Modeling Isothermal and Non-Isothermal Flows in Porous Media

MODELING ISOTHERMAL AND NON-
ISOTHERMAL FLOWS IN POROUS MEDIA
by
Ehsan Mohseni Languri
A Dissertation Submitted in
Partial Fulfillment of the
Requirements for the Degree of
Doctor of Philosophy
in Engineering
at
The University of Wisconsin-Milwaukee
May 2011
UMI Number: 3462824
All rights reserved
INFORMATION TO ALL USERS

The quality of this reproduction is dependent upon the quality of the copy submitted.
In the unlikely event that the author did not send a complete manuscript
and there are missing pages, these will be noted. Also, if material had to be removed,
a note will indicate the deletion.
UMT
Dissertation Publishing
UMI 3462824
Copyright 2011 by ProQuest LLC.
All rights reserved. This edition of the work is protected against
unauthorized copying under Title 17, United States Code.
ProQuest LLC
789 East Eisenhower Parkway
P.O. Box 1346
Ann Arbor, Ml 48106-1346
by
A Dissertation Submitted in
Partial Fulfillment of the
Requirements for the Degree of
Doctor of Philosophy
in Engineering
at
The University of Wisconsin-Milwaukee
Graduate School Approval Date
ii
ABSTRACT

By
The University of Wisconsin-Milwaukee, 2011
Under the Supervision of Professor Krishna M Pillai
A complete understanding of the physics of flow and heat transfer
phenomena in porous media is vital for accurate simulation of flow processes in
industrial applications. In one such application pertaining to liquid composite
molding (LCM) for manufacturing polymer composites, the fiber preforms used in
LCM as reinforcements are limited not only to the single-scale porous media in
the form of random fiber-mats, but also include dual-scale porous media in the
form of woven or stitched fiber-mats. The conventional flow physics is not able to
model the resin filling process in LCM involving the dual-scale porous media. In
this study, the flow in dual-scale porous media is studied in order to predict the
permeability of these fiber mats. The effect of aspect ratio of the fiber preform on
the accuracy and flow during permeability estimation in single- and dual-scale
porous media is analyzed experimentally and numerically.
iii
Flow of liquid in a free channel bounded on one side by porous medium is
studied next, and two well-known boundary conditions of stress continuity and
stress jump at the interface of the two regions are evaluated numerically. A point-
wise solution for Stokes flow through periodic and non-periodic porous media
(made of cylindrical particles) adjacent to the free channel is presented using the
finite element based CFD software COMSOL. The efficacy of the two interfacial
conditions is evaluated after volume averaging the point-wise velocity using a
long averaging volume, also called the representative elementary volume or REV,
and then comparing such a volume-averaged velocity profile with the available
analytical solution. The investigation is carried out for five different porosities at
three different Reynolds numbers to cover a wide range of applications. The
presence of randomly-placed cylinders during the creation of non-periodic porous
media damps out spatial fluctuations in the averaged velocity observed in periodic
porous media. The analytical solutions obtained after applying the stress-
continuity and stress-jump boundary conditions are found to work well at low
porosities, which is in contradiction with the results achieved earlier by other
researchers.
The traditional approach of using averaged equations in the regions of sharp
gradients in porous media to describe flow and transport is theoretically untenable
and perhaps inaccurate. A novel ensemble averaging method is being proposed to
test the accuracy of the volume averaged or smoothed description of flows in
iv
porous media in the regions of sharp gradients. In the new method, the flow in a
certain arrangement of particles (called a realization) is averaged using a small
unit cell, much smaller than the REV. Then such an averaged flow variable is
further averaged over a whole gamut of randomly-generated particle realizations.
This method of averaging will allow one to simulate the ensemble averaged flow
in the sharp-gradient regions of porous-media flows and thus compare the result
with the flow predicted by the traditional volume-averaged set of flow equations.
First the accuracy of the ensemble averaging method was tested by comparing the
permeability of an artificially generated porous medium obtained by the proposed
method against the permeability predicted by some established theoretical models
of permeability. The proposed method was found to be quite accurate. Later the
ensemble average method was applied to the open-channel porous-medium
interface region characterized by a sharp gradient in the flow velocities. It was
discovered that the volume averaged description of such flows, characterized by
the use of the Brinkman equation along with the stress-continuity and stress-jump
conditions, is quite accurate for a range of Reynolds numbers.
The non-isothermal transport during flow in porous media is examined next.
The main focus in this area of research is the thermal dispersion term found in the
heat transfer equation for single- and dual-scale porous media. Most of the
previous efforts on modeling the heat transfer phenomena in porous media were
devoted to isotropic porous media. However, for the anisotropic porous media
v
widely in many industrial applications (such as the woven or stitched fiber mats
used in LCM), not much research on the dispersion tensor is available. A new
combined experimental/numerical approach to estimating the dispersion tensor is
introduced and applied for both isotropic (single-scale) and anisotropic (dual-
scale) porous media. The equivalence between the heat and mass transfer is
exploited and a 1 -D flow experimental setup is employed to study the spreading of
a dye. Later the mathematical model for such a spreading of concentration
(equivalent to the temperature) around a point input in a constant velocity field is
solved using the finite element based software COMSOL. Thus obtained
numerical spreading pattern is fitted onto the experimentally observed one using
the dispersion matrix (tensor) as a fitting parameter. A few cases of single- and
dual-scale porous media are studied and the dispersion tensors are reported for
each individual case.
A chapter-wise summation of this Ph.D. thesis proceeds as follows. The
first chapter is allocated to study the effect of aspect ratio on the permeability
measurement for both single- and dual-scale porous media. In the same chapter,
flow-front of the test fluid during mold-filling in a 1-D mold for both random and
unidirectional fibrous porous media was studied experimentally and numerically.
In the next chapter, an evaluation of the boundary conditions employed to describe
the free-channel porous-medium interface is proposed after using a long REV
close to the interface to average the numerically simulated point-wise flow inside
both periodic as well as non-periodic porous media. Later, a new ensemble
vi
averaging method is introduced which can be used to study the efficacy of the
volume averaged equations in the regions of sharp gradients. In the last chapter,
the non-isothermal flow of liquid through the porous media studied with emphasis
on estimating the thermal dispersion term.
VH/u
Major Professor Date
vn
TABLE OF CONTENTS
CHAPTER 1. INTRODUCTION
1.1. Literature Review on Unsaturated Flow through Dual-scale Porous 5
Media
1.2. The Scope of Research 8
References 11
Chapter 2. EFFECT OF ASPECT RATIO ON MEASURED PERMEABILITY
AND FLOW-FRONT PROGRESS
2.1 Permeability Measurements 16
2.1.1 1-D Flow Experiment 17
2.1.2 Radial Flow Experiment 18
2.2 Effect of Preform Aspect Ratio on the Permeability Measured through 19
the 1-D Flow Experiment
2.2.1 Theory 21
2.2.2 Experiment 25
2.2.3 Results and Discussions 26
2.3 Effect of Preform Aspect Ratio on the Unidirectional Flow Front 31
during Mold-filling in the LCM
2.3.1 Results and Discussion 33
2.4 Conclusion 37
vni
References 38
CHAPTER 3 Evaluation of Boundary Conditions at the Clear-Fluid Porous-Medium
Interface
3.1 Introduction 43
3.2 Design of Numerical Experiments 47
3.2.1 Numerical Solution 49
3.3 Open Channel Flow Model for the Theoretical Solution 50
3.4 Theoretical Solution of the Porous Medium Flow 51
3.5 Stress-Continuity Interfacial Condition 51
3.6 Stress-Jump Interfacial Condition 54
3.7 Results and Discussions 56
3.8 Summary and Conclusion 70
References 72
Chapter 4. PREDICTING THE AVERAGE FLOW VARIABLES IN POROUS
MEDIA USING ENSEMABLE AVERAGING METHOD
4.1 Introduction 76
4.2 A Conceptual Description of the Ensemble Averaging Method 81
4.3 Validation 88
4.3.1 Permeability Estimation in a Porous Medium 88
4.3.1.1 Permeability Measurement Using Ensemble Averaging 91
Method
4.3.1.2 Results and Validation 93
4.3.1.3 New Permeability Models 97
ix
4.3.2 Clear-Fluid and Porous Media Interface Problem 99
4.3.2.1 Results and Validation 103
4.4 Conclusion
References 108
CHAPTER 5 NON-ISOTHERMAL MODELING OF POROUS MEDIA FLOW
5.1 Introduction 114
5.2 Proposed Work 116
5.3 Governing Equations of Non-Isothermal Flow in Porous Media 117
5.3.1 Macroscopic Level 117
5.3.2 Microscopic Level 121
5.4 Dispersion Tensor 121
5.4.1 Dispersion in Isotropic Porous Media 124
5.4.2 Dispersion in Anisotropic Porous Media 128
5.5 A New Combined Experimental/Numerical Methodology to Assess 133
Dispersion Tensor
5.5.1 Theory 132
5.5.2 Experimental Measurements and Devices 134
5.5.3 Numerical Modeling 139
5.5.4 Combined Experimental/Numerical Methodology 139
5.6 Results and Discussions 141
5.7 Summary and Conclusion 149
References 151
CHAPTER6 SUMMARY, CONTRIBUTION, AND FUTURE WORK
6.1 Summary 156
6.2 Contributions 158
6.3 Future Work 159
APPENDIX 161
CV 163
XI
LIST OF FIGURES
Number Description Page
1.1 Representative elementary volume (REV) 2
1.2 Some types of woven mats used in LCM 6
1.3 Schematic view of flow of fluid through dual-scale porous medium 7
2.1 Schematic view of 1 -D flow experiment 18
2.2 Schematic view of radial flow experiment 19
2.3 Schematic description of the three 1-D flow experiments 23
2.4 Radial flow experiment at flow-rate of 0.00256 ml/s 26
2.5 The pressure history for three different orientations for the aspect 27
ratio 1
2.6 Permeability measurement error, K, - K 2 , versus the aspect ratio 28
2.7 ^ , .,. K.-K, , . 29

Permeability measurement error, — , versus the aspect ratio
2.8 Setup for the 1-D flow experiment 32
2.9 Unsaturated flow front at different mat orientations for aspect ratio 33
of 4
2.10 Flow front of the filling process in the dual-scale porous media for 35
(a): Numerical modeling at different time steps (second) (b):
Experimental results
2.11 Figure 2.11 Unsaturated flow-front at four different aspect ratios 36
xu
of 1, 2, 3 and 4, from left to right
2.12 Pressure history profile for both cases of single scale and dual 36
scale porous media
3.1 Modeling creeping (Stokes) flow through the hybrid (open-channel 48
porous-medium) region: (A) A schematic of the velocity profile through
the channel and the porous medium, (B) The geometry of the hybrid
region with a periodic porous medium; (C) The geometry of the hybrid
region with a non-periodic, random porous medium. The last two figures
show typical REVs (Representative Elementary Volumes) employed for
averaging flow velocities inside the periodic and irregular porous media.
3.2 Effect of increase in Reynolds number on the streamline pattern 58
observed adjacent to the open-channel porous-medium interface for
the periodic porous medium. The porosity for the porous medium is
e = 0.6.
3.3 Streamline pattern at the interface of clear-channel and porous wall 59
for non-periodic porous media at f = o.5 a n d Re = l x i o 3 .
3.4 Plots of normalized average-velocity inside the periodic porous 60
medium at three different Re numbers for five different porosities: a)
£ = 0.5,b) s = 0.6,c) s = 0.7, d) s = 0.8 ande) a = 0.9.
3.5 An example of a weakly-random arrangement of cylinders employed 62
for the creation of an irregular porous medium adjacent to the
xiii
channel. The porosity is 0.7 (i.e., s = 0.7) for the porous medium.
3.6 Schematic view of six different REVs used for the REV-size 63
independence study (not in the actual dimensions).
3.7 Results of the REV size-independency analysis for the irregular 64
porous medium at £=0.5 for the six REVs listed in Table 3.3.
3.8 Results of the mesh independence study. (Details of the three grids 65
or meshes are given in Table 3.4.)
3.9 Repeatability of the solution by comparing three different random 66
realizations' results
3.10. Comparison of average velocity using FEM for both regular and 67
random cases with the Brinkman models at Re= 1 x 10"4 for
A)e = 0.5 ; B) e = 0.6 , Q e = 0.7 , D) e = 0.8 a n d E) * = 0-9
3.11 Actual velocity profile versus dimensionless channel height for 8=0.6 68
and Re=lx 10-4.
3.12 Velocity contours for the non-periodic porous media at E = 0.5 and 69
Re = lxio-3: A) velocity contour and B) volume-avearged velocity
3.13 Effect of variation in porosity of the irregular porous medium on the 70
distribution of the volume averaged velocity near the open-channel
porous-medium interface.
xiv
4.1 Variation of averaged quantities with respect to the REV size 77
4.2 Dimensionless vertical velocity and temperature profiles versus 79
similarity variable for natural convection on a vertical heated surface
filled with porous media (Cheng and Minkowycz, 1977).
4.3 Liquid composite molding process with hot walls of mold on sides 80
and the cold resin.
4.4 Anticipated variation in (¥{) through an increase in; a) Volume, b) 82
Number of realizations
4.5 Figure 4.5 Schematic view of random locations of cylinders inside 84
cells in different realizations
4.6 Fixed number of particles totally inside a cell, case 1 86
4.7 A typical distribution of a fixed number of particles inside a cell 87
for case 2.
4.8 (a): The geometry of the problem, (b): Details of a sample 91
realization with square unit cells each including a solid cylinder.
4.9 (a) Two different realizations randomly selected (b) Square 95
arrangement of periodic fiber mat array.
4.10 Comparison of the obtained permeability values from ensemble 95
average method with six other results.
4.11 Dependency of the permeability value on the number of realization 99
used for estimating the ensemble average.
xv
4.12 Creeping flow through the hybrid clear-fluid-porous medium 100
interface: (a) schematic view and (b) Ideal geometrical model
4.13 Velocity distribution in the clear-fluid porous medium interface; 104
( a ) Re = 0.01 a nd (b) Re = 0.001
4.14 Figure 4.13. a) Dependency of the Ensemble average method on 106
number of realizations; b) Part A; c) Part B.
5.1 Schematic view of the initial impregnation of a dual-scale porous- 118
medium by a liquid [19]
5.2 Mixing as a result of obstruction 123
5.3 Dispersion of a point injection displaced a distance L 125
5.4 The 1-D flow experiment setup: A) open mold cavity and B) the 135
assembled mold
5.5 The flow injection and flow-meter setup 136
5.6 Measuring devices for A) Density, B) Viscosity, and C) Porosity 137
5.7 Effect of dispersion in the form of an elliptical dye expansion in a 138
single-scale porous medium
5.8 Schematic view of the geometry, the boundary and initial 139
conditions
5.9 Effect of pure molecular diffusion of thermal dispersion by time 141
5.10 Random fiber-glass mat as a representative of single-scale porous 142
media
xvi
5.11 Matching the experimental and numerical results for a single-scale 143
porous medium (random fiber mat) at Q=2ml/s at three different
time frames of A) 7 s, B) 33 s and C) 51 s
5.12 The longitudinal and transverse increases of the initial circular 145
dye-patch with time
5.13 Stitched fiber-glass mat as a representative of dual-scale porous 145
media.
5.14 Comparison of the experimental and numerical modeling for an 146
anisotropic but periodic medium at Q = 2 ml/s : A) experimental
results at t = 0 s, B) numerical results at t = 0 s, C) experimental
results at t = 35 s, D) numerical results at t = 35 s, E) experimental
results at t = 71 s and F) numerical results at t = 71 s
5.15 Unidirectional fiber-glass as a representative of dual-scale porous 148
media
5.16 Unidirectional fiber-glass in A) initial unsaturated flow and B) 148
saturated flow
5.17 Experimental observation of flow of tracer through a dual-scale 149
porous medium made of the unidirectional fiber mats
Al Schematic view of locating the solid cylinder at the center of the 161
unit cell in order to generate the periodic porous media
A2 Schematic view of location of the solid cylinder at horizontal and 162
xvii
vertical distances from the center of unit cell in order to generate
the non-periodic porous media
A3 No overlapping for the irregular porous media 162
xviii
LIST OF TABLES
2.1. Details of experiments 25
2.2. Coefficients of two equations fitted to Kx - K2 error data. 30
23 K —K 31
Coefficients of two equations fitted to— error data.
K
\
3.1. 50
Details of volume averaging moving step for each porosity.
3' 2 ' Values of effective viscosity JU', coefficient /?, and slip coefficient a at 57
different porosities
3.3. 63
Details of the six REVs used in the REV-size independence study.
3.4. Details of the elements and nodes of three different meshes employed to 66
conduct the mesh independence study
4.1. Five different permeability models 94
4.2. 96
Details of permeability values at different porosities for each model.
4.3. Error's details for each model's results based on the numerical method's 97
results.
xix
NOMENCLUTURE
Roman Letters
A Area [m2]
b A mapping parameter
c Specific heat coefficient [J/kg.K]
D Diffusivity [m2/s] and Diameter [m]
f A position vector function
h Channel height [m]
i An integer number
K Permeability [m2]
k Dispersion tensor [m2/s]
L Length of channel [m]
m,n Parameters
N An integer number
q" Heat conduction [w/m2]
S A sink term
T Temperature [K]
t Time [s]
V Volume [m3]
Pe Pecklet number
Re Reynolds number
Greek Letters
XX
a Thermal diffusivity [m2/s]
P A coefficient in Brinkman equation
5 Thermal conductivity [W/mK]
5c Boundary layer thickness [m]
e Porosity
I An axis
P Density [kg/m3]
H Viscosity [pa.s]
9 Porosity
V A flow variable
a A function
Subscripts
ave Average
eff Effective
f Fluid and fiber
g Gap
P Pressure
R Reaction
real Realization
s Solid
t tow
Superscripts
clear Clear flow
xxi
porous Porous-media flow
Other Symbols
<> Volume-averaged quantity
< >f Pore-averaged quantity
xxii
ACKNOWLEDGMENTS
I would like to express my deepest gratitude to my advisor, Professor Krishna
Pillai, for his continuous guidance, patience, support, care. Professor Pillai provided an
excellent atmosphere for research and teaching during these years. I am proud of having
such an exceptional supervisor and teacher in my Ph.D. study. His high standards on
research and teaching and his commitment towards students made it a privilege to work
with him.
I sincerely appreciate my dissertation committee members: Professor Michael
Lovell, Professor Sam Helwany, Professor Ilya Avdeev, and Professor Arash Mafi for
their support, comments and insights. I am also grateful for their time and efforts in
evaluating this research work. I would like to thank Professor Tien-Chien Jen, Professor
John Riesel, Mr. Dan Beller, for their support, help and motivation during these years.
I am indebted to my B.Sc. and M.Sc. supervisors and teachers, Professor Hessam
Taherian, Professor Davood Domairry Ganji, Professor Mofid Gorji Band-Pay, Professor
Kamel Hooman and Professor Halimi who cheered me up and supported me during last
ten years of study.
I would like to thank my friends, Dr. Reza Masoodi, Rahi Abouk, Nima Jalali,
Sahar Hosseini, Sepideh Maleki, Reza Jahani, Dr. Hua Tan, Saman Beyhaghi, Andrew
Vechart, Mike Verhagen, Grover Bennett, Russell Moore, Ehsan Ghotbi and Dr.
Mohammad Habibi who were always willing to help and give their best suggestions and
motivation.
xxiii
I would also like to thank my parents, Ali Mohseni Languri and Mahdokht Tirgar,
my sisters, Elham, Elahe and Azadeh, and in-law family. They were always supporting
me and encouraging me with their best wishes and support.
Above all, I would like to thank my wife, Maedeh Sedaghati. She was always there
cheering me up and stood by me through the good times and bad. This success was not
possible without her support and strength during last two years of my study.
xxiv
1
Chapter 1
INTRODUCTION
Composite materials have been widely used in industries such as
automotive, shipbuilding and civil due to their light weight, high strength, high
corrosion resistance and flexibility in design. Such composite materials are made
of polymer resins as matrix and glass fibers or carbon fibers as reinforcements for
the plastics. There is a variety of manufacturing processes for making polymer
composites, but liquid composite molding (LCM) is one of the most widely used.
The LCM process includes several technologies, such as Resin Transfer Molding
(RTM), Vacuum-Assisted Resin Transfer Molding (VARTM), and Seemann
Composites Resin Infusion Molding Process (SCRIMP). The following major
steps are common for all LCM processes: first the reinforcing fibers are shaped
similarly to the final shape, which is called preform, placed in a closed mold [1].
Then matrix materials, in the form of liquid polymer, infiltrate the preform either
by injection or by vacuum. The matrix material later undergoes a solidification
process wherein the thermoset type of resins cure with the help of an increased
temperature. The final parts are removed when the solidification process is
complete.
There is a great need for numerical modeling of these processes for
industrial applications. The preforms are assumed to be a porous medium in this

2
case. A volume averaging needed in order to average flow variables like
temperature over a representative volume at the pore level. The averaging volume
should be sufficiently large so that it can capture any fluctuation in that region.
This volume is called the representative elementary volume (REV), Fig. 1.1. The
REV must be far smaller than the size of the entire domain; otherwise the resulting
average through the whole domain cannot represent what happens at any
individual points in the domain. On the other hand, the size of the REV must be
larger than the size of single pore that each REV includes a sufficient number of
pores to have a meaningful averaging process [2].
Figure 1.1 Representative elementary volume (REV)
By a further assumption of fully saturated flow behind the flow front of the resin,
one can model the mold filling stage of LCM using the Darcy's law as
3
<jf>=--.V<pf>f (1.1)
M
where <vf > is the volume-averaged velocity of resin in the preform, < pf >f is
the pore-averaged pressure of resin, K is the permeability tensor for fiber preform
and |i is the resin velocity. In the case of isotropic porous media, the permeability
tensor become a scalar and the resin can be assumed incompressible. Therefore,
continuity equation can be written as
V.<vf>=0 (1.2)
By substituting Eqn. (1.1) into Eqn. (1.2), one can obtain an elliptic partial
differential equation (the Laplace equation) that has only pressure as the unknown
variable.
V.(K t .V f ) = 0 (1.3)
Tucker [3] used the volume averaging method to derive the energy equation for
the single- scale porous media, as
{erPrCpf +E s p,C p ,J^(T) + pfCp,f(vf).V{T} = V.(Klh,f +Klh,s).V(T) + £fPr H R (f £ ) f (1.4)
+ Uf(vr).K;'.(vf)
where ef and es are the volume fraction of the fluid and solid, pf and pi are
densities of the fluid and solid, CpJ and Cpsare specific heats of the fluid and
solid, T is temperature for both the fluid and solid phases obtained through the
local thermal equilibrium assumption, HRis the heat of reaction per unit mass of
the fluid, and fc is the reaction rate for curing of the resin. KthJ and Kths, the
4
effective thermal conductivities of the fluid and solid phases, are expressed as
Kth,f =k f e f § + ^ j n f s b d A - ^ ^ JvfbdV (1.5)

V V
f A„ f V,
K ^ = k , e . 5 + ^ ^ Jnf.bdA (1.6)
where kf and ki are thermal conductivities of fluid and solid phases, Vf and Vs
are the volume of fluid and solid phases in REV, and Aîs the interfacial area
between fluid and solid phases within the REV. The vector function bthat
transforms the gradient of the average temperature into the local variations of the
deviation is given by
f f =b.V(T) (1.7)
The next step to modeling the whole plastics processing process is to set the
balance equation for the chemical reaction of the resin which may also called by
curing. The balance equation for chemical reaction, as derived by Tucker [3], is
ef | ( c f > f +(vf).V(cf)f =V.{Df.V(cf)f}+ef(fc)f (L8)
where cf is the dimensionless extent of chemical reaction and D f , the effective
diffusivity tensor as following
Df =D f £ f 8 + ^ J n h f d A - ^ - JvffdV (1.9)
V V
f Afs f Vf
where / is a vector function of position within the fluid phase that transforms the
average concentration gradients into the point-wise concentration deviation

5
c f =f.V(c f ) f (1.10)
It should be noted that partially saturated flow front may occur due to the
geometrical characteristics of some fiber mats. In such situation, the numerical
modeling for single-scale porous media is not applicable and the importance of
numerical modeling for dual-scale porous media is often felt.
1.1 Literature Review on Unsaturated Flow through Dual-scale
Porous Media
Some recent studies [4-13] have shown that resin flow through certain types
of porous material cannot be modeled using the conventional flow physics. The
micro-structure of fiber mats, such as woven or stitched fabrics, is the main reason
for such a discrepancy between the experiments and numerical predictions of mold
filling (Fig. 1.2). This is a fact that the fiber tows made of thousands of individual
fiber filaments either woven or stitched. The gap within the intra-tow region is of
the order of micrometers. The distance between the fiber bundles, called inter-tow
region, is of the order of millimeters. The difference in the orders of magnitude of
pore length-scales in a medium is assumed the main reason for partial saturation of
such porous media. These types of porous media are classified under the category
of dual-scale porous media.

6
Biaxial Iteaue Triaxial Weaue Knit Nultiaxial Multilayer

Harp Knit
Figure 1 2 Some types of woven mats used in LCM
In dual-scale porous media, the injected resin passes through the inter-tow
channels without impregnating due to the high flow-resistance inside tows. Hence,
such flow is much quicker than the flow impregnating the tows. This delay in tow
impregnation leads to partially saturated region in the flow-front, Fig. 1.3. This
dual-scale nature of a preform has certain influences on the mold filling process
such as void formation [4-6, 9 and 14-17], creation of visible partially-saturated
region [4, 7-11, 13], and dropping of inlet-pressure history under constant
injection rate [7-13, 18].
The flow model in dual-scale porous media can be divided into me two
main categories based on their length scale. One is the tow-impregnation model
and the second is the global impregnation model. The former focuses on the void
formation mechanism due to its micro-macro flow interaction, while the latter
studies the flow characteristics during mold filling. In the tow-impregnation
models, a single computational domain studies the flows in the inter-tow and intra-
tow, separately.
Fully saturated area Partially saturated region Unsaturated area
Macro-flow front
OnO QOOOo^
O O O O O O r, O o °
O O O Q O O O eft O
Single fiber
O n oO 0 ° o ° 0 ° n O
o°o°oO o°o°o
u
oo o 0 o°o o o 0
Figure 1.3 Schematic view of flow of fluid through dual-scale porous medium
The computational domain distinguishes the intra-tow and inter-tow regions
through the mesoscopic fabric architecture with certain number of fiber tows and
their adjacent ieter-tow regions [6-14]. On the other hand, the global impregnation
models do not need detailed local (the tow scale level) flow information- a lumped
quantity is used to study the mass exchange between the inter-tow and intra-tow
regions. The delay in impregnation within the tows at the micro-level acts as sinks
of liquid in the macroscopic flow field. The modified continuity equation at

8
macroscopic level was proposed by including a non-zero sink term [4, 8, 11, 15,
19 -27]
V.v = -S (1.11)
where S is the sink term, equal to the volumetric rate of resin absorption by
tows per unit volume. One can combine the above equation with Darcy's law to
complete the modeling of dual-scale porous flow using the global impregnation
model.
1.2 The Scope of Research
The primary goal of this research is to advance the understanding the
transport phenomena in isothermal and non-isothermal fluid flow through the
porous medium.
In chapter 2, the focus is on the important term of permeability in dual-scale
porous media. The conventional methods for measuring the permeability have
been introduced. The effects of some parameters such as aspect ratio of the
preform in both steady and unsteady states have been studied experimentally.
In chapter 3, a comprehensive study is conducted to model flow near the
interface of a free channel and a porous-medium channel. A review of different
boundary conditions available in the literature has been done which mainly
focuses on regular arrangements for solid phases in the porous medium. In this
study, flows in both the regular and random arrangements were studied
numerically. Both cases were compared to see the deviation from the flow in an
9
ideal regular arrangement.
Chapter 4 is devoted to the development of a new method of ensemble
averaging to predict the average flow parameters in the porous-medium problem.
We propose a new method to estimate the flow parameters like velocity,
permeability, etc. while avoiding the conventional representative elementary
volume (REV) approach. In the REV method, the size of the REV has always been
problematic since it is difficult to quantify the size of this volume, especially when
the problem involves a sharp gradient of a flow variable. Using the proposed
ensemble averaging method, one can confidently deals with porous-media
problems without worrying about the size of the REV, especially in complex
geometries. Two problems have been solved using the ensemble averaging method
and the results are compared to the available results of the REV method to show
the simplicity and the accuracy of the method.
Chapter 5 focuses on the thermal transport during fluid flow in porous
media where the governing equation for non-isothermal flow is considered. One of
the main issues with LCM in the non-isothermal flow problems of dual-scale
porous media found is prediction of the dispersion tensor. Most of the previous
efforts in this area have been devoted to random fiber mats (a single-scale porous
medium). In many industrial applications, however, the dual-scale porous media in
the form of woven and stitched fiber mats are used. A new technique is proposed
in this dissertation to predict the dispersion tensor in the thermal governing
equation in dual-scale porous media. This method is a combination of a finite-

10
element based solution and an experimental method to find the dispersion tensor.
Knowing the dispersion tensor in dual-scale porous media, one can numerically
model the heat transfer aspects of flow through porous media in the composite
industry.
Finally, in chapter 6, the summary is presented.
$
11
References
1. Rudd, CD., long, A.C., Kendall, K.N. and Mangin C.G.E., Liquid
molding technologies, Woodhead Publishing Ltd, 1997.
2. Parnas, R.S., A.J. Salem, T.A.K. Sadiq, et al., Interaction between
Micro- and Macro-Scopic Flow in RTM Preforms. Composite
Structures, 1994. 27(1-2): p. 93-107.
3. Tucker III, C.L. and Dessenberger, R.B., Governing equations for
flow and heat transfer in stationary fiber bed, Flow and rheology in
polymer composites manufacturing, Elsevier, Amsterdam, pp. 257-
323, 1994.
4. Parnas, R.S. and F.R. Phelan, Jr., Effect of Heterogeneous Porous
Media on Mold Filling in Resin Transfer Molding. SAMPE Quarterly,
1991. 22(2): p. 53-60.
5. Sadiq, T.A.K., S.G. Advani, and R.S. Parnas, Experimental
Investigation of Transverse Flow through Aligned Cylinders.
International Journal of Multiphase Flow, 1995. 21(5): p. 755-774.
6. Binetruy, C , B. Hilaire, and J. Pabiot, Interactions between Flows
Occurring inside and Outside Fabric Tows During RTM. Composites
Science and Technology, 1997. 57(5): p. 587-596.

12
7. De Parseval, Y., K.M. Pillai, and S.G. Advani, Simple Model for the
Variation of Permeability Due to Partial Saturation in Dual Scale
Porous Media. Transport in Porous Media, 1997. 27(3): p. 243-264.
8. Slade, J., K.M. Pillai, and S.G. Advani, Investigation of Unsaturated
Flow in Woven, Braided and Stitched Fiber Mats During Mold-Filling
in Resin Transfer Molding. Polymer Composites, 2001. 22(4): p. 491-
505.
9. Dungan, F.D. and A.M. Sastry, Saturated and Unsaturated Polymer
Flows: Microphenomena and Modeling. Journal of Composite
Materials, 2002. 36(13): p. 1581-1603.
lO.Breard, J., Y. Henzel, F. Trochu, et al., Analysis of Dynamic Flows
through Porous Media. Part I: Comparison between Saturated and
Unsaturated Flows in Fibrous Reinforcements. Polymer Composites,
2003. 24(3): p. 391-408.
ll.Babu, B.Z. and K.M. Pillai, Experimental Investigation of the Effect
of Fiber-Mat Architecture on the Unsaturated Flow in Liquid
Composite Molding. Journal of Composite Materials, 2004. 38(1): p.
57-79.
12.Kuentzer, N., P. Simacek, S.G. Advani, et al., Permeability
Characterization of Dual Scale Fibrous Porous Media. Composites

13
Part A: Applied Science and Manufacturing, 2006. 37(11): p. 2057-
2068.
13.Tan, H., T. Roy, and K.M. Pillai, Variations in Unsaturated Flow with
Flow Direction in Resin Transfer Molding: An Experimental
Investigation. Composites Part A: Applied Science and
Manufacturing, 2007. 38(8): p. 1872-1892.
14.Binetruy, C , B. Hilaire, and J. Pabiot, Tow Impregnation Model and
Void Formation Mechanisms During RTM. Journal of Composite
Materials, 1998. 32(3): p. 223-245.
15.Chan, A.W. and R.J. Morgan, Tow Impregnation During Resin
Transfer Molding of Bi-Directional Nonwoven Fabrics. Polymer
Composites, 1993. 14(4): p. 335- 340.
16.Chang, C.-Y. and L.-W. Hourng, Numerical Simulation for the
Transverse Impregnation in Resin Transfer Molding. Journal of
Reinforced Plastics and Composites, 1998. 17(2): p. 165-182.
17.Lekakou, C. and M.G. Bader, Mathematical Modelling of Macro- and
Micro- Infiltration in Resin Transfer Moulding (RTM). Composites -
Part A: Applied Science and Manufacturing, 1998. 29(1-2): p. 29-37.
18.Pillai, K.M. and S.G. Advani, A Model for Unsaturated Flow in
Woven Fiber Preforms During Mold Filling in Resin Transfer

14
Molding. Journal of Composite Materials, 1998. 32(19): p. 1753-
1783.
19.Phelan, F.R., Jr. Modeling of Microscale Flow in Fibrous Porous
Media. 1991. Detroit, MI, USA: Publ by Springer-Verlag New York
Inc., New York, NY, USA.
20.Pillai, K.M., Governing Equations for Unsaturated Flow through
Woven Fiber Mats. Part 1. Isothermal Flows. Composites - Part A:
Applied Science and Manufacturing, 2002. 33(7): p. 1007-1019.
21.Pillai, K.M. and M.S. Munagavalasa, Governing Equations for
Unsaturated Flow through Woven Fiber Mats. Part 2. Non-Isothermal
Reactive Flows. Composites Part A: Applied Science and
Manufacturing, 2004. 35(4): p. 403-415.
22.Wang, Y. and S.M. Grove, Modelling Microscopic Flow in Woven
Fabric Reinforcements and Its Application in Dual-Scale Resin
Infusion Modeling. Composites Part A: Applied Science and
Manufacturing, 2008. 39(5): p. 843-855.
23.Wang, Y., M. Moatamedi, and S.M. Grove, Continuum Dual-Scale
Modeling of Liquid Composite Molding Processes. Journal of
Reinforced Plastics and Composites, 2008: p. 1469-1484.

15
24.Jadhav, R.S. and K.M. Pillai, Numerical Study of Heat Transfer
During Unsaturated Flow in Dual-Scale Porous Media. Numerical
Heat Transfer; Part A: Applications, 2003. 43(4): p. 385-407.
25.Pillai, K.M. and R.S. Jadhav, A Numerical Study of Nonisothermal
Reactive Flow in a Dual-Scale Porous Medium under Partial
Saturation. Numerical Heat Transfer; Part A: Applications, 2005.
47(2): p. 109-136.
26.Williams, J.G., C.E.M. Morris, and B.C. Ennis, Liquid Flow through
Aligned Fiber Beds. Polymer Engineering & Science, 1974. 14(6): p.
413-419.
27.Gutowski, T.G., Z. Cai, S. Bauer, et al., Consolidation Experiments
for Laminate Composites. Journal of Composite Materials, 1987.
21(7): p. 650-669.
16
Chapter 2
EFFECT OF ASPECT RATIO ON MEASURED
PERMEABILITY AND FLOW-FRONT
PROGRESS
2.1 Permeability Measurements
Permeability is described as the ease with which a fluid flows in a porous
medium-material where a larger permeability indicates the fluid flows more easily.
Measuring the permeability of a porous medium is performed by using the Darcy's
law, Eqn. (1.1), along with flow experiments. Two famous methods for measuring
the permeability of a preform are the 1-D flow experiment and the radial flow
experiment [1-28]. Both of the mentioned methods can be carried out under
either steady state or transient conditions. The saturated permeability is the
measured permeability where the preform is fully wetted by the flowing fluid
under steady-state conditions. The unsaturated permeability is the permeability
measured using the transient mold-filling flow which is marked by partial
saturation of the fiber mat near the liquid-front. Some researchers have shown that
unlike saturated permeability, the unsaturated permeability is not only a function
of a fabric's architecture, porosity and diameter, but also a function of the fluid
properties as well as the flow-rate [6,10-12,15,19-21,23,25,28].

17
2.1.11-D Flow experiment
The 1-D flow experiment is conducted in a rectangular mold with its length
much larger than the width. The sample preform is placed inside the mold and the
test fluid is injected either at constant flow-rate or under constant injection-
pressure along the mold length through the preform [1-5, 7-12, 17-25, 27]. In
order to obtain the in-plane permeability tensor of an anisotropic porous medium,
one needs to conduct the 1-D flow experiments three times with different
orientations with respect to flow direction [9,10,19-21,4,25,27]. A schematic view
of 1-D flow experiment setup is shown in the Fig. 2.1. The unidirectional flow is
created within the preform inside the mold cavity when the test liquid is injected
from one end. The mold consists of two aluminum plates that sandwiches the
optical grade- acrylic sheet between them. The 1" thick acrylic sheet has very high
flexural-stiffness to minimize deflections under the proposed high-pressure test
conditions. The mold cavity is a 3/8" deep and 7" wide pocket which is sealed
with two o-rings along the cavity edge where the acrylic plate rests. There are
thirty five evenly spaced bolts which clamp the mold's plates to ensure the mold's
sealing and avoid plate deflection due to high pressure.

18
computer
8
^ ^ f *Woodata stepper-motor
\ Gear ounu
cant rallied bjf
stepper motor
piston
cyl-lndei:
Figure 2.1 Schematic view of 1-D flow experiment
2.1.2 Radial Flow Experiment
Radial flow experiment is mainly injection of a test liquid from an inlet port
centered in the mold where it creates a radial flow in the mold cavity [6, 13-28].
For obtaining the permeability tensor using this method, one single experiment is
enough and involves observing the shape of the radial flow-front during the fluid
impregnation [13-28]. A schematic view of the radial flow experimental setup is
shown in Fig. 2.2. The mold is composed of three parts. The top and bottom parts
are 53cmx53cmx2.54cm-transparent polycarbonate plates. The center part is a
0.9525 cm thick steel spacer. The two thick transparent plates sandwich the steel
spacer forming a cavity for packing the fiber mat. Sixteen equally spaced nuts and
19
bolts along the edges ensure enough compression of the fiber mat. Four pressure
transducers placed on the top surface of the setup read pressure at four different
locations.
Figure 2 2 Schematic view of radial flow experiment
2.2 Effect of Preform Aspect Ratio on the Permeability Measured
through the 1-D Flow Experiment
Numerous studies have reported the measured permeability values for
various types of fiber mats. This physical quantity has been measured by several
measurement techniques, such as 1-D flow and radial flow methods [1-9]. The
microstructure of most structural fiber mat materials consists of a network of fiber
bundles called tows. Arrangements of these tows range in configuration from a
simple, unidirectional alignment to complex weaves and braids. Similarly, the
steady-state (saturated) flow method and the transient (unsaturated) flow method
20
have been used as two separate ways of estimating the permeability [1-9].
Although measurement of permeability of different kinds of fiber mats is
crucial for an accurate mold-filling simulation for an LCM process, very little
effort has been made to investigate the accuracy of permeability measurement
using the 1-D flow method. Lundsrtom et al. [8] used the multi-cavity parallel
flow technique to measure the in-plane permeability of anisotropic fiber mats.
They reported that two ratios can influence the accuracy of 1-D flow basic
permeability measurements method. One of these ratios is the ratio of preform's
length L to its width W (henceforth to be called the aspect ratio LIW). The other
is the ratio between two in-plane principal permeabilities, K{ and K2 (henceforth
to be called the anisotropy ratio K{ /K2). These two ratios influence the accuracy
of the theoretical solution on which evaluation of the permeability through the 1-D
flow experiment is based. Lundstrom et al. [8] ran numerical simulations to
estimate the error in the effective permeability estimation through this
unidirectional approximation. Their results showed that the accuracy of the
permeability measurement techniques depends on the aspect ratio, the anisotropy
ratio, and the angle of the principal permeability direction with respect to the flow
direction. Tan and Pillai [29] ran numerical mold-filling simulations in a
rectangular 1-D flow mold for a given permeability tensor under constant
injection-rate and uniform inlet-pressure conditions to investigate how the inlet
pressure history and the flow-front patterns are influenced by the aspect ratio, the
anisotropy ratio, and the principal permeability angle. Their results showed that
21
these three factors have strong influence on the inlet pressure history as well as the
final steady-state pressure achieved. They also found that the simulated flow-
fronts differ significantly (were observed to be tilted) from the straight flow-fronts
predicted by the 1-D flow model.
In this reported work, experimental mold-filling investigations were carried
out in a rectangular 1-D flow mold under constant injection rate condition to study
the effect of the aspect ratio, ratio of the preform's length to its width, on the
accuracy of permeability measured through the 1-D flow experiment.
2.2.1 Theory
For an anisotropic fiber mat, the permeability is a tensor rather than a scalar
quantity. However, during the 1-D flow in an anisotropic fiber-mat, the Darcy's
law is given as
<v>=-^V<R>f (2.1)
H
where < v > is the volume-averaged flow velocity, < Pf >f is the pore-averaged
liquid pressure, and Keffis the effective permeability along the flow direction in the
1-D mold. It is essential to know the relation between the effective permeability
and the principal values of the permeability tensor to determine all the components
of the permeability tensor.
Darcy's law given in Eqn. (1.1) can be expanded in a matrix form as

22
af
l<v>>( Ku K12 d\{
(2.2)
[< v 2 >J v'
JV
v' 3'
_ 21 ^22 _
3x2
where < v, > and < v2 > are the volume-averaged velocity components along the
flow (x,) and the transverse (x2) directions, respectively. It is assumed that the
unidirectional flow observed during the 1-D flow experiment implies that the
velocity along the x2 direction is uniformly zero in the flow domain, i.e.,
< v 2 >= 0 (2.3)
By substituting < v2 > from Eqn. (2.3) into Eqn. (2.2) and simplifying the
equation, the following relation between the effective permeability and
components of the permeability tensor can be established.
K"2
K — ir' 12 (2.4)
22
If the principal axis of the permeability tensor is orientated at an angle 6 to the
flow direction, all components of the permeability tensor, in terms of the two
principal permeability components and the orientations, are as follows:
K' =K,cos 2 e+K,sin 2 9
K' = K , sin 2 6+K,cos 2 9 (2.5)
K
i2 = K 2i = ( K 2 - K , ) sinG cosG
Here K, and K2 are the two in-plane principal permeability components and 6 is
23
the angle of orientation of K, with respect to the flow direction.
Finally, upon substituting Eqn. (2.5) in Eqn. (2.4), the effective permeability K
can be expressed as a function of K,, K2 and d as
l)2
Kerf =K,.cos : l + ^ t a n 2 9 - - ' (2.6)
K, K.
+1
K, tan2 9
Hence, three different mat-orientation experiments need to be done to determine
the principal permeabilities, KândK.,, and the orientation0, Fig. 2.3.
Ill, Y
I,X
Figure 2.3 Schematic description of the three 1-D flow experiments.
As shown in the Fig. 2.3, these three directions are defined as I, II and III for the
angle of#, = 0, du = 0 + 45° and 6m =0 + 90° respectively. Based on Eqn. (2.5),
the effective permeability along these three directions are

24
KXK2
K, (2.7)
KxSin 6,+K2Cos2dl
2
K
KXK2
A (2.8)
// Kx Sin 6u + K2Cos2du
2
KXK2
V (2.9)
Kx Sin 0,„ + K2Cos2em
2
By defining
A=K,+Kl" (2.10)
2
D=K'~K'" (2.11)
2
Using above relations, one can transform Eqns. (2.7)-(2.9) to
Kx =K,——cos20 (2.12)
1
' A-D
K2=Km^-^cos20 (2.13)
2
'" A + D
^=-tan-'{A2~D2--} (2.14)
2 K„D D
Four different aspect ratios were investigated experimentally to establish the
dependence of the accuracy of the 1-D flow based permeability measurement
technique on the aspect ratio. Because the higher aspect ratios should follow the 1-
D assumption < v2 >= 0 more closely than lower aspect ratios, results from the
higher aspect ratio experiments should be more accurate. To support this idea, a
radial flow experiment with the same bi-axial fiber mat was performed. The
25
completely circular flow front occurring in that experiment showed that AT, and
K2 should be the same. Using experimental data in Eqn. (2.11), one obtains three
equations with three unknowns, Ki,, K22 and 0, that are essential for constructing
the full permeability tensor.
2.2.2 Experiment
Details of the preform used in this investigation are presented in Table 2.1.
Bi-axial fiber mat was selected for the preform in the experiments. The test liquid,
motor oil SAE 10W-40, wetted the preform at a constant flow rate. The mold was
so designed that it can be used for different aspect ratios. Details for each aspect
ratio are presented in the Table 2.1. Nine layers of reinforcement were placed in
the cavity to give the desired value of fiber volume fraction. Three preforms
oriented in three different directions were prepared for each aspect ratio.
Table 2.1 Details of experiments
Aspect ratio Dimensions Layers Number of experiments
1.0 5"x5" 9 3
2.0 10"x5" 9 3
3.0 15 x5 9 3
4.0 20"x5" 9 3
A pressure transducer was coupled to a digital voltmeter for recording the pressure
26
history. In order to ensure the repeatability, we often conducted the experiments
multiple times to obtain more than one measured values. The results show a good
repeatability in experiments.
2.2.3 Results and Discussions
Recalling the assumption used in the 1-D flow experiment, < v2 >=0, it is
clear that the longer aspect ratios should lead to more accurate results in
permeability measurements. The radial flow experiment was done using a bi-axial
fiber mat. Completely circular flow-front of the liquid penetrating the preform
shows that the principal permeability components should be the same (i.e. Kx = K2
). The flow front for the radial flow experiment is shown in the Fig. 2.4.
Circularity of the flow front is demonstrated by the fact that the three pressure
transducers placed symmetrically and at equal distance from the central injection
hole.
Figure 2.4 Radial flow experiment atflow-rateof 0.00256 ml/s.

27
Running three experiments along three different orientations of fiber-mat,
one can estimate the principal components of the permeability tensor. Fig. 2.5
shows the pressure plot as a function of time for an aspect ratio of 1 for the 3
different directions. Similar results were observed for the other aspect ratios with a
slight difference that the pressure histories of 0 degree and 90 degree orientations
became more similar as the aspect ratio increased.
70-
Aspect Ratio: 1
0 Degree — 45 Degree - - - 90 Degree

60
50 -
«T
^ 40 _ _ —-
/ S —
3
(A
2 30
a.
20
§ J*
10
if
0 121
Time (Sec)
Figure 2.5 The pressure history for three different orientations for the aspect ratio 1.
Figs. 2.6 and 2.7 show the decrease in the permeability measurement error
with an increase in the aspect ratio. Two different error criteria are used here to
focus more on the effect of the longer aspect ratios decreasing the error
incorporated with the permeability measurement using 1-D flow experiment. In

28
Fig. 2.6, the difference between the two principal directions of the permeability is
used as a criterion for accuracy. According to the results from the radial flow
experiment, the two principal directions of the permeability should be the same,
but different 1-D flow experiments show that the difference between these two
values decreases with increasing aspect ratio. The difference between Kl and K2
is 1.17 e -9 for an aspect ratio of 1. Fig. 2.6 shows that difference decreases to a
low of 4.26 e-10 for an aspect ratio of 4 which indicates that the accuracy
increases with aspect ratios.
1.4E-09
1.1737E-09
1.2E-09
•J.02567E-09
1E-09 -
8E-10
CM
6E-10
^V 5.04731 E-10
4E-10 -
4.26465E-10
2E-10
1 2 3
Aspect Ratio
Figure 2.6 Permeability measurement error, K, - K 2 , versus the aspect ratio.
Fig. 2.7 discusses another accuracy criterion for the permeability

29
K —K
measurement using 1-D flow experiment. The parameter —! indicating the %
error is selected as another criterion. The plot shows that this error value falls
gradually as the aspect ratio increases. The same conclusion as before can be
obtained from this figure: higher aspect ratios lead to more accurate permeability
measurements in the 1-D flow apparatus.
Ky —K2
Figure 2.7 Permeability measurement error, , versus the aspect ratio.
Some error predictions have been done for other aspect ratios in 1-D flow
experiments. Using the data from the specified aspect ratios used in these
experiments and fitting two different curves, one linear and one quadric, to these
data, one can estimate the permeability measurement error for other aspect ratios.
30
For measuring the error, it is needed to replace the aspect ratio number by X, and Y
would then be the predicted error of permeability measurements for that aspect
ratio. The correlation coefficient, r, is also presented for each curve fitting to
ensure the accuracy of the prediction equations. Using the general linear equation
Y - aX + b and general quadric equation Y = a + bX + cX2 and fitting these two
equations to the A", - K2 error data, results are obtained as shown in Table 2.2.
Table 2.2 Coefficients of two equations fitted to A", - K2 error data.
Y = aX + b Y = a + bX+cX2
a 1.473 e-9 a 1.561 e -9
b -2.763 e -10 b -3.635 e -10
— — c 1.744 e-11
r 0.958 r 0.959
Fitting the same equations to the —l- error data obtained through the
experiments, the following results in Table 2.3 can be used for the error prediction
of other aspect ratios.

31
Table 2.3 Coefficients of two equations fitted to error data.
Y = aX +b Y = a + bX +cX2
a 1.485 e-9 a 2.660 e -9
b -0.366 e -10 b -1.541 e -10
— — c 0.235 e -11
r 0.855 r 0.986
2.3 Effect of Preform Aspect Ratio on the Unidirectional Flow-Front
during Mold-Filling in the LCM
The main apparent characteristics of unsaturated flow through dual-scale
porous media are:
(1) A drooping inlet-pressure history is observed for woven or stitched fiber mats
while the pressure history is linear for single-scale porous media.
(2) Unlike the full-saturation observed behind the resin front for single-scale
porous media, a partially saturated region behind the resin front is observed for
dual-scale porous media.
The presence of unsaturated region in dual-scale porous media is due to the
dual-length scale pore space created in porous media by fiber tows. Pores present
between tows in such fiber mats are of a higher size compared to the microscopic
pores present inside fiber tows. Such a dual-scale cavity causes the invading resin
32
to preferentially go around the fiber tows and fill up the large pores between tows
before penetrating the fiber tows. Such delayed impregnation of tow acts like a
sink of liquid in the macroscopic flow field. A sink model based on adding a
nonzero sink function to the equation of continuity is proposed to describe such an
unsaturated flow in dual-scale porous media [1-2, 7]:
V.v=-S (2.15)
Here S , the sink term, is equal to the volumetric rate of liquid absorption.
The presence of the sink term in the continuity equation successfully explains the
drop in the inlet-pressure history seen during the unsaturated flows in dual-scale
porous media.
Fig. 2.8 shows the setup used for the 1-D flow experiments with fiber
performs of different aspect ratios. Experiments are done for four different aspect
ratios with the same flow rate and same number of layers.
33
2.3.1 Results and Discussion
Unsaturated flow front of the resin in dual-scale porous media is presented
here for the aspect ratio of 4 at the three different orientations of 0, 45 and 90
degrees (Fig. 2.9). The slides were taken at the intervals of 20 seconds.
(A) 0 Degree
(B) 45 Degree
(C) 90 Degree
Figure 2.9 Unsaturated flow-front at different mat orientations for aspect ratio of 4.
Fig. 2.9 shows that the unsaturated flow-front for the orientation of the 45
34
degree of mat is straighter than the two other orientations of 0 and 90 degrees. The
two other mat orientation lead to more curved unsaturated flow-fronts at the fluid
flow in a main principal direction of the fiber mat.
Fig. 2.10 shows that the flow-front of the filling process in RTM vary
significantly for the single-scale and dual-scale porous media. The numerical
simulation is done through the PORE-FLOW® code owned by the Laboratory for
Flow and Transport Studies in Porous Media at University of Wisconsin-
Milwaukee. This is a finite element/control volume code to simulate resin flow in
single-scale and dual-scale porous media. This 2D/3D code can accept structured
and unstructured mesh. The flow can be governed by either Darcy's law or
Brinkman equation in porous media and Stokes/Navier-Stokes equations in fluid
flow problems. The simulation of the flow-front for the single-scale porous media
shows that the flow-front is straight and tilted while the flow-front for the bia-axial
fiber-mat reveal the curvature characteristic of the dual-scale porous media. It can
be easily concluded that one cannot easily use the single-scale analysis for the
dual-scale case.
flow front
141 00
131.64
122.29
112.93
'imii-ill'III
103.57
1 94.21
• ! J.
11. 11! 84.86
111 75.50
66.14
56.79
47.43
38.07
28 71
19 36
10.00
Figure 2.10 Flow-front of the filling process in the dual-scale porous media for (a): Numerical
modeling at different time steps (second) (b): Experimental results
the Fig. 2.1:

36
Figure 2.11 Unsaturated flow-front at four different aspect ratios of 1, 2, 3 and 4, from left to right.
Fig. 2.12 shows the differences in the pressure histories for the single-scale
and dual-scale porous media. It can be observed that the pressure profile is curved
for the bia-axial mat while it is linear for single-scale porous media. This plot
shows a large difference in the pressure histories of the single- and dual-scale fiber
mats.
- Experimental result of dual scale porous media
-Numerical result of single scale porous media
0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95 100 105 110 115 120 125 130 135 140

Time (sec)
Figure 2.12 Pressure history profile for both cases of single scale and dual scale porous media
37
2.4 Conclusion
Permeability measurement technique used in 1-D flow experiments has been
investigated experimentally for four different aspect ratios both in steady-state and
transient conditions. Two different error criteria have been defined to gauge the
accuracy of permeability measurement, and linear and quadric equations have
been fitted to those data to predict the error in permeability measurement due to
perform aspect ratio using a biaxial fiber mat in 1-D flow experiments. The
quadric equation fitted to the error data has a better correlation coefficient than the
linear equation. Results, using both criteria, indicate that permeability
measurement accuracy for biaxial fiber mat in ID flow experiments increases with
increasing aspect ratio. Curvature pressure history profile and curvature flow-front
show the huge difference between the single-scale and the dual-scale porous
media characteristics. Experimental results reveal that using higher aspect ratios of
the preform leads to a straighter flow-front. Numerical simulation for single-scale
porous media can not be used to predict inlet-pressure histories for dual-scale fiber
mats.
38
References
1. Gauvin, R., M. Chibani, and P. Lafontaine, Modeling of Pressure
Distribution in Resin Transfer Molding. Journal of Reinforced Plastics
and Composites, 1987. 6(4): p. 367-377.
2. Trevino, L., K. Rupel, W.B. Young, et al., Analysis of Resin
Injection-Molding in Molds with Preplaced Fiber Mats. Part 1:
Permeability and Compressibility Measurements. Polymer
Composites, 1991. 12(1): p. 20-29.
3. Gebart, B.R., Permeability of Unidirectional Reinforcements for
RTM. Journal of Composite Materials, 1992. 26(8): p. 1100-1133.
4. Young, W.B. and S.F. Wu, Permeability Measurement of
Bidirectional Woven Glass Fibers. Journal of Reinforced Plastics and
Composites, 1995. 14(10): p. 1108-1120.
5. Ferland, P., D. Guittard, and F. Trochu, Concurrent Methods for
Permeability Measurement in Resin Transfer Molding. Polymer
Composites, 1996. 17(1): p. 149-158.
6. Hammond, V.H. and A.C. Loos, The Effects of Fluid Type and
Viscosity on the Steady-State and Advancing Front Permeability
Behavior of Textile Preforms. Journal of Reinforced Plastics and
Composites, 1997. 16(1): p. 50-72.

39
7. Gebart, B.R. and P. Lidstrom, Measurement of in-Plane Permeability
of Anisotropic Fiber Reinforcements. Polymer Composites, 1996.
17(1): p. 43-51.
8. Lundstrom, T.S., B.R. Gebart, and E. Sandlund, In-Plane Permeability
Measurements on Fiber Reinforcements by the Multi-Cavity Parallel
Flow Technique. Polymer Composites, 1999. 20(1): p. 146-154.
9. Ma, Y.L. and R. Shishoo, Permeability Characterization of Different
Architectural Fabrics. Journal of Composite Materials, 1999. 33(8): p.
729-750.
lO.Shojaei, A., F. Trochu, S.R. Ghaffarian, et al., An Experimental Study
of Saturated and Unsaturated Permeabilities in Resin Transfer
Molding Based on Unidirectional Flow Measurements. Journal of
Reinforced Plastics and Composites, 2004. 23(14): p. 1515-1536.
11.Adams, K.L., B. Miller, and L. Rebenfeld, Forced in-Plane Flow of an
Epoxy Resin in Fibrous Networks. Polymer Engineering and Science,
1986. 26(20): p. 1434-1441.
12.Chan, A.W. and S.T. Hwang, Anisotropic in-Plane Permeability of
Fabric Media. Polymer Engineering & Science, 1991. 31(16): p.
1233-1239.
40
13.Chan, A.W., D.E. Larive, and R.J. Morgan, Anisotropic Permeability
of Fiber Preforms: Constant Flow Rate Measurement. Journal of
Composite Materials, 1993. 27(10): p. 996-1008.
14.Hammami, A., F. Trochu, R. Gauvin, et al., Directional Permeability
Measurement of Deformed Reinforcement. Journal of Reinforced
Plastics and Composites, 1996. 15(6): p. 552-562.
15.Parnas, R.S. and A.J. Salem, A Comparison of the Unidirectional and
Radial Inplane Flow of Fluids through Woven Composite
Reinforcements. Polymer Composites, 1993. 14(5): p. 383-394.
16.Parnas, R.S., A.J. Salem, T.A.K. Sadiq, et al., The Interaction between
Microscopic and Macroscopic Flow in RTM Preforms. Composite
Structures, 1994. 27(1-2): p. 93-107.
17.Parnas, R.S., J.G. Howard, T.L. Luce, et al., Permeability
Characterization. Part 1: A Proposed Standard Reference Fabric for
Permeability. Polymer Composites, 1995. 16(6): p. 429-445.
18.Ahn, S.H., W.I. Lee, and G.S. Springer, Measurement of the 3-
Dimensional Permeability of Fiber Preforms Using Embedded Fiber
Optic Sensors. Journal of Composite Materials, 1995. 29(6): p. 714-
733.
41
19.Wang, T.J., C.H. Wu, and L.J. Lee, In-Plane Permeability
Measurement and Analysis in Liquid Composite Molding. Polymer
Composites, 1994. 15(4): p. 278- 288.
20.Gauvin, R., F. Trochu, Y. Lemenn, et al., Permeability Measurement
and Flow Simulation through Fiber Reinforcement. Polymer
Composites, 1996. 17(1): p. 34-42.
21.Lekakou, C , M.A.K. Johari, D. Norman, et al., Measurement
Techniques and Effects on in-Plane Permeability of Woven Cloths in
Resin Transfer Moulding. Composites Part a-Applied Science and
Manufacturing, 1996. 27(5): p. 401-408.
22.Lai, Y.H., B. Khomami, and J.L. Kardos, Accurate Permeability
Characterization of Preforms Used in Polymer Matrix Composite
Fabrication Processes. Polymer Composites, 1997. 18(3): p. 368-377.
23.Parnas, R.S., K.M. Flynn, and M.E. DalFavero, A Permeability
Database for Composites Manufacturing. Polymer Composites, 1997.
18(5): p. 623-633. 28.
24.Han, K.K., C.W. Lee, and B.P. Rice, Measurements of the
Permeability of Fiber Preforms and Applications. Composites Science
and Technology, 2000. 60(12- 13): p. 2435-2441.

42
25.Lundstrom, T.S., R. Stenberg, R. Bergstrom, et al., In-Plane
Permeability Measurements: A Nordic Round-Robin Study.
Composites - Part A: Applied Science and Manufacturing, 2000.
31(1): p. 29-43.
26.Kim, S.K. and I.M. Daniel, Observation of Permeability Dependence
on Flow Rate and Implications for Liquid Composite Molding.
Journal of Composite Materials, 2007. 41(7): p. 837-849.
27.Parnas R.S. and Salem A.J. (1993), A comparison of the
unidirectional and radial flow of fluids through the woven composite
reinforcements, Polymer composites, 14: 383-397.
28.Han K.K., Lee C.W. and Rice B.P. (2000), Measurements of the
permeability of fiber preforms and application, Composites science
and technology, 60 (12): 2435-2441.
29.Tan H. and Pillai K.M., (2007), Effects of fiber-mat anisotropy on 1-D
mold filling in LCM: A numerical investigation, Polymer composites,
DOI 10.1002/pc.20463.
43
Chapter 3
Evaluation of Boundary Conditions at the Clear-
Fluid Porous-Medium Interface
3.1 Introduction
Transport phenomena in a fluid channel bound by a porous medium layer
are of great importance in a wide range of industrial, environmental and medical
applications. Among the several simulation applications, one can mention
modeling of liquid composite molding (a composites manufacturing process), fuel-
cell transport, filtration of contaminants, drying processes, ground water pollution,
and overland-flow infiltration into soil during rainfalls, as well as applications in
the medical field such as simulation of flows in biological tissues, drugs and
nutrients transport in the body, etc. Transfer of shear stress from the open channel
into the interstitial fluid domains of the porous medium leads to the formation of a
boundary layer inside the porous medium at the open-channel porous-medium
interface. Though Beavers and Joseph [1] had demonstrated that mass flow-rate
through open channels bound by porous walls is higher than for channels bound
by solid walls, thereby inferring the existence of such a boundary layer indirectly,
it was only recently that Tachie et al. [2] showed conclusively through
44
experiments that such a boundary layer region does exist.
In order to solve this problem numerically, one need to establish appropriate
governing equations in the clear-fluid and porous-medium domains. The Navier-
Stokes equations are used to model the flow through the free channel, while the
Brinkman-Forchheimer equation (a.k.a. the Darcy-Brinkman-Forchheimer
equation) is used to model the fluid flow through the porous wall [3-7]. The latter
models the volume-averaged flow through a porous medium close to its clear fluid
boundary. For the low particle-based Reynolds-number flows (i.e. Re < 1), one
can neglect inertial effects in the Brinkman-Forchheimer equation and reduce it to
the well-know Brinkman equation [1]. Additional simplifications can be
introduced by neglecting the second-order viscous term in the Brinkman equation
for the flow of fluid far away from the interface; then the Brinkman equation
reduces to the well-known Darcy's law for modeling flows in ordinary porous
media.
The boundary conditions at the clear-fluid porous-medium interface have
been investigated by several researchers in the past [8-15]. In 1967, Beavers and
Joseph [1] introduced the slip-velocity boundary condition at this interface, which
has later been used by other researchers [8-11] to solve such problems
numerically. Ochoa-Tapia and Whitaker [12-14] introduced a different interfacial
boundary condition which assumed a jump in stress at the interface. They used the
rigorous volume averaging method to show the inconsistency in the stress

45
estimation at the clear-fluid/porous-medium interface. Valdes-Parada et al. [15]
developed the work done by Ochoa-Tapia and Whitaker further to provide an
expression of the stress jump boundary condition at the free channel/porous wall
interface that does not involve adjustable coefficients. They proposed an
approximate methodology to estimate the mixed stress tensor of the stress jump
boundary condition. In another work, they [16] used the same boundary condition
to model the mass transfer at the clear-flow/porous-medium interface. Recently,
Chandesris and Jamet [17,18] developed the stress jump boundary condition using
the matched asymptotic expansion method. They used a two-step up-scaling
approach for three different interface descriptions to identify the characteristic
length scales associated to each level of interface. Later in 2009, the same authors
[19] used the identical method to recover the empirical slip coefficient of Beavers
and Joseph [1] through the multi-scale approach. Duman and Shavit [20] used the
apparent interface concept to develop the stress jump boundary condition at the
clear-fluid/porous media interface. The stress jump boundary condition introduced
by Ochoa-Tapia and Whitaker [12] is very sensitive to the value of the parameter
P and exact location of the interface. They used an alternative method to avoid
the necessity of finding the exact location of the interface; their method needs only
one measurable parameter, the maximum velocity or the flow-rate. Tan et al. [21]
used boundary element method (BEM) to compare the three different interfacial
boundary conditions: Beavers-Joseph's slip-velocity condition, Brinkman's stress-
continuity condition and Whitaker's stress jump condition. They found that the
46
stress-jump condition gives more accurate results for high porosities; however, all
the three conditions are inaccurate at lower values of porosities. Later, Tan and
Pillai [22] used the Galerkin finite element based simulation to model flow at the
clear channel/porous media channel interface using both slip flow and stress jump
boundary conditions. They used a second order adjustable tensor to implement the
stress jump condition for three dimensional flows.
Some of the above listed papers (Larson and Higdon [10,11], Tan et al. [21])
as well as the papers listed later on (Sahraoui and Kaviany [23]) pertain to the
use of flow simulations in a virtual porous medium to either test the efficacy of
various boundary conditions or to estimate relevant parameters for them. Most of
these efforts have been done on periodic models of porous media that assume an
ideal, square-lattice arrangement of cylinders within a rectangular domain. On the
other hand, a non-periodic porous medium with a random placement of cylinders
is closer to the real porous media found in nature. To the best of our knowledge,
no ejfort has been made till date to investigate the boundary conditions adjacent
to a free channel flow on non-periodic or irregular porous-media. Our focus in
this chapter is to explore the efficacy of different boundary conditions at the clear-
fluid/porous-medium interface in a non-periodic porous medium. Our hope is that
such an approach to test the accuracy of the various boundary conditions is more
realistic and should yield results that are observed in nature as well. For the
purposes of comparison, we will also present the simulation results involving the
47
periodic porous media.
3.2 Design of Numerical Experiments
In this chapter, we will restrict our analysis to the creeping or Stokes flow
since the majority of flows involving porous media fall in that regime. Such a flow
will be studied adjacent to a porous medium in a domain to be called a 'hybrid'
domain (Fig. 3.1 A). In order to model the hybrid region including both periodic
and non-periodic porous regions, two different rectangular geometries have been
created. The porous medium is made of a square array of cylinders (representing
the solid phase) for the periodic case (Fig. 3.1 B), and is made of a random
arrangement of cylinders for the non-periodic case (Fig. 3.1 C). In both cases,
cylinders of radii a=10 |^m are used and the depth of the clear fluid region is h=20
um. The porosities of these geometries can be changed from 0.5 to 0.9 by varying
the inter-cylinder spacing, e.
In order to create a periodic porous medium with a specific porosity,
cylindrical-solid phase particles are located at fixed horizontal and vertical
distances from each other in the rectangular geometry.
On the other hand, to create a random distribution of cylinders in the non-
periodic porous medium at a specific porosity, a FORTRAN code is developed to
randomly place the solid-phase cylinders inside the rectangular geometry; the code
estimates the horizontal and vertical distances between cylinders using a random
number generator after ensuring no overlap of cylinders (see appendix for details).
48
ys///y/y/ss//!&///y//y^^ Y//y////A'Awy//y/AV^^^^
channel
(A)
wall
000gQOOj000O0QQ0Oi30O0QQ0O0O00QQ
Inlet oooooôiôaeooôTDiDcToooo^Trcraooeo" outlet
oooooooooooooooooooooooooooooo w
oooooooooooooooooooooooooooooo
wall
(B)
wall
ooooo0o0o0o0o009Pn^P°oo'ooOoooo outlet
Inlet
Hb o o o b o o o o o
OOO O O O Q O O o o o o w
^ _ o o o o o p o OOOOoo
OOOQOO Q O Q O OO o o p o o o o Q O O Q O O OOO
wall
(C)
Figure 3.1 Modeling creeping (Stokes) flow through the hybrid (open-channel porous-medium) region: (A)
A schematic of the velocity profile through the channel and the porous medium, (B) The geometry of the
hybrid region with a periodic porous medium; (C) The geometry of the hybrid region with a non-periodic,
random porous medium. The last two figures show typical REVs (Representative Elementary Volumes)
employed for averaging flow velocities inside the periodic and irregular porous media.
The no-slip boundary condition is applied at the top and bottom walls ( y =
49
h, y = -W) along with the cylinder surfaces. A pressure difference Ap is imposed
on the two vertical boundaries of the inlet and outlet. The resultant pressure-driven
flow experiences the pressure gradient of — = 1— along the flow direction. For
Ax m
the sake of consistency and simplicity, velocity and channel's height are
normalized by the interface velocity and cylinder radius, respectively.
3.2.1 Numerical Solution
The Stokes equation governs the flow of fluid through the clear channel as well as
through the interstitial spaces between cylinders inside the porous-medium. These
governing equations are solved through the finite element method (FEM) based
software, COMSOL, which employs the Galerkin based weak formulation. The
generated numerical result furnishes the point-wise velocity distribution within the
pore space of our computational domain.
Since one has to deal with averaged quantities within a porous medium, the
volume averaging method is applied to average the point-wise velocity distribution
within the porous region. A rectangular representative elementary volume (REV)
with a fixed size is used to average the flow velocities in the domain (see Fig. 3.1).
The averaged velocity is assigned to the centroid of the REV. The size of the REV
is kept fixed while moving it along the y direction. The steps of moving the REV
along the y direction depends on the porosity of the porous medium. As mentioned
earlier, the radius of cylinders remain constant for all porosities while the vertical
and horizontal distances between cylinders change by changing the porosities.

50
Setting each moving step along the y direction equal to a quarter of the vertical
distance between cylinders for the case of periodic media, one can simply
calculate the moving steps for individual porosity, Table 3.1.
Table 3.1 Details of step size for moving REVs corresponding to each porosity.
Porosity (e) 0.5 0.6 0.7 0.8 0.9
Moving step (J) 6.3x10"" 7.0xl0" 6 8.1xl0" 6 9.9X10"6 1.4xl0" 5
Following describes unit cells and moving steps, the imaginary interface
line between the clear channel and porous region is defined at the fifth unit cell
from the bottom of geometry for both periodic and non-periodic cases. In the
periodic case, there is no overlapping of solid cylinders with the imaginary line is
allowed. On the other hand, there is a chance for overlapping of solid-phase and
interface-line for the non-periodic case due to the randomly placed cylinders.
3.3 Open Channel Flow Model for the Theoretical Solution
The domain of the flow problem can be divided into two sections: the clear-
fluid region on top and the homogeneous porous-medium region at the bottom.
The continuity and the Stokes equations governing the steady laminar creeping
flow of a Newtonian fluid through the empty channel are
^ =0 (3.1)
dx
(^2 32 \
a u o u =0 (3.2)
dx
51
where u is velocity, p is pressure and ju is viscosity. The No-slip boundary
condition is used on the top impermeable boundary of the channel. The interface
boundary condition of the clear-fluid region with the porous media region is
discussed below.
3.4 Theoretical Solution of the Porous Medium Flow
In order to solve the interface problem, the Stokes and Brinkman equations
are used to model the steady, one-dimensional creeping flow in the channel and
porous medium, respectively. The interface between the free channel and porous
medium is located at y = 0. The width of the channel is taken to be h. By
considering both of the well-known boundary conditions of stress continuity and
stress jump at the interface, we have the following classical results.
3.5 Stress-Continuity Interfacial Condition
The Brinkman equation can be used to model the flow of fluid through the
porous media when the pore-scale Reynolds number is less than 1 [24,25]. The
superiority of the Brinkman equation over the classical Darcy law is due to its
second-order viscous term which can model the non-linear velocity profile,
especially close to boundaries with gradients [22]. The volume-averaged form of
the continuity and Brinkman equation are described, respectively, as

52
3 f Id2 ~\ [i

•+ H ,2 •£- = 0 (3.4)
3x 3y
K
where is the volume-averaged velocity, * is the pore-averaged pressure,
while P and ^ are the effective viscosity and permeability, respectively. The
permeability K can be either a scalar for isotropic porous medium or a tensor for
anisotropic porous medium [22]. For our case, the permeability corresponds to the
flow perpendicular to cylinders. Though the dependency of P on the porous
medium characteristic and fluid viscosity has been studied by some researchers
[23,26-27], n' = n is assumed for the sake of simplicity.
In this case, the continuity of velocity boundary condition mandates that
uclear\i=porous\i (3.5)
where superscripts clear and porous indicate the regions of clear-fluid and
porous medium, respectively. The normal and shear stress continuity at the
interface is also conducted—the former results in the equality of pressure,
while the latter results in the equality of shear stress that is established in the
present case through the relation
j clear •. , _ _ porous
du . 1a 
~T '«=7—1— '« (3-6)
dy e dy
By considering the stress continuity interfacial condition, the velocity and

53
stress are assumed to be continuous at the interface of the clear-fluid and the
porous-medium domains. The following velocity profiles in the clear fluid and
porous medium regions are obtained for the case of the pressure-driven creeping
flow [28]:
2 \
K_dp
u{y)-- 1+ M y 2j^4=+y for 0<y<h (3.7)
M JK 2ju dx K\MJK K
(u)(y) = {u)i + ({u)D -(u)i j 1 - „ p ^ | Z - ^ L for -oo< v < 0 (3.8)
where the interface velocity, <«>,, is defined as
K d(p)f
2 i^-^+hl
2ju dx M 4K K (3.9)
<«>.=- H' h
1+
M 4K
Defining the boundary-layer thickness Sc as the width of the region inside the
porous medium near the interface where (u) > 1.0l(u) , one may obtain the boundary
layer thickness through the relation («)| _ =1.01(K)D [23] as
50 * l - 2
K (3.10)
H' h
1+
M JK
According to this relation, one can estimate the boundary-layer thickness to be on

54
the order of K05 (which in turn, is on the order of diameter of cylinders used in the
present study, and hence quite small).
There is no boundary layer inside the porous medium modeled with Darcy's law,
while a boundary-layer regime exists inside the porous medium modeled with
Brinkman equation due to the nonlinear term in the Brinkman governing equation.
Thus the two approaches can be correlated.
3.6 Stress-Jump Interfacial Condition
To model flow inside porous medium near the clear-fluid porous-medium
interface more accurately, Ochoa-Tapia and Whitaker [12] used the volume
averaging method to propose the continuity equation and a more accurate form of
the Brinkman governing equation as
d
=0 (3.11)
dx
3 f \l(d2
^=0 (3.12)
dx e1 3y K
where e is the porosity of porous medium. The new equation, to be called
the modified Brinkman equation in the present chapter, is similar to the original
Brinkman equation given in Eqn. (3.4), except for the fact that it dispenses with
the parameter effective viscosity -" which is often used as a fitting parameter in
boundary flows. A comparison of the two equation reveals that ^' = it .

e
The idea that the stress may not be continuous at the interface as the solid
55
may be bearing part of the shear stress transmitted from the channel fluid is an old
one. Ochoa-Tapia and Whitaker used the rigorous volume averaging method to
derive a mathematical relation to describe the stress jump at the clear-fluid porous-
medium interface. For the one-dimensional flow parallel to the interface, the
stress-jump interfacial condition simplifies to
1 d(u) du
e dy ydT "dy ?=o +
=£<->**> (3.13)
where /? is a dimensionless coefficient whose value is on the order of one.
The second boundary condition is the continuity of velocity at the interface,
Eqn. (3.5). Using the stress-jump boundary condition given by Eqn. (3.13) as well
as the continuity of velocity condition, solutions to the Stokes and modified
Brinkman equations, Eqns. (3.2) and (3.12), can be worked out to be
K_dp_ for 0<y<h (3.14)

u{y) = (u) 1 +
^
-p 2ju dx •Je JK K
("}O0 = <"), +(<«>„ -("),) l-exp for - oo < y < o (3.15)
where the interface velocity, (u) , is given as
K d(Py ( 2 h . h2^
2jU dx 4E4K + -K (3.16)
w =•
h
re -p
I+
Once again, using the boundary layer definition, (M)| = l.oi(w) , the
> ' Iy=—§ \ ' D
56
thickness of boundary-layer in the porous region can be found to be
U2 2
h | 2hj3 .
50
<?.= *ln ^KK /KJK / (3.17)
One should note that the above solutions involve volume averaged quantities and,
therefore are applicable to the porous medium only.
It is easy to obtain the solution of the modified Brinkman Equation with the
stress-jump boundary condition by assuming /?=0 (stress continuity) and
substituting e with,«///' in Eqns. (3.14)-(3.17).
3.7 Results and Discussions
In order to compare the theoretically predicted flow variables using various
boundary conditions with the FEM simulated ('experimental') velocity, the
effective viscosity fi' and the adjustable coefficient (3 must be known. The
parameter fx' and /? are computed by adjusting ju' or /? to match the total flow-rate in
the channel calculated from Eqns. (3.7)-(3.8) and Eqns. (3.14), (3.15),
respectively, with that of the FEM results is similar to the experimental procedure
employed by Beavers and Joseph to calculate their slip coefficient [1]. The
5
Beavers-Joseph slip coefficient gets the form a = (uV/O , when the flow-rate
predicted by the Darcy law with the slip-velocity boundary condition at the
interface is equated to the flow-rate predicted by the Brinkman equation in
conjunction with the stress-continuity condition. Values of numerically estimated

57
p' I p., P and a are listed in Table 3.2 based on our calculations. The numerical
results for the periodic case are identical to the values estimated by Hua et al. [21].
Table 3.2 Values of effective viscosity^', coefficient/?, and slip coefficient a at different
porosities
e = 0.5 e = 0.6 e = 0.7 e = 0.8 £ = 0.9
IX'Ip 1.2 1.7 3.2 6.3 14.4
p 0.3 -0.1 -1.0 -1.8 -2.2
a 1.1 1.3 1.8 2.5 3.8
In order to understand the micro-scale flow patterns inside the porous
region near the open-channel porous-medium interface, streamlines for the
pressure-driven channel flow over the periodic porous medium are plotted in the
Fig. 3.2 at four different Reynolds numbers for the porosity £ - 0.6.
Re = 1x10^ Re = lxl0" 5
58
Re slxKT 4 Re = lxlO -3
Figure 3.2 Effect of increase in Reynolds number on the streamline pattern observed adjacent to the open-
channel porous-medium interface for the periodic porous medium. The porosity for the porous medium is
£ = 0.6.
The Reynolds number is calculated based on the appropriate characteristic
length inside the porous region. The cylinder diameter, 2a, is considered as the
characteristic length and the characteristic velocity, the pore-averaged velocity
inside the porous region, is calculated from the Darcy law by employing the
pressure gradient between the two ends of the channel. Density of p = 900kg/m3
and viscosity of // = 220xi0"3/5a.s are assumed to be the fluid properties (close to
the motor oil properties) for entire problem.
The streamlines above the imaginary interface (i.e., the tangent to the upper
layer of cylinders) shows a slight perturbation in the free channel flow due to the
presence of cylinders below. The streamline pattern below the interface is marked
by the presence of identical recirculation patterns between cylinders in the first
row, which is caused by the transmittal of momentum through shear stress from
the open channel to the porous medium. The recirculation pattern remains constant
despite a four orders of magnitude increase in the flow velocity. However,

59
streamlines at a higher Reynolds numbers, such as Re = lxlO~3, are more dense
compared to the lower Reynolds-number plots. It is clear that the pattern of
streamlines do not change with the Reynolds numbers in the Stokes flow regime.
Streamline pattern for a typical irregular configuration is plotted in Fig.
2-45 25 2,55 26 205 27 2,5 2a 295 2 <J 2 95 3 i05 J! 315 32 325
Figure 3 3 Streamline pattern at the interface of clear-channel and porous wall for non-periodic porous
media at e = 0 5 add Re - 1 x 10 ^
One should note that due to the irregularity, no trend is expected. Fig. 3.3
shows that at the interface, the resistance for the flow is higher in the porous
region and hence, more streamline exist in the clear-channel compared to the
porous region.
In Fig. 3.4, the volume averaged velocity inside the periodic porous
medium at the interface is shown for five different porosities at three different
Reynolds numbers (At any porosity, there are three different average velocity
60
profiles corresponding to three different Reynolds numbers.)
0.1
-0.1
-0.3 £=0.6
e = 0.5
-0.5 ©0
-*>-Re=1E-5 -0,7
_—--*£-^ • - v
©0 — Re=1E-5
-*~Re=1E-4
«-Re=1E-4 • •- Ra=1E-3
-0.9
•- Re=1E-3
4.E-03 6.E-03 4.E-03 6.E-03

u/u,'Interlace
(a) (b)
-- — - " 1
0.1 -- - - - -
<f
-0.1
e = 0.7 £ = 0.8
oo -0.3
o o
<x'° oo —»-Re=1E-5
-0.5
•0.7
c S)°'
o o —-Re=1E-5
:
^-Re=1E-4 -»-Re=1E-4
3
• • - Re=1E-3 • • • Re=1E-3
-0.9
-1 1
•05 5.E-03 1.E-02 2.E-02 2.E-02 3.E-02 1.E-05 l.E-02 2.E-02 3.E-02 4 E-02 5.E-02
U/Uintertaco
(c) (d)
0.1
-0.1
-03
£ = 0.9
O o
-0.5
o o
-0.7 -»-Re=1E-5
—>-R8=1E-4
-0.9 • • - Re=1E-3
-1.1
1.E-05 2.E-02 3.E-02 5.E-02 6.E-02 8.E-02
U/UintBr)Hce
(e)
Figure 3.4 Plots of normalized average-velocity inside the periodic porous medium at three different Re
numbers for five different porosities: a) £ = 0.5 , b) £ = 0.6,c) £ = 0.7,d) £ = 0.8 and e)
£ = 0.9.
The x-direction velocity is averaged using the REV shown in Fig. 3.1 B,
61
while the average velocity is normalized with the help of the interface velocity in
Fig. 3.4. We observe that the velocity varies in a sinusoidal manner—this happens
because the thin REV shown in Fig. 3.1 B alternates between row of cylinders and
empty space during its translation along the y axis. (The former leads to a dip in
the velocity while the latter is responsible for the peak.) One can study the effect
of an increase in the Reynolds number on velocity profile in the porous medium:
as expected, the average velocity increases with an increase in the Reynolds
number. However, one can note that the normalized velocity at higher porosity
(such as e = 0.9) is not increased by the increase in Reynolds number. The reason is
that the Darcy velocity is almost similar to the interface velocity at high porosity
values and hence, increase in Reynolds number does not lead to an increase in the
normalized velocity (see Fig. 3.4).
The geometry of an irregular arrangement of the cylinders to model the non-
periodic porous media at porosity 0.7 is presented in Fig. 3.5. The volume-
averaged velocity, obtained using the REV shown in Fig. 3.1C on this weakly-
disordered porous medium, is expected to have smaller fluctuations compared to
the periodic porous medium. This is because arbitrary variations in local velocity
fields caused due to randomness in cylinder locations is expected to damp out the
possible fluctuations in the averaged flow caused due to the presence of 'channels'
between cylinder rows in a periodic porous medium. The horizontal rows of
narrow cells, which constitute REVs corresponding to various y locations within

62
the porous region (Fig.3.5), were employed to volume average the flow velocity
within the porous medium.
&Mdii *MlM •it&i m± >4
• •
• • • • • • • • • • • • • • • • • • • • <
Figure 3.5 An example of a weakly-random arrangement of cylinders employed for the creation of an
irregular porous medium adjacent to the channel. The porosity is 0.7 (i.e., £ = 0.7) for the porous
medium.
The REV independency check was conducted on the irregular case to
ensure that the obtained results are independent of the REV size. This is because a
too-small REV causes fluctuations in the averaged quantity as it moves about
within the porous medium; hence the REV should be sufficiently big so that no
such spurious fluctuations are witnessed within a homogeneous porous medium
[29]. (It is obvious that there is no need tc conduct such an analysis for the
periodic porous media due to its periodic boundary conditions for Stokes flow.)
Using this analysis for the non-periodic porous medium, we plan to optimize the
REV size such that it is rendered suitable for averaging.
Fig. 3.6 shows the sample REV sizes used in this REV-size independence
test. Details of the six REVs used for the purpose of such a test are presented in
Table 3.3.
63
REV 1
,d 3^^tS5CK3Bw«jlî^Ji
^ - Q O O O O'p Q oQ
REV 2
OPOOOP eff£o REV 3
000o0000obo "oonno^/ê,-—^^
•©OQ w —^~^-~
rtOO OOOQ REV 4
Do PopoOo 0 , 9 0 0c ^ 0 "dPOOQOo OOP OOP REV 5
OOOpOQ Q O Q O o o o t o - g g i & o o - o Q O O O O Q O O P
Figure 3.6 Schematic view of six different REVs used for the REV-size independence study (not in the
actual dimensions).
The largest REV called REV 1 contains 34 cells (Note that the cell size
varies by change in porosity, due to change in the distance between solid
cylinders.) The smallest REV is called REV 6 and has 4 cells.
Table 3.3. Details of the six REVs used in the REV-size independence study.
REV Number of cells Area ( m 2 )
REV1 34 5.36e-9
REV 2 28 4.41e-9
REV 3 22 3.47e-9
REV 4 16 2.52e-9
REVS 10 1.58e-9
REV 6 4 6.30e-10
Fig. 3.7 shows the effect of change in REV size on variations in the normalized
volume-averaged x-direction velocity. One can see that REV 1 and REV 2 seem to
achieve an optimum size for the REV as they minimize fluctuation while at the
same time attain convergence in the velocity plot (especially away from the
interface). The REV 6 is the smallest and perhaps the worst REV for averaging
64
because the velocity distribution diverges significantly from the other, larger-REV
solutions. Because of these results, all the upscaling done in the present chapter is
done using REV 2.
Y/a -3
O.E+00 5.E-02 1.E-01 2.E-01 2.E-01 3.E-01 3.E-01

U/Ujnterface
Figure 3.7 Results of the REV size-independence analysis for the irregular porous medium at £ = 0.5 for
the six REVs listed in Table 3.3.
Due to a large velocity gradient at the interface, a flat line can be observed
near the clear-channel porous-medium interface. This reconfirms the fact that
velocities in the free channel and porous region have several orders-of-magnitude
difference and cannot be modeled through the same governing equation.
Mesh independence analysis was also conducted to ensure that our
numerical results are independent of the number of elements and nodes employed
65
in the finite element mesh. The effects of three different meshes, the details of
which are listed in the Table 3.4, were analyzed while estimating the volume
averaged velocity for our geometry. The normalized velocity distribution along the
y direction has been plotted for the three meshes in Fig. 3.8. Results show a
0.000E+00 -r
-2.000E-05
-«-Grid1
— a - Grid 3
-4.000E-05
- ©- Grid 2
Y/a
-6.000E-05
-8.000E-05
-1.000E-04
O.E+00 5.E-12 1.E-11 2.E-11 2.E-11 3.E-11 3.E-11

U/Uj n t e r f a c e
Figure 3.8. Results of the mesh independence study. (Details of the three grids or meshes are
given in Table 3.4.)
negligible dependence on mesh size and one can use the smaller mesh to minimize
computational time while maintaining sufficient accuracy. Repeatability of this
method in the non-periodic porous media has been analyzed by comparing the
results of three different realizations (Fig. 3.9). The pore-averaged velocity is
plotted along the channel including the interface for three randomly generated
porous geometries.
66
Table 3.4 Details of the elements and nodes of three different meshes employed to conduct the
mesh independence study.
Grid Number of elements Number of nodes

1 54177 29779
2 82313 44301
3 3467328 1755930
The pore-scale results show a similar trend for velocity profile for all three
realizations. This comparison indicates that long enough REV in an irregular
arrangement of solid phase for all different realizations damp out local fluctuations
and upscale to yield similar trends.
. —^» — — 1 ~ _ -
E
a.
•«©- Realization 1
Q.
O - • • Realization 2
- A - Realization 3
c
n
0.0E+00 4.0E-11 8.0E-11 1.2E-10 1.6E-10 2.0E-10

Pore-Averaged Velocity (m/s)
Figure 3.9 Repeatability of the solution by comparing three different random realizations' results
Effect of porosity variation on the normalized velocity distribution along the
y direction is compared with the two analytical solutions in Fig. 3.10. All the
computations were done for the non-periodic case at Re = lxl0" 4 . Through Fig.
3.10, one can observe that the higher porosity cases have a higher normalized
67
-2
£=0.5 Re=lxl0 4
-4 • e = 0.6 Re = lxlfJ- 4
—e— Brinkman Eq Stress Continuity Condition
Y/a -6 Bnnkman Eq. Stress Continuity Condition
'v —a— Brinkman Eq Stress Jump Condition -8 Bnnkman Eq Stress Jump Condition
- ©» Regular Arrangement
-10 « Regular Arrangement
~~°- Irregular Arrangement
-12
*- - Irregular Arrangement
0 000 0.001 0.002 0.003 0.004 0.005 0.006 0.007 0.006 0.009 (
-14
U/Uintortace
0.00 0- ,/„ 0.02
B
2
0
-2
e = 0.7 Re = lxl0-<
-4
£ = 0.8 Re = 1x10-"
Y/a
-6
Bnnkman Eq. Stress Continuity Conqit Y/a Bnnkman Eq Stress Continuity Cond tion
-8
Brinkman Eq Stress Jump Condition Qr--a * Bnnkman Eq Stress Jump Condition
Regular Arrangement 10
's-A
-12 «~*y-s Regular Arrangement
Irregular Arrangement 1
ijl-o
-14 — *- — Irregular Arrangement
O.E+00 2.E-03 4.E-03 6.E-03 8.E-03 l.E-(
0.00 0.01 nm oiw nnd not 0.06 0.07 0.08
 f /,„ t e r ( a e e ,/ ,„,.„,„
£= 09
R e = l « 10 •*
Y/a
—e—Brinkman Eq Stress Continuity Condition
—i>- Brinkman Eq Stress Jump Condition
-•**• Regular Arrangement
- o~ Irregular Arrangement
0.05 010 0.15 0.20 0.25 0 30 0.35 0.40 0.45 0.50
, / MM,,
Figure 3.10 Comparison of average velocity using FEM for both regular and random cases with the
Brinkman models at Re= lxlO"4 for A) £ = 0.5 ; B) £ = 0.6 Q £ = 0.7 D) £ = 0.8 and E) £ = 0.9
velocity as well as a smoother transition to interface velocity. As expected, there

68
are smaller fluctuations in velocity for the irregular arrangement of cylinders
compared to the regular one. It is clear that the irregular arrangements of cylinders
helps to damp out the fluctuations in the averaged velocity due to the 'blocking' of
any clear flow path between rows of cylinders.
Similar to Fig. 3.10 which shows normalized velocity versus normalized
height, Fig. 3.11 depicts the actual velocity profile on the logarithmic scale versus
normalized channel height. In this plot, one can see the actual velocity at the
interface of clear channel and porous medium rather than the normalized velocity.
-2
E=0.6
-4 Re=l<104
Y/a
-8 —B—Brinkman Eq. Stress Continuity Condition
—a •Brinknian Eq. Stress Jump Condition

-10
- e -regular modified
-12 - o -Irregular Arrangement
-14
1. E-12 l.E-10 l.E-08 l.E-06 l.E-04 l.E-02 1.E+00
Pore-averaged velocity (m/s)
Fig. 3.11 Actual velocity profile versus dimensionless channel height for £=0.6 and R e = l x l 0 " .
The comparison agrees with the fact that the results of weakly disordered
cylinders are in a better agreement with the Brinkman's model in conjunction with
the stress-jump condition. The main reason for the better, less 'oscillatory'
69
prediction of the irregular arrangement, compared to the regular arrangement, is
the damping out of velocity fluctuations in long REVs due to blocking of gaps
between cylinder rows.
Fig. 3.12 depicts the velocity contour and volume-averaged velocity profile
for the whole geometry for the non-periodic porous media at £ = 0.5 and
Re = lxlCT3. The position of the imaginary interface between clear region and
porous wall is clear in both volume-averaged and contours plots.
65
*I***W
45
'""••i*
*s
,gf 1?
:r* 00
"
v*" ^
O.E+00 5.E-07 l.E-06
A B
Figure 3.12 Velocity contours for the non-periodic porous media at e — 0.5 and Re = lxlO"3: A) velocity
contour and B) volume-avearged velocity
Effect of variation in porosity on volume averaged velocity near the
interface for irregular porous media is plotted in Fig. 3.13.

70
/
Re = lxl0" 4
^ e = 0.5
^ e = 0.6
• * - e = 0.7
-"- e = 0.8
-*>- e = 0.9
1 1
0.04 0.06 0.08 0.10

"'înterface
Figure 3.13 Effect of variation in porosity of the irregular porous medium on the distribution of the volume
averaged velocity near the open-channel porous-medium interface.
As expected, the higher porosity leads to a higher normalized velocity inside the
porous region.
3.8 Summary and Conclusion
Two important interfacial boundary conditions at the interface of free
channel-porous medium are evaluated using a numerical simulation of point-wise
flow based on the finite element method. Two types of porous media modeled by
regular and irregular cylinder arrangement are assumed for the lower porous
channel. While developing the analytical solution, the Stokes and Brinkman
equations are employed for modeling flow through the free channel and the porous
71
medium, respectively. The analytical solutions of the Brinkman equation with the
stress-jump and stress-continuity boundary conditions are presented in details and
compared with the results obtained in this chapter for periodic and non-periodic
cases. Mesh independence and REV independence checks were conducted to
ensure the accuracy of the results. The numerical results of the periodic porous
media at different porosities show that the Darcy velocity gets closer to the
interface velocity for higher-porosity porous media compared to the lower-
porosity one. The numerical results of velocity for non-periodic porous media
were found to be more damped (closer to the Brinkman equation analytical
solutions) compared to the sinusoid velocity profile observed for the periodic
porous media. Increase in porosity of the non-periodic porous media yields better
results in terms of predicting the interface velocity as compared to periodic porous
media.
72
References
[1] Beavers GS, Joseph DD. Boundary condition at a naturally permeable wall. J
Fluid Mech 1967;30:197-207.
[2] Tachie MF, James DF, Currie IG. Velocity measurements of a shear flow
penetrating a porous medium. J Fluid Mech 2003;493:319-43.
[3] Joseph DD, Nield DA, Papanicolaou G. Nonlinear equation governing flow in
a saturated porous medium. Water Res Res 1982;18:1049-52.
[4] Hsu CT, Cheng P. Thermal dispersion in a porous medium. Intl J Heat Mass
Transfer 1990;33:1587-97.
[5] Kladias N, Prasad V. Experimental verification of Darcy-Brinkman-
Forchheimer flow model for natural convection in porous media. J
Thermophysl991;5:560-76.
[6] Vafai K, Kim SJ. Fluid mechanics of the interface region between a porous
medium and a fluid layer-an exact solution. Intl J Heat Fluid Flow 1990; 11:254-6.
[7] Nield DA. The limitations of the Brinkman-Forchheimer equation in modeling
flow in a saturated porous medium and at an interface. Intl J Heat Fluid Flow
1991;12:269-72.
[8] Saffman PG. On the boundary condition at the surface of a porous medium.
Stud Appl Maths 1971;50:93-101.
[9] Taylor GI. A model for the boundary condition of a porous material: part 1. J
Fluid Mech 1971;49:319-26.
[10] Larson RE, Higdon JJL. Microscopic flow near the surface of two-
73
dimensional porous media, part 1: axial flow. J Fluid Mech 1986;166:449-72.
[11] Larson RE, Higdon JJL. Microscopic flow near the surface of two-
dimensional porous media, part 2: transverse flow. J Fluid Mech 1987; 178:119-
36.
[12] Ochoa-Tapia J A, Whitaker S. Momentum transfer at the boundary between a
porous medium and a homogeneous fluid I: theoretical development. Int J Heat
Mass Transfer 1995;38:2635^16.
[13] Ochoa-Tapia J A, Whitaker S. Momentum transfer at the boundary between a
porous medium and a homogeneous fluid II: comparison with experiment. Int J
Heat Mass Transfer 1995;38:2647-55.
[14] Ochoa-Tapia J A, Whitaker S. Momentum jump condition at the boundary
between a porous medium and a homogeneous fluid: inertial effects. J Porous
Media 1998;1:201-17.
[15] Valdes-Parada FJ, Goyeau B, Ochoa-Tapia JA. Jump momentum boundary
condition at a fluid-porous dividing surface: derivation of the closure problem.
Chem Eng Sci 2007;62(15):4025-39.
[16] Valdes-Parada FJ, Goyeau B, Ochoa-Tapia JA. Diffusive mass transfer
between a microporous medium and an homogeneous fluid: jump boundary
conditions. Chem Eng Sci 2006,61(5): 1692-704.
[17] Chandesris M, Jamet D. Boundary conditions at a planar fluid-porous
interface for a poiseuille flow. Int J Heat Mass Transfer 2006;49(13-14):2137-50.
[18] Chandesris M, Jamet D. Boundary conditions at a fluid-porous interface: an a

74
priori estimation of the stress jump coefficients. Int J Heat Mass Transfer
2007;50(17-18):3422-36.
[19] M. Chandesris and D. Jamet, Jump Conditions and Surface-Excess Quantities
at a fluid /Porous Interface: A Multi-scale Approach, Transport in Porous Media,
Volume 78, Number 3, 419-438.
[20] Tomer Duman and Uri Shavit, An Apparent Interface Location as a Tool to
Solve the Porous Interface Flow Problem, Transport in Porous Media, Volume 78,
Number 3, 509-524.
[21] Tan H, Chen X, Pillai KM, Papathanasiou TD. Boundary conditions at the
interface between the clear-fluid and porous medium domains'. In: Ninth
international conference on flow processes in composite materials, Montreal;
2008.
[22] Hua Tan, Krishna M. Pillai, Finite element implementation of stress-jump and
stress-continuity conditions at porous-medium, clear-fluid interface, Computers &
Fluids 38 (2009) 1118-1131.
[23] Sahraoui M, Kaviany M. Slip and no-slip velocity boundary conditions at
interface of porous, plain media. Int J Heat Mass Transfer 1995;35:927^-3.
[24] Reddy JN, Gartling DK. The finite element method in heat transfer and fluid
dynamics. New York: CRC; 2000.
[25] Donea J, Huerta A. Finite element methods for flow problems. England: John
Wiley & Sons Ltd.; 2003.
[26] Martys N, Bentz DP, Garboczi EJ. Computer simulation study of the effective
75
viscosity in Brinkman's equation. Phys Fluids 1994;6:1434-9.
[27] Kaviany M. Principles of heat transfer in porous media. New York: Springer;
1991.
[28] Neale G, Nader W. Practical significance of Brinkman's extension of Darcy's
law: coupled parallel flows within a channel and a bounding porous medium. Can
J Chem Engng 1974;52:475-8.
[29] Bear J., Dynamics of fluid in porous media, Dover Publications, 1988.
76
Chapter 4
PREDICTING THE AVERAGE FLOW
VARIABLES IN POROUS MEDIA USING
ENSEMBLE AVERAGING METHOD
4.1 Introduction
Most of the current studies in the area of porous media are based on using a
suitable representative elementary volume, REV, which is a domain used for
averaging the variables [1]. The problem associated with using REV is that the
size of the REV is needed to be defined for each type of problem. For example,
using a smaller REV for those parts of the problem with larger gradient of changes
while bigger REV for some sections of the computational domain with no sudden
changes. There are no definite standards and specified criteria for the selecting
size of the REV: converged variables tend to oscillate if very small sample is used
while the oscillations begin to dampen out with an increase in the sample size.
Eventually if the sample size is large enough, we begin to get a consistent reading,
and such an REV is called a suitable REV (Fig. 4.1).

77
Domain of
microscopic effects
Inhomogeneous
Averaged
quantity Homogeneous
P,y.(vf),e
REV's volume
Figure 4.1 Variation of averaged quantities with respect to the REV size
Recently, some work has been done for estimating the optimum size of REV
[1-3]. Wang et al. [1] used a new methodology based on estimating the 3-D
hydraulic conductivity tensor for a fractured rock mass to determine the REV size.
They studied the directional hydraulic conductivity of different cubic blocks and
calibrated the hydraulic parameters to quantify the REV size [1]. Nordahl and
Ringrose [2] proposed a criterion based on a normalized standard deviation, to
determine the REV. They showed that the size of the REV depends on both the
property measured and the correlation lengths of the lithological elements. Based
on this, they identified three flow upscaling regimes that each requires a different
method for upscaling. A new insight in to the determination of REV for
permeability estimation at the lithofacies scale and its relation to sedimentological
parameters is given [4].
The REV must be much smaller than the size of the entire domain;
otherwise the resulting averaged description through the whole domain cannot
78
represent what happens at individual points in the side domain. On the other hand,
the size of the REV must be sufficiently larger than the size of a single pore such
that each REV includes a sufficient number of pores to generate a meaningful
average [4]. By assuming ^ t o be flow variable in a computational domain, the
main assumption in using the REV technique is that the values of *P are not
supposed to change significantly in that REV while in porous media problems,

cm m
specially in industrial porous media of smaller dimensions (i.e. from to ),
strong gradients in ¥ renders the use of REVs problematic.
One of the practical examples of such a problem is the boundary layer
formation (Fig. 4.2, [5]) during the convective heat transfer in porous medium
adjacent to a heated vertical flat plate. A thin thermal boundary layer is formed
when the Raleigh number, Ra, takes large values. The governing equations for 2-D
boundary layer [5] can be expressed as
3<«> 8 < v> (4.1)

dx dy
K d= '> Pg/3(<T>-T„)
M dx
d<P'> (4.3)
= 0,
dy
3<T> d<T> d<T> d2<T> (AA\

a + +<v> = am ;— VT-^V
6t dx dy dy
where the subscript °° denotes the reference value at a large distance from the
heated boundary and P denotes the difference between the actual static pressure
and the local hydrostatic pressure.

79
Porouss Medium | T = T,
w
T=T
Figure 4.2 Dimensionless vertical velocity, (a), and temperature profiles plotted against the similarity
variable r|, (b), for natural convection adjacent to a vertical heated surface placed in porous media (Cheng
and Minkowycz, 1977).
The Oberbeck-Boussinesq approximation is assumed in the governing
equations and the Darcy law is valid for this problem. One can see that a large
gradient in the velocity and temperature is present in a very thin layer in the
porous media region adjacent to the heated wall. The question arises as to whether
it is possible to study the velocity and temperature variations in that large-gradient
thin-layer region through the volume averaged equations, Eqns. (4.1) to (4.4). Is
there any other way to avoid the assumption of constant property of a fluid in the
REV while averaging? (Note that the fluid properties are assumed constant within
an REV during averaging, while it is clear that a large temperature gradient in the
boundary layer will induce a sharp gradient in the viscosity as well.

80
Another example of weakness of using REV approach in industrial
problems with larger gradients is modeling the flow of resin in dual-scale porous
media found in the liquid composite molding process, LCM, where the resin flow
happens in thin molds (typically few millimeters) with fiber bundles of the size of
about a millimeter [6-8]. The main question arising in this problem is that are the
average temperature equations valid in such thin geometries where the temperature
values change significantly across the thickness of such a thin mold [6-8]. (Note
that often a temperature difference on the order of 100 °C exist between mold
walls and the incoming resin flowing between the walls.
'cold ^
•hot
Figure 4.3 Liquid composite molding process with hot walls of mold on sides and the cold resin.
In this chapter, a new method to be called the ensemble averaging method
will be introduced for testing the accuracy of averaged equations in the region of
sharp gradients. The main limitation of volume averaging method happens in
regions with large gradients in flow variables such as velocity, temperature, etc.
The situation gets worse when a property depends on another property such as the
fluid viscosity being a function of temperature, i.e., // = ju(T). The overall aim of
this work is to use weakly disordered porous media to test the accuracy of the
averaged equations.
81
In the current chapter, we seek to first use the ensemble averaging method
for the estimation of permeability of our porous medium at different porosities.
Later we use the method to obtain the average-velocity profile at the porous-media
clear-channel flow interface and compare it with the results of previous models
available for such flows. Using this technique, one can easily find if the governing
equations obtained for much larger system with clearly defined REVs are
applicable to regions of porous media with sharp gradients.
4.2 A Conceptual Description of the Ensemble Averaging Method
Using the concept of ensemble averaging available in statistical mechanics
[9], the mean of a quantity that is a function of a porous system's micro-state is
deduced from the ensemble of possible states based on the random distributions on
its solid and liquid phases. A random medium is characterized by an ensemble of
disordered states. Therefore, the mathematical expression of variables varies from
realization to realization. However, the mean of all realizations given for a
physical quantity is independent of the ensemble chosen. In general, the ensemble
average value of some physical quantity such as x at a state / can be defined by
Eqn. (4.5) [9] as
(*(?,)) = i«n—y> (,, a,) (4-5)
A realization in our case is one arrangement of randomly placed cylindrical
fibers to recreate porous medium. Each ensemble average of a flow variable is

82
obtained by volume averaging the flow variables over several realizations. The
phase average of a flow variable ¥ can be described as:
(4.6)
W^J*'^
where V and Vf are the total volume of REV and the pore volume within the
REV, respectively. Subscript/refers to the fluid phase present inside pores. It can
be noted that as a small averaging volume moves around in an REV, the particles
'move' about in a random manner, creating a new realization in the averaging
volume. Thus splitting an REV into N small averaging volumes is perhaps
equivalent to creating N random realizations and then averaging the flow
variables in a small averaging volume within each realization.
*/> *,)
(a) (b)
Figure 4.4 Anticipated variation in (*Pf \ through an increase in a) Volume of REV and b) Number of
realizations
One can rewrite Eqn. (4.6) as

83
v(v,)=\vfdv (4-7)
v
,
The following analysis can be done for two general cases. In case one, each cell
has a specific number of particles or cylinders all of which remain inside the cell.
In case two, assuming a fixed number of particles in each cell, at the most a half of
each cylindrical particle can be located outside the cell borders. By assuming N
cells inside the REV, Eqn. (4.7) can be rewritten as

V (*¥,) = \mfdV+ ft,dV+...+ l^fdV (4.8)
V V V
A f2 f»
such that the REV volume can be split into N cell volumes as
V=V,+V2+... + VN (4.9)
Similarly, the total pore volume within the REV can be split into pore volumes
within the cell as
V
f=VA+Vf>+- + V
fN (4-10)
By expressing Eqn. (4.8) as
v(^ / ) = ^ jyfdv+^- [vfdv+...+^- fydv (4-H)

V V V
\ V, l V,2 N V,N
and then dividing both sides of Eqn. (4.11) by the total volume, we get
\ fl v vlv{ ' v vlvifj ' v vNV{ !
Assuming all cells volumes used for averaging to be equal, one can write
84
V =v = =v =v (4.13)
vl — v2 — ... — v N — v c
Using Eqn. (4.13) in Eqn. (4.12), we get
(4.14)
<*/>-£ J«F,dv+i- ft,dv+...++- ftfdv
Knowing that
(4.15)
V N
and on defining
(4.16)
hlvlv
one can rewrite the Eqn. (4.8) as
(4.17)
Now, instead of assuming each cell to be a part of the REV, we assume each cell
corresponds to the cell used in a realization for averaging, Fig. 4.5.
• • • •
•
• • •""
" • •
• •
• # / A,
,A '•'" •
•
' *'l' -'•.
A Realizations
• •
§u~ 'A
•
A"
r ' > "i •
• •
• %"' Realization 2 |
•
• •
• •
Realization 1
Figure 4.5 Schematic view of random locations of cylinders inside cells in different realizations
85
In other words,
(4.18)
such that
haf) = Cell phase average for ith realization = — f^rfV (4.19)

c
v,,
Fig. 4.4 shows that as V increases, the OPJ is expected to converge to a fixed
value. Similarly, same pattern is likely to happen for (*¥f) when N , the number
of realizations, increases.
We will conduct a similar analysis here to relate the intrinsic phase average
(uff) of a flow quantity \\r to its ensemble average of the cell intrinsic averages
lwf\ / . By introducing the intrinsic phase average of x¥{ as
W-jrJ*/^'/ V,
(4.20)
one can rewrite the above equation as
(4.21)
which can be further changed to
Vfh¥jY = frfdV + jy¥fdV+...+ \VfdV (4.22)
/ vf v. , vf , vf ,
(4.23)
^yfhaf)f =-*- J*, dv+-^ Jyf d v+...+-^ j^ f dv
V v v
f> v,, f2 v [2 fN vfN
Now, we can compare two possible cases that can happen in a unit cell.
86
Case 1:
If a cell has fixed number of cylinders and all cylinders remain inside the cell
completely (Fig 4.6), then it is easy to show that
Figure 4.6 Fixed number of particles totally inside a cell, case 1
V =Vf =... = Vf =Vf (4.24)
and hence
Vf=VA+Vf2+- + V
f„=N-Vfc (4.25)
On using Eqns. (4.24) and (4.25), one can show that
(4.26)
Vf vf Vf V
f N
On further using Eqn. (4.26) in Eqn. (4.23), it is easy to see that
frfdv+±jyfdV+^+J- frfdV
W-£ r
/ . Vf, f V
>H /» v„ (4.27)
where
v/ 1 (4.28)
If we define
87
<+«.,+•-«,„] (4.29)
\ // N
The final form of Eqn. (4.27) reduces to
(N , , ^
*,)'=- (4.30)
N
where
*y.)f - Cell instrinsic phase-average for ith realization= J_ |\p jy (4.31)

/. K,
From Eqn. (4.29), it is clear that an increase in the number of realizations will lead
to the convergence of ( ^ ) to a fixed value.
Case 2:
We now assume that a cell has a fixed number of particles, such that up to a half
of any particle can go outside the cell boundary during the generation of a
realization of our porous medium (Fig. 4.7).
Figure 4.7 A typical distribution of a fixed number of particles inside a cell for Case 2.
88
In such situation, Vf,Vf^,...,Vf may or may not be equal to each other. If the
cells are created by translation of a cell within a REV, then
v
/=^/,+^+- + ^ (4.32)
n • >u * V, «Vf (i.e., Vc <<: v

), we get
f f
On assuming that -
V,
-L~. 1 (4.33)
Vf ~ N
Using Eqn. (4.33) with Eqn. (4.23), it is easy to obtain
[xV,dV + — \*¥,dV +...+— hf.dV (4.34)

*, > / - L —
"/, v h v v
f„ v, v
h h n
after imposing an approximation that
V
/ i ~Vf2 a
-VjN (4.35)
N has to be large enough to obtain the following approximate relation
(4.36)
Unlike the exact relation, Eqn. (4.30), for Case 1, the intrinsic phase average is
only approximately equal to the averaging of the cell intrinsic-phase averages.
This approximation is likely to improve with a increase in the several theoretical
models based on idealized geometries are available in the literature.
4.3 Validation
4.3.1 Permeability Estimation in a Porous Medium

89
The constant of proportionally between the flow-rate and the pressure
gradient in the Darcy law in a porous medium is related to permeability.
Permeability measurement is one of the main concerns in the study of porous
media. Permeability is typically a function of the geometry of the porous medium.
In 1927, Kozeny [10] developed the first model for permeability based on
an idealized porous-medium geometry with tortuous capillaries. Then Carman [11]
used the theory of specific surface area of the medium particles to modify
Kozeny's model. The resulting expression became the famous Kozeny-Carman
equation [12] which expresses permeability as function of porosity and
particle/fiber diameter (More information on some older permeability models can
be obtained from [4, 12]). In recent years, numerous research efforts have been
devoted to the development of a permeability model. Brushke and Advani [13]
proposed a relation for permeability for the case of fluid flow across a periodic
array of cylinders in 1993. Papathanasiou [14] did a computational investigation
on the effect of macro and micro scale porosities on the effective transverse
permeability of square arrays of permeable fiber tows. He described the
permeability as a power law function of the effective intra-tow porosity. In 2006,
Chen and Papathanasiou [15] developed a model for the axial permeability of
fibrous media by considering the underlying microstructure and its variability
numerically. They used unit cells consisting of 600 randomly placed fiber cross-
sections for modeling Stokes flow in a unidirectional fiber array. They considered
90
a large REV to represent a large spectrum of disordered fiber distributions. Monte-
Carlo (MC) procedure, similar to the method used here, was used for generating
an ensemble of non-overlapping disks [16] and Stokes flow through a large
number of unidirectional, disordered fiber arrays simulated.
One of the applications of the proposed ensemble averaging method as
applied to the study of flow and transport in porous media is to estimate
permeability. A simple 1-D flow through a porous medium is being modeled and
solved using this new technique (Fig. 4.8a). The boundary conditions for the
shown rectangular geometry are uniform velocity at the inlet, zero gage-pressure
at the outlet, and symmetry conditions on sides. For modeling this problem via the
ensemble averaging technique, the computational domain is divided into square
unit cells, in which each unit cell contains a solid cylinder inside it (Fig. 4.8b).
Using the ensemble averaging technique, one should solve for different
realizations such that each realization should have a random location for the
cylinder inside any square unit cell. The computational domain is then solved for
velocity and pressure values in each unit cell via the Stokes equation using the
finite element method based commercial software COMSOL [17].
The permeability value for each realization will be estimated through the
Darcy Law, Eqn. (4.40), as applied to the macro-flow geometry shown in Fig.
4.8a. The average of different realization is called the ensemble average of

91
permeability. Applying this technique to problems of flow in thin porous media is
recommended.
Symmetry boundary condition Outlet pressure

Uniform inlet condition
velocity
(a)
0.09"
0.08
0.07
D o0ooo o
O Oy • 0 o
o o o
0.06
0.05
oo o o o o
Y X
0.04 O
0.03
0.02 O o o o o oo
0.01
0
o o oo o o c
X
-0.01 .
0.02 0.04 0.06 0.08 0.1 0.12
(b)
Figure 4.8 (a): The geometry of the problem, (b): Details of a sample realization with square unit cells each
including a solid cylinder.
On applying this method, we anticipate that one can easily converge to
correct results without any concern about the REV size vis-a-vis the domain.
4.3.1.1 Permeability Measurement Using Ensemble Averaging
Method
92
We recall the definitions of the phase average y¥f) and the intrinsic phase
average ( ^ ) for any flow property ^ as
(«P)=i[Wv (4.37)
<*,)'-£#* (4-38)
By assuming an isotropic porous medium in this 1-D flow case, the continuity
equation and the Darcy law can be written, respectively, as
V.(uf) = 0 (4.39)
(S ) = _K V ( P f ) f (4.40)
n
On substituting Eqn. (4.40) into Eqn. (4.39), one can easily show from the Laplace
equation for average pressure and imposed pressure boundary conditions that the
pressure gradient should be linear with respect to x and can be shown to be
* w _<',>;-w: (4.41)
dx
where (Pf) and (Pf) are the intrinsic phase average pressure from two different
unit cells at the same horizontal level at the inlet and outlet, respectively, and L is
93
the horizontal distance of these two unit cells. By substituting Eqn. (4.41) in Eqn.
(4.40), one can easily find the relation for permeability through the Darcy law as
(4 42)
K=-A-r'(«r)i -
fc>;-<Pr>:
u.) is the mean value of the x-component velocity of the phase averaged fluid
' I ave
u
velocity in the porous medium, i.e., /„ \ =\ f/i + \ u '/z where (Uf) and /„ \ are the
averaged velocities in control volumes (cells) close to the inlet and outlet,
respectively. The same governing equation and boundary conditions apply to all
different realizations, Fig. 4.8a. The ensemble average value for permeability can
be obtained using different realizations as
1 ^
TJZ<«
N W
r
f
>
ave.i
Kave = M L < r M ' = =f (4.43)
f 2,i-X
i=l
f>^
where Krealn represents the value of permeability obtained from the n*
realization.
4.3.1.2 Results and Validation
For the sake of validation, the ensemble average result is compared with
several models for permeability including the Brushke-Advani [18], Gebart [19],
Davies [20], Kozeny Karman [10], Rumpf and Gupte [20] (see Table 4.1).
94
The finite element method is also used to calculate the permeability for the
reference case of square arrangement of cylinders using the commercial software
ofCOMSOL4.1.
Table 4.1 Five different permeability models

Model Formula Comment
•, 4
2 2
l2=-(l-£)
K 1(1-/ ) tan-'cVjJ) e n
Bruscke-Advani [18]
r2"3 V (J/
VT^ 7 2+1)
r is the fiber radius
^min = 1 " ^ for square

K. 16 ll-£mn %
Gebart[19]
arrangement
Davies [20] K—
64(1 - £)^{l + 56(1 -ef]
M ,5 = - for sphere
d
3
C £
Kozeny Carman [10] K—
M2(\-£)2
Ms=— for cylinder
d
Dp2 is surface average

p
Rumpf and Gupte[20] K—
5.6
sphere diameter
The square arrangement involves placing circular cross-sections of the fibers
in a square array in the rectangular region of Fig. 4.9b.

95
(a) (b)
Figure 4.9 (a) Two different realizations randomly selected, (b) Square arrangement of periodic fiber mat
array.
Figure 4.10 shows the results of different permeability estimation methods for the
square arrangement of fibers in porous medium. The validation is done for the
range of porosities from 0.4 to 0.9 with 10 realizations. Note that the permeability
is rendered dimensionless using the diameter of cylinders.
•Ensemble Average Method

0.500 Square Array
Gebart
— - - Kozeny-Karman
— - Davies
0.375 — - Rumpf-Gupte
—x— Brushke-Advani
K/D2
0.250
0.125
0.3 0.4 0.5 0.6 0.7 0.8 0.9
Figure 4.10 Comparison of the obtained dimensionless permeability values from ensemble average method
with six other results.
96
There is a perfect agreement between our results and other results for porosities
from 0.4 to 0.6 and then, the ensemble results getting closer to the square-array
numerical results for porosities from 0.6 till 0.8. The ensemble average
permeability for the porosity of 0.9 is almost in the middle of the other methods'
predictions.
In Table 4.2, details of the permeability values for different porosities
through different methods are listed. One can easily notice that the ensemble
results are very close to the square array results for all porosities.
Table 4.2 Details of permeability values at different porosities for each model.
Porosity
Permeability 0.4 0.5 0.6 0.7 0.8 0.9
Ensemble Average 3.37E-07 1.02E-06 2.12E-06 3.73E-06 6.65E-06 1.31E-05
Square array 3.63E-07 9.58E-07 1.96E-06 3.44E-06 6.21E-06 1.15E-05
Gebart [19] 1.54E-07 5.27E-07 1.33E-06 9.97E-07 3.17E-06 1.45E-05
Kozeny Karman [10] 3.62E-07 8.49E-07 8.15E-07 1.32E-06 1.97E-06 1.12E-05
Brushki- Advani [18] 1.60E-07 5.63E-07 1.48E-06 1.17E-06 4.03E-06 1.77E-05
Davies [20] 5.02E-07 9.00E-07 1.76E-06 1.26E-06 4.00E-06 1.55E-05
Rumpf and Gupte [20] 2.26E-07 6.43E-07 1.40E-06 8.33E-07
The error of each method's result based on deviation from the squared array
result is tabulated in Table 4.3. The last column represents the average error of
97
each result through the all porosities. It shows that the ensemble averaging method
has the average error of 8.6% which is quite good in comparison to the other
methods.
Of the permeability models considered, the Davies and the Kozeny Karman
methods give the most accurate results with respect to the squared array result
with the average error of 31.5% and 33.7% respectively.
Table 4.3 Error's for each permeability model with respect to the square array results.
Porosity
Error with respect to
square array 0.4 0.5 0.6 0.7 0.8 0.9 Error Average
Ensemble Average 7.3% 6.5% 8.1% 8.5% 7.1% 14.4% 8.6%
Gebart 57.6% 45.0% 32.0% 71.0% 49.0% 26.4% 46.8%
Kozeny Karman 0.4% 11.4% 58.4% 61.7% 68.3% 2.2% 33.7%
Brushki- Advani 55.9% 41.2% 24.5% 66.0% 35.1% 54.7% 46.2%
Davies 38.1% 6.0% 10.3% 63.5% 35.5% 35.6% 31.5%
Rumpf and Gupte 37.8% 32.8% 28.4% 75.8% 43.7%
4.3.1.3 New Permeability Models
Two new models of permeability can be proposed through the curve fitting of
results obtained from our proposed method in this section. One can estimate the
ensemble-average permeability value using
^V=7.3xl0-'f5740 (4.44)
D)
98
where permeability, K, is a power function of fiber's diameter, Df, and porosity,s.
This formula has a very low standard error (0.0151502) and very good correlation
coefficient (0.9961584) which clearly shows that the curve fitting is very accurate.
Another useful form of the permeability relation is an exponential function of the
form
— = 5.4X1(TV340£ (4.45)
D
)
which has a very low standard error value of 0.0199310 and an accurate
correlation coefficient value of 0.9933420. These two fitted relations were
introduced in this chapter for the first time and one can use these relations for
permeability estimation from the porosity and fiber diameter data.
Figure 4.11 shows the dependence of the ensemble averaging method on the
number of realizations to obtain accurate results. It is obvious that increasing the
number of realization leads to a more accurate result where the ensemble results
are compared with the semi-empirical Kozeny-Carman [10] method.

99
3.66E-07
3.64E-07
<^W>s-^<>>i^>i<>*-<<S*a^-^<^^
3.62E-07
R 3.60E-07
£ • One Realization
a 3.58E-07 • Two Realizations
A Three Realizations
3.56E-07
x Four Realizations
OH
x Five Realizations
3.54E-07
QlSquare Array
3.52E-07
0 1 2 3 4 5 6
Number of realizations
Figure 4.11 Dependence of the ensemble-average permeability value on number of realizations used for
estimating the ensemble average.
4.3.2 Clear-Fluid and Porous Media Interface Problem
Problems of fluid flow at the boundary of the porous medium and clear
channel flow are of the great importance in industrial applications [21]. One can
use the Stokes equation, Eqn. (4.46), and the Darcy's law, Eqn. (4.47) to model
the clear flow region and the flow in the isotropic porous medium, respectively:
-Vp + nV2u = 0 (4.46)
(u f )=-fv( Pf ) f (4.47)
where u and (ij represent the point-wise and volume-averaged velocities

100
while p and (p)f represent the point-wise and pore-averaged pressure,
respectively. K is the isotropic permeability of the porous medium and ju is the
fluid viscosity. Figure 4.12 shows the geometry of the porous media-clear fluid
boundary problem. There is the no-slip boundary condition at the y = h , on the
wall, and the constant pressure in the inlet (x - 0) and zero pressure at the outlet (
x = L). We have assumed both boundary conditions of stress continuity and stress
jump for the interface area.
H
X Imaginary Interface
(a) (b)
Figure 4.12 Creeping flow through the hybrid clear-fluid-porous medium interface: (a) schematic view and
(b) Ideal geometrical model
The slip velocity boundary condition at the interface problem modeled by
Beavers and Joseph [22] as:
±U-£e<»-M.> (4-48)
Here u is the x-direction tangential channel velocity beyond the interface for
y>0, M(0) is the tangential channel velocity at the interface y=0, (u)D is the
101
tangential volume-averaged velocity inside the porous media calculated by the
Darcy law, K is the permeability and a is the slip coefficient. Then, Saffman [23]
estimated the slip coefficient mathematically and showed that the ex depends on
the location of the interfacial boundary. Sahraoui and Kaviany [24] showed the
dependency of the slip coefficient on more factors such as porosity, Reynolds
number, channel size and interface location.
The Brinkman equation, Eq. (4.49), is a partial differential equation with a
second-order viscous term acts as a momentum equation for the flow through the
porous media. Using the Brinkman equation, Eqn. (4.49), for conserving the
momentum governing fluid flow in the porous medium, both the velocity and the
shear stress are often assumed to be continuous at the interface.
-v(py+Mv^-^=o (4.49)
Here ju and ju' represent the fluid viscosity and the effective viscosity,
respectively. By using the Brinkman equation in the hybrid (porous medium- clear
channel) region, it is assumed that stress is continuous at the interface [25-31]. The
solution in terms of velocity distributions in clear channel and porous medium
channel can be described as
K_dp
«(>')={"), i + . l * - - y 2l^Û-y fOr 0 < y < h (4.50)
M JK, 2H dx fi 4K Kj
(14)(y) = ( M ) i + ( ( M ) f l - ( „ > j l - e x ^ - ^ for - oo < y < o

(4.51)
where the interface velocity is defined as

102
K d(p)f
(4.52)
H 2(i dx VM- V K + K 1/ 1 + -\(^ V K
In 1995, Ochoa-Tapia and Whitaker [32] proposed the following equation for
transport of flow through the porous media close to the free channel-porous
medium interface:
-V<„)'+|v»-|(u) = 0 (4.53)
They have found that there is continuity of velocity at the clear-flow porous media
interface while there may not be such continuity for the stress. They used the
rigorous volume average method, to derive an expression for stress jump at this
interface. Using the condition of (u)\ =1.0l(w) , the boundary layer thickness
can be given by
S.=.
£ ta V ^ (4.54)
£ . . h ( I
l+
7K{Te-ft
where /? is the stress continuity coefficient (i.e, for /? = 0, stress is continuous at
the interface). Ochoa-Tapia and Whitaker introduced the stress-jump boundary
condition at y = 0 for the 1-D parallel flow case as
ld(u) du P
€ dy j=0"
"dy
y=0*
(4.55)
Knowing the ft to be of the order of one and using the Eqn. (4.55), the velocity
profiles for the Stokes and modified Brinkman governing equations can be
103
derived:
K^dp
u{y) = (u) 1 +
4s )4K 2/n dx K4e JK K (4.56)
{u){y) = {u)i + (( M ) D -( M )î-ex^-^]

(4.57)
The interface velocity is defined as
K d{p)' [ 2 h h2
2// dx \-[e 4K K (4.58)
( « > , = •
Many researchers used the stress-jump boundary condition which gives good
results [33-39]. Tan and Pillai [40] recently implemented a finite element solution
for modeling flows near the porous medium-clear fluid interface.
4.3.2.1 Results and Validation
Results of the Ensemble averaging method is compared with the solution of
the Brinkman equation with both the stress-continuity interfacial condition [41]
and the stress jump interfacial condition [32, 42-43]. The averaging of the flow
variables (i.e., velocity) were done through rows of unit cells in a same height. The
ensemble average of several realizations gives the ensemble averaging result for
geometry studied. These comparisons have been done for two different Reynolds
numbers of Re =0.01 and Re =0.001 , Fig. 4.13.

104
Y[m]
0.12 0.14
-*- Ensemble Average
-•- Brinkman_Slress Continuity
-*- Brinkman_Stress Jump
Velocity [m/s]
(a)
Re = 0.001
-*- Ensemble Average
-*- Brinkman_Stress Continuity, Re=0.001
-*- Brinkman_Stress Jump, Re=0.001
Velocity [m/s]
(b)
Figure 4.13 Velocity distribution in the clear-fluid porous medium interface: (a) Re = 0.01 and (b)
Re = 0.001
The results of the velocity profile along the thickness of the hybrid channels
for the cases of Re = 0.001 and Re = 0.01 are in a good agreement with the solution
of the Brinkman equations [32, 41-42]. Fig. 4.14 shows the dependence of the
105
ensemble averaging method on the number of realizations to obtain reasonable,
converged results.
velocity [m/t]
(a)
velocity [m/s]
(b)
106
•0 05
0 Oi *00
\)l
l»
III
I II
—•— Ensemble Average
4*<H»
i ii --B--1 Realization
ii
- -A- • 2 Realizations
-•K--3 Realizations
T T
I I - - * - - 4 Realizations
I I
I I
I I - -»•- • 5 Realizations
I I
I III —&— 6 Realizations
I
I
- -o- • 7 Realizations
I II
i n - — • 8 Realizations
i i
i i hi
i «i
velocity [m/s]
(C)
Figure 4.14 a) Dependency of the Ensemble average method on number of realizations; b) Part A;
c)PartB.
It can be observed that having only one or two realizations can cause
unrealistic fluctuations inside the porous medium. As the number of realizations
increases, the Ensemble average results become more accurate. It can be seen that
the solution for this problem was obtained accurately with maximum 8 realizations
after attaining a suitably converged solution.
4.4 Conclusion
In this study, a new averaging technique through the concept of ensemble
averaging is introduced and applied to study of two problems of flow in porous
media. The important benefit of this method is that the weakness of REV's size
dependence in the volume averaging method is omitted and it yields an approach

107
to evaluate the effectiveness of governing equations established using the volume
averaging method in regions with sharp gradients of a parameter. In the first
application, ten realizations were studied and averaged to estimate the
permeability of a rectangular domain filled with porous medium for different
porosities. The ensemble average results validated by comparison with five
different models for permeability, show good agreement. In the second
application, the ensemble averaging technique is used to model the boundary layer
of porous medium-clear fluid interface with two different boundary conditions as
an example of a problem with a large gradient of variables in a small domain. The
results are compared with the available results in the literature and show a
reasonable agreement. Such reasonable agreements in both applications show the
strength of the proposed ensemble average method as a powerful mathematical
tool to evaluate the effectiveness of the conventional volume-averaging based
governing equations in porous media with sharp gradients.

108
References
1. Wang M., Kulatilake P.S.W., UM J., Narvaz J., Estimation of REV size and
three-dimensional hydraulic conductivity tensor for a fractured rock mass
through a single well packer test and discrete fracture fluid flow modeling,
International journal of rock mechanics and mining sciences,
2002, vol. 39, no7, pp. 887-904.
2. Kjetil Nordahl and Philip S. Ringrose, Identifying the Representative
Elementary Volume for Permeability in Heterolithic Deposits Using
Numerical Rock Models, Mathematical Geosciences, Volume 40, Number
7/October, 2008.
3. Torquato S., Random heterogeneous materials. New York: Springer; 2001.
4. Jacob Bear, Dynamics of fluids in porous media, Dover publications, INC.,
1972.
5. Donald A. Nield, Adrian Bejan, Convection in porous media, Springer; 3rd
edition.
6. Ratle F., Achim V., Trochu F., Evolutionary operators for optimal gate
location in liquid composite moulding, Applied soft computing, Vol.9,
Issue 2, pp:817-823.
7. Luthy T., Ermanni P., Flow monitoring in liquid composite molding in
linear direct current sensing technique, Polymer composites, Vol.24, Issue
2, pp:249-262.
109
8. Umer R., Bickerton S., Fernyhough A., Modelling liquid composite
moulding preocesses employing wood fiber mat reinforcements, Key
engineering materials, Vol.334-335, pp:113-l 16.
9. Norman Davidson, Statistical mechanics, McGraw-Hill book company,
INC. 1962.
10. Kozeny, J., Uber kapillar leitung das wassers im bpden, Sitzungsberichte
wiener akademie der wissenschaft, Abt Ha 136, 271-276 (1927).
11. Carman, P. C , Fluid flow through granular beds, Trans Inst. Chem. Eng.
15,150-166(1937).
12. Scheidegger, A. E., Physics of flow through porous media,3rd ed.,
University of Toronto press, Toronto, 1974.
13.M.V. Brushke and S. G. Advani., Flow of generalized Newtonian fluids
across a periodic array of cylinders, J. Rheo., 37 (3) 1993.
14. T.D. Papathanasiou, On the effective permeability of square arrays of
permeable fiber tows, Int. J. Multiphase Flow, Vol 23, No. 1, pp.81-92,
1997.
15.Xiaoming Chen and T.D. Papathanasiou, Micro-scale modeling of axial
flow through unidirectional disordered fiber arrays, Composites science and
technology, 67 pp. 1286-1293 (2007).
16.Torquato S., Random heterogeneous materials. New York: Springer; 2001.
17. Www.comsol.com
110
18.M. V. Brushki and S.G. Advani, Flow of generalized Newtonian fluids
across a periodic array of cylinders, J. Rheol. 37 (3), 479-498 (1993).
19. B. R. Gebart, Permeability of unidirectional reinforcements for RTM,
Journal of composite material, Vol. 26, No. 8 (1992) 1100-1133.
20. Porous Media Fluid transport and pore structure, F. A. 1. Dullien, Academic
press, INC., 1991.
21.D.D. Joseph and L. N. Tao, Lubrication of a porous bearing-Stokes'
solution, J. Appl., 753-760 (1966).
22.G.S. Beavers and D.D. Joseph, Boundary conditions at naturally permeable
wall, J. Fluid Mech., 30, 197 (1967).
23.P.G. Saffman, on the boundary condition on the surface of a porous
medium, Stud. Appl. Maths, 50, 93 (1971).
24. M. Sahraoui and M. Kaviany, Slip and No-Slip Velocity Boundary
Conditions at Interface of Porous, Plain Media, Int. J. Heat Mass Transfer.
35, 927 (1992).
25. Salinger AG, Aris R, Derby JJ. Finite element formulation for large-scale
coupled flows in adjacent porous and open fluid domains. Int J Numer
Meth Fluids 1994; 18: 1185-209
26. Gartling DK, Hilckox CE, Givler RC. Simulation of coupled viscous and
porous flow problems. Comput Fluid Dyn 1996: 7:23-48.

Ill
27.Neale G, Nader W. Practical significance of Brinkma's extension of
Darcy's law: Coupled parallel flows within a channel and bounding porous
medium. Can J Chem Engng 1974;52:475-8.
28.Martys N, Bentz DP, Garboczi EJ. Computer simulation study of the
effective viscosity in Brinkman's equation. Phy Fluids 1994;6:1434-9.
29. Sahraoui M, Kaviany M. Slip and no-slip velocity boundary conditions at
interface of porous, plane media. Int J Heat Mass Transfer 1995;35:927-43.
30. James DF, Davis AM. Flow at the interface of a model fibrous porous
medium. J Fluids Mech 2001;426:47-72
31. Costa VAF, Oliveira LA, Baliga BR, Sousa ACM. Simulation of coupled
flows in adjacent porous and open domain using a control-volume finite -
element method. Numer Heat Transfer, Part A 2004;45:675-97.
32. Ochoa-Tapia JA, Whitaker S, Momentum transfer at a boundary between a
porous media and homogenous fluid I: theoretical development. Int J Heat
Mass Transfer 1995;38:2635-46
33. Valdes-Parada FJ, Goyeau B, Ochoa-Tapia JA. Jump momentum boundary
condition at a fluid-porous dividing surface derivation of the closure
problem. Chem Eng Sci 2007; 62(15):4025-39.
34. Valdes-Parada FJ, Goyeau B, Ochoa-Tapia JA. Diffusive mass transfer
between a microporous medium and an homogenous fluid: jamp boundary
conditions. Chem Eng Sci 2006; 61 (5): 1692:704.

112
35.Kuznetsov AV. Influence of the stress jump condition at the porous-
medium/clear fluid interface on a flow at a porous wall. Int. Commun.
36. Kuznetsov AV. Analytical investigation of Coette flow in a composite
channel partially filled whith a porous medium and partially filled with a
clear fluid. Int. J Heat Mass Transfer 1998;41:2556-60
37. Chandesris M, Jamet D. Boundary conditions at a planar fluid-porous
interface for a poiseuille flow. Int. J Heat Mass Transfer 2006;49(13-
14):2137-50.
38. Chandesris M, Jamet D. Boundary conditions at a fluid-porous interface: an
a prior estimation of the stress jump coefficients. Int. J Heat Mass Transfer
2007;50(17-18):3422-36.
39. Tan H, Chen X, Pillai KM, Papathanasiou TD. Boundary conditions at the
interface between the clear-fluid and porous medium domains. In: Ninth
international conference on flow processes in composite materials,
Montreal; 2008.
40. Hua Tan, Krishna M Pillai, Finite element implementation of stress-jump
and stress-continuity conditions at porous medium, clear-fluid interface,
Computers and Fluids, Volume 38, Issue 6, June 2009, Pages 1118-1131.
41. G. Neale and W. Nader, Practical significance of Brinkman's extension of
Darcy's law: coupled parallel flows within a channel and a bounding
porous medium, Can J. Chem. Eng., 52, 475-478 (1974).

113
42. Ochoa-Tapia J A, Whitaker S, Momentum transfer at a boundary between a
porous media and homogenous fluid II: Comparison with experiment. Int J
43. Ochoa-Tapia J A, Whitaker S, Momentum jump condition at the boundary a
porous media and a homogenous fluid: inertial effects. J Porous Media
1998; 1:201-17.
114
Chapter 5
NON-ISOTHERMAL MODELING OF POROUS
MEDIA FLOW
5.1 Introduction
One of the most important manufacturing processes of polymer composites
is the liquid composite molding (LCM), which includes several other technologies
such as: resin transfer molding (RTM), vacuum assisted resin transfer molding
(VARTM) and Seeman composite resin infusion molding (SCRIM) [1]. In such
technologies, the composites are created by impregnation of a preform with resin
injected into the mold's inlet. Some thermoset resins may undergo the cross-
linking polymerization, called curing reaction, during and after the mold-filling
stage. Thus, the heat transfer effect due to exothermal polymerization reaction of
resin may not be neglected in the mold-filling modeling of LCM. One should keep
in mind that the viscosity of the resin depends on the temperature as well as the
degree of cure. Therefore, the physics of the flow is coupled with the curing of
resin. This shows the importance of heat transfer and curing governing equations
in the non-isothermal flow in both single-scale and dual-scale porous media.

115
Generally, the energy balance equations can be derived using two different
approaches: (1) two-phase or thermal non-equilibrium model [2-5] and (2) local
thermal equilibrium model [6-18]. There are two different energy balance
equations for resin and fiber separately in the two-phase model, and the heat
transfer between these two equations occur via the heat transfer coefficient. In the
thermal equilibrium model, we assume that the resin and fiber reach local
thermodynamic equilibrium once the fiber is impregnated. Therefore, only one
energy equation is needed as the thermal governing equation [3,5]. Firstly, we
consider the heat transfer governing equation for the simple situation of isotropic
porous media. Assume that radioactive effects, viscous dissipation, and the work
done by pressure are negligible. We do further simplification by assuming the
thermal local equilibrium that Ts =Tf =T where Tsand rfare the solid and
fluid phase temperature, respectively. A further assumption is that there is a
parallel conduction heat transfer taking place in solid and fluid phases.
Taking the average over a representative elementary volume (REV) of the porous
medium, we have the following governing equation for the solid and fluid phases
(l-(p)(pc)sd<ls> =(\-<p)V.(ksV<Ts >°) + {\-<p)q: (5.1)

at
< f>
(P{pcP)f d( +(pcP)fv.V<Tf >f = (pV.(kfV<Tf >f) + (pqmf (5.2)
where c is the specific heat of the solid and cpis the specific heat at constant
pressure of the fluid, kis the thermal conductivity coefficient and q"is the heat
116
production per unit volume. Volume averaging of any arbitrary function OF) is
shown by <*F> operator and it can be taken either over fluid phase or solid phase
using <*F>f and <¥> s notations, respectively. By assuming the thermal local
equilibrium, setting Ti =Tf = T, one can add Eqns. (5.1) and (5.2) to obtain a single
equation for the volume averaged temperature as
(pc)m ^ p + (pc)f v.V <T >= V.(kmV <T» + q"m (5.3)
where (pc)m, km and q"m are the overall heat capacity per unit volume, the overall
thermal conductivity, and the overall heat conduction per unit volume of the
porous medium, respectively. They are defined as follows:
{pc)m = (\-<p){pc)s+(p{pcp)f (5.4)
km=<\-<p)k,+<pkf (5.5)
q><\-<P)q:+9qmf (5-6)
5.2 Proposed Work
In this chapter, our aim is to focus on the thermal dispersion term arising in
the governing equation for non-isothermal flow through porous media. The
thermal dispersion term exists as a result of both the micro-structure of the porous
material as well as the heat convection effect. A combined experimental/numerical
approach is introduced for the estimation of the dispersion tensor in the heat-
transfer governing equation of non-isothermal flow through porous media. The
experimental setup and the finite element method (FEM) procedure will be
117
defined in the following sections. Using the experiment data in numerical
simulation leads to an accurate estimation of dispersion tensor in the single-scale
and dual-scale porous media. This is first such attempt to explore thermal
dispersion term through a combined experimental/numerical method, especially in
dual-scale porous media characterized by a complex micro-structure.
5.3 Governing Equations of Non-Isothermal Flow in Porous Media
5.3.1 Macroscopic Level
Pillai and Munagavalasa [19] used a volume averaging method with the
local thermal equilibrium assumption to derive a set of energy and species
equations for dual-scale porous medium. The schematic view of the initial
impregnation of such a medium by a liquid is presented in the Fig. 5.1. Unlike the
single scale porous media, there is an unsaturated region behind the moving flow-
front in dual-scale porous media. The reason for such partial saturation can be
mentioned as the flow resistance difference between the gap and the tows where
the flow goes faster in the gaps rather than the wicking into the tows. Pillai and
Munagavalasa [19] applied the volume averaging method to the dual-scale porous
media. Using woven fiber mat in the LCM, they considered the fiber tows and
surrounding gaps as the two different phases.

118
Resin front
/
Sink effect ^ Gas bubble
Figure 5.1 Schematic view of the initial impregnation of a dual-scale porous-medium by

a liquid [19]
The point-wise microscopic energy balance and species equations for resin
inside the gaps is proposed at first, and then the volume average of these equation
is taken. This way, they developed the macroscopic energy and species balance
equations.
The macroscopic energy balance equation in dual-scale porous medium is given
by
£
P.C
g^p.i *Tdtt(T*Y+ (v*>-v(r*Y = V K *- v ( r *>' + e.p.H*f<+ Q«»-Q-* (5 7)
-
where the pg and Cp are the resin density and specific heat respectively. Tg is
the temperature of resin in the gap region, e g is gap fraction, HR is the heat
reaction and fc is the reaction rate. The egpgHRfc term represents the heat source
term caused due to the exothermic curing reaction of a thermosetting resin. The
term Kth is the thermal conductivity tensor for dual-scale porous medium defined
119
as
K =k e 5+
* g s v In^dA-£sv&I^SdV (5 8)
-
A gl g
where k g , 8 and vg are the thermal conductivity of the resin, a unit tensor, and
the fluctuations in the gap velocity with respect to the gap averaged velocity,
respectively. The vector b relates temperature deviations in the gap region to the
gradient of gap-averaged temperature in a closure. Considering the temperature
closure formulation as fg = b.V < Tg >s , the local temperature deviation is related
to the gradient of the gap-averaged temperature through the vector b [19].
Qconvin the Eqn. (5.8) is the heat source term due to release of resin heat prior to
the absorption of surrounding tows given by
Qconv=PgCP,gSg[(Tg)8-(Tg)8t] (5.9)
where Sg , the sink term and areal average of temperature on the tow-gap
interface, <Tg>gt, are expressed as
(5.10)
and
W-7-JT.dA
g
' (5.11)
Qcond is the heat sink term caused by conductive heat loss to the tows and is given
by
120
Qoo„c=^jk g (-VT 8 ).n gt dA (5.12)
Using the analogy between heat and mass transfer to derive the gap-averaged cure
governing equation following the Tucker and Dessenberger [6] approach, one can
derive the following equation for the pore-averaged resin cure:
e
* f (c«}' +
( v
g )- V
( c
8 ) S
= V
- D
- V
( 0 E + e
s f c +Mco„v - M d l f f (5.13)
8
Here is the degree of cure in a resin for which the value of 0 and 1 correspond
to the uncured and fully cured resin respectively, D is the diffusivity tensor for the
gap flows and is given by
D = D ieg 5 + ^ | n g t b d A - ^ - J v g b d V (5.14)
where D,is the molecular diffusivity of resin. In Eqn. (5.13), Mconv the convective
source created due to the release of resin cure when absorbing into tows as a
results of the sink effect, is given by
Mconv=Sg[(cg)8-(cg)8t] (5.15)
where /c g ) 8 , the areal average of temperature on the tow-gap interface, is
expressed as
gt
c-r=-J-Jc,dA
*/ ~ T ~ J Vg "~ (5 16)
-
Mdiff, the cure sink term created as a result of the diffusion of cured resin into
121
tows, is given by
M diff =^|D 1 (-Vc g ).n gt dA (5.17)
It should be noted that the only way to compute theQconv,Qcond, Mconvand Mdiffis
by solving for flow and transport inside the tows.
5.3.2 Microscopic Level
Phelan et. al. [20] showed that the conventional volume averaging method can be
directly used to derive the transport equation for thermo-chemical phenomena
inside the tows for single-scale porous media. The final derivation for microscopic
energy equation is
[e.fccj, +(l-e t XpC P ) f ]-^- + (pCP)Ivt.VTt = V.KtM.VTt +etp,HRfc (5.18)
where the subscript t refer to tows. The microscopic species equation is given by
e t - ^ - + vt.Vct =V.e,Dt.Vc, +etfc (5.19)

dt
The complete set of microscopic and macroscopic energy and species equations as
well as the flow equation should be solved to model the unsaturated flow in a
dual-scale porous medium.

5.4 Dispersion Tensor
In some cases, a further complication arises in the thermal governing
equation due to the presence of thermal dispersion [21] which happens due to the
spreading of heat at the pore scale. Such spreading is mainly due to the molecular
122
diffusion of heat as well as due to the hydrodynamic mixing caused by the random
motion of fluid in a porous medium. Greenkorn [22] mentioned the following nine
mechanisms for most of the thermal spreading at pore scale:
1. Molecular diffusion: important in the case of sufficiently long time scales.
2. Mixing due to obstructions: The flow channels in porous medium are
tortuous means that fluid elements starting a given distance from each other
and proceeding at the same velocity will not remain the same distance apart
after a certain time, Fig. 5.2.
3. Existence of autocorrelation in flow paths: Knowing all pores in the porous
medium are not accessible to the fluid after it has entered a particular fluid
path.
4. Recirculation due to local regions of reduced pressure: The conversion of
pressure energy into kinetic energy gives a local region of low pressure.
5. Macroscopic or megascopic dispersion: Due to non-idealities which change
gross streamlines.
6. Hydrodynamic dispersion: Macroscopic dispersion is produced in capillary
even in the absence of molecular diffusion because of the velocity profile
produced by the adhering of the fluid wall.
7. Eddies: Turbulent flow in the individual flow channels cause the mixing as
a result of eddy migration.
8. Dead-end pores: Dean-end pore volumes cause mixing in unsteady flow.
The main reason is that as the solute rich front passes the pore, diffusion
123
into the pore occurs due to molecular diffusion. After the front passes, the
solute will diffuse back out and thus, disperse.
9. Adsorption: It is an unsteady-state phenomenon where a concentration front
will deposit or remove material and therefore tends to flatten concentration
profiles.
\ —
I
)
II
I
< 1
J
Figure 5.2 Mixing as a result of obstruction
Generally, the first seven mechanisms are included in the dispersion coefficient.
Rubin [23] generalized the thermal governing equation as
(pel ~ + (pc)f v.VT = V.(k,VT) + C (5.20)
where k m is a second-order tensor called the dispersion tensor.
Two dispersion phenomena that have been extensively studied in as part of the
transport phenomena in porous media, are the mass and thermal dispersions. The
former involves the mass of a solute transported in a porous medium, while the
latter involves the thermal energy transported in the porous medium. Due to the
124
similarity of mass and thermal dispersions, they can be described using common
dimensionless transport equation as
a
< ^ •/+„ \ a < " > _ 1 3 f D, 3(ft)
vW
(5.21)
30 " 3X,. Pe3X, ° », ,
where (£2) is either the volume averaged concentration for mass dispersion or the
volume averaged dimensionless temperature for thermal dispersion, 0 is
dimensionless time, (£/,•) is averaged velocity vector, Pe is Peclet number, Dy is
dispersion tensor of 2n order. It should be noted that Pe = — for mass dispersion
andPe-— for thermal dispersion where u and L are characteristic velocity and
a
length, respectively. D and a are molecular mass and thermal diffusivities,
respectively.
5.4.1 Dispersion in Isotropic Porous Media
Most studies on the dispersion tensor so far have been focused on the isotropic
porous media. Nikolaveskii [24] obtained the form of dispersion tensor for
isotropic porous media by analogy to the statistical theory of turbulence. Bear [25]
obtained a similar result for the form of the dispersion tensor on the basis of
geometrical arguments about the motion of marked particles through a porous
medium. Bear studied the relationship between the dispersive property of the
porous media as defined by a constant of dispersion, the displacement due to a
uniform field of flow, and the resulting distribution. He used a point injection
125
subjected to a sequence of movements. The volume averaged concentration of the
injected tracer, C0, around a point which is displaced a distance L = utin the
direction of the uniform, isotropic, two dimensional field of flow from its original
position is considered in his research.
C(x,y;x 0 ,y 0 ) = m n (5.22)
27i7tx oy | 2CTV 2a,
where L is the distance of mean displacement, u is the uniform velocity of flow, t
is the time of flow, ax and cy are standard deviations of the distribution in the x
and y directions, respectively and, finally m and n are the coordinates of the point
(x,y) in the coordinate system centered at (£,r|) given by m = x - ( x 0 + L ) and
n = y - y 0 , Fig. 5.3. This figure shows a point injection as a result of subsequence
movement where initially circle tracer gets an elliptic shape at L = ut.
m H
(><.y)
fi— y
1—i—Ts— * ' ' 7~ ', 7 " "/
Figure 5.3 Dispersion of a point injection displaced a distance L
Standard deviations are defined by ax =(2DIL)05 and a =(2DnL)05 where D, and

126
D„ are the longitudinal and transverse constants of dispersion in porous media,
respectively. One should note that the D! and D„ used in the Bear work depend
only upon properties of the porous medium such as porosity, grain size,
uniformity, and shape of grains1. From Eqn. (5.22), it follows that, after a uniform
flow period, lines of the similar concentration resulting from the circular point
injection of the tracer take the ellipse shape centered at the displaces mean point
and oriented with their major axes in the direction of the flow:
" 4 + ^ =1 (5.23)
Bear conjectured that the property which is defined by the constant of dispersion,
DiJkl, depends only upon the characteristics of porous medium and the geometry of
its pore-channel system. In a general case, this is a fourth rank tensor which
contains 81 components. These characteristics are expressed by the longitudinal
and lateral constants of dispersion of the porous media. Scheidegger [26] used the
dispersion tensor Dy in the following form
Dâ^Yr (5-24)
M
where v is the average velocity vector, vk is the kth component of velocity vector,
aijhn is a fourth rank tensor called geometrical dispersivity tensor of the porous
medium. Bear demonstrated how the dispersion tensor relates to the two constants
He also considered a case when at the end of period t , , the velocity, remaining horizontal, is suddenly
changed from u , to u r
127
for an isotropic medium: «,, = longitudinal dispersion , and a±= transversal
dispersion2. Scheidegger [26] has shown that there are two symmetry properties
for dispersivity tensor
a a a n d a
Vbn= j*m ,ihn=av* (5'25)
Therefore, only 36 of 81 components of fourth rank tensor al]kl are independent.
For an isotropic porous medium, the dispersivity tensor must be isotropic. An
isotropic fourth rank tensor can be expressed as
««*» = *3,<L + P5Jim + Y5J3k (5.26)
where a, (3, y are constants and 51} is Kronecker symbol. Because of symmetry
properties expressed by Eqn.(5.23), we get
P= Y (5.27)
So the dispersivity tensor can be written as
at =aSS, +B(5,S +5 8A (5.28)

v f
ijkm tj km r* \ tk jm tm jk J
On substituting Eqn. (5.28) into Eqn. (5.24), we can obtain the dispersion tensor as
If we define a,_=alvl, all-a1=2yfflvl and n, =v,/lvl (n, is the mean flow direction), then
dispersion tensor Dl} can be written as
D
,j=aAj+{an-ai)n,nj ( 5 - 3 °)
2
The longitudinal direction is along the mean flow velocity in porous media, whereas the transverse
direction is perpendicular to the mean flow velocity.
128
From Eqn. (5.28), it is quite clear that the three principle directions of dispersion
tensor D are orthogonal to each other (due to the symmetry of Di}), and one
principle direction is along the mean flow direction (n) and the other two are
perpendicular to the mean flow direction. Therefore, for isotropic medium, the
dispersion tensor can be expressed by longitudinal and transverse dispersion
coefficients. If we consider the mean flow is along x-axis, Dtj can be written as
<h 0 0
D = 0 a± 0 (5.31)
0 0 a,
Therefore, transport equation can be written as
3(a) . ,3(n)_ I [ a2(Q) d2(a) 1 d2(a.)

flu — L y L + ax ——r - + ax (5.32)
de ^ xl ax, pe oXl oX2 oX 3 j
5.4.2 Dispersion in Anisotropic Porous Media
It has been shown that one of the principle axes of the dispersion tensor in
isotropic porous medium is along the mean flow direction. Unlike the isotropic
media, there are nine independent components in the dispersion tensor for the case
of anisotropic porous media. Bear [25] noted that the dispersion problem in a non-
isotropic material still remains unsolved. He suggested to distinguishing between
various kinds of anisotropics and doing statistical analysis with different frequency
functions for the spatial distribution of channels in each case. Unless some specific
types of porous media, such as the axisymmetric or transversely isotropic porous
media are considered, it is not possible to simplify the form of dispersion tensor.
129
In 1965, Poreh [27] used the theory of invariants to develop a dispersion tensor for
the axisymmetric porous media. The average properties of an axisymmetric porous
medium which affect the macroscopic dispersion pattern are invariants to rotation
about given line. He establish the general form of D^ with two arbitrary vectors R
and S as
DIJRlSJ = BlSIJRlSJ + B2vIvJRlSJ+B3AiAJRlSJ + B4vlRlAJSJ + B5ZlR,vJSJ (5.33)
where X. is the axis of symmetry, Bu B 2 , B3, B 4 , and B 5 are arbitrary functions of v2
and vklk. For arbitrary R and S and symmetric DtJ, one can have
DtJ = V „ + B2vlVj + B^k, + B4 (v,^ + Xy,) (5.34)
The dimensionless form of Eqn. (5.32) is obtained as
f?\ ' I A
7 = GA+G2 v,vy+G3^+G4 (VAJ+^VJ) (5.35)
D, \D0J \P*J
where G\, G2, G3 and G4 are dimensionless functions of (vl/D0) , (vl/v) . D0 is the
molecular diffusivity coefficient, / is the length characterizing the size of pores,
and v is the kinematic viscosity. The Final form of the dispersion tensor for
axisymmetric porous media can be given as
(*\ ^
Dn A+A2 D 2
,
*„+A V,Vj + 44+A ^ M,+4v,) (5.36)
v°oy V u
o J vA,v
where p, and p4 are dimensionless numbers, p 2 , P3 and P5 are even functions of
cos 0), and P6 is an odd function of cos GO
By assuming no motion within an axisymmetric medium, D^ is simplify to

130
^ =A3,+M^ (5-37)
Eqn.(5.37) indicating that one of the principal axes of D1} is, in this case, co-
directional with X. He followed these arguments for sufficiently large Reynolds
number, the dispersivity tensor for axisymmetric porous medium can be expressed
as
D e,v v e. (v X + v X)
-f- = ex8t) + ^ + eMj + ' J ' ' (5-38)
IV V V
where s,, E 2 , £ 3 , and £4 are parameters determined by the dimensionless
geometry of the medium and depends slightly on the Reynolds number and value
of cosft).On should note that the above derivation, primarily based on symmetry
considerations, can not reveal the scalar nature of general dispersivity tensor. Bear
[25] also noted that his analysis was based on an unproved assumption that D
may be expanded in a power series.
In 1967, Whitaker [28] applied pointwise volume-averaging method for transport
equation in anisotropic porous media and obtained the dispersion tensor DtJ as
D„ = D0 (Su + RB,J) + Clkjvk + Elkmivkvm (5.39)
where the second order tensor Bl} is a function of tortuosity vector
T = JQrijds (5.40)
s
The third-order tensor Clkj is given by

131
Cikj= VTTTVT (5-41)

5(v,)a
v ^ j
And the fourth-order tensor Eikmj is introduced as
a 3 (nv
v
;/
^ = ;T. , _ ^ (5-42)
a(a)
3<vt>a(v->a 3x
V , 7
where Q. is deviation of concentration or temperature from the average and v; is
the velocity deviation:
Q. = fi -(Q) f and v, = v,. - (v,.}f (5.43)
One should keep in mind that from Eqns. (5.41) and (5.42), we know that Cikj and
Ejhnj are completely symmetrical.
On comparing Eqn. (5.38) with Eqn. (5.24), one can note that there are both
third- and fourth-order symmetric tensors associated with velocity in the
Whitaker's derivation, while Nikolaveskii [24], Bear [25], and Scheidegger's [26]
derivations only contain fourth-order symmetric tensor. Patel and Greenkorn [29]
suggested that Whitaker's expression for dispersion tensor is the correct one for
anisotropic media. It is shown in section 5.4.1. that there are two distinct
components of dispersion tensor for isotropic medium while the Whitaker's
expression for dispersion tensor resulted in only one component for isotropic
media.
The diffusion term becomes less important at higher velocities and as a result,
132
Eqn. (5.39) reduces to
D ~ C i v, + E .v, v (5.44)
ij îkjk ikmj k m
For isotropic media, the tensors Clkj and Elkm] must be isotropic. Hence, Clk] =0, and
Elkmj is a linear combination of the Kronecker deltas as expressed in Eqn. (5.26).
Since Elkm] is completely symmetric, Eqn. (5.42), the tensor Elkm] can be shown to
be
E. =a(S.S +S S.+SS. ) (5.45)

ikmj \ ik mj im kj ij km }
Therefore Eqn. (5.39) reduces to
D y =a(2v,v,+3,|v| 2 ) (5.46)
Assuming 1-D flow in Cartesian coordinate frame where
v{ = u v2 = v3 = 0 (5.47)
then, eqn. (5.46) can be written as
"3 0 0"
D = au 0 1 0 (5.48)
0 0 1
Equation 5.48 shows that the longitudinal coefficient of dispersion tensor in this
case, is three times the transverse coefficient. Greenkorn [29] showed
experimentally that the ratio J\ I Dx varies approximately from lower value of 3 to
the higher value of 60. He showed that this ration is a function of the flow
velocity.
This ratio of longitudinal to transverse dispersion coefficients is shown to be a

133
function of the velocity of flow. Although the Whitaker's method is at variance
with Greenkorn experimental results, it still does give the correct lower limit result
for a homogeneous, uniform, isotropic medium.
5.5 A New Combined Experimental/Numerical Methodology to Assess

Dispersion Tensor
5.5.1 Theory
A new experimental-numerical method for predicting the dispersion tensor is
introduced and applied in this study. Using this method, one can estimate the
thermal dispersion tensor for different types of fiber mats characterized as single-
scale and dual-scale porous media. In this method, a numerical modeling comes
along with an experimental setup where the combination of both results leads to
prediction of the dispersion tensor. Since mass and thermal dispersions are very
similar in the transport mechanism in porous media, they can be described using
the dimensionless transport equations, Eqn. (5.21). Due to the similarity of the
transport equations for temperature and concentration, we are employing the
analogy between the concentration governing equation (in our experimental setup)
and the temperature governing equation (in our numerical modeling) to identifying
the dispersion tensor. In this prediction, the growth of the initially circular tracer
(dye) with time is studied. The results of the dye expansion at different time
frames are compared with results from our numerical model. Once the numerical
model's result is adjusted the same as the one from experiments at a specific time,
134
the dispersion tensor is obtained from the corresponding tensor components used
in the numerical simulation. Therefore, the dispersion tensor in the numerical
modeling is the filling parameter used in order to match the numerical results with
the results obtained from the experiments.
5.5.2 Experimental Measurements and Devices
The experimental section of this combined experimental/numerical method is done
using the 1-D flow experiment setup. In order to run the setup, a stack of fiber-
glass mat (called preform) is placed inside the mold cavity (Fig. 5.4 A). Two sides
of the preform in the horizontal plane are sealed using the removable glue in order
to avoid the leaking and race tracking along the sides during the experiment. The
mold is then closed and sealed tightly. The top layer of the mold is made out of a
Polycarbonate material which is clear and allows observation of the flow behavior
in the porous media. The test fluid is injected from one end of the mold and an
outlet is designed at another end of the mold. A pressure transducer is connected at
the top of the mold to record the inlet pressure history (see Fig. 5.4 B).
Figure 5.4 The 1-D flow experiment setup: A) open mold cavity and B) the assembled mold
A video camera is located on the top of the mold with the clear top mold to record
the flow process during the experiments. The camera equipped with a digital timer
136
and a measuring bar located along the mold length and facing to the camera are
employed to estimate the velocity of tracer liquid at different times. A syringe
filled with the dye (Methylene Blue) is placed at the flow inlet to inject the colored
dye into the steady flow of liquid through the fiber mats.
The test fluid injection system is designed to inject the liquid at a constant volume
flow-rate that can be monitored through a computer. The flow injection system
consists of a steel cylinder floating in a larger polycarbonate cylinder filled with
the test liquid. The floating cylinder is connected to a level gage for monitoring
the level of test liquid inside the large cylinder. A feedback system using the level
gage is employed on this flow-meter setup in order to keep the injection at a
steady level for a constant flow-rate experiment (see Fig. 5.5).
Figure 5 5 The flow injection andflow-metersetup
The test liquid is composed of 50% corn syrup and 50% water. The corn syrup is
white and hence, the blue dye injected at the inlet of the mold has a good color
137
contrast with the test liquid. The density of the corn syrup and water mixture is
measured by weighing a specific volume of liquid using a weighing balance, Fig.
5.6 A. The dynamic viscosity of the test liquid is obtained by using a Brookfield
viscometer, Fig. 5.6 B. The porosity of the porous material inside the mold cavity
is estimated by submerging the porous material inside a graduated cylinder filled
with water, Fig. 5.6 C (the increment of the water level caused by inserting of the
porous material into the water gives us the solid volume of the porous material. On
the other hand, by knowing the volume of the mold cavity, one should be able to
calculate the porosity of the porous medium inside the mold cavity using the
r , Mold volume-Solid volume . _ , . . . . ^ .

formula: = ). These physical measurements are conducted
Mold volume
several times and averaged in order to have a reasonable property value.
(A) (B) (C)

Figure 5.6 Measuring devices for A) Density, B) Viscosity, and C) Porosity.
Once the whole 1-D flow experiment setup is ready, it can be run at different flow
138
rates with different porosities as well as different porous media micro-structures
corresponding to the single and dual-scale porous media. The blue dye is injected
when the test liquid is passing through the porous material under steady-state
condition. The dispersion of injected dye is visible due to the contrast of its color
with the test liquid's color. According to the literature, the initial circular spot of
injected dye expands to an elliptical shape with time for the homogeneous single-
scale porous medium. Fig. 5.7 shows the ellipse generated as a result of dispersion
in a single-scale porous medium (this picture shows details of the dye injection
into a fluid moving through the random fiber-glass mats).
0JOHNSOIVJ LEVEL f i T O O L . Ts.

I «• C-* *.*. r * ? H l ta ft to RT
Figure 5.7 Effect of dispersion in the form of an elliptical dye expansion in a single-scale porous medium.
In order to complete this combined experimental and numerical method, the
numerical results should accompany the experimental data so that the dispersion
terms can be estimated.

139
5.5.3 Numerical Modeling
Our numerical results are needed to match with the experimental data in
order to estimate the dispersion tensor associated with the flow through the fibrous
porous media. The transient energy equation is solved by means of the finite
element method using COMSOL multiphysics software. Fluid properties as well
as the porous-material porosity are used as parameters to solve the transient energy
equation. Boundary conditions and initial conditions are assumed for the geometry
created (see Fig. 5.8).
The initial temperature (or initial concentration) distribution in the form of a
circle is assumed inside the geometry, close to the inlet. The test fluid (corn syrup-
water mixture) is injected from left inlet boundary which flows out from the right
outlet boundary.
Corrective heat flux

Convective heat flux
\imei; (Outlet)
k Initial temperature (initial

concentration)
Figure 5.8 Schematic view of the geometry, the boundary and initial conditions.
5.5.4 Combined Experimental/Numerical Methodology
The dye can be injected either under the steady-state condition when the
140
porous medium is fully wet and the test fluid is passing at a constant flow-rate or
under transient conditions when the porous medium is partially wet and the
injected dye is carried forward by the liquid front. The injected dye initially in the
form of a circle is converted downstream along with the injected liquid.
Depending on the micro-structure of the fiber mat, the flow rate, the porosity, etc.,
the expansion pattern of the initial circle of dye varies widely from one type of
fiber mat to another. For a single-scale porous medium (random fiber mat), the
initially circle of dye will change to an elliptical shape when the dye moves along
the mold direction. As long as the major axis of the elliptical tracer is along the
flow direction and the minor axis of the ellipse is perpendicular to it, one should
be able to model this dispersion as a tensor.
A finite element method (FEM) is used to model the problem. A
rectangular geometry (Fig. 5.8) is taken with one end as the inlet and the other end
as the outlet. The initial circular tracer at a location close to inlet is defined in the
FEM by defining some elements initially with a known temperature (or
concentration). The temperature governing equation is used in the FEM and the
thermal conductivity value, K , in this governing equation is the dispersion tensor.
One can simply change the values for the tensor K in the form, of 2x2 matrix in
order to fit the experimental data by comparing the dye expansion with time in
both the experimental and numerical results.

141
5.6 Results and Discussions
Study of Thermal Dispersion in Single-Scale Porous Media
The dispersion in porous media occurs due to the combined effects of
molecular diffusion and convection of flow in pore spaces. In order to have a
better understanding of the effect of convection and molecular diffusion, effect of
each term is studied individually. The Fig. 5.9 shows the effect of molecular
diffusion on thermal dispersion in porous media where there is no convection
during heat transfer. As it is expected, the initial temperature (or concentration)
disperses in its initial location as there is no convection. A 2x2 unit matrix is
assumed as the dispersion coefficient for this case which leads to a full circular
expansion due to similar diffusion in both horizontal and vertical directions.
o © o Q
t=0s t=6s t=13 s t=40s

Figure 5.9 Effect of pure molecular diffusion of thermal dispersion by time
To gain physical insights of the aforementioned method, the dispersion results of
the experimental modeling and finite element method are compared. Constant
flow-rate of Q=2 ml/s through the random fiber mat, a single-scale porous medium
(Fig. 5.10) with porosity of s=0.8 at three different time frames are considered
here.
=6 "*•
5e
Figure 5 10 Random fiber-glass mat as a representative of single-scale porous media
In order to estimate the dispersion tensor in such a single-scale porous medium,
002 0 0 02 0 04 0O6 0 08 01 012 014 016 019 02
- ^ f
0 02 0 002 0 0 1 0 0 6 0.08 01 012 014 0 16 0 18 02

143
E
002 0 002 004 006 0 08 01 012
F
01-1 016 018 02 I
Figure 5.11 Matching the experimental and numerical results for a single-scale porous medium (random
fiber mat) at Q=2ml/s at three different time frames of A) 7 s, B) 33 s and C) 51 s.
the experimental observation has to be matched with the numerical simulation
results. Three different time frames are assumed for this analysis.
The mentioned combined experimental/numerical method tries to estimate
the dispersion tensor by matching the experimental data for longitudinal and
transverse terms of dispersion by those in the numerical simulation. Results of the
proposed experimental/numerical approach give the dispersion tensor of D=
for the random porous media as a representative of a

(J o.looc — oJ
single-scale porous medium, Fig. 5.11. For this single-scale porous medium, the
flow direction along the mold horizontal direction and the corresponding vertical
direction are the longitudinal and transverse directions, respectively for the
dispersion tensor. Our results for three different flow-rates (Q=lml/s, Q=2 ml/s
and Q=3 ml/s) show that the dispersion tensor remains the same for this type of
porous media.
144
For the sake of validation, a comparison has been done between our results and
the results from an available model [32]as
u A t 2
n = Lim
D„ j - '
2x
*— (5.49)
Eqn. (5.49) models the longitudinal dispersion coefficient of the injected tracer
particles at the inlet of the porous medium sample. In this equation, U is the mean
tracer velocity which covers distance x in At seconds. In order to validate our
results with this model, two different time frames were selected and the velocity,
At, and the distanced traveled are substituted in Eqn. (5.49).
The porous medium is assumed to be Isotropic for this section. The tracer is
moving with a speed u=0.0002m/s at the initial location of x=0.070 m at t= 36 s to
the next location of x=0.090 at t=40 s. Following Eqn. (5.49), the longitudinal
dispersion is D = 3.20 xlO "7 while the longitudinal dispersion predicted by our
combined experimental/numerical method is D=3.19 xlO"7. One can see that there
is a good match between our results and those obtained from Eqn. (5.49) for the
longitudinal dispersion in the random fiber mat.
A study on the changing dimensions of the expanding tracer-generated ellipse in
the considered single-scale porous medium is presented in Fig. 5.12. This picture
shows the increase in the longitudinal and transverse dimensions of the initial
circular dye-batch with time. One can notice a rapid expansion along the
longitudinal direction compared to the transverse direction due to the present
convection effects along the flow direction.

145
_ — — -s
0 Lon
' Longitudinal Dimension (L)
— B - Trar
Transverse Dimension (D)
20 40 60 80 100
Time (sec)
Figure 5.12 The longitudinal and transverse increases of the initial circular dye-patch with time.
Study of Thermal Dispersion in Dual-Scale Porous Media
Similar study was carried out on a dual-scale porous medium for the purpose of
estimation of the dispersion tensor.
Figure 5.13 Stitched fiber-glass mat as a representative of dual-scale porous media.
In this study, a procedure similar to the one applied for the single-scale case, is
applied to a cavity filled with dual-scale porous media, Fig. 5.13. Experimental
146
observations along with the numerical simulation results are matched in order to
estimate the dispersion tensor.
i^:
1
<Hm 6 » bCt ti UM W»
U
B
i
I
E F
Figure 5.14 Comparison of the experimental and numerical modeling for an anisotropic but periodic
medium at Q = 2 ml/s : A) experimental results at t = 0 s, B) numerical results at t = 0 s, C) expenmental
results at t = 35 s, D) numerical results at t = 35 s, E) experimental results at t = 71 s and F) numerical
results at t = 71 s.
147
An important characteristic of the stitched fiber-glass (as a type of dual-scale
porous media used in the experiments) is its symmetry along the tows and the
corresponding vertical direction. This anisotropic porous medium is assumed as a
periodic porous medium in the two directions. Analysis on this type of porous
media through the experiments and computer simulations are presented in Fig.
5.14.
Based on the symmetry present in this anisotropic periodic porous medium,
the dispersion of heat (or concentration) has to be symmetric with respect to the
direction of the flow. The prediction for this case gives
D _ r3.186£ - 7 0 1
\- 0 1.062E-8J'
Another anisotropic porous medium studied in this chapter is the unidirectional
glass-fiber mat shown in Fig. 5.15. The dual-porosity effect in this unidirectional
porous medium causes an unsaturated flow-front to appear during the mold-filling
process. The micro-structure of the unidirectional mat is such that the flow
initially moves between tow bundles, Fig. 5.16 A. Once the channel between the
tow bundles is filled with the test fluid, it begins to wet the tow bundles from sides
which slows the liquid front. However, one tows are saturated, the injected liquid
flows in the gap region only (Fig. 5.16 B).

148
Figure 5.15 Unidirectional fiber-glass as a representative of dual-scale porous media.
Based on aforementioned reasons, the actual flow-rate in channels under steady-
state conditions is much more than the Darcy velocity.
channel
-Q/ > > U
^ A ~~ Darcv
cs, channel cs,cavity (5.50)
•> • >
I
• >
•*
• »
Wet *• •*
\ [ '
X •A:"•A.A
Wet
Dry
A B
Figure 5.16 Unidirectional fiber-glass in A) initial unsaturated flow and B) saturated flow.
Results obtained from experiments confirm this theoretical prediction. The
velocity of the dye inside the channels between the tow bundle (ucAa„„ep0.068 m/s)
is an order of magnitude larger than theoretical Darcy velocity (uDarcy=0.0050

149
m/s). Fig. 5.17 depicts the tracer movements through the unidirectional mat
channels with a much higher velocity compared to the Darcy velocity.
t=0 s t=2 s t=6 s

Figure 5.17 Experimental observation of flow of tracer through a dual-scale porous medium
made of the unidirectional fiber mats
According to above-mentioned reasons, the dispersion tensor for this type of
fiber glass is not estimated here. The Darcy's law is not obeyed in the entire
domain. This type of problem should be handled separately by assuming two
flows (i.e., gap and tow region flows) with different velocities.
5.7 Summary and Conclusion
In this chapter, the non-isothermal flow through porous media is studied
experimentally and numerically. An effort was made to quantify the thermal
dispersion effect which enhances heat transfer in porous-media flows. A combined
experimental/numerical approach to estimate the thermal dispersion during the
non-isothermal flow in porous media is described. Both single-scale and dual-
scale porous media were considered. Studies were done over three different mass
flow-rates. The estimated longitudinal thermal dispersion value in single-scale
porous media is compared with available results in the literature. The validation
150
study showed a very good match, and hence, motivated us to explore this method
for the dual-scale porous media. Preferential flow through channels frustrated our
efforts to apply the proposed experimental/numerical approach to model thermal
dispersion in the unidirectional stitched mat.

151
References
1. Rudd CD, Long AC, Kendall KN, Mangin CGE. Liquid molding
technologies. Woodhead Publishing Ltd; 1997.
2. Chan, A.W. and S.-T. Hwang, Modeling Nonisothermal Impregnation of
Fibrous Media with Reactive Polymer Resin, Polymer Engineering &
Science, 1992. 32(5):p. 310-318.
3. Chiu, H.-T., B. Yu, S.C. Chen, et al., Heat Transfer During Flow and Resin
Reaction through Fiber Reinforcement. Chemical Engineering Science,
2000. 55(17): p. 3365-3376.
4. Lee, L.J., W.B. Young, and R.J. Lin, Mold Filling and Cure Modeling of
RTM and Srim Processes. Composite Structures, 1994. 27(1-2): p. 109-120.
5. Lin, R.J., L.J. Lee, and M.J. Liou, Mold Filling and Curing Analysis Liquid
Composite Molding. Polymer Composites, 1993. 14(1): p. 71-81.
6. Tucker, C.L. and R.B. Dessenberger, Governing Equations for Flow
through Stationary Fiber Beds, in Flow and Rheology in Polymer
Composites Manufacturing, S.G. Advani, Editor. 1994, Elsevier.
7. Lam, Y.C., S.C. Joshi, and X.L. Liu, Numerical Simulation of the Mould-
Filling Process in Resin-Transfer Moulding. Composites Science and
Technology, 2000. 60(6): p. 845-855.

152
8. Wu, C.H., H.-T. Chiu, L.J. Lee, et al., Simulation of Reactive Liquid
Composite Molding Using an Eulerian-Lagrangian Approach. International
Polymer Processing, 1998(4): p. 398-397.
9. Bruschke, M.V. and S.G. Advani, Numerical Approach to Model Non-
Isothermal Viscous Flow through Fibrous Media with Free Surfaces.
International Journal for Numerical Methods in Fluids, 1994. 19(7): p. 575-
603.
10. Dessenberger, R.B. and C.L. Tucker, Thermal Dispersion in Resin Transfer
Molding. Polymer Composites, 1995. 16(6): p. 495-506.
ll.Kang, M.K., W. Lee, II, J.Y. Yoo, et al, Simulation of Mold Filling
Process During Resin Transfer Molding. Journal of Materials Processing
and Manufacturing Science, 1995. 3(3): p. 297-313.
12.Liu, B. and S.G. Advani, Operator Splitting Scheme for 3-D Temperature
Solution Based on 2-D Flow Approximation. Computational Mechanics,
1995. 16(2): p. 74-82.
13. Mai, O., A. Couniot, and F. Dupret, Non-Isothermal Simulation of the
Resin Transfer Moulding Process. Composites - Part A: Applied Science
and Manufacturing, 1998. 29(1-2): p. 189-198.
14.Ngo, N.D. and K.K. Tamma, Non-Isothermal '2-D Flow/3-D Thermal'
Developments Encompassing Process Modeling of Composites:
Flow/Thermal/Cure Formulations and Validations. American Society of

153
Mechanical Engineers, Applied Mechanics Division, AMD, 1999. 233: p.
83-102
15. Shojaei, A., S.R. Ghaffarian, and S.M.H. Karimian, Simulation of the
Three- Dimensional Non-Isothermal Mold Filling Process in Resin Transfer
Molding. Composites Science and Technology, 2003. 63(13): p. 1931-
1948.
16. Shojaei, A., S.R. Ghaffarian, and S.M.H. Karimian, Three-Dimensional
Process Cycle Simulation of Composite Parts Manufactured by Resin
Transfer Molding. Composite Structures, 2004. 65(3-4): p. 381-390.
17. Young, W.-B., Three-Dimensional Nonisothermal Mold Filling
Simulations in Resin Transfer Molding. Polymer Composites, 1994. 15(2):
p. 118-127.
18. Young, W.-B., Thermal Behaviors of the Resin and Mold in the Process of
Resin Transfer Molding. Journal of Reinforced Plastics and Composites,
1995. 14(4): p. 310.
19.Pillai, K.M. and M.S. Munagavalasa, Governing Equations for Unsaturated
Flow through Woven Fiber Mats. Part 2. Non-Isothermal Reactive Flows.
Composites Part A: Applied Science and Manufacturing, 2004. 35(4): p.
403-415.
20.Phelan, F.R., Jr. Modeling of Microscale Flow in Fibrous Porous Media.
1991. Detroit, MI, USA: Publ by Springer-Verlag New York Inc., New
York, NY, USA.

154
21. Donald A. Nield and Adrinan Bejan, Convection in Porous Media,
Springer, 3rd edition.
22. Robert A. Greenkorn, Flow phenomena in porous media, Marcel Dekker,
INC., New York and Besel, 1983.
23. Rubin, H., Heat dispersion effect on thermal convection in a porous
medium layer. J. Hydrol. 21, 1074,P: 173-184.
24. Nikolaevskii, V.N., Convective diffusion in porous media, Journal of
applied mathematics and mechanics, 23, 6, 1492-1503, 1959.
25. Bear, J., On the Tensor Form of Dispersion in Porous Media. Journal of
Geophysical Research, 1961. 66: p. 1185-1197.
26. Scheidegger, A.E., General Theory of Dispersion in Porous Media. Journal
of Geophysical Research, 1961. 66(10): p. 3273-3278.
27.Poreh, M., The Dispersivity Tensor in Isotropic and Axisymmetric
Mediums. Journal of Geophysical Research, 1965. 70: p. 3909-3913.
28.Whitaker, S., Diffusion and Dispersion in Porous Media. AIChE Journal,
1967. 13(3): p. 420-427.
29.Patel, R.D. and R.A. Greenkorn, On Dispersion in Laminar Flow through
Porous Media. AIChE Journal, 1970. 16(2): p. 332-334.
30. Beavers, G.S. and Joseph D.D., Boundary condition at a naturally
permeable wall, J. Fluid Mech,. Vol. 30, pp. 197-207, 1967.

155
31.Sahraoui, M. and Kaviany, M., Slip and no-slip boundary condition at
interface of porous plain media, Int. J. Heat Mass Transfer, Vol. 35, pp.
927-943, 1992.
32.Bouchaus J.P., GorgeJ.P. Bouchaud, A. Georges, Anomalous diffusion in
disordered media: statistical mechanisms, models and physical applications,
Phys. Rep. 195 (1990) 127-293.

156
Chapter 6
SUMMARY, CONTRIBUTION, AND FUTURE
WORK
6.1 Summary
In this Ph.D. thesis, transport and flow phenomena in porous media were
studied from several different points of view. We have first studied the effect of
preform aspect ratio on the permeability measurement in both single- and dual-
scale porous media under steady and transient conditions. Both 1-D flow and
radial flow experiments were conducted. Four different aspect ratios were
considered for both the experimental study as well as the numerical modeling. It is
concluded the longer aspect ratio leads to more accurate results which agrees with
the previous studies. Two different criteria were introduced in this work to capture
error in permeability measurements and two models for these errors were obtained
by curve fitting linear and quadratic equations to experimental data. Transient
mold-filling model showed huge differences in the pressure profile and flow-front
shape between the single- and dual-scale porous media. The mold filling
simulation with PORE-FLOW® and experimental data emphasized the fact that
flow in dual-scale porous media can not be predicted by the simulation based on
157
the single-scale flow physics.
The interface boundary condition problem between the clear-flow channel
and porous wall studied in a separate chapter. The main focus of this study is the
boundary condition estimation at the interface for both periodic and non-periodic
porous media. The results have been carried out with different REVs to ensure the
accuracy in the results. Our finite element method's results are then compared
with available results in the literature.
One of the chapters of this proposal is dedicated to a new method proposed
in this work on predicting different flow variables in porous media independent of
the REV size. The ensemble averaging method successfully proposed and applied
to some problems in porous-media flow. Using this method, one can solve
problems of flow in porous media without any concern about the size of the REV
especially with complex geometry or in a problem with sharp gradients in some
parameters. Two different practical applications of such method are presented and
solved. The comparison results showed a reasonable agreement between the
ensemble averaging method and the other available results.
Heat transfer aspect of a flow through porous media studied with particular
stress in thermal dispersion term. The governing equations are derived and
presented for both micro and macro scales. Most of the work done in the
dispersion so far is dedicated to the single-scale porous media. In this work, a

158
combined experimental/numerical approach is proposed to predict the dispersion
tensor in both single-scale and dual-scale porous media. Using this method, one
can predict the dispersion tensor correctly in different types of porous media and
use this term in non-isothermal flow through porous media such as the thermal
modeling of different composite plastics processing.
6.2 Contributions
The selected scientific contributions resulting from this Ph.D. thesis, which
have either already been published in conference proceeding or to be submitted for
publication in technical journal papers, are as follows:
• Study of the aspect ratio variations on the accuracy of permeability
estimation under transient and steady-state conditions of flow through
porous media.
• Proposing two new models based on our experimental data for the
estimation of error in permeability estimation through 1-D flow
experiments.
• Emphasizing the fundamental difference between the flow modeling of
single- and dual-scale porous media through the experimental and
numerical comparisons of the flow front evaluation.
• Evaluating the accuracy of various boundary conditions at the clear
channel-porous medium interface.

159
• First numerical study on the efficacy of boundary conditions at the interface
of a non-periodic porous medium and clear channel by employing a weakly
disordered porous medium created by random distribution of cylinders.
• A new method based on ensemble averaging for flow variable estimation in
porous media flows with sharp gradients.
• Successfully applying the proposed ensemble averaging method in two
industrial applications for validation purposes.
• A new combined experimental/numerical approach for estimating the
thermal dispersion term during non-isothermal flows in porous media.
• Successfully employing the new proposed technique for the estimation of
the thermal dispersion tensor for flows in dual-scale porous media.
6.3 Future Work
Isothermal and non-isothermal problem of flow through porous media
have many industrial and medical applications, each requiring a special
treatments. In this Ph.D. thesis, appropriates theories were developed for the
study of flow through porous media in some of these applications. The
following are the suggestions for improvement in the current work for future
studies:
• One can develop a better numerical modeling of the role of aspect ratio
on the measurement of permeability of porous media in order to have a

160
better comparison under both the steady-state and transient flow
conditions.
• Experimental modeling of the clear-channel porous-media interface
problem through several velocimetry techniques in order to have an
exact experimental results for the velocity profile at the interface (One
may use a simple color- tracer technique for the observation of flow of a
liquid at the interface of the clear channel and porous medium).
• Applying ensemble averaging method to more applicable, large-flow-
gradient problems (such as study of natural convection in porous media
adjacent to a hot flat plate, porous media flows with temperature
dependent viscosity, forced convective heat transfer during the flow of
resin through fiber-glass mats in resin transfer molding process, etc.)

161
Appendix
The irregular porous medium is made out of fixed number of cylinders is
treated as the solid phases of porous medium in the lower channel. The cylinders
are located randomly. A program written in FORTRAN randomly generates the
horizontal and vertical locations of the cylinder centers for each unit cell inside the
porous region. The simple case is that all cylinders located at the centers of the
unit cells, such that a periodic porous media is generated (see Fig.Al).
Figure Al Schematic view of locating solid cylinders at the center of the unit cells in order to generate a
periodic porous media.
On the other hand, in order to locate the cylinders randomly, Ax and Ay
defined as the horizontal and vertical distances of a cylinder's center from the
unit-cell center, respectively, are generated randomly (see Fig. A2).
An important constraint in this process is that no overlapping of cylinders is
allowed. In this case, each cylinder is allowed to go beyond the unit cell at
maximum half of the cylinder ( Axmax = Cylinder radius or Aymax=Cylinder

radius).
Figure A2 Schematic view of location of the solid cylinder at horizontal and vertical distances from the
center of unit cell in order to generate the non-periodic porous media.
Further, if location of the randomly generated cylinders overlapped, another
random location for the overlapped cylinder is generated so that the "no
overlapping" rule is implemented (see Fig. A3).
Figure A3 No overlapping rule applied to two cylinders generated in wo adjoining unit cells which
generating the irregular porous medium.
163
EHSAN MOHSENILANGURI
Address: 4042 North Wilson Drive, Phone: (414)364 5561
Unit 4 Email: Ehsan@uwm.edu,
Milwaukee, WI 53211 mohseni esn@vahoo.com
Education
2007- Ph.D. University of Wisconsin-Milwaukee (UWM), USA, Mechanical

2011 Engineering
•
Dissertation Title: Isothermal and Non-Isothermal Rows in
Porous Media
• Advisor: Associate Professor Krishna M. Pillai
2005- M.Sc. University of Mazandaran (UMI), Iran, Mechanical Engineering
2007 • Thesis Title: Study of solar airs collector in Building Energy
Analysis
• Advisor: Associate Professor Hessam Taherian
2001- B.Sc. University of Mazandaran (UMI), Iran, Department of Mechanical
2005 Engineering
• Senior Project Title: Experimental Study on Electro-Magnetic
Row-meter
• Advisor: Associate Professor Hessam Taherian
Publications
• Mohseni Languri E., Moore D. R., Masoodi, R., Pillai, K., Sabo, R., An Approach to
Model Resin Flow Through Swelling Porous Medium using Natural Fibers, 10th
International Conference on Flow Processing in Composite Materials (FPCM 10), July
2010, Ascona, Switzerland.
• Mohseni Languri E., Bennett III G. L., Masoodi R., Pillai K. M., A Reference Porous
Medium Made by Rapid Prototyping as a Calibration Tool for Permeability, 10th
International Conference on Flow Processing in Composite Materials (FPCM 10), July
2010, Ascona, Switzerland.
• Mohseni Languri E., Tan H. and Pillai K. M., Effect of Preform Aspect Ratio and
Anisotropy on the Transient 1-D Mold Filling in LCM, 9th International Conference
on Flow Processes in Composite Materials (FPCM 9), 2008, Montreal, Canada.
• Tan H., Mohseni Languri E. and Pillai K. M., Permeability prediction of dual-scale
fibrous structure using the unit cell, 9th International Conference on Row Processes in
Composite Materials (FPCM 9), 2008, Montreal, Canada.
• Mohseni Languri E., Vechart A., Tan H. and Pillai K. M., "Effect of Preform Aspect
Ratio on Permeability measured through ID Flow Experiments, 9th International
164
Conference on Flow Processes in Composite Materials (FPCM 9), 2008, Montreal,

Canada.
• Mohseni Languri E. and Pillai K.M., Estimating the Average Flow Variables in
Porous Media using Ensemble Averaging Method, Journal of Porous Media, 2010.
• Mohseni Languri E. and Pillai M. K., Numerical Study of Stress-Jump and Stress-
Continuity Conditions at Random Porous-Medium-Clear Fluid Interface Problems,
Physics of Fluid, 2010.
• Mohseni Languri E. and Pillai M. K., Study of Thermal Dispersion Term in Dual-
Scale Porous Media, International Journal of Heat and Mass Transfer, 2010.
• Ganji D.D., Jannatabadi M., Mohseni Languri E., Application of He's Variational
Iteration Method to Nonlinear Jaulent-Miodek Equations and Comparing it with
ADM, Journal of Computational and Applied Mathematics, Vol. 207 , Issue 1,
Pages 35-45, 2007.
• Ganji D.D., Nourollahi M., Mohseni Languri E., Application of He's Methods to
Nonlinear Chemistry Problems, Computers & Mathematics with applications,
Vol. 54, Issues 7-8, Pages 1122-1132, 2007.
• Ganji D.D., Jannatabadi M., Mohseni Languri E., The Comparison of Homotopy
Perturbation Method with Adomian's Decomposition Method in Coupled Jaulent-
Miodek Nonlinear Equations, Far East Journal of Applied Mathematics, Vol. 29,
Issue 1, Page 19-30, 2007.
Teaching Experience
2010 Fall Introduction to Fluid Mechanics (Adjunct Instructor) UWM

2009 Summer Introduction to Fluid Mechanics (Adjunct Instructor) UWM
Spring Engineering Fundamentals II (Teaching Assistant) UWM
2008 Fall Fundamental of Physics (Teaching Assistant) UWM
Fall Engineering Fundamentals I (Teaching Assistant) UWM
2007 Fall Introduction to Fluid Mechanics (Grader) UWM
Spring Thermodynamic Laboratory (Lab Assistant) UMI
2006 Fall Fluid Mechanics Laboratory (Lab Assistant) UMI
Spring Fluid Mechanics Laboratory (Lab Assistant) UMI
2005 Fall Fluid Mechanics Laboratory (Lab Assistant) UMI
Fall Thermodynamic Laboratory (Lab Assistant) UMI
Major Professor Date

W/AOI

Modeling Isothermal and Non-Isothermal Flows in Porous Media

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Modeling Isothermal and Non-Isothermal Flows in Porous Media

Uploaded by

Copyright:

Available Formats

MODELING ISOTHERMAL AND NON-

ISOTHERMAL FLOWS IN POROUS MEDIA

Ehsan Mohseni Languri

Partial Fulfillment of the

Requirements for the Degree of

The University of Wisconsin-Milwaukee

All rights reserved

INFORMATION TO ALL USERS

ISOTHERMAL FLOWS IN POROUS MEDIA

Ehsan Mohseni Languri

Partial Fulfillment of the

Requirements for the Degree of

The University of Wisconsin-Milwaukee

Graduate School Approval Date

MODELING ISOTHERMAL AND NON-

ISOTHERMAL FLOWS IN POROUS MEDIA

A complete understanding of the physics of flow and heat transfer

phenomena in porous media is vital for accurate simulation of flow processes in

industrial applications. In one such application pertaining to liquid composite

porous media is analyzed experimentally and numerically.

conditions is evaluated after volume averaging the point-wise velocity using a

three different Reynolds numbers to cover a wide range of applications. The

presence of randomly-placed cylinders during the creation of non-periodic porous

porosities, which is in contradiction with the results achieved earlier by other

The traditional approach of using averaged equations in the regions of sharp

gradients in porous media to describe flow and transport is theoretically untenable

and perhaps inaccurate. A novel ensemble averaging method is being proposed to

test the accuracy of the volume averaged or smoothed description of flows in

certain arrangement of particles (called a realization) is averaged using a small

further averaged over a whole gamut of randomly-generated particle realizations.

permeability of an artificially generated porous medium obtained by the proposed

method against the permeability predicted by some established theoretical models

ensemble average method was applied to the open-channel porous-medium

interface region characterized by a sharp gradient in the flow velocities. It was

discovered that the volume averaged description of such flows, characterized by

conditions, is quite accurate for a range of Reynolds numbers.

The non-isothermal transport during flow in porous media is examined next.

combined experimental/numerical approach to estimating the dispersion tensor is

exploited and a 1 -D flow experimental setup is employed to study the spreading of

a dye. Later the mathematical model for such a spreading of concentration

(equivalent to the temperature) around a point input in a constant velocity field is

each individual case.

A chapter-wise summation of this Ph.D. thesis proceeds as follows. The

unidirectional fibrous porous media was studied experimentally and numerically.

In the next chapter, an evaluation of the boundary conditions employed to describe

the free-channel porous-medium interface is proposed after using a long REV

both periodic as well as non-periodic porous media. Later, a new ensemble

on estimating the thermal dispersion term.

1.1. Literature Review on Unsaturated Flow through Dual-scale Porous 5

1.2. The Scope of Research 8

Chapter 2. EFFECT OF ASPECT RATIO ON MEASURED PERMEABILITY

AND FLOW-FRONT PROGRESS

2.1 Permeability Measurements 16

2.1.1 1-D Flow Experiment 17

2.1.2 Radial Flow Experiment 18

2.2 Effect of Preform Aspect Ratio on the Permeability Measured through 19

the 1-D Flow Experiment

2.2.3 Results and Discussions 26

2.3 Effect of Preform Aspect Ratio on the Unidirectional Flow Front 31

during Mold-filling in the LCM

2.3.1 Results and Discussion 33

CHAPTER 3 Evaluation of Boundary Conditions at the Clear-Fluid Porous-Medium

3.2 Design of Numerical Experiments 47