Excerpt from the Proceedings of the COMSOL Conference 2007, Boston

The Simulation of Pore Scale Fluid Flow with Real World

Geometries Obtained from X-Ray Computed Tomography
Walter Fourie*1, Rajab Said2, Philippe Young3, David L. Barnes1
1
University of Alaska, Fairbanks, Department of Civil and Environmental Engineering, Water
and Environmental Research Center
2
Simpleware, Ltd. UK
3
University of Exeter, UK, School of Engineering, Computer Science and Mathematics
*Corresponding author: PO Box 751018, UAF, Fairbanks, AK, 99775, [email protected]

Abstract: Previously it has been shown become well established, however, the mere
mathematically that Darcy’s Law for flow in visualization of what goes on inside the samples
porous media can be derived from the should only be the first step in scientific
fundamental Navier-Stokes equation. With the discovery. The data retrieved from such analyses
recent improvement of X-ray Computed can now allow the researcher to match
Tomography and robust modeling tools, it is now previously unseen phenomena to preconceived
possible to show that pore scale fluid flow doctrines that may have been around for
modeled by the Navier-Stokes equation can be hundreds of years.
used to derive macro parameters of Darcy’s Law,
such as the hydraulic conductivity. The model In 1856 Henry Darcy conducted research on
allows calculation of isotropy, tortuosity and fluid flow through sand, and his results became
dispersivity of the soil in all directions. These the fundamental equations by which we calculate
parameters which are used in the structural ground water flow (Darcy, 1856). These well
descriptions of the Darcy’s phenomological Law, accepted phenomological equations, Darcy’s
can now be directly calculated and compared to Law, have been shown to have an analytical
experimental results found in the laboratory. foundation, in that it can be derived (with
This study uses a 3D model of a coarse sand assumptions) from Stoke’s Law (e.g. Bear, 1972,
scanned at 4.3 um resolution. The scan data is Ene, 1990). Thus, even at the pore scale, it
converted into high quality tetrahedral meshes should be possible to define ground water flow
using the Image-Based Meshing tools developed through the Navier-Stokes equations. The
by Simpleware Ltd whilst simulation of the fluid purpose of this paper is to show that this is
flow is carried out using the Comsol possible. The real world geometry of a coarse
Multiphysics package. sand was obtained through X-Ray Computed
Tomography and then imported into Comsol to
Keywords: Computed Tomography, Hydraulic show that the microscopic modeling of flow
Conductivity, Finite Element Modeling mirrors the macroscopic results that can be
obtained through regular laboratory analyses and
1. Introduction calculation of flow through Darcy’s Law.

The advent of non-destructive examination 2. Image-Based Mesh Generation

using 3D scanning techniques (e.g. X-Ray
Computed Tomography) has caused great 2.1 Scan and Image Processing
excitement in the scientific world. Although such
scanning techniques were initially developed for The process of constructing an accurate 3D
the medical field at a macro-level, they are volumetric representation requires a series of 2D
increasingly being adopted in various research slices (typical provided by CT scanning) of the
fields (on both macro and micro levels) and used specimen under consideration. The quality of
for qualitative and quantitative analysis (Abdul- these slices/ images has a direct impact on
Aziz et al., 2006; Watson et al., 2006; Tabor et subsequent steps in the overall procedure.
al., 2007). The investigation of physical Automated tools for noise reduction (e.g.
phenomena at the micro-scale level (e.g. micro- Curvature Flow Filter or Metal Artifact
structural behavior of natural and man-made Reduction algorithm) smoothing and contrast
materials, micro-fluid dynamics, etc.) have sharpening (e.g. Recursive Gaussian Filtering or
 Excerpt from the Proceedings of the COMSOL Conference 2007, Boston

Binarisation Filter) can be very useful. 3. Materials and Methods

Alongside other data manipulation tools such as
3D cropping and padding, re-sampling and The soil used in the study was a coarse sand
rescaling. Segmentation is the process of with a mean diameter of 0.55 mm and the scans
identifying, within the 3D matrix/image, what were conducted in a SkyScan 1172 X-Ray
pixels belong to each object of interest. The Scanner at a resolution of 4.3 μm. The two
process can be fully automated by utilizing tools dimensional reconstructions were performed
such as “threshold” - by which an object is with SkyScan’s NRecon software and were
identified by specifying minimum and maximum imported into Simpleware’s ScanIP after which
values of signal strength. Other algorithms such it was resampled to 15 μm resolution. A slice
as Floodfill and Level Set Methods can also be view of one of the reconstructed images is shown
used for automated or semi-automated in Figure 1, with the area extracted for the study
segmentation. Unfortunately, in some cases bounded by the white square. Resampling was
particularly with poor quality images/scans, less necessary to decrease the size of the final mesh
automated techniques such as the manual brush to a manageable size. At this point the sample
painting may have to be used. was 0.81 x 0.81 x 0.81 mm in size. The
background images where then thresholded to
2.2 Mesh and Finite Element Model find the available air space and the soil particles.
Generation Due to the high resolution of the scans, the
thresholding operation included extremely fine
The segmented object(s) are then simultaneously pore networks that would not significantly
meshed based on an orthotropic grid intersected attribute to fluid flow. The fine flow paths were
by interfaces defining the boundaries. In effect a erased through an erosion/dilation morphological
base Cartesian mesh of the whole volume operation in ScanIP, with a structuring element
defined by the sampling rate is tetrahedralised at of 1 pixel in the x, y and z direction. Figure 2
boundary interfaces based on cutting planes and Figure 3 shows the reconstruction of the soil
defined by interpolation points. Smooth particles before and after the resampling.
boundaries are obtained by adjusting the
interpolation points in one, or a combination, of
two ways: by setting points to reflect partial
volumes or by applying a multiple material anti-
aliasing scheme. The process results in either a
mixed tetrahedral/hexahedral mesh or a pure
tetrahedral mesh and incorporates an adaptive
meshing scheme.

The adaptive meshing scheme preserves the

topology but reduces the mesh density where
possible towards the interior of the mesh by
agglomerating hexahedra into larger hexahedra
and generating transitional tetrahedra. The
approach is fully automated and robust, creating
smooth meshes with low element distortions
regardless of the complexity of the segmented
data. The approach adopted allows for an
arbitrary number of different volumes to be Figure 1. The soil particles used for the study. The
meshed. As neighboring sub-domains share a grain particles show up as a brighter color. Water can
common cutting surface, this ensures a node to also be seen in between the particles; however for the
node correspondence at the boundaries between purposes of this study they were not included in the
different meshed volumes, thus trivially analysis. All areas not occupied by soil grains were
satisfying the geometrical constraints at the considered void space.
boundary.
 Excerpt from the Proceedings of the COMSOL Conference 2007, Boston

The Incompressible Navier-Stokes application

mode was used to model the fluid flow through
the pore space. The density and viscosity of
water was used as the fluid and a pressure
gradient of 0.1 Pa was established as the
boundary conditions in the directions of fluid
flow. The other fluid boundaries were specified
as slip/symmetry and the sand grain surfaces
were specified as no slip. A stationary solution
was calculated with the GMRES Linear System
Solver and Incomplete LU preconditioner. The
program was run on a Linux system with 2 Dual
Core AMD Opteron Processors and 8 GB of
RAM. The MKL BLAS library was used to run 2
of the processors in parallel.
Figure 2: The original soil structure before it was
resample to 15 μm resolution.
Three simulations were conducted with the fluid
flow in each principle direction, x, y and z. For
each simulation the in and outflow boundaries
were integrated with respect to the velocity
vector to find the total volume of water flowing
through the model.

To validate the solution the soil sample was

tested in the laboratory with the standard
permeability test (ASTM, 2006).

4. Results

4.1 Hydraulic Conductivity

Each of the simulations solved for ~1.5 million

degrees of freedom and took about 10 hours to
Figure 3: The soil sample after it was resampled to complete. Boundary integration with respect to
15 μm and then smoothed. velocity gave the volumetric flow, Q [m3/s], and
this was substituted into Darcy’s law to find, K
[m/s], the hydraulic conductivity. The total flow
through the in and out flow conformed to one
another around 3%. Darcy’s law is shown in
Equation 1.

Q = KA
dh
(1)
dl
with A the cross sectional area [m2], h the
pressure head difference [m] and l [m] the
sample length. The model results are shown in
Figure 5. The calculated hydraulic conductivities
and the result of the laboratory permeability test
are shown in Table 1.
Figure 4. The complex network of air voids meshed
by ScanFE and then exported to a Comsol mesh file.
 Excerpt from the Proceedings of the COMSOL Conference 2007, Boston

Figure 5: The modeling results for the fluid flow in the three principle directions (a) in the x-direction, (b) in the y-
direction and (c) in the z-direction. The brighter the colors, the higher the velocity.

Table 1. Results summary of the hydraulic

conductivities through simulation and in the Figure 6 shows a particle trace of massless
laboratory. particles through the sample super-imposed on
the soil sample. The brighter the color, the longer
the travel time of the individual particle. From
Model flow directions Hydraulic conductivity
this plot the effective length of each traveling
[cm/s] particle can be calculated and substituted into
equation 2 to find the tortuosity. For this sample
x - direction 0.036 the average tortuosity was equal to 0.505.

y - direction 0.084

z - direction 0.046

Laboratory test 0.026

4.2 Tortuosity and Dispersivity

In transport through porous media, the fluid as

well as the potential contaminants does not
follow a straight line as it flows through the
media. Instead, the fluid follows a tortuous path
around the particles, which impacts the diffusion
and the advection of the fluid and contaminants.
The tortuousity of a soil can then be defined in
terms of the effective length of the path of the Figure 6: A particle trace plot super-imposed on the
fluid, Le: soil sample. The brighter colors indicate longer travel
times.
2
⎛ L⎞
τ = ⎜⎜ ⎟⎟ (2) Due to the tortuous paths within a porous
⎝ Le ⎠ medium, a solute introduced into the flow path
will disperse in the direction of the general flow
Where τ is the tortuousity and L is the length of as well as normal to it, as flow converges and
the sample (Bear, 1972). diverges continually. To account for this
phenomenon, Bear (1972) introduced the
hydrodynamic dispersion coefficient, D, which
 Excerpt from the Proceedings of the COMSOL Conference 2007, Boston

contains terms for advective dispersion and Cattle, 2003). The Kozeny-Carman equation is
molecular dispersion, both of which are still used ubiquitously due to its ease of use,
dependent on the medium’s tortuosity. In the though several parameters have to be estimated.
laboratory or in the field, the hydrodynamic The results from the numerical model allow
dispersion coefficient is calculated from tracer direct calculation of the permeability (Bear,
tests and for this model the same idea can be 1972):
used to generate graphs from which the
dispersion coefficient can be calculated, as n3
shown in Figure 7. k = coτ (3)
M s (1 − n )
2

Where n, is the porosity, Ms is the specific

surface area and co is the Kozeny constant (the
only value that still has to be estimated). The
porosity and specific surface area are calculated
by SkyScan’s CTAn software and value of co is
assumed as 0.5. The tortuousity has been
calculated previously as 0.505 and this gives a
hydraulic conductivity, K, equal to 0.129 cm/s,
which is within one order of magnitude of the
experimental results. In Kozeny’s derivation he
neglects the velocity components normal to the
general direction of flow. However, as can be
Figure 7: Experimental results derived from the seen in Figure 6 even in this small sample
numerical model compared to the Ogata and Banks
convergence and divergence of flow occurs
(1961) solution to the advection dispersion equation,
with D = 1e-8. which decreases the hydraulic conductivity of
the soil. Hence is it is logical that the Kozeny-
Carman equation over-predicts the hydraulic
5. Discussion
conductivity.
The modeling results show hydraulic
The accepted definition of dispersivity of a soil
conductivities that are the same order of
is scale dependent, mostly because of the
magnitude in all directions. This is to be
heterogeneity thereof. The results of the
expected as the sand is a homogenous isotropic
analytical solution of Ogata and Banks shown in
medium. The modeling results also show
Figure 7 may be a reasonable approximation of
hydraulic conductivity that is the same order of
the numerical results, but unfortunately for some
magnitude as the laboratory test result. The small
things field experiments can never be replaced.
difference between the results can be attributed
Even so, as technology improves scientists and
to the relatively small sample size. In a porous
engineers are able to delve deeper into the
medium it is important to choose a sample size
physical and chemical phenomena of the natural
that reflects the physical conditions of the whole,
world and ultimately will be able to tie together
namely a representative elementary volume. In
the phenomena that are apparent at the
this case the sample was slightly too small, but
molecular, micro and macro level.
software constraints limited the size of the
sample. Even so, the results indicate that it is
possible to calculate the hydraulic conductivity, a 6. Conclusions
macro-scale parameter, by simulation on the
micro-scale. The last 20 years has shown remarkable
technological advances in both the software and
Structural explanations of Darcy’s Law was first hardware industries. Scientists and engineers are
introduced by Kozeny in 1927 (Kozeny, 1927) now able to investigate and solve problems of
and was followed by numerous others of varying ever greater complexity. It does not seem that
complexity (e.g. Fair & Hatch, 1933; Vervoort & long ago when Charbeneau stated:
 Excerpt from the Proceedings of the COMSOL Conference 2007, Boston

7. Ene, H. I., Application of the

“Because of our fundamental inability to homogenization method to transport in
describe the advection transport at the pore scale, porous media, in Dynamics of Fluids in
however, we are forced to describe this transport Hierarchical Porous Media, J.H. Cushman,
on the average sense.” (Charbeneau, 1990). Editor, Academic Press (1990)
8. Fair, G.M. and Hatch, L.P., Fundamental
That “fundamental inability” has now become a factors governing the streamline flow of
“fundamental ability”. In less then 20 years from water through sand, Journal of the American
that statement, we are now able to investigate Water Works Association, 25, 1551-1565,
pore scale phenomena and show that a (1933)
simulation with the Navier-Stokes equation 9. Kozeny, J., Über kapillare Leiting des
mirrors results from Darcy’s Law. Other Wassers in Boden, Sitzungsberichte der
parameters such as the tortuosity and dispersivity Akademie der Wissenschaften, Wien, 136,
can also be directly calculated, rather than 271-306 (1927)
estimated. 10. Ogata, A., and Banks, R.B., A solution of
the differential equation of longitudinal
In the future we will be able to examine ever dispersion in porous media, United States
larger samples whilst analyzing them at an ever Geological Survey, Proffesional Paper no
smaller scale, thus tying together the molecular, 411-A (1961)
micro and macro levels. For this to happen 11. Tabor, G., Young, P.G., Beresford West, T.,
would require not only advances in the hardware Benattayallah, A., Mesh construction from
available, but also the software. Integration of medical imaging for multiphysics
visualization software and analysis software is an simulation: Heat transfer and fluid flow in
essential step towards improved scientific complex geometries, Engineering
understanding of the natural world. Applications of Computational Fluid
Mechanics, 2, 126-135 (2007)
6. References 12. Vervoort, R.W. and Cattle, S.R., Linking
hydraulic conductivity and tortuosity
1. Abdul-Aziz, A., Saury, C., Bui Xuan, V. and parameters to pore space geometry and pore-
Young, P., On the material characterization size distribution, Journal of Hydrology, 272,
of a composite using micro CT image based 36-49 (2003)
finite element modeling, Smart Structures 13. Watson, I.G., Lee, P.D., Dashwood, R.J.,
and Materials and NDE, 26 February-2 and Young, P., Simulation of the
March 2006 San Diego. Bellingham: SPIE, mechanical properties of an aluminum
2006. matrix composite using X-ray
2. ASTM D2434-68, Standard Test Method for microtomography, Metallurgical and
Permeability of Granular Soils (Constant Materials Transactions A, 37A, pp 551-558
Head) (2006) (2006)
3. Bear, J., Dynamics of Fluid in Porous
Media, American Elsevier Publishing 9. Acknowledgements
Company, (1972)
4. Berre, C., Fok, S.L., Marsden, B.J., Babout, The lead author would like to thank the Institute
L., Hodgkins, A., Marrow, T.J., and of Northern Engineering, the Inland Northwest
Mummery, P.M., Numerical modeling of the Research Alliance, the Water and Environmental
effects of porosity changes on the Research Center, the National Institute of Water
mechanical properties of nuclear graphite, Resources and the The Alaska University
Journal of Nuclear Materials, 352, pp 1-5 Transportation Center all for their financial
(2006) support. The lead author would also like to thank
5. Charbeneau, R.J., Groundwater Hydraulics Simpleware for the use of their software and
and Pollutant Transport, Prentice Hall, their technical support.
(1990)
6. Darcy, H., Les Fontaines Publiques de la
Ville de Dijon, Victor Dalmont, (1856)