no technique which yields satisfactory results under general hydrogeologic conditions and for practically feasible
discretization. Numerical solution techniques can be classified as Eulerian, Lagrangian and mixed Eulerian-Lagrangian. In
the well-established Eulerian approach, equation (10) is solved on a stationary grid. Common solution techniques are finite
difference, finite element or finite volume methods. These kinds of methods are well suited for solving parabolic equations
like the flow equation or the diffusion equation. However, for advection-dominated problems grid-based methods suffer from
SPE 113343 7
numerical dispersion and numerical oscillations. To avoid these problems, severe stability constraints restricting the grid
spacing and the size of the time step have to be met. One of the effective approaches to overcome this problem is the random
walk particle method. The adaptivity of particle methods and the straightforward physical interpretation of their results make
them a promising alternative to established grid-based methods in the field of solute transport dynamics in groundwater. The
interest in particle methods is big due to the fact that accurate solutions not suffering from numerical dispersion and artificial
oscillations and they can be obtained at a competitive computational cost.
The random particle is a marked fluid particle as it wanders through the porous medium selecting elemental pores for
each step according to the prescribed probability function. Each passage of a particle through an individual pore is a step in
random walk. The properties derived here will be fulfilled after a particle has completed a very large number, n, of
statistically independent steps. At each junction, the probability of path choice must be related to the probability of the
existence of a pore given by equation (3), and to the fluid velocity at a junction. The possibility that we are considering is that
the choice is proportional to the velocity, v , given by equation (4). The probability of path choice is assumed as,

dP
M
v D
dP
M
q
dE
4
2

= =
(11)

where, qis volumetric flow rate. Substituting equations (3) and (4) into equation (11), after applying the normalization (since
the integral over dE must equal unity) one can get,

d dD D f D l d l f
D
dE
2 4
4
sin . ) ( ). ( ) ( .
2

= (12)

Displacement of a marked particle after n steps is a random variable with components parallel to the axes,

=
=
E
n
E
n
dE l n Y
dE l n X

cos
sin
(13)

Substituting equation (12) into equation (13), and performing integration results in 0 =
n
Y , 3 8 = l n X
n
, and the time
for n steps in random variable,


= =
E E
n
dE v l tdE n T ) ( (14)

By substituting equations (4) and (12) into equation (14) and performing integration, one gets
=
4 2 2
3 8 D v D l n T
x n
. This leads to the result that a particle following the most probable path is transported
through the porous medium with a velocity, V D D v T X
x n n
= =
2
2 4
, close but not exactly equal to the Darcy
velocity. For randomly orientated uniform and isotropic network model the factor 3 8 changes to 3 2 . Since the variance
of X is the sum of the variances of the longitude displacement of the individual steps; hence,

( ) ( )
2 2 2 2 2
2
) sin sin ( ) (
x
E E
n n
l n dE l l n dE x x n x x n X X = = = =


(15)
Substituting equation (12) and integrating result in, ( )
2 2 2 2
9 64 4 3 = l l
x
. Similarly, the variance of Y is the
sum of transverse displacement of the individual particle (since 0 =
n
Y ),
( ) ( )
2 2 2 2 2 2
2
) sin (
y
E E
n n n
l n dE l n dE y n y n Y Y Y = = = = =


(16)
This yields,
2 2 2
4 = l l
y
. Also, the variance of T is ( )
2 2 2
) (
T x n n
v l n T T = and the covariance of X

and T

8 SPE 113343
is
2 2 2
) ( ) )( (
xT x n n n n
v l n T T X X = . Let us define the following dimensionless coordinates and time,
2 1
n l
X X
n n


= (17)
2 1
n l
Y Y
n n


= (18)
x
n n
v n l
T T
2 1


= (19)
It should be noticed that ,

and have means equal to zero and variances equal to
2 2
,
y x
and
2
T
, respectively. It
follows from Central Limit Theorem of statistics that the probability distribution of , are asymptotically normal and
statistically independent as n . So far we have examined the displacement and time after nsteps, where n is a fixed
number. In order to calculate the dispersion, it is necessary to know the probability distribution of the displacement after a
given time T in which T is in general large compared with
x
v l . Substituting the derived value of
n
T in equation (19) we
get,


=
4
2 2
2
1
2 1
3
8
D
D n
v n l
T
x


(20)
The subscript n in T
n
is dropped for the sake of simplicity.
The longitude and transverse dispersion coefficients are defined in terms of variance of the longitude and transverse
displacements and the time required for such displacements as,

T
T V X
D
L
2
) (
2

=
(21)
T
Y
D
T
2
2
=
(22)

where, the
2
) ( T V X and
2
Y are the variance of the average displacements in the x and y directions, respectively.
Combining equations (18) and (20), denoting the transverse displacement after time T byY and solving for Y result in,

2
1
2 2
4
2 2
4 2
1
2 2
4
8
3
8
3
2
1
8
3




= T l v
D
D
l
D
D
T l v
D
D
Y
x x

(23)
It should be noted that, 0 =
n
Y and the subscript n in Y
n
has been dropped for the sake of simplicity. Then,
2
2 2
4 2
8
3




= l v
D
D
T
Y
x

(24)
The expression for the longitudinal dispersion coefficient is obtained by substituting the value of
2
y
, variance of , into
equation (24). This produces,

SPE 113343 9
x T
v
l
l
D
D
D




=
2
2 2
4
64
3
(25)
For an ensemble of randomly oriented uniform and isotropic pores the factor 64 3 in equation (25) changes to 16 3 .
Substituting the value of
n
X into equation (17) and denoting the transverse displacement after time T by X ,
+ = l
n
n l X

3
8
2
1
(26)
Combining equation 18 with equation 26 and after some manipulating one can get,
( )



=
2
1
2 2
4
2 2
4
8
3
T l v
D
D
T v
D
D
X
x x

(27)
or,
( )

2
1
2 2
4
2
1
8
3
l v
D
D
T
T V X
x

(28)
and then,
[ ]

2
16
3
2
) (
2 2
2 2
4 2
+


=

x
v l
D
D
T
T V X

(29)
which is equal to,
[ ]
xT x T x
v l
D
D
T
T V X

2
16
3
2
) (
2 2
2 2
4 2
+


=


(30)
The covariance can be found as,

dE t t x x
l
v
v l n
t t x x
n
v l n
T T X X
E
x
x x
n n n n
) ( ) (
) )( ( ) )( (
2 2 2


=


=


=

(31)
After performing integration,



=
4
2 2
2 2
2
9
64
3
2
D
D
l
l

(32)
The Central Limit Theorem can not be applied to the variable evaluated for each particle. It is then not possible to
prove that T T V X 2 ) (
2
tends asymptotically to a value independent of T as is usually the case for random processes
occurring in the nature. It might be mentioned that T should be sufficiently large. What are the statistical properties of the
displacement of a marked fluid particle whose velocity is given by equation (4) for all value of t ? To answer this question, it
is necessary to get help from the statistical properties of the random walk in a more detail. The probability distribution of
is given by Markoff's theorem (Chanderarasekhar, 1943).

10 SPE 113343

+

= '
2
1
) (
'
dt Ae F
it

(33)
where, A is characteristic function of defined as,
dE
n l
t t v
it A
E
x
n



=
2
1
1
) (
' exp (34)
As a first order approximation,


=
2
2
) (ln 2
exp
ln 2
1
) (
A A
F

(35)
For being normally distributed with mean zero and variance
2
T
defined by equation (36),

= =
l
T v
D
D
D
D
A
x
T
.
8
3
ln
7
4
3
8
) (ln
2
2 2
4
2
4
2 2
2 2

(36)
The expression for the longitudinal dispersion coefficient is obtained from equations (30), (32) and (36) and the value of
2
x

as,
x
x
L
v l
D
D
l
T v
D
D
D
D
D
D
D
D
l
l
D




=
2 2
4
2
2 2
4
2
4
2 2
4
2 2
2 4
2 2
2
2
16
3
.
8
3
ln
21
32
1
2
9
64
3
4
4
3



(37)
Both dispersion coefficient expressions, equations (25) and (37), are definitely depending on the probability distribution
function of the model. One condition which must be satisfied for this model to be applicable to the actual dispersion of a
material quantity is that the amount of dispersion that takes place being large in comparison with the dispersion due to
molecular diffusion acting alone under static conditions. That is, the effective diffusivity must be large in comparison with
the molecular diffusivity. Molecular diffusion is driven by concentration gradients and is important only when flow velocities
are too small.



SPE 113343 11

(a) (b)
Figure 4. Comparison of predicted dispersion coefficients (solid lines) with
(a) experimental data of the micromodel, (b) experimental data of glass beads porous media

The model results for longitude and transverse dispersion as a function of fluid velocity are plotted in Figure 4a together
with experimental data of micromodel (Ghazanfari, 2007), and in Figure 4b together with the experimental data of 0.025 cm
glass beads (Koch and Barady, 1985). In the latter case, the average pore length is assumed to be one-half of the bead
diameter. It is obvious that the prediction of dispersion coefficients using equations (25) and (37) are in close agreement with
the measured data. For the experimental data presented here, molecular diffusion becomes controlling mechanism for the
fluid velocities lower than 0.001 cm/s. That is why no data were compared with the model prediction for the fluid velocities
below this level. As mentioned earlier, the model assumes that the amount of dispersion that takes place being large in
comparison with the dispersion due to molecular diffusion acting alone. Similarly, there is a limiting velocity (0.01 cm/s for
the data presented here) below which diffusion controls the transversal dispersion.

Discussions
In the random particle method, the transport equation is not solved on fixed grid but on individual particles. Since
particles are not lost nor destroyed, the method conserves mass exactly. The underlying assumption is that a particle with a
sufficiently long path length in a specific system will experience all the conditions encountered in the ensemble of systems,
and hence a temporal average over a single path is equally representative of a spatial average over an ensemble of path
lengths in different systems. Dispersive transport is simulated by tracking the particles along their trajectories. Such a method
is physically appealing, stable and easy to implement. However, in many applications a grid may be employed for the
definition of the velocity and dispersion characteristics in the modeled domain. In addition, interpretation of the particle
distribution as a concentration field may require a grid or some other means for converting the spatial distribution of particles
to concentration values. Also, one may argue that, as the detail description of grain skeleton is not at hand, little information
is available on the probability distribution of particle displacement. This seems even more complex.
The close agreement observed between prediction of dispersion coefficients and experimental data shown in Figures 4a
and 4b, when dispersion is large in comparison with the molecular diffusion, confirms that the assumptions of theory and
statistical network model, even though is simple, are not unreasonable. When molecular diffusion being negligible, empirical
equations for transverse dispersion coefficient shows that this coefficient is proportional to the average velocity and pore
space characteristics. This is what essentially shown by Equation (25). The proportionality of the longitude dispersion
coefficient with the product of the average velocity and its logarithm shown in equation (37), has also been observed by other
investigators (Saffman, 1959; Koch and Barady, 1985). Also, as mentioned earlier, Equation (9) indicates that the
permeability-porosity ratio is a function of average pore diameter squared and the foregoing discussion leads to an interesting
possibility to determine average pore throat size of porous media using permeability and porosity measurements. That is,
measurements of dispersion coefficients may reveal more information about the pore size characteristics.
It is well known that dispersion is always anisotropic even in the case of an isotropic medium and at least one order of
magnitude larger in flow direction than orthogonal to the flow direction. This is essentially what is shown by equations (25)
and (37) as plotted in Figures 4a and 4b. In general, for unconsolidated porous media it is expected that the ratio of longitude
to transverse dispersion varying from 3 to 10 (De Josselin de Jong, 1958; Grane and Gardner, 1961). Reported literature data
(Grane and Gardner, 1961) reveal a slightly velocity dependence of the ratio of the longitude to the transverse dispersion
coefficients. This is essentially what is concluded by dividing equation (25) by equation (37) and it can also be deduced by
observing the plots presented in Figures 4a and 4b.
The derivation presented here assumes that the direction of successive pores through which a random particle moves are
statistically independent, and there are reasons based on continuity arguments for believing that this assumption may not be
valid for the transverse dispersion, although reasonable for the longitude dispersion. The transverse dispersion may therefore
12 SPE 113343
be up to one order of magnitude less than that given by equation (25).

Conclusions
In this study a micromodel which its grains and pores are non-uniform in size, shape and distribution is considered as
porous medium. The transparent nature of the micromodel makes it potentially attractive for studying the pore-scale transport
phenomena. In order to relate the macroscopic properties, e.g. dispersion and permeability, to the structure of porous model,
the pore size as well as pore length distributions of micromodel were measured by applying an image analysis technique. A
physically-based two-dimensional random network model of the micromodel has been constructed which the non-uniformity
is governed by the measured probability distribution functions. The random particle method was then applied to correlate and
to predict dispersion coefficients based on probabilistic approaches. Predictions from the derived models for both longitude
and transverse dispersion coefficients are in close agreement with the experimental data under the condition that the
molecular diffusion is negligible. Also, an estimation of permeability-porosity ratio was presented which was in agreement
with the measured data. Despite the simplicity of the proposed network model, its accurate predictions provide some
confidence that it can be considered as a reasonable approximation of the complex nature of porous media with reliable
predictive capability.

References
Aris, R.1956. On the Dispersion of a Solute in a Fluid Flowing through a Tube. Proc. R. Soc. London, Ser. A 235, 67.
Bear, J. 1972. Dynamics of Fluids in Porous Media. American Elsevier, New York.
Bear, J. and Bachmat, Y. 1990. Introduction to Modeling of Transport Phenomena in Porous Media, Kluwer
Academic Publishers, Dordrecht.
Bijeljic B. and Blunt M. 2006. Pore Scale Modeling and Continuous Time Random Walk Analysis of Dispersion in
Porous Media. Water Resour. Res. 42W01202, doi:10.1029/2005WR004578.
Buckley, J. 1991. Multiphase Displacement in Micromodels, in Interfacial Phenomena in Petroleum Technology,
edited by N. Morrow, Marcel Decker, New York.
Chanderarasekhar, S.1943.Stochastic Problems in Physics and Astronomy. Rev. Mod. Phys. 15, 1.
Chiang, C. Y., Wheeler, M. F.and. Bedient, P. D. 1989. A Modified Method of Characteristics Technique and Mixed
Finite Element Method for Simulation of Groundwater Solute Transport. Water Resour. Res. 25, 1541.
Davis J.A. and Jones, S.C.1968 Displacement Mechanisms of Residual Solutions. J. Petrol. Technol. 20, 1415.
De Josselin de Jong, G. 1958. Longitudinal and Transverse Diffusion in Granular Deposits", Trans. American
Geophys. Union 39, 67.
Dullien F.A.L. 1992. Porous Media: Fluid Transport and Pore Structure, 2nd Edition, Academic Press, New York.
Fatt, I. 1956a, b, c.The Network Model of Porous Media, I, II, III, "Capillary Pressure Characteristics; Dynamic
Properties of a Single Size Tube Network ; Dynamic Properties of Networks with Tube Radius Distribution. Trans. AIME.
207, 144.
Ghazanfari, M.H., Rashtchian, D., Kharrat, R., Voussughi S.2007. Capillary Pressure Estimation of Porous Media
Using Statistical Pore Size Function. Chem. Eng. Tech. 30, 862.
Ghazanfari, M.H. 2007. Prediction of Multiphase Flow Properties in Porous Media Using Micromodel Experiments
and Pore-Scale Modeling. PhD Thesis, Sharif University of Technology.
Ghazanfari, M.H., Khodabakhsh, M., Rashtchian, D., Kharrat, R., and Vossoughi, S. 2006. Unsteady State Relative
Permeability and Capillary Pressure Estimation of Porous Media, XVI International CMWR Conf., Denmark.
Grane.F.E. and Gardner, G.N.F. 1961. Measurements of Transverse Dispersion in Granular Media. J. Chem. Eng.
Data 6,283.
Greenkorn, R.A. and Kessler D.P. 1970. Dispersion in Heterogeneous, Non-Uniform Anisotropic Porous Media, in
Flow through Porous Media, American Chem. Soc., Washigton, D.C.
Hassan, A.E. and Mohamed, M.M. 2003 On Using Particle Tracking Methods to Simulate Transport in Single-
Continuum and Dual Continua Porous Media. J. Hydrology. 275, 242.
Kinzelbach, W. 1988. The Random Walk Method in Pollutant Transport Simulations, in Groundwater Flow and
Quality Modelling edited by E. Custodia, G. A. NATO ASI series C. Math. Phys. Sci. 224, 227.
Koch. D.L., and Barady, D.F. 1985. Dispersion in Fixed Beds. J. Fluid Mech. 154, 399.
LaBolle, E. M., Fogg, G. E.and Tompson, A. F. B.1996. Random-Walk Simulation of Transport in Heterogeneous
Porous Media: Local Mass Conservation Problem and Implementation Methods", Water Resour. Res. 32, 583.
Man, H. N., and Jing, X. D. 2000. Pore Network Modeling of Electrical Resistively and Capillary Pressure
Characteristics, Transp. Porous Media. 41, 263.
Mattax, C. C. and, Kyte, J.R. 1961. Ever See Water Flood?. Oil Gas J. 59,115.
McKellar, M., and Wardlaw, N. 1982. A Method of Making Two-Dimensional Glass Micromodels of Pore Systems. J.
Can. Pet. Technol. 21, 39.
Neuman, S. P.1984 Adaptive Eulerian-Lagrangian Finite Element Method for Advection- Dispersion", Int. J. Numer.
Methods Eng. 20, 321.
SPE 113343 13
Prickett, T., Naymik, T. and Lohnquist, C. 1981. A Random Walk Solute Transport Model for Selected Groundwater
Quality Evaluations. Illinois State Water Survey, Bulletin 65.
Saffman, P.G. 1959. A Theory of Dispersion in a Porous Medium", J. Fluid Mech. 6, 321.
Sahimi, M.1993, Flow Phenomena in Rocks: from Continuum Models to Fractals, Percolation, Cellular Automata, and
Simulating Annealing. Rev. Mod. Phys. 65, 1393.
Scheidegger, A.E.1954.Statistical Hydrodynamics in Porous Media. J. Appl. Phys. 25, 994.
Taylor, G.I.1953. Dispersion of Soluble Matter in Solvent Flowing Slowly Through a Tube. Proc. R. Sot. London Ser.
A 219, 186.
Taylor, G.I.1954. Conditions under which Dispersion of a Solute in a Stream of Solvent Can Be Used to Measure
Molecular Diffusion. ibid 225, 473.
Uffink, G.1990. Analysis of Dispersion by the Random Walk Method. PhD thesis, Technische Universiteit, Delft.
Wilson, J. L. 1994. Visualization of Flow and Transport at the Pore Level, in Transport and Reactive Processes in
Aquifers, edited by T. H. Dracos and F. Stauffer, Rotterdam, The Netherlands.
Wolfsberg, A. V., and Freyberg, D. L.1994 Efficient Simulation of Single Species and Multi Species Transport in
Groundwater with Local Adaptive Grid Refinement. Water Resour. Res. 30, 2979.

You might also like