Permeability Estimation through Micro-Macro Coupling in

Particulate and Fibrous Porous Media

Bamdad Barari, Saman Beyhaghi and Krishna M. Pillai

Laboratory for Flow and Transport Studies in Porous Media
Department of Mechanical Engineering, University of Wisconsin-Milwaukee

1. Abstract
The closure formulation, developed as a part of the derivation of Darcys law proposed by
Whitaker [1], is used to develop a novel numerical method for estimating the permeability of a
porous medium with a given pore-level microstructure. The permeability of two distinct porous
media created from cellulose nano-fibers (CNF) and sintered polymer beads were then estimated
numerically. In order to use real micrograph in such simulations, 2D SEM pictures of the CNF
and polymer-wick porous media were considered. The falling head permeameter was used for
measuring the experimental permeability in order to test the accuracy of the permeability results
obtained by numerical simulation. The Permeability values were also compared with the
theoretical models of Kozeny-Carman and Rumpf-Gupte, and a good agreement was observed. A
good agreement between the numerical permeability results and the results obtained from the
experimental/theoretical means confirm the accuracy and usefulness of the proposed method in
estimating in this crucial property through an effective micro-macro coupling.
Keyword: Porous media, permeability, closure formulation, cellulose nano-fiber, polymer wick

2. Introduction
The problem of flow and transport of fluids in porous media has been studied extensively in the
last hundred years. In the last few decades, the volume averaging method has emerged as a
credible method to develop the macroscopic governing equations for modeling fluid flow in
porous media [1]. Using these equations, one can predict the volume-averaged velocity, pressure,
liquid density, and species concentrations inside any porous medium [2-5].
The basic idea in volume averaging is that the equations for flow and transport in porous
media can be extracted using spatial averaging with the help of representative elementary
volume or REV [6-8]. Efforts towards developing the volume averaging method were made by
several researchers during last few decades. Whitaker provided perhaps the most comprehensive
derivation of volume averaging theorems based on geometric ideas [9, 10]. Later, Whitaker [1]
outlined in detail the volume averaging method and development of the closure formulations for
Darcys law and mass transport in single- and multi-scale porous media. Gray et al. used a
generalized function to formulate and prove the volume averaging theorems [11]. The volume
averaging method as a powerful technique for upscaling was subsequently used by several
authors. Hassanizadeh and Gray [12] applied the volume averaging theorem for developing
continuum models for multiphase flows in porous media. Bennethum and Cushman [13] used the
multiscale hybrid mixture theory along with the volume averaging method to develop continuum
equations for swelling porous media. In general, the size of the REV must be much greater than
the nominal size of the solid particles and the characteristic inter-particle distances. Further
studies on upscaling was done by Gray and Miller [14] where averaging of quantities from
micro- to macro-scale using a thermodynamically constrained approach was conducted.

A great advantage of the volume average method is the development of an effective micro-macro
coupling, i.e., the coefficients of the macroscopic equations are derived from solving closure
formulations in unit cells of real porous media, thus incorporating the effects of pore-scale
geometry in the overall upscaled equations. As a result, physical properties such as tortuosity,
dispersion, and permeability are estimated after including the features of the microscopic flow,
transport and geometry. This approach was used to propose single- and two-equation models for
conduction and convection in porous media [15, 16]. A theoretical derivation of the governing
equations for the single-phase flow in swelling porous media was investigated by Pillai [17]
where novel forms of the mass balance (continuity) and the momentum balance (Darcy's law)
were developed using the volume averaging method. Bear and Bachmat [18] comprehensively
presented the volume averaging equations for single-phase, multi-phase and multicomponent
systems using some unique and rather complicated proofs of the averaging theorems. Dormieux
et al. [19] estimated the tortuosity after using the homogenization method by assuming the
porous medium to be consisting of parallel constant-thickness horizontal and vertical channels.
In most studies on scaling up done till now using the closure formulation, a simple square
shaped unit-cell with circular solid phase is most often considered [20]. Kim et al. used different
aspect-ratio rectangular unit-cells with rectangular inclusions for this purpose as well [21]. Davit
et al. [22] worked on local non-equilibrium models for mass transport in dual-porosity, dualphase porous media where they observed that the volume-averaging-method formulation with
the closure and the method of matching spatial moments were equivalent in the one-equation
non-equilibrium case. Golfier et al. [23] investigated mass transport in a porous medium
containing bio-film to develop the Darcy-scale equations and the effective dispersion tensor
using a closure formulation was solved for both cases of simple and complex unit cells.
3

The preforms made from cellulose nano-fibers (CNFs) were one of the two porous materials
used in this study. CNFs are made purely from cellulose molecules and have very good
mechanical property compared to other natural fibers, and even carbon or glass fibers. Recent
studies establish that the films or "nano-paper" made from CNFs are the strongest man-made,
cellulosic materials [24]. Recently, CNFs have begun receiving serious consideration as potential
reinforcement materials. The present research springs from the efforts by the authors on using
CNF preforms for making CNF composites by infiltrating them with thermosetting polymers
[25]. CNF preforms used in this study were prepared according to the technique developed by
Saito et al. [26].
The other porous medium used in this paper is the polymer wick. These kinds of wicks,
made from sintering polymer beads, are often used for delivering active liquids such as
fragrances and insecticides from a reservoir to ambient air. Such wicks have been used
previously in several research efforts: for example, in the estimation of the tortuosity tensor
using closure formulation in Beyhaghi and Pillai [27], in the development of a new model for
wicking based on the single-phase Darcys law after using the capillary suction pressure at the
moving liquid-air interface in Masoodi et al. [28].
In the present study, the closure formulation, developed as a part of the proof for Darcys
law by Whitaker [1], is used to measure the permeability through the two above-mentioned
porous media with realistic, irregular pore structures. Scanning Electron Microscopy (SEM) is
used to take micrographs of the cross sectional areas of the CNF preform and polymer wick. The
SEM micrographs are then transformed to binary pictures (containing only two colors: one
representing the solid phase, and the other the void region) and then they are imported into
COMSOL multi-physics software. The governing equations for the closure formulation along
4

with the assigned boundary conditions are solved within a unit cell placed at the center of the
micrograph. The permeability tensor can then be obtained from integrating the field of mapping
variables inside the unit-cell. In order to establish the independence of permeability tensor from
the unit-cell size and consequently find a proper size of the unit cell for our permeability
calculations, four concentric square-shaped unit cells with different areas are chosen. For the
CNF sample, an additional step is done. After estimating the size of the optimum REV, the
permeability is calculated as an average of the permeability obtained for four different REV
locations in the CNF micrograph. Finally, for both the CNF and wick specimens, the
permeability obtained by the numerical simulation is compared with the experimental
permeability obtained using the falling head permeameter and the theoretical/empirical formulae
for permeability from literature such as the Kozeny-Carman relation.
NOMENCLATURE
B

Closure or mapping variable for velocity

Vf

Pore volume or fluid volume

Closure or mapping variable for pressure

Velocity closure variable in terms of permeability
Pressure closure variable in terms of permeability
Arbitrary continuous function

V
x

REV volume

d
f
g

I
K

L
li

n
p

Position vector

R
Sfs
Sfe

Radius

Permeability tensor
Porous sample's thickness in experiment
i 1, 2,3 Lattice vectors
Unit vector normal to fluid-solid interface
Pressure

Greek Symbols
Arbitrary flow variable

Acceleration due to gravity

Unity tensor

Height of liquid during experiment

Gradient vector
Laplace operator
Fluid density
Fluid viscosity
Porosity

Subscripts
Fluid phase
f
fs fluid-solid
fe Fluid entrance-exit

fluid-solid interface area on REV

fluid entrance and exit area on REV

B
P

Burette
Porous

Time
Velocity

Initial

3. Theory
3.1. Permeability Estimation using Closure Formulation of Volume Averaging Method
The formulation developed by Whitaker [1], based on the volume averaging method, is used in
this theoretical model employed for estimating the porous-medium permeability. The flow
variables in the pore region of a representative elementary volume (REV) are averaged and the
values are used in the macroscopic flow-field. Two types of averaging can be defined for an
arbitrary flow variable as follow.
Phase Average:

1
V

Vf

dV

(1)

Intrinsic Phase-Average (also known as Pore Average):

1
Vf

Vf

dV

(2)

Here V represents the REV volume and V f represents pore volume (or fluid volume in case of
single-phase flow) within the REV of a porous medium. For low-Reynolds number flows, the
boundary value problem for solving the velocity and pressure fields associated with the Stokes
and continuity equations can be expressed as

0 p f f g 2 v f

(3)

.v f 0
B.C.1 v f 0 on S

(4)

B.C.2 v f f (r, t ) on S

(5)

fs

fe

(6)

where S

fs

represents the fluid-solid interface area while S

fe

stands for the fluid entrance and

exit areas in the REV. Using the volume averaging method [1], one can apply the volume
averaging theorems to all terms of Eqns. (3) and (4) to obtain the macroscopic momentum
balance equation in the form of Darcys law. During this process of upscaling, the pressure and
velocity terms are decomposed in terms of their intrinsic phase averages, and , and
corresponding perturbations, p f and v f , as

pf pf

pf

(7)

vf vf

vf

(8)

The use of Eqns. (7) and (8) during the volume averaging process results in the following
equation set to solve for pressure and velocity perturbations:

0 p f f 2 v f

1
Vf

n
.

I
p

v
dS
fs

f
f
f

S fs

.v f 0

(9a)

(9b)
f

B.C.1

vf vf

on S

fs

(9c)

B.C.2

v f f (r, t ) on S

fe

(9d)

Note that Eqns. (9a) and (9b) result from the momentum and continuity equations. In order to
close the system of equations, p f and v f are defined in terms of the corresponding averaged
quantities as

pf f bf . vf
vf Bf . vf

(10)

On substituting Eqn. (10) in Eqn. (9a) and (9b), the governing equations for the closure
variables, and can be developed as

0 b f f 2 f

1
Vf

n
.

I
b

dS

fs

f
f
S fs

f 0

(11a)

(11b)

The closure formulation allows the permeability tensor to be estimated in terms of the closure
variables as
1
Vf

n . I b
fs

f dS f K 1

(12)

S fs

Use of this definition in Eqn. (11a) results in

0 b f 2B f f K 1

(13)

At this point, b f and B f can be replaced by d f and D f [1] through the transformations

d f f 1b f .K
D f f 1B0f .K

(14)

B0f B f I
On using the expressions given by Eqn. (14) in Eqn. (13) transforms the pore level momentum
equation as

0 d f 2D f I

(15)

and the pore level continuity equation can be stated as

D f 0

(16)

Note that the closure formulation described by Eqns. (15) and (16) is similar to the Stokes
boundary value problem that can be solved in an REV [1].
In Stokes flow, the pressure and velocity fields within the REV are controlled by
boundary conditions Eqns. (5) and (6) at the fluid-solid interface and at the entrance and exit
regions of REV. The first boundary condition on fluid-solid interface, Eqn. (9c), in terms of the
closure variables can be expressed as

B.C. 1

Df 0

on S

fs

(17)

The second boundary condition given by Eqn. (9d) is problematic since it entails the use of an
unknown function, f, to assign deviations at the entrance of the REV [1]. The problem was made
tractable by considering a periodic unit cell instead of a full-fledged REV. The corresponding
periodicity conditions for the velocity and pressure deviations can be described in terms of the
closure variables as
D f (r li ) D f (r )

(18a)

d f (r li ) d f (r )

(18b)

These periodicity conditions state that the opposite sides of a square-shaped unit cell have the
same values of the D and d components in the considered 2-D domain. Here r represents the
position vector of any point on the unit-cell boundary, while li are the lattice vectors with
{1, 2} that express the spatially periodic nature of the unit cell in the 2-D space. Enforcing
the velocity deviations to add to zero within the REV results in the average-value condition of
9

the form
Bf

(19)

This is consistent with definition of the spatial deviation as defined in the volume-averaging
formulation [1]. On using this result with the definition of permeability tensor stated in Eqn.
(12) and the definitions of the closure variables given in Eqn. (14), we can estimate the
permeability tensor in much simpler form as
K f Df

(20)

The mapping variable D f is first estimated within the unit cell using the closure formulation
given by Eqns. (15) to (19), then D f is integrated in the pore region of the unit-cell to obtain K
through Eqn. (20).
3.2. Experimental method for estimating permeability
The falling-head permeameter is employed to measure the permeability of porous medium using
a liquid level at the inlet that is gradually decreasing with time due to the leakage of liquid
through the sample. As shown schematically in Fig. 1, the continuously decreasing head at the
inlet implies that the inlet pressure driving the flow is decreasing. As a result, the Darcy velocity
inside the porous sample is decreasing with time as well. If the initial height of water column in
the burette is x0 at time t = 0, and the height at each subsequent time t is x during the
experiment, then the permeability of the porous sample, K, can be estimated through the formula

K gR p 2
x0
ln( )
t
x
L p RB 2

10

(21)

where L p and R p are the porous sample's length and radius, respectively; RB is the radius of the
burette; and are the liquid density and viscosity, respectively. If the ordered pairs of (x,t)
are recorded during the experiment and the ln(

x0
) vs. t plot developed, the permeability can be
x

calculated from the slope of the plot.

x0
x

Fig. 1. A Schematic of Falling-Head Permeameter

3.3. Theoretical/empirical model for permeability

As mentioned in many well-known texts on porous media, several permeability models (both
empirical and theoretical) have been developed for different types of porous media such as
particulate (soil-like or sand-like) media, fibrous media, etc. Dullein [29] categorized
permeability models into different types based on flow patterns and porous-media structure that
included particulate or fibrous beds, tube-flow models, network models, etc. Most of the
permeability formulas or models listed in literature are of the form

K Df 2 ( f )

11

(22)

where ( f ) is a function of the porosity and D f is the mean particle or fiber diameter. In
general, predictions from these correlations cannot match the measured permeability perfectly;
however, by choosing a model that fits the case, the obtained permeability values will be within
an order-of-magnitude of the measured values [30, 31]. In the present paper, the Cozeny-Carman
[29] and Rumpf- Gupte [29] empirical correlations are used to estimate permeability for the two
porous media considered, i.e., the polymer wick and the CNF porous sample. These relations for
the permeability are:

K Df 2
K Df 2

f3

Cozeny Carman

180(1 f )2

(23)

1
f 5.5 , K 1 for 0.35 f 0.7
5.6 K

Rumpf Gupte

4. Results and discussion

4.1 Initial validation of Closure Formulation Method
In order to evaluate the accuracy of our proposed simulation technique for estimating the
permeability using COMSOL, results of a recent study by Lasseux et al. on a similar
determination of the permeability of an idealized porous medium were used [32]. A simple
square-shaped unit cell (with side of length l) with two different square-shaped solid particles in
the center (representing porosities of 0.5 and 0.75) was considered. The set of closure equations
defined in Eqns. (15) and (16), subject to the boundary conditions Eqns. (17) and (18), were then
solved within the void regions of the two unit cells. As shown in Table (1), there is a very good
agreement between the dimensionless permeability values for the unit cells predicted in [32] and
in the present study. This clearly establishes the accuracy of our method.

12

Table (1): A comparison between the results of this study and those of Lasseux et al. in terms of
dimensionless permeability K* [32]

K* (= K/l2)
Porosity

Lasseux et al., 2011

Present work

0.5

0.002386

0.002301

0.75

0.013023

0.013022

4.2 Permeability of porous medium made from CNF

Most of the earlier efforts of using the closure formulation were confined to regular idealized 2D
or 3D geometries that could be recreated by repeating a unit cell. Kim et. al. [21] considered a
regular 2D porous medium made of a unit cell consisting of rectangular particles, while Quintard
[33] studied a 3D periodic medium with rectangular objects within a unit cell. It was observed
that the 3D model was closer to the experiments as compared to the 2D effort. In spite of all
these previous investigations with idealized porous media, the closure formulation based
approach was never tested on real porous media with irregular, non-ideal pore space.
Here we will use the closure formulation to estimate the permeability of CNF sample that
appears to be isotropic from micrographic studies. As seen in the SEM pictures of CNF (Fig.2),
the random distribution of CNF solid phase indicates that the closure formulation cannot be
solved in a unit cell due to the lack of periodicity in the CNF solid-phase distribution. However,
Whitaker had discovered that the use of a unit cell is permissible for such purposes because the
error arising through the use of the periodic boundary conditions is confined to the borders of the
unit cell and is not significant [1]. Therefore, in order to solve the closure equations in our porelevel geometry, SEM micrographs of different cross-section of the CNF sample were taken.
13

These cross-sections were selected in the vertical and horizontal planes, and because the CNF
sample was determined to be isotropic from the study of such micrographs, one of the
micrographs (Fig. 2) was selected for the purpose of solving the closure-formulation equations.

Fig. 2. SEM micrograph of a representative sectioned plane from the CNF sample

Using some sort of image processing software, it is possible to create a 3D geometry of porespace for our CNF sample by stacking different cross-sections from consecutive parallel planes
and interpolating between them. However, in the present study, a 2D micrograph selected from
inside of the CNF sample is used as a porous-medium geometry to solve COMSOL in. In order
to use the geometry for computations, we assumed that brighter color fibers were closer to the
top surface of micrograph while darker ones belong to the lower levels. Then if a cross flow were
to occur, the brighter color surfaces will act as obstacles (particles) while the darker surfaces can
be treated as pores. Using this assumption, the SEM micrograph of the CNF sample can be
converted into a binary picture for use with COMSOL for further analysis. The 2-D geometry
extracted from the CNF micrograph is shown in Fig. 3.

14

(a)

(b)

Fig. 3. (a) The CNF micrograph used in our permeability calculation. (b) The corresponding
solid (white) and pore (dark) regions of a reconstructed 2-D binary region extracted from the
micrograph, which is used to define the domain geometry for COMSOL.
Note that for 2-D samples considered in the present study, the permeability tensor can be
obtained through Eqn. (20) as
D
11
K f
D
21

where each of the bracketed Dij

f
f

D12
D22

f
f

(24)

terms represents the intrinsic phase-average value of that

particular Dij element, obtained after integrating it in the entire void region of the reconstructed
binary micrograph. Here f represents the porosity that is obtained by dividing the total void
area with the total area of the micrograph. Different elements of the permeability tensor, Eqn. 24,
were estimated with COMSOL after the governing equations, Eqns. 15 to 19, were solved within
the void region of the domain. After all these calculations, the final form of the permeability
tensor for our sample CNF is obtained as

2.61 0.23 0.38 0.07

K 1011
(m2)
0
.
21

0
.
05
3
.
28

0
.
11

15

(25)

The closeness of the diagonal terms in this matrix form of the permeability tensor for this 2D
medium indicates that our CNF sample can be characterized as being almost isotropic. It is also
interesting to note the presence of non-zero diagonal terms that are fairly big and are on the order
of 10% of the main diagonal terms. This clearly indicates that the x (horizontal) and y (vertical)
coordinates considered in Fig. 3(b) are not the principal or material directions of the porous
medium. (If the x and y directions match the principal directions, the permeability-tensor matrix
acquires a pure diagonal form.) One may also note that the matrix is not symmetrical, which
contrasts with the almost universal observation that the permeability tensor is symmetric for a
majority of porous media. At this moment, we do not have any explanation for this divergence
from the conventional wisdom we merely state our results as it is. (Though there is some
anecdotal evidence from co-workers in this area, which suggests that this lack of symmetry is not
uncommon when one tries to estimate the permeability tensor using similar upscaling methods.)
In porous media literature, it is well established that any volume averaged quantity, such
as the permeability, should be independent of the size of the REV such that any changes in this
size should not lead to significant changes in that quantity. We applied the same criterion to our
unit-cell in order to find the proper size of the unit cell in our permeability calculations. Four
concentric square-shaped unit cells with different areas were chosen and the K11 components of
the permeability tensors obtained from these unit-cells were compared to each other.

16

Fig. 4. Four different unit-cells considered to check the insensitivity of the permeability tensor to
unit-cell size.

Unit cell size study

3
Permeability [m2]

2.5
2
1.5
1
0.5
0
0.2

0.4

0.6

0.8

Unit cell size/total geometry size

Fig. 5. Permeability component K11 for different unit cells of varying sizes

It is clear from Fig. 5 that though K11 does increase constantly with the size of the considered
unit cell, it seems to have converged to the value of 2.61e-11 m2. This effect could be explained
by the periodic boundary conditions applied on the boundaries of unit cells, as stated through
Eqn. (18). The influence of the periodic boundary condition, which is caused due to the violation
of the actual non-periodic nature of the porous medium, is known to be large in small regions
near the unit-cell boundaries [1]. However, this effect wanes as one move away from the
17

boundaries in larger unit cells. Therefore, the permeability results obtained from the smaller unitcells is less accurate compared to the values obtained from the larger unit-cells. As a result, the
convergence of the permeability results for larger unit cells represents an asymptotic approach to
a true permeability, and hence using the entire geometry as a unit cell is appropriate for solving
the closure formulation.
After estimating the size of the final REV, we also wanted to ascertain if the
permeability values do not vary much as the REV is moved around in Fig. 4. This can be
construed as a condition for establishing the homogeneity of the CNF sample. As shown in Fig.
5, the permeability converged to 2.5e-11 m2 approximately, which allows us to conclude that the
unit-cell size with area 60% (or 0.6) of the entire rectangular micrograph would be a good choice
for the REV in the present geometry. After thus establishing the size of the smallest acceptable
REV, we proceeded to calculate the permeability at four different locations in the micrograph by
translating this reduced REV to those locations. These different selected locations of the reduced
REV yield the values of the porosity, a macroscopically averaged quantity estimated through the
image processing capability of COMSOL as the ratio of the void area within the REV to the total
REV area, to be between 0.57 and 0.63. This is a fairly converged result and the overall porosity,
equated to the average of all these different values, was found to be 0.6. By using this porosity
value in our closure-formulation calculations, the average permeability was numerically
estimated to be 2.5e-11 m2. The scatter in this numerical permeability, as reported in Table (3), is
obtained from the maximum and minimum of the set of permeabilities obtained from the
translation of REVs.
After measuring the CNF permeability using the Closure formulation, the falling-head
permeameter was used to estimate the permeability of the CNF sample. In experiments, the
18

height of the liquids was measured as a function of time and the permeability was calculated
using Eqn. (21) by fitting a linear curve on the data points (Section 3.2). Experiments were done
using Decane and Hexane. (These two alkanes, unlike water, were used since the cellulose does
not dissolve in them.) Some of the physical properties of the working fluids and CNF sample are
presented in Table (2).
Table (2): Physical properties of the CNF and the working fluid.
Decane

Hexane

0.000859 Pa.sec

0.000297 Pa.sec

730 kg/m3

655 kg/m3

CNF sample radius

0.004 m

0.004 m

Syringe radius

0.013 m

0.013 m

CNF sample length

0.03 m

0.03 m

Viscosity
Density

The ln(

x0
) vs t plots for these two different working fluids are presented in Fig. 6. The slope of
x

the linear part of the diagram is used in the estimation of the permeability. The permeability
obtained from the experiments is compared with the numerical estimation in Table (3).

Decane

Hexane
1.5

y = 0.001x + 0.0688
R = 0.9957

Ln(x0/x)

Ln (x0/x)

1.5

0.5
0

y = 0.001x - 0.0128
R = 0.9477

1
0.5
0

500

1000

1500

Time (sec)

500

1000

Time (sec)

19

1500

2000

(a)

(b)

Fig. 6. Relationship between the ratio of CNF liquid-column heights and time for (a) Decane (b) Hexane.

In order to examine the accuracy of the permeability found using the closure formulation and
experimental method, the empirical correlation by Rumpf and Gupte [29] was used to estimate a
theoretical, true value of permeability. As stated in Eqn. (22), such a permeability is expressed
as a function of the porosity and the mean particle diameter. In this study, the porosity and mean
particle diameter were estimated by using COMSOL multiphysics software in the entire
rectangular micrograph of Fig. 3b. The porosity was calculated through its in-built image
processing software by dividing the void/empty area of the micrograph by the total area of the
rectangular micrograph. In order to find the mean particle diameter, we assumed the solid
particles to be circular and through dividing the total solid area by total number of particle, the
average area, and consequently the mean diameter, of one particle could be measured. The
permeability results obtained using the empirical formulations are presented in Table (3).

Table (3): Comparison of the permeability values as obtained from the closure-based numerical
simulations, experiments, and a theoretical model. Units are 10-11 m2
Simulation
K11= 2.61+/-0.23

Experiment with Decane Experiment with Hexane

K=3.8+/-0.24

K=1.8+/-0.05

K22= 3.28+/-0.11

20

Theory (Rumpf and Gupte)

K=3.23

As listed in the table (3), there is a good agreement between the permeability values obtained
from the experiments with Decane as the working fluid on the one hand, and the values of the
diagonal elements of the permeability tensor as obtained from the closure formulation on the
other. In addition, the permeability value obtained using the Rumpf and Gupte empirical
correlation are on the same order of magnitude as the permeability from the rest of the methods,
which confirms the accuracy of our simulations.
4. 3 Permeability of particulate porous medium made from sintered polymer beads
In order to examine the validity of the closure formulation for estimation of the permeability
tensor for particle-based media, porous cylindrical wicks sintered from micron size polymer
beads were selected, and SEM micrographs were obtained from thin slices taken from different
cross sections of the wicks. After some manipulations, the micrographs were imported into
COMSOL Multiphysics software, and the governing equations (Eqns. 15 through 18) were
solved in the void region in order to calculate different elements of the permeability tensor. A
sample micrograph and the corresponding binary figure are shown in Fig. 7. The numerical
simulations were carried out for three different samples and as before, the following form of the
permeability tensor was obtained after using Eqns. 20 and 24:
3.45 0.52 0.19 0.09
K 1011

0.07 0.04 3.8 0.02

(m2)

(26)

We note that compared to the CNF material, the diagonal terms of the permeability matrix are
much closer to each other. Hence one can conclude that industrially produced wick through the
sintering of polymer beads is much more isotropic that the CNF sample produced in a lab. Also
note that the off-diagonal terms are much smaller than those observed for the CNF material
considered earlier, which indicates that the considered x (horizontal) and y (vertical) directions in
21

Fig. 7(b) are much closer to the principal or material direction of the polymer wick. But once
again, we notice a lack of symmetry in the permeability tensor, which we report without any
explanation.

(a)
(b)
Fig. 7. (a) A sample micrograph for permeability calculation, and (b) the
corresponding transformed binary picture obtained using COMSOL.

In addition to the computational analysis, the Falling head Permeameter was used as before to
find the height of the liquid column within the wick as a function of time, and to eventually
estimate permeability. Some of the physical properties of the wick and the working liquid used in
this experiment are gathered in Table (4).
Table (4): Physical properties of the wick and the working fluid
Liquid substance

Dodecane (C12H26)

Density (Kg/m3)

753

Viscosity (mPa.s)

1.33

Wick material

Polyethylene

Wick Length (mm)

74.5 0.2

Cross-section Diameter (mm)

6.4 0.1

Particle size (m) (appx.)

17525

22

As shown in Fig. 8, there is a good linear relationship between Ln(x0/x) and time, and therefore,
using Eqn. (21) as before, the permeability was calculated. The experiments were repeated three
times and the scatter bars show the range of Ln(x0/x) values obtained.

Fig. 8. Relationship between the ratio of liquid-column heights and time for a polymer wick
using Dodecane.

Table (5): Comparison of the permeability values as obtained from numerical simulations,
experiments and theory (units are in 10-11 m2)
Simulation
K11=3.45
K22=3.8

Experiment

Theory (Kozeny-Carman)

K~2.7

K~6.28

As listed in Table (5), there is a good agreement between the permeability value obtained from
the falling head permeameter and the values of the diagonal elements of the permeability tensor
as obtained from the closure formulation. While the theoretical model yields a permeability value
on the same order of magnitude as the experiments, the simulation provides the more accurate
results.
5. Summary and Conclusions
In this study, the closure formulation employed by Whitaker to derive Darcys law using the
volume averaging method [1] was used to estimate the permeability tensor for porous media
23

found in CNF sample and sintered polymer wicks. The main advantage of the proposed closureformulation method is its direct estimation of the permeability tensor, unlike the Stoke-Flow
simulation method that requires simulating flow along three different directions in the pore space
to resolve the full tensor. The SEM micrographs of the CNF sample and polymer wicks were
used to create 2D binary pictures representing the solid and pore phases. The permeability tensor
was estimated by solving the partial differential equations associated with the closure
formulation within the pore space of a unit cell. From the closeness of the K11 and K22 terms in
the 2-D permeability tensor, the two considered porous materials were found to be almost
isotropic, the polymer wick being more so than the CNF sample. The presence of non-diagonal
terms indicated that the chosen x and y axis in the micrographs did not match the principal or
material directions. The permeability tensor was observed to be slightly non-symmetric, once
again, less so for the polymer wick than for the CNF sample. The convergence in the estimated
permeability values was established by varying the size of unit cells as well as by translating the
unit cells in the binary domain. The numerical permeability thus estimated was compared with
the experimental permeability obtained using the falling-head permeameter. A good agreement
between the two demonstrated the accuracy of the closure formulation and the resulting
simulation.

In

addition,

the

permeability

values

were

also

compared

with

two

theoretical/empirical (Kozeny-Carman and Rumpf and Gupte) permeability relations obtained

from the literature and a reasonable match was observed, thereby establishing the accuracy of the
proposed method again.
6. Future Work
Despite introducing an improvement to the permeability estimation process for realistic, irregular
porous materials, the 2D micrographs cannot perfectly represent the 3D pore structure of the
24

considered porous media. Even though the permeability values from the 2D simulation were
close to the experimental data, it would be more accurate to perform a full 3D analysis by
reconstructing the porous media from a series of 2D micrographs for parallel planes using a
micro-CT scanner. Though such a simulation will be computationally challenging due to high
memory and storage requirements, it will be worth the effort to undertake this step. In addition, it
is recommended to repeat the simulation and experiment for a variety of CNF and wick samples
so as to have a wider range of porosity and particle size variations in the porous media
considered, and hence have a more rigorous validation of Whitakers closure-formulation
approach for estimating permeability.

7. Acknowledgement
The authors gratefully acknowledge the financial support extended by the Research Growth
Initiative (RGI) program of the graduate school at University of Wisconsin-Milwaukee.

8. References
1.

Whitaker, S., The method of volume averaging. Vol. 13. 1998: Springer.

2.

Slattery, J.C., Flow of viscoelastic fluids through porous media. AIChE Journal, 1967. 13(6): p.
1066-1071.

3.

Whitaker, S., Diffusion and dispersion in porous media. AIChE Journal, 1967. 13(3): p. 420-427.

4.

Whitaker, S., Advances in theory of fluid motion in porous media. Industrial & engineering
chemistry, 1969. 61(12): p. 14-28.

5.

Anderson, T.B., Jackson, R., A fluid mechanical description of fluidized beds. Ind Eng Chem
Fundam, 1967. 6: p. 527-39.

25

6.

He, Y., Sykes, J. F., On the spatial-temporal averaging method for modeling transport in porous
media. Transport in porous media, 1996. 22(1): p. 1-51.

7.

Mls, J., On the existence of the derivative of the volume average. Transport in porous media,
1987. 2(6): p. 615-621.

8.

Gray, W.G., General conservation equations for multi-phase systems: 4. Constitutive theory
including phase change. Advances in water resources, 1983. 6(3): p. 130-140.

9.

Howes, F.A., Whitaker, S., The spatial averaging theorem revisited. Chemical engineering
science, 1985. 40(8): p. 1387-1392.

10.

Whitaker, S., A Simple Geometrical Derivation of the Spatial Averaging Theorem. Chemical
engineering education, 1985. 19(1).

11.

Gray, W.G., Lee, P. C. Y., On the theorems for local volume averaging of multiphase systems. Int
J Multiphase Flow 1977. 3: p. 333-40.

12.

Hassanizadeh, M., Gray, W. G., General conservation equations for multi-phase systems: 1.
Averaging procedure. Advances in Water Resources, 1979. 2: p. 131-144.

13.

Bennethum, L.S., Cushman, J. H., Multiscale, hybrid mixture theory for swelling systemsI:
balance laws. International Journal of Engineering Science, 1996. 34(2): p. 125-145.

14.

Gray, W.G., Miller, C. T., Thermodynamically constrained averaging theory approach for
modeling flow and transport phenomena in porous medium systems: 1. Motivation and overview.
Advances in Water Resources, 2005. 28(2): p. 161-180.

15.

Hsu, C.T., A closure model for transient heat conduction in porous media. Journal of heat
transfer, 1999. 121(3): p. 733-739.

16.

Nakayama, A., Kuwahara, F., Sugiyama, M., Xu, G., A Two-Energy Equation Model for
Conduction and Convection in Porous Media. Int J Heat Mass Transfer 2001. 44(22): p. 4375-9.

17.

Pillai, K.M., Single-Phase Flows in Swelling, Liquid-Absorbing Porous Media: A Derivation of

Governing Equations using the Volume Averaging Method. Journal of Porous Media, 2013. (In
press).
26

18.

Bear, J., Bachmat, Y., Introduction to modeling of transport phenomena in porous media. Vol. 4.
1990: Springer.

19.

Dormieux, L., Kondo, D.,Ulm, F. J., Microporomechanics. 2006: John Wiley & Sons.

20.

Quintard, M., Bletzacker, L.,Chenu, D.,Whitaker, S., Nonlinear, multicomponent, mass transport
in porous media. Chemical Engineering Science, 2006. 61(8): p. 2643-2669.

21.

Kim, J.H., Ochoa, J. A.,Whitaker, S., Diffusion in anisotropic porous media. Transport in porous
media, 1987. 2(4): p. 327-356.

22.

Davit, Y., Quintard, M.,Debenest, G., Equivalence between volume averaging and moments
matching techniques for mass transport models in porous media. International Journal of Heat
and Mass Transfer, 2010. 53(21): p. 4985-4993.

23.

Golfier, F., Wood, B. D.,Orgogozo, L,Quintard, M.,Bus, M., Biofilms in porous media:
development of macroscopic transport equations via volume averaging with closure for local
mass equilibrium conditions. Advances in water resources, 2009. 32(3): p. 463-485.

24.

Ankerfors, M., Lindstrm, T. NanoCellulose Developments in Scandinavia. in Paper and Coating

Chemistry Symposium (PCCS). 2009.

25.

Barari, B., Pillai, K. M., Qamhia, I. I.,El-Hajjar, R.,Ellingham, T. K. ,Turng, L-S, Sabo, R.,
Mechanical Testing Of Scaled Cellulose Nano-Fiber Based Composites Made Using Micro-Rtm
Process, in 19th International Conference On Composite Materials. 2013: Montreal, Canada.

26.

Saito, T., Hirota, M.,Tamura, N.,Kimura, S.,Fukuzumi, H.,Heux, L.,Isogai, A., Individualization
of nano-sized plant cellulose fibrils by direct surface carboxylation using TEMPO catalyst under
neutral conditions. Biomacromolecules, 2009. 10(7): p. 1992-1996.

27.

Beyhaghi, S., Pillai, K.M., Estimation of Tortuosity and Effective Diffusivity Tensors using
Closure Formulation in a Sintered Polymer Wick during Transport of a Nondilute, Multicomponent Liquid-Mixture, . Special Topics & Reviews in Porous Media An International
Journal, 2011. 2(4): p. 267-82.

27

28.

Masoodi R., P., K.M., Varanasi PP., Darcys Law based Models for Liquid absorption in Polymer
Wicks. AIChE J. , 2007. 53: p. 2769-82.

29.

Masoodi, R., Pillai, K. M., Wicking in porous materials: traditional and modern modeling
approaches. 2012: CRC Press.

30.

Barari, B., Pillai, K.M., Calibration of onedimensional flow setup used for estimating fabric
permeability using three different reference media. Polymer Composites, 2014.

31.

Vernet, N., Ruiz, E.,Advani, S.,Alms, J. B.,Aubert, M.,Barburski, M.,Barari, B.,Beraud, J.

M.,Berg, D. C.,Correia, N.,Danzi M.,Delavire T.,Dickert M.,Fratta, C. D.,Endruweit,
A.,Ermanni, P.,Francucci, G.,Garcia J. A.,George A.,Hahn C.,Klunker F.,Lomov S. V.,Long
A.,Louis B.,Maldonado J.,Meier R.,Michaud V.,Perrin H,Pillai K. M.,Rodriguez E.,Trochu
F.,Verheyden S.,Weitgrefe M.,Xiong W,Zaremba S.,Ziegmann G., Experimental determination of
the permeability of engineering textiles: Benchmark II. Composites Part A: Applied Science and
Manufacturing, 2014. 61: p. 172-184.

32.

Lasseux, D., Arani, A.A., Abbasian Ahmadi, A., On the stationary macroscopic inertial effects
for one phase flow in ordered and disordered porous media. Physics of Fluids (1994-present),
2011. 23(7): p. 073103.

33.

Quintard, M., Diffusion in isotropic and anisotropic porous systems: Three-dimensional

calculations. Transport in Porous Media, 1993. 11(2): p. 187-199.

28