pore-scale imaging and lattice boltzmann modeling …
TRANSCRIPT
The Pennsylvania State University
The Graduate School
College of Earth and Mineral Sciences
PORE-SCALE IMAGING AND LATTICE BOLTZMANN MODELING OF SINGLE-
AND MULTI-PHASE FLOW IN FRACTURED AND MIXED-WET PERMEABLE
MEDIA
A Dissertation in
Energy and Mineral Engineering
by
Christopher James Landry
© 2013 Christopher James Landry
Submitted in Partial Fulfillment
of the Requirements
for the Degree of
Doctor of Philosophy
May 2013
ii
The dissertation of Christopher James Landry was reviewed and approved* by the following:
Zuleima T. Karpyn
Associate Professor of Petroleum and Natural Gas Engineering
Dissertation Adviser
Chair of Committee
Li Li
Assistant Professor of Energy and Mineral Engineering
Russell T. Johns
Professor of Petroleum and Natural Gas Engineering
Maria Lopez de Murphy
Associate Professor of Civil Engineering
Luis Ayala
Associate Professor of Petroleum and Natural Gas Engineering
Associate Department Head for Graduate Education
*Signatures are on file in the Graduate School
iii
ABSTRACT
Three investigations of pore-scale single-phase and multiphase flow in fractured porous media
and mixed-wet porous media are presented here. With an emphasis on validating and utilizing
lattice Boltzmann models in conjunction with x-ray computed microtomography.
The objective of the first study is to investigate fracture flow characteristics at the pore-scale,
and evaluate the influence of the adjacent permeable matrix on the fracture’s permeability.
We use X-ray computed microtomography to produce three-dimensional images of a fracture in
a permeable medium. These images are processed and directly translated into lattices for
single-phase lattice Boltzmann simulations. Three flow simulations are presented for the imaged
volume, a simulation of the pore space, the fracture alone and the matrix alone. We show that
the fracture permeability increases by a factor of 15.1 due to bypassing of fracture choke points
through the matrix pore space. In addition, pore-scale matrix velocities were found to follow a
logarithmic function of the distance from the fracture. Finally, our results are compared
against previously proposed methods of estimating fracture permeability from fracture
roughness, tortuosity, aperture distribution and matrix permeability.
In the second study we present a pore-scale study of relative permeability dependence on the
strength of wettability of homogenous-wet porous media, as well as the dependence of relative
permeability on the distribution and severity of wettability alteration of porous media to a mixed-
wet state. A Shan-Chen type multicomponent multiphase lattice Boltzmann model is employed
to determine pore-scale fluid distributions and relative permeability. Mixed-wet states are
iv
created – after pre-simulation of homogeneous-wet porous medium – by altering the wettability
of solid surfaces in contact with the non-wetting phase. To ensure accurate representation of
fluid-solid interfacial areas we compare LB simulation results to experimental measurements of
interfacial fluid-fluid and fluid-solid areas determined by x-ray computed microtomography
imaging of water and oil distributions in bead packs (Landry et al. 2011). The LB simulations
are found to match experimental trends observed for fluid-fluid and fluid-solid interfacial area-
saturation relationships. The relative permeability of both fluids in the homogenous-wet porous
media is found to decrease with a decreasing contact angle. This is attributed to the increasing
disconnection of the non-wetting phase and increased fluid-solid interfacial area of the wetting
phase. The relative permeability of both fluids in the altered mixed-wet porous media is found to
decrease. However the significance of the decrease is dependent on the connectivity of the
unaltered solid surfaces, with less dependence on the severity of alteration.
In the third study we present sequential x-ray computed microtomography (CMT) images of
matrix drainage in a fractured sintered glass granule pack. Sequential imaging captured the
capillary-dominated migration of the non-wetting phase front from the fracture to the matrix in a
brine-surfactant-Decane system. The sintered glass granule pack was designed to have minimal
pore space beyond the resolution of CMT imaging, so that the pore space of the matrix
connected to the fracture could be captured in its entirety. The segmented image of the pore
space was then directly translated to a lattice to simulate the transfer of fluids between the
fracture and the matrix using lattice Boltzmann (LB) modeling. This provided us an opportunity
to validate the modeling technique against experimental images at the pore-scale. Although the
surfactant was found to alter the wettability of the originally weakly oil-wet glass to water-wet,
v
the fracture-matrix fluid transfer is found to be a drainage process, showing little to no counter-
current migration of the oil-phase. The LB simulations were found to closely match
experimental rates of fracture-matrix fluid transfer, equilibrium saturation, irreducible wetting
phase saturation and fluid distributions.
vi
TABLE OF CONTENTS
List of Tables ………………………………………………………………….. ix
List of Figures …………………………………………………………………. ix
Acknowledgements ……………………………………………………………
xv
CHAPTER 1: INTORDUCTION TO PORE-SCALE STUDIES ……………. 1
CHAPTER 2: SINGLE-PHASE LATTICE BOLTZMANN …………………
SIMULATIONS OF PORE-SCALE FLOW IN FRACTURED
PERMEABLE MEDIA
9
2.1 Summary .………………………………………………………………. 9
2.2 Introduction ……………………………………………………………. 9
2.3 Materials and Methods ………………………………………………… 17
2.3.1 X-ray Computed Microtomography Scanner …..……………….. 17
2.3.2 Fractured Permeable Media ….…...……………………………… 18
2.3.3 Image Processing ….…………………………………………….. 20
2.3.4 Fracture Pore Space Identification ...…………………………….. 22
2.3.5 Surface Area and Volume ….…………………………………….. 24
2.3.6 Aperture and Fracture Roughness .………………………………. 24
2.3.7 Lattice Boltzmann Simulations .…………………………………. 28
2.4 Results ..……………………………………………………………….. 31
2.4.1 Fracture and Matrix Pore Space Geometry .……………………… 31
2.4.2 LB Simulation Results ……..…………………………………….. 33
2.4.3 Fracture Permeability Estimation ……………………………….. 36
2.5 Conclusions ….……………………………………………………….. 43
vii
2.6 References …………..………………………………………………… 46
CHAPTER 3: RELATIVE PERMEABILITY OF HOMOGENOUS-WET ….
AND MIXED-WET POROUS MEDIA AS DETERMINED BY
PORE-SCALE LATTICE BOLTZMANN MODELING
55
3.1 Summary ..……………………………………………………………… 55
3.2 Background ….………………………………………………………….. 56
3.3 Materials and Methods ….………………………………………………. 64
3.3.1 Experimental Measurements …………………………………….... 64
3.3.2 Single-Phase BGK Lattice Boltzmann Model …………………….. 65
3.3.3 Shan-Chen Multicomponent Lattice Boltzmann Model …………. 67
3.3.4 LB Model Implementation ………………………………………… 69
3.4 Results …………………………………………………………………… 72
3.4.1 Fluid-Fluid and Fluid-Solid Interfacial Areas …………………….. 72
3.4.2 Pore Aperture and Fluid Distribution ……………………………… 77
3.4.3 Two-Phase Flow in a Circular Pore ……………………………….. 79
3.4.4 Relative Permeability of Homogenous-Wet States ………………. 81
3.4.5 Relative Permeability of Mixed-Wet States ………………………. 84
3.5 Conclusions …………………………………………………………….. 92
3.6 References ……..…………………………………………………………
95
CHAPTER 4: LATTICE BOLTZMANN MODELING AND ………………....
4D X-RAY COMPUTED MICROTOMOGRAPHIC IMAGING OF
FRACTURE-MATRIX FLUID TRANSFER
106
4.1 Summary ………………………………………………………………. 106
4.2 Introduction ……………………………………………………………... 107
4.3 Materials and Methods ………………………………………………….. 110
viii
4.3.1. Fractured Porous Medium ………………………………………… 110
4.3.2 Procedure ………………………………………………………….. 111
4.3.3 CMT Imaging …………………………………………………….... 113
4.3.4 Image Processing …………………………………………………... 113
4.4 Results ………………………………………………………………….... 115
4.4.1 Pore Space Analysis ………………………………………………... 115
4.4.2 Immiscible Displacement …………………………………….……. 121
4.4.3 LB Simulation of Fracture-Matrix Fluid Transfer ………….……... 127
4.4.4 Pore-Scale Experimental and Model Results ………………..…….. 130
4.5 Conclusions ……….……………………………………………………... 137
4.6 References …….…………………………………………………………. 140
CHAPTER 5: SUMMARY OF FINDINGS AND RECOMMENDATIONS ….
FOR FUTURE WORK
148
ix
LIST OF TABLES
Table 2.1. Pore space geometry and fracture characteristics.
…………………33
Table 2.2. Summary of LB permeability measurements and fracture
permeability estimates.
…………………41
Table 4.1. Temperature cycles for sintering of glass granules.
…………………110
LIST OF FIGURES
Figure 1.1 An example segmented CMT image showing the pore
space voxels in white (A). Lattice Boltzmann models
simulate fluids as swarming particles (B) on a discrete
lattice, where each pore voxel is represented by a LB
fluid node (C). The complex solid boundaries are also
retained by translating the CMT image 1-to-1 voxel-to-
node with the solid boundaries being translated to
bounce-back nodes, shown in yellow (D).
…………………4
Figure 2.1. Illustration of the double-sided ‘log-splitting’ approach
used to propagate a rough fracture in the porous
polyethylene rod.
…………………19
Figure 2.2. Demonstration of simple thresholding image
segmentation. At the left is a raw CMT image slice, the
raw data is cropped (shown as dashed lines) to remove
portions of fracture that were in contact with the splitting
tool. The middle plot shows the frequency of CT
registration numbers, and Gaussian fits to the solid and
pore phases in the image, the intercept of these Gaussian
fits is used to segment the image. At the right is a
segmented image, in which gray represents the pore
space and white represents the solid or grain.
…………………21
x
Figure 2.3. Example slices from the fracture space identification
procedure. The left image shows the pore space (light
gray) after 8 cycles of erosion, at which point all of the
matrix pore space is removed, this locates the fracture
(dark gray). The middle image shows the pore space in
contact with the fracture space after 2 cycles of reversed-
out erosions, the erosion cycle one short of the fracture
pore space becoming connected with the matrix pore
space; this identifies the starting point for expanding the
fracture pore space out. The right image shows the
fracture pore space after 6 cycles of expansion, the
expansion cycle one short of the fracture pore space
making contact with the grains or solid, this defines the
fracture.
…………………23
Figure 2.4. Illustration of the measure of the vertical aperture of the
fracture ( ), the perpendicular aperture ( ), the fracture
profile distances from edge of image ( ) and the
bandwidth window ( ). The light gray is the matrix pore
space, and the dark gray is the fracture.
…………………24
Figure 2.5. Fracture aperture distribution. The small initial peak is a
result of the matrix pore aperture distribution impacting
the aperture distribution of the fracture at fracture
boundaries. The distribution is overall log-normal,
consistent with fractures in natural permeable media.
…………………26
Figure 2.6. Poiseuille flow through a channel with a radius
velocity profile, comparison of LB results
and analytical solution. These results show the LB
simulation is closely approximating flow in the small
channel.
…………………31
xi
Figure 2.7. Determination of Hurst coefficient using the variable
bandwidth method. Data points are shown for four
slices, the linear function shown ( ) was
determined from 16 equidistant slices spanning the range
of the simulated pore volume. The Hurst exponent is the
slope of the fitted linear function shown.
…………………32
Figure 2.8. Normalized x-velocity glyph visualization of the
fracture-alone (above) and the all-pore (below)
simulations; the glyphs are larger and farther to the red
side of the spectrum for greater velocities.
Th28maximum normalized velocity shown in the gly29s
is .
…………………35
Figure 2.9. Mean normalized velocities ( ) as a function of
distance from the fracture ( ), and logarithmic function
fits.
…………………27
Figure 3.1. Segmented CMT image of the bead pack (A) with solid
voxels colored white, and the corresponding lattice of the
LB model (B) with the bounce-back nodes in blue.
…………………71
Figure 3.2. Specific fluid-fluid interfacial areas as a function of
wetting phase saturation for the LB simulations and
experimental CMT images.
…………………76
Figure 3.3. Fractional fluid-solid interfacial areas as a function of
wetting phase saturation for the LB simulations and
experimental CMT images, the wetting fluid-solid
interfacial areas are shown in black and the non-wetting
fluid-solid interfacial areas are shown in gray.
…………………76
Figure 3.4. Pore aperture and fluid distribution of initial …………………78
xii
homogenous-wet states.
Figure 3.5. Relative permeability in a circular pore with radius = 11
; LB simulation results and semi-analytical solution
curves.
…………………81
Figure 3.6. Relative permeability determined by LB simulations for
the initial homogenous-wettability states. The relative
permeability of the wetting and non-wetting phase are
shown in black and gray, respectively.
…………………82
Figure 3.7. Images of the LB lattice with initial homogenous-
wettability (A), the fluid distribution of the non-wetting
phase at the end of the simulation of the initial
homogenous-wettability with and (B),
and the alteration of the bounce-back densities of the
nodes in contact with the non-wetting fluid (C). This
image also summarizes the three parameters determining
the mixed-wet state, initial wettability (A), saturation of
alteration (B), and severity of alteration (C).
…………………86
Figure 3.8. Relative permeability determined by LB simulations for
the mixed-wet states of the originally homogenous-wet
states, (A) , (B) , (C)
, and (D) , with
increasing severity of alteration. The relative
permeability of the wetting and non-wetting phase are
shown in black and gray, respectively.
…………………88
Figure 4.1. Image of the fracture with the ROI highlighted (A), and
the corresponding mean fracture aperture and contact
area profiles with the ROI indicated by arrows.
…………………116
xiii
Figure 4.2. Image of the matrix pore space showing the fracture
saturated with water phase and the matrix saturated with
the oil phase, the ROI volume is highlighted (A). Also
the corresponding porosity (B) and mean aperture (C)
profiles with the ROI indicated with arrows.
…………………118
Figure 4.3. Example 0.1183 cm
3 image of sphere-packing method
used to measure pore aperture distribution. The
individual pore apertures are measured as the radius of
the spheres seen in this image. The red, purple, teal,
green, yellow, and light green represent pore aperture
measurements between ,
, ,
, , and
mm, respectively. The
background shows the solid in black and the pore space
in white.
…………………120
Figure 4.4. The fracture aperture distribution (left) and the pore
aperture distribution (right) for the portion of the core to
be imaged sequentially during water invasion of the
matrix.
…………………121
Figure 4.5. Image of oil phase (white), water phase (black), glass
(gray) contact in the fracture after initial water phase
injection. The contact angle appears in this image to be
large, near neutral wetting conditions.
…………………122
Figure 4.6. CMT images from the sequential scanning of the
surfactant water phase injection showing a volume
rendering (left) and a slice from a height of 1.0704 cm
(right). In the volume renderings the water phase is dark
orange, the solid is light orange and the oil phase is
white. In the slices the water phase is light gray, the
glass is dark gray, and the oil phase is white.
…………………124
xiv
Figure 4.7. Non-wetting phase saturation of the right and left matrix
as determined from sequential CMT imaging. Early time
refers to displacement before the interfacial tension
reduction has stabilized. The linear fits shown represent
a displacement following the rate function,
⁄
.
…………………126
Figure 4.8. Volume images of the 270x150x150 voxel3 (0.4256 x
0.2365 x 0.2365 cm3) sample volumes of the pore space
for the left and right matrix to be translated 1-to-1, voxel
to lattice node, for lattice Boltzmann simulation (A). LB
simulation setup showing the 5 lattice layer thick
pressure boundaries, with the fracture represented as the
non-wetting phase occupied pressure boundary (B).
…………………128
Figure 4.9. Non-wetting phase saturation of the right and left matrix
as determined from sequential CMT imaging and LB
drainage simulations. The only variable between the LB
simulations for the right and left matrix is the pore space
geometry.
…………………131
Figure 4.10. Comparison of experimental (left) and LB simulation
(right) vertical saturation profiles for the right (A, B) and
left (C, D) matrix, and horizontal saturation profiles for
the right (E, F) and left (G, H) matrix.
…………………133
Figure 4.11. Sample images of the LB simulations of fracture-matrix
transfer. The non-wetting phase saturation front and
simulation pore space are shown, the wetting phase is
occupies the pore space not occupied by the non-wetting
phase.
…………………135
xv
ACKNOWLEDGEMENTS
This material is based upon work supported by the National Science Foundation under
Grant No. 0747585 and the Marathon Alumni Centennial Graduate Fellowship. I would
like to thank Dr. Orlando Ayala of the University of Delaware for providing computational
resources, and for offering his expertise and time, many of the simulations shown here
would not have been possible without his help. I would also like to thank Soheil Saraji
of Piri Research Group at the University of Wyoming for providing interfacial tension
measurements. I would also like to thank my doctoral committee for their thoughtful
comments, and insightful discussion. Last I would like to thank my advisor, Dr. Zuleima
Karpyn, who had the unenviable task of keeping me on task.
1
CHAPTER 1: INTRODUCTION TO PORE-SCALE STUDIES
The goal of studying multiphase fluid flow at the pore-scale in porous media and fractured
porous media is to improve fluid transport prediction at the field-scale. Generally prediction of
fluid transport at the field scale is determined by models that discretize the formation of interest
– attributing macroscale flow properties to individual gridblocks – and simulating flow under
varying scenarios. Macroscale flow properties are conventionally determined by measuring fluid
flow in cores take from the formation of interest. Pore-scale studies attempt to elucidate the
connection between pore space characteristics and fluid mobility, with the intention of
determining macroscale flow properties to be utilized in field-scale fluid transport prediction.
Accurate determination of macroscale flow properties such as permeability, relative permeability
– saturation relationships and capillary pressure – saturation relationships are fundamental to
accurate field-scale fluid transport prediction. There are numerous techniques used to measure
macroscale flow properties of porous media and fractured porous media, all of which determine
macroscale flow properties by measuring fluid flux through cores taken from the formation of
interest under varying applied boundary conditions. Although direct measurement of macroscale
flow properties is straight forward there are downsides to these methods. For one they require
cores that can be expensive to procure, and often these cores only offer a small sampling of a
vast heterogeneous formation. The measurements themselves are also expensive, and in the case
of fractured porous media, measuring flow in the fracture and porous matrix separately can be
very difficult, if not impossible. Most importantly, these methods provide measurements, but no
2
insight into the cause for resulting macroscale flow measurements. The porous media is treated
as essentially a “black-box”.
Pore-scale studies peer into this black-box to elucidate the relationship between pore space
characteristics and resulting macroscale fluid flow. Although the physics of multiphase fluid
flow is fairly well understood, the complexity of the solid boundary conditions encountered in
pore spaces render any possibility of analytical solutions of fluid flow to be beyond practicality.
Complex non-invasive imaging techniques, such as x-ray computed microtomography (CMT)
allow us to not only image pore spaces at resolutions great enough to capture the shape of
individual pores, but also the distribution of fluids within these pores. Being able to image pore
spaces and fluid distributions at the pore-scale provides insight into multiphase fluid flow, as in
we are able to peer into the black-box, but this does not alleviate some of the issues we
encountered with conventional measurements of multiphase fluid flow. It is still expensive, and
requires cores.
Pore-scale models offer us a powerful tool to not only gain insight into the connection between
pore space characteristics and resulting macroscale fluid flow. Pore-scale models use either
imaged pore spaces or synthesized pore spaces to simulate fluid flow at the pore-scale. There are
many different types of pore-scale models in existence, including classic computational fluid
dynamics (discrete solutions of Navier-Stokes equation), pore-network models, smoothed
particle hydrodynamics, the level-set method and lattice Boltzmann (LB) models. The best
choice of model to use when investigating fluid flow at the pore-scale is dependent on the
objective of the investigation, here we employ lattice Boltzmann models. The complexity of the
3
solid boundaries simulated in LB models is only limited by computational power, or in the case
of imaged pore spaces, the resolution of the image. This is an improvement over more
commonly used pore-network models which represent the pore space as a network of discrete
geometric shapes (i.e. cylinders, triangular tubes, square tubes, spheres etc.). Briefly, LB models
can be described as simulating fluids as swarming particles on a discrete lattice. A detailed
description of the model used in the investigations presented here can be found in Chapter 3.
The complex solid boundaries of pore spaces imaged by CMT can be translated 1-to-1 voxel-to-
node onto LB lattices To better visualize what is meant by maintaining the complexity of the
solid boundaries of pore space in LB models please refer to figure 1.1. In this figure we show a
CMT image of a pore space and the corollary LB lattice. LB models simulate fluid flow in a
purely local manner, no long range information is required during the evolution of the model,
each node only requires knowledge of adjacent nodes. This means LB models have an inherent
parallelism, and simulations can carried out across many processors using parallel computing
techniques. One of the greatest challenges encountered in the implementation of LB models is
the computational expense. Although LB models are inherently parallel and simulations can be
spread across hundreds, if not thousands of processor cores, the computational demand is so
great that multiphase simulations on 1003 node
3 lattices can require upwards of a week on 200
cores to reach convergence. However, it should be noted the only expense is computational. It
is possible to run numerous scenarios at only the cost of computer time.
4
Swarming Particles Discrete LBE
A)
B) C)
D)
Figure 1.1. An example segmented CMT image showing the pore space voxels in white (A). Lattice
Boltzmann models simulate fluids as swarming particles (B) on a discrete lattice, where
each pore voxel is represented by a LB fluid node (C). The complex solid boundaries are
also retained by translating the CMT image 1-to-1 voxel-to-node with the solid boundaries
being translated to bounce-back nodes, shown in yellow (D).
There are two main objectives of the three investigations presented here in chapters 2, 3, and 4.
The first is to validate multi-phase pore-scale fluid distributions predicted by LB models against
those imaged experimentally. One of the main goals of pore-scale studies is to better understand
the connection between pore space characteristics and resulting macroscale flow, considering
5
this, it is imperative that we validate LB prediction of pore-scale fluid distribution. LB models
have in many previous investigations been used to determine macroscale flow properties in the
same way as is done conventionally, by imposing pressure boundaries and monitoring fluid flux.
In these studies LB models have been shown to accurately predict macroscale flow properties.
For further information regarding previous investigations into prediction of macroscale flow
properties by LB please refer to subsection 3.2 Background. There are however only a handful
of studies (subsection 3.2 Background) that have attempted to compare LB model pore-scale
fluid distributions against experimentally imaged pore-scale fluid distributions. The second main
objective is to take advantage of the intimate knowledge of fluid velocity and distribution offered
by the LB model to perform numerical investigations into single-phase and two-phase fluid flow
that would be difficult if not impossible to perform in a physical laboratory setting.
With these objectives in mind we present three investigations. In chapter 2 we take advantage of
the intimate knowledge of velocity (velocity is known at every LB fluid node) offered by the LB
model to investigate single-phase flow in fractures bounded by a porous matrix. We simulate
flow in a fracture bounded by porous walls and the same fracture bounded by impermeable
walls, to investigate the effect porous walls have on fracture permeability. Because we have
intimate knowledge of velocity we simulate the flow in our fracture-matrix coupled system, and
measure the flow in the fracture exclusively. This would be nearly impossible in a conventional,
physical experimental system, because it is extremely difficult to separate the measurement of
fluid flow in the fracture from that of the matrix. In this investigation we also compare the our
results to existing fracture permeability estimations based on fracture characteristics, such as the
mean aperture, standard deviation of the aperture distribution, wall roughness, tortuosity and
6
contact area. Since many formations of interest are fractured, and fluid flow is often dominated
by flow in fractures in these formations, accurate estimations of fracture permeability are
imperative to the success of field-scale fluid transport prediction.
In the third chapter we engage in validation of LB model prediction of fluid distribution by
comparison to experimental results, and investigate fluid flow in mixed-wet states using the LB
model. Experimental pore-scale fluid distribution data sets are sparse, which is one of the
reasons so few investigations have attempted to compare model and experimental pore-scale
fluid distributions. We use an experimental data set we previously published (Landry et al.
2011) to compare to our LB model results. And after presenting an argument for validation of
our methods through comparison to experimental measurements, we engage in an investigation
of wettability alteration to mixed-wet states and the effect this has on the relative permeability of
the porous medium. We take advantage of the control of solid surface wettability (the adhesion
force of every LB bounce-back node is tunable) offered by the LB model to investigate two-
phase flow in mixed-wet porous media. Although numerical studies into the relative
permeability of mixed-wet porous media using pore-network models has existed for over twenty
years (subsection 3.2 Background), this is the first thorough investigation of two-phase flow in
mixed-wet porous media using LB to the author’s knowledge. Also much like in the first
investigation, this investigation would be very difficult to perform in a conventional, physical
experimental system, due to the difficulty in engineering the wettability of physical porous
media.
7
In the fourth chapter we again engage in validation of LB model prediction of pore-scale fluid
distribution by comparison to experimental results. In this investigation we image and simulate
fracture-matrix fluid transfer at the pore-scale in a surfactant-brine-oil fluid system. Many
formations of interest are fractured and the transfer of fluids from the high-permeability, low-
storage fractures to the low-permeability, high-storage matrix is of great interest. Predicting this
migration requires a rigorous investigation into the mechanics that determine the transfer of
fluids to and from the matrix. Understanding the physics of this process can be greatly enhanced
by investigating fluid transfer at the interface of the fracture and matrix at the pore-scale. Many
previous investigations at the pore-scale have used two-dimensional etched-glass micromodels to
investigate fracture-matrix fluid transfer (see subsection 4.2 Introduction). We present a first of
its kind three-dimensional study of fracture-matrix fluid transfer at the pore-scale. We
sequentially image fracture-matrix fluid transfer at the pore-scale in a three-dimensional
fractured porous medium using CMT. This is a novel use of three-dimensional CMT imaging,
and the first time to the author’s knowledge that fracture-matrix fluid transfer has been captured
in four dimensions at the pore-scale. We compare pore-scale fluid distributions from this
experiment to LB model results to both validate the method of simulation and gain insight into
the physics of fracture-matrix fluid transfer in a surfactant-brine-oil system.
The porous media used in these investigations is all synthetic, there are two reasons we do not
use natural porous media in our investigations. First, the CMT imaging system has limited
resolution. To compare LB and experimental results it is imperative that the model system is
representative of the porous medium imaged, if there are pore spaces beyond the resolution of
the CMT imaging system, they may affect the fluid flow in the experimental system in a way that
8
cannot be represented in the simulation. Second, natural porous media can contain
heterogeneous surfaces that we cannot differentiate using the imaging techniques at our disposal.
Small differences in the wettability of natural porous media due to heterogeneous composition or
surface precipitation may affect fluid flow in the experimental system in a way we cannot
translate to the simulation. Investigations like those presented here, help to build a foundation of
validation of pore-scale modeling techniques, and highlight the strengths and limits of the pore-
scale models used. The ultimate goal is to apply these techniques to natural porous media, and
free the determination of macroscale flow properties from the “black-box” of conventional
measurement techniques.
9
CHAPTER 2: SINGLE-PHASE LATTICE BOLTZMANN SIMULATIONS OF PORE-
SCALE FLOW IN FRACTURED PERMEABLE MEDIA
2.1 SUMMARY
The objective of this work is to investigate fracture flow characteristics at the pore-scale, and
evaluate the influence of the adjacent permeable matrix on the fracture’s permeability. We use
X-ray computed microtomography to produce three-dimensional images of a fracture in a
permeable medium. These images are processed and directly translated into lattices for
single-phase lattice Boltzmann simulations. Three flow simulations are presented for the imaged
volume, a simulation of the pore space, the fracture alone and the matrix alone. We show that
the fracture permeability increases by a factor of 15.1 due to bypassing of fracture choke points
through the matrix pore space. In addition, pore-scale matrix velocities were found to follow a
logarithmic function of the distance from the fracture. Finally, our results are compared
against previously proposed methods of estimating fracture permeability from fracture
roughness, tortuosity, aperture distribution and matrix permeability.
2.2 INTRODUCTION
Current understanding of the physical phenomena concerning fluid flow through fractures is
limited. Fluid flow in fractured formations depends on both the fracture system and the
adjacent permeable rock, also called the rock matrix. Fractures control the overall conductivity
10
of the rock while the permeable matrix provides fluid storage capacity. Studying fracture
structures, flow through fractures and fracture-matrix interactions presents significant
challenges with important applications in areas such as geo-environmental remediation, geo-
hazard mitigation, geothermal exploitation, and hydrocarbon production. In particular,
development of unconventional hydrocarbon resources, such as coalbed methane, gas shales and
tight sands, depends on our ability to represent fracture flow and fracture-matrix interactions for
the design of recovery strategies and performance predictions. The deliverability of these
unconventional hydrocarbon resources is highly controlled by the connectivity of natural
fracture network, the accessibility of fractures to the matrix storing hydrocarbons and the
fracture permeability. Of fundamental interest to the study of flow in fractured permeable media
is the determination of fracture permeability from fracture and matrix pore space geometry.
The study of flow in fractures is, for the most part, limited to fractures in rock with
impermeable fracture walls. The simplest approximation of single phase fluid flow in
fractures with impermeable walls is given by the cubic law, the analytical solution to the Navier-
Stokes equation for viscous flow through wide smooth parallel plates (no-slip walls). Given
the wall is wide enough (Width >> aperture) and the length across which a pressure head is
applied is long enough (Length > Width), the cubic law is defined as:
(
)
where is the flow rate through the fracture, w is the fracture width, is the fracture
aperture, is the dynamic viscosity, is the pressure gradient, and in the context of Darcy’s
11
equation fracture permeability ( ), is defined as, ⁄ . The simplest modification
to the cubic law is to replace the aperture, with the mean aperture ( ), this is referred to as the
alternate cubic law (ACL). This simple definition of flow through fractures does not include
important factors influencing fluid flow in natural fracture geometries, such as, fracture
roughness, asperities, anisotropies and aperture distribution. These factors have been shown to
greatly influence flow in fractures (Auradou et al. 2005, Keller et al. 1995, Konzuk and
Kueper 2004, Nolte et al. 1989, Tsang and Witherspoon 1981, Tsang and Witherspoon
1983, Witherspoon et al. 1980). As the effect of these factors became apparent, empirical and
semi-analytical modifications of the cubic law were made in the hopes of developing an
estimation of fracture permeability from fracture geometry statistics. Witherspoon et al. 1980
studied flow across smooth and rough non-contacting surfaces and introduced a modification
to the fracture permeability that included the fracture roughness,
where is a surface roughness factor ranging from 1.04 to 1.65. After the introduction of fractal
geometry, it became apparent fractures possess self-affine fractal properties regardless of the
host rock (Brown and Scholz 1985, Crandall et al. 2010, Poon et al. 1992, Schmittbuhl et al.
1995). The fractal property of fractures is often described by its Hurst exponent ( ), related to
the fractal dimension ( ) by . With this knowledge, investigators began to study
flow in synthetically developed fractures that followed fractal geometries. The study of flow
through rough fractures can be difficult to quantify and observe experimentally, thus
investigators turned to recent developments in pore-scale fluid flow models, and computational
12
fluid dynamics (CFD) to numerically simulate flows in complex pore spaces. A variety of pore
scale models have been developed, here we focus on lattice Boltzmann (LB) methods
(McNamara and Zanetti 1988), and to a lesser extent its forbearer, lattice gas cellular automata
(LGCA) (Frisch et al. 1986) methods.
Lattice techniques hold great promise in their application to the study of flow in permeable
media; they are capable of simulating flow in any complex media – given a lattice of adequate
resolution is used – as well as being easily used in parallel processing. LB and LGCA methods
have been used to study single-phase flow in both permeable media (e.g., Boek and Venturoli
2010, Ferreol and Rothman 1995, Genabeek and Rothman 1996, Koponen et al. 1997, Maier et
al. 1998, Manz et al. 1999, Pan et al. 2004) and fractures (e.g., Auradou et al. 2005, Boutt et
al. 2006, Drazer and Koplik 2002, Eker and Akin 2006, Kim et al. 2003, Madadi and Sahimi
2003, Nazridoust et al. 2006, Stockman et al. 1997, Zhang and Kang 2004, Zhang et al. 1996).
Zhang et al. 1996 studied single-phase flow through synthesised self-affine fractal fracture
using LCGA; they proposed an empirical estimation of fracture permeability as a
function of mean aperture:
with a value of of 2.67, 3.15 and >4 for fractures with a Hurst exponent of 0.8, 0.3 and -0.5.
A value of 2 would be expected for flow between parallel plates, the value of is expected to
increase monotonically with roughness. Eker and Akin 2006 also investigated single phase flow
in synthesized self-affine fractal fractures with Hurst exponents ranging from -0.21 to -0.50
13
using LB methods, and following the work of Zhang et al. 1996 proposed to be a linear
function of fractal dimension,
These empirical predictions only take the fracture roughness into consideration, other factors
governing fracture geometry such as asperities and aperture distribution must also be included in
a complete prediction for the estimation of fracture permeability. Zimmerman and Bodvarsson
1996 attempted to relate the effective hydraulic fracture aperture ( ) to the statistics of aperture
distribution. The effective hydraulic fracture aperture is a measured property of flow through a
fracture, it is the equivalent aperture for flow between smooth plates for a measured flow rate
(Q),
(
)
The relation between ( ) and the statistics of aperture distribution was derived from the
application of Reynold’s lubrication equation. As they noted, asperities can be viewed as a part
of the aperture distribution, however, it is convenient to treat the contact areas or asperities
separately from the aperture distribution. The is proportional to the fracture permeability
(
⁄ ), thus their prediction for the relation of the to fracture statistics is also a
definition of fracture permeability:
(
)
14
is the standard deviation of the fracture aperture distribution and C is the fractional
contact between fracture walls. The applicability of numerical results from synthesised
fractures has merit, but validation of these methods requires studying real fracture geometries.
Imaging of fractures has become simplified by the employment of sophisticated three
dimensional imaging techniques, such as X-ray computer microtomography (CMT).
The use of X-ray CMT in the study of flow in permeable media and fractures has become
widespread, offering a non-invasive method to image complex pore geometries and fractures in
three dimensions. High resolution images of pore space are often used in conjunction with pore-
scale models to investigate pore scale flow mechanisms that are difficult to measure and
observe. A large variety of permeable media has been imaged using CMT, including macro
permeable clays (Heijs et al. 1995) soils (Perret et al. 1999), bead packs (Culligan et al.
2006), unconsolidated sands (Brusseau et al. 2006, Wildenschild et al. 2005), sandstones
(Auzerais et al. 1996, Nakashima et al. 2004) and limestones (Okabe and Blunt 2004). Many
fractures have also been studied using CMT, including single brittle fractures in Berea
sandstone (Karpyn et al. 2007), single planar artificial fractures of increasing aperture in
limestone (Ketcham et al. 2010), lab-scale hydraulic fractures in limestone (Renard et al.
2009), natural fractures in limestone (Wennberg et al. 2009), and single brittle fractures in
granite (Johns et al. 1993, Keller et al. 1999), to name a few.
Nazridoust et al. 2006 studied flow in a single tensile fracture in Berea sandstone imaged by
Karpyn et al. 2007. They proposed studying flow in fractures in a manner analogous to the
15
study of pressure losses due to friction in pipes and ducts. The resistance to flow is measured as
a friction factor, analogous to fracture permeability. For flow between smooth parallel plates
the friction factor (f) is defined as,
They introduced a new friction factor dependant on fracture tortuosity and aperture distribution,
and fitted to CFD results for flows of varying velocity,
( )
where ⁄
is the Reynolds fracture number and is the kinematic viscosity; this can be
related to the fracture permeability by the pressure drop,
(
)
where is the corrected mean fracture aperture, defined as, , is the dynamic
viscosity, and is the tortuosity,
16
where is the length of the tortuous actual path of flow and is the length of the fracture,
arriving at a semi-analytical estimation of fracture permeability,
( )
Crandall et al. 2010 also studied flow in a single fracture in Berea sandstone imaged by CMT
using CFD. They used the same image of a fracture as Nazridoust et al. 2006, and attempted to
include matrix effects on flow by making the CFD grid adjacent to the fracture permeable.
Following the work of Nazridoust et al. 2006 they redefined the friction factor for the fracture to
include the flow through the matrix,
( )
⁄
and equivalently the fracture permeability,
( (
) ⁄ )
( )
where is the matrix permeability and is the width of the matrix. Overall they found the
friction factor decreased with increasing matrix permeability, and the fracture permeability is
directly proportional to the matrix permeability.
17
As of yet, there is no agreed upon approach for estimation of fracture permeability from
fracture geometry and the flow behavior of the rock matrix. This is a fundamental issue, and a
key building block for adequate description and simulation of fluid flow through fractured
reservoirs. To the author’s knowledge, this is the first pore-scale study of flow in fractured
permeable media in which a three-dimensional rough fracture is explicitly interconnected with
the pore network of the adjacent permeable rock. We use CMT to image the fractured
permeable media at resolutions high enough to capture the solid surface curvature of the
grains; hence we are able image the entirety of the pore space including that of the matrix.
This image is translated directly into a lattice to be used by LB simulations of single phase
flow. To study the effect of the matrix on flow in the fracture, and vice versa, three simulations
are presented. One in which the entirety of the pore space is simulated (fracture in permeable
rock), one in which the matrix pore space is removed and the flow is simulated in the fracture
alone (analogous to a fracture in impermeable rock), and lastly five simulations are run on
portions of the pore space that do not contain any part of the fracture and averaged to determine
the permeability of the matrix alone. The results of these simulations are compared to the
empirical and semi-analytical estimations of fracture permeability discussed above.
2.3 MATERIALS AND METHODS
2.3.1 X-ray Microtomography Scanner
The pore space of the fractured permeable medium was imaged using X-ray (CMT).
Scanning took place at Penn State’s Center for Quantitative X-ray Imaging (CQI). X-ray CMT
imaging is a non-invasive imaging technique that produces a three-dimensional grid of voxels
18
– much like pixels for a two-dimensional image – each of which contain an integer value CMT
registration number. The scan of the core in the scan presented here is composed of 800 slices,
each of these slices being 1024 x 1024 voxels. Each slice of the three-dimensional grid of CMT
registration numbers is produced by a mathematical reconstruction of 1440 two-dimensional
images taken perpendicular to the axis of the core over the course of a full 360° core rotation.
CMT registration numbers produced by the mathematical reconstruction are dependent on the
relative X-ray attenuation, or X-ray opaqueness, of the material occupying each voxel. X-ray
attenuation by a material is a function of the density and apparent atomic number of the
material. These images are segmented to identify the material occupying each voxel, to be
further elaborated on in the image processing section. The half inch diameter cores in this
investigation result in CMT scans with a voxel resolution of 13.606 x 13.606 x 12.405 μm3, the
latter represents the thickness of the slice. However, to directly translate the CMT image of the
pore space to a LB simulation requires a cubic lattice, thus the images are treated as having
cubic voxels with side lengths of 13.606 μm.
2.3.2 Fractured Permeable Media
Porous polyethylene rods manufactured by Small Parts Inc., were used as the permeable
medium. These rods are 0.127 cm in diameter and formed by sintering medium/coarse,
angular/subangular polyethylene grains into rod form. This creates a permeable structure similar
to a medium/coarse, well-sorted unconsolidated sand or sandstone. A synthetic permeable
medium is used in lieu of a natural permeable medium to ensure the pore space of the matrix
is resolvable. Polyethylene is a malleable material, thus a double- sided ‘log-splitting’
19
approach was used to create a fracture in the porous polyethylene rods (Figure 2.1), instead of
more conventional techniques (i.e., modified Brazilian test) used to induce fractures on natural
rock. Two splitting tools are driven along the desired fracture plane, not down the center of the
rod, but along the sides. This results in a rough fracture propagating between the splitting tools.
The outer portions of the fracture that were in contact with the spitting tool are cropped from
the image and not used in the LB simulation. This crop results in a total pore space lattice
volume of 520 x 520 x 800 voxel3.
Figure 2.1. Illustration of the double-sided ‘log-splitting’ approach used to propagate a rough fracture
in the porous polyethylene rod.
20
2.3.3 Image processing
The raw, cropped CMT images are processed and segmented for analysis. The raw images
are first filtered using a 3 x 3 x 3 kernel median filter to remove salt-and-pepper type noise. To
quantify measurements of the pore space, the filtered images are segmented into the phases
present, pore and solid. The contrast between the solid phase and pore space is sufficient for
application of simple thresholding. Simple thresholding segments the image by declaring all
voxels below a threshold value as belonging to one population and all those above to the other.
The threshold value is determined by fitting Gaussian curves to a frequency plot of the CT
registration numbers (Figure 2.2) (Mees et al. 2003). The frequency plot contains three peaks,
the first is associated with the pore space, and the second two are associated with the solid,
with respect to increasing CT registration number. The intercept of the two Gaussian curves
for the respective populations is used as the threshold value, here at a CT registration number
of 17323. The two peaks of the solid represent the two different polyethylene particulates used
in the construction of the permeable rod.
21
Figure 2.2. Demonstration of simple thresholding image segmentation, at the left is a raw CMT image
slice, the raw data is cropped (shown as dashed lines) to remove portions of fracture that
were in contact with the splitting tool. The middle plot shows the frequency of CT
registration numbers, and Gaussian fits to the solid and pore phases in the image, the
intercept of these Gaussian fits is used to segment the image. At the right is a segmented
image, in which grey represents the pore space and white represents the solid or grain.
The segmented images contain some sporadic solid voxels misidentified by the segmentation
procedure in the pore space. To remove these sporadic voxels, the segmented 3D image of the
pore space undergoes a percolation. A percolation finds all like voxels in contact. First a
percolation is conducted on the solid voxels. The contact points between individual grains are
beyond the resolution of the CMT images, thus this percolation finds all the connected solid
voxels or ‘true’ solid voxels. The ‘negative’ of the connected solid voxels is then considered
the pore space. To remove any isolated pores from the image of the pore space a second
percolation is carried out on the negative of the connected solid voxels. The second percolation
finds all pore space voxels in contact, ensuring the pore space to be simulated does not
contain any unnecessary, non-interacting (isolated pores) pore space. This prepares the pore
space for direct interpretation into a lattice for use by LB simulations.
22
2.3.4 Fracture Pore Space Identification
The identification of the fracture is non-trivial; unlike previous investigations of fractures, the
distinction between the fracture and matrix is not readily apparent. To identify the pore space
associated with the fracture pore space – also simply referred to as the fracture – a sequence of
erosions, expansions and percolations were performed on the segmented images of the pore
space. A cycle of erosion substitutes all the pore space edge voxels with solid voxels, thus
removing a single layer of voxels from the pore space volume. Edge voxels are defined as those
with voxel face contact between the two phases present, namely solid and pore. Expansion is the
opposite of erosion.
The first step in the identification of the fracture pore space is the isolation of the locality of
the fracture. If the fracture aperture is generally greater than the greatest distance between
grains in the pores of the matrix, the fracture can be located by eroding the pore space until a
single pore is remaining. This point was reached after eight cycles of erosion for the pore space
presented here. This single pore only captures the core of the fracture pore space, to identify
the entirety of the fracture this pore must be dilated out. Each cycle of erosion results in a loss
of detail, to minimize this effect, the core of the fracture pore space is reversed-out by two
cycles of erosion. It is important to note that this is not two cycles of expansion. We have the
images preceding each erosion; we can retain loss of detail to these two cycles of erosions. The
dilation cannot continue in this reversing-out manner. The next reversed-out erosion opens up
the pore throats connecting to the matrix pore space, to continue in this manner would
erroneously begin to identify the matrix pore space as fracture pore space. Instead, at this point
23
we perform cycles of expansion until the fracture pore space contacts the solid surface. After
seven cycles of expansion the dilated fracture pore space overlaps solid voxels, thus six cycles
of expansion defines the fracture. This sequence is summarized in Figure 2.3.
Figure 2.3. Example slices from the fracture space identification procedure. The left image shows the pore space (light grey) after eight cycles of erosion, at which point all of the matrix pore space is removed, this locates the fracture (dark grey). The middle image shows the pore space in contact with the fracture space after two cycles of reversed-out erosions, the erosion cycle one short of the fracture pore space becoming connected with the matrix pore space; this identifies the starting point for expanding the fracture pore space out. The right image shows the fracture pore space after six cycles of expansion, the expansion cycle one short of the fracture pore space making contact with the grains or solid, this defines the fracture.
The flow in the matrix adjacent to the fracture is influenced by the high rate of flow in the
fracture, and differs significantly from the flow in the rest of the matrix. To investigate the
extent of this effect, the flow in the pore space adjacent to the fracture pore space is analyzed.
To measure this flow, the voxels adjacent to the fracture must be identified as a function of
distance from the fracture. To do so, the fracture pore space is expanded to identify voxels at
equal spatial distance from the fracture throughout the entirety of the pore space. These
expansions identify voxels as a function of distance from the fracture, and provide a measure of
flow velocity as a function of distance from the fracture.
24
2.3.5 Surface Area and Volume
The segmented images are used to measure surface areas and volumes of grains and pore space.
Volumes are calculated as the sum of all solid and pore voxels ( ). The porosity of the
matrix ( ) is defined as, ( )⁄ and the fracture porosity ( ) is defined as,
⁄ , where is the facture volume. The surface areas are determined using the image
processing software Avizo Fire 6.3. This software uses a modified marching cubes algorithm to
“wrap” a tetrahedral surface around the segmented image data for the solid phase. This surface
quantifies the solid surface area ( ). This is normalized to the bulk permeable volume of the
matrix to find the specific solid surface area ( ) defined as, ⁄ , where is the
bulk volume of the matrix defined as, .
2.3.6 Aperture and Fracture Roughness
It is assumed that the slices of the CT scan run orthogonal to the vertical plane of the
fracture; the measurement of apertures and fracture roughness are taken per slice of the scan.
There are two manners in which aperture can be measured, the perpendicular aperture (hp)
and the vertical aperture (h) (Figure 2.4). The apertures measured here are vertical apertures,
albeit the perpendicular aperture is not subject to fracture orientation, as the vertical aperture
is, considering the velocities reported here follow a Cartesian coordinate system defined by
the images, the vertical aperture was considered a more appropriate method of measurement.
The aperture distribution is shown in figure 2.5; the small initial peak is a result of the matrix
pore aperture distribution impacting the aperture distribution at fracture boundaries. The
25
fracture boundaries are defined by the matrix pore space, this will define the lower end of the
fracture aperture distribution. This effect on the aperture distribution becomes negligible as the
fracture aperture to matrix pore aperture ratio increases, but for our fracture the effect is
readily apparent. The distribution is overall log-normal, consistent with fractures in natural
permeable media (Karpyn et al. 2007, Keller et al. 1999, Ketcham et al. 2010). In this study,
fracture apertures below the maximum pore aperture found in the matrix (~0.33 mm) are
considered fracture contact areas. Pore apertures are measured in the same manner as fracture
apertures. The definition of contact area used here is,
where is the number of fracture apertures measured that are less than the maximum vertical
aperture of the matrix pores and is the total number of measured fracture apertures.
26
Figure 2.4. Illustration of the measure of the vertical aperture of the fracture (h), the perpendicular
aperture (hp), the fracture profile distances from edge of image (Bj, Bj+s) and the
bandwidth window (s). Note, The light gray is the matrix pore space, and the dark gray is
the fracture.
Figure 2.5. The fracture aperture distribution, the small initial peak is a result of the matrix pore
aperture distribution impacting the aperture distribution of the fracture at fracture
boundaries. The distribution is overall log-normal, consistent with fractures in natural
permeable media.
0
0.005
0.01
0.015
0.02
0.025
0.03
0 0.5 1 1.5 2 2.5
Fre
qu
ency
Fracture Aperture (h) (mm)
27
The fracture profile is defined by the fracture pore space in individual two-dimensional slices of
the image. The roughness of the fracture profiles of individual slices is characterized by the
Hurst exponent ( ) also referred to as the roughness exponent or self-affine exponent. We use
the variable bandwidth method to determine the Hurst exponent ( ) of the fracture profile for
individual slices. We assume our fracture is well described by fractional Brownian motion, such
that a scaling relationship exists between the size of the sampling window ( ) and the standard
deviation of the fracture profile ( ):
The Hurst exponent can be determined from the slope of a log-log plot of standard deviation
( ) vs. sampling window size ( ). The standard deviation is calculated according to a
modified form of (Dougan et al. 2000):
√
∑ ( )
where is the number of datapoints contained in the dataset, j is the index of the fracture profile
height ( ) (Figure 2.4), and is a count of the number of samples that contain an asperity,
wherein a measurement of is not possible. The assumption that a fracture is well described by
fractional Brownian motion and the variable bandwidth method is an appropriate method for
the measurement of the Hurst exponent has been found to be generally true for fractures in
natural rocks by previous investigators (Crandall et al. 2010, Renard et al. 2009). The
tortuosity ( ), previously defined, requires a measurement of the mean length of a streamline
flowing through the fracture ( ), this is found using the programme Paraview. The stream
tracer function seeds a particle at the inlet and progress the particle in the velocity field. The
28
mean length of a streamline is determined from 50 streamlines seeded at regular intervals along
the fracture inlet.
2.3.7 Lattice Boltzmann Simulations
Most petroleum engineers are accustomed to the ‘top-down’ methods of simulating fluid flow,
in which partial differential equations defining fluid flow are discretized using finite-
difference, finite-element and finite-volume techniques. The LB method has its roots in
LCGA methods, which simulate flow using a ‘bottom-up’ approach, in which, fluids are
simulated by a swarm of particles moving along discrete directions on a lattice. The streaming
and collision of these particles are governed by a set of rules such that the time-averaged motion
of the particles is consistent with the Navier-Stokes equations. The LCGA method is subject to
numerical instability, to avoid instability the LB method replaces these particles with averages
referred to as particle distribution functions. For a review of LB methods, see Succi 2001 and
Sukop and Thorne 2006. A detailed description of the LB model is given in chapter 3.
The lattice of LB methods is well suited for direct pore space lattice construction from CMT
images. Each voxel is directly translated to a single node in the Boltzmann lattice. The pore
voxels are represented by nodes on which ‘particles’, represented by particle distribution
functions, can freely move upon, the solid voxels are represented by non-interacting nodes, and
the solid boundary voxels are represented by bounce-back nodes. Bounce-back nodes are
essentially algorithm devices that return any exiting ‘particles’ back into the pore lattice domain
with the opposite momentum. Here, we use the half-way bounce-back condition on the nodes
at the solid surface, this ensures a microscopic, and hence, a macroscopic no-slip condition
29
halfway between the pore nodes and the bounce-back nodes (He et al., 1997). A density is
associated with each node, this gives the system mass, and as a result pressure. The LB
simulation presented here simulates a quasi-incompressible fluid, wherein the lattice node
density ( ) is directly proportional to the lattice pressure ( ), following from the ideal
gas equation of state. The ratio of the determining forces involved in flow, namely the
inertial and viscous forces, are conventionally defined by Reynolds numbers. The physical and
lattice Reynolds numbers ( ) are defined as,
and,
where is the reference velocity, taken as the average velocity of the flow to be
simulated, is the reference length taken as the length of the simulated volume, is the
physical dynamic viscosity, is the maximum lattice velocity, is the lattice resolution,
is the lattice kinematic viscosity and is the lattice density. Here, a water flow with a
is simulated, a stokes flow. The lattice kinematic viscosity was set to 1/6
(generally accepted to be the most numerically stable lattice kinematic viscosity), and the
resolution was set equal to 520, corresponding to CMT image resolution. To impose a flow on
the lattice the pressure at the inlet and outlet are set to constant values. To set the pressure at
the inlet and outlet, the densities of the nodes at these boundaries are maintained throughout the
simulation. Since the density is related to the pressure by the equation of state, this ensures the
pressures will remain constant at these boundaries (Zou and He 1997). All other boundaries are
treated as periodic, as in, when a particle exits a boundary it directly enters the boundary exactly
30
opposite of where it exited. The ‘particles’, driven by the pressure (lattice density) gradient
between the inlet and outlet, undergo a single stream and collision step during each iteration of
the LB simulation until steady state flow is reached. Each lattice site has an associated
macroscopic velocity vector with velocity components reported in the three Cartesian
directions ( , where ). Steady-state flow is reached when the mean
lattice velocity magnitude converges.
The open source LB code Palabos v0.7r3 was used to perform the LB simulations. Simulations
were carried out on Penn State’s HPC lion-X cluster. The LB simulations used a D3Q19
lattice with Bhatnagar-Gross-Krook (BGK) dynamics. Pan et al. (2006) investigated the
accuracy of LB simulations of flow in sphere packs using multiple-relaxation-time (MRT),
single-relaxation-time (SRT) and BGK dynamics. They found for a lattice viscosity of 1/6
(used here) the BGK model was as accurate as the more computationally expensive MRT
model. To validate the simulation, a Poiseuille flow was simulated through a narrow
cylindrical channel with a radius ( ) of 0.33 mm. The analytical solution to the velocity profile
for Poiseuille flow is,
The velocity profiles for the analytical solution and LB simulation results for Poiseuille flow
are shown in figure 2.6.
31
Figure 2.6. Poiseuille flow through a channel with a radius a = 0.33(mm) velocity profile, comparison
of LB results and analytical solution. We observe that the LB simulation is closely
approximating flow in the small channel.
2.4 RESULTS
2.4.1 Fracture and Matrix Pore-Space Geometry
A summary of all measured properties of the pore space are summarised in Table 2.1. The
matrix porosity (φm) was found to be ~0.32, significantly lower than the values of 0.40–0.50
reported by Prodanovic et al. 2006, which suggests inconsistency in the manufacturing
process of these rods. The fracture porosity (φf) of the pore space is 0.243; the volume of the
fracture is significant, and the fracture is expected to dominate flow. The roughness of the
fracture was measured for 16 slices evenly distributed along the vertical extent of the fracture
(every 0.652 mm). The Hurst exponents were determined for each of these slices; figure 2.7
shows a log-log plot of deviation vs. window size for some of these slices. The Hurst exponent
ranges from 0.51 to 0.69, with a mean Hurst exponent (ε) of 0.612; this is a nominal value of
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
-0.33 -0.27 -0.21 -0.15 -0.09 -0.03 0.03 0.09 0.15 0.21 0.27 0.33
u (
m/s
)
x (m)
LB sim
Analytical
32
roughness for permeable media. The mean Hurst exponent of 0.612 is similar to Hurst exponents
measured for tensile fractures in Berea sandstone (0.38 to 0.45) (Crandall et al. 2010), and single
hydraulically induced fractures in annular cores of Fountainbleu sandstone (0.4 to 0.5) (Renard
et al. 2009). The mean tortuosity was determined from the tortuosity measurement of 50
streamlines, and found to be 0.52. This is a relatively high tortuosity, the tortuosities measured
for a single tensile fractures in Berea sandstone ranged from 0.29 to 0.30 (Nazridoust et al.
2006) and 0.034 to 0.064 (Crandall et al. 2010). The differences in tortuosity measured by
Nazridoust et al. 2006 and Crandall et al. 2010 are a result of different methods for determining
actual length of streamlines. Crandall et al. 2010 determined the actual length from seeded
streamlines in the same manner used here, Nazridoust et al. 2006 assumed the fracture profile
was representative of the tortuosity.
Figure 2.7. Determination of Hurst coefficient using the variable bandwidth method, the data points
are shown for four slices, the linear function shown (ε = 0.612) was determined from 16
equidistant slices spanning the range of the simulated pore volume. The Hurst exponent
is the slope of the fitted linear function shown.
0
0.5
1
1.5
2
2.5
0 0.5 1 1.5 2 2.5
log
(σfp
)
log (s)
Slice 200
33
2.4.2 LB Simulation Results
The objective of running the three LB simulations is to study flow in a fractured permeable
medium, a fractured impermeable medium and an unfractured permeable medium, three LB
simulation results are presented respectively, one of the entire pore space (fractured permeable
media), one of the fracture alone (fractured impermeable media), and an average of five
simulations of the matrix alone (permeable media). The velocities are normalized to the mean x-
velocity ( ), also the Darcy velocity, for each voxel with index ,
| |
| |
Table 2.1 Pore Space Geometry and Fracture Characteristics
34
A three-dimensional velocity glyph visualization is presented in figure 2.8, this visualization
illustrates the channeling of flow in the fractures for both simulations. In general, the flow
through the fracture is significantly increased by the permeable matrix. Also, the flow in the
fracture-alone simulation shows a single readily apparent high velocity choke point, while the
flow in the fracture – aided by the permeable matrix – is more widely dispersed over the fracture.
The matrix only LB simulations were run on 200 x 200 x 400 voxel portions of the pore space
that did not contain any part of the fracture; the average of five of these simulations was used to
determine matrix flow properties. The permeability was determined from the mean x-, or Darcy
velocities,
The flow in the fracture for the all-pore space simulation dominates the flow, with 88% of the
flow occurring in the fracture with a velocity 4.24 times the mean x-velocity. The permeability
of the fracture with a permeable matrix is 15.1 times greater than the fracture alone. The
permeability of the matrix in the all-pore space simulation is also increased by the proximity of
the fracture by 2.14 times. However, if the simulation included a greater extent of the matrix –
as in a greater distance between fractures – this effect would dissipate. As in, if a greater extent
of the matrix was given this apparent increase in permeability would drop to unity. Keep in
mind the periodic boundaries imply that the flow simulated is the equivalent of imposing a
distance between ‘fractures’ equal to the dimension of the simulated volume orthogonal to flow.
35
Figure 2.8. Normalized x-velocity glyph visualization of the fracture-alone (above) and the all-pore
(below) simulations; the glyphs are larger and farther to the red side of the spectrum for
greater velocities. The maximum normalized velocity shown in the glyphs is .
To explore the flow adjacent to the fracture, the velocities are plotted as a function of distance
36
from the fracture in figure 2.9. The velocities appear to decrease logarithmically as a function of
distance from the fracture following the form,
( ) ( )
where is the velocity in the x-, y- or z-direction and is the distance from the fracture. This
is a purely empirical fitting of data. The high velocities in the matrix immediately adjacent to
the fracture persist for only a short distance, a distance on the order of a single grain diameter.
The previously mentioned apparent increase in the matrix permeability in the all-pore simulation
is a result of these high velocities.
37
Figure 2.9. Mean normalized velocities ( ) as a function of distance from the fracture ( ),
and logarithmic function fits.
2.4.3 Fracture Permeability Estimation
A summary of the LB simulation results and fracture permeability estimates from previously
mentioned investigations are presented in Table 2.2. The fracture permeability determined from
LB simulation results for the all-pore simulation and the fracture-alone simulation are 31.9e-9
and 2.11e-10 m2, respectively. The ACL determines an estimate of fracture permeability from
the mean aperture, ignoring all other aspects of the fracture geometry and matrix permeability.
38
The ACL overestimates the fracture permeability in the all-pore simulation and the fracture-
alone simulation by 11.1 and 168 times, respectively. The ACL is not expected to give a good
estimate of permeability for our fracture. The ACL estimate decreases in accuracy with the
greater influence of unaccounted geometry determining flow in the fracture, such as the aperture
distribution, tortuosity and fracture roughness. If the fracture wall separation was significantly
increased, the tortuosity would decrease and the roughness would interfere with a smaller
percentage of the flow. Witherspoon et al. 1980 fit an empirical prediction of fracture
permeability to measurements from flow experiments. They determined the fracture permeability
from the mean aperture and a surface roughness factor, here set to the highest value
( ) of the range of values suggested. Witherspoon et al. 1980 overestimates the
fracture permeability in the all-pore simulation and the fracture-alone simulation by 6.71 and
102 times, respectively. Witherspoon et al. 1980 was determined empirically from flow
experiments, thus it is subject to the nature of the fractures used to fit it. If these fractures were
not as rough and tortuous as ours we would expect these characteristics to have less of an
effect on their flow experiments, and more closely approximated by the ACL. Zhang et al.
1996 determines the fracture permeability from the mean aperture and a power factor, here set to
a value ( ) linearly interpolated from the values determined for fractures with a
Hurst exponent of 0.80 ( ) and 0.30 ( ). This estimate, like the estimate of
Witherspoon et al. 1980, was determined empirically, and unlike Witherspoon et al. 1980,
from LCGA simulations on synthetic fractures, instead of flow measurements. Zhang et al. 1996
underestimates the fracture permeability in the all-pore simulation by 3.85 times, and
overestimates fracture permeability in the fracture-alone simulation by 3.94 times. The
simulated flow in Zhang et al. 1996 took place in a fracture with impermeable walls, thus their
39
estimate of fracture permeability is most appropriately tested by the fracture-alone LB
simulation, and the overestimate is a result of the differences in the fractures simulated by them
and the fracture imaged here. The fractures simulated by Zhang et al. 1996 had no induced
tortuosity, nor varying aperture. Tortuosity and varying aperture within a fracture will increase
the resistance to flow, without consideration of these fracture characteristics we would expect an
overestimation of permeability. However, considering the simplicity of their proposed estimation
of fracture permeability it is a vast improvement over ACL and Witherspoon et al. 1980. Eker
and Akin 2006 proposed the β value of the Zhang et al. 1996 is a linear function of the fractal
dimension, as we see in table 2.2 their estimate is by far the most inaccurate. This is not
surprising, considering this linear function was fit to numerical flow simulations on
synthesised fractures in impermeable rocks with abnormally high fractal dimensions
( ). Their linear function for does not fit the results of Zhang et al. 1996,
nor ours, suggesting a linear function of determined from the fractal dimension alone will
not accurately estimate fracture permeability outside from those for which it was fit.
The fracture permeability estimate of Zimmerman and Bodvarsson 1996, unlike the above
estimations of fracture permeability, is semi-analytically derived. Their estimate is only
applicable to fractures in impermeable rock, and underestimates the fracture permeability for
the fracture-alone LB simulation by approximately 48%. The underestimation indicates that
characterizing the fracture by the mean and standard deviation of the fracture aperture alone does
not provide enough information to accurately predict the fracture permeability.
Nazridoust et al. 2006 derived a semi-analytical prediction for fracture permeability that assumes
40
the flow in the fracture can be treated in much the same manner as flow in pipes.
Conventionally in pipe and duct flow, pressure losses are a function of friction factors, which
are dependent on flow rate. The cubic law, which was the basis for previously evaluated
fracture permeability estimations, assumes Stoke’s or laminar flow. The smooth, parallel plate
solution used by Nazridoust et al. 2006 includes inertial effects, thus the resistance to flow is
dependent on flow rate. The fracture permeability estimation of Nazridoust et al. 2006 is
dependent on the flow rate, fracture aperture mean, standard deviation and tortuosity. They
fitted their function for the friction factor to CFD simulations performed on two-dimensional
CMT images of fractures in Berea sandstone. These two-dimensional fracture profiles
contained no asperities. Their derivation does not account for permeable fracture walls, thus
only the fracture-alone LB simulation is comparable. Nazridoust et al. 2006 overestimate the
permeability in the fracture-alone LB simulation by 1.33 times, but it is the best estimate of those
tested. This overestimation can be attributed to the significant amount of asperities in our
fracture, which are not accounted for by Nazridoust et al. 2006, and in particular the effect these
asperities have on a three-dimensional fracture.
41
Table 2.2. Summary of LB Permeability Measurements and Fracture Permeability Estimates
Notes: Summary of LB results ( ) and fracture permeability
estimates, where - Alternate Cubic Law, - (Witherspoon et al., 1980), -
(X. Zhang et al., 1996), -(Eker and Akin, 2006), - (Zimmerman and Bodvarsson,
1996), - (Nazridoust et al., 2006) and - (Crandall et al., 2010); is the only
estimate that considers matrix permeability, and the only estimate appropriate for the all-
pore LB simulation.
Crandall et al. 2010 built upon the work of Nazridoust et al. 2006 by adding the matrix
permeability to the prediction of fracture permeability. They used the same CMT images of a
fracture in Berea sandstone as Nazridoust et al. 2006 and performed two-dimensional CFD
simulations with a homogenous, isotropic fracture wall with permeability ranging from 2E-16 to
2E-12 mm2 (0.2 to 2000 mD). Since their derivation accounts for a permeable fracture wall,
42
only the all-pore LB simulation is comparable. Crandall et al. 2010 underestimates the fracture
permeability in the all-pore simulation by 6.72 times. Although the prediction for the LB
simulation of the fracture alone (Nazridoust et al. 2006) is off by a relatively small percentage
(33%), the prediction for the all-pore space fracture permeability (Crandall et al. 2010) is off by
672%. Here are some explanations for this large difference. First, Crandall et al. 2010
validated their derivation of fracture permeability from CFD results for matrix permeability well
below our matrix permeability of 4.92E-11mm2 (49.7 D). Also, there is a fundamental
difference in the nature of the matrix for our simulations and theirs. Our simulation directly
simulates an actual matrix pore space, this allows for fluid to channel in and out of the fracture
through the matrix, helping to bypass fracture choke points. This will greatly increase the
flow in the fracture, as we observe from our results. The matrix in Crandall et al. 2010 does
not allow for this mechanism of bypass. In CFD the rock matrix is divided up into a fine-
grid of finite-volumes, and each of these volumes are homogenous. This form of simulation
assumes the finite-volumes are on the order of the representative elementary volumes (REV) of
flow for the permeable medium being simulated. The fracture pore space does not have a REV,
but the matrix does, and it’s REV is dependent on the grain size of the permeable media. The
REVs of sandstone, and the matrix simulated in the all-pore LB simulation presented here, are
many orders of magnitude larger than the finite volumes used by Crandall et al. 2010.
Crandall et al. 2010 use of CFD is arguably a fair representation of flow in fractured
permeable media with much smaller REVs.
43
2.5 CONCLUSIONS
A single rough fracture was propagated along the axis of a porous polyethylene rod; this
fractured permeable medium was imaged using CMT. The fracture pore space identification
process, consisting of a succession of erosions, percolations and expansions on the segmented
processed CMT images of the pore space, presented here appears to capture the fracture well.
The fracture was found to have a log-normal aperture distribution and a Hurst exponent of
0.612, similar to tensile fractures in Berea and Fountainbleu sandstone. Three LB simulations
were performed on the imaged pore space, one of all-pore space imaged (fracture in permeable
media), one of the fracture alone (fracture with impermeable walls), and an average of five
simulations on portions of the pore space that did not contain any part of the fracture to be
representative of flow in the matrix only. The fracture permeability was increased by a factor of
15.1 by the presence of the highly permeable matrix, and the matrix permeability was increased
by a factor of 2.14. However, this increase in the matrix permeability is a result of the presence
of the fracture, and is dependent on fracture spacing. The x-, y-, and z-velocities were found to
be a logarithmic function of the distance from the fracture. Although developing a generalised
form of this equation was beyond the scope of this paper.
Fracture permeability estimates from a variety of previous investigations were compared to our
LB simulation results; all of these estimates assume an impermeable fracture wall, except for
Crandall et al. 2010. The ACL does not account for fracture roughness, tortuosity, asperities
and aperture distribution, and has been widely accepted to grossly overestimate the fracture
permeability. The ACL overestimated the fracture permeability by two orders of magnitude.
44
Zhang et al. 1996 suggest permeability is a power function of mean aperture and a fitted power
factor β. Although this will provide a good estimate of permeability if a value of β is found
that fits the fracture flow measurements, it has no power of prediction without a function
for β dependent on fracture characteristics. Eker and Akin 2006 proposed a linear function for β
dependent on the fractal dimension of the fracture wall roughness; however their prediction for β
did not fit the results of Zhang et al. 1996, nor ours. A linear function for β dependent only
on the fracture wall fractal dimension is inadequate. Zimmerman and Bodvarsson 1996 derived
a semi-analytical prediction of fracture permeability dependent on the mean fracture aperture,
fracture aperture deviation and fractional contact area from the Reynold’s lubrication
equation. The permeability predicted by this function underestimated fracture permeability
approximately 48%, indicating accurate prediction of fracture permeability requires more
information than the mean and standard deviation of the fracture aperture provide. Nazridoust et
al. 2006 derived a semi-analytical function of fracture permeability dependent on mean
fracture aperture, fracture aperture standard deviation and flow rate from the parallel-plate flow
solution for flow with significant inertial effects. Their prediction of fracture permeability
provided the closest match to LB simulations, overestimating the fracture permeability by a mere
33%. The slight overestimation can be attributed to neglecting the effect of asperities.
Crandall et al. 2010 built upon the work of Nazridoust et al. 2006 to include the effect of matrix
permeability on flow in fractured permeable media. Their fracture permeability prediction
underestimated the fracture permeability in the LB simulation of the fractured permeable
medium by a factor of 6.72 times. As discussed, the simulations used in Crandall et al. 2010
assumed the flow in the matrix can be represented by homogenous and isotropic finite
volume elements. However, they used REVs that are much smaller than the REVs of
45
sandstones and the permeable media used in our simulations. Our permeable medium is
simulated at the pore-scale without any assumptions made regarding the matrix pore space.
This allows for bypassing of fracture choke points through the pore spaces of the matrix adjacent
to the fracture. This dynamic is not possible in CFD simulations. It is for these reasons flow in
the fracture was so much greater in our simulations.
LB methods, and other pore-scale models are tools that give us insight into fluid flow through
permeable media at scales that are very difficult to observe; particularly in the study of
fractured permeable media, for which measuring the flow in the fracture alone would prove
very difficult. It is for this reason validating our results would require more indirect methods,
such as measuring the mean permeability of the entire matrix-fracture system and comparing
this permeability to simulation results. However, this investigation was not meant to be an
evaluation of the accuracy of LB methods, the evaluation of the accuracy of single-phase LB
models has been addressed by many of the previous researchers mentioned. A comprehensive
prediction of fracture permeability from fracture and matrix characteristics for fractures in
permeable rock that includes all determining factors, including the flow behavior of the matrix,
remains elusive. However, pore-scale studies such as this one provide a tool to help minimize
unrealistic assumptions made in more simplistic representations of flow in fractured
permeable media.
46
2.6 REFERENCES
Auradou, H., Drazer, G., Hulin, J.P. and Koplik, J. (2005) ‘Permeability anisotropy induced by
the shear displacement of rough fracture walls’, Water Resources Research, Vol. 41, No. 9,
W09423, DOI: 10.1029/2005WR003938.
Auzerais, F.M., Dunsmuir, J., Ferreol, B.B., Martys, N., Olson, J., Ramakrishnan, T.S.,
Rothman, D.H. and Schwartz, L.M. (1996) ‘Transport in sandstone: a study based on three
dimensional microtomography’, Geophysical Research Letters, Vol. 23, No. 7, pp.705–708.
Boek, E. S. and Venturoli, M. (2010) ‘Lattice-Boltzmann studies of fluid flow in porous media
with realistic rock geometries’, Computers and Mathematics with Applications, Vol. 59, No.
7, pp.2305–2314.
Boutt, D.F., Grasselli, G., Fredrich, J.T., Cook, B.K. and Williams, J.R. (2006) ‘Trapping
zones: the effect of fracture roughness on the directional anisotropy of fluid flow and colloid
transport in a single fracture’, Geophysical Research Letters, Vol. 33, L21402, DOI: 10.1029/
2006GL027659.
Brown, S.R. and Scholz, C.H. (1985) ‘Broad bandwidth study of the topography of natural
rock surfaces’, Journal Geophysical Research, Vol. 90, No. B14, pp.12575–12582.
47
Brusseau, M.L., Peng, S., Schnaar, G. and Costanza-Robinson, M.S. (2006) ‘Relationships
among air-water interfacial area, capillary pressure, and water saturation for a sandy porous
medium’, Water Resources Research, Vol. 42, W03501, DOI: 10.1029/2005WR004058.
Crandall, D., Bromhal, G. and Karpyn, Z.T. (2010) ‘Numerical simulations examining the
relationship between wall-roughness and fluid flow in rock fractures’, International Journal of
Rock Mechanics and Mining Sciences, Vol. 47, No. 5, pp.784–796.
Culligan, K.A., Wildenschild, D., Christensen, B.S.B., Gray, W.G. and Rivers, M.L. (2006)
‘Pore-scale characteristics of multiphase flow in porous media: a comparison of air-water and
oil-water experiments’, Advances in Water Resources, Vol. 29, No. 2, pp.227–238.
Dougan, L.T., Addison, P.S. and McKenzie, W.M.C. (2000) ‘Fractal analysis of fracture: a
comparison of dimension estimates’, Mechanics Research Communications, Vol. 27, No. 4,
pp.383–392.
Drazer, G. and Koplik, J. (2002) ‘Transport in rough self-affine fractures’, Physical Review
E, Vol. 66, 026303, DOI: 10.1103/PhysRevE.66.026303.
Eker, E. and Akin, S. (2006) ‘Lattice Boltzmann simulation of fluid flow in synthetic
fractures’,Transport in Porous Media, Vol. 65, No. 3, pp.363–384.
48
Ferreol, B. and Rothman, D.H. (1995) ‘Lattice-Boltzmann simulations of flow-through
Fontainebleau sandstone’, Transport in Porous Media, Vol. 20, No. 1, pp.3–20.
Frisch, U., Hasslacher, B. and Pomeau, Y. (1986) ‘Lattice-gas automata for the Navier-
stokes equation’, Physical Review Letters, Vol. 56, No. 14, pp.1505–1508.
Genabeek, O.V. and Rothman, D.H. (1996) ‘Macroscopic manifestations of microscopic
flows: phenomenology from simulation’, Annual Review of Earth and Planetary Sciences, Vol.
24, No. 1, pp.63–87.
He, X., Zou, Q., Luo, L. and Dembo, M. (1997) ‘Analytic solutions of simple flows and analysis
of nonslip boundary conditions for the lattice Boltzmann BGK model’, Journal of Statistical
Physics, Vol. 87, Nos. 1/2, pp.115–136.
Heijs, A.W.J., Delange, J., Schoute, J.F.T. and Bouma, J. (1995) ‘Computed-tomography as a
tool for nondestructive analysis of flow patterns in macroporous clay soils’, Geoderma, Vol.
64, Nos. 3/4, pp.183–196.
Johns, R.A., Steude, J.S., Castanier, L.M. and Roberts, P.V. (1993) ‘Nondestructive
measurements of fracture aperture in crystalline rock cores using X-ray computed-
tomography’, Journal of Geophysical Research-Solid Earth, Vol. 98, No. B2, pp.1889–1900.
49
Karpyn, Z.T., Grader, A.S. and Halleck, P.M. (2007) ‘Visualization of fluid occupancy in a
rough fracture using micro-tomography’, Journal of Colloid and Interface Science, Vol. 307,
No. 1, pp.181–187.
Keller, A.A., Roberts, P.V. and Blunt, M.J. (1999) ‘Effect of fracture aperture variations on
the dispersion of contaminants’, Water Resources Research, Vol. 35, No. 1, pp.55–63.
Keller, A.A., Roberts, P.V. and Kitanidis, P.K. (1995) ‘Prediction of single-phase transport
parameters in a variable aperture fracture’, Geophysical Research Letters, Vol. 22, No. 11,
pp.1425–1428.
Ketcham, R.A., Slottke, D.T. and Sharp, J.M. (2010) ‘Three-dimensional measurement of
fractures in heterogeneous materials using high-resolution X-ray computed tomography’,
Geosphere, Vol. 6, No. 5, pp.499–514.
Kim, I., Lindquist, W.B. and Durham, W.B. (2003) ‘Fracture flow simulation using a finite-
difference lattice Boltzmann method’, Physical Review E, Vol. 67, No. 4, 046708, DOI:
10.1103/PhysRevE.67.046708.
Konzuk, J.S. and Kueper, B.H. (2004) ‘Evaluation of cubic law based models describing
single-phase flow through a rough-walled fracture’, Water Resources Research, Vol. 40, No.
2, W02402, DOI: 10.1029/2003WR002356.
50
Koponen, A., Kataja, M. and Timonen, J. (1997) ‘Permeability and effective porosity of
porous media’, Physical Review E, Vol. 56, No. 3, pp.3319–3325.
Madadi, M. and Sahimi, M. (2003) ‘Lattice Boltzmann simulation of fluid flow in fracture
networks with rough, self-affine surfaces’, Physical Review E, Vol. 67, No. 2, 026309, DOI:
10.1103/PhysRevE.67.026309.
Maier, R.S., Kroll, D.M., Kutsovsky, Y.E., Davis, H.T. and Bernard, R.S. (1998) ‘Simulation
of flow through bead packs using the lattice Boltzmann method’, Physics of Fluids, Vol. 10,
No.1, pp.60–74.
Manz, B., Gladden, L.F. and Warren, P.B. (1999) ‘Flow and dispersion in porous media:
lattice-Boltzmann and NMR studies’, AICHE Journal, Vol. 45, No. 9, pp.1845–1854.
McNamara, G.R. and Zanetti, G. (1988) ‘Use of the Boltzmann equation to simulate lattice-
gas automata’, Physical Review Letters, Vol. 61, No. 20, pp.2332–2335.
Mees, F., Swennen, R., Van Geet, M. and Jacobs, P. (2003) Applications of X-ray
Computed Tomography in Geosciences, p.57, Geological Society Special Publications, London.
Nakashima, Y., Nakano, T., Nakamura, K., Uesugi, K., Tsuchiyama, A. and Ikeda, S.
(2004) ‘Three-dimensional diffusion of non-sorbing species in porous sandstone: computer
51
simulation based on X-ray microtomography using synchrotron’, Journal of Contaminant
Hydrology, Vol. 74, Nos. 1/4, pp.253–264.
Nazridoust, K., Ahmadi, G. and Smith, D.H. (2006) ‘A new friction factor correlation for
laminar, single-phase flows through rock fractures’, Journal of Hydrology, Vol. 329, Nos. 1/2,
pp.315–328.
Nolte, D.D., Pyraknolte, L.J. and Cook, N.G.W. (1989) ‘The fractal geometry of flow paths
in natural fractures in rock and the approach to percolation’, Pure and Applied Geophysics,
Vol. 131, Nos. 1/2, pp.111–138.
Okabe, H. and Blunt, M.J. (2004) ‘Prediction of permeability for porous media
reconstructed using multiple-point statistics’, Physical Review E, Vol. 70, No. 6, 066135,
DOI: 10.1103/ PhysRevE.70.066135.
Pan, C., Luo, L-S. and Miller, C.T. (2006) ‘An analysis of lattice Boltzmann schemes for
porous medium flow simulation’, Computers and Fluids, Vol. 35, Nos. 8–9, pp.898–909.
Pan, C.X., Prins, J.F. and Miller, C.T. (2004) ‘A high-performance lattice Boltzmann
implementation to model flow in porous media’, Computer Physics Communications, Vol.
158, No. 2, pp.89–105.
52
Perret, J., Prasher, S.O., Kantzas, A. and Langford, C. (1999) ‘Three-dimensional
quantification of macropore networks in undisturbed soil cores’, Soil Science Society of
America Journal, Vol. 63, No. 6, pp.1530–1543.
Poon, C.Y., Sayles, R.S. and Jones, T.A. (1992) ‘Surface measurement and fractal
characterization of naturally fractured rocks’, Journal of Physics D, Applied Physics, Vol. 25,
No. 8, pp.1269–1275.
Prodanovic, M., Lindquist, W.B. and Seright, R.S. (2006) ‘Porous structure and fluid partitioning
in polyethylene cores from 3D X-ray microtomographic imaging’, Journal of Colloid and
Interface Science, Vol. 298, No. 1, pp.282–297.
Renard, F., Bernard, D., Desrues, J. and Ougier-Simonin, A. (2009) ‘3D imaging of
fracture propagation using synchrotron X-ray microtomography’, Earth and Planetary Science
Letters, Vol. 286, Nos. 1/2, pp.285–291.
Schmittbuhl, J., Schmitt, F. and Scholz, C. (1995) ‘Scaling invariance of crack surfaces’, Journal
of Geophysical Research, Vol. 100, No. B4, pp.5953–5973.
Stockman, H.W., Li, C. and Wilson, J.L. (1997) ‘A lattice-gas and lattice Boltzmann study
of mixing at continuous fracture junctions: importance of boundary conditions’, Geophysical
Research Letters, Vol. 24, No. 12, pp.1515–1518.
53
Succi, S. (2001) The Lattice Boltzmann Equation for Fluid Dynamics and Beyond, Oxford
University Press, USA.
Sukop, M. and Thorne, D.T. (2006) Lattice Boltzmann Modelingan Introduction for
Geoscientists and Engineers, Springer-Verlag, Berlin, Heidelberg.
Tsang, Y.W. and Witherspoon, P.A. (1981) ‘Hydromechanical behavior of a deformable
rock fracture subject to normal stress’, Journal of Geophysical Research, Vol. 86, No. 10,
pp.9287–9298.
Tsang, Y.W. and Witherspoon, P.A. (1983) ‘The dependence of fracture mechanical and fluid-
flow properties on fracture roughness and sample-size’, Journal of Geophysical Research, Vol.
88, No. 3, pp.2359–2366.
Wennberg, O.P., Rennan, L. and Basquet, R. (2009) ‘Computed tomography scan imaging
of natural open fractures in a porous rock; geometry and fluid flow’, Geophysical Prospecting,
Vol. 57, No. 2, pp.239–249.
Wildenschild, D., Hopmans, J.W., Rivers, M.L. and Kent, A.J.R. (2005) ‘Quantitative analysis
of flow processes in a sand using synchrotron-based X-ray microtomography’, Vadose Zone
Journal, Vol. 4, No. 1, pp.112–126.
54
Witherspoon, P.A., Wang, J.S.Y., Iwai, K. and Gale, J.E. (1980) ‘Validity of cubic law for fluid-
flow in a deformable rock fracture’, Water Resources Research, Vol. 16, No. 6, pp.1016–1024.
Zhang, D.X. and Kang, Q.J. (2004) ‘Pore scale simulation of solute transport in fractured
porous media’, Geophysical Research Letters, Vol. 31, No. 12, L12504, DOI: 10.1029/
2004GL019886.
Zhang, X., Knackstedt, M.A. and Sahimi, M. (1996) ‘Fluid flow across mass fractals and self-
affine surfaces’, Physica A: Statistical Mechanics and its Applications, Vol. 233, Nos. 3/4,
pp.835–847.
Zimmerman, R.W. and Bodvarsson, G.S. (1996) ‘Hydraulic conductivity of rock
fractures’, Transport in Porous Media, Vol. 23, No. 1, pp.1–30.
Zou, Q. and He, X. (1997) ‘On pressure and velocity boundary conditions for the lattice
Boltzmann BGK model’, Physics of Fluids, Vol. 9, No. 6, pp.1591–1598.
55
CHAPTER 3: RELATIVE PERMEABILITY OF HOMOGENOUS-WET AND MIXED-
WET POROUS MEDIA AS DETERMINED BY PORE-SCALE LATTICE
BOLTZMANN MODELING
3.1 SUMMARY
We present a pore-scale study of the dependence of relative permeability dependence on the
strength of wettability of homogenous-wet porous media, as well as the dependence of relative
permeability on the distribution and severity of wettability alteration of porous media altered to a
mixed-wet state. A Shan-Chen type multicomponent multiphase lattice Boltzmann model is
employed to determine pore-scale fluid distributions and relative permeability. Mixed-wet states
are created – after pre-simulation of homogeneous-wet porous medium – by altering the
wettability of solid surfaces in contact with the non-wetting phase. To ensure accurate
representation of fluid-solid interfacial areas we compare LB simulation results to experimental
measurements of interfacial fluid-fluid and fluid-solid areas determined by x-ray computed
microtomography imaging of water and oil distributions in bead packs (Landry et al. 2011). The
LB simulations are found to match experimental trends observed for fluid-fluid and fluid-solid
interfacial area-saturation relationships. The relative permeability of both fluids in the
homogenous-wet porous media is found to decrease with a decreasing contact angle. This is
attributed to the increasing disconnection of the non-wetting phase and increased fluid-solid
interfacial area of the wetting phase. The relative permeability of both fluids in the altered
mixed-wet porous media is found to decrease. However the significance of the decrease is
56
dependent on the connectivity of the unaltered solid surfaces, with less dependence on the
severity of alteration.
3.2 BACKGROUND
Macroscale multiphase flow properties in porous media (i.e. relative permeability and capillary
pressure) are strongly dependent on wettability. The wettability of a porous medium is initially
determined by its mineral composition. However, brines and oils can alter the wettability of
porous media, such that the multiphase flow properties of a rock can be significantly dependent
on saturation history. Subsequently, the majority of consolidated and unconsolidated porous
media have a mixed or fractional wettability. The process occurs in three steps:
1) Rocks initially develop wettability as a result of mineral-water interactions.
2) Migration of non-aqueous phases into the rock alters the wettability of the rock.
3) Subsequent cycles of displacement further alter the wettability of the rock.
Wettability is conventionally measured using the Amott-Harvey method or the USBM method.
Both measure an index of wettability from displacement experiments. Indices range from -1-to-1
for perfectly oil-wet to perfectly water-wet conditions. At the pore-scale, wettability is described
by contact angles. The contact angle is measured through the aqueous phase and ranges from 0°
to 180°, for perfectly water-wet to perfectly oil-wet, respectively. Common reservoir porous
media includes siliclastic rocks (i.e. sandstone) primarily composed of quartz, but also
containing significant amounts of feldspars, calcite, dolomite and clays; and carbonate rocks (i.e.
limestone) composed primarily of calcite, dolomite, clays and organics.
57
Conventional measurement of oil recovery and macroscale flow properties from mixed-wet
reservoir rock cores demonstrate some of the important trends observed in field-scale variations
of saturation history and wettability. Janhunandan and Morrow 1995 investigated changes in
wettability and oil recovery from Berea cores aged with two different crude oils and varying
initial water saturations (Swi), aging temperatures and brine compositions. Regardless of oil type
and brine composition the Amott index of the rock increased monotonically with Swi. Jerauld
and Rathmell 1997 observed the same trend in oil-wetness and connate water saturation for
Prudhoe Bay sandstone cores composed primarily of quartz. Core wettability trended more oil-
wet with the decrease in connate water saturation above the oil-water contact line of the
reservoir. With more oil present in the rock, a greater portion of the pore space will be in contact
with the oil, resulting in larger surface area of the pore space subject to wettability alteration.
Both studies also demonstrated the dependence of wettability alteration on the brine
composition, oil composition and aging temperature. Schembre et al. 2006 measured the
wettability and oil recovery of cores from a diamtomite (carbonate) reservoir. The Amott indices
were found to increase significantly from 0.2-0.55 to 0.55-0.85 with temperature increases from
45 to 230 C, and correlate with the percent content of water-wet clays. The increase in the
Amott indices were also found to increase oil recovery. The temperature dependence indicates
variation in the temperature of a reservoir will affect the wettability of the rock.
Once a fluid is in contact with a mineral surface in the porous media the alteration in wettability
is controlled by fluid-mineral surface chemistry. As summarized by Buckley and Liu 1998 the
mechanisms involved in crude oil/brine/solid interaction, include polar interactions between the
polar functional groups in the oil and polar surface sites on the mineral, surface precipitation,
58
acid/base reactions and ion binding. One of the most common minerals found in siliclastic rocks
is quartz. The surface interactions between quartz and brine depend on both the pH and ionic
strength of the brine. The affinity of brine to quartz in the presence of an oil phase is
proportional to the amount of ion adsorption. Ion adsorption increases with increasing ionic
strength and decreases for pHs in the proximity of the pH of zero-point surface charge (pHzpc =
1.5-4.0) (Barranco et al. 1997). The same behavior is observed in coal, with the strongest water
wettability found at alkaline and acidic conditions. Chaturvedi et al. 2009 demonstrated the
effect pH has on the wettability, and subsequently, the relative permeability of coal. At brine
pHs of 10, and 7, the intersection point of relative permeability curves were Sw = 0.65 (water-
wet) and Sw = 0.43 (weakly oil-wet), respectively. Carbonate wettability is also sensitive to brine
pH and composition, as demonstrated in Yu et al. 2009 chalk cores experienced very little
spontaneous imbibition of reservoir brine, while seawater altered the wettability to water-wet and
spontaneously imbibed the core. Evje and Hiorth 2011 modeled these results with the inclusion
water-rock chemistry, specifically the dissolution and precipitation of calcite, magnesite and
anhydrite in the presence of Mg2+
and SO42-
. They attributed the increase in wettability to the
presence of SO42-
in the seawater brine, which increased the dissolution of calcite, exposing new
water-wet surfaces. One of the most important mechanisms of wettability alteration by crude
oils is the adsorption and deposition of asphaltenes on mineral surfaces. Asphaltenes can render
initially water-wet mineral surfaces oil-wet. Saraji et al. 2010 measured the adsorption of
asphaltene to calcite, dolomite and quartz by asphaltene containing oils under flow conditions.
The asphaltene was found to adsorb the most to calcite, and to lesser extent quartz and dolomite.
Wettability was not measured, but it was concluded that the amount of asphaltene adsorbed was
sufficient to form a monolayer on all minerals. The alteration of mineral wettability is controlled
59
by mineral-fluid interactions. Knowledge of individual mineral-fluid surface chemistry provides
insight into which minerals in a porous medium are being altered. Tabriziky et al. 2011
measured changes to contact angles and wettability indices of quartz, kaolinite and calcite after
aging in presence of oil with and without asphaltene content, and brines containing either MgCl2
or Na2SO4. Wettability was measured using BET H2O surface adsorption, and contact angles
measured on flat surfaces of calcite and quartz to investigate the dependence of wettability
alteration on ionic species and the presence of asphaltene. Quartz was neutral wet for oil without
asphaltenes and distilled water, and weakly oil wet for oil with asphaltenes and distilled water.
SO42-
had a negligible effect; however, Mg- significantly altered quartz to a water wet state in the
presence of asphaltenes, and an oil-wet state without the presence of asphaltenes. Kaolinite was
neutral wet for oil with and without asphaltenes and distilled water. SO42-
, as with quartz, had a
negligible effect. Mg- significantly altered kaolinite to a water wet state in the presence of oil
with and without asphaltene. Calcite was water wet in presence of oil with and without
asphaltene and distilled water. SO42-
altered the wettability to a weakly water-wet state, and Mg-
altered the wettability further to a weakly oil-wet state. Studies such as these demonstrate the
complexity involved in fluid/mineral interactions and the resulting alteration of wettability.
There are two factors to consider in the study of wettability alteration, the chemistry of alteration
and the distribution of fluids in the pore geometry with respect to mineral constituents of the
porous media. There are few studies that have attempted to measure contact angles at the pore
scale within natural porous media, where mineral heterogeneity exists in individual pores. Robin
et al. 1995 imaged the distribution of oil and water phases in a reservoir sandstone and carbonate
before and after wettability alteration using cryo-SEM. The images provided qualitative
60
evidence of individual mineral wettability alteration within individual pores. Many of the trends
in mineral wettability alteration discussed above are found in these images. Although these
types of images confirm the assumptions of bulk studies, they do not provide much insight into
the 3D pore geometry that governs the movement of fluids. Non-destructive 3D imaging of pore
geometry and fluid distributions using x-ray computed microtomography (CMT) provides us
with a powerful tool to image and analyze pore geometries and fluid distributions in these
geometries. Many investigators have used CMT to image a variety of fluid distributions in
unconsolidated materials (Culligan et al. 2006, Brusseau et al. 2006, Brusseau et al. 2008,
Constanza-Robinson et al. 2008, Al-Raoush et al. 2009, Lebedeva and Fogden 2011) and rock
(Coles et al. 1998, Turner et al. 2004, Prodanovic et al. 2007, Iglauer et al. 2011,, Silin et al.
2011, Kumar et al. 2012). Armed with knowledge of the fluid/mineral wettability alteration and
pore geometry, pore scale models can be employed to investigate the macroscale multiphase
flow properties of mixed-wet porous media.
The focus of this paper is to further elucidate the relationship between mixed-wet states and
relative permeability. In the past two decades pore network models have been successfully used
to model wettability alteration, multiphase fluid flow of mixed-wet states, and saturation history
dependent relative permeability and capillary pressure hysteresis (Blunt 1997a, Blunt 1998a,
Blunt et al. 2002, Dixit 1996, Dixit 1998, Al-Futaisi and Patzek 2004, Hui and Blunt 2000,
Jackson et al. 2003). Kovscek et al. 1993 introduced a physically based model for wettability
alteration within individual pores; each pore is represented by a capillary tube with a star-shape
cross-section. The tube is initially occupied by the wetting phase, during drainage the wetting
phase is displaced by a piston-like mechanism by the non-wetting phase. The non-wetting phase
61
will now occupy the center of the pore, leaving the corners occupied by the wetting phase. For
the surface of the pore to make contact with the non-wetting phase the capillary pressure must
overcome a critical capillary pressure at which the wetting film destabilizes to molecular
thickness. Once contact is made the surface of the pore in contact with the oil is altered in its
strength of wettability. This concept was further extended to squares (Blunt 1997a, Blunt
1997b), triangles and lenses (Oren et al. 1998). As is noted in Blunt 1997, the disjoining
capillary pressure is not known for each individual pore, pore network models cope with this by
assigning a wettability alteration to a fraction of the pores occupied by oil. These fractions can
then be varied to observe resulting multiphase flow behavior. The extraction of representative
pore networks from 3D images was introduced by Oren et al. 1998, and has led to numerous
studies of multiphase flow in a variety of porous media imaged by CMT and modeled by pore
network models. In mixed-wet media there are three general possible modes of assigning
fractional wettability to pores - the largest pores, the smallest pores, or at random (Oren and
Bakke 2003, Hoiland et al. 2007). The choice of contact angles representing the wettability can
also be uniform or distributed at random over a range. With these free parameters almost any
wetting state can be modeled (Fenwick and Blunt 1998, Hui and Blunt 2000, Jackson et al. 2003,
Al-Futaisi and Patzek 2004, Speiteri et al. 2008, Svirsky et al. 2007, Zhao et al. 2010). There
are concerns regarding the use of pore network models for the study of wettability alteration and
its effect on multiphase flows. The simplification of the pore space into capillary tubes of
varying shape is not predisposed to the inclusion of knowledge of the mineral composition of the
pores. The surface chemistry is highly dependent on accurate representation of fluid/mineral
interfacial areas, which may be difficult to ascertain without proper inclusion of the mineral
composition of pores.
62
In this paper we use a more rigorous pore scale modeling approach. Lattice Boltzmann (LB)
methods can directly use 3D images of pore space by a one-to-one correlation of image voxels to
lattice sites. The accurate representation of any pore geometry is then only a question of lattice
resolution. This makes the model very appealing over pore network models. However, it is far
more computationally expensive, requiring on the order of 104-5
more cpu time than an
equivalent pore network model (Vogel et al. 2005). The computational needs of LB are
compensated by its inherent parallelism – in most models each node is only aware of its
neighboring nodes. Multiphase LB models have been validated against experimental
measurements of capillary pressure in bead packs (Pan et al. 2004, Schaap et al. 2007, Porter et
al. 2009) and sandstones (Ramstad et al. 2010, Ramstad et al. 2012), as well as experimental
measurements of relative permeability in sphere packs (Ghassemi and Pak 2009, Hao and Cheng
2010) and sandstones (Boek and Venturoli 2010, Ramstad et al. 2010). Investigations of mixed
wettability using LB are limited to the work of Hazlett et al. 1998. In this investigation a CMT
image of a reservoir sandstone was translated directly into a lattice for LB multiphase
displacement simulations for a strongly water-wet sandstone and a mixed-wet sandstone. The
mixed wet state was determined by simulating a drainage to irreducible water saturation. At the
end of the drainage, the pore walls in contact with the non-wetting phase are altered in
wettability. The measurements of capillary pressure and relative permeability matched
experimentally observed trends. Although there has been a fair amount of validation of LB
methods via matching experimental measurements of macroscale flow properties, few have
compared fluid distributions imaged using CMT with LB model results. Sukop et al. 2008
compared two phase fluid distributions in a sand pack imaged using CMT to distributions
63
resulting from LB simulations. The saturation per slice was found to match experimental results
well, but there was no analysis presented of interfacial area measurements. Porter et al. 2009
compared interfacial fluid-fluid areas measured by CMT to LB displacement simulations and
found a good match for drainage simulations.
Our objective is to investigate the evolution of wettability alteration as a result of saturation
history and the initial state of wettability, as well as the sensitivity of relative permeability to this
evolution. Proper pore-scale simulations of wettability alteration require accurate predictions of
fluid-solid interfacial areas. To test the ability of the LB model to predict these interfacial areas
we will first compare CMT images of oil/brine fluid distributions in water-wet and weakly oil-
wet bead packs from Landry et al. 2011 to the results of LB simulations. The bead packs can be
viewed as simple analogs of water-wet siliclastic porous media and weakly oil-wet carbonate
porous media. There are three variables of interest in our investigation of relative permeability
– saturation relationships and their dependence on wettability, the initial wetting state of the
porous media, the saturation at which wettability alteration takes place, and the degree of
alteration. As was done in the work of Hazlett et al. 1998 solid surfaces in contact with the non-
wetting fluid, after fluid distribution is established by homogenous-wet LB simulations, will be
altered to create a mixed-wet porous medium. Relative permeability measurements on the
initial and mixed-wet states will then be carried out using the LB model. We will be able to
evaluate the extent to which the LB model can recreate pore scale distributions of immiscible
fluids, and compare relative permeability measurements of mixed-wet states. In this simplified
scenario, the initial wettability, saturation of alteration and severity of alteration control the
mixed-wet state of the porous media. The only variable in our simulations is wettability, as a
64
result of these three parameters. Natural porous media is subject to a far more complex
wettability alteration process, dependent on numerous variables related to the mineral-fluid
interactions. However, the resulting evolution of relative permeability – saturation relationships
can be investigated without such considerations.
3.3 MATERIALS AND METHODS
3.3.1 Experimental Measurements
The validation of pore-scale models at the pore-scale is limited in the literature. Here we
compare LB simulation results to CMT images of brine/oil distributions in strongly water-wet
(glass) and weakly oil-wet (polyethylene) bead packs from Landry et al. 2011. Both bead types
were spherical with a size range between 0.425 and 0.600 mm, and packed in a column 25.4 mm
in diameter and ~80-90 mm in length. The fluids used were a 4-8% NaI brine and kerosene, with
dynamic viscosities of 1.00 and 2.43 cP, respectively. The bead packs were initially fully
saturated with brine, followed by capillary-dominated displacement by kerosene, then brine. At
the end of each displacement CMT images were taken with a voxel resolution of 0.0260 x 0.0260
x 0.0292 mm3 and 0.0259 x 0.0259 x 0.0274 mm
3, for the glass and polyethylene bead packs,
respectively. Due to the relatively large pore aperture of these porous media gravity plays a
significant role in the distribution of fluids in these bead packs. This results in a classic
transition-zone saturation profile in both bead packs at the end of drainage, from which fluid
distributions can be observed at varying fluid saturations. From these images specific interfacial
fluid areas were measured, as well as blob size and surface area distributions. These
65
measurements of volume and area will be compared in this paper to the results of LB
simulations.
3.3.2 Single-Phase BGK Lattice Boltzmann Model
The LB method is a mesoscopic method based on microscopic particle dynamics that provides
numerical solutions to macroscopic hydrodynamics. In short, fluids are modeled as swarms of
streaming and colliding particles. Each node on the lattice contains a particle distribution
function, , where is the index of each discrete velocity, is the location of the node on
the lattice, and is time; here we use a D3Q19 lattice (Qian et al 1992). The D3Q19 particle
distribution function has 19 discrete velocities, , including a zero velocity and 18 velocities
pointing to neighboring nodes. The particle distribution function is also referred to as the density
function – the macroscopic density at each node is represented by a distribution of densities of
particles moving along each discrete velocity. The particle distribution function evolves in time
according to the lattice Boltzmann equation,
[
]
where is the relaxation parameter, is a discrete time step ( , for one time iteration, in
lattice units 1 t.u.) and is the equilibrium particle distribution function. The left hand
side of the equation represents the streaming of particles by passing the particle distribution at
each discrete velocity to the respective neighboring node. The right-hand side of the equation
represents the collision of particles as a partial relaxation to the equilibrium particle distribution.
The relaxation parameter (also known as the collision interval) is representative of the rate of
particle collisions, and is related to the kinematic viscosity, , of the lattice by, ,
66
where is the speed of sound of the lattice, or propagation speed, equal to √ . The discrete
lattice speed unit is the ratio of the lattice spacing – equal to 1 lattice length unit –
and the time step. Here we use Bhatnar-Gross-Krook (BGK) dynamics, meaning the relaxation
parameter is defined by a single value. Although it has been shown that BGK dynamics results
in viscosity-dependent permeability measurements (Pan et al. 2004), as opposed to more
computationally demanding dynamics (e.g. Multiple-Relaxation Time dynamics), for our
purposes BGK dynamics will suffice. The D3Q19 equilibrium particle distribution function
can be calculated as
[
]
where is the weight of each discrete velocity, is the density of , and is the
macroscopic velocity moment of . The discrete velocities are defined as follows,
[ ]
[
]
with weights ⁄ , ⁄ , ⁄ . The
density is obtained by summing the particle densities, ∑ , and the macroscopic velocity
is obtained by summing the particle momentum and dividing by density , ∑ ⁄ . To
impose external forces, , on the lattice (i.e. gravity), momentum is added to the macroscopic
velocity in the following way,
where is the macroscopic velocity calculated from the particle distribution function prior to
collision. Each time iteration of the LB equation proceeds in three steps. First, the particle
67
distributions are streamed to respective nodes. Second, the macroscopic density and velocity are
determined from these new particle distributions, and when present, with the inclusion of an
exterior force as stated above. Third, the equilibrium particle distribution is obtained and the
particle distribution undergoes collision. There are three types of lattice nodes, fluid nodes,
bounce-back nodes and solid nodes. The nodes on which fluids move are fluid nodes, the nodes
on the surface of solids are bounce-back nodes, and nodes on which no fluids exist are solid
nodes. The bounce-back nodes act as algorithmic devices that return incoming streamed particle
densities with the opposite momentum, producing a no-slip condition halfway between the fluid
node and the bounce-back node.
3.3.3 Shan-Chen Multicomponent Lattice Boltzmann Model
We employ the Shan-Chen Multicomponent (MC) LB model to simulate two-phase immiscible
fluid flow (Shan and Doolen 1995, Martys and Chen 1996). The Shan-Chen MC LB model is
one of the least computationally demanding multiphase LB models, and for this reason one of the
most commonly used multiphase LB models. Two immiscible fluids are simulated on the lattice
by representing each fluid phase with its own particle distribution function, , where
is the index for each component particle distribution function. The component particle
distributions interact via a pseudo-potential interparticle force ( ) defined as,
∑
where is the parameter controlling the strength of the interparticle force, when is positive
the force is repulsive. Only the nearest-neighbor nodes are considered in the calculation of the
68
interparticle force. The adhesion force between components and solid surfaces is created by
imposing fictitious component densities, , on bounce-back nodes,
∑
where is an indicator function that denotes the existence of a bounce-back node, as in,
when is occupied by a bounce-back node. Conventionally, negative values
of are used for wetting fluids and positive values of are used for the non-wetting
fluids, with . The fictitious bounce-back densities are not densities in the
same manner as fluid nodes, they are simply values that control the strength of adhesion. These
forces are included in the lattice Boltzmann equation in the same manner as the exterior force,
The total momentum of all particles at each lattice node most be conserved by the collision
operator, thus the particle distribution functions for both components must be included in
calculations,
∑
∑
The surface densities of bounce back nodes ( ) in a Shan-Chen D3Q19 two
component system (wetting and non-wetting) can be estimated for a given contact angle by the
following equation proposed by Huang et al. 2007,
69
where is the density of the wetting fluid and is the dissolved density of the non-
wetting fluid in the wetting fluid (for values of sufficient to segregate component fluids the
dissolved density is very low). The lattice pressure at each node is determined by the D3Q19
Shan-Chen MC LB model equation of state,
[ ]
[ ]
All LB simulations presented here were executed using the open source code Palabos.
3.3.4 LB Model Implementation
There are a collection of parameters that are chosen by the user to define a LB model simulation
of fluid flow. The resolution of the lattice is an important factor when designing a LB
simulation, optimally the resolution should be as fine as computing resources allow. However,
due to the immense computational needs of LB simulations it is often the case that computing
resource limits and time constrictions determine the resolutions of LB fluid flow simulations in
porous media. Also, the necessary resolution to adequately represent flow in a porous media is
dependent on the pore structure, thus determining an adequate resolution for simulations merits
further investigation. Given our computing resources we chose to translate our CMT images
with a 1-to-1 correlation, voxel-to-node. This ensures we are using a resolution great enough to
capture the fluid-fluid and fluid-solid interfacial areas measured by CMT, and results in lattice
sizes within the limits of our computing resources. The imaged voxel resolution of the glass
bead pack is 0.0260 x 0.0260 x 0.0292 mm3, to impose a 1-to-1 correlation of voxel-to-node the
raw image of the pore geometry is resampled to 0.0260 x 0.0260 x 0.0260 mm3 and segmented
into pore and solid voxels. The pore voxels of the segmented image volume are translated as
70
lattice nodes, and the solid voxels that are in contact with pore voxels are translated as bounce-
back nodes, with all other solid voxels being ignored in the lattice (Figure 3.1). The resolution of
the imaged voxels determines the physical size of a lattice length unit, , here being the side
length of a CMT voxel, and the physical mass of lattice mass unit, , being the volume of a
CMT voxel times the density of the physical fluid being simulated.
Other parameters control the density and viscosity ratio of the fluids and the wettability of the
solid surfaces. Throughout the LB simulations presented here the sum of the lattice density of
the components, , is set to 1.0 . The time relaxation parameter , is set to
1 for both fluids; the purpose of this being two-fold. The relaxation parameter determines the
viscosity of each component, and thus the viscosity ratio of the fluids. Our experimental fluids
are kerosene and water with a viscosity ratio, ⁄ , of 2.3 and 0.43 for the glass bead
pack and polyethylene bead pack (Note: in the glass bead pack the kerosene is the non-wetting
phase, in the polyethylene bead pack the water is the non-wetting phase) displacements,
respectively. The capillary number, , is a dimensionless measure of the ratio of viscous to
capillary forces, and is defined as,
where is the dynamic velocity, is the fluid velocity in the direction of flow, and is the
interfacial tension. The experimental displacements were very low, on the order of
. Given the interfacial tension forces are so much greater than the viscous forces; we
assume simulating fluids with a viscosity ratio of 1 will suffice for our needs. Also the use of
BGK dynamics has been shown to result in viscosity-dependent permeability outside time-
relaxation parameters of 1 (Pan et al. 2006). A time relaxation parameter of 1 is both
71
numerically stable and representative of the physical system to be modeled. The interfacial
tension of the lattice is determined by the value of , we use a value of 1.8, as suggested by
Huang et al. 2007, giving a lattice interfacial tension of 0.100. In the SC MC LB model, values
of lower than 1.8 decrease the contrast of the component fluids, while higher values result in
an undesirable compression of fluid components (Schaap et al. 2007, Huang et al. 2007). As
was previously mentioned, the desired contact angle can be determined by equation proposed by
Huang et al 2007 for given bounce-back node densities and . We do not know
exactly what the contact angles are in our experimental system, we are only aware of the fact that
the glass bead pack is water-wet and the polyethylene bead pack is weakly oil-wet. To compare
LB results to experimental results we will simulate a range of contact angles.
Figure 3.1: Segmented CMT image of the bead pack (A) with solid voxels colored white, and the
corresponding lattice of the LB model (B) with the bounce-back nodes in blue.
72
3.4 RESULTS
3.4.1 Fluid-Fluid and Fluid-Solid Interfacial Areas
The objective of this investigation is to study flow in mixed-wet porous media, it is important
that the model can adequately capture the interfacial areas between solids and fluids. To validate
our method we compare the fluid surface and interfacial areas determined by LB simulations to
experimental CMT measurements. As previously mentioned and illustrated in figure 3.1 a 1-to-1
correlation of CMT voxel to lattice node is used. The image chosen was 1003 voxel
3, or ~5
beads to a side. Larger lattices up to 2003 voxel
3 were also used in initial simulations, but the
results were similar to results using the smaller 1003 voxel
3 lattices at an 1/8 of the computational
cost. In our steady-state simulations the initial density of each fluid component is uniformly set
at a desired saturation with a sum of 1, , and an exterior force of
⁄ is applied to both component fluids to initialize concurrent flow
resulting in capillary-dominated flow ( ). This setup allows the fluids to segregate
without out further input within the pore space. The disadvantage of this setup is it allows fluids
access to pores that would not be accessible in displacement processes, and also lead to a more
scattered non-wetting fluid distribution than what would be observed experimentally. All other
parameters of simulation are the same as those used in the simulation of the circular pore, with
the exception of the bounce-back densities ( ).
To compare the fluid distributions of the LB simulation to experimental images we measure the
specific fluid surface areas and specific interfacial areas. For details on the experimental
measurement of these areas refer to Landry et al. 2011. To measure these areas and volumes of
73
component fluids from the LB results the lattice must be first segmented to identify the fluid
occupying each node, here the fluid occupying the node, , is defined as the component
fluid with the greatest density,
The volume of the fluid component is,
∑
where the . The volume of the solid is the product of the sum of all the
bounce-back and non-interacting nodes and . The wetting-phase, non-wetting phase, and
solid surface areas, , are measured using a modified marching cubes algorithm,
this algorithm wraps a triangular-mesh around the segmented data (Dalla et al. 2002). The
specific fluid and solid surface areas, , are defined as, ⁄ , where is the
volume of the whole lattice. The specific solid surface area of the lattice, .
One of the drawbacks of using the LB model to simulate strongly-wetting porous media is the
occurrence of a monolayer or 1-node thick wetting phase film surrounding all bounce-back
nodes. This film is a consequence of using a mesoscopic model founded on particle dynamics;
the pseudo-potential affinity of the wetting phase on nodes adjacent to the bounce-back nodes is
too strong to be dislodged by the non-wetting phase. This monolayer can exaggerate the
measurement of the wetting-phase surface areas, which will have an effect on interfacial areas.
Very little flow occurs in these monolayers, so although they may exaggerate wetting phase
surface areas, they should not have a significant effect on flow.
74
We are interested in comparing three specific interfacial areas, the specific fluid-fluid interfacial
area, the non-wetting fluid-solid interfacial area and the wetting fluid-solid interfacial area. The
specific fluid-fluid interfacial area is the total area of the fluid-fluid interfaces (meniscus)
normalized to bulk volume and is defined as (Dalla et al. 2002),
The specific fluid-fluid interfacial area is important to the study of multiphase flows; it is here
that energy and mass are transferred between fluid phases. The specific wetting phase-solid
interfacial area, , and the specific non-wetting phase-solid interfacial area, are defined
as, . The solid surface area varies in the images, therefore we normalize the
specific fluid-solid interfacial areas to the specific solid surface area, giving us the fractional
wetting fluid-solid interfacial area, , and the fractional non-wetting fluid-solid interfacial
area, , defined as
⁄ .
The wettability alteration that will be investigated here will occur on bounce-back nodes in
contact with the non-wetting phase, and the fraction of the surface that will be altered is equal to
the fractional non-wetting fluid-solid interfacial area. Four contact angles are considered
, resulting from setting ,
respectively. A comparison of LB and experimental measurements of specific fluid-fluid
interfacial areas can be found in figure 3.2. Qualitatively, the LB simulations show a distinctive
curve trending towards a maximum at wetting phase saturations between 0.3 and 0.5. This same
curve is observed in the experimental measurements from the glass and polyethylene bead packs,
and has also been reported by both previous experimental and numerical investigations (Reeves
and Celia 1996, Brusseau et al. 2006, Culligan et al. 2006, Schaap et al. 2007, Joekar-Niaser et
75
al. 2009, Porter et al. 2009, Landry et al. 2011). We also observe an increase in specific fluid-
fluid interfacial area with decreasing contact angles, as would be expected, and is also reflected
in the experimental measurements of the weakly oil-wet polyethylene bead pack and the
moderately water-wet glass bead pack. A rough estimation by visual inspection from these
curves would suggest that the contact angles of the glass and polyethylene bead packs are near
and , respectively. However, we cannot make a quantitative comparison of
model and experimental results without knowledge of the experimental contact angles. In
general, the LB simulations capture the trends of the relationship between saturation and
interfacial fluid-fluid areas. The LB simulations also agree with the trends of the experimental
measurements of the fractional fluid-solid interfacial areas, as shown in figure 3.3. The
fractional non-wetting phase interfacial area decreases with decreasing contact angles, this is a
result of the wetting phase dominating occupation of the smaller pore spaces and forcing the
non-wetting phase into the largest pores where its fractional interface with the solid decreases.
This relationship affects the fraction of a solid surface that is in contact with a wettability altering
fluid, and also the location of the fluid in the pore space (i.e. the small pores where clay minerals
or organics may be found, or the large pores where multiple mineral constituents may be found.)
76
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
2
2.2
2.4
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
[mm
-1]
Sw
0.1
0.2
0.3
0.4
P
G
[
]
Poly beads
Glass beads
Figure 3.2: Specific fluid-fluid interfacial areas as a function of wetting phase saturation for the LB
simulations and experimental CMT images.
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Xfs
[]
Sw
0.10.20.30.4PG
Poly beads
Glass beads
Figure 3.3: Fractional fluid-solid interfacial areas as a function of wetting phase saturation for the LB
simulations and experimental CMT images, the wetting fluid-solid interfacial areas are
shown in black and the non-wetting fluid-solid interfacial areas are shown in gray.
3.4.2 Pore Aperture and Fluid Distribution
77
To measure the pore aperture distribution of the pore space, first a skeleton or medial axis, of the
pore space must be constructed. The skeleton is constructed by thinning algorithms (Lee et al.
1994) using the software Avizo Fire 6.3. The skeleton of the pore space is a connected network
of voxels that represent the midpoint between solid walls in the pore space. Each voxel in the
skeleton can be thought of as the center of a sphere that makes contact with the wall in at least
two places. There is a significant amount of literature regarding the subject of pore space
topology description (Thovert et al. 1993, Lindquist et al. 1996, Bakke and Oren 1997,
Prodanovic et al. 2006). The pore aperture is the maximum radius of a sphere that can occupy a
pore. To determine the pore aperture distribution, a “sphere packing” of sequentially smaller
spheres are strung along the skeleton. First, the voxels of skeleton are labeled with their distance
to the pore wall. The labeled skeleton is then searched for voxels that fall within a user-defined
range. These voxels then become the center of a sphere – with a radius equal to the distance to
the closest pore wall – placed in the pore space, and all voxels of the pore space that fall within
this sphere are labeled. The pore aperture distribution can then be determined by simply
summing the labeled voxels. The fluid distribution can also be easily measured by masking the
segmented fluid occupation ( ) with the pore aperture-labeled pore space. The pore
aperture distribution and fluid distribution are presented in figure 3.4. We can see from the fluid
distribution that decreasing the contact angle pushes the non-wetting phase into the larger pores.
This is a result of the increasing affinity of the wetting phase for the solid surfaces, resulting in
the wetting phase dominance of smaller pore spaces. This can have a substantial effect on the
dependence of fluid mobility on wettability alteration. If only the larger pores that would be
78
occupied by the non-wetting phase in a homogenous-wet state are altered the effect on the
relative permeability of the fluids will be minimal.
0
0.05
0.1
0.15
0.2
0.25
0.3
0 0.026 0.052 0.078 0.104 0.13 0.156 0.182 0.208
Volu
me
-weig
hte
d f
raction
Pore aperture [mm]
0
0.2
0.4
0.6
0.8
1
1.2
0 0.026 0.052 0.078 0.104 0.13 0.156 0.182 0.208
Non-w
ett
ing p
hase f
luid
fra
ction
Pore aperture [mm]
0.1
0.2
0.3
0.4
Figure 3.4: Pore aperture and fluid distribution of initial homogenous-wet states.
79
3.4.3 Two-phase Flow in a Circular Pore
To test our model against a known semi-analytical solution for two-phase immiscible flow we
simulate flow in a circular pore. Two-phase immiscible flow is commonly described using an
empirical extension of Darcy’s law,
where the mean velocity of each fluid parallel to the direction of flow, , is a function of the
dynamic viscosity, , pressure gradient, , absolute or single-phase permeability, , and the
relative permeability defined as , where is the effective permeability. In a
circular pore where the wetting fluid is distributed as an annulus along the walls of the pore and
the viscosity ratio is one, the wetting phase relative permeability is a function of the wetting
phase saturation, , and the non-wetting phase relative permeability is a function of
non-wetting phase saturation, ⁄ (Ramstad et al. 2010). To simulate two-
phase flow in a circular pore at varying saturations the wetting phase fluid component is initially
distributed as an annulus along the pore walls with the non-wetting phase fluid component
occupying the nodes in the center in correlation with the desired saturation. The walls of the
pore are strongly-wetting to maintain the annular distribution of the wetting phase. Otherwise,
under weakly-wetting conditions, the wetting phase will separate and lose its annular shape, the
annular shape being an assumption of the semi-analytical solution of immiscible two-phase flow
in a circular pore. To determine the relative permeability of component fluids in the LB
simulation the momentum in the direction of flow of each fluid is summed over all pore nodes at
the given wetting phase saturations,
80
∑
and normalized to the fully saturated momentum,
To impose concurrent flow in a circular pore with a radius of 11 nodes an exterior force of
⁄ is applied to both component fluids, and periodic boundaries
are imposed at the inlet and outlet. The system is considered to have reached steady-state when
the convergence criteria were met,
A choice of was chosen, based on numerical experiments. In figure 3.5 are the results
of the LB simulation and the semi-analytical solutions for two-phase flow in a circular pore.
The difference between the semi-analytical solution and the LB simulations is due to the discrete
nature of the rendering of the circular pore and the resolution. Previous investigations have
shown multicomponent LB models are increasingly more accurate with finer resolutions
(Ramstad et al. 2010).
81
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 0.2 0.4 0.6 0.8 1
kr
S(w)
Figure 3.5: Relative permeability in a circular pore with radius = 11 ; LB simulation results and
semi-analytical solution curves.
3.4.4 Relative Permeability
The relative permeability of the porous media is determined by LB simulations in the same
manner as described in the section Two-phase flow in a circular pore. With the simulation
volume being bounded by walls perpendicular to flow, and as was used in the circular pore
simulations, periodic boundaries are applied in the direction flow (i.e. the inlet and outlet).
However, as was described in Martys and Chen 1996, the total momentum of the component
lattices did not display a linear relationship with exterior forcing below . To
determine the relative permeability exterior forces between
⁄ were used. The relative permeability curves for the initial homogenous-wettability
82
porous media are shown in figure 3.6. At low wetting phase saturations the non-wetting phase
relative permeability exceeds one and the wetting phase relative permeability is slightly negative.
This has also been reported in the preliminary results of Boek and Ventourilli 2010. These
saturations are at or below experimentally observed irreducible wetting phase saturations of
and for the glass and polyethylene beads respectively, and are unlikely to
occur in displacement processes. We are able to measure the relative permeability at these
saturations using the LB simulations due to the employment of a steady-state setup which allows
for any initial saturation. However, some of these simulations may not have a physical corollary.
-0.1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
1.2
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
kr
Sw
0.10.20.30.4
Figure 3.6: Relative permeability determined by LB simulations for the initial homogenous-wettability
states. The relative permeability of the wetting and non-wetting phase are shown in black
and gray, respectively.
83
Generally, figure 3.6 shows decreasing the contact angle also decreases the relative permeability
of both phases at all saturations. The variation is small due to the high porosity and connectivity
of the pore space. Also, as would be expected, the crossover point for the relative permeability
curves occurs at . The relative permeability of the fluids is governed by two competing
mechanisms that act to inhibit and promote fluid flow, namely, fluid-solid interfacial area and
connectivity, respectively. Greater fluid-solid interfacial area leads to greater resistance to flow,
while increased connectivity leads to greater mobility. The dominant factor being the
connectivity of the phase, which results in the general trend of increasing phase relative
permeability with increasing phase saturation. Unlike the fluid-solid interfacial area, we do not
have a simple method of measuring the connectivity of a fluid phase, but given that we have
measurements of the solid-fluid interfacial areas we can infer differences in relative permeability
that cannot be attributed to trends in the solid-fluid interfacial areas are a result of connectivity.
Also the connectivity correlates with the fluid distribution, with the more connected wetting
phase occupying smaller pores, as was shown in figure 3.4.
How the wetting phase and non-wetting phase respond to these competing mechanisms can be
very different. In figure 3.6 we can see the net effect on the non-wetting phase relative
permeability is an overall decrease as the fluids become more disconnected (occupy larger
pores). Although there is less fluid-solid interfacial area this effect is dominated by the decrease
in connectivity. Unlike the non-wetting phase, the wetting phase relative permeability shows
very little dependence on the wetting strength of the porous medium for . We know that
there will not be a decrease in the connectivity of the wetting phase as the contact angle is
decreased, thus the small decrease in mobility between and the other homogenous-wet
84
states can only be attributed to an increase in fluid-solid interfacial areas. It is not surprising that
the non-wetting phase relative permeability is more greatly influenced by connectivity, being the
less connected phase, and the wetting phase relative permeability is more greatly influenced by
the fluid-solid interfacial area, being the generally more connected phase. However, regarding
the non-wetting phase relative permeability, a similar previous investigation reported the
opposite effect from what is reported here. Li et al. 2005 simulating two-phase flow in a
homogenous sphere pack using a multiple-relaxation-time Shan-Chen type multicomponent LB
model reported an increase in the non-wetting phase relative permeability for neutral wet states
( ) over strongly wet states ( ). This could be attributed to differences in the pore
space geometry resulting in the prevalence of increased non-wetting phase mobility due to the
decrease in solid-fluid interfacial area, over the decrease in mobility associated with decreased
connectivity as was observed here. The homogenous sphere pack used by Li et al. 2005 may
have a generally narrower distribution of pore apertures, meaning fewer large pore spaces for the
non-wetting phase to occupy, increasing the dependence of the non-wetting phase relative
permeability on the solid-fluid interfacial area. The interaction of these competing mechanisms
is sensitive to small differences in pore space geometry. In the mixed-wet states individual pores
can have mixed-wettability resulting in the competition of these mechanisms within each pore.
3.4.5 Relative Permeability of Mixed-Wet States
As was previously stated, the bounce-back density of the nodes in contact with the non-wetting
phase after fluid distribution is established in the homogenous-wetting simulations, are altered to
create a mixed-wet porous medium (figure 3.7). The mixed-wet state is the result of three
85
parameters, the initial homogenous-wettability, the saturation at which alteration takes place, and
the severity of alteration. Here we will alter the wettability of each of four homogenous-wetting
states at a wetting phase saturation near 0.5, with four increasingly severe alterations, resulting in
sixteen mixed-wet states. The fictional bounce-back density of nodes in contact with the non-
wetting phase are altered to preferentially wet the non-wetting phase with contact angles
( ). To avoid confusion, the initial
wetting phase is from here on referred to as the wetting phase of the mixed-wet states, and used
as the wetting phase saturation in figures. Also, the mixed-wet states will be referred to
according to the following indexing, , where is the for the initial wettability, and
is the for the altered wettability (i.e. would refer to a state that was initially
wet, then altered as previously stated to ).
The initial homogenous-wet states will use the same indexing as the mixed-wet states absent a
superscript. There are numerous mixed-wet states that could be considered using this
framework, we limit this investigation to studying the mixed-wet state resulting at .
These mixed-wet states are the result of wettability alteration to the homogenous-wet states at
, and therefore the mixed-wet states are dependent on the fluid distribution of the
phases at . The wetting phase at this saturation is described as occupying the smaller
pore spaces (figure 3.4), and is the more well connected phase. Thus the surfaces altered by the
non-wetting phase are disconnected, with an increasing fraction of the surface altered (increasing
fractional solid-fluid interfacial area) with increasing contact angle.
86
Figure 3.7: Images of the LB lattice with initial homogenous-wettability (A), the fluid distribution of the
non-wetting phase at the end of the simulation of the initial homogenous-wettability with
and (B), and the alteration of the bounce-back densities of the nodes in
contact with the non-wetting fluid (C). This image also summarizes the three parameters
determining the mixed-wet state, initial wettability (A), saturation of alteration (B), and
severity of alteration (C).
The relative permeability curves and the fractional solid-fluid interfacial areas for the mixed-wet
states are shown in figure 3.8. Recapping the observations of the homogenous-wet states, the
non-wetting phase was found to decrease with decreasing contact angle due to decreasing
connectivity, while the wetting phase was found to decrease with decreasing contact angle due to
increasing solid-fluid interfacial area. In general, the wettability alteration has little effect on the
wetting phase relative permeability. The fractional wetting phase solid-fluid interfacial areas of
the mixed-wet states are generally lower than those of the mixed-wet states. Unlike in the
homogenous-wet states this decrease does not correlate with an increase of the wetting phase
relative permeability. We can conclude that any differences in the wetting phase relative
permeability of mixed-wet states cannot be attributed to differences in the solid-fluid interfacial
area. Instead the insensitivity of the wetting phase relative permeability to wettability alteration
can be attributed to the fluid distribution of the wetting phase at the saturation of alteration. At
87
the wetting phase has developed a well-connected pathway (occupies smaller pores,
has a high fractional solid-fluid interfacial area). The solid in contact with the wetting phase
remains unaltered, thus, this connected path remains in the mixed-wet states. For the
mixed-wet states there is some decrease in the wetting phase relative permeability observed for
the mixed-wet state. This decrease is limited to the
mixed-wet state as a result of its
poorer connectivity of the established unaltered wetting phase pathways in comparison to those
of the ,
, and mixed-wet states. But even this small decrease only exists for a
mixed-wet state suffering a severe alteration. Overall, the wetting phase relative permeability is
unaffected by the wettability alteration, the established unaltered wetting phase pathways that
encourage the wetting phase to occupy the smaller pores dominates the mobility of the wetting
phase.
88
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
kr
Sw
0.1-0.00.10.20.3
-0.1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
1.2
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
kr
Sw
0.1-0.00.10.20.3
-0.1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
1.2
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
kr
Sw
0.1-0.00.10.20.3
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
kr
Sw
0.1-0.00.10.20.3
-0.1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
1.2
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
kr
Sw
0.1-0.00.10.20.3
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
kr
Sw
0.1-0.00.10.20.3
-0.1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
1.2
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
kr
Sw
0.1-0.00.10.20.3
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
kr
Sw
0.1-0.00.10.20.3
A)
C)
E)
G) H)
F)
D)
B)
89
Figure 3.8: Relative permeability and fractional fluid-solid interfacial areas determined by LB
simulations for the mixed-wet states of the originally homogenous-wet states, (A, B)
, (C, D) , (E, F) , and (G, H) ,
with increasing severity of alteration. The relative permeability and fractional fluid-solid
interfacial areas of the wetting and non-wetting phase are shown in black and gray,
respectively.
The non-wetting phase relative permeability is significantly altered in the mixed-wet states, and
as in the case of the wetting phase, the most significant effect is seen for the mixed-wet
states. For the ,
, and mixed-wet states the severity of the wettability alteration
has little effect on the mobility of the non-wetting phase, but the existence of a wettability
alteration does. This can be understood as a result of solid-fluid interaction on altered solid
surfaces. These surfaces are no longer dominated by the initial wetting phase, resulting in an
increase of the non-wetting phase fluid-solid interfacial area and a decrease in the non-wetting
phase mobility. The increase in the non-wetting phase relative permeability correlates with an
increase in the non-wetting phase fluid-solid interfacial area. In general, the non-wetting phase
becomes pinned by its affinity to these altered solid surfaces, the lubricating effect the wetting-
phase has on the non-wetting phase mobility is lost. The non-wetting phase responds to the
strength of alteration much as the wetting phase responded in the homogenous-wet states to
increasing strength of wettability. Unlike the ,
, and mixed-wet states, for the
mixed-wet states the severity of alteration appears to be proportional to the decrease in the
non-wetting phase relative permeability. This also can be considered in the context of the fluid
distribution of the homogenous-wet , being close to neutrally-wet the non-wetting phase
occupies smaller pore spaces and has a greater fluid-solid interfacial area than the more strongly
90
wetting scenarios. Thus, the alteration of the surface occurs not only on a larger fraction of the
solid surface, but also in the smaller pore spaces – much like the wetting phase in the
homogenous-wet states, the non-wetting phase relative permeability decreases due to the greater
fluid-solid interfacial areas. This increase in fluid-solid interfacial areas is reflected in the
measurements of non-wetting fluid-solid interfacial areas (figure 3.8). Incremental increases in
the non-wetting fluid-solid interfacial area for the mixed-wet states resulted in
incrementally decreasing non-wetting phase permeability. The non-wetting fluid-solid interfacial
area of the ,
, and mixed-wet states increases significantly with the existence of
an alteration, but the increase does not respond to increasing severity of alteration, which is
reflected in the relative permeability curves. Overall the non-wetting phase responded to the
wettability alteration with a decrease in relative permeability as a result of increasing non-
wetting fluid-solid interfacial areas.
Had the wettability alteration occurred at high wetting phase saturations we would expect the
same result to a lesser extent, simply due to less of the solid surface being altered. Wettability
alteration at lower wetting phase saturations would have a somewhat increasing effect, however,
not as much as one may at first think. The extreme example of altering the entire surface when
fully saturated with the non-wetting phase would result in a simple role reversal of the wetting
and non-wetting phases. This would suggest that at lower wetting phase saturations of
wettability alteration we could expect the wetting phase relative permeability to “rebound” and
begin to respond by increasing with decreasing fluid-solid interfacial area. However, saturations
below irreducible wetting phase saturation are unlikely to occur. At saturations near the
irreducible wetting phase saturation the wetting phase will dominate the smaller pore spaces.
91
And as occurred in our simulations, wettability alteration will not occur in these smaller pore
spaces. The wetting phase will continue to be pinned to these smaller pore spaces, and generally
will not experience any increase in mobility, as one might intuitively expect when considering
the extreme example of the entire surface being altered. At these saturations the non-wetting
phase will show a slightly greater effect than what we see here at , however the
differences will not be great. In our simulations of mixed-wet states the non-wetting phase
relative permeability has been significantly decreased by the wettability alteration. There is
very little mobility left to be lost.
Comparable studies to this one are limited for the most part to numerical experiments using pore
network modeling (Blunt et al. 2002, Valvatne and Blunt et al. 2004, Zhao et al. 2010, Gharbi
and Blunt 2012). There are a few notable differences between PNMs and LB models. As was
described in the background section, pore network models use networks of capillary tube
elements to represent the pore space topology, while LB models use direct translations of CMT
images to construct the lattice. This also means that wetting films are simulated by very
different means. In PNMs the wetting phase is simulated as thin films, allowing for very low
wetting phase saturations to maintain connectivity. Consequently in mixed-wet scenarios it is
possible for a pore to contain a thin film of one phase sandwiched between the other phase
occupying the center and corners of the pore (Blunt et al. 2002). The wetting phase films in a
lattice Boltzmann simulation are limited in their thickness to the resolution of the lattice, and
may or may not appear depending on the resolution of the lattice and the strength of the bounce-
back density ( ) used. Also PNMs simulate displacements, unlike here, where our
simulations are steady state. The PNM study of mixed-wettability in six different types of
92
limestone of Gharbi and Blunt 2012 offers a fair comparison to this work. This work
investigated the mixed-wet states of PNMs constructed from CMT images of limestone cores.
They also found a significant reduction in the non-wetting phase (oil) relative permeability for
wettability alteration fractions of 0.25 and 0.5 – similar to solid surface fractions altered at
here. However, they also report a significant reduction in the wetting phase relative
permeability, not seen here. This can be attributed to not only differences in the pore space
geometry, but also the method in which fluid films are simulated. At our wetting phase
remains well connected for homogenous-wet states , , and , and as was
previously stated there remains well-connected unaltered solid-surface for the wetting-phase to
be established, ensuring an insignificant effect on the wetting phase relative permeability for the
mixed-wet states. In the mixed-wet states of Gharbi and Blunt 2012 much of the wetting phase
existed as thin films, resulting in very low wetting phase permeability. The inability to simulate
fluid films beyond the resolution of the lattice is one of the weaknesses of the LB method.
However the ability of the LB method to simulate fluids in realistic pore geometries makes it a
very attractive technique.
3.5 CONCLUSIONS
We have presented a study of the dependence of relative permeability on wettability for both
homogenous-wet and mixed-wet states using LB modeling. Comparison between LB simulation
and experimental results show the LB simulations capture the trends of fluid-fluid and fluid-solid
interfacial areas well. Since our mixed-wet states are created by altering the wettability of the
93
solid in contact with the non-wetting phase it is important that these interfacial areas are
accurately captured to create realistic mixed-wet states.
The homogenous-wet states display a decrease in relative permeability for both phases with
decreasing contact angles. This is attributed here to the competition of two mechanisms that
increase and decrease fluid flow. With decreasing contact angle the non-wetting phase becomes
more disconnected and occupies larger pore spaces. This occupation results in less non-wetting
fluid-solid interfacial areas which acts to increase the mobility of the non-wetting phase,
however, the decreasing connectivity dominates this effect resulting in lower non-wetting phase
relative permeability. The decrease in the wetting phase relative permeability with decreasing
contact angle is a result of greater fluid-solid interfacial area which acts to decrease the mobility
of the wetting phase; any increases in connectivity of the wetting phase cannot overcome this
effect, due to the pinning of the wetting phase to the smaller pore spaces.
The pores of the mixed-wet state are themselves mixed, the interaction between the two
competing mechanisms controlling flow will also compete within individual pores. This can
lead to a rich assortment of fluid mobility behavior as a result of varying the wettability alone.
In this investigation we find that the connectivity of the wetting phase at the saturation of
wettability alteration is critical to the significance of the effect the alteration to a mixed-wet state
will have on the wetting phase mobility. In general, the severity of alteration is inconsequential
if the unaltered solid surface is well connected. As is the case for initial homogenous-wet states
with a at The non-wetting phase relative permeability demonstrates a
significant decrease in mobility proportional the severity of alteration when the solid surfaces
94
being altered includes a significant portion of the smaller pore spaces. As is the case for initial
homogenous-wet state with a at For initial homogenous-wet states with a
at the resulting mixed-wet states all show a significant decrease in the non-
wetting phase mobility, but no dependence on the severity of the alteration, even when the solid
surface is altered to neutrally wet. Overall the most important factor governing non-wetting
phase mobility in mixed-wet states is the fluid-solid interfacial areas, greater interfacial fluid-
solid surface areas lead to decreases in mobility. The responses to a wettability alteration at a
wetting phase saturation of 0.5 shown here, are comparable to what we would expect for
wettability alteration at lower wetting phase saturations due to the pinning of the wetting phase
to the smaller unaltered pore spaces.
The framework presented here to study the evolution of wettability alteration and its effect on the
relative permeability of two-phase flow in a simple porous medium could easily be applied to
natural consolidated and unconsolidated porous media. Our initial wettability states are
homogenous, for natural porous media the initial wettability will be determined by the mineral
constituents and their locations in the pore space geometry. We present simulations of varying
alteration severity, for natural porous media the level of wettability alteration is a result of
complex fluid-mineral interactions and would likely require bulk studies of wettability alteration
for individual mineral constituents. Given this knowledge can be determined using imaging and
laboratory techniques, LB simulations can be easily included to estimate fluid flow properties,
such as relative permeability, for a wide assortment of scenarios.
95
3.6 REFERENCES
Al-Futaisi, and Patzek T.W. Secondary imbibition in NAPL-invaded mixed-wet sediments.
Journal of Contaminant Hydrology 2004; 74:61-81. doi:10.1016/j.jconhyd.2004.02.005
Hui M., and Blunt M.J. Wettability on three-phase flow in porous media. Journal of Physical
Chemistry B 2000; 104:3833-3845.
Alipour Tabrizky V., Hamouda A. A., and Denoyel R. Influence of magnesium and sulfate ions
on wettability alteration of calcite, quartz, and kaolinite: Surface energy analysis. Energy and
Fuels 2011; 25:1667-1680. doi:10.1021/ef200039m.
Al-Raoush R.I. Impact of wettability on pore-scale characteristics of residual nonaqueous phase
liquids. Environmental Science and Technology 2009; 43:4796-4801. doi:10.1021/es802566s
Bakke S., and Oren P.E. 3-D pore-scale modeling of sandstones and flow simulations in the pore
networks. SPE Journal 1997; 2(2):136-149. doi:10.2118/35479-PA
Barranco Jr. F.T., Dawson H.E., Christener J.M., and Honeyman B.D. Influence of aqueous pH
and ionic strength on the wettability of quartz in the presence of dense non-aqueous-phase
liquids. Environmental Science and Technology 1997; 37:676-681.
96
Boek E.S., and Venturoli M. Lattice-Boltzmann studies of fluid flow in porous media with
realistic rock geometries. Computers and Mathematics with Applications 2010; 59:2305-2314.
doi:10.1016/j.camwa.2009.08.063
Blunt M.J. Effects of heterogeneity and wetting on relative permeability using pore level
modeling. SPE Journal 1997; 2(1):70-87. doi:10.2118/36762-PA
Blunt M.J. Pore level modeling of the effects of wettability. SPE Journal 1997; 2(4):494-510.
doi:10.2118/38435-PA
Blunt M.J., Jackson M.D., Piri M., and Valvatne P.H. Detailed physics, predictive capabilities
and macroscopic consequences for pore network models of multiphase flow. Advances in Water
Resources 2002; 25:1069-1089.
Brusseau M.L., Peng S., Schnaar G., and Constanza-Robinson M.S. Relationships among air-
water interfacial area, capillary pressure, and water saturation for a sandy porous medium.
Water Resources Research 2006; 42:W03501. doi:10.1029/2005WR004058
Brusseau M.L., Janousek H., Murao A., and Schnaar G. Synchrotron X-ray microtomography
and interfacial partitioning tracer test measurements of NAPL-water interfacial areas. Water
Resources Research 2008; 44:W01411. doi:10.1029/2006WR005517
97
Buckley J.S. and Liu Y. Some mechanisms of crude oil/brine/solid interactions. Journal of
Petroleum Science and Engineering 1998; 20:155-180. doi:10.2118/37230-PA
Coles M.E., Hazlett R.D., Spanne P., Soll W.E., Muegge E.L., and Jones K.W. Pore level
imaging of fluid transport using synchrotron X-ray microtomography. Journal of Petroleum
Science and Engineering 1998; 19:55-63.
Constanza-Robinson M.S., Harrold K.H., and Lieb-Lappen R.M. X-ray microtomography
determination of air-water interfacial area-water saturation relationships in sandy porous media.
Environmental Science and Technology 2008; 42: 2949-2956. doi:10.1021/es072080d
Culligan K.A., Wildenschild D., Christensen B.S.B., Gray W.G. and Rivers M.L. Pore-scale
characteristics of multiphase flow in porous media: A comparison of air-water and oil-water
experiments. Advances in Water Resources 2006; 29:227-238.
doi:10.1016/j.advwatres.2005.03.021
Dalla E., Hilpert M., and Miller C.T. Computation of the interfacial area for two-fluid porous
medium systems. Journal of Contaminant Hydrology 2002; 56:25-48.
Dixit A.B., McDougall S.R., and Sorbie K.S. A pore-level investigation of relative permeability
hysteresis in water-wet systems. SPE Journal 1998; 3(2):115:123. doi:10.2118/37233-PA
98
Evje S., and Hiorth A. A model for interpretation of brine-dependent spontaneous imbibition
experiments. Advances in Water Resources 2011; 34:1627-1642.
doi:10.1016/j.advwatres.2011.09.003
Fenwick D.H., and Blunt M.J. Network modeling of three-phase flow in porous media. SPE
Journal 1998; 3(1):84-96. doi:10.2118/38881-PA
Gharbi O., and Blunt M.J. The impact of wettability and connectivity on relative permeability in
carbonates: A pore network modeling analysis. Water Resources Research 2012; 48:W12513.
doi:10.1029/2012WR011877
Ghassemi A., and Pak A. Numerical study of factors influencing relative permeabilities of two
immiscible fluids flowing through porous media using lattice Boltzmann method. Journal of
Petroleum Science and Engineering 2011; 77:135-145. doi:10.1016/j.petrol.2011.02.007
Hao L., and Cheng P. Pore-scale simulations on relative permeabilities of porous media by
lattice Boltzmann method. International Journal of Heat and Mass Transfer 2010; 53:1908-1913.
doi:10.1016/j.ijheatmasstransfer.2009.12.066
Hazlett R.D., Chen S.Y. and Soll W.E. Wettability and rate effects on immiscible displacement:
Lattice Boltzmann simulation in microtomographic images of reservoir rocks. Journal of
Petroleum and Science Engineering 1998; 20:167-175
99
Hoiland L.K., Spildo K., and Skauge A. Fluid flow properties for different classes of
intermediate wettability as studied by network modeling. Transport in Porous Media 2007;
70:127-146. doi:10.1007/s11242-006-9088-x
Huang H., Thorne D.T., Schaap M.G., and Sukop M.C. Proposed approximation for contact
angles in Shan-and-Chen-type multicomponent multiphase lattice Boltzmann models. Physical
Review E 2007; 76:066701. doi: 10.1103/PhysRevE.76.066701
Iglauer S., Paluszny A., Pentland C.H., and Blunt M.J. Residual CO2 imaged with X-ray
microtomography. Geophysical Research Letters 2011; 38:L21403.
doi:10.1029/2011GL049680
Jackson M.D., Valvatne P.H., and Blunt M.J. Prediction of wettability variation and its impact
on flow using pore- to reservoir-scale simulations. Journal of Petroleum Science and
Engineering 2003; 39:231-246. doi:10.1016/S0920-4105(03)00065-2
Jadhunandan P.P., and Morrow N.R. Effect of wettability on waterflood recovery for crude-
oil/brine/rock systems. SPE Reservoir Engineering 1995; 10(1):40-46. doi:10.2118/22597-PA
Jerauld G.R., and Rathmell J.J. Wettability and relative permeability of Prudhoe Bay: A case
study in mixed-wet reservoirs. SPE Reservoir Engineering 1997; 12(1):58-65.
doi:10.2118/28576-PA
100
Joekar-Niaser V., Hassanizadeh, S.M., and Leijnse A. Insights into the relationships among
capillary pressure, saturation, interfacial area and relative permeability using pore network
modeling. Transport in Porous Media 2008; 74:201-219. doi:10.1007/s11242-007-9191-7
Kovscek A.R. Wong H., and Radke C.J. A pore-level scenario for the development of mixed
wettability in oil reservoirs. AIChE Journal 1993; 39(6):1072-1085.
Kumar M., Fogden A., Senden T., and Knackstedt M. Investigation of pore-scale missed
wettability. SPE Journal 2012; 17(1):20-30. doi: 10.2118/129974-PA
Landry C.J., Karpyn Z.T., and Piri M. Pore-scale analysis of trapped immiscible fluid structures
and fluid interfacial areas in oil-wet and water-wet bead packs. Geofluids 2011; 11:209-227.
doi: 10.1111/j.1468-8123.2011.00333.x
Lebedeva E.V., and Fogden A. Micro-CT and wettability analysis of oil recovery from
sandpacks and the effect of waterflood salinity and kaolinite. Energy and Fuels 2011; 25:5683-
5694. doi:10.1021/ef201242s
Lee T.C., Kashyap R.L. and Chu C.N. Building skeleton models via 3D medial surface/axis
thinning algorithms. CVGIP, Graphical Models and Image Processing 1994; 54:462-478.
Li H., Pan C., and Miller C.T. Pore-scale investigation of viscous coupling effects for two-phase
flow in porous media. Physical Review E 2005; 72:026705. doi:10.1103/PhysRevE.72.026705
101
Lindquist B., Lee S.M., Jones K.W., and Spanne P. Medial axis analysis of void structure in
three-dimensional tomographic images of porous media. Journal of Geophysical Research 1996;
101(B4):8297-8310.
Martys N.S., and Chen H. Simulation of multicomponent fluids in complex three-dimensional
geometries by the lattice Boltzmann method. Physical Review E 1996; 53(1):743-750.
Oren P.E., Bakke S., and Arntzen O.J. Extending predictive capabilities to network models.
SPE Journal 1998; 3(4):324-336. doi:10.2118/52052-PA
Oren P.E., and Bakke S. Reconstruction of Berea sandstone and pore-scale modeling of
wettability effects. Journal of Petroleum Science and Engineering 2003; 39:177-199.
doi:10.1016/S0920-4105(03)00062-7
Pan C., Hilpert M., and Miller C.T. Lattice-Boltzmann simulation of two-phase flow in porous
media. Water Resources Research 2004; 40:W01501. doi:10.1029/2003WR002120
Pan C., Luo L.S., and Miller C.T. An evaluation of lattice Boltzmann schemes for porous
medium flow simulation. Computers & Fluids 2006; 35:898-909.
doi:10.1016/j.compfluid.2005.03.008
102
Porter M.L., Schaap M.G., and Wildenschild D. Lattice-boltzmann simulations of the capillary
pressure-saturation-interfacial area relationship for porous media. Advances in Water
Resources; 32:1632-1640. doi:10.1016/j.advwatres.2009.08.009
Prodanovic M., Lindquist W.B., and Seright R.S. 3D image-based characterization of fluid
displacement in a Berea core. Advances in Water Resources; 30:214-226.
doi:10.1016/j.advwatres.2005.05.015
Prodanovic M., Lindquist W.B., and Seright R.S. Porous structure and fluid partitioning in
polyethylene cores from 3D x-ray microtomographic imaging. Journal of Colloid and Interface
Science 2006; 298:282-297. doi:10.1016/j.jcis.2005.11.053
Qian Y.H., d’Humieres D., and Lallemand P. Lattice BGK Models for Navier-Stokes Equations.
Europhysics Letters 1992; 17(6):479-484. doi:10.1209/0295-5075/17/6/001
Ramstad T., Oren P.E., and Bakke S. Simulation of two-phase flow in reservoir rocks using a
lattice Boltzmann method. SPE Journal 2010; 15(4):917-927. doi: 10.2118/124617-PA
Ramstad T., Idowu N., Nardi C., and Oren P.E. Relative permeability calculations from two-
phase flow simulations directly on digital images of porous rock. Transport in Porous Media
2012; 94:487-504. doi:10.1007/s11242-011-9877-8
103
Reeves P.C., and Celia M.A. A functional relationship between capillary pressure, saturation,
and interfacial area as revealed by a pore-scale network model. Water resources research 1996;
32(8):2345-2358.
Robin M., Rosenberg E., and Fassi-Fihri O. Wettability studies at the pore level: A new
approach by use of cryo-SEM. SPE Formation Evaluation 1995; 10(1):11-19.
doi:10.2118/22596-PA
Saraji S., Goual L., and Piri M. Adsorption of asphaltenes in porous media under flow
conditions. Energy and Fuels 2010; 26:6009-6017. doi:10.1021/ef100881k.
Schaap M.G., Porter M.L., Christensen B.S.B, and Wildenschild D. Comparison of pressure-
saturation characteristics derived from computed tomography and lattice Boltzmann simulations.
Water Resources Research; 43:W12S06. doi:10.1029/2003WR002120
Schembre J.M., Tang G.Q., and Kovscek A.R. Wettability alteration and oil recovery by water
imbibition at elevated temperatures. Journal of Petroleum Science and Engineering 2006;
52:131-148. doi:10.1016/j.petrol.2006.03.017
Shan X., and Doolen G. Multicomponent lattice-Boltzmann model with interparticle interaction.
Journal of Statistical Physics 1995; 81:379-393.
104
Silin D., Tomatsu L., Benson S.M., and Patzek T.W. Microtomography and pore-scale modeling
of two-phase fluid distribution. Transport in Porous Media; 86:495-515. doi:10.1007/s11242-
010-9636-2
Spiteri E.J., Juanes R., Blunt M.J., and Orr Jr. F.M. A new model of trapping and relative
permeability hysteresis for all wettability characteristics. SPE Journal; 13(3):277-288.
doi:10.2118/96448-PA
Sukop M.C., Huang H., Chen L.L., Deo M.D., Oh K., and Miller J.D. Distribution of multiphase
fluids in porous media: Comparison between lattice Boltzmann modeling and micro-x-ray
tomography. Physical Review E 2008; 77:026710. doi: 10.1103/PhysRevE.77.026710
Svirsky D.S., van Dike M.I.J., and Sorbie K.S. Prediction of three-phase relative permeabilities
using a pore-scale network model anchored to two-phase data. SPE Reservoir Evaluation and
Engineering; 10(5):527-538. doi:10.2118/89992-PA
Thovert J.F., Salles J., and Adler P.M. Computerized characterization of the geometry of real
porous media: their discretization, analysis and interpretation. Journal of Microscopy 1993;
170(1):65-79.
Turner M.L., Knufing L., Arns C.H., Sakellariou A., Senden T.J., Sheppard A.P., Sok R.M.,
Limaye A., Pinczewski W.V., and Knackstedt M.A. Three-dimensional imaging of multiphase
flow in porous media. Physica A 2004; 339:166-172. doi:10.1016/j.physa.2004.03.05
105
Vogel H.J., Tolke J., Schulz V.P., Krafczyk M., and Roth K. Comparison of a Lattice-
Boltzmann model, a full-morphology model, and a pore network model for determining capillary
pressure-saturation relationships. Vadose Zone Journal 2005; 4:380-388.
doi:10.2136/vzj2004.0114
Yu L., Evje S., Kleppe H., Karstad T., Fjelde I., and Skjaeveland S.M. Spontaneous imbibition
of seawater into prefentially oil-wet chalk cores – Experiments and simulations. Journal of
Petroleum Science and Engineering 2009; 66:171-179. doi:10.1016/j.petrol.2009.02.008
Zhao X., Blunt M.J., and Yao J. Pore-scale modeling:effects of wettability on waterflood oil
recovery. Journal of Petroleum Science and Engineering; 71:169-178.
doi:10.1016/j.petrol.2010.01.011
106
CHAPTER 4: PORE-SCALE LATTICE BOLTZMANN MODELING AND 4D X-RAY
COMPUTED MICROTOMOGRAPHY IMAGING OF FRACTURE-MATRIX FLUID
TRANSFER
4.1 SUMMARY
We present sequential x-ray computed microtomography (CMT) images of matrix drainage in a
fractured sintered glass granule pack. Sequential imaging captured the capillary-dominated
migration of the non-wetting phase front from the fracture to the matrix in a brine-surfactant-
Decane system. The sintered glass granule pack was designed to have minimal pore space
beyond the resolution of CMT imaging, so that the pore space of the matrix connected to the
fracture could be captured in its entirety. The segmented image of the pore space was then
directly translated to a lattice to simulate the transfer of fluids between the fracture and the
matrix using lattice Boltzmann (LB) modeling. This provided us an opportunity to validate the
modeling technique against experimental images at the pore-scale. Although the surfactant was
found to alter the wettability of the originally weakly oil-wet glass to water-wet, the fracture-
matrix fluid transfer is found to be a drainage process, showing little to no counter-current
migration of the oil-phase. The LB simulations were found to closely match experimental rates
of fracture-matrix fluid transfer, equilibrium saturation, irreducible wetting phase saturation and
fluid distributions.
107
4.2 INTRODUCTION
The displacement of fluids in fractured rock by transfer of fluids between the high-permeability
low-storage fractures and low-permeability high-storage surrounding rock is a common process
that occurs during hydrocarbon recovery, carbon dioxide injection, and groundwater
contamination. During an injection into, or flooding of, a fractured rock, fluid migration is
determined by the complex interaction of capillary and viscous forces that control flow in the
fractures, as well as by the transfer of fluids to and from the matrix. Predicting this migration
requires a rigorous investigation into the mechanics that determine the transfer of fluids to and
from the matrix. Understanding the physics of this process can be greatly enhanced by
investigating fluid transfer at the interface of the fracture and matrix at the pore-scale.
Experimental studies of fracture-matrix fluid transfer at the pore-scale have been limited, for the
most part, to etched-glass micromodels and two-dimensional glass bead models. Micromodels
are transparent, two-dimensional representations of pore networks (Wan et al. 1996,
Karadimitriou and Hassanizadeh 2012). Using these models, a wide variety of multiphase flow
processes involving fracture-matrix transfer have been investigated in recent years including
spontaneous imbibition into water-wet media (Rangel-German and Kovscek 2006, Hatiboglu and
Babdagli 2008, Hatiboglu and Babdagli 2011), CO2 flooding of naturally fractured oil-wet
reservoirs at immiscible, near-miscible, and miscible pressure (Er et al. 2010), continuous gas,
water and alternating gas-water injection in oil-wet naturally fractured reservoirs (Dehghan et al.
2012), and multiphase flow in multiple intersecting fractures (Buchgraber et al. 2012). These
108
studies have provided invaluable insight into the physics involved in multiphase fracture-matrix
fluid transfer; however, their limitation to two-dimensions assumes flow in the missing third
dimension will have minimal impact on fluid flow. In the investigation presented here, the third
dimension is retained by utilizing X-ray computed microtomography (CMT), which is a non-
invasive three-dimensional imaging technique. This allows us to image fracture-matrix fluid
transfer of a three-dimensional porous media at the pore-scale. Also, images obtained of the
pore space can be directly inputted into pore-scale lattice Boltzmann models for further
investigation of the mechanics controlling flow.
X-ray computed microtomography (CMT) has become a commonly used method of imaging
pore spaces and fluid distributions of unconsolidated sands and beadpacks (Culligan et al. 2006,
Brusseau et al. 2006, Brusseau et al. 2008, Constanza-Robinson et al. 2008, Al-Raoush et al.
2009, Lebedeva and Fogden 2011), rock (Coles et al. 1998, Turner et al. 2004, Prodanovic et al.
2007, Iglauer et al. 2011, Silin et al. 2011, Kumar et al. 2012) and fractures (Karpyn et al. 2007)
at the pore-scale in three dimensions. Additionally, numerous investigations using CMT have
imaged artificially induced (Ketcham et al. 2010) and naturally occurring fractures in rocks
(Keller et al. 1999, Renard et al. 2009, Wennberg et al. 2009). Due to advances in CMT imaging
technology within the past two decades, it is now possible to image pore spaces and fluid
distributions at the pore scale in three dimensions. However, carrying out experiments using
CMT is time-consuming and costly. Robust, pore-scale models offer the possibility of simulating
a wide variety of scenarios in order to improve our predictive capabilities regarding fracture-
matrix fluid transfer, at the cost of only computational time. Here, we focus on the use of lattice
Boltzmann (LB) models, although numerous pore-scale modeling techniques currently exist
109
(Meakin and Tartakovsky 2009). Multiphase LB models have been validated against
experimental measurements of capillary pressure in bead packs (Pan et al. 2004, Schaap et al.
2007, Porter et al. 2009) and sandstones (Ramstad et al. 2010, Ramstad et al. 2012), as well as
experimental measurements of relative permeability in sphere packs (Ghassemi and Pak 2009,
Hao and Cheng 2010) and sandstones (Boek and Venturoli 2010, Ramstad et al. 2010).
Although there has been a fair amount of validation of LB methods via matching of experimental
measurements of macroscale flow properties, few have compared fluid distributions imaged
using CMT with LB model results. Sukop et al. 2008 compared two-phase fluid distributions in
a sand pack imaged using CMT to distributions resulting from LB simulations. The saturation
per slice was found to match experimental results well. Porter et al. 2009 compared interfacial
fluid-fluid areas measured by CMT to LB displacement simulations, and found a good match for
drainage simulations.
We are interested in investigating fracture-matrix fluid transfer at the pore-scale using
experimental, sequential, three-dimensional CMT imaging of fluid transfer and lattice Boltzmann
modeling. A synthetic core was manufactured in order to contain a minimal amount of pore
space unresolvable by CMT imaging, and was subsequently fractured to serve as our fractured
porous medium. This fractured core is more closely analogous to a real rock than two-
dimensional micromodels, and it also has the previously stated advantage of allowing fluids to
move in three directions. One of the advantages of using LB models in conjunction with CMT is
the ease of translating images of the pore space into a simulation lattice. Here, we simply
translate the CMT image of the pore space 1-to-1, image voxel to lattice node, thus retaining the
complex solid boundary conditions of the pore space in the model. Also, by manufacturing the
110
core to have minimal pore space unresolvable by CMT, we can be confident that the modeled
pore space is a close representation of the physical pore space observed during the experiment.
4.3 MATERIALS AND METHODS
4.3.1 Fractured Porous Medium
The fractured porous medium is a sintered glass granule pack, fractured using a modified
Brazilian test. Crushed glass was sieved for granules ranging from 115 to 210 μm in diameter.
These granules were placed in a 7.62 cm diameter, 10.2 cm long mullite tube and sintered
following the temperature cycle summarized in table 4.1. Cores with a diameter of 1.27 cm were
drilled from these sintered glass granule packs. A single, tensile fracture was induced along the
axis of the core, and the two resulting halves of the core were glued along the edges of the
fracture to ensure the two halves are immobile. The resulting fractured core is 3.12 cm in length
and ~1.5 cm in diameter.
Table 4.1: Temperature cycles for sintering of glass granules.
Step Time [h] Cumulative Time
[h] Temperature [C]
Increase 8 8 25 → 600
Hold 1.5 9.5 600
Increase 1 10.5 600 → 645
Hold 1 11.5 645
Decrease 1 12.5 645 → 600
Hold 1.5 14 600
Decrease 16 30 600 → 25
111
4.3.2 Procedure
The core was placed between the inlet and outlet of the core holder and encased in fuel-resistant
Teflon heat-shrink tubing to create an airtight seal at both the inlet and outlet of the core holder.
The core abuts the face of the inlet and outlet ports. This face was designed with crisscrossing
channels in order to uniformly spread the injected/received fluid across the face of the inlet/outlet
ports for even fluid delivery/recovery to the core. The core and core holder assembly was placed
vertically in the CMT scanner under vacuum. The core was then scanned in its entirety to
acquire a dry image of the pore space. Following the dry scan, the core was presaturated with
the oil phase, 40% wt iodododecane in decane. The decane was doped with iodododecane in
order to create contrast between phases in the CMT images. The fully oil-saturated core was
then scanned in its entirety to acquire an image of the connected pore space. After scanning, the
oil phase was displaced by injection of the water phase, 5%wt NaCl DI water. The interfacial
tension between the water and oil phase was measured using the sessile drop method, and found
to be 42.86 mN/m. The flow rates throughout the experiment were maintained at a rate of 0.1 to
1.0 ml/min. The balance of viscous and capillary forces are commonly measured by the
capillary number, ⁄ , a dimensionless measure of viscous and capillary forces. The
displacements here have capillary numbers between 10-5
and 10-7
. All displacements in the
experiment were considered to be capillary-dominated.
The CMT scanner is equipped to provide real-time radiographs (2D side-view images) of a 0.532
cm length of the core during water injection. During the water injection no water was observed
entering the matrix. At the end of the initial water phase injection, the core was scanned in its
112
entirety, confirming that the water phase had only succeeded in displacing the oil phase from the
fracture. Although glass is often considered a water-wet material, the presaturation with the oil
phase and exposure to x-rays rendered the core weakly oil-wet. The water phase, being the non-
wetting phase, occupies only the largest portions of the pore space necessary to connect the inlet
with the outlet, i.e. the fracture. Due to the vertical orientation of the core and the difference in
densities of the fluids, there is a difference in the pressure ( ) of the heavier water-occupied
fracture and lighter oil-occupied matrix, equal to , where and are the
densities of water and oil, is the gravitational constant, and is the distance from the top or
inlet to the core. Because there are no viscous forces driving the water phase into the matrix,
only capillary and gravitational forces are acting in the transfer of fluid to and from the matrix.
The capillary pressure ( ) is the difference in pressure between the non-wetting ( ) and
wetting phase ( ). In our system the non-wetting phase occupies the fracture, while the wetting
phase occupies the matrix, thus . For the non-wetting phase to penetrate the matrix, the
difference in pressure due to gravity forces must overcome the capillary entry pressure ( ).
The capillary entry pressure is the capillary pressure necessary for the non-wetting phase to
invade a pore occupied by the wetting phase. The capillary entry pressure is directly
proportional to the interfacial tension, and can be lowered by decreasing the interfacial tension.
In order to lower the interfacial tension of our fluid pair, 2 ml/l of dish soap was added to the
water phase, lowering the interfacial tension to 0.30 mN/m. The surfactant-water phase is
injected at the same rates as the water-phase injection. Once the surfactant-water phase
injection is commenced the surfactant-water phase mixes with the water phase already occupying
the fracture. Once the surfactant becomes established in the fractured porous medium, the
113
interfacial tension of the system will drop and the surfactant water-phase will begin to displace
the oil phase (wetting phase) from the matrix. During the injection of the surfactant water phase,
a 0.506 cm section of the core was sequentially scanned, producing 14 3D images of the water
phase entering the matrix. This volume of the core is referred to as the region of interest (ROI)
from this point on. Each scan takes approximately 10 minutes, with time in between scans
ranging from 1 to 20 minutes. The injection is not stopped during the scans. The movement of
the water phase into the matrix was slow enough that the moving interface did not appreciably
blur our images. The sequential scanning of the water phase invasion was ended after 220
minutes, at which time the core was scanned in its entirety.
4.3.3 CMT Imaging
X-ray computed microtomography has become a popular technique to non-invasively image the
pore spaces of porous media in 3D. A CMT “scan” involves taking a sequence of 2-D
radiographs of a rotating sample and mathematically reconstructing them into a 3-D image
composed of voxels. Each voxel is much like a pixel in a 2D image, and contains a CMT
registration number, which is a relative measure of the x-ray attenuation of the material
occupying a particular voxel. Images reported here were taken at Pennsylvania State
University’s Center for Quantitative x-ray Imaging (CQI). Scans of the core in its entirety
resulted in images 1024x1024x1980 voxel3, while sequential scans during the water phase
invasion of the matrix were 1024x1024x321 voxel3. Only a section of the core could be
monitored during transient imaging because imaging of a larger section would have come at a
114
loss to the overall number of images that could be acquired. All scans have a voxel resolution of
0.015762 x 0.015762 x 0.015767 mm3, with the latter representing height, i.e. slice thickness.
4.3.4 Image Processing
The CMT images must be further processed in order to determine meaningful information. First,
all images are filtered using a 3x3x3 kernel median filter to remove salt and pepper type noise.
The CMT registration numbers of the three phases present, i.e. glass, oil phase, and water phase,
are significantly different, allowing for identification of the phase occupying each voxel.
Segmentation here was accomplished through simple thresholding. Other, more complex
thresholding techniques, such as bi-level thresholding with indicator kriging, were attempted, but
did not appear to result in any visually apparent improvement in segmentation. In simple
thresholding, two phases are segmented according to a threshold CMT value. Voxels with CMT
registration numbers below the threshold are labeled as one phase and those with CMT
registration numbers above the threshold are labeled as the other phase. To segment our three
phase images, a two-step segmentation process was used. First, the glass and pore spaces were
segmented in the presaturation image using a threshold determined by fitting a Gaussian curve to
the peak in the CMT registration number frequency distribution corresponding to the glass. The
threshold was taken as the CMT registration number corresponding to where the Gaussian curve
has a frequency value below 1% of the peak value; all voxels with CMT registration numbers
above this value are considered to belong to the pore space. This segmentation was then used to
mask the three phase images. By subtracting out the glass, only two phases, the oil phase and
water phase, are left. These two phases were then segmented in the same manner as the
115
segmentation of the glass from the pore space. With the images subsequently segmented into
glass, oil phase, and water phase, measurements of the pore space and fluid distribution could be
made.
However, before measurements associated with the matrix and fracture could be taken, the pore
space had to be labeled as belonging to either the fracture or the matrix. The fracture and matrix
pore spaces are connected, and the distinction between them is not obvious. To identify the pore
space associated with the fracture and the matrix, a process of image erosion, isolation, and
dilation, detailed in a previous study of ours (Landry and Karpyn 2012), was used. The matrix
measurements are taken from the portion of the matrix parallel to the fracture and within 0.5 cm
of the fracture face. After segmentation of the images and labeling of the pore space, volume
measurements could be made from the image.
4.4 RESULTS
4.4.1 Pore Space Analysis
During the surfactant fluid penetration of the matrix, it became clear that the matrix on either
side of the fracture was appreciably different, and that these differences resulted in significantly
different rates of water phase invasion into the matrix. To measure the characteristics of the
matrix separately, for what will be referred to from here on as the right and left sides, a line of
division was determined from the center point of the fracture aperture measurements. In our
images, the fracture is oriented perpendicular to image voxels, therefore we found it appropriate
to measure the so-called “vertical aperture” ( ), i.e. the distance between the fracture walls. An
116
image of the fracture, and a corresponding mean fracture aperture profile, are shown in figure
4.1. The mean fracture aperture was measured for each slice. The ROI is also highlighted in
figure 4.1.
Figure 4.1: Image of the fracture with the ROI highlighted (A), and the corresponding mean fracture
aperture and contact area profiles with the ROI indicated by arrows.
Overall, the fracture profile shows two wavelengths of decreasing and increasing mean aperture,
with the two minimums corresponding to significant contact areas between the fracture walls.
The portion of the core to be imaged sequentially during the surfactant water phase injection of
the matrix contains, at its center, a large “asperity”, meaning that the aperture is smaller than the
average pore aperture and is thus indistinguishable from the matrix using the labeling method
A) B)
117
described here. This portion of the core was chosen because it is located near the center and
contains a portion of the fracture that has a significantly varying aperture and the occurrence of
an asperity. One of the objectives of this study is to investigate the dependence of fracture-
matrix fluid transfer on the fracture aperture and the existence of asperities.
The porosity of the right and left matrix was calculated by summing the voxels of the
segmented, labeled pore space images from the presaturation scan. Calculating the porosity for
each slice, a profile of the porosity is presented in figure 4.2. The porosity of the core varies
from ~0.1 to ~0.35, with the lowest porosities being found at the bottom and top of the core.
Also, the porosity of the left matrix is generally slightly greater than that of the right matrix as a
result of the sintering process. The top, bottom, and sides, with the left matrix being closer to the
center, of the sintered sample reached greater temperatures, resulting in greater compaction. The
porosity of the left and right matrix of the portion of the core to be sequentially imaged is 0.3030
and 0.2698, respectively. Although the porosity gives a bulk measure of the volume of pore
space present, and provides information in regards to the level of sintering compaction, the
displacement of the wetting phase in the matrix by the non-wetting phase in the fracture is not
dependent on the porosity. Instead, this displacement is strongly dependent on the pore aperture
of the matrix. The porosity and the pore aperture are not necessarily related, it is both possible to
have a high porosity, small pore aperture porous medium and a low porosity, large pore aperture
porous medium. Another objective of this study was to design a synthetic 3D porous medium
that could be imaged with minimum unresolvable pore space, thereby allowing very accurate and
complete measurements of pore space geometry, such as the matrix pore aperture distribution.
118
Figure 4.2: Image of the matrix pore space showing the fracture saturated with water phase and the
matrix saturated with the oil phase, the ROI volume is highlighted (A). Also the
corresponding porosity (B) and mean aperture (C) profiles with the ROI indicated with
arrows.
A significant amount of literature regarding the description of pore space topology exists
(Thovert et al. 1993, Lindquist et al. 1996, Oren and Bakke 1997, Prodanovic et al. 2006).
Although there are numerous measurements that can be made regarding the pore space topology,
we focused on the pore aperture distribution. In order to measure the pore aperture distribution
of the pore space, first a skeleton, or medial axis, of the pore space must be constructed by
thinning algorithms (Lee et al. 1994) using the software Avizo Fire 6.3. The skeleton of the pore
space is a connected network of voxels that represents the midpoint between solid walls in the
pore space. Each voxel in the skeleton can be thought of as the center of a sphere that makes
contact with the wall in at least two places. The pore aperture is the maximum radius of a
A) B) C)
119
sphere that can occupy a pore. To determine the pore aperture distribution, a “sphere packing”
of sequentially smaller spheres are strung along the skeleton. First, the voxels of skeleton are
labeled with their distance to the pore wall. Next, the labeled skeleton is searched for voxels that
fall within a user-defined range. These voxels then become the center of a sphere – with a radius
equal to the distance to the closest pore wall – placed in the pore space. The algorithm treats all
voxels that fall within this sphere as solids in order to avoid overlapping measurements of pore
aperture between spheres from different aperture ranges. An example image of the pore
apertures is shown in figure 4.3. The mean pore aperture is defined here as the volume-weighted
mean pore aperture,
∑
where represents all pore apertures measured, is the pore aperture, and is the volume
weight of each pore aperture measured,
∑
The weight of each pore aperture is equal to the volume ( ) of the associated sphere divided by
the sum of all pore aperture sphere volumes. The mean pore aperture profile generally follows
the trends seen in the porosity profile, with the largest pore apertures occurring in the middle of
the core for both of the matrixes, and the left matrix pore apertures being generally greater than
the right (figure 4.2). The fracture and pore aperture distributions for the ROI are given in figure
4.4. The fracture and pore aperture distributions are lognormal, similar to what is observed in
CMT images of Berea sandstone (Prodanovic et al. 2007) and fractures in Berea sandstone
(Karpyn et al. 2007). The mean pore aperture of the left and right matrix is 0.0770 and 0.0721
(mm), respectively.
120
Figure 4.3. Example 0.1183 cm
3 image of sphere-packing method used to measure pore aperture
distribution. The individual pore apertures are measured as the radius of the spheres
seen in this image. The red, purple, teal, green, yellow, and light green represent pore
aperture measurements between , ,
, , , and mm,
respectively. The background shows the solid in black and the pore space in white.
121
Figure 4.4. The fracture aperture distribution and the pore aperture distribution for the portion of the
core to be imaged sequentially during water invasion of the matrix.
4.4.2 Immiscible Displacement
After the core is presaturated with the oil phase, the water phase without the addition of a
surfactant is injected. The oil phase is nearly completely displaced from the fracture, with water
phase saturations in the fracture, , while the matrix has yet to be invaded by the
water phase. Only a small amount of the oil phase is retained in the fracture near the top of the
core, and as is evident in the image presented in figure 4.5, the contact angle through the oil
phase is quite large, close to neutral wetting conditions. The thorough displacement of the oil
phase from the fracture by the water phase can be attributed to the weakly oil-wet condition. The
affinity of the wetting phase for the solid is too weak for the wetting phase to be retained in the
smaller apertures of the fracture. Also, as was previously mentioned, the non-wetting water
phase does not invade the matrix during the initial water phase injection due to capillary forces,
which can be decreased by the addition of a surfactant.
122
Figure 4.5. Image of oil phase (white), water phase (black), glass (gray) contact in the fracture after
initial water phase injection. The contact angle appears in this image to be large, near
neutral wetting conditions.
At the beginning of the surfactant water phase injection, the non-wetting water phase is already
established in the fracture. Thus, the fracture-matrix fluid transfer occurs in what has been
previously referred to as the “instantly-filled” regime (Rangel-German and Kovscek, 2006), i.e.
fracture-matrix fluid transfer does not occur until after the displacing fluid has been established
in the fracture, as opposed to the “filling” regime in which fracture-matrix transfer occurs
concurrently with the advancing displacement in the fracture. The regime of fracture-matrix
fluid transfer is often determined by the velocity at which the displacing front advances in the
fracture, with a faster moving front resulting in an “instantly-filled” fracture. The flow presented
here is very slow; however, it represents an instantly-filled fracture-matrix transfer due to the
delay in the injection of surfactants required for fluids to penetrate the matrix.
123
A few example images from the sequential scanning of the surfactant water phase injection of
the matrix are shown in figure 4.6. During the sequential scanning there are some small amounts
of the oil phase that appear in the fracture (figure 4.6). However, the total volume of the oil
phase in the fracture is much smaller than the amount of the oil phase being displaced from the
matrix suggesting there is very little, if any, movement of the oil phase from the matrix to the
fracture. The displaced oil phase from the matrix migrates in the direction of flow through the
matrix to the outlet. The small amount of oil found in the fracture appears to result from the
snap-off of oil already in, or near, the fracture (figure 4.6). This suggests that the surfactant is
also acting to alter the wettability of the solid to water-wet. The surfactant of the soap used is
sodium dodecyl sulfate (SDDS), an anionic surfactant, which has been shown to decrease the
water phase contact angle with quartz (Zdziennicka et al. 2009). We would expect that the
functional surface sites of quartz would be similar to the glass used here, leading to the observed
wettability alteration to water-wet.
124
Figure 4.6. CMT images from the sequential scanning of the surfactant water phase injection
showing a volume rendering (left) and a slice from a height of 1.0704 cm (right). In the
volume renderings the water phase is dark orange, the solid is light orange and the oil
phase is white. In the slices the water phase is light gray, the glass is dark gray, and the
oil phase is white.
125
The effect this wettability alteration has on the transfer of fluids is discussed in the following
pore-scale analysis. The non-wetting phase saturation of the matrix ( ) as a function of the
square root of time ( ⁄ ) is shown in figure 4.7. At the start of the surfactant water phase
injection, the surfactant must first mix with the water phase already present, and establish a
monolayer at the interface of the oil and water phase, before interfacial tension will be
significantly reduced. As the interfacial tension is reduced, we observe an increasing rate of
wetting-phase displacement, referred to in figure 4.7 as “early time”, which stabilizes at a near
constant rate, , with rate being a linear function of ⁄ , ⁄ , where
⁄ is
the time the displacement is initiated. This stable displacement of brine-surfactant-Decane is the
focus of this study, considering the majority of the displacement takes place after the surfactant
has been established, and the interfacial tension significantly reduced. The ⁄
for the
displacement is not a set number due to the slowly decreasing interfacial tension at early time.
Instead, we can determine an apparent start time for the stable displacement as determined by
fitting the rate function by linear regression. Fitting experimental measurements for ⁄
resulted in correlation coefficients of 0.9893 and 0.9801 for the right and left matrix, respectively
(figure 4.7). Including earlier times resulted in lower correlation coefficients, and excluding
⁄
did not result in significantly greater correlation coefficients. Therefore we can be
fairly confident that the stable displacement occurs in this time frame. The apparent ⁄
is 4.48
and 4.32 (min1/2
) for the right and left matrix, respectively. Both sides are in agreement on the
apparent ⁄
. The is 0.0787 and 0.0551 (min-1/2
) for the right and left matrix, respectively.
There is a significant difference in rate observed between the right and left matrix. Given that
the only variable between the right and left matrix is the pore space geometry, the significance of
126
the effect matrix pore space geometry has on fracture-matrix fluid transfer is highlighted. To
further understand the physics involved in this displacement, we compared the experimental
pore-scale fluid distributions to the fluid distributions of a pore-scale LB model simulating a
capillary-dominated drainage of the right and left matrix – separately but equally – in which the
only variable in the simulation of the right and left matrices will be pore space geometry.
Figure 4.7. Non-wetting phase saturation of the right and left matrix as determined from sequential
CMT imaging. Early time refers to displacement before the interfacial tension reduction
has stabilized. The linear fits shown represent a displacement following the rate function,
⁄
.
Early
Time
Stable
Displacement
127
4.4.3 LB Simulation of Fracture-Matrix Fluid Transfer
For details of the LB model used refer to chapter 3. To model the surfactant water phase
movement from the fracture to the matrix a 270x150x150 voxel3 (0.4256 x 0.2365 x 0.2365 cm
3)
sample volume (figure 4.8) of the pore space from each matrix is translated 1-to-1, voxel to
lattice node, for lattice Boltzmann simulation. This translation maintains the complexity of the
pore space solid boundary conditions – the image is directly translated into fluid and bounceback
nodes. The size of the image volume to be simulated is limited by the prohibitive computational
requirements of the LB model. Simulating the entire matrix volume scanned would be optimal,
although, it should be noted, that the sample volume used is larger than the representative
elementary volume determined by porosity and specific solid surface area measurements not
shown here. The simulation lattices are bound by solid walls orthogonal to the direction of flow,
and pressure boundaries 5 lattice layers thick at the ends. These pressure boundaries represent
the non-wetting phase occupied fracture, and the pushback of the capillary force of the wetting
phase occupied matrix. Images of the sample pore space used in LB simulations, and an
illustration of the LB simulation set up is given figure 4.8.
128
Figure 4.8: Volume images of the 270 x 150 x 150 voxel3 (0.4256 x 0.2365 x 0.2365 cm
3) sample
volumes of the pore space for the left and right matrix to be translated 1-to-1, voxel to
lattice node, for lattice Boltzmann simulation (A). LB simulation setup showing the 5
lattice layer thick pressure boundaries, with the fracture represented as the non-wetting
phase occupied pressure boundary (B).
The wettability of the core was observed previously to be very weak with a contact angle in the
range of 75° to 90°. Using the equation of Huang et al. 2007 given above, a , and a
total component density, , this contact angle can be approximated by using
. Setting the pressure boundaries is less exact; experience and
numerical experiments were performed to determine the pressure difference ( ) required for
the non-wetting phase to traverse the simulation volume, resulting in a general estimation of
0.033 (mulu-1
tu-2
). Typically, the capillary pressure of the simulation is scaled ( ⁄ )
using the physical and lattice interfacial tensions in accordance with Laplace’s law,
Non-wetting
phase Pressure
Boundary
(Fracture)
Wetting phase
Pressure Boundary
(Matrix)
Matrix:
Initially occupied by
wetting phase
Direction of DisplacementB)A)
129
where is the length scaling equal to the resolution of the image, and, and are the
physical and lattice interfacial tensions respectively. A lattice pressure difference of 0.033
(mulu-1
tu-2
) represents a physical pressure difference of 6.33 Pa, and is less than the physical
pressure difference experienced by the fluids in the portion of the core sequentially scanned.
This suggests the process of fracture-matrix fluid transfer is indeed a drainage process and not
the result of wettability alteration followed by imbibition. Had the simulation required a
pressure difference significantly greater than the pressure difference experienced in the
experiment, it would suggest the drop in the interfacial tension alone could not account for the
initialization of fracture-matrix fluid transfer. The simulation runs until the convergence
criterion ( ), defined here as,
is met. Based on numerical experiments not shown here, a convergence criteria of 0.001 was
chosen. The LB simulations each required near iterations (
) to
reach equilibrium, completing the displacement.
Time scaling of LB simulations ( ⁄ ) is vague, typically dimensionless
measurements of the forces involved (i.e. capillary number, Reynolds number) are matched to
those used in the simulation to determine a reasonable value of the time scaling (Sukop and
Thorne 2006) or chosen, as described by Hatiboglu and Babadagli 2008, “carefully” to match
scaling of the interfacial tension, viscosity and density between the LB and experimental system.
Time-scaling in the manner described by Hatiboglu and Babadagli 2008 resulted here in a gross
underestimation of the time-scaling on the order of 103 - 10
4. The displacements described here
130
occurred over a period of a few hours, whereas those described by Hatiboglu and Babadagli
2008 occurred in a matter of seconds. The LB models employed are similar, and thus the time-
scaling is of a similar magnitude, which was shown to be appropriate for the displacements
described in Hatiboglu and Babadagli 2008, but not those presented here. We determined the
time-scaling by matching the physical time of stable displacement in the experimental system to
the lattice time required for the simulation to reach equilibrium. Thus, the time-scaling is
determined by the measured time of displacement in the experiment, and not derived from the
simulation alone. This highlights one of the weaknesses of these types of LB simulations; alone
they cannot predict the rate of fracture-matrix fluid transfer for slow displacements. Unlike the
physical fracture-matrix transfer, there is no delay in the start of the displacement in the LB
simulations. To compensate for this, the LB simulation results are shifted by an amount equal to
the mean
of the right and left matrix. The stable physical displacement required is roughly
to reach equilibrium, giving a time-scaling of
⁄ .
4.4.4 Pore-Scale Experimental and Model Results
One of the main objectives of this study was to evaluate the accuracy with which the LB
simulation captures fluid distribution at the pore-scale. As was mentioned in the Introduction,
there are very few investigations that have compared LB simulation results with experimental
results at the pore-scale. Also, due to the employment of a surfactant that appears to alter the
wettability of the solid surface, we were interested in determining if the fracture-matrix fluid
transfer can be described by a drainage process, and if the differences observed in the fracture-
131
matrix transfer rate ( ) of the right and left matrix can be attributed to differences in the pore
space geometry. A comparison of as a function of for the experimental and LB results
is presented in figure 4.9.
Figure 4.9. Non-wetting phase saturation of the right and left matrix as determined from sequential
CMT imaging and LB drainage simulations. The only variable between the LB
simulations for the right and left matrix is the pore space geometry.
The LB simulation of the right matrix fits the experimental results very well; however, the
simulation of the left matrix underestimates the rate of fracture-matrix fluid transfer and,
ultimately, the equilibrium non-wetting phase saturation. This could be attributed to simulating
too low of a pressure difference; at higher capillary pressures the non-wetting phase saturation is
expected to be greater. However, the right matrix simulation correctly reproduced the
132
equilibrium saturation. It is also possible that the size of the pore space translated into lattices
for LB simulation was too small. Returning to figure 4.6, we observe the existence of elongated
pore spaces with lengths on the order of 0.10 – 0.15 cm or 63 - 95 voxels ( ). It is possible that
our simulations, which have side lengths of 150 voxels ( are too small to effectively access
these larger pore spaces during the simulated displacement, we further comment on this in the
next section. Determination of appropriate lattice sizes and resolutions for LB simulations
continue to be subjects of discussion, and with increasing computational accessibility, these
issues can be addressed. Although the LB model underestimates the fracture–matrix fluid
transfer for the left matrix, it does correctly predict the higher of the left matrix over the right
matrix for the first half of the displacement. This suggests that the stable displacement of the
wetting phase from the matrix by the non-wetting phase occupied fracture as a result of the
introduction of an interfacial tension reducing and wettability altering surfactant can be described
by a drainage process, without the consideration of wettability alteration. This is not surprising
considering a surface cannot be altered until the non-wetting phase fluid makes contact with the
surface, and this surface contact is controlled by a drainage-like displacement.
133
A) B)
C) D)
E) F)
G) H)
Experimental Model
134
Figure 4.10. Comparison of experimental and simulated vertical saturation profiles in the matrix (left
and right) adjacent to the fracture for the right (A, B) and left (C, D) matrix, and horizontal
saturation profiles for the right (E, F) and left (G, H) matrix.
To further investigate how effective the LB simulations were at replicating the experimentally
observed pore-scale fluid distributions we compare vertical and horizontal saturation profiles.
The vertical (saturation as a function of height) and the horizontal (saturation as a function of
distance from fracture) for the experimental and LB model are presented in figure 4.10.
Experimental and LB results are reported in dimensionless height ( ⁄ ) and length
( ⁄ ) units, where the reference height and length are the respective height and lengths
of the analyzed CMT image ( ) and LB simulation volume
( ). We also show some snapshots from the LB simulation
for the right and left matrix in figure 4.11.
135
Figure 4.11: Sample images of the LB simulations of fracture-matrix transfer. The non-wetting phase
saturation front and simulation pore space are shown, the wetting phase is occupies the
pore space not occupied by the non-wetting phase.
Left Matrix Right Matrix
136
The experimental vertical saturation profiles (figure 4.10A, C) do not display any dependence on
the fracture aperture, even with the presence of a sizable asperity. However, if the asperity were
larger we would expect the matrix adjacent to the asperity to experience some delay in the arrival
of the saturation front from nearby open portions of the fracture. Similar uniformity of the
vertical saturation profiles is seen in the LB simulations (figure 4.10B, D), but due to small size
of the LB simulation volumes the saturation front is less uniform, a result of piston-like
displacement occurring in individual pores. We can be fairly confident the capillary-dominated
fracture-matrix fluid transfer presented here has very little dependence on the fracture aperture,
given the fracture is occupied by the displacing phase.
The LB simulation for the left matrix (figure 4.10D) has low non-wetting phase saturations near
the bottom of the simulation volume. The blockage of the saturation front in the lower portion of
the simulation volume is a result of its small size. This becomes more apparent in the images of
the LB simulations shown in figure 4.11. As was previously stated, LB simulations are
computationally demanding, limiting the size of the simulation volume. The blockage of the
saturation front indicates that the volume simulated may be too small, and could contribute to the
underestimation of the fracture-matrix fluid transfer. Furthermore, the higher rate of
displacement in the left matrix was captured by the LB simulation at first, until the lack of
accessibility to the lower portion of the simulation volume slowed the displacement rate. The
experimental horizontal saturation profiles (figure 4.10E, G) show a saturation front that is 0.30
– 0.40 cm in length, with an irreducible wetting phase saturation near 0.90-0.95. The LB model
horizontal saturation profiles (figure 4.10F, H) roughly match the saturation front lengths of the
experimental measurements, but more importantly they accurately predict an irreducible wetting
137
phase saturation near 0.90-0.95. Also we see here that the non-wetting phase saturation front for
the left matrix reached the edge of the sample. In the LB simulation of the left matrix only the
top part of saturation front transited the simulation volume, causing the overall vertical saturation
profiles to underestimate experimental results. Again the mismatch can be attributed to
simulating to small of a volume. Considering this, we do not believe any mismatches between
the model and experimental results should be attributed to shortcomings of the LB method, but
rather are simply a result of a rough estimation of boundary conditions. Also issues involving
the size of the sample volume used to simulate fracture-matrix fluid transfer are a result of our
limited computational capacity.
4.5 CONCLUSIONS
Here we have presented a pore-scale investigation of fracture-matrix fluid transfer. A surfactant
water phase displacement of an oil-saturated oil-wet fractured synthetic porous medium was
sequentially imaged using 3D CMT imaging. The synthetic porous medium was designed to
have minimal pore space beyond the resolution of the CMT imaging used, allowing us to observe
the motion of the displacement front at the pore-scale. The displacement occurred over a period
of approximately 3 hours allowing us to image a 0.5043 cm length of the core 14 times to
produce 4D data of fracture-matrix fluid transfer. Although the displacement was not stopped
during sequential imaging, the displacement was slow enough that there was no appreciable
blurring of the 3D images. A surfactant was introduced to lower the interfacial tension between
the water and oil phase, and consequently the capillary entry pressure, to initiate fracture-matrix
fluid transfer. Images showed the surfactant altered the wettability of the porous medium
138
causing small amounts of the wetting phase in and near the fracture to snap-off and be
transported in the fracture. But, almost all of the non-wetting phase was displaced out the top of
core. There was very little, if any, counter-current migration of the wetting phase. The
surfactant was introduced into a fracture with the non-wetting water phase already established,
thus there was a delay in the stabilization of the rate of fracture-matrix fluid transfer as the
surfactant established itself at the oil-water interface. After this delay the displacement stabilized
and the non-wetting phase saturation of the right and left matrix correlated linearly with the
square-root of time. Although the differences in the porosity and pore aperture distribution of
the right and left matrix were small, the fracture-matrix transfer rate of the left matrix was
significantly greater than the right matrix. This emphasizes the importance of pore space
geometry, as this is the only variable that exists between the right and left matrix. Although the
matrix pore space geometry is an important factor determining the migration of the displacing
front, the movement of the displacing front showed very little dependence on the fracture
aperture, and the existence of an “asperity”.
One of the objectives of this investigation was to evaluate the use of a Shan-Chen type LB
models by making pore-scale comparisons of fluid distributions imaged using CMT and model
results, which are lacking in the literature. We present LB drainage simulations of a sample
volume of the left and right matrix to simulate the experimentally observed fracture-matrix fluid
transfer. Typical methods of scaling time in accordance with the interfacial tension results in a
3-4 order of magnitude underestimate of the time scaling. Here we scale time to match
experimentally observed physical time. This highlights one of the weaknesses of these types of
LB simulations; alone they cannot predict the rate of fracture-matrix fluid transfer for slow
139
displacements. The LB drainage simulation of the right matrix closely matched the
experimentally observed equilibrium saturation, displacement rate, irreducible wetting phase
saturation, as well as the vertical and horizontal saturation profiles. The LB drainage simulation
of the left matrix closely matched the experimentally observed displacement rate for the first half
of the displacement and the irreducible wetting phase saturation. However, the simulation LB
simulation of the left matrix underestimated the equilibrium saturation, which was reflected in an
underestimation of the displacement rate for the second half of the displacement and a pore-scale
fluid distribution match that was not as convincing as the simulation of the right matrix. This
underestimation is likely a result of using too small of a simulation volume, which limited access
to some of the larger elongated pores of the left matrix. We do not believe any mismatches
between the model and experimental results should be attributed to shortcomings of the LB
method. Due to the prohibitive computational demands of multicomponent LB models the
sample volumes of the image used in the LB model was significantly smaller than the ROI,
considering the simulations only contained on the order of 100 grains they replicated
experimental results fairly well. This successful replication also implies that although the
surfactant altered the wettability of the porous medium to water-wet the fracture-matrix fluid
transfer can be described as a drainage process.
140
4.6 REFERENCES
Al-Raoush R.I. Impact of wettability on pore-scale characteristics of residual nonaqueous phase
liquids. Environmental Science and Technology 2009; 43:4796-4801. doi:10.1021/es802566s
Bakke S., and Oren P.E. 3-D pore-scale modeling of sandstones and flow simulations in the pore
networks. SPE Journal 1997; 2(2):136-149. doi:10.2118/35479-PA
Boek E.S., and Venturoli M. Lattice-Boltzmann studies of fluid flow in porous media with
realistic rock geometries. Computers and Mathematics with Applications 2010; 59:2305-2314.
doi:10.1016/j.camwa.2009.08.063
Brusseau M.L., Peng S., Schnaar G., and Constanza-Robinson M.S. Relationships among air-
water interfacial area, capillary pressure, and water saturation for a sandy porous medium.
Water Resources Research 2006; 42:W03501. doi:10.1029/2005WR004058
Brusseau M.L., Janousek H., Murao A., and Schnaar G. Synchrotron X-ray microtomography
and interfacial partitioning tracer test measurements of NAPL-water interfacial areas. Water
Resources Research 2008; 44:W01411. doi:10.1029/2006WR005517
141
Buchgraber M., Al-Dossary M., and Ross C.M., and Kovscek A.R. Creation of a dual-porosity
micromodel for pore-level visualization of multiphase flow. Journal of Petroleum Science and
Engineering; 86-87:27-38.
Coles M.E., Hazlett R.D., Spanne P., Soll W.E., Muegge E.L., and Jones K.W. Pore level
imaging of fluid transport using synchrotron X-ray microtomography. Journal of Petroleum
Science and Engineering 1998; 19:55-63.
Dehghan A.A., Ghorbanizadeh S., and Ayatollahi Sh. Investigating the fracture network effects
on sweep efficiency during WAG injection process. Transport in Porous Media 2012; 93:577-
595. doi: 10.1007/s11242-012-9970-7.
Er V., Babadagli T., and Xu Z. Pore-scale investigation of the matrix-fracture interaction during
CO2 injection in naturally fractured oil reservoirs. Energy Fuels 2010; 24:1421-1430. doi:
10.1021/ef901038v.
Ghassemi A., and Pak A. Numerical study of factors influencing relative permeabilities of two
immiscible fluids flowing through porous media using lattice Boltzmann method. Journal of
Petroleum Science and Engineering 2011; 77:135-145. doi:10.1016/j.petrol.2011.02.007
142
Hao L., and Cheng P. Pore-scale simulations on relative permeabilities of porous media by
lattice Boltzmann method. International Journal of Heat and Mass Transfer 2010; 53:1908-1913.
doi:10.1016/j.ijheatmasstransfer.2009.12.066
Hatiboglu C.U. and Babadagli T. Pore-scale studies of spontaneous imbibition into oil-saturated
porous media. Physical Review E 2008; 77:066311. doi: 10.1103/PhysRevE.77.066311.
Hatiboglu C.U. and Babadagli T. Experiment and visual analysis of co- and counter-current
spontaneous imbibition for different viscosity ratios, interfacial tensions, and wettabilities.
Journal of Petroleum Science and Technology 2011; 70:214-228. doi:
10.1016/j.petrol2009.11.013.
Huang H., Thorne D.T., Schaap M.G., and Sukop M.C. Proposed approximation for contact
angles in Shan-and-Chen-type multicomponent multiphase lattice Boltzmann models. Physical
Review E 2007; 76:066701. doi: 10.1103/PhysRevE.76.066701
Iglauer S., Paluszny A., Pentland C.H., and Blunt M.J. Residual CO2 imaged with X-ray
microtomography. Geophysical Research Letters 2011; 38:L21403.
doi:10.1029/2011GL049680
Karadimitriou N.K. and Hassanizadeh S.M. A review of micromodels and their use in two-phase
flow studies. Vadose Zone Journal 2012; 11(3): doi: 10.2136/vzj2011.0072.
143
Karpyn Z.T., Grader A.S., and Halleck P.M. Visualization of fluid occupancy in a rough
fracture using microtomography. Journal of Colloid Interface Science 2007; 307(1):181-187. doi:
10.1016/j.jcis.2006.10.082
Keller, A. A., Roberts, P. V. and Blunt, M. J. Effect of fracture aperture variations on the
dispersion of contaminants. Water Resources Research 1999; 35(1):55-63.
Ketcham R.A., Slottke D.T., and Sharp Jr. J.M. Three-dimensional measurement of fractures in
heterogeneous materials using high-resolution X-ray computed tomography. Geosphere
6(5):499-514. doi: 10.1130/GES00552.1
Kumar M., Fogden A., Senden T., and Knackstedt M. Investigation of pore-scale missed
wettability. SPE Journal 2012; 17(1):20-30. doi: 10.2118/129974-PA
Landry C.J., Karpyn Z.T., and Piri M. Pore-scale analysis of trapped immiscible fluid structures
and fluid interfacial areas in oil-wet and water-wet bead packs. Geofluids 2011; 11:209-227.
doi: 10.1111/j.1468-8123.2011.00333.x
Lebedeva E.V., and Fogden A. Micro-CT and wettability analysis of oil recovery from
sandpacks and the effect of waterflood salinity and kaolinite. Energy and Fuels 2011; 25:5683-
5694. doi:10.1021/ef201242s
144
Lee T.C., Kashyap R.L. and Chu C.N. Building skeleton models via 3D medial surface/axis
thinning algorithms. CVGIP, Graphical Models and Image Processing 1994; 54:462-478.
Lindquist B., Lee S.M., Jones K.W., and Spanne P. Medial axis analysis of void structure in
three-dimensional tomographic images of porous media. Journal of Geophysical Research 1996;
101(B4):8297-8310.
Martys N.S., and Chen H. Simulation of multicomponent fluids in complex three-dimensional
geometries by the lattice Boltzmann method. Physical Review E 1996; 53(1):743-750.
Meakin P. and Tartakovsky A.M. Modeling and simulation of pore-scale multiphase fluid flow
and reactive transport in fractured and porous media. Reviews of Geophysics 2009; 47:RG3002.
doi: 10.1029/2008RG000263
Pan C., Hilpert M., and Miller C.T. Lattice-Boltzmann simulation of two-phase flow in porous
media. Water Resources Research 2004; 40:W01501. doi:10.1029/2003WR002120
Porter M.L., Schaap M.G., and Wildenschild D. Lattice-boltzmann simulations of the capillary
pressure-saturation-interfacial area relationship for porous media. Advances in Water
Resources; 32:1632-1640. doi:10.1016/j.advwatres.2009.08.009
145
Prodanovic M., Lindquist W.B., and Seright R.S. Porous structure and fluid partitioning in
polyethylene cores from 3D x-ray microtomographic imaging. Journal of Colloid and Interface
Science 2006; 298:282-297. doi:10.1016/j.jcis.2005.11.053
Prodanovic M., Lindquist W.B., and Seright R.S. 3D image-based characterization of fluid
displacement in a Berea core. Advances in Water Resources 2007; 30:214-226.
doi:10.1016/j.advwatres.2005.05.015
Qian Y.H., d’Humieres D., and Lallemand P. Lattice BGK Models for Navier-Stokes Equations.
Europhysics Letters 1992; 17(6):479-484. doi:10.1209/0295-5075/17/6/001
Ramstad T., Oren P.E., and Bakke S. Simulation of two-phase flow in reservoir rocks using a
lattice Boltzmann method. SPE Journal 2010; 15(4):917-927. doi: 10.2118/124617-PA
Ramstad T., Idowu N., Nardi C., and Oren P.E. Relative permeability calculations from two-
phase flow simulations directly on digital images of porous rock. Transport in Porous Media
2012; 94:487-504. doi:10.1007/s11242-011-9877-8
Rangel-German E.R. and Kovscek A.R. A micromodel investigation of two-phase matrix-
fracture transfer mechanisms. Water Resources Research 2006; 43:W03401. doi:
10.1029/2004WR003918.
146
Renard F., Bernard D., Desrues J., and Ougier-Simonin A. 3D imaging of fracture propagation
using synchrotron X-ray microtomography. Earth and Planetary Science Letters 2009: 286:285-
291. doi: 10.1016/j.epsl.2009.06.040
Schaap M.G., Porter M.L., Christensen B.S.B, and Wildenschild D. Comparison of pressure-
saturation characteristics derived from computed tomography and lattice Boltzmann simulations.
Water Resources Research; 43:W12S06. doi:10.1029/2003WR002120
Shan X., and Doolen G. Multicomponent lattice-Boltzmann model with interparticle interaction.
Journal of Statistical Physics 1995; 81:379-393.
Silin D., Tomatsu L., Benson S.M., and Patzek T.W. Microtomography and pore-scale modeling
of two-phase fluid distribution. Transport in Porous Media 2011; 86:495-515.
doi:10.1007/s11242-010-9636-2
Sukop, M. and Thorne, D. T. (2006) Lattice Boltzmann Modeling an Introduction for
Geoscientists and Engineers, Springer-Verlag.
Sukop M.C., Huang H., Chen L.L., Deo M.D., Oh K., and Miller J.D. Distribution of multiphase
fluids in porous media: Comparison between lattice Boltzmann modeling and micro-x-ray
tomography. Physical Review E 2008; 77:026710. doi: 10.1103/PhysRevE.77.026710
147
Thovert J.F., Salles J., and Adler P.M. Computerized characterization of the geometry of real
porous media: their discretization, analysis and interpretation. Journal of Microscopy 1993;
170(1):65-79.
Turner M.L., Knufing L., Arns C.H., Sakellariou A., Senden T.J., Sheppard A.P., Sok R.M.,
Limaye A., Pinczewski W.V., and Knackstedt M.A. Three-dimensional imaging of multiphase
flow in porous media. Physica A 2004; 339:166-172. doi:10.1016/j.physa.2004.03.059
Wan J., Tokunaga T.K., Tsang C.F., and Gudmundur S. Bodvarsson. Improved glass
micromodel methods for studies of flow and transport in fractured porous media. Water
Resources Research 1996; 32(7):1955-1964.
Wennberg, O. P., Rennan, L. and Basquet, R. Computed tomography scan imaging of natural
open fractures in a porous rock; geometry and fluid flow. Geophysical Prospecting 2009;
57(2):239-249. doi: 10.1111/j.1365-2478.2009.00784.x
Zdziennicka A., Szymczyk K., and Janzuk B. Correlations between surface free energy of quartz
and its wettability by aqueous solutions of nonionic, anionic and cationic surfactants. Journal of
Colloid and Interface Science 2009; 340:243-248. doi :10.1016/j.jcis.2009.08.040
148
CHAPTER 5: SUMMARY OF FINDINGS AND RECOMMENDATIONS FOR FUTURE
WORK
In these investigations we have highlighted some of the strengths of the LB model, and its ability
to elucidate the relationship between pore space characteristics and macroscale two-phase fluid
flow. Our comparisons of pore-scale fluid distributions between the model and experiment are
promising. There are however many questions that can be addressed by future investigations.
In the first investigation we found that the presence of porous walls can greatly enhance the
permeability of the fracture. We also found that fracture permeability estimation using
previously proposed formulations dependent on fracture characteristics did a generally poor job
of predicting the permeability of the fracture simulated. The determination of fracture
permeability is greatly simplified when the existence of the porous walls has an insignificant
effect. We can expect that as the ratio of the fracture aperture to pore aperture increases the
presence of the porous walls will have less of an effect, until at a critical ratio it will have no
effect. The question then arises what is this critical ratio? One should also consider that there
are many different types of fractures that exist across vast length scales (10-6
to 104 m). At what
length scales and what type of fractures does the presence of a porous wall significantly affect
flow? It would be optimal if a formulation could be determined that estimated fracture
permeability from the fracture characteristics, but the results seen here are not promising.
However considering there are many types of fractures, does there exist a “class” of fractures for
149
which the estimation of fracture permeability from the fracture characteristics is practical. We
can speculate that fractures with a large fracture aperture to pore aperture ratio would fall into
this “class”. Future investigations that collected a variety of fracture images could simulate flow
using LB and begin to find the limits of fracture permeability estimation from fracture
characteristics. Although it should be noted that it is unlikely that a robust fracture permeability
estimation from fracture characteristics could be found for fractures like the one presented here
that a small fracture aperture to pore aperture ratio – the solid boundaries of porous matrix-
fracture system a too complex to be distilled into a simple formulation.
In the second investigation we found that the relative permeability of both phases in
homogenous-wet porous media decreased with increasing wetting strength. It would be
interesting to compare this to experimental measurements of relative permeability. There are
many techniques that can be employed to change the wettability of beads, performing
conventional relative permeability measurements of homogenous-wet bead packs would help to
further connect the results of the second investigation to experimental measurements.
In the third investigation we found that LB simulation matched the experimental results for the
right matrix well, but not the left matrix. This was attributed to size of sample image translated
into a LB lattice being likely too small. However, it is also possible that the issue was the
underestimation of the pressure difference imposed in the simulations. It would be easy to
determine which of these were true by simulating the fracture-matrix fluid transfer on a larger
sample image. But what would be of greater interest to the community is a generalized study of
lattice sizes and resolutions used in LB simulations. As in, what is the representative elementary
150
volume required to properly simulate flow? One way to answer this question would be to use a
pore space image of a semi-homogenous medium, such as a bead pack, and simulate increasingly
larger volumes until the macroscale flow results stabilize.
VITA: Christopher James Landry
EDUCATION:
Doctor of Philosophy, Energy and Mineral Engineering
The Pennsylvania State University, University Park, Pennsylvania May 2013
Dissertation Title: Pore-scale imaging and lattice Boltzmann modeling of single- and multi-
phase flow in fractured and mixed-wet permeable media.
Advisor: Dr. Zuleima Karpyn
Master of Science, Petroleum and Natural Gas Engineering
The Pennsylvania State University, University Park, Pennsylvania
December 2009
Thesis Title: Experimental pore-scale analysis of fluid interfacial areas in oil-wet and water-wet
bead packs.
Advisor: Dr. Zuleima Karpyn
Bachelor of Science, Geology & Physics
Western Michigan University, Kalamazoo, Michigan
December 2006
PEER-REVIEWED PUBLICATIONS:
1) Landry C.J. and Karpyn Z. (In preparation) “Lattice Boltzmann modeling and 4D x-ray
computed microtomogrpahy imaging of fracture-matrix fluid transfer”.
2) Landry C.J. and Karpyn Z. (In preparation) “Relative permeability of homogenous-wet and
mixed-wet porous media as determined by pore-scale lattice Boltzmann modeling”.
3) Landry C.J. and Karpyn Z. (2012) “Single-phase lattice Boltzmann simulations of pore-scale flow in fractured permeable media” International Journal of Oil, Gas and Coal
Technology, 5(2/3), 182-206.
4) Landry C.J., Karpyn Z. & Piri M. (2011) “Pore-scale analysis of trapped immiscible fluid
structures and fluid interfacial areas in oil-wet and water-wet bead packs” Geofluids,
11(2), 209-227.
5) Landry C.J., Koretsky C., Das S. & Lund T. (2009) “Surface complexation modeling of Co(II)
on mixtures of hydrous ferric oxide, quartz and kaolinite” Geochimica et Cosmochica
Acta, 73 , 3723-3737
AWARDS:
1) “Second Place Society of Petroleum Engineers International Student Paper Contest – Master’s
Division” Society of Petroleum Engineers Annual Technical Conference, September
2010, Florence, Italy.
2) “First Place Society of Petroleum Engineers Regional Student Paper Contest – Master’s
Division” Eastern/Mid-Continent/Rockies SPE Regional Paper Contest, March 2010,
Pennsylvania State University, State College PA, U.S.A.
3) “Second Runner-up Graduate Student Paper Contest” Canadian International Petroleum
Conference, June 2009, Calgary Canada.
4) “2008 Marathon Alumni Centennial Graduate Fellowship” College of Earth and Mineral
Sciences, October 2008, Pennsylvania State University, State College PA, U.S.A.
5) “Outstanding Student Paper in the Division of Geochemistry” American Chemistry Society,
2006 National Meeting, Adsorption of metals to geomedia symposium, Atlanta GA U.S.A.
6) “2005-2006 Arts & Sciences Research and Creative Activities Award” College of Arts &
Sciences, Western Michigan University, Kalamazoo MI U.S.A.