PALABOS 3D Boltzmann Method For Fluid Flow Modeling Porous Rocks

Application and Assessment of the Lattice
Boltzmann Method for Fluid Flow Modeling in

Porous Rocks
vorgelegt von
Diplom Geophysiker
Matthias Halisch
aus Rotenburg/Wmme
von der Fakultt VI - Planen Bauen Umwelt

der Technischen Universitt Berlin
zur Erlangung des akademischen Grades
Doktor der Naturwissenschaften

- Dr.rer.nat. -
genehmigte Dissertation
Promotionsausschuss:
Vorsitzender: Prof. Dr. Gerhard Franz

Berichter: Prof. Dr. Ugur Yaramanci
Berichter: Prof. Dr. Holger Steeb
Berichter: Dr. Stefan Wessling
Tag der wissenschaftlichen Aussprache: 19. Mrz 2013
Berlin 2013
D 83
We have found it of paramount importance that in order to progress, we
1
must recognize our ignorance and leave room for doubt.
1
Richard P. Feynman ( 1918, 1988)
For Susanne
Acknowledgements
Working almost three years that intensly on this thesis gave me a tremendous
amount of new insights and memories. It would not have been possible to
complete this work without countless people, colleagues and friends. First, I
would like to thank my director Prof. Dr. Ugur Yaramanci for accepting and
supporting this doctoral thesis. Furthermore, I would like to thank my super-
visors, Prof. Dr. Holger Steeb and Dr. Stefan Wessling for many fruitfull talks,
discussions and exchange of ideas, pushing my thesis straight forward to the
finishing line.
I would also like to thank my supervisor, Dr. Thomas Wonik. Your comments,
your reviewing of my writings and your countless and sometimes ruthless pun-
ches have been an essential motivation for my work over the past three years! I
also owe a debt of gratitude to Dr. Manfred Wuttke for his support concerning
all topics related to CFD and linux, as well as to Juliane Herrmann for many
useful hints, comments and help, concerning the creation of figures.
I also owe a great debt of gratitude to the staff of palabos, exceptionally

to Mr. Jonas Latt, for the great and uncomplicated support for helping me
setting up the Lattice Boltzmann Simulation! Last but not least, I would li-
ke to thank my dearest colleagues, Cornelia Mller, Esther Vogt and Henrike
Baumgarten, for their priceless support of laughs, talks and coffee during so
many rock bottom moments of mental motivation and disturbance. Just stay
as you are!
v
Abstract
Matthias Halisch, Leibniz Institute for Applied Geophysics, PhD thesis 2012
Application and Assessment of the Lattice Boltzmann Method for Fluid

Flow Modeling in Porous Rocks - simulation of transport properties by
using real pore geometries derived by high resolution micro CT imaging
Modeling physical properties and phenomena of natural rocks forms an important

field of work for modern petrophysics. In recent years, the fast and non-destructive
X-ray computed micro tomography has pushed the knowledge and understanding of
fluid dynamics in complex pore systems extensively forward.
This PhD thesis is in accordance to this wide and complex field of research and
gives a comprehensive overview about state of the art transport modeling in porous
media. New approaches for high precisely modeling aspects in real pore geometries
are shown. Conducive to this thesis, theoretical background on fluid dynamics, mod-
ern CT imaging and modeling techniques is introduced first. Afterwards, a detailed
and interdisciplinary rock characterization is performed, before results of transport
modeling on a synthetic and on real pore networks are given.
Special attention is given onto two main topics. On the one hand, the essential
workflow to obtain high resolution 3D data sets for the advanced pore network ex-
traction is introduced in detail. Since modern CT devices are able to scan samples
with micron to submicron resolution, even very small pore structures can be taken
into account for modeling purposes. For this, a special network extraction algorithm,
which is able to preserve these small sample features with high accuracy during grid
generation, is tested.
vii
On the other hand, an advanced method to evaluate the representiveness of transport
modeling phenomena, depending on the voxel size of the scanned data, is introduced.
The so derived representative elementary volumes (REV) are then used to apply the
Lattice Boltzmann Method (LBM) for the estimation of fluid transport properties of
these structures. Due to this numerically easy but elegant approach, even structures
of arbitrary complexity (e.g. coarse and fine pore networks, secondary porosities,
clay related porosities, etc.) can be investigated and taken into account on different
scales as well.
Additionally, comparative simulations with a commonly used modeling technique

are performed and evaluated in close combination with the results of the Boltzmann
transport modeling on different scales. Finally, results of each partition of this thesis
are combined, assessed, discussed and summarized.
Kurzfassung
Matthias Halisch, Leibniz-Institut fr Angewandte Geophysik, Doktorarbeit 2012
Anwendung und Bewertung der Gitter Boltzmann Methode fr die Mo-

dellierung von Fluid Transport in porsen Gesteinen - Simulation von
Transporteigenschaften unter Verwendung realer Porenraumgeometrien
aus hochauflsenden Micro-CT Untersuchungen
Die Modellierung physikalischer Eigenschaften und Phnomene von natrlichen Fest-

gesteinen stellt inzwischen einen wichtigen Arbeitsbereich innerhalb der modernen
Gesteinsphysik dar. In den letzten Jahren hat dabei die Rntgen-Computertomografie
als schnelles, zerstrungsfreies sowie bildgebendes Verfahren die Grenzen des Ver-
stndnisses der Fluid Dynamik in komplexen Systemen signifikant vorangetrieben.
Diese Promotionsschrift reiht sich konsequent in die komplexe Thematik mit ein
und gibt zunchst einen umfassenden Einblick in den Stand der Technik von Str-
mungssimulationen in porsen Medien, bevor neue Anstze fr hochgenaue Modell-
rechnungen in echten Porennetzwerken aufgezeigt werden. Zielfhrend werden dabei
zunchst die theroretischen Grundlagen bezglich der Fluid Dynamik, dem bild-
gebenden Verfahren sowie der Modellierungstechnik gegeben. Diesem theoretischen
Teil folgt eine ausfhrliche, interdisziplinre Gesteinscharakterisierung, bevor die Er-
gebnisse der Strmungssimulationen an synthetischen und realen Porennetzwerken
dargestellt werden.
Besonderes Augenmerk dieser Arbeit liegt dabei auf zwei thematischen Bereichen.
Zum Einen, wird der essentielle Arbeitsablauf zur Gewinnung hochaufgelster 3D
Datenstze fr die Extraktion der in-situ Porennetzwerke analysiert. Aufgrund der
hohen erreichbaren Auflsung moderner CT-Systeme (wenige Mikrometer, teils einige
hundert Nanometer) knnen erstmalig auch sehr kleine Porenstrukturen fr die Mo-
ix
dellierung aufbereitet werden. Dazu wird ein verbesserter Extraktions-Algorithmus
getestet, der in der Lage ist, die segmentierten Strukturen mit hoher Genauigkeit
und Zuverlssigkeit in numerische Gitter zu transferieren.
Zum Anderen wird ein neues Verfahren zur berprfung der Reprsentativitt von
Transportmodellierungen eingefhrt. Die so ermittelten reprsentativen Volumina
werden im Anschluss daran mittels der Gitter Boltzmann Methode auf ihre Trans-
porteigenschaften hin untersucht. Aufgrund dieser numerisch sehr einfachen, aber
dennoch eleganten Methode knnen beliebige Strukturen (z.B. grobe und feine Po-
rennetzwerke, sekundre Porositten, Tonporositten, etc.) auf beliebigen Skalen mit
hoher Auflsung modelliert und gleichfalls charakterisiert werden.
Ergnzend dazu werden vergleichende Simulationen mit gebruchlichen Software-

paketen evaluiert und mit den Ergebnissen der Boltzmann Modellierung auf ver-
schiedenen (Mikro-) Skalen gegenbergestellt. Abschlieend werden die Ergebnisse
aller Teilbereiche zusammengefasst, beurteilt und diskutiert.
Contents
Acknowledgements v
Abstract vii
Kurzfassung ix
1 Introduction 1
1.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Pore Scale Fluid Flow Models . . . . . . . . . . . . . . . . . . . . . . 2
1.3 CT Imaging and Pore Network Extraction . . . . . . . . . . . . . . . 3
1.4 Pore Scale Simulation Techniques . . . . . . . . . . . . . . . . . . . . 4
1.5 Motivation and Chapter Description . . . . . . . . . . . . . . . . . . 5
2 High Resolution Micro CT Imaging 7

2.1 Basic Theory and Principle . . . . . . . . . . . . . . . . . . . . . . . 7
2.1.1 X-ray Generation . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.1.2 X-ray Attenuation . . . . . . . . . . . . . . . . . . . . . . . . 9
2.1.3 Basic Principle of CT Scanning . . . . . . . . . . . . . . . . . 10
2.1.4 Image (Spatial) Resolution and Sharpness . . . . . . . . . . . 11
2.2 3D Image Reconstruction . . . . . . . . . . . . . . . . . . . . . . . . 13
2.3 The nanotom 180 CT Scanner . . . . . . . . . . . . . . . . . . . . . 15
2.4 Sample Preparation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.5 Imaging Workflow . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3 Basics of Fluid Dynamics 19

3.1 Darcys Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.2 Prerequisites for the Navier-Stokes Equations . . . . . . . . . . . . . 21
3.2.1 Reynolds Transport Theorem . . . . . . . . . . . . . . . . . . 21
3.2.2 Conservation of Mass . . . . . . . . . . . . . . . . . . . . . . . 21
xi
3.2.3 Conservation of Momentum . . . . . . . . . . . . . . . . . . . 23
3.2.4 Conservation of Energy . . . . . . . . . . . . . . . . . . . . . 24
3.3 Derivation of the Navier-Stokes Equations . . . . . . . . . . . . . . . 25
3.3.1 The Navier Equations . . . . . . . . . . . . . . . . . . . . . . 25
3.3.2 The incompressible Navier-Stokes Equations . . . . . . . . . . 29
3.3.3 The compressible Navier-Stokes Equations . . . . . . . . . . . 30
4 The Lattice Boltzmann Method (LBM) 31

4.1 The Boltzmann Equation . . . . . . . . . . . . . . . . . . . . . . . . 31
4.2 Lattice Gas Automata (LGA) . . . . . . . . . . . . . . . . . . . . . . 34
4.3 Lattice Boltzmann Equation (LBE) . . . . . . . . . . . . . . . . . . . 35
4.4 Lattice Boltzmann Models (LBm) . . . . . . . . . . . . . . . . . . . . 38
4.4.1 The 2D Nine Velocity LBm (D2Q9) . . . . . . . . . . . . . . 38
4.4.2 The 3D Nineteen Velocity LBm (D3Q19) . . . . . . . . . . . 40
4.5 Setting up a Simulation with Palabos . . . . . . . . . . . . . . . . . . 42
4.5.1 About Palabos . . . . . . . . . . . . . . . . . . . . . . . . . . 42
4.5.2 Workflow to set up a Simulation with Palabos . . . . . . . . . 43
5 Rock Sample Characterization 49

5.1 Basic Geological and Mineralogical Rock Description . . . . . . . . . 50
5.1.1 Bentheimer Sandstone (S1) . . . . . . . . . . . . . . . . . . . 50
5.1.2 Obernkirchen Sandstone (S2) . . . . . . . . . . . . . . . . . . 52
5.1.3 Flechtingen Sandstone (S3) . . . . . . . . . . . . . . . . . . . 54
5.2 Petrophysical Rock Characterization . . . . . . . . . . . . . . . . . . 56
5.2.1 Bentheimer Sandstone (S1) . . . . . . . . . . . . . . . . . . . 57
5.2.2 Obernkirchen Sandstone (S2) . . . . . . . . . . . . . . . . . . 60
5.2.3 Flechtingen Sandstone (S3) . . . . . . . . . . . . . . . . . . . 63
5.3 Combined Rock Evaluation . . . . . . . . . . . . . . . . . . . . . . . 66
6 The REV Challenge - a Classical and a New Approach 71

6.1 Classical REV Approach . . . . . . . . . . . . . . . . . . . . . . . . . 71
6.1.1 Classical REV Analysis for S1 . . . . . . . . . . . . . . . . . . 74
6.2 Advanced REV Approach . . . . . . . . . . . . . . . . . . . . . . . . 80
6.2.1 Advanced REV Analysis for S1 . . . . . . . . . . . . . . . . . 81
7 Quantification of Sample Inhomogeneity 91
7.1 S1 CT Miniplug Inhomogeneity Analysis . . . . . . . . . . . . . . . . 93
8 Pore Network Extraction 99

8.1 Basics of Advanced Network Extraction . . . . . . . . . . . . . . . . 99
8.1.1 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
8.1.2 Pore Network Extraction Workflow . . . . . . . . . . . . . . . 104
8.2 Benchmark Sample Pore Network . . . . . . . . . . . . . . . . . . . . 106
8.3 Bentheimer Sandstone (S1) Pore Network . . . . . . . . . . . . . . . 108
8.4 Obernkirchen Sandstone (S2) Pore Network . . . . . . . . . . . . . . 110
8.5 Flechtingen Sandstone (S3) Pore Network . . . . . . . . . . . . . . . 112
9 Modeling Fluid Flow with In Situ Geometries 115

9.1 Classical Finite Element (FE) Approach . . . . . . . . . . . . . . . . 115
9.2 Results of the Lattice Boltzmann Simulation . . . . . . . . . . . . . . 121
9.2.1 Results for the S1 Sandstone . . . . . . . . . . . . . . . . . . 121
9.3 Combined Petrophysical and Modeling Results . . . . . . . . . . . . 141
10 Conclusions and Outlook 145

10.1 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
10.2 Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
A Petrophysical Data 151
B Generated Meshes for CFD Modeling 161
C Lattice Boltzmann Source Code 171

List of Figures
2.1 Processes responsible for X-ray photon emission: Bremsstrahlung (A)

and high-energy transitions within the electron shell of an atom (B). 8
2.2 Basic assembly and functionality of a X-ray tube. . . . . . . . . . . . 9
2.3 Exemplified X-ray attenuation through a homogeneous material. Mod-
ified after Buzug (2005). . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.4 Exemplified X-ray attenuation through a multi-component material.
Modified after Buzug (2005). . . . . . . . . . . . . . . . . . . . . . . 10
2.5 Basic setup of a CT scanning device: X-ray source (left), CNC stage
and specimen (middle) and flat panel detector unit (right). . . . . . . 11
2.6 Schematic diagram of geometrical limits controlling image resolution. 12
2.7 Schematic illustration of the influence of the focal spot size on the
penumbra effect. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.8 Overview of processes involved from image acquisition to image re-
construction. Modified after Buzug (2005). . . . . . . . . . . . . . . . 14
2.9 The nanotom 180 high resolution CT device from GE Sensing and
Inspection Technologies, productline phoenix|x-ray. . . . . . . . . . . 16
3.1 Darcys experimental set-up for determination of (rock) permeability.

Modified after Batchelor (1994). . . . . . . . . . . . . . . . . . . . . . 20
3.2 Definition of appearing normal and shear stresses for the y-faces of a
small control volume 4V . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.3 Exemplary diagram of the only possible surface forces that contribute
in y-direction on a small control volume. . . . . . . . . . . . . . . . . 27
4.1 Basic collisions of the FHP-I model (Luo, 2003). . . . . . . . . . . . . 34

4.2 Exemplary evolution scheme of the FHP-I model from time t (red) to
t + 4t (green). Modified after Luo (2003). . . . . . . . . . . . . . . . 35
4.3 The 2D Lattice Boltzmann model, showing the Cartesian lattice and
velocities. Modified after Sukop and Thorne Jr. (2010). . . . . . . . . 38
xv
4.4 Schematic view of the D2Q9 x- and y-velocity components. Modified
after Sukop and Thorne Jr. (2010). . . . . . . . . . . . . . . . . . . . 39
4.5 Schematic workflow of velocity vector implementation to derive the
D3Q19 Lattice Boltzmann model. . . . . . . . . . . . . . . . . . . . . 41
4.6 Exemplary pore network geometry for the Lattice Boltzmann simula-
tion with Palabos. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
4.7 Syntax to start the grid rescaling with Palabos. The first part (light
gray box) initiates the parallel computing, the second part (light yel-
low box) starts the rescaling process. Next, the input.stl file is defined
(light blue box) and the fit direction is set (0 = x-direction, 1 = y-
direction, 2 = z-direction, dark green box). After defining the correct
fit length (light green box), the output file needs to be addressed (light
red box). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
4.8 Syntax to start the Lattice Boltzmann simulation with Palabos. . . . 45
4.9 Examples of the output files derived with Palabos. From left to right:
geometry including inflow and outflow zone, velocity and vorticity field
after 4,000 iterations. . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
4.10 Exemplary visualization of the pressure drop of the tutorial sample
geometry with Paraview after 4,000 iterations. . . . . . . . . . . . . . 47
4.11 Workflow to set up a Lattice Boltzmann simulation with Palabos. . . 48
5.1 Exemplary thin sections of the S1 specimen set. . . . . . . . . . . . . 51

5.2 Exemplary SEM images of a S1 specimen. . . . . . . . . . . . . . . . 52
5.7 Representative image of a S1 laboratory scale sample (side- and topview)
including a schematic view of the CT miniplug size and sampling po-
sition. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
5.8 Exemplary T2 relaxation time spectrum and average percentage of
effective porosity for the S1 specimen set. . . . . . . . . . . . . . . . 59
5.9 Exemplary representative pore throat distribution derived from S1
mercury injection experiments. . . . . . . . . . . . . . . . . . . . . . 60
sition. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
5.11 Exemplary T2 relaxation time spectrum and average percentage of
effective porosity for the S2 specimen set. . . . . . . . . . . . . . . . 62
5.12 Exemplary representative pore throat distribution derived from mer-
cury injection experiments of the S2 specimen set. . . . . . . . . . . 62
sition. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
5.14 Exemplary T2 relaxation time spectra and average percentage of ef-
fective porosity for the S3 specimen set (red: low , orange: high
sample). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
5.15 Exemplary representative pore throat distribution derived from mer-
cury injection experiments of the S3 specimen set. . . . . . . . . . . 65
5.16 Porosity - Permeability ( - kkl ) relationship for all specimen sets. . 67
5.17 Permeability - BET surface (kkl - Spor ) relationship for all specimen
sets. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
5.18 Average percentages of porosity related to NMR T2 spectra cut off
values for all specimen sets. . . . . . . . . . . . . . . . . . . . . . . . 68
6.1 The basic REV principle, shown by the variation of porosity (n) within
the neighborhood of a point P (x0 ) as a function of the averaging
volume (U). Modified after Bear and Bachmat (1990). . . . . . . . . 72
6.2 Estimation of the S1 REV by using the classical approach from Bear
(1988). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
6.3 Estimation of the S2 REV by using the classical approach of Bear
(1988). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
6.4 Estimation of the S3 REV by using the classical approach of Bear
(1988). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
6.5 Directional porosity pseudo logs of the S1 CT sample. . . . . . . . . 82
6.6 Variogram analysis for the directional porosity distribution of the S1
CT sample. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
CT sample. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
CT sample. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
7.1 Directional S and resulting I for a variety of subvolume edge sizes
of the S1 CT specimen. . . . . . . . . . . . . . . . . . . . . . . . . . 94
8.1 Tetrahedral base case for: (A) cube with all vertices inside the object
(i.e. no iso-surface running through the model), and (B) a cube with
a single iso-surface passing through it (Young et al., 2008). . . . . . . 101
8.2 Improvement of hex-tet ratio and mesh coarsening techniques illus-
trated on a small cylindrical volume: (A) voxel based mesh (Lego
brick mesh); (B) tetrahedralized mesh using VoMaCs; (C) tetra-
hedralized with internal hexahedral grid and partial resorbed surface
hexahedra; (D) same as before but with EVoMaC algorithm used in
Simpleware (Young et al., 2008). . . . . . . . . . . . . . . . . . . . . 102
8.3 Qualitative examples for grid accuracy and topology preservation within
Simpleware. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
8.4 Generalized workflow for the phase segmentation and pore network
extraction with Simpleware. . . . . . . . . . . . . . . . . . . . . . . . 105
8.5 2D gray value phase image (left hand side) and resulting segmentation
histogram (right hand side) of the benchmark sample. . . . . . . . . 107
histogram (right hand side) of the S1 sandstone CT sample. . . . . . 109
9.1 General workflow for fluid flow modeling with Comsol. . . . . . . . . 116
9.2 Visualization of the velocity magnitude (left hand side) and of the
pressure field (right hand side) at the outflow face of the benchmark
sample, simulated with Comsol. . . . . . . . . . . . . . . . . . . . . . 118
9.3 Visualization of the pressure drop amongst the benchmark pore net-
work, calculated with Palabos. . . . . . . . . . . . . . . . . . . . . . 119
9.4 Visualization of the vorticity field of the benchmark sample after
115,000 iterations (left hand side: x-direction; middle: y-direction;
right hand side: z-direction), calculated with Palabos. . . . . . . . . 119
9.5 Visualization of the normalized velocity field of the benchmark sam-
ple after 115,000 iterations (left hand side: x-direction; middle: y-
direction; right hand side: z-direction), calculated with Palabos. . . . 120
9.6 Visualization of the pressure drop amongst the feldspar pore network
within the S1 specimen. . . . . . . . . . . . . . . . . . . . . . . . . . 122
9.7 Visualization of the vorticity field of the S1 feldspar pore network after
right hand side: z-direction). . . . . . . . . . . . . . . . . . . . . . . . 123
9.8 Visualization of the velocity field of the S1 feldspar pore network after
9.9 Visualization of the pressure drop amongst the clay pore network
within the S1 specimen. . . . . . . . . . . . . . . . . . . . . . . . . . 124
9.10 Visualization of the vorticity field of the S1 clay pore network after
9.11 Visualization of the velocity field of the S1 clay pore network after
9.12 Visualization of the pressure drop amongst the REV pore network of
the S1 specimen. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
9.13 Visualization of the vorticity field of the S1 REV pore network af-
ter 96,000 iterations (left hand side: x-direction; middle: y-direction;
9.14 Visualization of the velocity field of the S1 REV pore network af-
9.15 Visualization of the pressure drop amongst the clay pore network of
the S2 specimen. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
9.16 Visualization of the vorticity field of the S2 clay pore network after
9.17 Visualization of the velocity field of the S2 clay pore network after
9.18 Visualization of the pressure drop amongst the channel pore network
of the S2 specimen. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
9.19 Visualization of the vorticity field of the S2 channel pore network
after 23,000 iterations (left hand side: x-direction; middle: y-direction;
9.20 Visualization of the velocity field of the S2 channel pore network af-
9.21 Visualization of the pressure drop amongst the REV pore network of
the S2 specimen. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
9.22 Visualization of the vorticity field of the S2 REV pore network af-
9.23 Visualization of the velocity field of the S2 REV pore network af-
9.24 Visualization of the pressure drop amongst the pore network of the S3
specimen. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
9.25 Visualization of the vorticity field of the S3 pore network after 158,000
iterations (left hand side: x-direction; middle: y-direction; right hand
side: z-direction). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
9.26 Visualization of the velocity field of the S3 pore network after 158,000
iterations (left hand side: x-direction; middle: y-direction; right hand
side: z-direction). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
A.1 Histograms of petrophysical data for the S1 specimen set. . . . . . . 157

B.1 Generated REV mesh for the benchmark sample. . . . . . . . . . . . 162

B.2 Generated REV mesh of the S1 sample. . . . . . . . . . . . . . . . . 163
B.3 Generated feldspar porosity mesh of the S1 sample. . . . . . . . . . . 164
B.4 Generated clay porosity mesh of the S1 sample. . . . . . . . . . . . . 165
B.5 Generated REV mesh of the S2 sample. . . . . . . . . . . . . . . . . 166
B.6 Generated clay porosity mesh of the S2 sample. . . . . . . . . . . . . 167
B.7 Generated channel porosity mesh of the S2 sample. . . . . . . . . . . 168
B.8 Generated fracture porosity mesh of the S3 sample. . . . . . . . . . . 169
List of Tables
5.1 Overview of the applied petrophysical methods and sample quantities

for all three specimen sets. From top to bottom: porosity, density,
apparent and Klinkenberg permeability, NMR T2 spectra and porosity,
specific sample surface. . . . . . . . . . . . . . . . . . . . . . . . . . . 57
5.2 Overview of the ranges and mean values of the basic petrophysical
parameters of the S1 specimen set. . . . . . . . . . . . . . . . . . . . 58
8.1 Main segmentation threshold (void space - matrix) and meshing pa-
rameters of the benchmark sample. . . . . . . . . . . . . . . . . . . . 106
rameters of the S1 sandstone CT sample. . . . . . . . . . . . . . . . . 108
9.1 Main input and simulation parameters for the FE and LBM simulation
with the benchmark sample (the gap indicates not necessary input
data, DoF = Degree of Freedom). . . . . . . . . . . . . . . . . . . . . 117
9.2 Main parameters for the permeability estimation from the benchmark
sample. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
9.3 Main parameters for the permeability estimation of the S1 sample. . 121
xxiii
9.6 Summary of the permeability values derived by numerical and petro-
physical laboratory experiments (ksim = simulated permeability; kcor
= corrected simulated permeability; klab = sample laboratory perme-
ability; kset = average specimen set permeability). . . . . . . . . . . 142
9.7 Aberration between calculated and measured k-values for the S1 REV
and S2 REV, respectively. . . . . . . . . . . . . . . . . . . . . . . . . 144
A.1 Basic petrophysical properties (from left to right: grain density, poros-
ity, Klinkenberg permeability, susceptibility, mass and pore volume
specific BET surface) of the Bentheimer sandstone (S1) specimen set. 152
specific BET surface) of the Obernkirchen sandstone (S2) specimen set.154
specific BET surface) of the Flechtingen sandstone (S3) specimen set. 156
Chapter 1
Introduction
1.1 Overview
Petrophysics forms an essential fragment within the network of applied geosciences

and its related disciplines. Unlike many other fields of geoscience, geophysics - and
especially petrophysics - is not solely limited to phenomenological work. In fact, the
observed phenomena are not only described, they are assessed and explained with
complex physical processes, mathematical approaches and algorithms. The interpre-
tation of geophysical data derived at almost any scale and any field of application
needs the distinct knowledge of material (rock) properties and their complex corre-
lations.
Therefore the accurate determination of petrophysical parameters and flow prop-

erties in porous media is of great importance. It is possible to perform laboratory
experiments to obtain these macroscopic properties, for example, permeability, poros-
ity and specific surface area, that are required to determine fluid flow behavior in
total. The knowledge of these fundamental petrophysical properties is an indispen-
sible prerequisite for research. However, these investigations are often difficult or
even impossible to perform due to a lack of time, financial resources and / or sample
quantity and quality, respectively. Because of this, the setup and usage of numerical
and analog models has been of importance to understand the measured data and the
observed phenomena since the very beginning of modern petrophysical research.
1
2
1.2 Pore Scale Fluid Flow Models

One of the earliest model within the field of fluid transport at the pore scale is the
so called capillary tube model, which has been developed by Carman in 1937 and
which was published in 1956 (Carman, 1956). This model has been consequently
improved to take rock complexity more and more into account. As a result, the
idealized single capillary channel has been replaced with varying pore geometries by
introducing the so called Kozeny-Carman cross section shape-factor, leading to a
new understanding of fluid flow properties (Engelhardt and Zemann, 1960; Brace,
1977; Gueguen et al., 1986).
The first numerical models for fluid transport at the pore scale have been simi-
lar to the analytical ones as described before. The researchers attempted to simulate
natural pore networks by either using equivalent geometry approaches - similar to
Carmans - or by creating discrete or randomly distributed artificial, mostly spheri-
cal packings (Jerauld and Salter, 1990; Dullien, 1992; Martyns and Garboczi, 1992;
Lowry and Miller, 1995). Later on, these simplified models have been consequently
replaced by 2D models derived and digitally processed from thin sections or scan-
ning electron microscopy images (Auset and Keller, 2004; Kameda and Dvorkin,
2004; Fauzi and Bahri, 2005; Cloete, 2006). Nevertheless, it is obvious that two-
dimensional models poorly represent natural rocks because they are unable to repro-
duce spatial distribution and interconnectivity of real pore networks.
An alternative has been developed with the so called image based approach which
has been recently described by several authors (Keehm et al., 2001; Keehm, 2003;
Kameda and Dvorkin, 2004; Zhang et al., 2011). Within this approach, first 2D
image data sets are derived like it is done for the classical two-dimensional mod-
eling (e.g. by thin sections or X-ray Computed Tomography). Then the statistical
properties of the 2D pore space (e.g. porosity) are calculated from these images.
Next it is assumed that a three-dimensional pore space of this rock type obeys the
same statistical properties as the investigated 2D void space. Finally, specialized
geostatistical algorithms are used to create 3D pore structures. Regardless of the
promising results of these models compared to experimental lab data, they are only
applicable on homogeneous types of reservoir rocks.
Consequently, fully three-dimensional modeling at the pore scale is the one method
to describe and understand fluid flow for homogeneous and inhomogeneous as well
Introduction 3
as for isotropic and anisotropic rocks with best possible results. Hence, there have
been many research efforts to take this topic into account. Comparable to the 2D
modeling, two main branches have developed: First, by generating artificial pore
geometries similar to natural rocks (Bryant et al., 1993; Bakke and ren, 1997; Enz-
mann, 2000; Keehm et al., 2001). Second, by extracting in situ pore geometries and
rebuilding them into equivalent models (Blunt, 2001; Al-Kharusi and Blunt, 2007;
Dong, 2007). Again, the same challenges occur as within 2D modeling. Artificial 3D
geometries only deliver a good approximation of the real fluid flow behavior of nat-
ural rocks and fail if complex and inhomogeneous structures are observed. Contrary
to this, rebuilt equivalent 3D geometries can fit fluid flow behavior with remarkable
results but without any respect to in situ pore space topology and to very small pore
geometry features. So far there are only very few 3D approaches which deal with the
original pore geometries derived by modern three-dimensional imaging techniques
(Kameda, 2004; Dvorkin et al., 2009; Tlke et al., 2010).
1.3 CT Imaging and Pore Network Extraction

For many years, the only way to determine the interior structure of a rock sample was
to create thin sections. This technique is not only time-consuming, also a valuable
sample is lost by the destructive process. To prevent sample loss as well as sustained
loss of information, X-ray Computed Tomography (CT) is a promising way to in-
vestigate rock cores fast, non-destructively and in three dimensions. Major steps in
important hardware components like open micro- and nanofocus X-ray tubes on the
one side, and the development of highly efficient flat panel detectors on the other side
allowed the development of very versatile and high resolution laboratory CT systems
(Buzug, 2005; Brunke et al., 2008). Electromagnetic focusing of the electron beam
gave access to the generation of X-ray beams with an emission spot diameter down
to one micron or even smaller. This small beam diameter is essential for CT exam-
ination with voxel-sizes in the sub-micron range. These characteristics with respect
to spatial resolution principally allow CT measurements, which can compete with
many absorption contrast setups at synchrotron radiation facilities (Brunke et al.,
2008).
Originating from medical imaging and materials science, CT has coherently and
recently found its way into the large spectrum of geoscientific applications as a non-
destructive and truly 3D imaging technique for qualitative and especially quantitative
4
core analysis (Mees et al., 2003; Carlson et al., 2003; Akin and Kovscek, 2003; Knack-
stedt et al., 2004). As an outcome of this technique, several methods for pore network
extraction have been developed and the most important ones either use medial axis
or maximum ball algorithms in particular (Lee and Kashyap, 1994; Lindquist et al.,
1996; Sheppard et al., 2005; Silin and Patzek, 2006; Al-Kharusi and Blunt, 2007).
These equivalent networks are widely accepted for conventional single- and multi-
phase transport modeling. So far, there are only very few approaches which use the
original pore networks for computational fluid dynamics (CFD), due to limitations
in 3D image resolution, network extraction without preservation of topology and
computational time and power (Knackstedt et al., 2004; Young et al., 2008; Zhan
et al., 2010; Tlke et al., 2010).
1.4 Pore Scale Simulation Techniques

As already introduced, performing numerical modeling and calculating fluid trans-
port properties in natural pore geometries is difficult to obtain. Hence, transport
properties like permeability or fluid velocity are by default derived experimentally
or by empirical relationships that have been observed for specific rock formations
over a long period of time. Laboratory experiments can be slow and of high cost,
whereas empirical relations are hardly transferable to other, more complex types of
(reservoir) rocks. Numerical simulation techniques have been consequently advanced
for this field of application. Most commonly used for computational fluid dynamics
are Finite Difference (FDM), Finite Volume (FVM) and Finite Element Methods
(FEM), which have been introduced and assessed by many authors generally and in
detail (Wesseling, 1991; Lege et al., 1996; Yeh, 2000; Rausch et al., 2005; Zienkiewicz
et al., 2008; Durran, 2010).
Regardless of the extensive use and the functionality of these methods, they do have
clear limitations of implementing differential equations with increasing complexity
of pore structures (Ladd, 1994; Saeger et al., 1995; Martys and Chen, 1996; Chang
and Alexander, 2008). These limitations either result in very extensive meshing
procedures to catch complex pore structures or in simplifying and rebuilding these
structures into an equivalent counterpart. Due to these limitating factors as well as
to recent advances in three-dimensional imaging techniques and image quality, the
Lattice Boltzmann Method (LBM) marks a promising and reasonable tool for fluid
transport.
Introduction 5
Traditional computational fluid dynamic (CFD) methods solve the conservation

equations of macroscopic properties numerically (i.e. mass, momentum, and en-
ergy). The LBM is based on a statistical - that means macroscopic - description of
a microscopic phenomenon (Chen and Doolen, 1998; Sukop and Thorne Jr., 2010).
It idealizes a fluid, which consists of small (fictive) particles. This fluid performs
consecutive propagation and collision processes over a discrete lattice mesh and there-
fore shows almost similar behaviour at a macroscopic scale compared to real fluids.
Other techniques (FEM, FDM) discretize the model and the governing equations
(top-down approach), whereas LBM recovers the governing equation (bottom-up
approach) from the rules of the discrete model (Succi, 2001; Keehm, 2003). This
pushes the LBM to the method of choice to handle complex pore geometries for
simulation of fluid transport.
1.5 Motivation and Chapter Description

As a result of the ongoing technical and instrumentational development of 3D imaging
techniques, network extraction algorithms and computational power, it is mandatory
to extend fluid flow modeling to the next level of resolution (micron and sub-micron
range). Furthermore, it is also mandatory to extend the complexity of used pore ge-
ometries as well as to increase the knowledge about fluid flow properties in complex
systems. Thus, this thesis places emphasis on two parts of advanced pore scale fluid
transport modeling: First, on the derivation, segmentation and extraction of real
pore networks from next generation high resolution micro CT imaging data.
Second, on implementing and assessing the Lattice Boltzmann Method for accu-
rate estimation and evaluation of fluid flow properties in complex porous media.
Chapter 2 provides the neccessary theoretical background information on high res-

olution micro CT imaging, including hardware functionality, sample preparation,
basics of image reconstruction and post scan processing to derive sharp (in terms of
image detail and image contrast) 3D data sets.
Chapter 3 and 4 introduce the basics of fluid dynamics and of the Lattice Boltz-
mann Method (LBM), starting from the theory of lattice gas automata, continuing
to the Boltzmann Equation and ending up with common models.
6
Neccessary information about basic petrophysical properties and mineralogical com-

position of the investigated rock samples are given in chapter 5.
Chapter 6 introduces and discusses two different methods to determine the Repre-
sentative Elementary Volume (REV) for the different types of sandstone that have
been used for this thesis. Chapter 7 complements the previous chapter by introduc-
ing a new technique for an advanced inhomogeneity analysis.
In chapter 8 the advanced pore network extraction as well as the meshing and mesh
optimization process is described in detail.
Chapter 9 includes the results of the benchmark simulation as well as the results,
interpretation and discussion of singlephase fluid flow modeling using the LBM.
Chapter 10 concludes and summarizes the major results of this thesis and takes
a brief look onto future prospects.
Chapter 2
High Resolution Micro CT

Imaging
For this thesis, several series of measurements on three different types of (porous)
sandstones have been carried out. Besides a detailed petrophysical characterization,
extensive 3D imaging with next generation and high resolution X-ray Computed To-
mography has been performed. This so called micro CT offers the possibility to
investigate rock samples non-destructively and in three dimensions. It can reveal
qualitatively interior structures, different types of porosity, pore space interconnec-
tivity and mineralogical components (e.g. matrix and cementation minerals).
In this chapter, the basic theory and working principle of high resolution CT imaging
and image reconstruction, as well as special demands on sample preparation and the
general scanning workflow will be introduced.
2.1 Basic Theory and Principle

Micro CT imaging is a complex technique and comprises of even more complex
physical and mathematical processes and operations. Therefore it is legitimate to
give a brief but nevertheless comprehensive theoretical overview.
7
8 High Resolution Micro CT Imaging
2.1.1 X-ray Generation

X-rays are electromagnetic waves with wavelengths between 108 m and 1013 m
and with photonic energies between 1 keV and 250 keV. They are generated when fast
electrons collide with (dens) matter. X-ray energy depends on the electron velocity
, which is proportional to the acceleration voltage Ua between anode and cathode.
When electrons hit the anode material - usually consisting of tungsten or a tungsten
alloy - primarily two atomic processes are responsible for X-ray photon emission:
First, by the so called Bremsstrahlung (also braking radiation or deceleration

radiation), which generates the continuous X-ray spectrum. X-ray photons are
emitted when electrons are decelerated and / or diffracted within the Coulomb
field of atoms (figure 2.1, A).
Second, by high-energy transitions within the electron shell of atoms and

molecules, which forms the characteristic X-ray spectrum. An incident elec-
tron can impart some of its energy to an electron at a lower shell so that this
electron is extracted. Next, an electron of a higher shell drops into the ener-
getic favorable lower shell and emits a characteristic photon. This process is
possible for different shell combinations (figure 2.1, B).
Figure 2.1: Processes responsible for X-ray photon emission: Bremsstrahlung

(A) and high-energy transitions within the electron shell of an atom (B).
During X-ray generation, about 99 % of the incoming energy is transformed to heat,

whereas only 1 % is converted into radiation. Figure 2.2 shows a schematic drawing
of the general assembly and functionality of a X-ray tube.
Basic Theory and Principle 9
Figure 2.2: Basic assembly and functionality of a X-ray tube.
2.1.2 X-ray Attenuation

Although X-rays obey a high but material dependent transmission potential, they
are attenuated during the transmission of the investigated object. This attenuation
is depending on sample thickness, sample density and sample composition and can
be physically described by Rayleigh scattering, by photoelectric effect, by Comp-
ton scattering, and by pair production. These effects are extensively described in
several specialized books (Gerthsen and Meschede, 2003; Demtrder, 2010) and are
therefore not closely introduced. The attenuation of X-rays transmitting through a
homogeneous material with a defined thickness of x can be exemplified as shown
in figure 2.3. The attenuation of the X-ray intensity I(x+x) after transmission
through matter results in
I(x + x) = I(x) (x) I(x) x, (2.1)
where is called attenuation coefficient.
Figure 2.3: Exemplified X-ray attenuation through a homogeneous material.

Modified after Buzug (2005).
This formula finally leads to the Lambert-Beers Law of attenuation for homogeneous
materials:
I(x) = I0 ex , (2.2)
with I0 as initial X-ray intensity and x as the length of the X-ray path through the
entire material.
It is obvious that natural rocks are not homogeneously composed. X-rays are trans-
mitted through different mineral phases, void space or even different types of fluid,
whereat all these materials hold different attenuation coefficients i (figure 2.4). This
leads to an overall intensity Im through the multi-composed material:
Im = I0 (1 1 x) (1 2 x) (1 i x) (1 m x). (2.3)
This equation could be further simplified and finally leads to:

m
P
i x
Im = I0 e i=1 , (2.4)
where the increment i reflects an individual material component with a corresponding

attenuation coefficient i over a discrete path length x. For the investigated rock
samples of this thesis, the amount of phases is limited and the most distinguishable
ones are void space and main matrix forming minerals, respectively.
Figure 2.4: Exemplified X-ray attenuation through a multi-component mate-

rial. Modified after Buzug (2005).
2.1.3 Basic Principle of CT Scanning

High resolution CT scanning of rock samples is similar to conventional (medical) CT
scanning. The most significant difference is that in case of geoscientific CT scanning
the sample turns on its own axis, whereas in medical applications, source and detec-
tor are moving and the patient stays fixed. The main components of all CT devices
are: the X-ray tube, the computerized numerical control unit (CNC) for positioning
Basic Theory and Principle 11
and rotation of the sample, and the special detector unit. Besides, high-performance
computers are connected to control and monitor the CT device (X-ray source as well
as scanning parameters) for image data acquisition and image data reconstruction,
respectively.
Figure 2.5 illustrates the basic setup of a CT scanning device. In this case, the
X-ray tube is a so called microfocus tube, which is essential for high resolution
scanning. The rock sample is specially prepared and fixed on the CNC. This highly
precise sample stage rotates the object stepwise in sections considerably smaller than
one degree. This controls the amount of possible projections to be accumulated and
therefore directly influences scan quality. For the detector, commonly charge-coupled
devices (CCD) are used, whilst recently complementary metal oxide semiconductors
(CMOS) have been installed more often. The reason for the change from CCD to
CMOS is the significantly better signal to noise ratio as well as an increased detection
speed and a bigger detector area.
Figure 2.5: Basic setup of a CT scanning device: X-ray source (left), CNC
stage and specimen (middle) and flat panel detector unit (right).
2.1.4 Image (Spatial) Resolution and Sharpness

The term image resolution is commonly used if CT scanning is discussed. This
term is not as unambiguously as it seems and therefore it is neccessary to point out
the correct definitions. Figure 2.6 shows the geometrical limits which define image
magnification, voxel resolution and image sharpness. The image magnification M is
defined as the ratio of focus detector distance (FDD) to focus object distance (FOD):
M = F DD/F OD. (2.5)
The so called voxel resolution V is now defined as the ratio of pixelsize P on the
detector to image magnification M :
V = P/M. (2.6)
Figure 2.6: Schematic diagram of geometrical limits controlling image resolu-

tion.
So, the larger the FDD, the larger the magnification of the object. Furthermore, the
larger M, the smaller the voxelsize V (because P is constant). Unfortunately, with
increasing FDD, image contrast and sharpness are greatly decreased due to a higher
X-ray pathlength and a higher factor of unsharpness U, the so called penumbra
effect:
U = (M 1) F, (2.7)
where F equals the focal spot size of the X-ray source. As a result, F is the pre-
dominating factor, which effects image sharpness and consequentially image quality
and resolution. To avoid or at least to minimize the penumbra effect, focal spot size
needs to be reduced significantly (figure 2.7). In theory, it should be smaller than
the adjusted voxelsize to achieve optimal scanning results. Practically, modern high
resolution X-ray tubes have different types of magnetic lenses so that the focal spot
size can be adjusted for different sample sizes and scanning purposes.
3D Image Reconstruction 13
Figure 2.7: Schematic illustration of the influence of the focal spot size on the
penumbra effect.
2.2 3D Image Reconstruction

During the scanning process only two dimensional projections of the stepwise rotating
sample are collected. These 2D slices need to be reconstructed in order to derive a
three dimensional view of the investigated specimen. The main steps of 3D image
reconstruction are:
2D data acquisition,
extensive filtering processes,
the back projection process and
3D image visualization.
The complete process sequence is schematically shown in figure 2.8 and obeys com-
plex mathematical methods. After acquisition, horizontal image data are zero padded
and ramp filtered in the Fourier domain. Then, back projection process takes place,
where the filtered cone beam projections are transferred with appropriate weights.
It is remarkable that the basics of all modern imaging techniques have already been
developed in 1917 - without any practical field of application - by the austrian math-
ematician Johann Radon. Nevertheless, 3D back projection technique for cone beam
geometry has not been developed until the 1990s (Grangeat, 1990). By then, only
parallel beam image reconstruction has been possible. Therefore cone beam back
projection will be briefly described due to its great impact on modern CT imaging.
Figure 2.8: Overview of processes involved from image acquisition to image

reconstruction. Modified after Buzug (2005).
The first step of the reconstruction technique is to weight the 2D projections:
Xwc {, (, ())} = cos()X c {, (, ())} , (2.8)
where Xwc equals the weighted (X-ray) cone beam transformation tensor and the
attenuation coefficient for different projection angles ( , , ); and are projec-
tion plane vectors and is the direct distance vector to the detector plane.
Next, it is necessary to calculate the partial derivation G in direction of the main

detector axis x and y therefrom:
c
Gx (, ) = X {, (, ())} (2.9)
x w
and
c
Gy (, ) = X {, (, ())} . (2.10)
y w
A major step is the calculation of line integrals upon the detector plane and weight-
ing of the result with the cone beam factor 1/ cos2 () to get access to the radial
derivation of the Radon transformation p, :
Z
1
p, () = Gx (, ) sin() + Gy (, ) cos()d. (2.11)
cos2 ()

The nanotom 180 CT Scanner 15
This equation needs to be transformed and interpolated onto vertical meridian Radon
planes, so that the filtered back projection h could be applied:
Z/2
1 2 p, ()
h [X p {, (, )}] = 2 sin()d. (2.12)
4 2
/2
Finally, back projection within horizontal planes through integration on the plane
angle of rotation can be applied to receive the 3D data:
Z2
1
f (x, y, z) = h [X p {, (, )}] d. (2.13)
2
=0
More details about the extensive mathematical theory of 3D image reconstruction

can be found in Buzug (2005) and Grangeat (2009).
2.3 The nanotom 180 CT Scanner

The CT device that has been used for imaging in this thesis is a nanotom 180
device (figure 2.9), located in the petrophysics laboratory at the Leibniz Institute
for Applied Geophysics in Hannover, Germany. It is manufactured by GE Sensing
and Inspection Technologies and is part of the productline of phoenix|x-ray, located
in Wunstorf, within the vicinity of Hannover. The nanotom 180 has been especially
designed for the investigation of small (rock) samples, to achieve 3D image data
sets of high quality and with very high resolution (Brunke et al., 2008). This de-
vice is equipped with a special water-cooled nanofocus X-ray tube (180 kV, 15 W),
which allows several modes of operation depending on sample size and habit for best
possible results. The minimum focal spot size is about 0.6 m which results in a
detail detectability of about 0.2 m. In practice, it is possible to derive voxel reso-
lutions smaller than 1 m and to achieve detail detectabilities of about 350 - 700 nm.
The detector is a Hamamatsu CMOS flat panel detector with a pixel size of 50
m and a field of view (FOV) of 2300 x 2300 pixel (5-Megapixel). The FOV can
be laterally (virtually) extended by a factor of three so that even larger samples can
be investigated with high resolution. The included acquisition and reconstruction
software datos|x allows parallelized image processing, resulting in reconstruction
times for a 2000 x 2000 x 2000 voxel volume of less than 15 minutes. Reconstruction
is perfomed on multiple grapical processing units (GPU, nvidia GeForce 8000, 1 GB
RAM) of a fourfold DELL IntelCore2Quad cluster (3.5 GHz, 16 GB RAM each).
Figure 2.9: The nanotom 180 high resolution CT device from GE Sensing and
Inspection Technologies, productline phoenix|x-ray.
2.4 Sample Preparation
The maximum sample size for imaging with the nanotom 180 is approximately 150
mm x 120 mm (height x diameter) and the maximum sample weight can reach 2 kg.
Practically, small samples with an ideal rotational symmetry (e.g. small cylindrical
samples) are needed to derive high resolution data.
For this work, special CT imaging samples have been drilled out of laboratory scale
cores (30 mm in diameter, 40 mm in length) after detailed core analysis. The attain-
able diameter of the CT miniplugs is strongly dependent on sample composition
and cementation. In practice, diameters of 3.5 to 5.0 mm for poorly cemented sand-
stones and 1.5 to 2 mm for dense and well cemented sandstones are possible to
achieve. These small samples are dried carefully (since they have been drilled under
wet conditions) and are fixed to the top of thin acrylic glas rods (3 mm in diameter).
The acrylic rods are small enough to get close to the X-ray source and still have an
adequate stiffness to prevent blurring effects. Samples are fixed with a special fast
drying glue, which does not penetrate into the pore space and simultaneously allows
easy sample displacement without severe sample damaging.
Imaging Workflow 17
2.5 Imaging Workflow

In this section a brief introduction of the basic workflow for high resolution CT imag-
ing with the nanotom 180 device is given.
I. Sample Positioning
For high resolution CT scans, samples need to be placed as close as possible to the
X-ray source, whereas the detector needs to be moved as close as possible to the
sample. Only with these prerequisites done, best possible results in terms of resolu-
tion, contrast and sharpness can be derived. Additionally, it is no longer necessary
to retain a small area of clear void space (air) on both sides of the specimen for a
distinct phase differentiation. For this study, a sample diameter of 3.5 mm enables a
voxel resolution as small as 1 m, which is reasonable for small pore space features
in dense and highly cemented rocks.
II. X-ray Parameter Configuration

In a next step, X-ray parameters like tube voltage and X-ray filament current need
to be configured. Tube voltage controls the radiography of the sample and filament
current controls the phase contrast in general. For this device, tube voltage should
be choosen to achieve a minimum gray value (GV) within the sample image on the
detector (likely the most dense mineral phase) in the range of 180 - 200 GV. Con-
trary to this, filament current should be increased until a target power of 1.5 W or
an image GV of 3600 is reached. For the investigated rock samples, tube voltage
ranges in between 50 - 75 kV and filament current in between 90 - 120 mA.
III. Detector Calibration

Detector calibration needs to be applied in order to remove inhomogeneities on the
background images. These inhomogeneities are small variations in the X-ray beam
as well as a temperature dependent detector shift, resulting in non-uniform response
times. For this purpose, two blank images are required: First, with X-rays turned
off (so called offset calibration). Second, with activated X-ray beam (so called gain
calibration). An improper detector calibration may lead to severe circular artifacts
and therefore to 3D data sets of less quality. Additionally, detector acquisition time
should be bigger than the internal response time of the electronic devices. For this
study, a detector acquisition time of 1000 ms was adequate, since the dead time of
the detector unit amounts to 700 ms.
IV. Scan Parameter Configuration

Last, scanning parameters need to be configured: total number of projections, image
stack average and skip. At this point, a compromise between scanning time and
image information gain has to be made. All investigated samples have been scanned
with 1500 projections, 8 average and 1 skip, resulting in a total scan time of 230
minutes per sample to ensure equal quality for all investigated samples.
Chapter 3
Basics of Fluid Dynamics
Generally, fluid dynamics is a sub-discipline of fluid mechanics and handles the field
of fluid flow. It has been a wide field of research for several centuries, probably start-
ing with the ancient Principle of Archimedes. The biggest impact has been derived
by scientists and engineers like Isaac Newton ( 1643), Leonhard Euler ( 1783),
Claude-Louis Navier ( 1836), Henry Darcy ( 1858), Jean Louis Marie Poiseuille
( 1869), Gotthilf Hagen ( 1884), George Gabriel Stokes ( 1903) and Osborne
Reynolds ( 1912), who greatly advanced the understanding and the mathematical
theory of fluid flow.
For this thesis, only the basic principles of fluid dynamics will be introduced since
several excellent textbooks exist, which include detailed mathematical and physical
descriptions (Bear, 1988; Wesseling, 1991; Batchelor, 1994). Special emphasis will be
given on the derivation and description of the Navier-Stokes equations, which form
the basics of fluid mechanics.
3.1 Darcys Equation

As far as fluid flow is mentioned within petrophysics, mostly the term of permeabil-
ity and furthermore Darcys law and Darcys equation are related to this topic. Due
to the significant impact on hydrology, subsurface flow and especially on laboratory
scale determination of (rock) permeability, it seems comprehensible to start from
this point into the topic of fluid dynamics.
Originally, Henry Darcy (working as an engineer for hydraulics) investigated wa-

ter flow in vertical homogeneous sand filters within the vicinity of Dijon (France).
19
20 Basics of Fluid Dynamics
Figure 3.1 allegorizes the original experimental set-up that has been used. Darcy
has shown empirically that the flow rate q is:
1. proportional to the streamed cross section area A;
2. proportional to a material dependent factor, called permeability k ;
3. proportional to the occurring pressure drop pinlet - poutlet = p (equivalent to

the water head height difference h1 - h2 = h) and
4. inversly proportional to fluid viscosity and to the length L of the porous

medium.
Figure 3.1: Darcys experimental set-up for determination of (rock) permeabil-

ity. Modified after Batchelor (1994).
These results are expressed in the following, most common form of Darcys equation:
q k p
v= = , (3.1)
A L
where v or also (more often and in the following used) ~u equals fluid velocity. Regard-
less of its impact, Darcys equation is only valid for slow, i.e. laminar, viscous fluid
flow on a macroscopic scale (Bear, 1988). Hence, Darcys equation is inappropriate
for modeling fluid flow on a microscopic (pore) scale. Furthermore, this experimental
law is - strictly speaking - a special case of the Navier-Stokes equations and can be
derived from them by the mathematical operation of homogenization. But before
the Navier-Stokes equations are described more in detail, several basic prerequisites
need to be introduced.
Prerequisites for Navier-Stokes Equations 21
3.2 Prerequisites for the Navier-Stokes Equations

The Navier-Stokes equations form a set of partial differential equations that are
derived from conservation laws, the transport theorem and the fundamental theorem
of calculus. Therefore it seems reasonable to take a closer look at a selection of
prerequisites to obtain the Navier-Stokes equations at the end.
3.2.1 Reynolds Transport Theorem

Reynolds transport theorem (also known as the Leibniz-Reynolds transport theo-
rem) is a generalization of the Leibniz integral rule which is an essential mathematical
tool for the theorem of calculus. In general, it can be reduced and simplified to the
following statement: What was there, plus what goes in, minus what comes out, is
equal to what is there. This theorem is used to describe the basic conservation laws
of continuum mechanics, since it is more convenient to calculate with distinct con-
trol volumes than to identify and follow a system of discrete fluid particles (Potter
and Foss, 1982). The composition of both (fluid volume and particle description) is
enabled by the transport theorem, which has been named after Osborne Reynolds.
The transport theorem states that for any volume V(t) and differentiable scalar
field , there is:
Z Z
d
dV = + ~u dV. (3.2)
dt t
V (t) V (t)
An extensive proof of this theorem can be found amongst others in (Bear, 1988) and
(Wesseling, 1991). Hence, the detailed derivation is disregarded.
3.2.2 Conservation of Mass

The law of conservation of mass is described by a partial differential equation and
states that the rate of change of mass in a random volume V(t) equals the rate of
mass production within this volume. This is mathematically expressed as follows:
Z Z
d
dV = dV. (3.3)
dt
V (t) V (t)
Here, (t,x) is the density of a material particle at time t and at position x, whereas
(t,x) equals the mass production rate at defined time and position per volume. For
most applications it is legitimate to assume = 0, since 6= 0 is only valid for
multiphase flow, where is considered separately for each fluid phase (Wesseling,
1991). Using Reynolds transport theorem (equation 3.2) leads to:
Z

+ ~u dV = 0. (3.4)
t
V (t)
Since this holds for every V(t), the integrand must equal zero so that the continuity
equation is derived:

+ ~u = 0. (3.5)
t
The term ~u is also known as the mass current density j. The law of mass conservation
can be rewritten as follows, if the product rule for divergence is applied:

+ ~u 4 + ~u. (3.6)
t
For the time dependent change of the density of a particle, which moves along a
distinct path x(t), this leads to:
d dx
(t, x(t)) = + 4 = ~u. (3.7)
dt t dt
This means that density changes with the divergence of the velocity field. If density
stays the same during motion, equation (3.5) leads from a mass continuity equation
to a simplified volume continuity equation:
~u = 0, (3.8)
which is also called the mass conservation law for incompressible flow.
At this point, a special remark should be given. Incompressibility is often mis-

leadingly assumed to be a fluid property. In fact, it is a flow property! Wesseling
(1991) has given a proper description, that compressibility only depends on the speed
of flow so that the following definitions can be applied:
If the velocity of flow is in order of speed of sound, flow is designated as

compressible.
If the velocity of flow is significantly smaller than the speed of sound, which
can be legitimately presumed for liquids, flow is designated as incompressible.
Prerequisites for Navier-Stokes Equations 23
3.2.3 Conservation of Momentum

The law of conservation of momentum is also known as Newtons second law. It
says that the rate of change of momentum of a volume equals the total force on this
volume. Applying this onto a specific volume V(t) leads to:
Z Z Z
d
u dV = Fb dV + Fs dS, (3.9)
dt
V (t) V (t) S(t)
where Fb and Fs are denoted as body and surface forces respectively. Exchanging
the material property in the transport theorem (equation 3.2) by u gives:
Z Z Z
u
+ (u u ), dV = Fb dV + Fs dS. (3.10)
t
V (t) V (t) S(t)
Surface forces Fs can also be expressed by:
Fs = n , (3.11)
with the stress tensor and the outward unit normal n on the surface dS. Applying
the divergence theorem, and substitution of the material property with the stress
tensor notation (3.11), equation (3.10) can be rewritten as:
Z Z
u
+ (u u ), dV = Fb + , dV. (3.12)
t
V (t) V (t)
Since this must be valid for every V(t), we can simplify (3.12) to:
u
+ (u u ), = , + Fb , (3.13)
t
which forms the law of conservation of momentum in Cartesian tensor notation.
3.2.4 Conservation of Energy

Conservation of energy is defined by the first law of thermodynamics: a change in the
internal energy of a closed thermodynamic system is equal to the difference between
the heat supplied to the system and the ammount of work done by the system on its
surroundings (Kondepudi, 2008). Applying this thermodynamic law on a material
volume V(t) gives:
Z
d
EdV = W + Q, (3.14)
dt
V (t)
where E is the energy per unit mass, W the rate of external work on the fluid in the
material volume, and Q is the rate of heat variation. Energy, rate of work and rate
of heat variation are defined as:
1
E = e + u u , (3.15)
2
Z n o
W = u fb + (u ) , dV, (3.16)
V (t)
Z
Q= {q + (T, ) , } dV, (3.17)
V (t)
with as thermal conductivity, T as temperature and q as heat rate per mass unit.
Again, if we use the transport theorem on the first law of thermodynamics, and if
furthermore the definitions of the rate of work W, and the rate of heat variation Q
are implemented, we get:
Z Z n
E o
+ (u E) , dV = (u ) , + (T, ) , +u fb + q dV.
dt
V (t) V (t)
(3.18)
Similar to the other conservation laws, this condition must be valid for all material
volumes V(t). With this assumption, the so called energy equation can be derived:
E
+ (u E) , = (u ) , + (T, ) , +u fb + q. (3.19)
dt
Derivation of the Navier-Stokes Equations 25
3.3 Derivation of the Navier-Stokes Equations

The Navier-Stokes equations form the fundamental mathematical model of fluid dy-
namics. Therefore it seems reasonable to introduce the derivation of this model in
detail. It is remarkable that - so far - it has not been proven that in three dimensions
solutions always exist (Navier-Stokes existance problem) or that if they do exist, they
do not contain any singularity (Navier-Stokes smoothness problem). The Clay Math-
ematics Institute (Cambridge University, USA) has called the Navier-Stokes problem
to be one of the seven most important problems in mathematics to be solved.
So far, most of the prerequisites for the derivation of the Navier-Stokes equations
have been given in the previous chapter: the basic transport theorem, the continuity
equation and accordingly the law of conservation of momentum. Hence, it is possible
to take a closer look onto the mathematical formalism leading to the Navier-Stokes
equations.
3.3.1 The Navier Equations

In order to derive the Navier-Stokes equations, it is neccessary to link the stress
tensor , as used before, directly to the fluid velocity (fluid motion) ~u. For this
purpose, the stress tensor needs to be described more in detail. Rewriting the tensor
to matrix form leads to:

xx xy xz

=
yx yy ,
yz (3.20)
zx zy zz
where represents shear stress and normal stress, respectively. In general, stress
is defined as force over an area. As an example, xx = Fx /Ax stands for the force in
x-direction that acts over an area in this direction (zy-plane).
As described in chapter 3.2.1, it is more appropriate to use control volumes than

to describe a system of millions upon millions of fluid particles. For this, we assume
and center a small volume 4V = 4x 4y 4z at position (x,y,z) and further assume
that the stress tensor is known at the center of this volume. Additionally, all
stresses are defined as positive, i.e. forces act on positive faces in positive direction,
or on negative faces in negative direction, respectively. For clarification, figure 3.2
exemplarily shows such an arrangement for the y-faces.
Figure 3.2: Definition of appearing normal and shear stresses for the y-faces
of a small control volume 4V .
For further sequel, and due to symmetry of the stress tensor , only one specific di-
rection is observed for derivation. The stress term on the negative coordinate y-face
can be described as:

yy 4y
yy 4x4z. (3.21)
y 2
The origin of the stress yy (x, y, t) is located at the center of the volume but since the
value of the stress on the y-face needs to be obtained, the term has to be expanded
in the y-direction by half of the volume edge length 4y:

4y
yy x, y ,z . (3.22)
2
For complementation, (3.22) needs to be expanded by using Taylor series about

yy (x, y, z) and its y-derivates at (x,y,z). By doing this, the shear value can be
obtained (Batchelor, 1994), so that the force in the y-direction can be assumed as a
first order approximation related to (3.21). Consequently, the next step is to derive
only the corresponding forces that contribute in the y-direction (instead of those that
act on the y-faces). Thus, figure 3.2 is modified, so that the only possible surface
forces in this direction on the control volume (4Fys ) are shown (figure 3.3).
Figure 3.3: Exemplary diagram of the only possible surface forces that con-
tribute in y-direction on a small control volume.
Then, 4Fys can be expressed by:

yy 4y yy 4y
4Fys = yy + 4x4z yy 4x4z
y 2 y 2

xy 4x xy 4x
+ xy + 4y4z xy 4y4z (3.23)
x 2 x 2

zy 4z zy 4z
+ zy + 4x4y zy 4x4y.
z 2 z 2
Since stresses have been assumed to be positive, equation (3.23) can be simplified
to:
yy 4y xy 4x
4Fys = 2 4x4z + 2 4y4z
y 2 x 2
(3.24)
zy 4z xy yy zy
+2 4x4z = + + 4x4y4z.
z 2 x y z
Contrary to this, obtaining the body force along y-direction is trivial:
4Fyb = By 4x4y4z. (3.25)
Thus, the momentum equation holds for the equivalent component:

X X
Fys + Fyb = m ay . (3.26)
With the definition of the mass m = 4V = 4x4y4z and acceleration as the

total derivative ay = DV /Dt, the momentum equation finally forms to:

xy yy zy v v v v
+ + + By = +u +v +w . (3.27)
x y z t x y z
Here, the first three terms represent the surface forces per unit volume, the fourth
term represents the body force per unit volume and the right hand side equals the
inertia force per unit volume. Due to symmetry effects, the second subscripts of the
stress terms always stay y, whereas the first subscripts - which represent the areas,
where forces act on - change from x to z for each successive term. Equivalently,
the derivates change for each term, so that the equations for the x- and z-direction
can be quickly complemented:

xx yx zx u u u u
+ + + Bx = +u +v +w ; (3.28a)
x y z t x y z

xy yy zy v v v v
+ + + By = +u +v +w ; (3.28b)
x y z t x y z

xz yz zz w w w w
+ + + Bz = +u +v +w . (3.28c)
x y z t x y z
These three notations are known as Naviers equations. Since there remain nine
unknown variables in this formalism, a set of constitutive relations needs to be applied
in order to make Naviers equations solvable. This is done by implementation of the
so called Stokes relations for normal stresses and shear stresses , which can be
expressed in Cartesian coordinates as follows:

v u
xy = yx = + , (3.29a)
x y

w v
yz = zy = + , (3.29b)
y z

u w
zx = xz = + , (3.29c)
z x
2 u
xx = p ~u + 2 , (3.29d)
3 x
2 v
yy = p ~u + 2 , (3.29e)
3 y
2 w
xx = p ~u + 2 , (3.29f)
3 z
where is the (dynamic) fluid viscosity and p is the pressure. A complete and
extensive derivation of Stokes relations can be found in Bear (1988).
3.3.2 The incompressible Navier-Stokes Equations

For geoscientific applications in general, and for pore scale fluid flow in particular,
incompressible flow can be assumed (Wesseling, 1991). As shown before, the term
~u may be set to zero, related to the continuity equation for incompressible flow
(equation 3.8). This condition can be substituted into (3.29), which then can be
substituted into Naviers equations (3.28):
2
u 2u 2u

p u u u u
Bx + + 2 + 2 = +u +v +w ; (3.30a)
x x2 y z t x y z
2
2v 2v

p v v v v v
By + + + = +u +v +w ; (3.30b)
y x2 y 2 z 2 t x y z
2
w 2w 2w

p w w w w
Bz + + + = +u +v +w . (3.30c)
z x2 y 2 z 2 t x y z
This triplet forms the so called incompressible Navier-Stokes equations (or Navier-
Stokes equations for incompressible fluid flow). By adding the continuity equation,
a set of four equations and four dependent (unknown) variables (u,v,w and p) is
derived, so that this fluid dynamic problem now becomes solvable.
For completeness, the commonly used vector form of the incompressible Navier-
Stokes equations is:
DV ~ p + 2 ~u.
~u = 0 and = B (3.31)
Dt
It is very often useful to express the Navier-Stokes equations (3.31) in dimensionless
form. For this, flow is supposed to have a characteristic length scale L and a char-
acteristic velocity scale U (Wesseling, 1991). Hence, dimensionless variables can be
introduced, so that by substitution into equation (3.31) the so called dimensionless
Navier-Stokes equation can be derived:
D~u ~ + 1 2 ~u.
= p + B (3.32)
Dt Re
Re is called Reynolds number and is defined as Re = ~uL/. The benefit of this
notation can directly be seen from equation (3.32): for Re 1, inertia is predominant
and for Re 1, viscosity (inner friction) is dominating.
3.3.3 The compressible Navier-Stokes Equations

If compressible flow is taken into account, the stress tensor notation needs to be
reconsidered. According to Newtons law of viscosity, stress components are related
to the first (or dynamic) coefficient of viscosity , and to the second coefficient of
viscosity (sometimes named bulk viscosity). Then, the stress tensor can be
described as:
ui uj
ij = + + ij ~u, (3.33)
xj xi
where ij is the Kronecker-Delta. The three normal and six shear stresses then
transform to:

v u
xy = yx = + , (3.34a)
x y

w v
yz = zy = + , (3.34b)
y z

u w
zx = xz = + , (3.34c)
z x
u
xx = 2 + ~u, (3.34d)
x
v
yy = 2 + ~u, (3.34e)
y
w
xx = 2 + ~u. (3.34f)
z
Applying the same mathematical formalism as before, the compressible Navier-Stokes

equations (or Navier-Stokes equations for compressible fluid flow) can be derived:

p u u v
Bx + 2 + ~u + +
x x x y y x
(3.35)
u w u u u u
+ + = +u +v +w .
z z x t x y z
The equations for the y- and z-component can be obtained equivalently. Again, and
for completeness, the compressible Navier-Stokes equations can be written in vector
form as:
~u
+ ~u ~u = p + ~u + (~u)T
t (3.36)
~
+ ( ~u) + B,
where (~u)T is the transpose of ~u.

Chapter 4
The Lattice Boltzmann Method

(LBM)
In this chapter, a detailed overview of the Lattice Boltzmann Method (LBM), its
origin and its derivation will be given. For classical CFD purposes, standard top
down and bottom up approaches to model the behavior of physical systems are
widely used (Chen et al., 1994; Succi, 2001; Sukop and Thorne Jr., 2010).
The top down approach starts with a continuum description of macroscopic phe-
nomena, provided by partial differential equations - such as the Navier Stokes equa-
tions for incompressible fluid flow, which have been introduced in chapter 3. Bottom
up approaches are based on the microscopic, i.e. particle description, which are pro-
vided by molecular dynamics. In this case, position and velocity of each particle in
the system are linked by solving Newtons equation of motion. These approaches
are more suitable for understanding fundamental microscopic interactions than for
understanding macroscopic dynamics (Chen et al., 1994).
The LBM can be generally described as a mesoscopic approach (Chen et al., 1994).
Nevertheless, starting point of of the LBM is the Boltzmann equation, which forms
the (microscopic) fundament of this method.
4.1 The Boltzmann Equation

The Boltzman equation - else known as the Boltzmann transport equation - forms
the fundamental differential equation of kinematic gas theory as well as of non-
31
32 The Lattice Boltzmann Method
equilibrium thermodynamics. It was named after the austrian physician Ludwig

Boltzmann. This equation is used to calculate and describe the statistical distribu-
tion of particles within a fluid. Starting point is the derivation of an equation for the
time evolution of a distribution (properly a density) function, which is defined as
f (~x, ~v , t)d~xd~v , (4.1)
where t equals time, and ~x and ~v the particle position and velocity, respectively.
Two assumptions now need to be considered. First, those particles described by the
distribution function f experience an external force F. Assuming the abscence of
collisions (i.e. the number of particles within the original volume element d~xd~v is
equal to the number of particles within the volume that has experienced an external
force), the density distribution function can be described as follows:
1
f (~x, ~v , t) = f (~x + ~v dt, ~v + F dt, t + dt). (4.2)
m
As a result it can be assumed that with abscence of collisions, particles within the
described space behave similar as particles of an incompressible fluid (Haug, 2006).
Second - and more obvious - it may be assumed, that collisions do occur. In this case,
particle density within the phase space volume d~xd~v changes, leading to a modified
balance equation:

f 1
f (~x, ~v , t) + dt = f ~x + ~v dt, ~v + F dt, t + dt . (4.3)
t coll m
(f /t)coll is called the collision term. By developing equation (4.3) after dt and
then taking the limit of dt against zero, the Boltzmann equation is derived (Qian
and Orszag, 1993):

f ~ 1 ~
= + ~v ~x + F ~v f (~x, ~v , t), (4.4)
t coll t m
with the so called collision and streaming term forming the left and the right hand
side of the equation, respectively.
So far, this result is without any practical use, since the collision term is still un-
defined. A key finding of Boltzmann was the development of a balance equation of
particle gain- and loss processes for the phase volume d~xd~v in the time interval dt.
This is the so called Stozahl or molecular chaos approach (Arkeryd, 1972a).
The loss of particles within dt caused by collisions is given by:
dN dt = f (~x, ~v , t)d~xd~v dt. (4.5)

The Boltzmann Equation 33
The gain of particles within dt caused by collisions is equivalently given by:
dN+ dt = f+ (~x, ~v , t)d~xd~v dt. (4.6)
With this, the collision term as introduced with equation (4.4) forms to:

f
= f (~x, ~v , t) + f+ (~x, ~v , t). (4.7)
t coll
To calculate f (~x, ~v , t), a volume element d~x and a particle with a velocity of (~v +
d~v , ~v ), which changes its velocity due to collision with another particle within the
same volume and with a different velocity (v~1 + dv~1 , v~1 ), needs to be considered.
Then, f (~x, ~v , t)d~xd~v within a time interval dt is formed by multiplication of different
terms:
by the probability of velocity change of both particles (~v ~v and ~v1 ~v1 )
given by (~v , ~v1 ~v , ~v1 )d~v d~v1 ;
by the relative flux of particles with the velocity ~v1 to particles with the velocity
~v , given by |~v ~v1 | f (~x, ~v , t)d~v1 ;
by total number of particles with velocity ~v within the volume element, given
by f (~x, ~v , t)d~xd~v .
This leads to:

Z Z Z
f (~x, ~v , t)d~xd~v = [(~v , ~v1 ~v , ~v1 )d~v d~v1 ]
~v1 ~v ~v1
(4.8)
[|~v ~v1 | f (~x, ~v , t)d~v1 ]
[f (~x, ~v , t)d~xd~v ] .
Due to symmetry and elastic collisions (conservation of momentum and energy),

the Boltzmann equation with molecular chaos assumption finally forms to (Arkeryd,
1972b):

~ 1 ~
+ ~v ~x + F ~v f (~x, ~v , t) = f (~x, ~v , t) + f+ (~x, ~v , t)
t m
Z Z Z
= d~v1 d~v1 d~v (~v , ~v1 ~v , ~v1 ) |~v ~v1 | (4.9)
~v1 ~v1 ~v
[f (~x, ~v , t)f (~x, ~v1 , t) f (~x, ~v , t)f (~x, ~v1 , t)] .
The Boltzmann equation holds the boundary value feature (for small medium free
path lengths) that it merges to the Navier-Stokes equation. More detailed derivation
and theoretical background can be found in literature (Arkeryd, 1972a,b; Haug,
2006).
4.2 Lattice Gas Automata (LGA)

The first successful application of the Boltzmann equation has been set up by the so
called Lattice Gas Automata (LGA), proposed by Frisch et al. (1986) and Wolfram
(1986). The LGA is constructed as a simplified, fictious molecular dynamic in which
space, time and particle velocities are all discrete related to Chen and Doolen (1998).
In general, it consists of a regular lattice with discrete particles residing upon the
lattice nodes. In case of the first LGA derived by Frisch et al. (1986), a lattice site
can be simultaneously occupied by six particles. Hence, this model is called the
six velocity or - related to their developers - FHP-I model. This model basically
consists of two steps: streaming (advection) and collision of the particles. The most
basic collisions of the FHP-I model are shown in figure 4.1.
Figure 4.1: Basic collisions of the FHP-I model (Luo, 2003).
Two particles colliding with opposite momenta will rotate their momenta 60 clock-
wise (or even counter-clockwise) with equal probability. Particle number, momentum
and energy are conserved within the collision process locally and exactely, com-
pared to Luo (2003). After definition of a set of Boolean variables ni (~x + t) with
i = (1, . . . , M ), which describe the particle distribution and where M equals the
number of directions of the particle velocities at each node, the evolution equation
of the LGA can be described as follows:
ni (~x + ~vi , t + 1) = ni (~x, t) + i (n(~x, t)) , (4.10)
with i = (0, 1, . . . , M ). The local particle velocities are given by ~vi , where the index
i denotes the distinct directions as shown in figure 4.1.
Lattice Boltzmann Equation (LBE) 35
Starting from an initial state, evolution at each time element takes place in two steps:
streaming (advection), in which each particle moves to the next (nearest) node
in the direction of its velocity, and
collision, which happens when particles arrive at a node and interact with each
other due to the previously defined collision rules.
Figure 4.2 illustrates this evolution for the fundamental FHP-I model in one time
step (from t to t+4t), related to Luo (2003). In this figure, red colored arrows
represent velocities at time t, green colored arrows correspond to time t + 4t, giving
a good overview about the two basic evolution sequences.
Figure 4.2: Exemplary evolution scheme of the FHP-I model from time t (red)
to t + 4t (green). Modified after Luo (2003).
4.3 Lattice Boltzmann Equation (LBE)

Up to this point, the fundamental ideas of the Boltzmann equation and of the LGA
have been presented. In a last step, these basics need to be transferred to the Lat-
tice Boltzmann equation (LBE) or Lattice Boltzmann Method (LBM), respectively.
Compared to pertinent literature, several ways to obtain the LBE from discrete veloc-
ity models or from the Boltzmann equation exist. This is also valid for the derivation
of the Navier-Stokes equation therefrom (Chen and Doolen, 1998; Luo, 2001; Succi,
2001; Luo, 2003).
Due to the kinetic approach of the Boltzmann equation it seems reasonable to derive
the LBE with a discrete kinetic equation approach. The best and probably most
complete and comprehensive approach is given by Chen and Doolen (1998). Hence,
only the most important steps are presented in this section. The starting point is
similar to the kinetic equation of the LGA as described in equation (4.10):
fi (x + vi 4x, t + 4t) = fi (x, t) + i (f (x, t)) , (4.11)
with i = (0, 1, 2, . . . , M ), and where fi equals the particle velocity distribution func-
tion along a discrete, here the ith direction. i is the collision operator, which
describes the rate of change of the particle velocity distribution fi . If 4x/4t = |vi |,
equations (4.10) and (4.11) hold the same discretizations.
As described, collision operator i must satisfy conservation of mass and momentum

as introduced in chapter 3:
X X
i = 0 and i vi = 0. (4.12)
i i
Chen and Doolen (1998) now assume, that the lattice spacing 4x and the time
increment 4t in equation (4.11) can be regarded as small values of the same order
(denoted by ), if only special physics is of interest. If now a Taylor expansion in
time and space is performed, a continuum form of the kinetic equation, which is
accurate to second order in , is derived:
fi 1 2 fi

fi 1 i
+ vi fi vi vi fi + vi + 2
= . (4.13)
t 2 t 2 t
By performing a so called Chapman-Enskog expansion, and furthermore by ex-

(eq)
panding fi to the local equilibrium distribution function fi and the local non-
(neq)
equilibrium distribution function fi , the macroscopic hydrodynamic equation
can be derived (Frisch et al., 1987). A significant step here is the development of a
linearized collision operator as cited by Chen and Doolen (1998):
i (t) Mij (eq)

= fi fj , (4.14)

where
i f (eq)
Mij = (4.15)
fi
forms the so called collision matrix, which describes the rate of change of velocity
and scattering in directions i and j (Higuera and Jimnez, 1989). Since this ma-
trix only depends on the angle between these directions, it has a limited and hence
straightforward set of values (Chen and Doolen, 1998).
Lattice Boltzmann Equation (LBE) 37
Because Mij still must satisfy conservation of mass and momentum, and further-
more under the assumption of a local particle distribution relaxation to equilibrium
state at a distinct rate , it can be expressed as:
1
Mij = ij . (4.16)

This so called single relaxation time scheme has been set up by Bhatnagar et al.
(1954), and has led to the development of the famous BGK-collision term therefrom:
i 1 (neq) 1 (1) (2)

= fi = fi fi . (4.17)

Finally, this leads to the lattice BGK-equation:
(eq)
fi fi
fi (x + vi , t + 1) = fi (x, t) . (4.18)

The development of the collision term and this equation was an essential milestone,
not only for simplification, but also for greatly reducing numerical noise, which has
been so far a significant problem of the classical LGA (Benzi et al., 1992; Succi,
2001). By using the continuum form of the kinetic equation (4.13), the following
equation to 0th order (0 )
(eq) (1)
fi (eq) fi
+ vi 1 fi = , (4.19)
t1
as well as the equation to 1st order (1 )
(1) (eq) (1) 1 (eq)
fi + fi + vi fi + vi vi fi
t1 t2 2
(4.20)
(eq) 1 2 (eq) 1 (3)
+ vi i f + f = fi
t1 i 2 t21 i
can be derived. Chen and Doolen (1998) have shown, that from these equations the
following mass and momentum equations can be developed:
u
+ u = 0 and + = 0, (4.21)
t t
where is the momentum flux tensor. This package of equations fulfills accuracy
to 2nd order in for the discrete kinetic particle distribution function (4.11). Since
the Navier-Stokes equations possess second order non-linearity, and furthermore by
using the previous assumptions, the final momentum equation results to:

ui
+ j ui uj = i p + j (i uj + j uj ) . (4.22)
t
This equals the Navier-Stokes equations, if the density variation is small enough
(Qian and Orszag, 1993).
4.4 Lattice Boltzmann Models (LBm)

Up to this point, the basic theory and principles of the LBM have been introduced.
It approximates a fluid by a discrete distribution of particles and simulates fluid flow
by two simple steps: streaming and collision following discrete rules. Compared to
Chen et al. (1994), the LBM acts as a mesoscopic fluid flow approach.
Finally, the two most important Lattice Boltzmann Models (LBm) will be intro-
duced:
the two-dimensional, nine velocity model (D2Q9), and
the three-dimensional, nineteen velocity model (D3Q19).
The 2D model will be shown since this has been the starting point for all further
LBm. The 3D model forms the basic setup that has been used for fluid flow modeling
in this thesis.
4.4.1 The 2D Nine Velocity LBm (D2Q9)

One of the most important applications of the Lattice Boltzmann Method is the
simple two-dimensional fluid flow model. Figure 4.3 shows the Cartesian lattice and
the velocities vx , respectively. The index x = (0, 1, . . . , 8) indicates the direction,
whereas v0 denotes resting particles.
Figure 4.3: The 2D Lattice Boltzmann model, showing the Cartesian lattice
and velocities. Modified after Sukop and Thorne Jr. (2010).
Lattice Boltzmann Models (LBm) 39
As it contains two dimensions and nine velocities, this scheme is known as the D2Q9
model. The nomenclature for the Lattice Boltzmann models was introduced first by
Qian et al. (1992) and has been commonly used since then. The first part of this
abbreviation always denotes the model dimension (2D or 3D), whereas the second
part describes the number of velocities.
The most fundamental units used for LBM are the so called lattice units (lu, as seen
in figure 4.3) and the time steps (ts). The assignment of velocities to the lattice
directions is straightforward. The velocity magnitude v1 - v4 of the main directions
(1 - 4) is one lu/ts, and the velocity magnitude v5 - v8 of the diagonal directions

(5 - 8) is 2 lu/ts. Related to Sukop and Thorne Jr. (2010), these velocities are
exceptionally convinient because all of their x- and y-components are either zero or
one (figure 4.4).
Figure 4.4: Schematic view of the D2Q9 x- and y-velocity components. Mod-
ified after Sukop and Thorne Jr. (2010).
The macroscopic velocity v then forms to an average of the lattice velocities vx ,

weighted by directional densities fx :
8 8
1X X
v= fx vx with = fx , (4.23)

x=0 x=0
as the macroscopic fluid density. With this, fluid motion can be described as shown
within the previous sections. An extensive derivation and more implementations for
this model can be found by Succi (2001) and by Sukop and Thorne Jr. (2010).
4.4.2 The 3D Nineteen Velocity LBm (D3Q19)

The three-dimensional LBm can be derived equivalently to the 2D model, although
the lattice structure is more complex. Three different spatial model types exist,
taking fifteen (D3Q15), nineteen (D3Q19) and twenty-seven (D3Q27) velocity di-
rections into account. The first type tends to be numerical unstable (Benzi et al.,
1992), whereas the last type presumes a lot more computational power (Sukop and
Thorne Jr., 2010). The best compromise and most common 3D lattice Boltzmann
model is formed by the second, the so called nineteen velocity model, which has been
used to perform fluid flow modeling for this thesis (Chen et al., 1994; Succi, 2001).
Figure 4.5 gives an overview of this special lattice structure. Similar to the 2D
model, the simplest starting point is formed by the zero velocity vector. Next, the
six nearest neighbors (v1 to v6 ) are added to the model (figure 4.5, step 1). Finally,
the twelve second nearest neighbors (v7 to v19 ) are appended (figure 4.5, step 2) so
that the complete nineteen velocity lattice model is derived (figure 4.5, step 3). The
implementation of the third nearest neighbors along the far diagonals of the lattice
would complete the D3Q27 Lattice Boltzmann model.
Lattice Boltzmann Models (LBm) 41
Figure 4.5: Schematic workflow of velocity vector implementation to derive

the D3Q19 Lattice Boltzmann model.
4.5 Setting up a Simulation with Palabos

4.5.1 About Palabos
For the Lattice Boltzmann modeling, a toolbox named Palabos has been used. The
Palabos toolbox is an open source library for general-purpose CFD, written in C++
(Latt et al., 2012). It is freely available under the terms of the open source general
public license AGPLv3 (http://www.gnu.de). Due to its straight forward library
structure, the source code could be easily adapted for a large variety of CFD re-
lated problems and challenges, e.g. as the simulation of thermal, or multiphase flow
through a porous media at pore-scale accuracy (Latt et al., 2012).
The code covers a wide range of functionalitiy of physics (e.g. Navier-Stokes equa-
tions, multiphase flows), of basic fluid models as the BGK-model (compare chapter
4.3), of straight-wall boundary conditions (e.g. Dirichlet or Neumann conditions for
the velocity or the pressure), of off-lattice boundary conditions as the voxelization
of highly detailed .stl files, and many other features. The implemented models are
D2Q9, D3Q13, D3Q15, D3Q19, and D3Q27, respectively.
The source code supports massively parallel CFD modeling with MPI for shared-
memory and distributed memory platforms. The code is fully benchmarked and
extensively documented (Latt et al., 2012).
The complete documentation and many other references may be found directly at
the Palabos homepage:
Main page, including Palabos users forum: www.palabos.org
Introduction: www.palabos.org/documentation/userguide/introduction.html
Users Guide: www.palabos.org/documentation/userguide/
Benchmarks: www.palabos.org/academia/usecases
Download: www.palabos.org/download-ql
Setting up a Simulation with Palabos 43
4.5.2 Workflow to set up a Simulation with Palabos

This section illustrates the general workflow to compute permeability and related
parameters (i.e. pressure, vorticity, fluid velocity) of a porous natural rock of arbi-
trary complexity. This workflow may also be found as a brief tutorial at the Palabos
homepage (www.palabos.org/tutorial). The setup of the simulation process consists
of four major steps as listed below and as summarized in figure 4.11:
1. Generation of the desired geometry.

In a first step, the geometry of the porous media structure needs to be seg-
mented and then extracted as described in detail in chapter 8. The so derived
3D model has to be saved as a so called .stl file, which contains the surface
information of the pore network structure in binary ASCII format. This is
neccessary, since the complex pore networks of natural rocks may result in
grids with several million cells, which therefore result in large data files (up
to gigabyte size). For this workflow tutorial, a small exemplary pore network
geometry of a natural sandstone has been used (figure 4.6).
Figure 4.6: Exemplary pore network geometry for the Lattice Boltzmann sim-
ulation with Palabos.
2. Rescaling of the geometry.

The generated geometry needs to be rescaled since the basic length unit used
in Palabos is [m]. The rescaling process can be achieved by using the Palabos
library scaleSTL, which can be found in the utility folder of the installed
software distribution. The scaleSTL module supports parallel computation for
the rescaling of the geometries, which is highly recommended for large (REV)
geometry files. The special syntax for parallelized rescaling is shown in figure
4.7. Finally, the rescaled model file needs to be copied into the correspondent
fluid flow simulation folder.
Figure 4.7: Syntax to start the grid rescaling with Palabos. The first part (light
gray box) initiates the parallel computing, the second part (light yellow box)
starts the rescaling process. Next, the input.stl file is defined (light blue box)
and the fit direction is set (0 = x-direction, 1 = y-direction, 2 = z-direction,
dark green box). After defining the correct fit length (light green box), the
output file needs to be addressed (light red box).
3. Simulating fluid flow through this geometry.

After rescaling of the geometry, and after copying the final grid file into the
designated program folder, the fluid flow simulation is ready to start. First,
the code needs to be compiled by using the correspondent makefile, which is
located within the simulation directory. Then the simulation can be started
as exemplarily shown by the special syntax in figure 4.8.
After setting up the relevant parameters for parallel computing (light gray
box), the simulation can be started (light yellow box). For this, the size of
the computational domain (light blue box), the inlet fluid velocity (blue box),
the fluid viscosity (light green box) and the input grid file (light red box) need
to be defined. The final input (red box) controls the option, whether the re-
sults are written in a special time step visualization file or not (1 = yes, 0 = no).
Figure 4.8: Syntax to start the Lattice Boltzmann simulation with Palabos.
The complete source code of the main program is listed in Appendix C, there-
fore only some basic funcionalities of the code are briefly introduced in the
following:
Line 1 - 45:
In the first lines, general definitions and settings are made, e.g. the D3Q19
Lattice Descriptor (line 14), the maximum number of iterations (line 26), and
the iteration step at which a 2D slice image of the simulation results has to be
written into a temporary directory of the program (line 27).
Line 46 - 84:
In this section, the rescaled grid file is loaded and transformed into lattice
coordinates.
Line 85 - 102:
Afterwards the lattice transformation, the computational domain is extended
in fluid flow direction (here: z-direction), to create an inflow and outflow buffer
zone (line 86). The creation of such a buffer zone guarantees that the particles
are able to uniformly enter the original grid domain, and therefore ensuring
numerical stability. In this case, inflow and outflow zone are set to 1/3 of the
original domain length in flow direction (figure 4.9, left hand side).
Line 103 - 255:

These lines contain the main part of the program code. Here, the boundary
conditions are set (line 103 - 114), and several output files (.gif, .vti and .vtk),
containing information on fluid velocity, vorticity and pressure gradient (line
133 - 194) are written into the temporary program directory (figure 4.9, right
hand side).
Figure 4.9: Examples of the output files derived with Palabos. From left to
right: geometry including inflow and outflow zone, velocity and vorticity field
after 4,000 iterations.
4. Post data processing and visualization.

Finally, the results of the simulation can be processed and visualized. For this
purpose, a special open source software named Paraview has been used (for
details please check: www.paraview.org). With this toolbox, the rescaled grid
can be directly combined with the 2D slices, which have been calculated by
the LBM code of Palabos (figure 4.10).
Figure 4.10: Exemplary visualization of the pressure drop of the tutorial sam-
ple geometry with Paraview after 4,000 iterations.
Figure 4.11: Workflow to set up a Lattice Boltzmann simulation with Palabos.

Chapter 5
Rock Sample Characterization
It is legitimate to assume that for a majority of geoscientific CFD applications mostly

porous and permeable rocks are investigated. As already pointed out in chapter 1.3
and 1.4, this fact has two main reasons: First, because most of this research is re-
lated to the investigation of classical porous hydrocarbon reservoirs. Second, due
to limitations in research time, 3D imaging and computational hardware. CFD is
performed either by using equivalent network models or by using modified in situ
models of porous rocks, mostly so called Berea or Fontainbleau type networks (Arns
et al., 2004; Dong, 2007; Zhan et al., 2010). For this thesis, a different approach has
been choosen to demonstrate the advances and accordingly the possibilities of high
resolution micro CT imaging as well as the complex grid generation and the prefer-
ential adaptation of the LBM for numerical pore networks of arbitrary complexity.
Hence, three different types of sandstone have been selected:
1. Bentheimer sandstone (S1), lower cretaceous. This rock is comparable to the

well known Berea and Fontainbleau type sandstone.
2. Obernkirchen sandstone (S2), middle cretaceous (Wealden). In principle com-

parable to the S1 rock type, the pore network of this sandstone differs moder-
ately from S1.
3. Flechtingen sandstone (S3), middle permian (Rotliegend). This rock greatly

differs from S1 and S2 - not only by petrophysical properties but also by
complexity of the pore network structures.
In this chapter, a brief introduction of the basic geological setting and mineralogical
composition will be presented. Special emphasis will be given on a detailed, CFD
related petrophysical sample characterization.
49
50 Rock Sample Characterization
5.1 Basic Geological and Mineralogical Rock De-

scription
The pore structure may be characterized by petrophysical investigations in detail,
nevertheless its development is a matter of (geological) deposition environment and
mineralogy. Therefore it is reasonable to briefly introduce these information. In fact,
small subsample sets have been created in order to perform classical thin section
analysis and Scanning Electron Microscopy (SEM) on exemplary rock samples.
5.1.1 Bentheimer Sandstone (S1)

The Bentheimer sandstone is part of the lower Valanginium sequence, which is associ-
ated to the lower cretaceous (approximately 140 - 134 mio years old). This sandstone
is segmented into three litho-stratigraphical units related to Kemper (1968): lower
Bentheimer sandstone, Romberger intermediate and upper Bentheimer sandstone
(also known as flasered sandstone).
Kortmann (1983) has classified this sandstone petrographically as a quartz sand-

stone with few clay and silt content. Grain size is varying between very fine and
coarse, whereas fine grain content is predominating. In general, components are
well rounded and well sorted, excluding partial content of stable and unstable heavy
minerals as well as feldspars. Stadtler (1998) gives a good comprehensive and more
detailed description of the S1 type sandstone.
Thin section analysis widely confirm these results. Figure 5.1 shows some exem-
plary results. On the left hand side, quartz (gray) and feldspar (checkered gray) can
be clearly identified. Quartz grains are sometimes sharp edged and mostly coarse,
indicating a late and short ranged depository. Predominating matrix grain size has
been determined to be in range of 250 - 450 microns. Small amounts of clay minerals
are allways allocated in the vicinity of partial dissoluted feldspar minerals. These
accumulations obviously appear in two different ways. First, as (partial) coatings
around quartz grains. Second, as fillings of interparticle porosity (right hand side of
figure 5.1).
Basic Geological and Mineralogical Rock Description 51
Figure 5.1: Exemplary thin sections of the S1 specimen set.
Results from SEM have shown that the few and locally distributed clay accumula-
tions mostly consist of illite, which form small spatial areas with low but intercon-
nected porosity. Figure 5.2 gives a comprehensive overview of the S1 SEM results,
including a general overview and a closer look onto exemplary mineral phases, pore
throats and surfaces. On the upper left side, an overview on the quartz matrix is
shown. Most of the grains are of similar size and shape, which indicates isotropic
physical behaviour. On the upper right side, a dissolved feldspar mineral including
its intragranular porosity is shown. The lower part of figure 5.2 shows some clayey
agglomerations, which locally influence pore space and specific sample surface as well.
Due to these results, pore space can be characterized as well interconnected. Void
space mostly consits of large pores (up to 60 - 80 m in diameter) and only few
small pores closely related to clay mineral accumulations. Partial dissolved feldspars
may add more complex but local porosity to the total pore network. Cementation
minerals are practically not existent, which also suggests high permeability.
Figure 5.2: Exemplary SEM images of a S1 specimen.
5.1.2 Obernkirchen Sandstone (S2)

The Obernkirchen sandstone is part of the upper Berriasium sequence (also known
als Wealden sequence, approximately 145 - 150 mio years old), which is associated
to the lower cretaceous, forming the lower transition to the jurassic age. Related
to Hafner (1987) this sandstone is characterized by a homogeneous and fine grained
rock structure, sometimes interrupted by thin clayey layers. Additionally, feldspar
minerals indicate strong mechanical influences, which leads to a higher degree of
degradation than for the S1 sandstone. Pore space could be filled by different kinds
of cementation (Brner et al., 2011).
Results from thin section and SEM investigations confirm these findings only to
some extent. Figure 5.3 shows exemplarily two different thin section views: on the
left side a close up upon the matrix, which is formed by quartz, including some frag-
mented feldspar (top) as well as zirconium (red) as accessory mineral. The overview
on the right side clearly shows different grain sizes of the quartzitic matrix. Hence,
this composition results in a very heterogeneous rock structure, which is a contrary
result related to Hafner (1987). It seems legitimate to assume that this heterogeneity
influences pore structure and interconnectivity to some extent.
Additionally, SEM images indicate almost no cementation but instead complex and
randomly distributed clay aggregations (figure 5.4). These aggregations sometimes
form coatings on the quartz matrix, but more often they are allocated within pore
interspaces, probably leading to a reduction of porosity and permeability. In sum-
mary, the pore space of the S2 sandstone is stronger affected by mineralogy than
the pore space of S1. Due to the higher degree of aggrillaceous alteration, pores
are significantly smaller. The random spatial distribution of these agglomerations
makes it difficult to describe a representative sample volume (REV) for CFD pur-
poses. The increased complexity and heterogeneity of the matrix obviously reduces
the interconnectivity of the pore network. Both, increased complexity as well as
matrix heterogeneity, have a strong effect on porosity and permeability.
5.1.3 Flechtingen Sandstone (S3)

The Flechtingen sandstone is part of the so called Rotliegend sequence (approxi-
mately 302 - 257 mio years old), which is associated to the lower permian age. Related
to Brle (1996), this sandstone is generally characterized by a distinct heterogeneity
throughout the entire sequence, caused by different depository environments. Mostly,
a fine layering of small and medium size grains on a small (millimeter) scale is pre-
dominating. The red color of the rock is a result of fine deposited hematite, caused
by the arid and desert like environment of the permian age. Cementation minerals
are often of calcitic and dolomitic nature, but baryte and clayey cementations are
also very common. Clay minerals are often represented by chlorite and illite, but
mostly by kaolinite. Due to complex diagenetic phenomena, mineralogy and grading
effects, resulting pore structures strongly vary on a small scale (Brle, 1996; Paech
et al., 2006).
These results have been confirmed by thin sections and SEM images on representa-
tive S3 samples. Figure 5.5 shows two exemplary thin section images of this specimen
set. The left side of the figure highlights a heavily, barytic cemented part of the pore
space. This cementation type is locally distributed so that diagenetic consolidation
is predominating, leading to small and narrow pores with poor interconnectivity.
Additionally, quartz grains are very often coated by clay minerals, which leads to an
ongoing reduction of porosity and permeability (figure 5.5, right hand side).
SEM imaging has confirmed that the clay minerals cause a significant reduction on
intergranular porosity (figure 5.6). Partial dissoluted feldspar minerals (mostly ap-
atite) very often form isolated, so called honeycomb cavities. Prevalently, the outer
layer of these structures is ruptured so that kaolinite, as a result of feldspar degra-
dation, can migrate into the narrow pore spacings and form additional, wide spread
cementation and coatings. Hence, fluid flow properties are in general a lot more af-
fected by mineralogy and geology for the S3 specimen set than for S1 and S2. These
results lead to the conclusion that the pore network is of random complexity. Void
space adheres a high quantity of small and narrow pores, including many distinct
and small scale heterogeneities. It can be expected that petrophysical properties
vary over a distinct value range for the S3 sandstone.
5.2 Petrophysical Rock Characterization

To understand and evaluate CFD results, a detailed scientific program of standard
and special core analysis (CAL / SCAL) has been carried out. In order to derive
a good statistical sample description, a total of 50 core plugs for each of the three
different sandstones has been prepared. Table 5.1 summarizes all of the applied
methods as well as the total sample number of core plugs used for the different types
of investigation. A complete list of all derived data values is given in the appendix
of this thesis (Appendix A). In this section, petrophysical data for each investigated
rock type will be discussed separately, before a combined and comparatative rock
type evaluation will take place.
Petrophysical Rock Characterization 57
method / parameter Bentheimer S1 Obernkirchen S2 Flechtingen S3

pycnometry / , d 50 50 50
permeametry / ka , kkl 50 50 50
NMR / T2 , 40 40 11
BET / Sm , Spor 45 44 43
Hg-injection 5 5 5
Table 5.1: Overview of the applied petrophysical methods and sample quan-
tities for all three specimen sets. From top to bottom: porosity, density, ap-
parent and Klinkenberg permeability, NMR T2 spectra and porosity, specific
sample surface.
5.2.1 Bentheimer Sandstone (S1)

For petrophysical core analysis, a total of fifty S1 core plugs (30 mm in diamater, 40
mm in length) has been prepared. Later on, a selection of samples has been used to
derive smaller, so called CT miniplugs (4 mm in diameter, 5 - 10 mm in length),
for high resolution imaging and pore network extraction (figure 5.7).
Figure 5.7: Representative image of a S1 laboratory scale sample (side- and

topview) including a schematic view of the CT miniplug size and sampling
position.
Porosity and Klinkenberg gas permeability have been determined for a basic rock
sample characterization. Altogether, S1 sandstone can be classified as a typical
reservoir sandstone: highly permeable and with a significant amount of porosity. In
this case, porosity averages to approximately 25 %, permeability to approximately
600 mD. As expected from thin section and SEM analysis, specific surface area
(BET surface) is quite low (approximately 0.45 m2 /g in average), due to the small
amount of clay and dissolved minerals within. The ranges and mean values of these
investigations are summarized in table 5.2, total parameter distributions are given
by histograms in appendix A.1. Compared with data from literature (Schn, 1983,
2004), the derived values are within the documented valuations. For a more detailed
and structural characterization of the pore space, a small SCAL program has been
applied, including mercury injection and nuclear magnetic resonance spectroscopy
(NMR) experiments. It is possible to describe the pore size distribution by evaluat-
ing transversal relaxation time spectra (T2 ) and by using empirical, rock specific cut
off times as described by Dunn et al. (2002).
parameter minimum value maximum value average value

[/] 0.20 0.23 0.215
ka [mD] 548 717 621
kkl [mD] 522 687 593
Sm [m2 /g] 0.20 0.81 0.45
Table 5.2: Overview of the ranges and mean values of the basic petrophysical
parameters of the S1 specimen set.
Figure 5.8 shows an exemplary T2 relaxation time spectrum of the S1 specimen set
including average percentage of effective porosity of all NMR investigations related
to the empirical cut off values. In average, 91 % of the distribution is related to cut
off times bigger than 33 ms, which is a clear indication for free movable water and
therefore of large and interconnected pore space. The small residual of the distribu-
tion is related to capillary and clay bound water (cut off values smaller 33 ms / 3
ms), which equals small and narrow pores. This result is coherent with the findings
from conventional mineralogical analysis of this sandstone.
Additionally, the results from capillary pressure experiments confirm these obser-
vations: the pore throat radii distribution indicates a maximum at throat sizes in
range of 10 - 25 m. This range also contributes most to permeability of the in-
vestigated samples (figure 5.9). At least 60 - 80 % of ambient gas permeability is
related to this throat range, whereas the resuming radii obviously do not contribute
to fluid flow at all. The low number of pore throats smaller than 6.3 m is also a
clear indication for a minor sample clay content, which is consistent to the results of
BET surface measurements, thin sectioning and SEM imaging analysis, respectively.
It is legitimate to assume that the pore network of the S1 specimen consists of

one well connected and dominant pore space with very few, spatial and randomly
distributed secondary networks, which are related to small clay agglomerations.
Compared to experimental results, these secondary networks seem to be connected
with the main network, although they apparently have no impact on macroscopic
fluid flow properties.
Figure 5.8: Exemplary T2 relaxation time spectrum and average percentage of

effective porosity for the S1 specimen set.
Figure 5.9: Exemplary representative pore throat distribution derived from S1

mercury injection experiments.
5.2.2 Obernkirchen Sandstone (S2)

As before, a total of fifty S2 core plugs has been prepared for petrophysical analysis.
Figure 5.10 shows size and geometry of the S2 standard plugs as well as size and
sampling location of the CT miniplugs. For a basic sample characterization, a CAL
and SCAL program - analog to the S1 sample set - has been applied.

position.
As observed within thin section analysis, the S2 sandstone features a very heteroge-
neous grain size distribution. Permeability is stronger affected and therefore lower
(approximately 16 mD in average), whereas specific surface area has increased (ap-
proximately 0.66 m2 /g in average), compared to the results of the S1 set. It is
interesting that - whilst porosity has slightly changed (approximately 16 % in aver-
age) - total changes in permeability are significant. Obviously, rock structure (i.e.
clay agglomerations and grain size heterogeneities) is predominating fluid flow prop-
erties. The ranges and mean values of these investigations are summarized in table
5.3, total value distributions are given by histograms in appendix A.2.

[/] 0.155 0.18 0.164
ka [mD] 11 65 18
kkl [mD] 9 59 16
Sm [m2 /g] 0.38 0.93 0.66
An exemplary NMR T2 distribution is shown in figure 5.11. The observed struc-

tural influence is now clearly visible: the amount of free movable water is reduced
to 70 % in average, whereas the amount of capillary bound water, related to smaller
pores within cementated and narrow (clayey) pore spacings, has increased to approx-
imately 26 % in average. This points to an increasing complexity in pore structure
and may be an indication for the development of a coarse and a fine part of the pore
network.
This result is consistent with results from pore throat radii measurements of the
S2 subsample set (figure 5.12). The main throat size - summarizing to 35 % of all
throats - is located in between of 3 - 6 m. Additionally, a lot of the pore throats
are in range of 0.25 - 2.5 m (approximately 30 %). Contrary to the main throat
sizes, these pores almost do not contribute to permeability, except for the throats in
range between 1 - 2 m. This may be an additional indication for the split up of the
pore network as pointed out before.
Figure 5.11: Exemplary T2 relaxation time spectrum and average percentage

of effective porosity for the S2 specimen set.
Figure 5.12: Exemplary representative pore throat distribution derived from

mercury injection experiments of the S2 specimen set.
5.2.3 Flechtingen Sandstone (S3)

The S3 specimen set also consists of fifty core plug samples. Figure 5.13 exemplarily
shows size and geometry of the S3 standard plug as well as size and sampling position
of the CT miniplugs.

position.
As seen from basic mineralogical characterization, S3 samples definitely are more

complex than the S1 and S2 specimen. Not only complexity, but also the increased
heterogeneity can be clearly evaluated from CAL and SCAL results. Outcomes from
standard CAL show great variability within the value range therefrom. Porosity
differs from approximately 6 % to 17 %, whereas permeability scatters within three
orders of magnitude. Nevertheless, average permeability is in the same order as the
mean k-value of the S2 specimen set (approximately 17 mD in average). Specific
surface area is greatly increased (approximately 1.5 m2 /g in average), due to the
mixture of weathering effects, cementation and clay minerals. The ranges and mean
values are summarized in table 5.4, total value distributions are given by histograms
in appendix A.3. Note that - due to the great variations - permeability has been
plotted as log kkl for a better comparative overview.

[/] 0.065 0.167 0.111
ka [mD] 0.15 166 19
kkl [mD] 0.10 154 17
log kkl [mD] -1.0 2.19 0.49
Sm [m2 /g] 0.94 2.19 0.49
Hence, results from NMR experiments are more complex to describe than results
from the other types of sandstone. Due to the high variability in porosity, T2 spectra
are also heavily influenced. Because of this, a spectra of a low (red curve) and
of a high (orange curve) sample are exemplary shown in figure 5.14. S3 samples
indicate a high amount of small pores (clay bound water, approximately 19 % in
average) as well as a distinct amount of medium range pores (capillary bound water,
approximately 29 % in average). The residual is related to free movable water, i.e.
larger and interconnected pores. This separation may point to a tripartite of the pore
network structures: a good connected main network, a smaller, partially separated
network and a fine network directly related to the clay agglomerations. Above all,
results from SEM deliver significant indications for this assumption.
On the other hand, findings from mercury injection experiments do not give any
indications for a separation of the pore network (figure 5.15). As exemplarily shown,
the number of small and very small pores (< 1.0 m / < 0.004 m) has greatly
increased - in average, they accumulate to (at least) one third of all measured pore
throats. Nevertheless, these throats obviously do not participate to permeability at
all. One fifth of the radii, ranging between 1 - 4 m, seem to form a small network
with a small but measurable contribution to total permeability (at least 5 % in aver-
age). Most of the fluid flow is associated to pore throats bigger than 4 m, therefore
it seems legitimate to assume that these pores form the main network.
Figure 5.14: Exemplary T2 relaxation time spectra and average percentage of

effective porosity for the S3 specimen set (red: low , orange: high sample).
Figure 5.15: Exemplary representative pore throat distribution derived from

mercury injection experiments of the S3 specimen set.
5.3 Combined Rock Evaluation

Finally, stand alone results of the mineralogical / geological description as well as
petrophysical rock characterizations are combined to form a conclusive and mean-
ingful overview for all three specimen sets. As seen, S1 and S2 sandstones are simple
and similar composed or structured in terms of mineralogy and internal pore space
configuration, respectively. Solely, the S2 specimens feature a higher degree of clayey
cementation and structural heterogeneity so that the resulting pore network appears
to be more complex than the network of S1. Contrary to this, the S3 sandstone
has a very complex and versatile mineralogical composition and internal structure
therefrom. The Flechtingen sandstone is heavily cemented and compacted and - ad-
ditionally - exhibits complex weathering as well as clayey structures, which greatly
increase the complexity of the pore network.
From a petrophysical point of view, S1 and S2 specimen behave similar, although

S2 samples contain a distinct amount of clay compared to S1 samples. For both
sandstones, porosity and permeability show a good correlation, which is a general
indication of rock sample homogeneity. As investigated, S1 sandstone features al-
most no significant amount of clay or cementation minerals, which leads to a very
pronounced -kkl relationship (figure 5.16). The Obernkirchen sandstone (S2) shows
some slight value scattering within this relationship as a result of the matrix het-
erogeneity and the preponderant clayey pore fillings. Contrary to S1, the permian
sandstone (S3) features a significant scattering in data values. As pointed out, this is
a result of the complex structured sample, including weathering, cementation effects
and a distinct clay mineral content, respectively.
A similar behavior can be observed for the relationship of permeability and BET
surface (figure 5.17; note: measured Sm (mass specific surface) has been transferred
to Spor (porosity specific surface) due to scaling issues). S1 and S2 sandstones show
narrow data value ranges, whereas data of the Flechtingen sandstone (S3) scatter
significantly. Again, S1 sandstone exhibits the clearest value relationship, whilst
Obernkirchen sandstone data slightly disperse. Related to the results from thin sec-
tioning and SEM, the increased BET surface values of the S2 sandstone are a result
of the significant grain size variations and of the increased clay content, respectively.
The distinct scattering of the S3 data are again a result of weathering, dissolution
of mineral phases and the extensive amount of clay minerals within.
Combined Rock Evaluation 67
Figure 5.16: Porosity - Permeability ( - kkl ) relationship for all specimen sets.
Figure 5.17: Permeability - BET surface (kkl - Spor ) relationship for all speci-
men sets.
Additionally, effects taking place on pore space can be observed by NMR (figure
5.18) and by mercury injection experiments (figures 5.9, 5.12, 5.15). Related to
the described empirical T2 cut off values for pore space characterization of (porous)
sandstones, the decrease in the number of large pores is directly linked to increasing
sample complexity. The T2 spectra exhibit good indications for the development
of subordinated pore networks within the entire pore space. As clearly seen from
the graphic, medium as well as small sized pores increase with ongoing cementation,
weathering and accumulation of clay. Additional indications for this network separa-
tion can be found by capillary pressure experiments, even though results therefrom
are not that explicit (especially for the S3 specimen set).
Figure 5.18: Average percentages of porosity related to NMR T2 spectra cut

off values for all specimen sets.
Combined Rock Evaluation 69
Related to the CFD challenge, rock sample characterization can be concluded as

follows:
The S1 sandstone is a homogeneous rock without any cementation or relevant

amount of clay minerals. Petrophysical parameters are not affected at all
by mineralogy and data values are within known ranges for this sandstone.
Pore space can be expected to consist of a homogeneous network with good
interconnectivity. This should also result in a quite small sample volume for
representative modeling of fluid flow at the pore scale.
The S2 sandstone still can be characterized as a homogeneous rock related to

the results of petrophysical investigations. Nevertheless, sample matrix and
therefore pore space are more complex than for the S1 rock type. Grain size
heterogeneities and especially the distinct amount of clay minerals lead to
smaller pores in general and to an increased amount of medium size cavities in
detail. Although pore network shows good interconnectivity, determination of
a representative modeling volume may be more difficult due to the allusively
development of a subordinated pore network.
S3 sandstone holds a complex mixture of different effects: compaction, weath-

ering and partial mineral phase dissolution, cementation and different types
of clay mineral accumulation. These effects have significant impact on petro-
physical properties, leading to distinct data value scattering, i.e. to a very het-
erogeneous rock behavior. Precisely, S3 sandstone samples range from dens
and tight to porous and permeable including a wide intermediate valuation
therefrom. Related to the CFD question, evaluation of the entire sandstone is
very difficult, since each sample holds an unique mixture of effects and param-
eter values. Hence, determination of a representative sample volume or pore
network of this rock type does not seem to be applicable or even reasonable.
Chapter 6
The REV Challenge - a Classical

and a New Approach
Besides the special demands on computational power, the adaptation of different

numerical codes and techniques, one main attempt of CFD is the predictive and
therefore representative modeling of petrophysical parameters in general, and of
hydraulic properties in particular. Although modeling of fluid flow through small
sample networks may increase the knowledge of fluid dynamics on a microscopic
scale, representative modeling of physics remains the most important task. In this
chapter, a short and basic introduction to the challenge of estimating a so called
Representative Elementary Volume (REV) will be given. Additionally, the assess-
ment of the classical REV estimation as well as a new and advanced method for the
REV chalenge will be demonstrated.
6.1 Classical REV Approach

The fundamental theory of the challenging REV estimation has been introduced by
Bear (1988). This approach has been developed and consequently enhanced over
the past years. The probably best publication related to this topic was written by
Bear and Bachmat (1990). Since the extensive theory can be reviewed there, only
the basic idea will be introduced. As a starting point, a very simple consideration
needs to be set up: which typical size or - better - which typical parameter may
characterize a porous medium (e.g. the three types of sandstone used for this thesis)
most completely? The answer is somehow trivial: porosity.
71
72 The REV Challenge - a Classical and a New Approach
As a geometrical, or structural porous medium property, it is a good parameter to

determine a REV, since for consolidated (porous) rocks a distinct correlation be-
tween porosity and other, CFD related properties, does exist (Schn, 2004; Tiab and
Donaldson, 2004). For unconsolidated rocks, a specific REV for each of them (e.g.
permeability) needs to be considered seperately (Bear, 1988).
Figure 6.1 reflects the main idea of this approach. If an infinitesimal small and
cubic elementary volume is considered, which may be randomly distributed within
the porous medium domain, porosity may vary between zero and one, because the
volume is either related to rock matrix (grain material) or to pore space.
Figure 6.1: The basic REV principle, shown by the variation of porosity (n)
within the neighborhood of a point P (x0 ) as a function of the averaging volume
(U). Modified after Bear and Bachmat (1990).
If the volume is increased step by step, porosity alternates between a maximum and
minimum value until they converge to each other. At a distinct point, the domain
of (microscopic) inhomogeneity is discontinued (figure 6.1, left hand side) and the
Classical REV Approach 73
porous media domain is entered (figure 6.1, middle section). In this range, property
variations are minimal and almost converge to the in-situ porosity value. Hence,
the material may be described as a homogeneous medium. In fact, it is possible
that heterogeneity may increase again, related to macroscopic volume features (e.g.
fractures) so that the function diverges again (figure 6.1, right hand side).
It should be mentioned that the minimum volume size Umin for a representative
elementary volume strictly depends on the specified domain accuracy (Bear and
Bachmat, 1990). For this thesis, all REV considerations are directly related to the
achieved voxel resolution of the CT scans (i.e. 1.75 m).
Practically, the classical REV estimation for the three different sandstones that have
been used for this thesis, has been applied as follows:
The high resolution 3D data set of each CT mini-plug has been used to derive
a maximized cubic sample volume. In case of the S1 - S3 specimen, it has been
possible to achieve a maximum edge length of 1750 voxel (= 3.0625 mm).
Next, randomly distributed and cubic subvolumes have been extracted and
analyzed for their effective porosity. The edge length of these subvolumes has
been stepwise increased, starting from 50 voxel and ending up at the maximum
edge size of 1750 voxel. Quantitative investigations have been performed with
AVIZO Fire, a special software distributed by the Visualization Science Group
(VSG, www.vsg3D.com).
Afterwards, the so derived porosity values are plotted against their correspond-
ing subvolume edge length, as shown within figure 6.1.
This procedure has been applied and repeated on each of the three different types
of sandstone. The edge length cut-off value for the REV determination has been
set individually for each sample volume. As reference, the effective laboratory scale
porosity value, derived by helium pycnometry, has been used.
6.1.1 Classical REV Analysis for S1

Figure 6.2 illustrates the results of the classic REV estimation for the S1 sandstone.
In principle, the behavior is very similar compared to the idealized case as described
in the previous section (6.1). For small volumes, i.e. volumes with an edge length
100 microns, porosity values range between zero (matrix only) and one (void space
only). Nevertheless, the -value distribution within these small reconstructed areas
is not homogeneous. In fact, mean porosity is very close to the laboratory reference
value. With increasing edge length, porosity decreases rapidly from the upper value
boundary, whereas the lower limit remains almost constant (= zero). At an edge
length of 200 voxel ( 350 m) the lower porosity limit increases slightly. In case
of the S1 sandstone, this threshold value can be correlated to the average grain size
of the investigated sample. As soon as the edge of the REV exceeds this length, it
does not fit completely into the grain / matrix material anymore. In point of fact,
this observation is coherent with the average grain size measured / derived by thin
section and SEM analysis (250 - 450 microns, compare chapter 5.1.1).
As a result, it seems legitimate to assume that this method can be used to generally
approximate the average grain size of porous and homogeneous sandstones within
an acceptable range of tolerance. With ongoing enlargement of the REV, porosity
scattering decreases continuously as predicted by Bear and Bachmat (1990), though
the rate of change of porosity also decreases markedly. Remarkably, mean and me-
dian values of porosity, starting at a REV edge length of 500 voxel ( 850 microns),
remain almost constant. This effect can be directly related to the homogeneity of
the Bentheimer sandstone. At edge lengths larger than 1000 voxel ( 1.75 mm),
changes become minimal. Therefore it seems legitimate to characterize this zone as
the predicted porous media domain for this rock type, depending on the specified
accuracy (voxel resolution) of this type of analysis.
Furthermore, it can be seen that porosity tends to be underestimated, although

scanning resolution is high. This is a side effect of the image, i.e. gray value based
phase segmentation, and should be taken into account, when results of transport
modeling are discussed. Based upon these results, the cut off value for the S1 REV
has been set to 1250 voxel with an image resolution of 1.75 microns.
Classical REV Approach
Figure 6.2: Estimation of the S1 REV by using the classical approach from Bear (1988).
75

Remarkably, the behavior is very similar to the idealized case as described by Bear
and Bachmat (1990), and also compared to the case of the S1 rock. Even though
the S2 sandstone is characterized by an increased complexity and heterogeneity in
structure and mineralogy as well (compare 5.1.2).
For small volumes ( 50 voxel edge length), porosity ranges between zero (matrix
only) and almost one (void space only). Due to the smaller average grain size and
the smaller average pore size, -value decrease is significantly higher than for S1. At
an edge length of 150 voxel ( 263 m), the lower porosity limit rises continuously.
Again, this observation shows a good correlation to the average grain size, which has
been derived by thin section and SEM analysis (250 - 350 microns, compare chapter
5.1.2), though expectations have been contrary to this result. Obviously, mineralog-
ical features (e.g. clay agglomerations) are less distinctive within a 3D volume than
observed within the 2D thin sections.
Porosity scattering decreases continuously with ongoing REV edge length growth.
As observed for the S1 sandstone, mean and median values remain almost constant
over a large edge size valuation. At a REV size of 10003 voxel, changes in porosity
become minimal. Again, this parameter tends to be underestimated compared to the
lab derived core plug reference value. Furthermore, this effect is more pronounced
than for the S1 type. Besides the difficulties of an image based segmentation, scan-
ning resolution must be taken into account, due to the significantly higher degree of
smaller, and therefore more complex structures (e.g. interparticle porosity).
A higher scanning resolution may lead to slightly better results since more and
smaller porosity features could be detected, but this can be only done at cost of
the maximal achievable REV edge length (compare chapter 2.1.4). Hence, it seems
feasible and applicable to keep the adjusted scanning resolution of 1.75 microns and
set the S2 REV to 12503 voxel, which is of similar size as for the S1 sandstone.
Figure 6.3: Estimation of the S2 REV by using the classical approach of Bear (1988).
77

Due to the very high complexity and heterogeneity of the rock structure and miner-
alogy, a conclusive REV estimation is expected to be very difficult compared to the
idealized case of Bear and Bachmat (1990).
As seen from S1 and S2, porosity also ranges between zero (matrix only) and one
(void space only) for small volumes ( 50 voxel edge size). By increasing the inves-
tigated volume edge size stepwise, drops very quickly from the upper boundary
to significantly lower values (figure 6.4). Additionally, a clear shift to lower average
porosity values per extracted volume can be detected. Compared to the extensive
amount of cementation and clay minerals within S3, this result seems to be conclu-
sive. Furthermore, the lower porosity boundary stays almost constant (0 - 1 %) until
a volume edge size of nearly 600 voxel ( 1.05 mm) is reached. Average grain size
has been determined to be in range of 250 - 350 m (compare chapter 5.1.3).
This also seems to be conclusive if the strong structural sample inhomogeneity is

considered. As seen from thin sectioning and SEM analysis, and also clearly visible
in figure 6.4, large areas of the S3 pore space are well cemented (right hand side of
slice image), whereas other regions are affected by microfractures (mid section of slice
image) and others by clayey agglomerations (left hand side of slice image). By this
means, the different types of cementation as well as two different types of porosity
(fractures and intergranular) make sure that an average grain size estimation does
not seem to be applicable. The lower -value boundary increases slightly, whereas
the upper value boundary indicates a similar behavior compared to S1 and S2.
In total, porosity is stronger underestimated than for the S2 type. Although the
effects responsible are the same as for the previous rock types (scanning resolution
and segmentation impreciseness), S3 inhomogeneity has a significantly higher impact
on these effects. Due to this it seems inappropriate to classify a REV for the entire
Flechtingen specimen set. Instead, it is reasonable to recommend an individual REV
calibration for such heterogeneous porous rocks. In case of S3 and as a compromise
between these effects, scanning resolution remained unaffected and the REV cut off
value has been set to 1000 voxel.
Figure 6.4: Estimation of the S3 REV by using the classical approach of Bear (1988).
79
6.2 Advanced REV Approach
As pointed out within the previous section, the approach of Bear (1988) seems to be
a feasible method to characterize as well as to estimate a representative elementary
sample volume for CFD purposes. Even in case of the distinct heterogeneous S3 CT
sample, porosity - as a function of REV subsample edge length - converges in a very
similar manner, compared to the the S1 and S2 type sandstone. This result is indeed
unexpected, especially if the significantly different types of pore networks of S1, S2
and S3 are taken into account. Therefore, it seems reasonable to take a closer look
on the spatial porosity distributions.
For this, porosity has been calculated slice by slice (slice thickness = 1 voxel =
1.75 m) for a variety of REV edge lengths of each sample in x-, y- and z-direction.
By doing this, it is possible to investigate porosity variations on a small scale and -
furthermore - to detect the REV size at which these variations become minimal. In
case of an ideal, i.e. isotropic rock, variations in each direction and despite of the
sample volume should be constant. Although specific rock types indicate isotropic
behavior on a lab scale, it seems obvious that this should not be mandatory assumed
on a microscopic scale. As seen in case of the S1, S2 and S3 CT samples, structural
features (like cracks, texture effects, etc.) do have some impact on the directional
porosity distribution within the observed rock volumes. In theory, inhomogeneity of
these variations should become smaller and smaller with increasing REV size, but
with greater sensitivity to small scale sample features.
At first, porosity variations have been calculated for each sample in x-, y- and z-
direction by using the segmented CT data in combination with the getslicedata
quantification module of the CT data analysis software AVIZO Fire (VSG, 2011).
Afterwards, a special statistical procedure is applied, the so called variogram analysis
(Cressie, 1993; Ernste, 2011). This type of analysis characterizes a spatial data set as
a function, which describes the degree of spatial dependence of a spatial random field
or statistic process. It is defined as the variance of the difference between field values
at two locations across realizations of the investigated field. In practice, porosity as
a function of REV size should indicate a saturation curve behavior, if representa-
tiveness in each direction is reached. Cressie (1993) gives a good and comprehensive
overview about this topic.
Advanced REV Approach 81
6.2.1 Advanced REV Analysis for S1

Figure 6.5 illustrates the calculated directional porosity logs for the S1 CT sample.
Although mean as well as median porosity remain almost constant in x-, y- and
z-direction, it can be clearly seen that porosity itself scatters remarkably along each
axis. The average variation is as high as 7 %, the biggest value can be observed in
x-direction (approximately 8.6 %). Variations in y- and z-direction are almost equal,
uniformly distributed and even smaller (approximately 6 %).
Especially for the yz-face (in x-direction) inflow and outflow porosity values greatly
vary compared to the general distribution along this axis. Such a minimum in poros-
ity can also be observed within the xz-face (in y-direction), although this feature lies
within the propagation direction (around slice 375). In x- and z-direction, a max-
imum occurs at this location. This clearly points to a structurally caused porosity
feature, resulting in elongated pore shapes within the yz-/xy-face. Due to the large
distance to the next comparable minimum on this curve (located at slice 1375), and
due to the results from chapter 5, it seems legitimate to assume that the observed
and randomly distributed clayey agglomerations cause this effect.
This is an important result for permeability tensor calculation and accordingly for
the estimation and quantification of permeability anisotropy. Although mean and
median -value - due to their integrative nature - confirm the REV size as it has
been estimated within section 6.1.1, the representative volume for directional inves-
tigations should be increased. This increase should be stopped, until the standard
deviation in x-direction converges to the y- and z-values.
The variogram analysis clearly points this result out (figure 6.6). Whereas the find-
ings of y- and especially z-direction indicate an asymptotic behavior, and therefore
a representative elementary volume edge size, the variogram of in x-direction still
has not reached the asymptotic value region. This means that structural features of
the pore space - possibly related to spatial randomly distributed clay agglomerations
- necessitate to extend the predicted REV size in this direction.
The REV Challenge - a Classical and a New Approach
Figure 6.5: Directional porosity pseudo logs of the S1 CT sample.

82
Advanced REV Approach
Figure 6.6: Variogram analysis for the directional porosity distribution of the S1 CT sample.
83

Figure 6.7 illustrates the calculated directional porosity logs for the S2 CT sample.
As for the S1 specimen, mean, as well as median -value remain almost constant for
each direction. Scattering of porosity along all three axes is less distinct than it is
for the previous sample. Again, variation in x-direction indicates the highest value
(approximately 6 %), whereas the variations in y- and z-direction are almost equal
and uniformly distributed (approximately 4 %).
Contrary to S1, porosity logs of the S2 sample are less noisy, i.e. the S2 sample does
not indicate any large features within the observed directional porosity distribution.
Nevertheless, for the x-direction two maxima (around slice 800 and slice 1250) and
one minimum (around slice 1125) do exist, which cannot be directly correlated to
any obvious porosity feature in y- and z-direction, respectively. This finding points
to a strictly local and spatially limited feature within the zy-face. Elongated pores
should have some effect to the other faces, as it has been observed for the S1 CT
sample before.
Due to these findings it may be legitimately assumed that the S2 type sandstone
seems to be a lot more isotropic than the S1 type specimen. This is also consis-
tent with results of the classical REV analysis as performed in section 6.1.2. The
estimated size of the representative elementary volume seems to be confirmed and
applicable for fluid flow modeling.
The variogram analysis additionally confirms this result, as visualized in figure 6.8.
All three variograms reach asymptotic behavior, including an identical correlation
length of 60 ( 120 voxel 210 m) for each direction, although variance in x-
direction is slightly higher than for y- and z-direction. Finally, it can be concluded
that the estimated S2 CT sample REV is feasible for fluid flow modeling purposes.

85
86
Advanced REV Approach 87

Figure 6.9 illustrates the calculated directional porosity logs of the S3 CT sample. In
case of this specimen, mean as well as median porosity values are almost equal for the
yz- and xz-face, whereas data of the xy-face clearly tend to lower values. However, the
range of porosity variation, which is around 4.6 %, indicates no significant differences
by a direct comparison of each directional log. The overall scattering is less distinct
than it has been observed for the S1 and S2 CT sample (compare figures 6.5 and 6.7).
Nevertheless, an indication of a distinct shift to higher porosities exists for x- and

y-direction (around slice 600). This observation is a lot more apparent for the xy-face
than for the other two faces. The standard deviation for the porosity distribution in
z-direction also features the highest value ( 1.4 %, compared to 1 % for the x-
and y-direction). Based upon the results of chapter 5, this finding can be directly
linked to a microfracture, which propagates in z-direction through the lower section
of the S3 specimen.
Compared to the S1 and S2 sandstone specimens, the S3 sample is characterized

through a high degree of (structural) complexity and therefore obviously obeys a
distinct anisotropy of fluid flow properties. In particular, the observed fracture in-
fluences the porosity distribution significantly and additionally complicates the esti-
mation of a meaningful representative elementary volume (compare section 5.2.3).
The calculated variograms clearly reflect these results (figure 6.10). For none of
the three directions an asymptotic diagram shape is derived. Especially the vari-
ogram in z-direction is obviously of linear shape, and therefore far away from any
representativeness. Nevertheless, the variogram analysis additionally and clearly in-
dicates the propagation direction of the microfracture (perpendicular to the xy-face).
For representative modeling purposes, this sample volume does not seem to be of an
appropriate size. The presence of microfractures complicates the REV analysis and
estimation for the S3 sandstone significantly.

88
89
Chapter 7
Quantification of Sample
Inhomogeneity
As described within the previous chapter, the advanced REV analysis approach
provides a significantly better understanding of sample homogeneity (inhomogene-
ity), isotropy (anisotropy) and fluid flow representativeness, respectively. It is even
possible to distinguish between primary and secondary porosity, i.e. between the
intergranular porous network and microfractures.
Due to the observed directional scattering of porosity, standard deviation seems

to be a sensitive numerical value to investigate scale, i.e. REV edge size, dependent
structural effects and transitions. For this, porosity standard deviation (S ) has
been calculated for a large set of subvolume edge lengths (250 to 1500 voxel) for
each of the CT specimen and for each spatial direction (x-, y- and z-axis). Theo-
retically, S should greatly decrease with increasing sample volume, until the lower
(constant) data limit, which solely depends on the scanning resolution, is reached.
The gap between each directional deviation value can be used to quantify the de-
gree of inhomogeneity and anisotropy of the pore network structure. For a better
quantification of the inhomogeneity, it seems reasonable to introduce a new statisti-
cal parameter: the so called inhomogeneity factor I (Ernste, 2011). This factor is
defined as follows:
S,max S,min 4S
I = = . (7.1)
S,min S,min
91
92 Quantification of Sample Inhomogeneity
S,max equals the maximum, and S,min the minimum observed directional standard
deviation, respectively. For similar directional deviation values S , 4S and there-
fore I tend to zero, indicating a homogeneous porous medium. The larger 4S , the
larger the structural and directional inhomogeneity of the sample. In theory, each
time the pore network is disturbed, either by a natural or an induced structural
feature, I should increase and then decrease again until the next sample volume
dependent feature is added.
Due to the results so far, the following observations may be previously assumed:
S of the S1 and S2 specimen should indicate a similar behavior.
Related to the variogram analysis, I of S2 should be smaller than for S1.
S3 specimen should indicate the largest scattering of S and therefore a high

degree of inhomogeneity.
S1 CT Miniplug Inhomogeneity Analysis 93
7.1 S1 CT Miniplug Inhomogeneity Analysis

Figure 7.1 illustrates the results of the advanced inhomogeneity analysis, which has
been performed for the Bentheimer (S1) CT miniplug. As expected, directional
porosity standard deviation decreases continuously, starting at 7.5 % for an edge
size of 250 voxel and ending up around 1.5 % for an edge size of 1300 to 1500 voxel
for the analysed volumina. Each deviation curve obeys a similar shape (continuous
decay), and it can be observed that scattering is less distinct for several specific edge
lengths than it is for others.
This observation is also clearly visible for the calculated inhomogeneity factor I .
For each of this specific edge sizes, I indicates a distinct minimum. The first one
is located at an edge size of 300 voxel (I300 = 0.275). This correlates very well to
the average grain size of the S1 specimen, as it has been determined as described
within chapter 5.1.1. This finding is also consistent with the result of the classical
REV analysis (chapter 6.1.1), where the increase of the lower porosity value has been
directly linked to the estimated average grain size of 250 to 300 voxel (approximately
400 to 500 m).
The second minimum (I750 = 0.375) is alocated at a subvolume edge length of 750
voxel (approximately 1300 m). So far, no specific sample feature has been linked
to the inhomogeneity increase following this edge size. A closer look at the results
of the scanning electron microscopy yields a possible and plausible correlation. This
distance equals the mean space between partially dissoluted feldspar minerals within
the S1 CT specimen. If this mineralogical feature is added, porosity inhomogeneity
is substantially increased before the overall scattering and inhomogeneity decreases
again.
Conclusively, the third minimum, which is located at an subvolume edge length

of 1325 voxel (I1325 = 0.125), can be related to the mean distance between the
arbitrary clay agglomerations within this sample (compare 5.1.1). Because the clay
agglomerations are very unfrequently distributed, and because of the fact that the
overall impact of clay pores on the inhomogeneity of the porous network is distintive,
the increase of inhomogeneity behind this cutoff value is significantly higher (by a
factor of 2.6), than it has been observed for the feldspar cutoff value (factor of 1.3).
Figure 7.1: Directional S and resulting I for a variety of subvolume edge

sizes of the S1 CT specimen.

Figure 7.2 illustrates the results of the advanced inhomogeneity analysis for the
Obernkirchen (S2) CT miniplug. As predicted - and similar to the results that have
been derived for the S1 sample - directional S decreases continuously. Starting from
Smax 4.4 % at an edge size of 250 voxel, S decays smoothly to Smin 0.75
% for an edge size of 1200 to 1500 voxel. Overall scattering is very low compared to
the results of the S1 sample, although again some specific cutoff / correlation lengths
can be detected.
Although scattering of S is not that distinctive as it has been obvserved for S1,
calculated I is still as sensitive to S changes as anticipated. Again, several specific
minima and regions can be clearly identified. For S1, the first minimum could be
directly linked to the average grain size of the rock matrix. In case of S2, this does
not seem to be feasible.
Due to the SEM results and the classical REV analyis (compare chapter 5.1.2 and
6.1.2), average grain size has been estimated to 200 m. However, the first min-
imum (I300 = 0.1) is located at 300 voxel edge size ( 525 m), which is directly
followed by another minimum (I500 = 0.1) at 500 voxel edge length ( 875 m)
nearby. These features may be related to both, different predominant pore shapes
(elongated pores) and small scale grain size variations, respectively.
The third minimum (I875 = 0.09) is alocated at an edge size of 875 voxel ( 1500
m). The successive inhomogeneity increase correlates with the observed mean dis-
tance between the mechanically influenced (cracked and therefore porous) feldspar
minerals within this sandstone (compare chapter 5.1.2). This feature has also been
observed for the S1 type sandstone, though the feldpar grains of S1 are a lot more
porous.
The last regional feature (I1125 = 0.0375) is added at a subvolume edge length
of 1125 voxel ( 1970 m) and can be correlated to the mean distance between the
clayey cemented pore spacings within the S2 CT sample. Because these agglomer-
ations are smaller, and additionally occur more often than within the S1 miniplug,
the observed correlation length is slightly shifted to a lower value.


Figure 7.3 illustrates the results of the I analysis for the Flechtingen (S3) CT mini-
plug. Obviously, a completely different result can be observed, although directional
porosity standard deviation curves basically show the expected decay (Smax 3 %
to Smin 1 %). Nevertheless, scattering between the three directional S curves
is significant, especially for the z-direction and for edge sizes larger than 400 voxel.
Calculated inhomogeneity I clearly reflects this trend. Only one minimum can
be detected (I350 = 0.325) at a subvolume edge size of 350 voxel ( 600 m). Com-
pared to the classical REV analysis (chapter 6.1.3), this edge size correlates with the
first increase of the lower porosity limit. However, due to the massive cementation
of the S3 sample, this size is not a measure of an average grain size, contrary to the
results of the S1 and S2 CT miniplugs.
For subvolume edge lengths larger than 350 voxel, the predominating influence of
the observed fracture is clearly visible. The crack propagates in z-direction which is
additionally indicated by the high average deviation value (Savg 2 %) compared
to the x- and y-direction. The calculated inhomogeneity is constantly high (Iavg
1.1) and gives no indication for any further correlation lengths of structural features
as described for the S1 and S2 specimen. Due to these results, it must be assumed
that it does not seem to be applicable to derive a feasible REV for the S3 CT mini-
plug.

Chapter 8
Pore Network Extraction
So far, equivalent networks, as described in chapter 1.3, are widely accepted and
used for conventional single- and multiphase transport modeling. Due to limitations
in 3D image resolution, computational power and proper network extraction algo-
rithms, only few approaches exist which use in situ geometries for CFD purposes
(Knackstedt et al., 2004; Young et al., 2008; Zhan et al., 2010; Tlke et al., 2010).
In this chapter, the basic theory of advanced image based pore network extraction
is introduced. Recent developments, especially within network extraction technol-
ogy, offer a significantly higher potential for computational fluid dynamics at the
pore scale than the equivalent approaches. Besides, a generalized workflow for the
advanced network extraction is given as well as a detailed characterization of the
derived benchmark and rock sample models.
8.1 Basics of Advanced Network Extraction

Recent advances in surface and volume mesh generation techniques have led to
promising chances for pore network extraction. These image based approaches con-
vert the three dimensional data directly into meshes suitable for (petrophysical)
modeling. Related to a review on network extraction by Young et al. (2008), these
techniques have several major advantages:
Robust mesh generation for topologies and materials of arbitrary complexity;
Generation of highly accurate meshes in which the geometric accuracy of mesh

domains is only dependent on the image accuracy (i.e. voxel resolution);
99
100 Pore Network Extraction
Ability of modeling material inhomogeneities by assigning material properties

based on the image signal strength (image gray value).
8.1.1 Theory
Due to these important developments it is self-evident that these techniques offer
great potential for geoscientific and petrophysical applications. For this thesis, a
special software, SIMPLEWARE by name, has been used to segmentate and de-
rive highly accurate networks from 3D imaging data. This software uses an advanced
algorithm, which includes the benefits as listed above, that is abbreviated with EVo-
MaC: Extended volumetric marching cubes.
This grid based method was first developed by Lorensen and Cline (1987) and has
been adapted for the generation of volume meshes by Mueller and Ruegsegger (1994)
and by Ahrenholz et al. (2006). The basic idea of this technique is to create a trian-
gular mesh that approximates the (segmented) iso-surface in a first step. In a next
step, the normal vectors to the surface are calculated at each vortex of the triangle
before continuing (marching) to the next cube (i.e. voxel in this case).
In cartesian (voxelized) space, up to eight possible parts meet at a vertex so that the
nominal number of base cases to be solved increases from 256 to 4096. By exploiting
symmetry effects, this number can be greatly reduced to 70 fundamental base cases
(Young et al., 2008). This ensures not just nodal but particulary face connectivity
at interfaces between tetrahedralized cubes, which is an important prerequisite to
avoid incompatible displacements in structural analysis. Figure 8.1 shows both, a
fully tetrahedralized cube (A) and an example tetrahedralization of a cube with an
iso-surface passing through (B).
Basics of Advanced Network Extraction 101
Figure 8.1: Tetrahedral base case for: (A) cube with all vertices inside the
object (i.e. no iso-surface running through the model), and (B) a cube with a
single iso-surface passing through it (Young et al., 2008).
This approach overcomes the limitations of conventional surface meshing, i.e. that
topology is preserved with very high accuracy. Resultant meshes can clearly be all
tetrahedral (tet) or of mixed hexahedral-tetrahedral (hex-tet) type. By converting
internal structure voxel to hexahedra, hex-tet ratio - as a quality feature of the
mesh - can be further improved. For base cases (all non surface elements cut by an
iso-surface) the classic eight-noded hexahedral topology is maintained, providing an
optimized mesh-quality. Simplified, this expresses that the mesh is as fine enough
where it should be (e.g. at the pore surface), and that it is coarse where it is allowed
to be (e.g. inner pore volume). Figure 8.2 exemplifies this major development in
volume mesh generation in comparison to older approaches.
Simpleware includes two special modules for advanced grid generation: +FE Grid
and +FE Free (Simpleware, 2010a,b). The first module automatically generates grids
with voxel size accuracy. Built in mesh optimization stringently preserves sample /
material surface smoothness and topology. The +FE Free module offers additional
functionalities for direct control of the global coarseness of the model. At zero setting
the surface is taken from results of +FE Grid meshing, the interior is filled using the
+FE Free algorithm.
If the coarseness control value is changed, the remeshing algorithm will be fully ac-
tivated and allowed to increase (coarsen) or decrease (densify) element edge lengths
based upon the current features. Both modules have been used to generate fully
(sample dependent) optimized grids. A qualitative example is shown in figure 8.3.
Figure 8.2: Improvement of hex-tet ratio and mesh coarsening techniques il-
lustrated on a small cylindrical volume: (A) voxel based mesh (Lego brick
mesh); (B) tetrahedralized mesh using VoMaCs; (C) tetrahedralized with
internal hexahedral grid and partial resorbed surface hexahedra; (D) same as
before but with EVoMaC algorithm used in Simpleware (Young et al., 2008).
Figure 8.3: Qualitative examples for grid accuracy and topology preservation
within Simpleware.
8.1.2 Pore Network Extraction Workflow

For pore network extraction of all investigated samples, a generalized workflow has
been used (schematically shown in figure 8.4). At first, the 3D image data must be
imported into the software, typically as a reconstructed raw data volume file. How-
ever, other commonly used formats are also supported (e.g. DICOM, image stack
files, etc.). After preferred gray scale value range is set, image data should be filtered
in order to increase information value. During this study, a combination of Gaussian
and Median filters turned out to be appropriate to reduce image noise and artifacts
as well as to increase edge sharpness (transition between matrix minerals and pore
space).
In case that image file size exceeds several gigabyte - which happens quite often
for high resolution -CT data - the image data files can be resized / resampled. The
practicability of this feature must be individually considered for each sample, since
resampling operations are always coherent with loss of voxel resolution, i.e. of sample
topology information.
Next, image based segmentation can take place. Depending on the modeling pur-
pose and the desired mesh type, either special threshold or floodfill tools can be
used. For classical fluid transport modeling, only the effective pore network has to
be extracted. Nevertheless, if coupled processes are focussed, also multiple sample
components can be segmented and converted into a multi-component grid.
Finally, the arbitrary segmented structure is converted into a highly detailed sur-
face or volume mesh (as described in the previous chapter), before the 3D grid is
ready for export to a CFD software.
Figure 8.4: Generalized workflow for the phase segmentation and pore network
extraction with Simpleware.
8.2 Benchmark Sample Pore Network

The pase segmentation of the benchmark sample has been the less challenging one
compared to the other investigated specimen, since only two phases (solid and void
space, figure 8.5 left hand side) do exist. Hence, two clearly visible gray value peaks
can be identified within the derived histogram (figure 8.5, right hand side). After
phase identification, the threshold between the void space and the silica beads has
been defined at 15,000 gray values (GV). This value has then been used for the
extraction of the void space within the special software Simpleware as described
within the previous sections. The so derived three dimensional pore space model has
been transferred into a volume mesh for the Lattice Boltzmann and Comsol simu-
lation. The main segmentation and meshing parameters are summarized in table 8.1.
grid name threshold # vol. elements # nodes grid size

REV 15,000 796,245 227,280 6.75 x 6.75 x 6.75 mm
Table 8.1: Main segmentation threshold (void space - matrix) and meshing
parameters of the benchmark sample.
Benchmark Sample Pore Network
Figure 8.5: 2D gray value phase image (left hand side) and resulting segmentation histogram (right hand side) of the benchmark
107
sample.
8.3 Bentheimer Sandstone (S1) Pore Network

The phase segmentation for the S1 specimen has been challenging, since density
variations (and therefore X-ray absorption coefficients) between the mineral phases
are small ( 0.1 - 0.2 g/cm3 ; figure 8.6, left hand side). Hence, the segmentation of
phases based upon the gray value histogram has been difficult to perform compared
to the benchmark sample, although the main peaks can be clearly identified (figure
8.6, right hand side).
Due to the complexity of pore structures within this sample (compare chapter 5),
three different structures / meshes have been extracted / generated: a large REV
mesh, a mesh of a pore network within a dissolved feldspar grain, and a mesh of a
very fine pore network, which is related to clay mineral agglomerations within the
primary pore network of the S1 sample (compare Appendix B). The main segmen-
tation and meshing parameters are summarized in table 8.2.

REV 4600 15,864,044 19,592,947 2.18 x 2.18 x 2.18 mm
feldspar 4650 514,282 120,813 415 x 352 x 520 m
clay 4635 4,297,878 978,334 375 x 336 x 357 m
parameters of the S1 sandstone CT sample.
Bentheimer Sandstone (S1) Pore Network
Figure 8.6: 2D gray value phase image (left hand side) and resulting segmentation histogram (right hand side) of the S1
109
sandstone CT sample.
8.4 Obernkirchen Sandstone (S2) Pore Network

As for the S1 sample, phase segmentation for the S2 sample has been even more
challenging, especially since this sample features smaller grains and pore spacings
compared to the S1 sandstone (figure 8.7, left hand side). As a result, phase peaks
are less distinct on the gray value histogram (figure 8.7, right hand side).
As before, some complex pore structures have been identifed within this sample
(compare chapter 5). Therefore three different structures / meshes have been ex-
tracted / generated: a large REV mesh, a small and fine mesh of a porous network,
which is related to clayey agglomerations, and finally a mesh of a network, which
features some kind of pore channel, caused by different dissolved minerals (compare
Appendix B). The main segmentation and meshing parameters are summarized in
table 8.3.

REV 6750 14,906,725 4,778,759 2.18 x 2.18 x 2.18 mm
clay 6675 919,718 276,356 334 x 423 x 247 m
channel 6735 149,327 44,429 305 x 441 x 215 m
Obernkirchen Sandstone (S2) Pore Network
111
8.5 Flechtingen Sandstone (S3) Pore Network

Due to the high complexity of the pore structures that have been observed within
the S3 specimen, it has not been possible to derive an appropriate representative
elementary volume (compare chapter 6 and chapter 7). Furthermore and due to dif-
ferent stadiums of dissolved mineral phases on a small length scale (several hundred
microns), no appropriate small scale pore structure (e.g. a clay related network)
could be identified. Additionally, the S3 specimen features a distinct microfracture,
which dominates most of the interconected pore network (compare chapter 5). Hence,
only a large mesh, which includes the fracture, has been generated for this specimen
(compare Appendix B). Phase segmentation has been difficult to perform, due to the
broadend density distribution of the partially dissolved mineral phases (figure 8.8,
left hand side).
Although some distinct phase peaks can be observed at low gray values within the
histogram (figure 8.8, right hand side), the phases with higher densities produce a
broad gray value spectrum without any distinct peaks. As a result, it seems le-
gitimate to assume that these mineral phases are stronger affected by dissolving
processes than the others. This seems to be a conclusive result compared to the
SEM and thin section analysis on this sample (chapter 5). The main segmentation
and meshing parameters are summarized in table 8.4.

fracture 5200 25,435,328 17,693,687 2.65 x 2.65 x 2.65 mm
Flechtingen Sandstone (S3) Pore Network
113
Chapter 9
Modeling Fluid Flow with In Situ

Geometries
As already mentioned, a wide range of investigations in the field of classic fluid

dynamics is computed with finite element (FE) and finite volume (FV) toolboxes,
respectively (compare chapter 1). Hence, it seems legitimate to draw a brief compari-
son between a classic FE and the Lattice Boltzmann approach. Following, the results
and advantages of the Lattice Boltzmann simulations in different pore structures of
arbitrary complexity are presented and discussed in detail.
9.1 Classical Finite Element (FE) Approach

For the comparison between the FE and the Lattice Boltzmann approach, the com-
monly used software Comsol 4.2a has been applied. Only a short overview of the
general workflow of fluid flow modeling with Comsol is presented, since this software
package is very well documented (e.g. Pryor (2012)). Similar to Palabos, Comsol
features different types of modules for different types of problems to be solved (e.g.
free and porous media flow module, Darcy flow module, etc.). Besides the general
method used by each program (FE vs. LBM), two main differences exist:
Comsol features extensive functionalities for the modeling of coupled processes

(THMC-coupling) in contrast to Palabos, which is a highly specialized tool
package, solely limited to fluid dynamics related questions.
Comsol directly features a graphical user interface (GUI), which provides a

feeling of confidence related to the setup of the simulation process, though
115
116 Modeling Fluid Flow with In Situ Geometries
it is not as intuitive as Palabos. This means, benchmarking is essential, since

the program code is not visible (blackbox behavior).
The workflow for setting up a FE simulation with Comsol is schematically shown in

figure 9.1. First, the space dimension of the simulation needs to be defined (here: 3D).
Next, the desired physics module has to be chosen. In case of coupled phenomena,
different modules may be selected and combined. For this type of simulation, only
the free and porous media flow module needs to be selected. Following, it has to be
defined, whether the simulation should be of stationary or of instationary type (here:
stationary). Then, several definitions and settings need to be applied: geometry
settings (import of the mesh), material definitions (e.g. the fluid inside of the pore
space) as well as global simulation settings (inflow / outflow zones, walls, boundary
conditions, etc.). After selecting the desired solver (direct / iterative / multigrid), the
simulation can be started. If convergence is reached, simulation stops automatically.
In case of numerical instability, input parameters may be varied for a possibly better
result.
Figure 9.1: General workflow for fluid flow modeling with Comsol.
Classical Finite Element (FE) Approach 117
For an exemplary comparison of both methods, the benchmark sphere model has
been chosen. The flow direction has been set to z-direction for both simulations.
The main input parameters are listed in table 9.1.
type boundary vin [m/s] viscosity [m2 /s] pressure [Pa] DoF
FE no slip 0.00001 / 50 909,000
LBM no slip 0.00001 1e-8 / 2,500,000
Table 9.1: Main input and simulation parameters for the FE and LBM simula-
tion with the benchmark sample (the gap indicates not necessary input data,
DoF = Degree of Freedom).
The results of the FE simulation are shown in figure 9.2, the results of the Lattice
Boltzmann modeling in figures 9.3, 9.4 and 9.5. The main result of the direct compar-
ison between these two methods is as theoretically expected. Although the geometry
is even simple compared to the pore network of the investigated sandstones, the
FE simulation has not been able to derive a stable, i.e. stationary and converging
solution for this model. Figure 9.2 shows the velocity magnitude and the pressure
distribution at the outflow face of the sphere model. Both, average velocity as well as
pressure distribution are significantly too small and the simulation (iterative solver)
ended without reaching convergence criteria or stationary conditions, respectively.
Contrary to this, the Lattice Boltzmann simulation did reach convergence criteria
and has derived a stable numerical solution. Figure 9.3 shows the pressure drop in
direction of the fluid flow. The pressure distribution is - compared to the result of
the FE simulation - very uniformly pronounced. The resulting vorticity field (figure
9.4) indicates an overall steady behavior. Areas of high vorticity can be directly
linked to narrow pore channels within this network, as it could be expected by the-
ory. Furthermore, these zones are allocated to higher fluid flow velocities, as shown
in figure 9.5. According to equation (3.1), all needed parameters for permeability
prediction are derived (table 9.2).
Modeling Fluid Flow with In Situ Geometries
Figure 9.2: Visualization of the velocity magnitude (left hand side) and of the pressure field (right hand side) at the outflow
face of the benchmark sample, simulated with Comsol.
118
Classical Finite Element (FE) Approach 119
Figure 9.3: Visualization of the pressure drop amongst the benchmark pore
network, calculated with Palabos.
Figure 9.4: Visualization of the vorticity field of the benchmark sample after
115,000 iterations (left hand side: x-direction; middle: y-direction; right hand
side: z-direction), calculated with Palabos.
Figure 9.5: Visualization of the normalized velocity field of the benchmark

sample after 115,000 iterations (left hand side: x-direction; middle: y-direction;
right hand side: z-direction), calculated with Palabos.
grid vin [m/s] viscosity [m2 /s] pressure [Pa] sample length [m]
benchmark 0.0001 1e-6 34.6 0.00675
Table 9.2: Main parameters for the permeability estimation from the bench-
mark sample.
With these values, permeability can be calculated to 1.88e-4 m2 . Obviously, this

result is not realistic. Due to the nature of the Lattice Boltzmann simulations, this
result needs to be recalculated. In general it can be assumed that the smaller the
Reynolds number (Re), the slower the Lattice Boltzmann simulation, due to the pre-
dominating friction term (compare equation 3.32). For deriving fluid flow with low
Reynolds number, inlet velocity needs to be very small. By doing this, the needed
time for the computation of the streaming term will increase significantly. To com-
pensate this numerical effect, permeability can be rescaled by adapting the inflow
velocity and scaling it with a constant factor. This fastens the simulation signifi-
cantly, but without changing the overall result! By rescaling with a factor of 1e-8,
permeability can be recalculated to 1.9 Darcy. Due to the grain size and pore
shape of the sphere packing, this is a feasible result.
Results of the Lattice Boltzmann Simulation 121
It can be concluded that with comparable boundary conditions, numerical settings

and temporal effort, the FE approach did not derive a stable result. Contrary, the
Lattice Boltzmann simulation reaches convergence and even produces a robust result
for this geometry. Nevertheless, this does not mean that the FE approach generally
fails for this type of simulation! Certainly, for these complex geometries and with
equal boundary conditions, the Lattice Boltzmann simulation really seems to be the
method of choice. Because of this, FE simulations for the sandstone geometries have
been renounced, due to the significantly higher degree of complexity of the pore
network structures.
9.2 Results of the Lattice Boltzmann Simulation

As seen in the previous section, FE simulations for fluid flow modeling with in situ
geometries does not seem to be a feasible approach. Therefore, the Lattice Boltz-
mann Method is the method of choice for this specific type of modeling. In the
following, the results for the variety of sandstone geometries, as shown in chapter 8
and Appendix B, are presented.
9.2.1 Results for the S1 Sandstone

For the S1 sandstone, three different pore network structures have been extracted.
First, the results of fluid flow simulation for the feldspar network and second for the
clay pore network are presented. Finally, the results for the REV simulation are
discussed. The relevant parameters for these simulations are listed in table 9.3.
feldspar 0.00001 1e-8 477,000 0.00052
clay 0.00001 1e-8 4.09 0.00357
REV 0.00005 1e-7 29 0.00218
Table 9.3: Main parameters for the permeability estimation of the S1 sample.
122 Modeling Fluid Flow with In-Situ Geometries
Feldspar Geometry
The pressure drop for the feldspar pore network geometry (figure 9.6) in flow direc-
tion is significantly high ( 4.7 bar). The reason for this effect can be directly found
within the geometry: The pore network forms a distinct channel structure in flow
direction so that most of the fluid is forced to flow through a single pore channel.
At this point, also the highest pressure gradient could be estimated (middle of figure
9.6).
Figure 9.6: Visualization of the pressure drop amongst the feldspar pore net-
work within the S1 specimen.
The vorticity and velocity field are consistent with this finding. Zones with high
vorticity and flow velocity can be found at the outflow face of this geometry as well
as inside the pore network of the dissolved feldspar mineral (figures 9.7 and 9.8),
where the flow is forced to go through. Obviously, these areas of high vorticity also
seem to be the zones, where most of the dissolution process of the feldspar mineral
takes place, compared to the structures as observed by the SEM (chapter 5).
Figure 9.7: Visualization of the vorticity field of the S1 feldspar pore network
after 146,000 iterations (left hand side: x-direction; middle: y-direction; right
hand side: z-direction).
Figure 9.8: Visualization of the velocity field of the S1 feldspar pore network
The permeability, calculated from the simulation data results in 1.6e-9 m2 . Again,
this value needs to be corrected, since the Reynolds number is still high ( 0.5), due
to the previously described numerical effect of the LBM. After rescaling of the aver-
age fluid velocity to a value of 1.45e-7 m/s, permeability decreases to 15 mD,
which is a realistic value related to the pore size and structure as observed by the
mineralogical investigations described in chapter 5.
Clay Geometry
The pressure drop for the clay gometry is visualized in figure 9.9. The general flow
behavior is similar to the behavior as it has been observed for the feldspar pore
network. The fluid is forced through narrow inflow and outflow channels, again
resulting in local areas of high fluid flow velocity and vorticity, respectively (figures
9.10 and 9.11).
Figure 9.9: Visualization of the pressure drop amongst the clay pore network
within the S1 specimen.
Nevertheless, pressure drop is by decades smaller ( 4e-5 bar), which indicates a

high permeability. Indeed, it is calculated to 1e-7 m2 . Due to the size of the
pore network structures in range of a few microns, this result is the opposite of the
expected data range, since this network is expected to have a very low permeability.
A close look at the Reynolds number shows that for this structure, laminar flow
conditions have not been derived with these settings (RE = 4.28). Because of this,
fluid velocity urgently needs to be rescaled. As seen in the petrophysics chapter,
fluid permeablility should be in range of about 1 mD or even less. Recalculation
with a fluid velocity of about 1e-14 m/s leads to a corrected permeability of 0.9
mD.
Figure 9.10: Visualization of the vorticity field of the S1 clay pore network
Figure 9.11: Visualization of the velocity field of the S1 clay pore network after
side: z-direction).
S1 REV Geometry
Finally, the REV pore network has been used for a representative modeling ap-
proach. The pressure drop is shown in figure 9.12. Pressure appears to drop con-
stantly amongst this geometry, which results in an average gradient of 0.00029 bar.
As expected, fluid velocity magnitude and vorticity correlate very well to the pore
structure. Pore throats are linked to higher velocity and vorticity values, whereas the
large pores indicate lower values at all (figures 9.13 and 9.14). Due to homogeneous
structure of this REV network (compare chapter 5 and chapter 6), these features
are uniformly distributed, leading to no recognizable preferential flow paths or even
flow traps within this model. Since the REV geometry has way more cells than the
significantly smaller feldspar or clay grid, inflow fluid velocity has been set to higher
speed in order to fasten the simulation. Of course, this leads to a high Reynolds
number (RE = 11) and generally to non laminar flow per definition. Therefore fluid
velocity needs to be rescaled by a factor of 1e-7, which leads to a permeability of
568 mD (5.68e-6 m2 without recalculation). This is a very good result compared to
the average laboratory k-value of the S1 specimen set (chapter 5.2.1). This result is
also congruent with REV estimation techniques and confirms the REV size for this
specific sandstone.
Figure 9.12: Visualization of the pressure drop amongst the REV pore network
of the S1 specimen.
Figure 9.13: Visualization of the vorticity field of the S1 REV pore network
Figure 9.14: Visualization of the velocity field of the S1 REV pore network

As for the S1 sandstone, three different pore geometries have been extracted for the
S2 sandstone (compare chapter 8 and Appendix B). First, the results of fluid flow
simulation through the clay network are presented. Following, results for the flow
through the pore channel structure are introduced, and finally the derived data of
the S2 REV simulation are presented and discussed. The main parameters for these
simulations are compiled in table 9.4.
clay 0.00005 1e-7 1380 0.000247
channel 0.00001 1e-8 0.2921 0.000215
REV 0.00001 1e-8 0.01754 0.00218
Clay Pore Network
The calculated pressure distribution for the S2 clay geometry is shown in figure
9.15. For this network, the pressure gradient is higher than for the S1 clay structure
(0.0138 bar). The general pressure distribution seems to be very homogeneous, which
indicates a good connectivity between the pores. This effect has not been observed
by this extent for the S1 clay network. Although both sandstones contain the same
type of clay and clayey structures. The magnitude of fluid vorticity as well as of
fluid velocity are again directly linked to narrow pores and pore throats, as shown
in figures 9.16 and 9.17. Additionally, two larger zones of high velocity and vorticity
could be observed at the outflow face of this grid. Obviously, a preferential flow path
forms the exit for most of the fluid at this point of the network.
Figure 9.15: Visualization of the pressure drop amongst the clay pore network
of the S2 specimen.
Despite of this good pore connectivity, the overall pressure drop is the highest one
observed for all S2 network structures. Nevertheless, resulting Reynolds number is
indicating laminar flow conditions (RE = 0.1235), confirming the connectivity as-
sumption. The permeability is calculated to 8e-9 m2 , which is again overestimated
by the Lattice Boltzmann simulation. After rescaling of the inflow velocity by a fac-
tor of 1e-7, permeability reaches 0.8 mD, which is an equal result as observed for
the estimated permeability of the S1 clay network.
Figure 9.16: Visualization of the vorticity field of the S2 clay pore network
Figure 9.17: Visualization of the velocity field of the S2 clay pore network after
side: z-direction).
Channel Geometry
The channel pore structure is formed by a mixture of dissolved mineral phases (com-
pare chapter 5) and is an interesting feature within the S2 pore network. It is
expected to act as a preferential flow path for the entire network structure, therefore
permeability is expected to be much higher than for the S2 REV network. In fact,
the calculated pressure drop (figure 9.18) is very low ( 2.9e-6 bar), although the
Reynolds number almost indicates laminar flow conditions (Re = 0.0669) for this
structure. A closer look to the velocity magnitude map (figure 9.20) confirms the
assumption of a preferential flow path, since the highest velocity magnitudes can be
found within the large pore spacings. Additionally, the vorticity field (figure 9.19)
gives a good indication for a fast, but non turbulent flow regime, because only very
few areas with higher vorticity exist.
Figure 9.18: Visualization of the pressure drop amongst the channel pore net-
work of the S2 specimen.
Figure 9.19: Visualization of the vorticity field of the S2 channel pore network
Figure 9.20: Visualization of the velocity field of the S2 channel pore network
Consistently, calculated permeability is in range of 2e-8 m2 . Nevertheless, this

value still is overestimated, resulting in a rescaled fluid velocity for recalculation of
the permeability by a factor of about 1e-5. With this, permeability yields to a value
of about 230 mD, which is an acceptable and reasonable value for this type of struc-
ture, related to the findings of chapter 5.
S2 REV Geometry
Finally, the S2 REV simulation has been performed upon the equivalent REV mesh.
The resulting pressure distribution is shown in figure 9.21. The average pressure gra-
dient equals to 1.75e-5 bar, which is a lower value compared to the result derived
for the S1 sandstone. This is a surprising finding: due to the smaller average pore
size, the pressure gradient was expected to be higher than for the S1 geometry (with
equal boundary settings).
Figure 9.21: Visualization of the pressure drop amongst the REV pore network
of the S2 specimen.
The velocity map (figure 9.23) also gives an indication for an overall slow fluid flow,
except for some few narrow pores, which are located at the inflow zone. Contrary to
this, vorticity (figure 9.22) is very distinct in y-direction through the pore network,
whereas vorticity in the two other directions solely correlates with the observed in-
flow structures. Obviously, narrow pore channels exist in the y-direction so that the
fluid flow is forced to go through this small pores, resulting in a very homogeneous
distribution of flow channels with higher vorticity and velocity, respectively.
Additionally, the previously observed zones with higher vorticity give an indica-
tion for non laminar flow in this vicinity. In fact, the average Raynolds number
here is larger than 1 (RE = 1.363), which confirms this assumption. Nevertheless,
the permeability for this sample has been calculated to 8 mD. This result is in
accordance with the results from the laboratory experiments (compare chapter 5).
This also confirms the estimated S2 REV size, derived by the classic and new REV
estimation techniques.
Results of the Lattice Boltzmann Simulation
Figure 9.22: Visualization of the vorticity field of the S2 REV pore network after 97,000 iterations (left hand side: x-direction;
middle: y-direction; right hand side: z-direction).
135
Modeling Fluid Flow with In-Situ Geometries
Figure 9.23: Visualization of the velocity field of the S2 REV pore network after 97,000 iterations (left hand side: x-direction;
136


Due to the observed structure of the S3 sample, solely one pore network mesh has
been generated, including the clearly identified microfracture (compare chapter 5
and chapter 6). The main modeling parameters for this simulation are summarized
in table 9.5.
Frac 0.00005 1e-8 116 0.00265
The calculated pressure field is visualized in figure 9.25 and gives in addition to the
fluid velocity map (figure 9.24), and the fluid vorticity (figure 9.26) a good impression
about the overall flow regime. Obviously, the fracture is predominating the fluid
transport within this network structure.
Figure 9.24: Visualization of the pressure drop amongst the pore network of
the S3 specimen.
The fracture obeys the highest pressure gradient (0.00116 bar) as well as the highest
velocity and vorticity magnitude, respectively. Nevertheless, the rest of the pore
network seems to be well connected to this fracture and participates to some extent
to fluid flow. Indications for this can be found in the velocity and vorticity figures,
where these values increase even outside the microfracture.
As already mentioned, it has not been possible to derive a REV grid for the S3
sample, due to this special pore network feature. Therefore, a permeability predic-
tion needs to be discussed reasonably, since most of the flow through this model
depends on the fracture. Concerning a representative fluid flow model, the impact of
the fracture should be significantly smaller, due to the low porosity and pore size and
structure. The permeability has been calculated to 3.5e-8 m2 out of the simulation
data. The Reynolds number indicates a turbulent flow regime (RE 13) so that
the average inflow velocity needs to be rescaled for permeability recalculation. By
doing this, a factor of 1e-5 delivers the best permeability fitting result of 350 mD.
This result is consistent and comparable with the findings of the S2 channel network
simulation and additionally with the fracture size as observed for the S3 sample in
chapter 5.
Results of the Lattice Boltzmann Simulation
Figure 9.25: Visualization of the vorticity field of the S3 pore network after 158,000 iterations (left hand side: x-direction;
139
Modeling Fluid Flow with In-Situ Geometries
Figure 9.26: Visualization of the velocity field of the S3 pore network after 158,000 iterations (left hand side: x-direction;
140
Combined Petrophysical and Modeling Results 141
9.3 Combined Petrophysical and Modeling Re-

sults
In this section, special emphasis on the results of the petrophysical laboratory in-
vestigations in combination with the results of the Lattice Boltzmann simulations
will be given. The most important parameter, which needs to be calculated from
numerical fluid flow simulations, is probably the permeability of the porous media
(compare chapter 1 and chapter 5). Therefore it seems reasonable to take a closer
look on the combination of all results, which have been derived by now.
The results of the permeability simulations and experiments are summarized in table
9.6. As described within the previous chapter, the speed of the Lattice Boltzmann
simulations can be scaled with inflow velocity. By doing this, most of the permeabil-
ity values are overestimated so that the fluid velocity needed to be rescaled in order
to derive reasonable k-values as well as the low Reynolds number criteria for laminar
flow conditions.
The permeability estimation for small geometries is mostly depending on a good

petrophysical knowledge of the entire sample set, with meaningful statistics as shown
in chapter 5. Certainly, these values obey the highest uncertainty, since it is not
possible to measure the permeability of these structures directly in the laboratory.
Because of this, velocity values have been rescaled for the Lattice Boltzmann simula-
tion in that way, so that the corrected values are conclusive with the results derived
from SEM and thin sectioning as well as from petrophysical statistics.
grid sandstone ksim [m2 ] kcor [mD] klab [mD] kset [mD]
clay S1 1.05e-7 0.9 (555) (598)
feldspar S1 1.6e-7 15.7 (555) (589)
REV S1 5.6e-6 568 555 589
clay S2 8.2e-9 0.82 (11) (16)
channel S2 2.3e-8 229 (11) (16)
REV S2 7.8e-15 8 11 16
Frac S3 3.5e-8 351 (1.2) (17)
Table 9.6: Summary of the permeability values derived by numerical and petro-
physical laboratory experiments (ksim = simulated permeability; kcor = cor-
rected simulated permeability; klab = sample laboratory permeability; kset =
average specimen set permeability).
By doing this, it can be observed that both clay geometries feature a very similar
permeability in range of 1 mD (table 9.6). This is a very conclusive result, since
SEM investigations have shown that both sandstone types contain the same type of
clay, featuring the same type of structure (clay booklets). Furthermore, the velocity
and vorticity maps of the clay geometries give a very good impression about the flow
regime inside of these structures. The inflow pressure for a fluid into these regions is
higher than for the rest of the regular porous network. This leads to the assumption
that the clay pore networks hold the fluid in the inside, until a higher pressure drop
is applied. Then, this type of network participates on the entire porous network as
it could be expected by theory.
The results for the feldspar and channel geometry are more difficult to interpret,
since these two structures are - due to their different origin - not directly compa-
rable. Nevertheless, the simulations deliver in first case some profound knowledge
on the complex interaction between different pore networks. As seen, pressure and
inflow velocity are in range of the values calculated for the REV network. This leads
to the result that both networks directly participate to overall fluid flow. In case
of the feldspar network, the type of flow obviously influences the dissolution process
and locality upon these minerals. Once a flow path is created, zones of high velocity
and especially vorticity occur, where the highest degree of mineral degradation could
be observed. This is additionally confirmed by the findings of the SEM imaging
(chapter 5).
Combined Petrophysical and Modeling Results 143
The permeability has been estimated to be in range of 15 mD. Compared to

the pore size of the overall S1 pore network, this seems to be an acceptable result
for the k-value inside such a dissolved mineral.
In case of the S2 channel structure it can be clearly assumed that this type of
local network acts as a preferential flow path for the fluid inside the equivalent REV
network. Although zones of high velocity occur in this structure, vorticity is small in
comparison. Contrary to the feldspar network, since here velocity and vorticity are
much more linked to each other. This is obviously an effect caused by the different
types of minerals. In this case, dissolved quartz forms most of this special structure.
Furthermore it can be assumed that hardly any precipitation process will take place
in this specific pore network, due to the high transport velocity of the fluid within.
The estimated permeability of 230 mD underlines the previous assumptions.
Similar to the S2 channel structure, the fracture dominates fluid flow in the S3 pore
network, although on an even larger scale. The simulation has shown that, although
the S3 network indicates good overall connectivity, the fracture indeed dominates
the flow regime and, as a further result, greatly increases the overall permeability.
On the laboratory scale, the equivalent S3 sample has a k-value of about 1 mD,
due to the extensive amount of clay coatings and cementation minerals (compare
chapter 5). On the smaller (milimeter) scale of this CT sample, this crack increases
permeability to 350 mD and forms a flow path herein, where dissolved mineral
phases are supposed to be transported and, in the end, even participated.
Up to this point it can be concluded that the Lattice Boltzmann simulations with in
situ structures do not solely deliver information on flow regime and permeability, but
especially a very good insight on small scale process understanding, such as (e.g.)
dissolution and precipitation processes and localities, respectively. Additionally, the
influence of such structures upon the REV pore networks may be estimated and
assessed.
This leads to the comparison of results of the fluid flow modeling and of the lab-
oratory experiments for the estimated representative elementary volumes of the S1
and S2 sandstones. Due to the nature of the Lattice Boltzmann simulations, inflow
velocity scales the speed of the overall numerical process. After rescaling, the perme-
ability for both REV samples has been recalculated. The results show a very good
fit compared to the laboratory experiments. The aberration between both values for
both sandstones is shown in table 9.7.
grid kcor [mD] sample origin klab [mD] aberration [%] / [mD]
S1 REV 568 BE 02 555 + 2.3 / +13
S2 REV 8 OK 60 11 - 27 / - 3
Table 9.7: Aberration between calculated and measured k-values for the S1
REV and S2 REV, respectively.
This result also shows that the new approach to estimate a porous media REV
delivers feasible representative grid sizes for this type of fluid flow modeling. In
combination with the results derived by the simulations with the special (small)
pore networks (e.g. the clay structures), a new undestanding of the interaction be-
tween these structures arises. Hence, it can be assumed that the much smaller clay
pore networks do not participate to overall fluid flow under normal flow (bound-
ary) conditions. The fluid stays trapped, until the overall pressure exceeds a specific
threshold value.
To elaborate on this point, it would probably make sense to assume that for a
mixture of fluids, these types of structures should be aimed for (e.g.) enhanced fluid
(e.g. oil) recovery. Furthermore, it is possible to predict areas of higher fluid trans-
port as well as of low or even creep flow. Additionally, zones where dissolution or
precipitation takes place can be assessed and probably predicted.
Chapter 10
Conclusions and Outlook
Finally, the results of all investigations are assessed and concluded. Last, an outlook
on reasonable future prospects and investigations is given.
10.1 Conclusions
The development of the Computed Tomography (CT) technology has been an ongo-
ing and rapid process during the last few years. The size of the CT scanners has been
greatly decreased, whereas spatial resolution has been consequently pushed forward
to the micrometer scale. Meanwhile CT systems are well established in the field of
geoscience and also for petrophysical investigations. Parallel to this, new techniques
for fluid flow modeling and especially for the extraction of highly detailed pore net-
works - and numerical grids therefrom - have been developed. Therefore, detailed
investigations towards the basics of -CT and of the CFD challenges have been per-
formed, including a new approach for the extraction and generation of highly detailed
in situ pore structures for numerical modeling. Furthermore, a new technique for the
estimation of representative elementary volumes (REV) in close combination with
classical mineralogical and petrophysical investigations has been introduced. Finally,
desired data and results have been discussed and assessed as a new joint approach
for next generation CFD and petrophysical modeling, respectively.
Besides a general introduction on state of the art challenges for fluid flow model-
ing and of state of the art CT technology, which form the fundamental motivation
for this thesis, a detailed introduction on next generation -CT imaging has been
focused first. Along the basic principle and theory of this imaging technique, it has
been shown that it is possible to derive 3D data sets with spatial resolution and
145
146 Conclusions and Outlook
image sharpness within the micron and even submicron range, respectively. With
these prerequisites it has been possible to scan, extract and evaluate very small pore
space features as clay induced porosities. Nevertheless, CT settings must be cho-
sen individually and reasonable for each single sample, in order to derive the best
possible agreement between image quality, scanned sample volume, scanning and
reconstruction time, and resulting file size of the 3D data set.
Next, basics on fluid dynamics have been introduced to set up the fundament for the
derivation of the Navier-Stokes equations by the Lattice Boltzmann Method (LBM).
Here, the theoretical advantages and disadvantages between classical approaches and
the LBM are assessed. By theory, this technique should be able to handle the struc-
tures of arbitrary complexity very well, and even practically this expectation has
been measured up. The D3Q19 model has been described, since this Boltzmann
model has been used for the following simulations with the Lattice Boltzmann tool-
box Palabos. The workflow and setup for this type of CFD is shown and discussed
in detail.
For this thesis, three different types of sandstones have been used to extract pore
structures of arbitrary complexity. Initially, these sandstones have been extensively
examined in terms of mineralogy and pore structure, in order to derive a profound
knowledge of the inner structures of these rocks. For each rock type, a total set
of fifty samples has been prepared for petrophysical experiments (e.g. permeability,
specific surface, porosity, etc.), consequently aiming at the physical characterization
of the void space. These results have been compared with the findings of the classical
mineralogical methods for a detailed joint sample characterization. The specimen
greatly differ in terms of pore size, clay amount dissolution and precipitation effects
as well as in terms of the pore structure. The S1 pore network is characterized by
large pores with very good connectivity, only slightly disturbed by very few clay
packages and dissolved feldspar minerals, respectively. Although the S2 network is
also well connected, pores are smaller and even more influenced by the clay min-
erals, which reduce overall permeability significantly. Besides, dissolved quartz can
be found, which may form preferential flow paths through the pore space. The S3
pore network is extensively influenced by clay minerals, cementation and weathering
effects. Furthermore, the occurance of a distinct microfracture makes the REV esti-
mation impossible, due to its significant influence on porosity and also permeability.
The S1 sandstone is the most porous and also most permeable one of the investigated
Conclusions 147
rocks. The S2 sandstone is in between S1 and S3, which is characterized by a wide

span of porosity and permeability values, respecitively. As a result, a variety of pore
structures has been identified for the extraction and generation of numerical meshes
for fluid flow modeling.
After this detailed sample characterization, a classical and a new approach to de-
termine a voxel resolution dependent representative elementary volume have been
introduced and assessed. Both methods are capable of deriving the correct REV
size, but in addition, the new approach could be used to calibrate the REV in terms
of directional modeling purposes (anisotropic fluid flow). The new developed in-
homogeneity analysis, which has been performed for all three types of sandstones,
greatly inreases the knowledge on different pore network structures, and furthermore
at which scale they may be observed. In close combination with the previous inves-
tigations, this new method has proven to be a very feasible and reliable tool for this
type of sample characterization.
Ongoing, basics on pore network extraction and grid generation for fluid flow mod-
eling have been described in detail. The generated grids are characterized by a high
grade of detail and pore structure preservation. With this, very small and complex
structures are made accessible for fluid flow modeling, since now the pore wall topol-
ogy - as a function of the derived voxel resolution - is fully preserved. The applied
software algorithm works excellently and provides a robust mesh optimization tool-
box in addition. With this, even large REV grids can be greatly optimized without
losing any significant pore structure information.
Finally, the results of the fluid flow simulations are presented, discussed and as-
sessed in detail and in close combination with the results from previous laboratory
investigations. A direct comparison between a classical FE toolbox and the Lattice
Boltzmann code has shown that, in fact under comparable hardware and numerical
boundary conditions, the FE method is not able to deliver a stable result, whereas
the Boltzmann code provides stable and reliable data - especially in terms of REV
modeling! The REV simulation results fit very well with the measured laboratory
data and, by this, confirm the new approaches on REV size and inhomogeneity esti-
mation and sample characterization. In addition, small scale pore structures, as clay
porosities, have been investigated and these results have increased the fundamental
understanding of processes (e.g. transport) at this scale.
In total, the combination of high resolution -CT and of the LBM forms a very
feasible and reasonable choice for fluid flow modeling. It allows representative mod-
eling and, in addition, allows to resolve small scale and even fracture pore structures
very nicely. This enables the LBM to be applied to any other type of rock and ob-
served pore structure in the future.
10.2 Outlook
The Lattice Boltzmann Method, in combination with high resolution -CT imaging
and classical mineralogical and petrophysical investigations has proven to be a mag-
nificently reliable tool for fluid flow simulations and (flow) process understanding at
the microscale. Further investigations will focus in the optimization of the 3D data
set reconstruction, e.g. by using massively parallelized reconstruction on graphical
processing units (gpu). With this, even larger data sets can be prepared for ongoing
processing purposes very rapidly.
The visualization of the derived simulation data is up to this point very limited,
due to the very large size of the timestep files (several gigabyte per timestep). For
this thesis, visualization it has been solely limited to mostly two dimensions. In
the future, the output files are going to be optimized and prepared for a fully 3D
visualization of all derived simulation data.
Consequently, new types of sandstones will be investigated so that - besides an

improved mineralogical and petrophysical understanding - fluid flow processes in a
large variety of pore networks can be assessed and compared. This, in combination
with a joint data analysis as presented in this thesis, will probably greatly increase
the knowledge upon these types of rocks. Furthermore, other types of rock, e.g. such
as carbonates, are going to be investigated and assessed in a similar manner. These
investigations are not only limited to scientific issues. In addition, such an advanced
reservoir characterization, including a greatly increased process knowledge on small
scales, may be used by (e.g.) by oil and gas industry for a significantly better reser-
voir understanding in terms of predictive (modeling) issues or even reservoir fluid
recovery.
Outlook 149
Finally, the Boltzmann code will be improved and enhanced so that multiphase fluid
flow simulations, especially with immiscible fluids, can be performed within these in
situ structures. A combination of this type of modeling with other fields of petro-
physics (e.g. Nuclear Magnetic Resonance Spectroscopy) seems to be a meaningful
future prospective. Coupled process modeling is strictly limited by the nature of the
Lattice Boltzmann Method. Nevertheless, coupling of thermal and fluid transport
on this micrometer scale will be certainly investigated.
Appendix A
Petrophysical Data
sample ID d [g/cm3 ] [%] kkl [mD] [106 SI] Sm [m2 /g] Spor [1/m]
BE 01 2.658 21.5 541 -4.20 0.404 3.65
BE 01A 2.651 21.3 562 -2.79 / /
BE 02 2.661 21.8 596 -0.47 0.327 3.12
BE 02A 2.651 21.5 555 -3.29 0.347 3.36
BE 03 2.653 21.2 585 -4.66 0.342 3.37
BE 03A 2.649 21.6 582 -0.93 0.203 1.95
BE 04 2.649 21.9 611 -5.63 0.320 3.02
BE 04A 2.650 21.7 605 -6.08 0.704 6.72
BE 05 2.654 22.5 642 -3.79 0.382 3.49
BE 05A 2.648 22.0 627 -5.16 / /
BE 06 2.652 22.2 635 -4.73 0.292 2.71
BE 06A 2.644 22.0 613 -2.82 0.364 3.42
BE 07 2.650 21.7 611 -4.71 0.420 4.01
BE 07A 2.655 21.9 620 -3.78 0.438 4.14
BE 08 2.681 22.2 625 -3.82 0.812 7.63
BE 08A 2.659 21.9 618 -3.30 0.507 4.82
BE 09 2.638 22.2 637 -3.75 0.468 4.33
BE 09A 2.640 22.5 645 -3.31 0.769 7.01
BE 10 2.649 21.9 587 -3.76 0.406 3.84
BE 10A 2.644 21.5 605 -3.27 / /
BE 11 2.674 21.8 594 -3.77 0.520 4.98
BE 11A 2.649 21.4 606 -4.66 0.427 4.16
151
152 Petrophysical Data
BE 12 2.653 21.0 579 -3.71 0.810 8.07

BE 12A 2.641 21.0 579 -3.70 0.292 2.91
BE 13 2.648 20.4 559 -2.77 0.530 5.48
BE 13A 2.646 20.6 553 -4.15 0.417 4.26
BE 14 2.648 20.4 555 -4.14 0.344 3.56
BE 14A 2.644 20.3 537 -3.68 0.556 5.78
BE 15 2.651 20.0 526 -5.04 0.452 4.78
BE 15A 2.637 20.1 537 -3.66 / /
BE 16 2.651 20.2 538 -4.16 0.509 5.33
BE 16A 2.656 20.6 548 -4.63 0.327 3.35
BE 17 2.646 20.5 533 -5.98 0.334 3.44
BE 17A 2.655 20.7 530 -4.64 0.541 5.51
BE 18 2.647 20.5 538 -7.86 0.629 6.45
BE 18A 2.647 20.6 557 -5.55 0.489 4.99
BE 19 2.639 20.1 522 -5.04 0.411 4.32
BE 19A 2.644 20.3 529 -3.67 0.433 4.51
BE 20 2.651 22.3 588 -3.32 0.324 2.99
BE 20A 2.645 22.3 674 -3.78 / /
BE 21 2.658 22.6 665 -3.32 0.344 3.14
BE 21A 2.647 23.0 686 -4.30 0.227 2.02
BE 22 2.650 21.0 540 -4.64 0.595 5.92
BE 22A 2.672 22.0 574 -3.80 0.620 5.88
BE 23 2.645 22.5 664 -4.72 0.358 3.25
BE 23A 2.654 22.2 653 -3.76 0.332 3.09
BE 24 2.650 22.2 653 -4.72 0.344 3.19
BE 24A 2.648 22.1 622 -2.82 0.340 3.18
BE 25 2.652 22.5 642 -3.78 0.475 4.35
BE 25A 2.666 22.4 652 -4.75 0.739 6.82
Table A.1: Basic petrophysical properties (from left to right:
grain density, porosity, Klinkenberg permeability, suscepti-
bility, mass and pore volume specific BET surface) of the
Bentheimer sandstone (S1) specimen set.
Petrophysical Data 153
OK 10A 2.676 16.2 11 -3.07 0.781 10.78
OK 10B 2.679 16.0 11 -2.65 / /
OK 20A 2.689 16.7 18 -1.77 0.654 8.79
OK 20B 2.677 16.5 15 -2.64 0.568 7.71
OK 30A 2.681 16.4 14 -2.20 0.576 7.90
OK 30B 2.674 16.2 17 -1.32 0.678 9.37
OK 40A 2.674 16.2 20 -0.88 0.543 7.50
OK 40B 2.679 16.6 17 -1.77 0.619 8.32
OK 50A 2.671 16.0 10 -3.92 0.817 11.50
OK 50B 2.676 16.1 11 -3.49 / /
OK 60A 2.675 16.1 11 -3.06 0.559 7.80
OK 60B 2.677 15.9 11 -3.93 0.403 5.70
OK 70A 2.678 16.0 11 -2.19 0.880 12.38
OK 70B 2.695 15.9 11 -3.08 0.598 8.55
OK 80A 2.688 17.8 47 -5.80 0.378 4.71
OK 80B 2.691 16.4 28 -3.52 0.437 6.02
OK 90A 2.691 16.2 13 -1.32 0.617 8.61
OK 90B 2.695 16.9 17 -0.44 0.634 8.39
OK 100A 2.683 16.3 19 -3.93 0.721 9.97
OK 100B 2.676 18.0 58 -2.21 / /
OK 110A 2.679 15.6 10 -1.76 0.668 9.66
OK 110B 2.675 15.5 12 -3.94 0.892 12.98
OK 120A 2.671 17.0 25 -2.21 0.613 8.01
OK 120B 2.675 16.2 20 -0.88 0.796 10.99
OK 130A 2.686 15.6 11 -2.63 0.704 10.25
OK 130B 2.678 16.4 35 -2.20 0./ /
OK 140A 2.694 15.7 9 -2.21 0.924 13.34
OK 140B 2.705 16.1 9 -1.33 0.531 7.46
OK 150A 2.672 16.3 13 -1.32 0.758 10.41
OK 150B 2.708 16.0 22 -3.09 / /
OK 160A 2.704 16.8 18 -1.33 0.563 7.54
OK 160B 2.702 16.4 12 -2.65 0.494 6.82
OK 170A 2.696 16.6 9 -3.97 0.621 8.41
OK 170B 2.695 16.7 9 -1.76 0.512 6.90

OK 180A 2.692 15.5 9 -4.39 0.440 6.47
OK 180B 2.681 15.7 10 -0.88 0.850 12.27
OK 190A 2.704 16.5 10 -1.76 0.738 10.13
OK 190B 2.716 16.7 12 -0.89 0.767 10.39
OK 200A 2.676 16.0 10 -2.19 0.736 10.35
OK 200B 2.704 15.8 12 -3.10 / /
OK 210A 2.705 16.5 10 -1.77 0.551 7.56
OK 210B 2.682 16.9 13 -2.20 0.711 9.36
OK 220A 2.683 16.6 9 -3.09 0.778 10.49
OK 220B 2.679 16.9 9 -1.76 0.690 9.09
OK 230A 2.684 17.2 15 -2.20 0.739 9.56
OK 230B 2.685 16.7 13 -0.44 0.931 12.48
OK 240A 2.697 16.7 12 0.00 0.556 7.48
OK 240B 2.702 16.5 10 -1.77 0.619 8.49
OK 250A 2.698 16.6 12 -4.85 0.536 7.29
OK 250B 2.707 16.2 10 -1.32 0.656 9.17
Obernkirchen sandstone (S2) specimen set.
Petrophysical Data 155
FL 10 2.682 15.0 7.28 240 2.62 39.99
FL 20 2.701 8.2 0.12 285 / /
FL 30 2.704 7.6 0.48 206 0.94 30.68
FL 40 2.705 10.3 18.22 389 1.57 36.83
FL 50 2.679 10.5 23.06 378 1.18 26.72
FL 60 2.683 12.1 8.62 201 / /
FL 70 2.678 15.3 108.62 266 1.27 18.78
FL 80 2.684 8.6 1.50 314 1.07 30.30
FL 90 2.664 11.5 32.18 421 1.39 28.47
FL 100 2.697 12.6 10.96 202 2.50 46.90
FL 110 2.699 14.1 22.90 207 / /
FL 120 2.690 11.0 0.29 301 1.39 30.39
FL 130 2.689 12.6 9.70 201 / /
FL 140 2.695 8.4 0.10 381 1.56 45.82
FL 150 2.712 10.4 1.21 253 1.17 27.52
FL 160 2.698 10.5 2.28 346 1.71 39.32
FL 170 2.693 10.4 2.09 337 1.47 34.16
FL 180 2.716 12.4 8.77 204 1.28 24.55
FL 190 2.743 16.7 77.62 206 2.19 29.91
FL 200 2.696 14.5 21.70 209 2.58 41.02
FL 210 2.696 7.6 0.46 288 1.44 47.30
FL 220 2.720 15.7 7.12 239 1.45 21.27
FL 230 2.713 10.2 0.66 247 1.35 32.14
FL 240 2.696 6.5 0.20 305 1.30 50.31
FL 250 2.709 12.3 8.28 205 1.37 26.45
FL 260 2.719 10.9 17.26 423 1.21 26.92
FL 270 2.731 12.4 1.87 220 1.57 30.50
FL 280 2.763 16.6 154.41 209 2.17 30.10
FL 290 2.733 16.4 141.93 209 1.54 21.43
FL 300 2.675 10.3 1.77 335 2.47 57.25
FL 310 2.713 13.4 8.00 212 1.17 20.61
FL 320 2.742 15.8 98.51 211 1.45 21.20
FL 330 2.707 8.5 0.91 212 1.57 45.79
FL 340 2.682 6.5 0.16 203 / /

FL 350 2.705 8.3 0.29 217 1.22 36.67
FL 360 2.714 14.1 27.19 207 1.68 27.74
FL 370 2.709 12.5 4.92 212 1.35 25.65
FL 380 2.696 7.7 0.30 205 1.91 38.41
FL 390 2.716 12.6 7.24 195 / /
FL 400 2.681 7.4 0.23 399 2.09 70.09
FL 410 2.696 11.1 3.31 332 1.57 33.89
FL 420 2.699 8.3 0.60 211 1.12 33.29
FL 430 2.678 10.8 1.67 240 / /
FL 440 2.675 12.2 2.06 194 1.19 22.93
FL 450 2.686 9.5 3.64 325 1.37 34.84
FL 460 2.682 8.2 0.42 299 1.22 36.54
FL 470 2.682 12.8 2.52 201 1.66 30.45
FL 480 2.684 8.2 0.80 207 1.05 31.66
FL 490 2.674 9.5 1.25 307 1.20 30.71
FL 500 2.675 8.6 0.30 296 1.14 32.31
Flechtingen sandstone (S3) specimen set.
Petrophysical Data
Figure A.1: Histograms of petrophysical data for the S1 specimen set.

157
Petrophysical Data

158
Petrophysical Data

159
Appendix B
Generated Meshes for CFD

Modeling
161
Generated Meshes for CFD Modeling
Figure B.1: Generated REV mesh for the benchmark sample.

162
Figure B.2: Generated REV mesh of the S1 sample.

163
Figure B.3: Generated feldspar porosity mesh of the S1 sample.

164
Figure B.4: Generated clay porosity mesh of the S1 sample.

165
166
Figure B.5: Generated REV mesh of the S2 sample.

Figure B.6: Generated clay porosity mesh of the S2 sample.

167
Figure B.7: Generated channel porosity mesh of the S2 sample.

168
169
Figure B.8: Generated fracture porosity mesh of the S3 sample.

170 Generated Meshes for CFD Modeling
Appendix C
Lattice Boltzmann Source Code
1 #i n c l u d e " p a l a b o s 3 D . h"
2 #i n c l u d e " p a l a b o s 3 D . hh"
3 #i n c l u d e " p a l a b o s 2 D . h"
4 #i n c l u d e " p a l a b o s 2 D . hh"
5 #i n c l u d e <v e c t o r >
6 #i n c l u d e <cmath>
7 #i n c l u d e 
8 #i n c l u d e <f s t r e a m >
9
10 u s i n g namespace p l b ;
11 u s i n g namespace s t d ;
12
13 typedef double T;
14 #d e f i n e DESCRIPTOR d e s c r i p t o r s : : D 3 Q 1 9 D e s c r i p t o r
15
16 T lx = 0. , l y =0. , l z =0.;
17 T dx = 0 . , dt = 0 . ;
18 p l i n t nx =0 , ny =0 , nz =0;
19
20 Array<T,3 > l o c a t i o n ( 0 . , 0 . , 0 . ) ;
21
22 plint N = 0;
23 T uInlet = 0.;
24 T kinematicViscosity = 0 . ;
25
26 plint maxIter = 250000;
27 plint imageIter = 200;
28
29 T uLB = 0 . 0 0 2 ;
30 T nuLB = 0 . , t a u = 0 . , omega = 0 . ;
31
32 b o o l writeFullVTK = f a l s e ;
33
34 M u l t i S c a l a r F i e l d 3 D s o l i d F l a g s = 0 ;
35 M u l t i B l o c k L a t t i c e 3 D <T, DESCRIPTOR> l a t t i c e = 0 ;
36 {
37 // These p a r a m e t e r s are for technical , internal usage .
38 plint margin = 2 ;
39 plint extraLayer = 0;
40 plint borderWidth = 1 ;
41 plint envelopeWidth = 1 ;
42 plint blockSize = 0;
43 Array<T,2 > xRange , yRange , zRange ;
44 plint l a y e r = margin + e x t r a L a y e r ;
45
46 // Load t h e STL f i l e and s c a l e / t r a n s l a t e it to lattice coordinates .
47 T r i a n g l e S e t <T> t r i a n g l e S e t ( s t l F i l e N a m e , DBL ) ;
48 DEFscaledMesh<T> defMesh ( t r i a n g l e S e t , 0 , 0 , margin , extraLayer ) ;
171
172 Lattice Boltzmann Source Code
49 defMesh . getMesh ( ) . computeBoundingBox ( xRange , yRange , zRange ) ;

50
51 l x = xRange [ 1 ] xRange [ 0 ] ;
52 l y = yRange [ 1 ] yRange [ 0 ] ;
53 l z = zRange [ 1 ] zRange [ 0 ] ;
54 dx = l x / (T)N ;
55 nx = ( p l i n t ) ( l x / dx + 0 . 5 ) ;
56 ny = ( p l i n t ) ( l y / dx + 0 . 5 ) ;
57 nz = ( p l i n t ) ( l z / dx + 0 . 5 ) ;
58
59 p c o u t << "The s i z e of t h e p o r o u s media is "
60 << l x << " m e t e r x "
61 << l y << " m e t e r x "
62 << l z << " m e t e r . " << s t d : : e n d l ;
63
64 p c o u t << "The r e s o l u t i o n of t h e c o m p u t a t i o n a l domain is "
65 << nx << " x " << ny << " x " << nz << s t d : : e n d l ;
66
67 l o c a t i o n = Array<T,3 >( xRange [ 0 ] , yRange [ 0 ] , zRange [ 0 ] ) ;
68 defMesh . getMesh ( ) . t r a n s l a t e ( l o c a t i o n ) ;
69 defMesh . getMesh ( ) . s c a l e ( 1 . / dx ) ;
70 defMesh . getMesh ( ) . i n f l a t e ( ) ;
71 TriangleBoundary3D<T> boundary ( defMesh ) ;
72
73 Box3D mediaDomain_LU( l a y e r , nx1+l a y e r , l a y e r , ny1+l a y e r , l a y e r , nz1+l a y e r ) ;
74 int flowType = v o x e l F l a g : : o u t s i d e ;
75 VoxelizedDomain3D<T> v o x e l i z e d D o m a i n (
76 boundary , flowType , mediaDomain_LU , borderWidth , envelopeWidth , blockSize );
77
78 M u l t i S c a l a r F i e l d 3 D f l a g M a t r i x _ t m p ( ( M u l t i B l o c k 3 D &) v o x e l i z e d D o m a i n . g e t V o x e l M a t r i x ( ) ) ;
79 s e t T o C o n s t a n t ( flagMatrix_tmp , f l a g M a t r i x _ t m p . getBoundingBox ( ) , 0);
80 s e t T o C o n s t a n t ( flagMatrix_tmp , voxelizedDomain . getVoxelMatrix ( ) ,
81 voxelFlag : : inside , f l a g M a t r i x _ t m p . getBoundingBox ( ) , 1);
82 s e t T o C o n s t a n t ( flagMatrix_tmp , voxelizedDomain . getVoxelMatrix ( ) ,
83 voxelFlag : : innerBorder , f l a g M a t r i x _ t m p . getBoundingBox ( ) , 1);
84
85 // The f u l l domain is e x t e n d e d i n zd i r e c t i o n to c r e a t e an inlet / outlet buffer zone .
86 Box3D f u l l D o m a i n ( 0 , nx 1 , 0 , ny 1 , nz / 3 , nz1+nz / 3 ) ;
87
88 s o l i d F l a g s = g e n e r a t e M u l t i S c a l a r F i e l d ( f u l l D o m a i n ) . r e l e a s e ( ) ;
89 Box3D copyDomain ( 0 , nx 1 , 0 , ny 1 , 1 , nz 2 ) ;
90 s t d : : auto_ptr<M u l t i S c a l a r F i e l d 3 D > i n v e r t e d (
91 add ( 1 , m u l t i p l y ( 1 , f l a g M a t r i x _ t m p ) ) );
92 copy ( i n v e r t e d , copyDomain , s o l i d F l a g s , copyDomain ) ;
93
94 Box3D s l i c e ( f u l l D o m a i n ) ;
95 s l i c e . x0 = s l i c e . x1 = nx / 2 ;
96 ImageWriter<T>(" l e e l o o " ) . w r i t e S c a l e d G i f (
97 " geometry " ,
98 copyConvert( e x t r a c t S u b D o m a i n ( s o l i d F l a g s , s l i c e )) );
99 ImageWriter<T>(" l e e l o o " ) . w r i t e S c a l e d G i f (
100 " inverted " ,
101 copyConvert( e x t r a c t S u b D o m a i n ( i n v e r t e d , s l i c e )) );
102 }
103 // S e t up t h e boundary c o n d i t i o n s : constant flow .
104 void flowSetup ()
105 {
106 O n L a t t i c e B o u n d a r y C o n d i t i o n 3 D <T, DESCRIPTOR> b o u n d a r y C o n d i t i o n
107 = c r e a t e L o c a l B o u n d a r y C o n d i t i o n 3 D <T, DESCRIPTOR> ( ) ;
108
109 Box3D bbox = l a t t i c e >getBoundingBox ( ) ;
110 Box3D i n l e t ( bbox ) ; i n l e t . z1 = i n l e t . z0 ;
111 Box3D o u t l e t ( bbox ) ; o u t l e t . z0 = o u t l e t . z1 ;
112
113 b o u n d a r y C o n d i t i o n >a d d V e l o c i t y B o u n d a r y 2 N ( i n l e t , l a t t i c e ) ;
114 b o u n d a r y C o n d i t i o n >a d d P r e s s u r e B o u n d a r y 2 P ( o u t l e t , l a t t i c e ) ;
115
116 i n i t i a l i z e A t E q u i l i b r i u m ( l a t t i c e , l a t t i c e >getBoundingBox ( ) , 1 . , Array<T, 3 > ( 0 . , 0 . , uLB ) );
117 setBoundaryVelocity ( l a t t i c e , inlet , Array<T, 3 > ( 0 . , 0 . , uLB ) );
118
Lattice Boltzmann Source Code 173
119 l a t t i c e > i n i t i a l i z e ( ) ;
120 delete boundaryCondition ;
121 }
122 void writeGif ( plint iter )
123 {
124 p c o u t << " W r i t i n g G i f . . . " << e n d l ;
125 const p l i n t xComponent = 0 ;
126 const p l i n t yComponent = 1 ;
127 const p l i n t zComponent = 2 ;
128
129 Box3D s l i c e ( l a t t i c e >getBoundingBox ( ) ) ;
130 s l i c e . x0 = s l i c e . x1 = nx / 2 ;
131 ImageWriter<T> i m a g e W r i t e r ( " l e e l o o " ) ;
132
133 i m a g e W r i t e r . w r i t e S c a l e d G i f ( c r e a t e F i l e N a m e ( " uz " , iter , 6) ,
134 computeVelocityComponent ( l a t t i c e , s l i c e , zComponent ) );
135 i m a g e W r i t e r . w r i t e S c a l e d G i f ( c r e a t e F i l e N a m e ( " uzNorm " , iter , 6) ,
136 computeVelocityNorm ( l a t t i c e , slice ) );
137 i m a g e W r i t e r . w r i t e S c a l e d G i f ( c r e a t e F i l e N a m e ( " omega_z " , iter , 6) ,
138 computeNorm ( c o m p u t e V o r t i c i t y (
139 computeVelocity ( l a t t i c e ) ), slice ) );
140 Box3D s l i c e x ( l a t t i c e >getBoundingBox ( ) ) ;
141 s l i c e x . y0 = s l i c e x . y1 = ny / 2 ;
142 i m a g e W r i t e r . w r i t e S c a l e d G i f ( c r e a t e F i l e N a m e ( " ux " , iter , 6) ,
143 computeVelocityComponent ( l a t t i c e , s l i c e x , xComponent ) );
144 i m a g e W r i t e r . w r i t e S c a l e d G i f ( c r e a t e F i l e N a m e ( " uxNorm " , iter , 6) ,
145 computeVelocityNorm ( l a t t i c e , slicex ) );
146 i m a g e W r i t e r . w r i t e S c a l e d G i f ( c r e a t e F i l e N a m e ( " omega_x " , iter , 6) ,
148 computeVelocity ( l a t t i c e ) ), slicex ) );
149 Box3D s l i c e y ( l a t t i c e >getBoundingBox ( ) ) ;
150 s l i c e y . z 0 = s l i c e y . z 1 = nz / 2 ;
151 i m a g e W r i t e r . w r i t e S c a l e d G i f ( c r e a t e F i l e N a m e ( " uy " , iter , 6) ,
152 computeVelocityComponent ( l a t t i c e , s l i c e y , yComponent ) );
153 i m a g e W r i t e r . w r i t e S c a l e d G i f ( c r e a t e F i l e N a m e ( " uyNorm " , iter , 6) ,
154 computeVelocityNorm ( l a t t i c e , slicey ) );
155 i m a g e W r i t e r . w r i t e S c a l e d G i f ( c r e a t e F i l e N a m e ( " omega_y " , iter , 6) ,
157 computeVelocity ( l a t t i c e ) ), slicey ) );
158 }
159 v o i d writeVTK ( p l i n t iter )
160 {
161 p c o u t << " W r i t i n g VTK . . . " << e n d l ;
162 if ( writeFullVTK ) {
163 VtkImageOutput3D<T> vtkOut ( c r e a t e F i l e N a m e ( " v t k " , iter , 6 ) , dx , location );
165 s l i c e . x0 = s l i c e . x1 = nx / 2 ;
166 vtkOut . w r i t e D a t a <f l o a t >( computeVelocityNorm ( l a t t i c e , s l i c e ) , " v e l o c i t y N o r m " , dx / d t ) ;
167 vtkOut . w r i t e D a t a <f l o a t >(add ( 1 . , c o m p u t e D e n s i t y ( l a t t i c e , s l i c e )) , " pressure " ,
168 1 . / 3 . dxdx / ( d t d t ) ) ;
169 vtkOut . w r i t e D a t a <3 , f l o a t >( c o m p u t e V e l o c i t y ( l a t t i c e , s l i c e ) , " v e l o c i t y " , dx / d t ) ;
170 vtkOut . w r i t e D a t a <3 , f l o a t >( c o m p u t e V o r t i c i t y ( c o m p u t e V e l o c i t y ( l a t t i c e , s lic e )) , " vorticity " ,
171 1 . / dt ) ;
172 }
173 VtkImageOutput3D<T> v t k O u t S l i c e ( c r e a t e F i l e N a m e ( " v t k _ s l i c e " , iter , 6 ) , dx , location );
175 s l i c e . x0 = s l i c e . x1 = nx / 2 ;
176 v t k O u t S l i c e . w r i t e D a t a <f l o a t >( computeVelocityNorm ( l a t t i c e , s l i c e ) , " v e l o c i t y N o r m " , dx / d t ) ;
177 v t k O u t S l i c e . w r i t e D a t a <f l o a t >(add ( 1 . , c o m p u t e D e n s i t y ( l a t t i c e , s l i c e )) , " pressure " ,
178 1 . / 3 . dxdx / ( d t d t ) ) ;
179 v t k O u t S l i c e . w r i t e D a t a <3 , f l o a t >( c o m p u t e V e l o c i t y ( l a t t i c e , s l i c e ) , " v e l o c i t y " , dx / d t ) ;
180 v t k O u t S l i c e . w r i t e D a t a <3 , f l o a t >( c o m p u t e V o r t i c i t y ( c o m p u t e V e l o c i t y ( l a t t i c e , s lic e )) , " vorticity " ,
181 1 . / dt ) ;
182 }
183 i n t main ( i n t argc , char argv [ ] ) {
184
185 p l b I n i t (& a r g c , &a r g v ) ;
186 g l o b a l : : d i r e c t o r i e s ( ) . s e t O u t p u t D i r ( " . / tmp / " ) ;
187
188 std : : s t r i n g stlFileName ;
174
189 try {
190 g l o b a l : : a r g v ( 1 ) . r e a d (N ) ;
191 g l o b a l : : argv ( 2 ) . read ( u I n l e t ) ;
192 g l o b a l : : argv ( 3 ) . read ( k i n e m a t i c V i s c o s i t y ) ;
193 g l o b a l : : argv ( 4 ) . read ( stlFileName ) ;
194 g l o b a l : : a r g v ( 5 ) . r e a d ( writeFullVTK ) ;
195 }
196 c a t c h ( P l b I O E x c e p t i o n& e x c e p t ) {
197 p c o u t << e x c e p t . what ( ) << s t d : : e n d l ;
198 p c o u t << "The p a r a m e t e r s for this program a r e :\n";
199 p c o u t << " 1 . Resolution .\ n ";
200 p c o u t << " 2 . Inlet Velocity [m/ s ] . \ n " ;
201 p c o u t << " 3 . Kinematic Viscosity [m^2/ s ] . \ n " ;
202 p c o u t << " 4 . STL f i l e name . \ n " ;
203 p c o u t << " 5 . S h o u l d VTK f i l e for full domain be w r i t t e n ?\ n " ;
204 p c o u t << " Example : " << ( s t d : : s t r i n g ) ( g l o b a l : : a r g v ( 0 ) ) << " 150 0 . 0 1 1 . e5 BE_Feldspat_c . s t l 1"
205 << s t d : : e n d l ;
206 exit (0);
207 }
208 p c o u t << "VOXELIZING THE DOMAIN . . . " << s t d : : e n d l << s t d : : e n d l ;
209 createDataStructure ( stlFileName ) ;
210
211 d t = uLB/ u I n l e t dx ;
212 nuLB = k i n e m a t i c V i s c o s i t y d t / ( dxdx ) ;
213 t a u = 3 . nuLB + 0 . 5 ;
214 omega = 1 . / t a u ;
215
216 p c o u t << " D i s c r e t e grid s p a c i n g : " << dx << " m e t e r s . " << s t d : : e n d l ;
217 p c o u t << " D i s c r e t e t i m e s t e p : " << d t << " s e c o n d s . " << s t d : : e n d l ;
218 p c o u t << " K i n e m a t i c v i s c o s i t y : " << k i n e m a t i c V i s c o s i t y << " m^2/ s = " << nuLB << " lattice units ."
219 << s t d : : e n d l ;
220 p c o u t << " R e l a x a t i o n t i m e t a u : " << t a u << s t d : : e n d l << s t d : : e n d l ;
221 p c o u t << "SETTING UP THE SIMULATION . . . " << s t d : : e n d l << s t d : : e n d l ;
222
223 l a t t i c e = g e n e r a t e M u l t i B l o c k L a t t i c e <T, DESCRIPTOR>( s o l i d F l a g s , 1 , new SmagorinskyBGKdynamics
224 <T, DESCRIPTOR>(omega , 0.14)). release ();
225 l a t t i c e >p e r i o d i c i t y ( ) . t o g g l e A l l ( t r u e ) ;
226 defineDynamics ( l a t t i c e , s o l i d F l a g s , l a t t i c e >getBoundingBox ( ) , new BounceBack
227 <T, DESCRIPTOR> ( 0 . 9 ) , 1);
228 flowSetup ( ) ;
229
230 p c o u t << "STARTING THE SIMULATION . . . " << s t d : : e n d l << s t d : : e n d l ;
231 T previousIterationTime = T( ) ;
232 // Loop o v e r main t i m e iteration .
233 for ( plint iT =0; iT<m a x I t e r ; ++iT ) {
234 g l o b a l : : t i m e r ( " mainLoop " ) . r e s t a r t ( ) ;
235 if ( iT%i m a g e I t e r ==0) {
236 w r i t e G i f ( iT ) ;
237 if ( iT %(5 i m a g e I t e r )==0) {
238 writeVTK ( iT ) ;
239 }
240 }
241 if ( iT%50==0) {
242 p c o u t << " s t e p " << iT
243 << " ; t=" << iT d t ;
244 p c o u t << " ; av e n e r g y ="
245 << s e t p r e c i s i o n ( 1 0 ) << g e t S t o r e d A v e r a g e E n e r g y <T>( l a t t i c e )
246 << " ; av r h o="
247 << s e t p r e c i s i o n ( 1 0 ) << g e t S t o r e d A v e r a g e D e n s i t y <T>( l a t t i c e ) << e n d l ;
248 p c o u t << "Time s p e n t d u r i n g previous iteration : "
249 <c o l l i d e A n d S t r e a m ( ) ;
253 p r e v i o u s I t e r a t i o n T i m e = g l o b a l : : t i m e r ( " mainLoop " ) . s t o p ( ) ;
254 }
255 }
Bibliography
Ahrenholz, B., Tlke, J., and Krafczyk, M. (2006). Lattice boltzmann simulations in
reconstructed parametrized porous media. International Journal of Computational
Fluid Dynamics, 20:369 ff.
Akin, S. and Kovscek, A. (2003). Applications of X-ray computed tomography in

the geosciences - Geological Society Special Publication No. 215, chapter three:
Computed tomography in petroleum research, pages 2338. Geological Society
London.
Al-Kharusi, A. and Blunt, M. (2007). Network extraction from sandstone and carbon-
ate pore space images. Journal of Petroleum Science & Engineering, 56:219231.
Arkeryd, L. (1972a). On the Boltzmann equation part I: Existence. Archive for

Rational Mechanics and Analysis, 45:116.
Arkeryd, L. (1972b). On the Boltzmann equation part II: The full initial value
problem. Archive for Rational Mechanics and Analysis, 45:1734.
Arns, C., Knackstedt, M., Pinczewski, W., and Martys, N. (2004). Virtual perme-
ametry on microtomographic images. Journal of Petroleum Science & Engineering,
45:4146.
Auset, M. and Keller, A. (2004). Pore-scale processes that control dispersion of

colloids in saturated porous media. Water Resources Research, 40.
Bakke, S. and ren, P.-E. (1997). 3-D Pore-Scale Modelling of Sandstones and Flow
Simulations in the Pore Networks. SPE Journal, 2:136149.
Batchelor, G. (1994). An Introduction to Fluid Dynamics. Cambridge University

Press, 14th edition.
Bear, J. (1988). Dynamics of Fluids in Porous Media. Dover Publications, Inc., 2nd
edition.
175
176
Bear, J. and Bachmat, Y. (1990). Introduction to Modelling of Transport Phenomena

in Porous Media. Kluwer Academic Publishers.
Benzi, R., Succi, S., and Vergassola, M. (1992). The Lattice Boltzmann Equation:
theory and applications. Physics Reports, 222(3):146 198.
Bhatnagar, P., Gross, E., and Krook, M. (1954). A model for the collision processes in
gases. I: small amplitude processes in charged and neutral one-component system.
Physical Review, 94:511525.
Blunt, M. (2001). Flow in porous media pore-network models and multiphase flow.
Colloid & Interface Science, 6:197207.
Brace, W. (1977). Permeability from resistivity and pore shape. Journal of Geo-
physical Research, 82(23).
Brle, C. (1996). Porenraumuntersuchungen ausgewhlter Rotliegend Sandstein-

proben Norddeutschlands unter besonderer Bercksichtigung der Porengeometrie.
PhD thesis, Universitt Wrzburg.
Brner, A., Bornhft, E., Hfner, F., Hug-Diegel, N., Kleeberg, K., Mandl, J.,
Nestler, A., Poschold, K., Rhling, S., Rosenberg, F., Schfer, I., Stedingk, K.,
Thum, H., Werner, W., and Wetzel, E. (2011). Steine und Erden in der Bundesre-
publik Deutschland, volume 1. Bundesanstalt fr Geowissenschaften und Rohstoffe
(BGR) in Zusammenarbeit mit den Staatlichen Geologischen Diensten der Bun-
deslnder.
Brunke, O., Brockdorf, K., Drews, S., Mller, B., Donath, T., Herzen, J., and Beck-
mann, F. (2008). Comparison between X-ray tube based and synchroton radiation
based micro CT. In Development in X-Ray Tomography VI, volume 7078. Inter-
national Society of Optics and Photonics (SPIE).
Bryant, S., King, P., and Mellor, D. (1993). Network Model evaluation of permeabil-
ity and spatial correlation in a real random sphere packing. Transport in Porous
Media, 11:5370.
Buzug, T. (2005). Einfhrung in die Computertomographie - Mathematisch-

physikalische Grundlagen der Bildrekonstruktion. Springer Verlag, 1 edition.
Carlson, W., Rowe, T., Ketcham, R., and Colbert, M. (2003). Applications of X-ray
computed tomography in the geosciences - Geological Society Special Publication
177
No. 215, chapter two: Applications of high-resolution X-ray computed tomography

in petrology, meteoritics and paleontology, pages 722. Geological Society London.
Carman, P. (1956). Flow of Gases Through Porous Media. Academic Press Interna-
tional.
Chang, Q. and Alexander, J. (2008). Application of Lattice Boltzmann Method -

Thermal Multiphase Fluid Dynamics. VDM Verlag Dr. Mller.
Chen, S. and Doolen, G. (1998). Lattice Boltzmann Method for Fluid Flows. Annual
Review of Fluid Mechanics, 30:329364.
Chen, S., Doolen, G., and Eggert, K. (1994). Lattice-Boltzmann Fluid Dynamics:
a versatile tool for multiphase and other complicated flows. Los Alamos Science,
(22):98111.
Cloete, M. (2006). Pore-scale modelling for fluid transport in 2D porous media.

Masters thesis, University of Stellenbosch.
Cressie, N. (1993). Statistics for Spatial Data, volume 1. John Wiley & Sons.
Demtrder, W. (2010). Experimentalphysik 3: Atome, Molekle und Festkrper.

Springer Verlag, 4th edition.
Dong, H. (2007). Micro-CT Imaging and Pore Network Extraction. PhD thesis,
Imperial College London.
Dullien, F. (1992). Porous Media - fluid transport and pore structure. Academic
Press Inc., 2nd revised edition.
Dunn, K., Bergman, D., and Latorraca, G. (2002). Nuclear Magnetic Resonance -
Petrophysical and Logging Applications, volume 32. Pergamnon.
Durran, D. (2010). Numerical Methods for Fluid Dynamics: with Aplications in

Geophysics. Springer Verlag, 2 edition.
Dvorkin, J., Derzhi, N., Fang, Q., Nur, A., Grader, A., Baldwin, C., Tono, H.,
and Diaz, E. (2009). From micro to reservoir scale: Permeability from digital
experiments. The Leading Edge, pages 14461453.
Engelhardt, W. and Zemann, J. (1960). Porenraum der Sedimente. Springer Verlag,

2 edition.
178
Enzmann, F. (2000). Modellierung von Porenraumgeometrien und Transport in ko-

rngesttzten porsen Medien. PhD thesis, Johannes-Gutenberg-Universitt Mainz.
Ernste, H. (2011). Angewandte Statistik in Geografie und Umweltwissenschaften,

volume 1. UTB Stuttgart.
Fauzi, U. and Bahri, A. (2005). Physical and Lattice Gas Automata Fluid Flow
Modeling in Real Porous Media. Indonesian Journal of Physics, 16(3):6770.
Frisch, U., dHumires, D., Hasslacher, B., Lallemand, P., Pomeau, Y., and Rivet, J.
(1987). Lattice gas hydrodynamics in two and three dimensions. Complex Systems,
1:649707.
Frisch, U., Hasslacher, B., and Pomeau, Y. (1986). Lattice-Gas Automata for the
Navier-Stokes Equation. Physical Review Letters, 56:1505 1508.
Gerthsen, C. and Meschede, D. (2003). Gerthsen: Physik. Springer Verlag, 22nd

edition.
Grangeat, P. (1990). Mathematical Methods in Tomography: Mathematical Frame-

work of Cone Beam 3D Reconstruction via the First Derivative of the Radon Trans-
form. Springer Verlag.
Grangeat, P. (2009). Tomography. Iste Publishing Company.
Gueguen, Y., David, C., and Darot, M. (1986). Models and time constants for
permeability evolution. Geophysical Research Letters, 13(5):460463.
Hafner, W. (1987). Obernkirchener Sandstein. Steinmetz und Bildhauer, 5:2933.
Haug, H. (2006). Statistische Physik - Glaichgewichtstheorie und Kinetik. Springer

Verlag.
Higuera, F. and Jimnez, J. (1989). Boltzmann approach to lattice gas simulations.

Europhysics Letters, 9:663668.
Jerauld, G. and Salter, S. (1990). The effects of pore-structure on hysteresis in

relative permeability and capillary preassure: pore-level modeling. Transport in
Porous Media, 5(2):103151.
Kameda, A. (2004). Permeability Evolution in Sandstone: Digital Rock Approach.

PhD thesis, Stanford University.
179
Kameda, A. and Dvorkin, J. (2004). To see a rock in a grain of sand. The Leading
Edge, pages 790792.
Keehm, Y. (2003). Computational Rock Physics: Transport Properties in Porous

Media and Applications. PhD thesis, Stanford University.
Keehm, Y., Mukerji, T., and Nur, A. (2001). Computational rock physics at the
pore scale: Transport properties and diagenesis in realistic pore geometries. The
Leading Edge, pages 180183.
Kemper, E. (1968). Einige Bemerkungen ber die Sedimentationsverhltnisse und

die fossilen Lebensspuren des Bentheimer Sandsteins (Valanginium). Geologisches
Jahrbuch, 86:89106.
Knackstedt, M., Arns, C., Limaye, A., Sakellariou, A., Senden, T., and Sheppard,
A. (2004). Digital Core Laboratory: Properties of reservoir core derived from 3D
images. In 2004 SPE Asia Pacific Conference on Integrated Modelling for Asset
Management, page 14.
Kondepudi, D. (2008). Introduction to Modern Thermodynamics. John Wiley &

Sons, 1st edition.
Kortmann, H. (1983). Sedimentologische Untersuchungen ber den Bentheimer

Sandstein (Valangin) der Umgebung von Bad Bentheim. Diplomarbeit, Univer-
sitt Kln.
Ladd, A. (1994). Numerical simulations of particulate suspensions via discretized

Boltzmann equation. Journal of Fluid Mechanics, 271:311339.
Latt, J., Malaspinas, O., Parmigiani, A., Kontaxakis, D., and Lagrava, D. (2012).
Palabos 1.0 Users Guide. FlowKit Ltd., Universite de Geneve, Switzerland.
Lee, T. and Kashyap, R. (1994). Building Skeleton Models via 3-D Medial Surface
/ Axis Thinning Algorithms. Graphical Models and Image Processing, 56(6):462
478.
Lege, T., Kolditz, O., and Zielke, W. (1996). Handbuch zur Erkundung des Un-
tergrundes von Deponien und Altlasten - Strmungs- und Transportmodellierung.
Number 2. Springer Verlag.
180
Lindquist, W., Lee, S.-M., Coker, D., Jones, K., and Spanne, P. (1996). Medial
Axis analysis of void structure in three-dimensional tomographic images of porous
media. Journal of Geophysical Research, 101:82978310.
Lorensen, W. and Cline, H. (1987). Marching cubes: A high resolution 3D surface

construction algorithm. In SIGGRAPH 87 Proceedings of the 14th annual confer-
ence on Computer graphics and interface techniques, volume 21. ACM SIGGRAPH
Computer Graphics, ACM New York.
Lowry, M. and Miller, C. (1995). Pore-scale modeling of nonwetting phase residual

in porous media. Water Resources Research, 31(3):455473.
Luo, L. (2001). Theory of the Lattice Boltzmann Method: Lattice Boltzmann Models
for non-ideal Gases. Technical Report ICASE 2001-8, National Aeronautics and
Space Administration (NASA), Hampton, Virginia (USA).
Luo, L. (2003). Lattice Boltzmann Methods for Computational Fluid Dynamics.
Martyns, N. and Garboczi, E. (1992). Length scales relating the fluid permeabil-
ity and electrical conductivity in random two-dimensional model porous media.
Physical Review B, 46:60806090.
Martys, N. and Chen, H. (1996). Simulation of multicomponent fluids in complex

three-dimensional geometries by the lattice Boltzmann method. Physical Review
E, 53:743750.
Mees, F., Swennen, R., van Geet, M., and Jacobs, P. (2003). Applications of X-ray
computed tomography in the geosciences - Geological Society Special Publication
No. 215, chapter one: Applications of X-ray computed tomography in the geo-
sciences, pages 16. Geological Society London.
Mueller, R. and Ruegsegger, P. (1994). Three dimensional finite element modeling of

non-invasively assessed trabecular bone structures. Journal of Medical Engineering
and Physics, 17:126133.
Paech, H.-P., Kampe, A., and Weyer, D. (2006). Stratigraphie von Deutschland -
Unterkarbon (Mississippium). Number 6. Deutsche Union der Geowissenschaften
- Deutsche Stratigraphische Kommission; Deutsche Union der Geologischen Wis-
senschaften - Deutsche Stratigraphische Kommission.
Potter, M. and Foss, J. (1982). Fluid Mechanics. Great Lake Press, 2nd edition.
181
Pryor, R. (2012). Multiphysics Modeling Using Comsol V4.A - First Principles Ap-
proach. Transatlantic Publishers.
Qian, Y., dHumires, D., and Lallemand, P. (1992). Lattice BGK models for Navier-
Stokes equation. Europhysics Letters, 17:479484.
Qian, Y. and Orszag, S. (1993). Lattice BGK models for the Navier-Stokes equation:
nonlinear deviation in compressible regimes. Europhysics Letters, 21:255259.
Rausch, R., Schfer, W., Therrien, R., and Wagner, C. (2005). Solute Transport Mod-
elling. An Introduction to Models and Solution Strategies. Gebrder Borntrger
Verlagsbuchhandlung, 2 edition.
Saeger, R., Scriven, L., and Davis, H. (1995). Transport processes in periodic media.
Journal of Fluid Mechanics, 299:115.
Schn, J. (1983). Petrophysik - Physikalische Eigenschaften von Gesteinen und Min-

eralen. Ferdinand Enke Verlag Stuttgart.
Schn, J. (2004). Physical Properties of Rocks - Fundamentals and Principals of

Petrophysics, volume 18. Elsevier.
Sheppard, A., Sok, R., and Averdunk, H. (2005). Improved Pore Network Extraction
Methods. In Proceedings of the International Symposium of the Society of Core
Analysts, Toronto 2005. Society of Core Analysts.
Silin, D. and Patzek, T. (2006). Pore space morphology analysis using maximal
inscribed spheres. Physica A, 371:336360.
Simpleware (2010a). simpleware - Reference Guide, Version 4.2. SIMPLEWARE

LTD., Innovation Centre, Rennes Drive, Exeter EX4 4RN, UK.
Simpleware (2010b). simpleware - Tutorial Guide, Version 4.2. SIMPLEWARE

LTD., Innovation Centre, Rennes Drive, Exeter EX4 4RN, UK.
Stadtler, A. (1998). Der Bentheimer Sandstein (Valangin, NW-Deutschland) : eine

palkologische und sequenzstratigraphische Analyse. Number 49 in Bochumer geol-
ogische und geotschnische Arbeiten. Ruhr-Universitt Bochum : Inst. fr Geologie.
Succi, S. (2001). The Lattice Boltzmann Equation for Fluid Dynamics and Beyond.
Oxford Scientific Publications.
182
Sukop, M. and Thorne Jr., D. (2010). Lattice Boltzmann Modeling - An Introduction

for Geoscientists and Engineers. Springer Verlag.
Tiab, D. and Donaldson, E. (2004). Petrophysics - Theory and Practice of Measur-

ing Reservoir Rock and Fluid Transport Properties. Elsevier - Gulf Professional
Publishing, second edition.
Tlke, J., Baldwin, C., Mu, Y., Derzhi, N., Fang, Q., Grader, A., and Dvorkin, J.
(2010). Computer simulations of fluid flow in sediment: From images to perme-
ability. The Leading Edge, pages 6874.
VSG (2011). AVIZO - Reference & Tutorial Guide, Version 7. VSG - Visualization
Sciences Group.
Wesseling, P. (1991). Principles of Computational Fluid Dynamics, volume 1 of

Springer Series in Computational Mathematics. Springer Verlag.
Wolfram, S. (1986). Cellular automata fluids I: Basic theory. Journal of Statistical

Physics, 45:471526.
Yeh, G.-T. (2000). Computational Subsurface Hydrology. Kluwer Academic Publish-

ers.
Young, P., Beresford-West, T., Coward, S., Notarberardino, B., Walker, B., and
Abdul-Aziz, A. (2008). An efficient approach to converting three-dimensional
image data into highly accurate computational models. Philosophical Transactions
of the Royal Society, 366:31553173.
Zhan, X., Schwartz, L., Toksz, M., Smith, W., and Morgan, F. (2010). Pore-
scale modeling of electrical and fluid transport in Berea sandstone. Geophysics,
75(5):135142.
Zhang, S., Della Maestra, F., Combaret, N., Klimentidis, R., Barthelemy, P., Albou,
R., and Lichau, D. (2011). The Analysis and Simulation of Rock Properties using
FIB-SEM and Virtual Material Studio. Technical report, Visualization Siences
Group (VSG) & ExxonMobil Upstream Research Company.
Zienkiewicz, O., Taylor, R., and Zhu, J. (2008). The Finite Element Method - Its
Basis and Fundamentals. Elsevier, 6 edition.

PALABOS 3D Boltzmann Method For Fluid Flow Modeling Porous Rocks

Hochgeladen von

Dokumentinformationen

Originaltitel

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

PALABOS 3D Boltzmann Method For Fluid Flow Modeling Porous Rocks

Hochgeladen von

Copyright:

Verfügbare Formate

Application and Assessment of the Lattice

Boltzmann Method for Fluid Flow Modeling in

von der Fakultt VI - Planen Bauen Umwelt

Doktor der Naturwissenschaften

Vorsitzender: Prof. Dr. Gerhard Franz

Tag der wissenschaftlichen Aussprache: 19. Mrz 2013

I also owe a great debt of gratitude to the staff of palabos, exceptionally

Application and Assessment of the Lattice Boltzmann Method for Fluid

Modeling physical properties and phenomena of natural rocks forms an important

Additionally, comparative simulations with a commonly used modeling technique

Matthias Halisch, Leibniz-Institut fr Angewandte Geophysik, Doktorarbeit 2012

Anwendung und Bewertung der Gitter Boltzmann Methode fr die Mo-

Die Modellierung physikalischer Eigenschaften und Phnomene von natrlichen Fest-

Ergnzend dazu werden vergleichende Simulationen mit gebruchlichen Software-

2 High Resolution Micro CT Imaging 7

3 Basics of Fluid Dynamics 19

4 The Lattice Boltzmann Method (LBM) 31

5 Rock Sample Characterization 49

6 The REV Challenge - a Classical and a New Approach 71

8 Pore Network Extraction 99

9 Modeling Fluid Flow with In Situ Geometries 115

10 Conclusions and Outlook 145

A Petrophysical Data 151

B Generated Meshes for CFD Modeling 161

C Lattice Boltzmann Source Code 171

2.1 Processes responsible for X-ray photon emission: Bremsstrahlung (A)

3.1 Darcys experimental set-up for determination of (rock) permeability.

4.1 Basic collisions of the FHP-I model (Luo, 2003). . . . . . . . . . . . . 34

5.1 Exemplary thin sections of the S1 specimen set. . . . . . . . . . . . . 51

A.1 Histograms of petrophysical data for the S1 specimen set. . . . . . . 157

B.1 Generated REV mesh for the benchmark sample. . . . . . . . . . . . 162

5.1 Overview of the applied petrophysical methods and sample quantities

Petrophysics forms an essential fragment within the network of applied geosciences

Therefore the accurate determination of petrophysical parameters and flow prop-

1.2 Pore Scale Fluid Flow Models

1.3 CT Imaging and Pore Network Extraction

1.4 Pore Scale Simulation Techniques

Traditional computational fluid dynamic (CFD) methods solve the conservation

1.5 Motivation and Chapter Description

Chapter 2 provides the neccessary theoretical background information on high res-

Neccessary information about basic petrophysical properties and mineralogical com-

High Resolution Micro CT

2.1 Basic Theory and Principle

2.1.1 X-ray Generation

First, by the so called Bremsstrahlung (also braking radiation or deceleration

Second, by high-energy transitions within the electron shell of atoms and

Figure 2.1: Processes responsible for X-ray photon emission: Bremsstrahlung

During X-ray generation, about 99 % of the incoming energy is transformed to heat,

Figure 2.2: Basic assembly and functionality of a X-ray tube.

2.1.2 X-ray Attenuation

I(x + x) = I(x) (x) I(x) x, (2.1)

where is called attenuation coefficient.

Figure 2.3: Exemplified X-ray attenuation through a homogeneous material.

This equation could be further simplified and finally leads to:

where the increment i reflects an individual material component with a corresponding

Figure 2.4: Exemplified X-ray attenuation through a multi-component mate-

2.1.3 Basic Principle of CT Scanning

2.1.4 Image (Spatial) Resolution and Sharpness

M = F DD/F OD. (2.5)

Figure 2.6: Schematic diagram of geometrical limits controlling image resolu-

2.2 3D Image Reconstruction

extensive filtering processes,

the back projection process and

i (t) Mij (eq)

i 1 (neq) 1 (1) (2)