Computers and Mathematics With Applications: Feng Xiao, Xiaolong Yin

Computers and Mathematics with Applications 72 (2016) 328–348
Contents lists available at ScienceDirect
Computers and Mathematics with Applications

journal homepage: www.elsevier.com/locate/camwa
Geometry models of porous media based on Voronoi

tessellations and their porosity–permeability relations
Feng Xiao, Xiaolong Yin ∗
Petroleum Engineering Department, Colorado School of Mines, Golden, CO 80401, USA
article info abstract

Article history: In this paper, we present methods that directly model the random structure of porous
Available online 19 October 2015 media using Voronoi tessellations. Three basic structures were generated and they cor-
respond to porous medium geometries with intersecting fractures (granular), intercon-
Keywords:
nected tubes (tubular), and fibers (fibrous). Fluid flow through these models was solved by
Voronoi tessellation
a massively parallelized lattice Boltzmann code. We established the porosity–permeability
Porous media
Lattice Boltzmann method relations for these basic geometry models. It is found that, for granular and tubular geome-
Parallel computing tries, the specific surface area is a critical structural parameter that can bring their poros-
porosity–permeability relation ity–permeability relations together under a unified Kozeny–Carman equation. A connected
Heterogeneity fracture network, superimposed on the basic Voronoi structure, increases the dimension-
Fibrous media less permeability relative to the Kozeny–Carman equation; isolated large pores (vugs), on
the other hand, decreases the dimensionless permeability relative to the Kozeny–Carman
equation. The Kozeny–Carman equation, however, cannot distinguish a heterogeneous
structure with an embedded partially penetrating fracture. The porosity–permeability re-
lation for fibrous geometries in general agrees with those established for simple-cubic,
body-centered cubic, and face-centered cubic models. In the dilute limit, however, the de-
pendence on the solid fraction is weaker in Voronoi geometries, indicating weaker hydro-
dynamic interactions among randomly interconnected fibers than those in the idealized
models.
© 2015 Elsevier Ltd. All rights reserved.
1. Introduction
A porous medium is a type of two-phase disordered material, in which one phase is solid and the other phase is pore space,
which must be continuous to allow flow and transport to take place. Artificial porous media include concrete, ceramics,
paper, porous semiconductors and polymer composites; natural porous media include coral, bones, organic tissues, soil and
rocks. Hydrocarbon-bearing rocks, in particular, are important resources, from which crude oil and natural gas are extracted.
In terms of geology, a hydrocarbon bearing rock can be sandstone, carbonate or shale. Sandstone is a clastic sedimentary
rock made up mainly by sand-size (62.5 µm–2 mm) minerals that commonly are quartz and/or feldspar; the sand grains
are cemented by crystallized silicates or calcium carbonates, the pore size is usually greater than 2 µm for conventional
sandstone reservoirs and ranges from 0.03 to 2 µm for tight-gas sandstone reservoirs. Carbonate is formed by many
physicochemical and biochemical processes (e.g. cementation, bioturbation, compaction and pressure solution, mineral
replacement and recrystallization, and dolomitization) that act on terrestrial sediments mixed with organic debris such
as shells or skeletons of mollusks and brachiopods; they may contain vuggy pores that are connected through interparticle
∗ Corresponding author.
E-mail address: xyin@mines.edu (X. Yin).
http://dx.doi.org/10.1016/j.camwa.2015.09.009
0898-1221/© 2015 Elsevier Ltd. All rights reserved.
F. Xiao, X. Yin / Computers and Mathematics with Applications 72 (2016) 328–348 329
porosity [1], and the pore size can vary from 10 nm to 10 cm [2]. Shale is a very fine-grained sedimentary rock that is formed
by silt and clay-size (1–62.5 µm) minerals such as quartz and feldspar; it has a high percentage of kerogen if it is organic-
rich and may also contain carbonate and pyrite. There are microfractures, pores in kerogen, and pores in minerals for gas
shales, and the pore size can vary from a few nanometers to several hundred nanometers. These descriptions are brief for
simplicity, and cannot cover all the geometric complexities. Of these rocks, the pore structure of shales and low permeability
sandstones are still under active investigations for better characterization (e.g. [3–8]). The study by Nelson [9] provides a
good summary of pore sizes in sandstones, tight sandstones and shales.
Flow and transport of hydrocarbon in these geological porous media are long-lasting research topics since the modern
petroleum industry emerged in the 19th century. Central to this subject is the law discovered by Darcy [10] that
treats the porous media as a continuum and relates flow velocity to pressure differential by the use of permeability or
hydraulic conductivity. Today we know that the Darcy’s law results from the upscaling of slow viscous laminar flow of an
incompressible Newtonian fluid. The practical importance of Darcy’s law stimulated researchers to derive the permeability
from the knowledge of the microstructure of porous media. Although the simplest way to characterize pore structures is the
void fraction, namely porosity, there have been many other approaches that have gone to greater details. Some are purely
empirical expressions that relate permeability to an effective grain diameter (e.g. [11–13]) or to packing- and sand-shape
factors (e.g. [14]); others are based on more quantitative models of the pore structure that may be further divided into two
classes: the cell model in which the fluid phase is continuous and solid phase is discontinuous ([15,16]), and the bundle or
network of capillary tubes in which the solid phase is continuous with interconnected fluid-filled pores. The well-known
Kozeny–Carman equation [17–19] was developed based on Poiseuille flow through a bundle of capillary tubes, and it yields
permeability as a function of porosity, specific surface area and a coefficient that takes into account the shape and tortuosity
of tubes. The widely used network model treats the porous medium as a network of pore bodies of varying sizes and pore
throats of varying cross sections that connect the pores. The first network model was composed of regular networks of
single-size or varying-size tubes proposed by Fatt [20–22]. In addition to these two approaches, there are also attempts
made to extract statistical information, such as statistical correlation functions, of the pore structure for permeability
estimation [23]. However, it is only possible to derive bounds on the permeability due to the limited information from
this statistical approach. This method, therefore, is known under the name of variational bounds. de Boer et al. [24] provides
a historical review including continuum theory and void fraction theory. The reference by Adler [25] includes more details
and references for these pioneering studies.
Besides the above mentioned intuitive methods, researchers have also applied other theories to study porous media. The
first is percolation theory introduced by Broadbent and Hammersley [26], and it deals with completely random materials
without any spatial correlation. The early studies of percolating structures were carried out on random square or cubic
lattice structures (e.g. [27]); later, 3D models were generated using continuum objects placed at random or regular positions
(e.g. [28]). For simulating fluid flow in porous media, networks of tubes [29,30] are mostly used, and usually studied is the
relation between the permeability and the critical percolation threshold (e.g. [31,32]). Golden [33] provides a complete
review of percolation models for porous media, and an in-depth introduction to percolation theory and its applications for
flow in porous media is available in Hunt and Ewing [34]. The second is fractals invented by Mandelbrot [35] for geometries
that can be characterized by statistical self-similarity. Fractal dimension has been used to quantify surface roughness and
its morphology for sandstones (e.g. [36]), carbonates (e.g. [37]) and fractures [38]. Fractal geometries can also be directly
applied to simulate porous media, for example, the Navier–Stokes equation in conduits (pores) with fractal perimeter was
solved for permeability (e.g. [39,40]). The references by Adler [25] and Sahimi and Yortsos [41] have good reviews on the
application of fractals to porous media.
Numerous investigations have also been conducted to reproduce the real microstructure of porous media, and they
are named after reconstruction. Several reconstruction techniques have been developed owning to the advances made in
spatial microscopy technology, numerical simulations aided by three-dimensional visualization and applications of methods
from disciplines such as statistical physics, spatial analysis and digital image analysis. The first reconstruction technique is
direct measurement of a 3D microstructure via X-ray computed tomography (CT) (e.g. [42,43]), focused ion beam-scanning
electron microscope (FIB-SEM) with submicron resolution (e.g. [44,3,6]), or energy-dispersive spectrometry (EDS) that is also
capable of giving mineralogical compositions (e.g. [45,46]). Wildenschild and Sheppard [47] provide a thorough and latest
review of these imaging techniques. The second reconstruction technique is to generate 3D microstructures from spatial
information measured on 2D images (e.g. [48]). This topic has been extensively studied largely due to the convenient and
inexpensive access to high-resolution 2D images such as thin sections from simple optical techniques or cross section images
from other advanced imaging techniques. The process consists of two steps: one step is to measure structural information
on 2D images such as the commonly used porosity and statistical correlation functions; the other step is to numerically
generate 3D images to match those measures using threshold Gaussian random fields (e.g. [49,50]), Boolean or germ-grain
model (e.g. [51]) and simulated annealing (e.g. [52–56]), and recently introduced Markov random field based method [57].
However, these reconstructed models from 2D images are still not the geometrical (pore size and shape) and topological
(pore connectivity) equivalents of the original sample [58,59], although they are equivalent in terms of the applied statistical
measures, they may generate poor connectivity when dealing with low porosity. Along the line of imaging and statistical
measures, recently there is an interesting work that a 0.4×0.4×17 m3 of karst carbonate was reconstructed with a resolution
of 5 mm from borehole imagery and caliper data [60]. The third reconstruction technique is to model the geological process
by which the porous medium was made [61–63]. By simulating the processes of sedimentation, compaction and cementation
330 F. Xiao, X. Yin / Computers and Mathematics with Applications 72 (2016) 328–348
starting from sphere packing, it is possible to reproduce the pore connectivity for homogeneous sandstones. However, it is
still difficult to model geological processes for complex rocks like carbonates and shales, for which the geological history
involves a large number of complex physicochemical processes, multi-scale pore structures, highly irregular or very fine
grains.
Along with the efforts in reconstruction is the development of quantitative measures that characterize a porous
medium geometry. These measures provide quantitative information on how similar the reconstructed geometry is to the
sample geometry. Early works in this area include the use of porosity, specific surface area, tortuosity and pore/grain size
[13,64,25]. These measures are still widely used today because of their accessibility through experiments and the popularity
of the Kozeny–Carman equation for permeability calculation. As previously mentioned, statistical methods based on spatial
and image analyses used to study the structure of disordered multiphase materials have been introduced to study porous
media. Among the various statistical measures, the two-point correlation functions are the most commonly used owing
to their versatility in characterizing all kinds of two-phase disordered materials and their accessibility from 2D images
before having to resort to 3D imaging data sets. Torquato [65] provided a good summary of various correlation functions
including n-point functions, linear-path functions and chord-length distribution functions. Other statistical methods include
linear or spherical contact distributions [66,67] and local porosity distribution and local percolation probability [68,69].
Among others, integral geometry [70] is a promising method to measure spatial structures. The main measures are the
four Minkowski functionals that respectively describe volume, surface area, curvature and connectivity, and they can easily
be computed from local properties of porous media [71,72]. Recently the Minkowski functionals have been extended to
tensorial quantities [73], which are able to quantify anisotropy and higher order surface structure.
While more and more insights of porous medium have been obtained from analysis of the geometries of specific samples,
researchers are still faced with several challenges. Different from artificial and biological porous media, geological porous
media are more complex not only due to difficulties in characterizing the complex microstructure of a specific sample in
laboratories but also due to the very small sampling size compared to the heterogeneities that are exhibited on all length
scales. A submicron resolution image that has captured some microstructural features of carbonates or shales is usually
smaller than one millimeter in size and is not able to reflect the variations on larger length scales. For this reason, most
modeling studies were made on the relatively simple and homogeneous sandstones. There are only limited studies on
modeling of vuggy carbonates [74–76,51,77] and there are none on shales. In terms of the correlation between porous media
geometry and flow/transport, although the Kozeny–Carman equation has been developed into various forms with fitting
coefficients [78–80], it is still intended for homogeneous, isotropic and granular porous media with moderate porosities;
with porosity and specific surface as the only parameters, it fails to predict permeability of complex porous media [81,82].
This reveals the engineering reality that additional measures of geological porous media geometry need to be sought and
correlated to flow and transport properties.
In this work, geometries based on Voronoi tessellation are used to build simple random pore structures. Using lattice
Boltzmann simulations, we characterized the porosity–permeability relations of these geometries, and studied the effect of
geometric complexity, such as fractures and vugs, on the porosity–permeability relation. A Voronoi tessellation is a specific
decomposition of a metric space based on a given set of points [83]. It subdivides a 2D or 3D space into convex polygons (2D)
or polyhedra (3D) by connecting the bisecting lines or planes that are between pairs of points. When the initial points are
given randomly, Voronoi tessellation gives a random division of space. Voronoi tessellations can be found in a large number
of applications [84]. When it comes to modeling of porous media, Voronoi models made of cells, networks, and confined
spheres have all been previously used. Winterfeld et al. [85] and Jerauld et al. [86,87] studied the percolation properties of
2D and 3D Voronoi networks. Vrettos et al. [88,89] used analytical methods to study gas transport properties of cylindrical
pores derived from 2D and 3D Voronoi network models. Arns et al. [59,90] applied Minkowski functionals to characterize
3D Voronoi network models and cell models, and the 3D voxelated images were generated by thickening the edges or facets
of Voronoi cells. Schröder-Turk et al. [73] used Minkowski tensors to characterize 3D sphere packing confined by a Voronoi
tessellation. In the literature, Voronoi tessellations were mostly used in their simple forms, namely edges and confined
spheres, although Arns et al. [59,90] stand as exceptions in which voxelized pore structures are created. Few works have
been conducted to study the permeability of pore structures modeled by Voronoi tessellations.
This paper is organized as follows: first, we present the method to generate basic isotropic pore structures from
Voronoi tessellations and the additional complexities that can be analytically introduced, such as anisotropy, vuggy porosity,
fractures, and pore size distribution. Then, after a brief introduction of the lattice Boltzmann method, its parallel computing
performance and validations, we present the porosity–permeability relation of three Voronoi geometry models and how
the presence of fractures or vugs affects the porosity–permeability relation.
2. Porous media geometry
2.1. Two-dimensional (2D) geometry models
Fig. 1 is an illustration of the generation of an isotropic, homogeneous, and periodic 2D geometry model in a square
domain. The process starts with a set of random points (Poisson points) that are distributed in the center of the domain
boxed by the red square. To enforce periodicity, these points were shifted to the four nearest neighbor domains in the x and
Fig. 1. Illustration of 2D homogeneous and isotropic porous media geometry generation. Left: Random points in the center domain and its images in the
neighbor domains; Middle: Periodic tessellation in the center domain; Right: The edges of the Voronoi tessellation are moved toward the centers of the
polygons to create a pore space in the form of an open-pore network.
y directions and the next four nearest neighbor domains in the diagonal directions. The Voronoi tessellation generated using
the original and the shifted points, as exemplified in Fig. 1, is strictly periodic in the center domain. Pore space between the
polygonal ‘‘grains’’ was created by applying analytical calculations to shift edges of a polygon toward its center. Then, the
intersection points of the shifted edges, determined analytically, form the vertices of the new polygonal ‘‘grain’’. Repeating
this calculation for all the polygons in the Voronoi tessellation leads to a pore network. The aperture of the pores (channels)
can be prescribed to vary according to a size distribution or as a function of spatial location. These geometries do not
contain any dead-end pores. They do not reflect the 3D connectivity of real porous media. However, they can be used as
computationally affordable alternatives to 3D geometry models to study the qualitative effect of pore geometry on flow and
transport [91,92]. In addition, these 2D textures can be directly molded into Polydimethylsiloxane (PDMS) and bounded to
glass to build high-precision microfluidic porous media micromodels [93,94].
2.2. Three-dimensional (3D) geometry models
3D periodic porous media geometry models were constructed from 3D Voronoi tessellations using analytical processes
that are similar to those for the 2D models except that the pore space can take two different conformations. When tangent
planes were used to reduce the bodies of polyhedra, pore space is created in the form of fractures between polyhedral
‘‘grains’’. We refer to this kind of geometry as granular. When the pore space is generated by assigning radii to the edges of
the 3D Voronoi cells, the resulted geometry is essentially a set of randomly inter-connected tubes. We thus refer to this kind
of geometry as tubular. These two geometry models, respectively, fall into the classical categories of cell model and network
model, which are the two extremes of random homogeneous porous media. The former is formed by a continuous fluid phase
and a discontinuous solid phase, and the latter is bicontinuous, with interconnected fluid-filled pores and a continuous solid
phase. Note that the latter model, when inverted, becomes a model for fibrous/foamy porous media such as extracellular
matrix in biology [95,96] and metal foams [97]. Fig. 2 shows an example of the granular porous medium model and an
example of the tubular porous medium model. Both were periodic, homogeneous, and isotropic and were created in a cubic
domain.
By varying the density of the initial random points and the size of the fractures or tubes, the porosity of the models
can be varied over a wide range. These analytical geometry models can be easily modified to introduce various features.
For example, the apertures of fractures in the granular model, as well as the radii of tubes in the tubular model, can vary
according to a size distribution or as a function of spatial location. For the tubular model, at the joints of cylinders, spheres can
be added to simulate a pore network with pore bodies connected by pore throats. The positions of the initial Poisson points
may be varied according to a density distribution. The orientation of the polyhedra in the granular geometry model may be
rotated to create more realistic pore shapes and grain–grain contacts. The fractures or tubes may be selectively removed to
control coordination number and pore connectivity. The granular and tubular models may be compacted along one or two
dimensions to simulate anisotropy, which is an important feature of geological porous media. By removing certain grains in
a granular geometry or introducing large spheres into a tubular network, the vuggy nature of carbonates can be qualitatively
represented.
Figs. 3 through 5 show the various geometries that can be obtained starting from the basic granular and tubular models.
Fig. 3 shows a geometry with a fracture aperture distribution and a geometry with a pore throat and pore body size
distribution. Pore size distribution is an important characteristic of sedimentary rocks, and it is often times available as a
log-normal distribution from petrophysics studies such as mercury injection, gas adsorption, or image analysis. Fig. 4 shows
examples of anisotropic but homogeneous porous media geometries. These geometries were generated by first creating
isotropic Voronoi tessellations in a domain with an aspect ratio of 1:1:3. Then, the Voronoi cells were scaled by a factor of
1/3 in the z direction, which leads to an anisotropy in the orientation statistics of the connections. Finally, the faces (for the
granular geometry) or the edges (for the tubular geometry) were assigned with an aperture width (granular) or diameter
(tubular) to create the pore space. Note that the anisotropy can also be introduced and controlled by assigning different
Fig. 2. 3D Voronoi-based geometry models: Left: A granular geometry model; Right: A tubular/fibrous geometry model.
Fig. 3. Granular and tubular geometries with aperture or tube size distributions.
aperture or tube size according to their orientations. The degree of anisotropy, which is measured by the permeability tensor,
can be tuned to match the experimental measurements. Pore-scale heterogeneity in the form of large vugs is common in
poorly sorted sandstones and carbonates where chemical dissolution has played significant roles in the development of
the pore system. Fig. 5 shows examples of pore-scale vuggy heterogeneity models. For a granular geometry, prescribed or
randomly selected grains can be removed to create vugs; for a tubular geometry, prescribed or randomly selected cylinder
joints can be assigned with spheres of larger diameters to create vugs. In the left panel of Fig. 5, the removed grains,
highlighted in red, generate a vuggy porosity of 5%. In the right panel of Fig. 5, by adding spheres of a uniform size to
randomly selected cylinder joints, a vuggy porosity of 4% was created. Clearly, these analytical operations applied to the
basic Voronoi tessellation can simulate any kind of heterogeneities seen in complex geological porous media.
As both granular and tubular geometries are comprised of simple shapes, such as polyhedra, cylinders and spheres,
analytical calculations based on computational geometry can be performed not only to control their shapes but also to
superimpose them through Boolean operators. For example, the pore space can be defined as a location that is either in a
granular pore system or in a set of fractures. Fig. 6 shows a composite pore system created this way. The aperture size of the
fractures is 5 times that of the matrix. This pore system is an analog to a fractured porous medium, which is an important
class of geological porous media. Fractures created by local stresses originated from plate tectonics, geothermal expansion,
Fig. 4. Anisotropic granular and tubular geometries.
Fig. 5. Granular and tubular geometries with vugs.
or kerogen maturation, when connected, provide preferential flow paths. The interaction between fractures and the porous
matrix that they cut through is the subject of many fundamental investigations.
For the basic homogeneous and isotropic granular and tubular geometries as shown in Fig. 2, the porosity is well
correlated to the dimensionless seeding density of the Poisson points. The dimensionless seeding density is defined as
N / (L/w)3 for the granular geometries or N / (L/d)3 for tubular geometries, where N is the number of initial Poisson points,
L is the cube size, w is the fracture aperture and d is the diameter of tubes. The relation of porosity to dimensionless seeding
density is presented in Fig. 7.
3. Parallelized lattice Boltzmann method
In this study, a parallelized lattice Boltzmann simulator is used to simulate single-phase fluid flow through the porous
media models. Different from the traditional computation fluid dynamics methods such as finite difference, finite element,
or finite volume, the lattice Boltzmann method solves the Navier–Stokes equation indirectly by simulating the evolution
of a simplified fluid molecular velocity distribution, the moments of which recover the Navier–Stokes equation. It handles
no-slip boundaries using a simple bounce-back scheme, which, due to its simplicity, is very attractive for fluid-dynamics
problems with complex geometries. Today, it has been successfully applied to a large number of fluid-dynamics problems,
Fig. 6. A synthetic fracture system and a fractured granular geometry.
Fig. 7. Relation of porosity to the dimensionless seeding density for basic granular and tubular geometries.
including single-phase flow, multiphase/multicomponent flow, reactive flow, particulate suspensions, compressible flow,
turbulent flow, thermal flow, viscoelastic flow and micro flow. Reviews of this method and its applications are available in
the literature [98–100].
The lattice Boltzmann method consists of two processes: ‘‘particle’’ propagation and ‘‘particle’’ collision. A particle is
defined as a collection of molecules that travel with the same velocity on a certain lattice. The transport equation for the
particle is
fα rj + cα 1t , t + 1t = fα rj , t + Ωa (f ) + 1tFα
   
(1)
where fα (rj , t ) is the vector of the particle velocity distribution function, the fluid nodes (rj ) are defined on a space-filling
cubic lattice, 1t is the time step that is taken as 1 in this study, α represents propagation pathways corresponding to cα
that is also called the set of discretized velocities, Ωa (f ) is an operator that yields the effect of the particle collision, Fα
is the discretized force term corresponding to each population in the distribution function. The post-collision population
fα rj , t + Ωa (f ) + 1tFα is then advected to location rj + cα 1t. The discretized force in Eq. (1) is given by

  
F · cα C : cα cα − cs2 I
Fα = w α + (2)
cs2 2cs4
√
where F is the external force, cs is the lattice speed of sound which equals 1/ 31x/1t, where 1x is the lattice grid spacing,
and C = (uF + Fu) [101]. wα is a set of weighting coefficients associated with the discretized velocities.
In the lattice Boltzmann method used in this study, the set of discrete velocities cα follows the standard D3Q19 scheme;
D3Q19 stands for three dimensions and 19 velocity quadratures:
(0, 0, 0) α=0

cα = (±1, 0, 0), (0, ±1, 0), (0, 0, ±1) α = 1, 2, . . . , 6 (3)
(±1, ±1, 0), (±1, 0, ±1), (0, ±1, ±1) α = 7, 8, . . . , 18.
The associated weighting coefficients are
1/3

 cα2 = 0
wα = 1/18 cα2 = 1 (4)
1/36 cα = 2.

 2
Compared with other choices such as D3Q15 and D3Q27, D3Q19 achieves a balance between computational reliability and
efficiency [102]. As to the boundary condition during particle propagation, the link-bounce-back scheme [103] is used to
recover the no-slip boundary condition, among many others [104,105]. This bounce-back scheme is first-order accurate
in the lattice spacing, and it uses a stair-case approximation to model a curved surface. However, compared with high-
order boundary conditions such as Bouzidi et al. [106], Ginzburg and d’Humieres [107], Latt et al. [108], Mei et al. [104], the
link-bounce-back scheme is computationally simple and robust. In the next section, a detailed resolution analysis will be
performed to assess the errors associated with boundary discretization in the link-bounce-back scheme.
Particle collision plays a central role in the lattice Boltzmann method. A collision scheme must satisfy the conservation
of mass and momentum, and also describes the relaxation process toward stress equilibrium, which, macroscopically,
yields the viscosity and constitutive equation of the fluid. The most common and the simplest collision model is the
Bhatnagar–Gross–Krook scheme (BGK, [109])
1 
Ωa (f ) = − fα rj , t − fαeq
 
(5)
τ
 
9 3
fαeq = wα ρ 1 + 3 (cα · u) + (cα · u)2 − u2 (6)
2 2
where τ is the relaxation time, and fαeq is the equilibrium distribution. Based on the gas kinetic theory, for isothermal systems,
the microscopic local density (ρ ) and local momentum (ρ u) and local Euler stress (ρ uu+pI ) at each node are calculated as
follows
 
ρ= fαeq = fα (7)
 
ρu = eq
cα fα = cα f α (8)
 
ρ uu+pI = cα c α fαeq = cα cα f α (9)
where p = ρ cs2 . BGK scheme is a Single-Relaxation-Time (SRT) collision model, and researchers have also developed
Multiple-Relaxation-Time (MRT, [110]) and Two-Relaxation-Time (TRT, [111]) collision methods. MRT introduces multiple
relaxation parameters to control the rates of relaxation of various moments of particle distribution. It has advantages over
the BGK for its stability and accuracy [112], although it is more computationally intensive than BGK. TRT is a simplified
expression of MRT that retains the simplicity of BGK and achieves the stability of MRT. In this study, MRT is used and it is
implemented by introducing a matrix in the transport equation
fα rj + cα δ t , t + δ t = fα rj , t − M−1 · S · M · fα rj , t − fαeq
       
(10)
  M
where  is a 19 × 19 integer transformation matrix for D3Q19 propagation scheme and it is used to transform
fα rj , t − fαeq into the moment space, S is the collision matrix in the moment space and it is a diagonal matrix with the
relaxation rates and M−1 is used to transform the post-collision terms out of the moment space. For detailed configurations of
M and S, the readers are referred to d’Humieres [110]. This collision scheme guarantees mass and momentum conservation,
but does not consider energy conservation, which is acceptable for an isothermal system.
A few cautions are needed to simulate incompressible laminar flows. This lattice Boltzmann method is second-order
accurate, and the compressibility error is proportional to the square of the Mach number Ma = u/cs . In this study, Ma is
kept below 0.1 in all simulation to simulate an essentially incompressible fluid. To ensure laminar flows with small Reynolds
number, the body force density which is equivalent to a macroscopic pressure gradient was kept small or the viscosity was
set to high values by adjusting the relaxation time in MRT.
Massive parallel computing on a distributed-memory cluster is critical for simulating three-dimensional porous media
flows of meaningful sizes, which often contain 106 –109 lattice nodes. In terms of parallelization, compared with traditional
Fig. 8. Parallel computing efficiency on Mira (left) and Titan (right).
computational fluid dynamics methods, LBM holds superiority in two aspects: (1) discretization of complex geometries
using simple structured lattices, and (2) the inherent locality of its scheme. In this study, we developed a parallel lattice
Boltzmann code using MPI (Message Passing Interface), a two-lattice implementation [113], a vectorized data structure
similar to a sparse matrix (e.g., [114]), an orthogonal recursive partitioning (e.g., [115]), and a collective communication
strategy. Performance tests were performed on Mira and Titan, two supercomputers that are representative of the major
and state-of-art architectures of high performance computing. Mira is located at the Argonne National Laboratory (IBM Blue
Gene/Q, 16 GB RAM and 16 1.6 GHz PowerPC A2 cores, 32 MB eDRAM ascent as a shared L2 cache to all the cores) and
Titan is located at the Oak Ridge National Laboratory (Cray XK7, 32 GB RAM and one 16-core 2.2 GHz AMD Opteron 6274
processor in each node, 8 × 2 MB shared L2 cache and 2 × 8 MB shared L3 cache for all cores). Titan and Mira are listed as
the second largest and the fifth largest supercomputers in the world as of June 2014 (http://www.top500.org). The tested
porous medium geometries are cubic arrays of solid spheres, and the size varies from 3203 1x3 to 25603 1x3 . Each solid
sphere is 91x in radius and is confined in a 203 1x3 elemental cube. The use of cubic arrays allows domain partitioning to
be conducted based on elemental cubes, which provides a perfect balance of workload and communication. This enables a
quantitative assessment of the performance of the communication algorithm.
As shown in Fig. 8, the code was tested using up to 262,144 cores of Mira, and 65,536 cores of Titan. The parallel computing
efficiency is defined as EP = (tN /N )/t1 , where t1 is the execution time by one core and tN is the execution time by N cores.
With the same geometry and the same number of cores, the code on Mira demonstrates a much better performance than on
Titan, as shown by the 25603 1x3 geometry on 65,536 cores. This can be attributed to the faster cores on Titan, the Pmax /core
(peak performance per core) of which is faster than that of Mira by a factor of 4. The code achieves 90% efficiency with
the 25603 geometry on 262,144 cores on Mira, and 81% efficiency with the 12803 geometry on 16,384 cores on Titan. For
a smaller number (<10,000) of cores, the decrease of efficiency reflects the effect of increasing ratio of message size to
workload, as shown by the tests of the 3203 1x3 geometry on both Mira and Titan. For a very large number (>30,000) of
cores, it reflects the effect of a very large number of messages sent by each core, and this communication overhead (decrease
of efficiency) becomes more obvious on Titan that has faster cores. Once the memory on each computing node is maximally
utilized to increase the workload, the communication overhead of this code is negligible on tens of thousands of cores. Thus,
this code can achieve a near ideal parallel speedup/efficiency on most clusters.
4. Validation in simple geometries
Poiseuille flows in a cylindrical tube and between two parallel plates were simulated and then validated against the
analytical solutions for incompressible laminar flows of a Newtonian fluid with no-slip boundary condition. Examples of
channel and tube geometries in voxel representations are shown in Fig. 9. In this section, the term resolution is defined as
the number of lattice nodes (1x) used to resolve the width of a fracture or the diameter of a tube. All simulations were
performed with a body force of 0.0001 (simulation units) and a viscosity of 6 (simulation units) to ensure laminar flow,
except that when the resolution is higher than 80 1x, where a higher viscosity of 38 (simulation units) is needed to maintain
the Stokes flow condition and to accelerate convergence toward the steady state. In all simulations, the fluid was initially
stationary and was gradually accelerated by the applied body force to reach the steady state.
As shown in Fig. 10, errors systematically decrease with increasing resolution. For flow between parallel plates, error
is reduced to below 1% at the resolution of eight. The errors for flows in tubes were all negative, due to the ‘‘roughness’’
Fig. 9. A horizontal tube, an inclined tube, and two parallel plates for Poiseuille flow simulations.
Fig. 10. Validation—flow between parallel plates and through cylindrical tubes.
that is inherent in the staircase representation of the geometry. For the tubes, rather high resolutions are needed to reduce
the errors to below 2% (20 for the horizontal tube and 40 for the inclined tube). Note that the trends in the errors can be
non-monotonic with the staircase representation of the geometries. In the D3Q19 propagation scheme, it is known that the
accuracy is improved when a surface follows the lattice directions [116].
5. Random geometries—lattice resolution and system size
When reporting numerically obtained permeability, one must assure that (1) the geometric features are sufficiently
resolved and the permeability is independent of grid resolution, and (2) the porous media is sufficiently large in dimension to
form an REV (Representative Elementary Volume) for permeability. In this section, numerical simulations were conducted
with various resolutions of the same granular, tubular and fibrous geometries to establish the optimal lattice resolution
for single-phase flow simulations in the Darcy/Stokes flow regime. We also conducted system size/aspect ratio tests to
determine the minimal number of Poisson points and system aspect ratio needed to establish statistically representative
permeability.
Tables 1 through 3 summarize the test results on the effect of lattice resolution. In these simulations, we used the
granular and tubular geometries shown in Fig. 2. The reported permeability was normalized by the fracture aperture or
the tube/cylinder diameter of the geometries, and the relative error was assessed with respect to the geometry with the
highest resolution. In these tables, it can be observed that the porosity quickly stabilized with increasing lattice resolution.
For the granular geometry, the relative error became less than 2% when aperture size became 10 1x; for tubular geometries,
interestingly, the relative errors were smaller than those of granular geometries at the same resolution, which is opposite to
the trends in Fig. 10. Additionally, these relative errors were also less than those of the horizontal and the inclined tubes. This
is, perhaps, an indication that errors due to application of the bounce-back over inclined surfaces of different orientations
cancel each other. With these data, we show that it is acceptable to use a relatively low resolution (6 1x with a relative error
of 1%) to resolve the tube diameter in the Darcy/Stokes flow regime. For the fibrous geometry generated from the inversion
of the tubular geometry, the relative errors also became negligible as the lattice resolution exceeded 6 1x. Based on the
relative errors presented in Tables 1 through 3, the resolution of 6 1x is used to resolve the minimum fracture aperture as
well as the minimum tube/fiber diameter in this study.
In the system size tests, we maintained the grid resolution (6 1x) and the seeding density of the Poisson points (five per
2003 of system size), but systematically increased the system size and varied the aspect ratio. For each system size/aspect
ratio, five sample geometries were generated. Table 4 shows the porosity and the dimensionless permeability, and the
Table 1
Effect of lattice resolution on the permeability of a granular geometry. Both the system size
and the fracture width w are in the unit of 1x.
Geometry System size w φ kD = k/w 2 εk = kD /kD,r − 1
Granular R1 60 × 60 × 60 2 29.98% 1.972E−02 1.4%
Granular R2 120 × 120 × 120 4 30.00% 1.884E−02 −3.1%
Granular R3 180 × 180 × 180 6 30.00% 1.890E−02 −2.8%
Granular R4 240 × 240 × 240 8 30.00% 1.901E−02 −2.2%
Granular R5 300 × 300 × 300 10 30.00% 1.912E−02 −1.7%
Granular R6 360 × 360 × 360 12 30.00% 1.920E−02 −1.3%
Granular R7 420 × 420 × 420 14 30.00% 1.926E−02 −1.0%
Granular R8 480 × 480 × 480 16 30.00% 1.932E−02 −0.7%
Granular R9 540 × 540 × 540 18 30.00% 1.936E−02 −0.5%
Granular R10 600 × 600 × 600 20 30.00% 1.940E−02 −0.3%
Granular R11 720 × 720 × 720 24 30.00% 1.945E−02 Reference
Number of Poisson points N = 40 for all geometries.
Table 2
Effect of lattice resolution on the permeability of a tubular geometry. Both the system size
and the tube diameter d are in the unit of lattice spacing 1x.
Geometry System size d φ kD = k/d2 εk = kD /kD,r − 1
Tubular R1 30 × 30 × 30 2 14.83% 2.088E−03 −0.8%
Tubular R2 60 × 60 × 60 4 14.92% 2.081E−03 −1.1%
Tubular R3 90 × 90 × 90 6 14.91% 2.083E−03 −1.0%
Tubular R4 120 × 120 × 120 8 14.91% 2.087E−03 −0.9%
Tubular R5 150 × 150 × 150 10 14.91% 2.085E−03 −0.9%
Tubular R6 300 × 300 × 300 20 14.91% 2.091E−03 −0.7%
Tubular R7 210 × 210 × 210 14 14.91% 2.092E−03 −0.6%
Tubular R8 240 × 240 × 240 16 14.91% 2.096E−03 −0.4%
Tubular R9 270 × 270 × 270 18 14.91% 2.099E−03 −0.3%
Tubular R10 300 × 300 × 300 20 14.91% 2.101E−03 −0.2%
Tubular R11 360 × 360 × 360 24 14.91% 2.105E−03 Reference
Table 3
Effect of lattice resolution on the permeability of a fibrous geometry. Both the system size
and the fiber diameter d are in the unit of lattice spacing 1x.
Geometry System size d φ kD = k/d2 εk = kD /kD,r − 1
Fibrous R1 30 × 30 × 30 2 85.17% 3.699E−01 14.5%
Fibrous R2 60 × 60 × 60 4 85.08% 3.307E−01 2.4%
Fibrous R3 90 × 90 × 90 6 85.09% 3.248E−01 0.5%
Fibrous R4 120 × 120 × 120 8 85.09% 3.232E−01 0.1%
Fibrous R5 150 × 150 × 150 10 85.09% 3.229E−01 0.0%
Fibrous R6 180 × 180 × 180 12 85.09% 3.229E−01 0.0%
Fibrous R7 210 × 210 × 210 14 85.09% 3.230E−01 0.0%
Fibrous R8 240 × 240 × 240 16 85.09% 3.230E−01 0.0%
Fibrous R9 270 × 270 × 270 18 85.09% 3.230E−01 Reference
coefficient of variance (ratio of standard deviation to mean) from the five samples. As shown in Table 4, for 40 or more Poisson
points, the coefficient of variance for both porosity and dimensionless permeability is smaller than 1%. The aspect ratio does
not affect the porosity and permeability, as shown by Granular S7. Since all stochastic geometries were generated from
Voronoi tessellation, the tubular and fibrous geometries should have a similar statistical behavior. Thus, for permeability
studies, N = 40 should lead to statistically representative porous media models.
6. Porosity–permeability relations
Several researchers have studied the percolating properties and the topological characteristics of porous medium
geometries represented by networks constructed from Voronoi tessellations [85–89,59,73], but few of them performed
direct numerical simulations to characterize the permeability. This section presents porosity–permeability relations for
random granular geometries with uniform fracture aperture and random tubular and fibrous geometries with uniform
tube/fiber diameter. These porous media are characterized by a single length scale. Therefore their porosity–permeability
relations, once properly non-dimensionalized, should only depend on the porosity. Such porosity–permeability relations
have not been characterized. Then, we present porosity–permeability data of heterogeneous vuggy and fractured geometries
to show the effect of pore structure on porosity–permeability relations. Comparisons will also be made between the
Table 4
Effect of system size, aspect ratio, and sample variation on permeability of granular geometries. CV stands for coefficient of variation.
Geometry System size Number of Poisson Mean porosity, φ Mean kD = k/w 2 CV of kD CV of
points, N φ
Granular S1 200 × 200 × 200 5 14.6% 4.78E−06 6.7% 1.9%
Granular S2 301 × 301 × 301 17 14.4% 4.78E−06 1.3% 1.6%
Granular S3 400 × 400 × 400 40 14.2% 4.63E−06 0.6% 0.9%
Granular S4 800 × 400 × 400 80 14.2% 4.65E−06 0.6% 0.5%
Granular S5 600 × 600 × 600 135 14.3% 4.67E−06 0.8% 0.4%
Granular S6 700 × 700 × 700 214 14.2% 4.67E−06 1.0% 0.8%
Granular S7 3200 × 400 × 400 320 14.2% 4.65E−06 0.5% 0.4%
Fracture aperture w = 6 for all geometries.
Fig. 11. Porosity–permeability relations of homogeneous and isotropic granular and tubular geometries.
simulation data and the Kozeny–Carman equation. In this section, all flows were driven by a low pressure gradient (0.0001
in lattice units) and the pore fluid has a high viscosity (6 in lattice units), such that the flow is in the Stokes/Darcy regime.
The simulation domains are all cubic in shape.
6.1. Homogeneous granular and tubular geometries
Figs. 11 and 12 show the simulation results obtained from homogeneous and isotropic tubular and granular geometries
for a wide range of porosities from 0.02 to 0.40. The permeability is reported in three dimensionless forms, k/w 2 for granular
geometries, k/d2 for tubular geometries, and ks2 that is applicable to both types of geometries. Here, w is the fracture
aperture, d is the cylinder diameter, and s is the specific surface area per unit solid volume.
Fig. 11 shows k/w 2 and k/d2 as functions of porosity φ . It is seen that granular geometries have higher dimensionless
permeability than tubular geometries of the same porosity.
Fig. 12 shows that the specific surface area is an important length scale that is capable of giving a consolidated
porosity–permeability relation applicable to both types of geometries. Specific surface area is an experimentally measurable
quantity (through gas adsorption) and an important geometric measure and length scale of a porous medium. For granular
geometries, calculation of the specific surface area can be analytically performed by summing up the areas of all polyhedron
facets (excluding those that are on the domain boundaries). For tubular geometries, as the geometries are made up by
randomly intersecting cylinders, an exact analytical calculation of the specific surface area is very difficult. Nonetheless, a
good estimation of the specific surface area for tubular geometries can be made from the porosity and the tube diameter.
The effective total length of tubes is calculated by pore volume divided by tube cross-sectional area, and the total surface
area of all the tubes is,
As = 4Vp /d (11)
where Vp is pore volume and d is diameter. The specific surface area is given by
s = As /Vs (12)
where Vs is solid volume. The well-known Kozeny–Carman equation [17–19] non-dimensionalizes the permeability using
the specific surface area, and presents ks2 as a function of porosity and a fitting coefficient that takes into account the
tortuosity of the medium.
1 φ3
k= . (13)
C (1 + φ)s2
Fig. 12. Porosity–permeability relations of homogeneous and isotropic granular and tubular geometries and comparison with the Kozeny–Carman
equation (Eq. (13)).
In Eq. (13), C is the Kozeny–Carman coefficient and φ is the porosity. Fig. 12 shows ks2 of the stochastic geometries and the
Kozeny–Carman equation with a Kozeny–Carman coefficient of 5 as reported in literature for uniform sphere packing [81].
The dimensionless permeability of granular geometries and that of tubular geometries are very close when the porosity
is less than 20%. However, at higher porosity, the dimensionless permeability of tubular geometries becomes about two
times of that of the granular geometries. This difference is caused by the more efficient transport of fluid through cylindrical
pores. It is well known that for a given cross-sectional area (related to porosity) and a given perimeter (related to the surface
area), the round cross-section has the highes t permeability. Fig. 12 is a direct verification of the Kozeny–Carman equation.
It shows that the equation is not only applicable to sphere packings, but also fracture/tubular networks when the specific
surface area is used to non-dimensionalize the permeability.
6.2. Effect of heterogeneity
Geological porous media have complicated pore structures and a single length scale is not adequate to describe them.
Fractures and vugs (very large isolated pore bodies) are perhaps the most distinctive pore-scale features that can be readily
observed from images of rocks. Fractures are highly conductive relative to the matrix, and they represent a larger length scale
that is topologically connected and therefore dominate the hydrodynamic conductivity. Vugs, on the other hand, represent a
larger length scale that is topologically isolated; the permeability can be, to a large extent, still controlled by the pore throats
between the vugs.
To understand the effect of these heterogeneous geometric features on the porosity–permeability relations, we simulated
flows through heterogeneous granular and tubular geometries with vugs or fractures. Table 5 lists these geometries and
their dimensionless permeabilities. G7 is a base granular geometry created with 30 Poisson points. It has 24.8% porosity,
fracture aperture size (width) is 6 1x, and the system size is 200 × 200 × 200 1x3 . G7-1 has a 20-lattice wide fracture
parallel to the flow direction (connected fracture) superimposed to the base structure of G7. G7-2 has the same fracture but
perpendicular to the flow direction. G7-3 was created by randomly removing several polyhedrons from G7. As expected, the
dimensionless permeability was greatly enhanced in G7-1. When the fracture is perpendicular to the flow direction, as in
G7-2, or when there are vugs present, as in G7-3, the dimensionless permeability presented the same behavior of dropping
to below the Kozeny–Carman curve (Fig. 13). However, comparing with G7 in terms of dimensional permeability reveals that
the introduced disconnected fracture and vuggy porosity still enhanced the flow; however the increase in the permeability
was offset by the reduction in the surface area, leading to a decreased dimensionless permeability. Vuggy porosity was then
introduced to a tubular geometry T7, which was created with 165 Poisson points, 10.0% porosity, a tube size (diameter) of 6
1x, and a system size of 200 × 200 × 200 1x3 . Spheres of radius 10 1x, 15 1x and 30 1x were inserted into T7, to generate T7-
1, T7-2 and T7-3, respectively. As shown in Table 5 and Fig. 13, the size of vugs did not affect the permeability appreciably,
because their fractions in the overall porosity are almost the same and because they are isolated. However, these vugs
reduced the surface area of the porous medium, causing the dimensionless permeability to drop below the Kozeny–Carman
equation.
In addition to the Voronoi tessellation based geometries, a Berea sandstone sample was also tested. The geometry was
constructed from 400 two-dimensional micro-CT images. The size of the model is 600 × 600 × 400 1x3 with 1x = 1.16 µm,
and the porosity is 10.8%. The x, y and z-permeabilities (51, 52 and 102 md) were obtained using the parallel LBM code and
the solid surface area (5.09 × 106 µm2 ) was obtained using a MATLAB program that computes the surface area [117].
As shown in Table 5 and Fig. 13, the dimensionless permeabilities for the Berea sample (B-x, B-y, B-z) are all below the
Kozeny–Carmen equation. Also plotted in Fig. 13 are data from Arns and Knachstedt [118] presented using three lines
that correspond to their average, upper bound, and lower bound. These data were also obtained from lattice Boltzmann
simulations for about 750 geometries that were sub-sampled from four Fontainebleau sandstone images in the size of
4803 1x3 with the resolution of 5.7 µm. Our Berea results are in agreement with [118]. The dimensionless permeabilities of
Fig. 13. Permeability of heterogeneous granular and tubular geometries.
Table 5
Permeability of heterogeneous granular and tubular geometries and a Berea sample. The units for permeability k and
specific area s are 1x2 and 1x−1 , respectively, except for Berea where they are µm2 and µm−1 . rv stands for vug
radius and φv stands for vug porosity.
Geometry φ k s ks2 Comments
G7 24.8% 5.18E−01 9.97E−02 5.15E−03 Base granular geometry

G7-1 32.20% 4.13E+00 1.07E−01 4.71E−02 Fracture aligned with flow, width = 201x
G7-2 32.20% 5.69E−01 1.07E−01 6.48E−03 Fracture perpendicular to flow, width = 201x
G7-3 37.90% 1.43E+00 9.93E−02 1.41E−02 Vuggy, φv = 13.1%
T7 10.0% 4.44E−02 7.38E−02 2.42E−04 Base tubular geometry
T7-1 12.60% 4.72E−02 6.92E−02 2.26E−04 Vuggy, rv = 101x, φv = 3.8%, 73 vugs
B-x 5.1E−02 1.1E−05

B-y 10.8% 5.2E−02 1.5E−06 1.2E−05 Berea geometry
B-z 1.02E−01 2.3E−05
these sandstone geometries are lower than those of homogeneous tubular and granular geometries because of the angularity
of sand grains and the heterogeneous structure.
From the above data, it is tempting to think that connected fractures should increase the dimensionless permeability
to above the Kozeny–Carman equation and disconnected fractures or vugs should shift the dimensionless permeability
to below the Kozeny–Carman equation. However, additional simulations show that the Kozeny–Carman equation is not
sufficient to be such an indicator. Starting from G7-1, we generated a few variants where the fracture length was shortened
to 3/4 (G7-1-3), 1/2 (G7-1-2), and then 1/4 (G7-1-1) of the size of the domain, as shown in Fig. 14. We also simulated the flow
in the complex fractured granular geometry presented in Fig. 6. Table 6 lists these geometries and their dimensional and
dimensionless permeabilities in all three directions. The dimensionless permeabilities are also plotted in Fig. 15. When the
single fracture length is increased toward the full length, the z-permeabilities approached that of G7-1, which is as expected
as the fracture fully penetrates the medium in the z direction; the y-permeabilities approached that of G7-2, which is also
reasonable as the fracture is always perpendicular to the y direction and always acts as a vug. What is more interesting is the
trend in the x-permeabilities, which represents the situation where the fracture is extending in the same direction of flow
and acts as a partially penetrating conduit. When the fractures are short (1/4 and 1/2 of the full length), the x-permeabilities
were close to the z-permeabilities and were below the Kozeny–Carman equation. When the fracture length becomes 3/4 of
the full length, the x-permeability became quite different from the z-permeability and came back on the Kozeny–Carman
curve. Though not tested, we believe that with further increase in the fracture length the x-permeability will approach G7-1.
The trend in the x-permeability shows that (1) the orientation of fractures relative to the flow direction is very important;
and (2) Kozeny–Carman equation is not a sufficient indicator of pore geometry as in this case a heterogeneous geometry with
a nearly-penetrating fracture showed the same dimensionless permeability as the homogeneous geometry. When it comes
to G-FN, the system with complex fractures that are connected in all three directions but with some dead ends and tortuosity
effects, as shown in Fig. 15, the dimensionless x-, y-, and z-permeabilities are all above the Kozeny–Carman equation. The
x- and y-permeabilities are smaller than z-permeability, due to the dead ends and the fracture tortuosity effect.
Fluid velocity distribution is a quantity directly related to advective transport through porous media. To quantify the
difference made in the fluid velocity distribution by fractures or vugs in the granular and tubular geometries, the probability
density function, skewness, and kurtosis were measured for the velocity field in four selected geometries. The probability
density function describes the normalized likelihood for a random variable to take on a given value. Skewness and kurtosis,
being the third and the fourth order moment of the distribution, describe how close the distribution is to a Gaussian.
The skewness is a measure of the extent to which a probability distribution leans to one side of the mean: a negative
Table 6
Permeability of fractured granular geometries. The units for permeability k and specific area s are 1x2 and
1x−1 , respectively.
Geometry φ k s ks2 Comments
26.5% x 5.64E−01 1.02E−01 5.81E−03

One simple fracture, 1/4 size of
G7-1-1 26.5% y 5.34E−01 1.02E−01 5.50E−03
the domain
26.5% z 1.22E+00 1.02E−01 1.26E−02
28.4% x 6.74E−01 1.03E−01 7.22E−03

G7-1-2 28.4% y 5.47E−01 1.03E−01 5.86E−03
the domain
28.4% z 2.12E+00 1.03E−01 2.26E−02
30.3% x 9.14E−01 1.06E−01 1.02E−02

G7-1-3 30.3% y 5.59E−01 1.06E−01 6.23E−03
the domain
30.3% z 3.03E+00 1.06E−01 3.38E−02
43.0% x 7.23E+00 1.22E−01 1.07E−01

Fig 6; channel width = 4.5 1x
G-FN 43.0% y 8.82E+00 1.22E−01 1.30E−01
and fracture width = 30 1x
43.0% z 2.78E+01 1.22E−01 4.10E−01
Fig. 14. Variants of a single fracture in a granular geometry.
Fig. 15. Permeability of fractured granular geometries.
skewness in the velocity distribution indicates that the distribution leans to the right and there are many fluid ‘‘particles’’
that move slower than the mean, and a positive skewness indicates that the distribution leans to the left and there are many
fluid ‘‘particles’’ moving faster than the mean. The kurtosis measures the peakedness of the probability density function.
A Gaussian distribution has a kurtosis of three. A higher kurtosis (more than three) indicates a more peaked probability
distribution than the Gaussian. The excess kurtosis is defined as the kurtosis minus three, which is the quantity reported in
this paper.
Fig. 16. Probability density function of velocity field for heterogeneous stochastic geometries.
Fig. 16 shows the probability density functions of fluid velocity components in four selected geometries: G7, a
homogeneous and isotropic granular geometry; G7-1, a fractured granular geometry with a single and fully-penetrating
fracture parallel to the x direction; T7, a homogeneous and isotropic tubular geometry; and T7-2, a vuggy tubular geometry
with 73 spherical vugs of radius = 10 1x. Table 7 shows the skewness and excess kurtosis of the distributions. The mean
flow was in the x direction.
As to the skewness, in the directions perpendicular to the mean flow direction (y and z), none of the geometries showed
significant skewness, indicating that distributions of the velocity components normal to the mean flow are nearly symmetric.
In the mean flow direction, all geometries showed positive skewness, which indicates that all distributions lean to the left.
As we compare G7-1 with G7, it is clear that the fracture in G7-1 increases the skewness due to the fact that the fluid in
the fracture is moving much faster than the fluid in the matrix. As we compare T7-2 to T7, it is noted that the addition of
vugs also increases the skewness. This reduction is due to the fact that fluid in the vugs is not flowing as fast as that in the
inter-vug connections.
Table 7 shows that the excess kurtosis of velocity distributions in G7 are quite close to that for Gaussian distributions in
all three directions. Between G7 and T7, the velocity distributions in the tubular geometry are more ‘‘peaked’’ than those in
the granular geometry. When the fracture is introduced, in G7-1 the velocity distributions in the y and z directions became
strongly peaked, because there is no longer strong flow in the y and z directions (the fracture has taken all of the flux).
Table 7
Skewness and excess kurtosis of fluid velocity distributions for homogeneous
and heterogeneous geometries.
Geometry Skewness Excess kurtosis
x y z x y z
G7 0.477 0.103 −0.035 −0.607 0.198 0.473

G7-1 1.287 0.108 0.146 −0.023 2.708 2.990
T7 1.654 −0.117 −0.039 2.994 1.926 2.281
T7-2 2.348 −0.153 0.224 7.127 4.559 7.960
Fig. 17. Higdon and Ford [119] fibrous models: SC (left), BCC (middle) and FCC (right).
Between T7 and T7-2, it is seen that vugs significantly increased the excess kurtosis, due to the higher fraction of nearly
stagnant fluids in the large vugs.
6.3. Fibrous geometries
Fibrous medium is an important kind of porous medium that is widely used in filtration, insulation, absorption, and
heat/mass exchangers. It is also found in biological systems (e.g. tissues, bones, inter-cellular matrix) and geological
formations (e.g. peats). In this study, fibrous geometries were created by inverting the binary images of tubular geometries,
and the solid volume fraction ranges from 0.01 to 0.40.
Higdon and Ford [119] studied the drag in fibrous media using three periodic models based on ordered networks of
cylindrical fibers—the simple cubic model (SC), the body-centered cubic model (BCC), and the face-centered cubic model
(FCC) (Fig. 17). By changing the radius of fibers, a range of porosities can be obtained. The fluid velocity field was computed
using the spectral boundary element method [120]. With the definition of the effective length, Higdon and Ford [119]
calculated the mean drag force per unit length of fiber F . F is related to the permeability of fibrous medium by the following
equation
π r2
 
F
= (14)
µU φs k
where µ is the fluid viscosity, U is the superficial velocity of the fluid through the fibers. F /µU is therefore the dimensionless
mean drag force per unit length of fiber. Using the solid volume fraction φs = 1 − φ where φ is the porosity, the effective
length of fibers L may be defined as π r 2 L = φs V where V is the total volume of the system. Then, with the definition of
the normalized length of fibers LD = Lr 2 /V , a dimensionless total drag force LD F /µU can be defined and calculated. In
the absence of hydrodynamic interaction among the fibers, it is expected that the dimensionless total drag force should
scale linearly with the solid volume fraction φs . Deviation from the linear scale can be used to assess the intensity of the
hydrodynamic interaction. In this study, we report both the dimensionless permeability k/r 2 and the dimensionless total
drag force LD F /µU.
In order to verify our lattice Boltzmann model and data analysis routines, we first simulated flows through one SC, one
BCC, and one FCC model and compared our results against those reported in [119]. The parameters of these models and their
dimensionless permeability and total drag, characterized using our code, are presented in Table 8. These simulations were
conducted in systems of size 100 × 100 × 1001x3 . The radii of the fibers are 91x for the SC model and 61x for the BCC and
FCC models. As shown in Fig. 18, our results agree very well with Higdon and Ford’s correlations.
Table 8
Dimensionless permeability and total drag of selected SC, BCC, and FCC models.
Geometry φ k/r 2 Normalized Mean drag, Total drag,
effective length, LD F /µU LD F /µU
SC 93.2% 4.15E+00 2.18E−02 1.11E+01 2.41E−01

BCC 92.9% 4.13E+00 2.26E−02 1.07E+01 2.42E−01
FCC 91.6% 3.18E+00 2.68E−02 1.17E+01 3.14E−01
Fig. 18. Dimensionless total drag force of fibrous geometries. Solid fraction is up to 0.09.
Fig. 19. Dimensionless permeability of Voronoi fibrous geometries.
We then carried out simulations through a set of eleven fibrous geometries. For geometries with solid fractions less than
0.09, we present the total dimensionless drag forces in Fig. 18. It can be observed that, in the dilute limit, random fibrous
network based on Voronoi tessellation shows lower drag forces than SC, BCC, and FCC models of the same solid fraction and
weaker hydrodynamic interactions. This, perhaps, is due to the lower average coordination number of the Voronoi fibrous
network (four) compared to those of the SC (six), BCC (eight), and FCC (twelve) models. When coordination number is high,
we expect stronger hydrodynamic interaction near the junctions of the fibers. When the solid fraction is less than 0.05, the
4/3
total drag force of the Voronoi network scales as φs . In Fig. 19, the dimensionless permeabilities of all fibrous geometries
were plotted as a function of porosity. Interestingly, aside from some differences in the solid fraction range of 0.05–0.15,
in general the permeability of the Voronoi network agrees very well with Higdon and Ford’s correlations on a logarithmic
scale.
7. Conclusions
This paper studies the porosity–permeability relations of basic random porous media models and the effect of geometric
heterogeneities. We developed a method based on the Voronoi tessellation to generate three basic types of random porous
medium geometries: granular, tubular, and fibrous. Fluid flow through these geometry models was solved using a massively
parallelized lattice Boltzmann code. We validated the code using analytical solutions for Poiseuille flows. Through a series of
tests, we determined the lattice resolutions and minimum sizes of the geometry models so that the permeability values are
grid-independent and statistically representative. It is found that a grid resolution of six lattice nodes for fracture aperture
or tube/fiber diameter can produce results that are within 3% of the true grid-independent permeability, and Voronoi
geometries generated using 30 or more Poisson points can give statistically representative porosity and permeability.
Porosity–permeability relations were first characterized for homogeneous and isotropic granular and tubular geometries
within the porosity range of 0.01–0.40. When the permeability was normalized by the specific surface area (per unit volume
of solid), the porosity–permeability relations for granular and tubular geometries are very close at low porosities (<20%)
and can be well represented by the Kozeny–Carman equation with a Kozeny–Carman coefficient of five. At higher porosity
(>20%), the tubular geometries have higher dimensionless permeability. This difference is likely caused by the more efficient
transport of fluid through the cylindrical pores.
Heterogeneities introduced in the forms of vugs or fractures caused deviations from the Kozeny–Carman equation.
Our data show that connected fractures significantly enhanced the flow. Systems with connected fractures appeared on
the porosity–permeability cross-plot as points above the Kozeny–Carman equation. Systems with fractures that are not
conductive along the flow direction and vugs, on the other hand, appeared as points below the Kozeny–Carman equation.
Exception to this general observation, however, was found when we applied a non-penetrating fracture along the direction
of flow. This fracture, on one hand, offered a low-resistance pathway. However, since it is not fully penetrating, it did not
increase the permeability as much as a fully penetrating fracture. As a result, this heterogeneous porous medium with a
partially penetrating fracture has the same dimensionless permeability as a homogeneous porous medium. Therefore, the
relative position of the dimensionless permeability to the Kozeny–Carman equation is not a sufficient indicator of pore
geometry. Other transport properties, for example, dispersion, may be used in conjunction with permeability and specific
surface area to infer geometric characteristics of the porous medium and this will be pursued in our future studies.
For fibrous geometries, we analyzed the dimensionless permeability and dimensionless total drag force in the solid
fraction range of 0.01–0.40. The dimensionless permeability of Voronoi fibrous geometries, in general, agrees well with
Higdon and Ford’s [119] correlations established for simple cubic (SC), body-centered cubic (BCC), and face-centered cubic
(FCC) fibrous models. In the dilute limit, the dimensionless total drag of the Voronoi models is less than those in the SC, BCC,
4/3
and BCC models. The hydrodynamic interaction between fibers in the Voronoi network follows a power law of φs in the
dilute limit, where φs is the fiber solid fraction. This scaling is weaker than that observed for SC, BCC, and FCC configurations.
Acknowledgments
This research is funded by the Research Partnership to Secure Energy for America 09122-29. XY also thanks the support
from the Coupled Integrated Multiscale Measurements and Modeling (CIMMM) Consortium of the Natural Gas and Oil
Institute (UNGI) of Colorado School of Mines and the consortium of Unconventional Reservoir Engineering Project (UREP).
The computational resource was provided by the Golden Energy Computing Organization at Colorado School of Mines.
We are also grateful to Argonne and Oak Ridge Leadership Computing Facilities for providing accommodations on code
performance testing.
References
[1] F.Jerry Lucia, Rock-fabric/petrophysical classification of carbonate pore space for reservoir characterization, AAPG Bull. 79 (9) (1995) 1275–1300.
[2] R.M. Sok, T. Varslot, A. Ghous, S. Latham, A.P. Sheppard, M.A. Knackstedt, Pore scale characterization of carbonates at multiple scales: Integration of
micro-CT, BSEM, FIBSEM, Petrophysics 51 (6) (2010) 379–387.
[3] M. Elgmati, H. Zhang, B. Bai, R. Flori, Q. Qu, Submicron-pore characterization of shale gas plays, Paper SPE 144050 Presented at North American
Unconventional Gas Conference and Exhibition, The Woodlands, Texas, June 14–16, 2011.
[4] U. Kuila, M. Prasad, Surface area and pore-size distribution in clays and shales, Paper SPE 146869 Presented at SPE Annual Technical Conference and
Exhibition, Denver, Colorado, October 30–November 2, 2011.
[5] D. Silin, T. Kneafsey, Gas shale: From nanometer-scale observations to well modeling, Paper SPE 149489 Presented at Canadian Unconventional
Resources Conference, Alberta, Canada, November 15–17, 2011.
[6] M. Curtis, C. Sondergeld, R. Ambrose, C. Rai, Microstructural investigation of gas shales in two and three dimensions using nanometer-scale resolution
imaging, AAPG Bull. 96 (4) (2012) 665–677.
[7] J.D. Walls, E. Diaz, T. Cavanaugh, Shale reservoir properties from digital rock physics, Paper SPE 152752 Presented at SPE/EAGE European
Unconventional Resources Conference and Exhibition, Vienna, Austria, March 20–22, 2012.
[8] M. Knackstedt, A. Carnerup, A. Golab, R. Sok, B. Young, L. Riepe, Petrophysical characterization of unconventional reservoir core at multiple scales,
Paper SPWLA 2012-242 Presented at SPWLA 53rd Annual Logging Symposium, Cartagena, Colombia, June 16–20, 2012.
[9] P.H. Nelson, Pore-throat sizes in sandstones, tight sandstones, and shales, AAPG Bull. 93 (3) (2009) 329–340.
[10] H. Darcy, Les fontaines publiques de la ville de Dijon, V. Dalmont, 1856.
[11] F. Seelheim, Methoden zur Bestimmung der Durchlässigkeit des Bodens, Anal. Bioanal. Chem. 19 (1) (1880) 387–418.
[12] W.C. Krumbein, G.D. Monk, Permeability as a function of the size parameters of unconsolidated sand, Trans. AIME 151 (1) (1943) 153–163.
[13] J. Bear, Dynamics of Fluids in Porous Media, American Elsevier, 1972.
[14] G.M. Fair, L.P. Hatch, Fundamental factors governing the streamline flow of water through sand, J. Am. Water Works Assoc. (1933) 1551–1565.
[15] J. Happel, Viscous flow in multiparticle systems: slow motion of fluids relative to beds of spherical particles, AIChE J. 4 (2) (1958) 197–201.
[16] H. Hasimoto, On the periodic fundamental solutions of the Stokes equations and their application to viscous flow past a cubic array of spheres, J.
Fluid Mech. 5 (02) (1959) 317–328.
[17] J. Kozeny, Ueber kapillare leitung des wassers im boden, Sitzungsber. Akad. Wiss. Wien 136 (1927) 271–306.
[18] P.C. Carman, Fluid flow through granular beds, Trans. Inst. Chem. Eng. 15 (1937) 150–166.
[19] P.C. Carman, Flow of Gases through Porous Media, Butterworths Scientific Publications, London, 1956.
[20] I. Fatt, The network model of porous media I. Capillary pressure characteristics, Trans. AIME 207 (1956) 144–159.
[21] I. Fatt, The network model of porous media II. Dynamic properties of a single size tube network, Trans. AIME 207 (1956) 160–163.
[22] I. Fatt, The network model of porous media III. Dynamic properties of networks with tube radius distribution, Trans. AIME 207 (1956) 164–177.
[23] S. Prager, Viscous flow through porous media, Phys. Fluids 4 (1961) 1477–1482.
[24] I.R. de Boer, I.W. Ehlers, A historical review of the formulation of porous media theories, Acta Mech. 74 (1–4) (1988) 1–8.
[25] P.M. Adler, Porous Media: Geometry and Transports, Butterworth-Heinemann, 1992.
[26] S.R. Broadbent, J.M. Hammersley, Percolation processes I. Crystals and mazes, Proc. Cambridge Philos. Soc. 53 (3) (1957) 629–641.
[27] S. Kirkpatrick, Percolation and conduction, Rev. Modern Phys. 45 (4) (1973) 574–588.
[28] I. Balberg, Recent developments in continuum percolation, Phil. Mag. Part B 56 (6) (1987) 991–1003.
[29] R. Lenormand, Flow through porous media: limits of fractal patterns, Proc. R. Soc. Lond. Ser. A Math. Phys. Eng. Sci. 423 (1864) (1989) 159–168.
[30] M. Blunt, P. King, Macroscopic parameters from simulations of pore scale flow, Phys. Rev. A 42 (8) (1990) 4780–4787.
[31] S. Feng, B.I. Halperin, P.N. Sen, Transport properties of continuum systems near the percolation threshold, Phys. Rev. B 35 (1) (1987) 197–214.
[32] B. Berkowitz, I. Balberg, Percolation theory and its application to groundwater hydrology, Water Resour. Res. 29 (4) (1993) 775–794.
[33] K.M. Golden, Percolation models for porous media, in: Homogenization and Porous Media, Springer, New York, 1997, pp. 27–43.
[34] A. Hunt, R. Ewing, Percolation Theory for Flow in Porous Media, Vol. 771, Springer Science & Business Media, 2009.
[35] B.B. Mandelbrot, The Fractal Geometry of Nature. 1982, Freeman & Co, San Francisco, 1982.
[36] C.G. Jacquin, P.M. Adler, The fractal dimension of a gas—liquid interface in a porous medium, J. Colloid Interface Sci. 107 (2) (1985) 405–417.
[37] D. Avnir, D. Farin, P. Pfeifer, Surface geometric irregularity of particulate materials: the fractal approach, J. Colloid Interface Sci. 103 (1) (1985)
112–123.
[38] C.A. Aviles, C.H. Scholz, J. Boatwright, Fractal analysis applied to characteristic segments of the San Andreas fault, J. Geophys. Res.: Solid Earth
(1978–2012) 92 (B1) (1987) 331–344.
[39] P.M. Adler, Transport processes in fractals III. Taylor dispersion in two examples of fractal capillary networks, Int. J. Multiph. Flow 11 (2) (1985)
241–254.
[40] P.M. Adler, Fractal porous media III: Transversal Stokes flow through random and Sierpinski carpets, Transp. Porous Media 3 (2) (1988) 185–198.
[41] M. Sahimi, Y.C. Yortsos, Applications of fractal geometry to porous media: a review, Paper SPE 20476 Presented at Annual Fall Meeting of the Society
of Petroleum Engineers, New Orleans, Louisiana, 1990.
[42] J.H. Dunsmuir, S.R. Ferguson, K.L. D’Amico, J.P. Stokes, X-ray microtomography: a new tool for the characterization of porous media, Paper SPE 22860
Presented at SPE Annual Technical Conference and Exhibition, Dallas, Texas, October 6–9, 1991.
[43] M.E. Coles, R.D. Hazlett, E.L. Muegge, K.W. Jones, B. Andrews, B. Dowd, P. Siddons, A. Peskin, P. Spanne, W.E. Soll, Developments in synchrotron X-ray
microtomography with applications to flow in porous media, SPE Reser. Eval. Eng. 1 (4) (1998) 288–296.
[44] B.J. Inkson, M. Mulvihill, G. Möbus, 3D determination of grain shape in a FeAl-based nanocomposite by 3D FIB tomography, Scr. Mater. 45 (7) (2001)
753–758.
[45] M. Curtis, R. Ambrose, C. Sondergeld, C. Rai, Transmission and scanning electron microscopy investigation of pore connectivity of gas shales on the
nanoscale, Paper SPE 144391 Presented at North American Unconventional Gas Conference and Exhibition, The Woodlands, Texas, June 14–16, 2011.
[46] G.R. Chalmers, R.M. Bustin, I.M. Power, Characterization of gas shale pore systems by porosimetry, pycnometry, surface area, and field emission
scanning electron microscopy/transmission electron microscopy image analyses: Examples from the Barnett, Woodford, Haynesville, Marcellus,
and Doig units, AAPG Bull. 96 (6) (2012) 1099–1119.
[47] D. Wildenschild, A.P. Sheppard, X-ray imaging and analysis techniques for quantifying pore-scale structure and processes in subsurface porous
medium systems, Adv. Water Resour. 51 (2013) 217–246.
[48] M.Y. Joshi, A class of stochastic models for porous media (Ph.D. diss.), University of Kansas, 1974.
[49] J.A. Quiblier, A new three-dimensional modeling technique for studying porous media, J. Colloid Interface Sci. 98 (1) (1984) 84–102.
[50] Z.R. Liang, C.P. Fernandes, F.S. Magnani, P.C. Philippi, A reconstruction technique for three-dimensional porous media using image analysis and
Fourier transforms, J. Pet. Sci. Eng. 21 (3) (1998) 273–283.
[51] B. Biswal, P.-E. Øren, R.J. Held, S. Bakke, R. Hilfer, Modeling of multiscale porous media, Image Anal. Stereol. 28 (2009) 23–34.
[52] C.L.Y. Yeong, S. Torquato, Reconstructing random media, Phys. Rev. E 57 (1) (1998) 495–506.
[53] C.L.Y. Yeong, S. Torquato, Reconstructing random media. II. Three-dimensional media from two-dimensional cuts, Phys. Rev. E 58 (1) (1998) 224–233.
[54] M.S. Talukdar, O. Torsaeter, M.A. Ioannidis, J.J. Howard, Stochastic reconstruction of chalk from 2D images, Transp. Porous Media 48 (1) (2002)
101–123.
[55] M.S. Talukdar, O. Torsaeter, M.A. Ioannidis, J.J. Howard, Stochastic reconstruction, 3D characterization and network modeling of chalk, J. Pet. Sci. Eng.
35 (1) (2002) 1–21.
[56] P. Čapek, V. Hejtmánek, L. Brabec, A. Zikánová, M. Kočiřík, Stochastic reconstruction of particulate media using simulated annealing: improving pore
connectivity, Transp. Porous Media 76 (2) (2009) 179–198.
[57] K. Wu, M. IJ Van Dijke, G.D. Couples, Z. Jiang, J. Ma, K.S. Sorbie, J. Crawford, I. Young, X. Zhang, 3D stochastic modelling of heterogeneous porous
media–applications to reservoir rocks, Transp. Porous Media 65 (3) (2006) 443–467.
[58] A.N. Diogenes, L. Santos, C. Fernandes, A.C. Moreira, C.R. Apollon, Porous media microstructure reconstruction using pixel-based and object-based
simulated annealing—comparison with other reconstruction methods, Eng. Térmica 8 (2009) 35–41.
[59] C.H. Arns, M.A. Knackstedt, K.R. Mecke, Characterising the morphology of disordered materials, in: Morphology of Condensed Matter, Springer, Berlin,
Heidelberg, 2002, pp. 37–74.
[60] M.C. Sukop, K.J. Cunningham, Lattice Boltzmann methods applied to large-scale three-dimensional virtual cores constructed from digital optical
borehole images of the karst carbonate Biscayne aquifer in southeastern Florida, Water Resour. Res. 50 (2014) 8807–8825.
[61] S. Bryant, M. Blunt, Prediction of relative permeability in simple porous media, Phys. Rev. A 46 (4) (1992) 2004–2011.
[62] S. Bakke, P.-E. Øren, 3-D pore-scale modelling of sandstones and flow simulations in the pore networks, SPE J. 2 (2) (1997) 136–149.
[63] M. Pilotti, Reconstruction of clastic porous media, Transp. Porous Media 41 (3) (2000) 359–364.
[64] F.A.L. Dullien, Porous Media: Fluid Transport and Pore Structure. Access Online via Elsevier, 1991.
[65] S. Torquato, Statistical description of microstructures, Annu. Rev. Mater. Res. 32 (1) (2002) 77–111.
[66] L. Muche, D. Stoyan, Contact and chord length distributions of the Poisson voronoi tessellation, J. Appl. Probab. (1992) 467–471.
[67] D. Stoyan, W.S. Kendall, J. Mecke, L. Ruschendorf, Stochastic Geometry and its Applications, Vol. 2, Wiley, Chichester, 1995.
[68] B. Biswal, C. Manwart, R. Hilfer, Three-dimensional local porosity analysis of porous media, Physica A 255 (3) (1998) 221–241.
[69] R. Hilfer, Review on scale dependent characterization of the microstructure of porous media, Transp. Porous Media 46 (2–3) (2002) 373–390.
[70] L.A. Santalo, Integral Geometry and Geometric Probability, in: Encyclopedia of Mathematics and Its Applications, vol. 1, Addison-Wesley, 1976.
[71] C.H. Arns, M.A. Knackstedt, W.V. Pinczewski, K.R. Mecke, Euler–Poincaré characteristics of classes of disordered media, Phys. Rev. E 63 (3) (2001)
031112.
[72] P. Lehmann, M. Berchtold, B. Ahrenholz, J. Tölke, A. Kaestner, M. Krafczyk, H. Flühler, H.R. Künsch, Impact of geometrical properties on permeability
and fluid phase distribution in porous media, Adv. Water Resour. 31 (9) (2008) 1188–1204.
[73] G.E. Schröder-Turk, W. Mickel, S.C. Kapfer, M.A. Klatt, F.M. Schaller, M.J.F. Hoffmann, N. Kleppmann, et al., Minkowski tensor shape analysis of cellular,
granular and porous structures, Adv. Mater. 23 (22–23) (2011) 2535–2553.
[74] H. Okabe, M.J. Blunt, Prediction of permeability for porous media reconstructed using multiple-point statistics, Phys. Rev. E 70 (6) (2004) 066135.
[75] H. Okabe, M.J. Blunt, Pore space reconstruction of vuggy carbonates using microtomography and multiple-point statistics, Water Resour. Res. 43 (12)
(2007) W12S02.
[76] B. Biswal, P.-E. Øren, R.J. Held, S. Bakke, R. Hilfer, Stochastic multiscale model for carbonate rocks, Phys. Rev. E 75 (6) (2007) 061303.
[77] S. Roth, B. Biswal, G. Afshar, R.J. Held, P.-E. Øren, L.I. Berge, R. Hilfer, Continuum-based rock model of a reservoir dolostone with four orders of
magnitude in pore sizes, AAPG Bull. 95 (6) (2011) 925–940.
[78] W.D. Carrier III, Goodbye, hazen; hello, Kozeny–Carman, J. Geotech. Geoenviron. Eng. 129 (11) (2003) 1054–1056.
[79] R.P. Chapuis, M. Aubertin, Predicting the Coefficient of Permeability of Soils Using the Kozeny–Carman Equation, École polytechnique de Montréal,
2003.
[80] P. Xu, B. Yu, Developing a new form of permeability and Kozeny–Carman constant for homogeneous porous media by means of fractal geometry,
Adv. Water Resour. 31 (1) (2008) 74–81.
[81] F.J. Valdes-Parada, J.A. Ochoa-Tapia, J. Alvarez-Ramirez, Validity of the permeability Carman–Kozeny equation: A volume averaging approach, Physica
A 388 (6) (2009) 789–798.
[82] K. Yazdchi, S. Srivastava, S. Luding, On the validity of the Carman–Kozeny equation in random fibrous media, Paper Presented at International
Conference on Particle-based Methods II, PARTICLES 2011, Barcelona, Spain, October 26–28, 2011.
[83] G.M. Voronoï, Nouvelles applications des paramètres continus á la théorie des formes quadratiques. Deuxième mémoire: recherches sur les
parallélloèdres primitifs, J. Reine Angew. Math. 134 (1907) 198–287.
[84] A. Okabe, B. Boots, K. Sugihara, S.N. Chiu, Spatial Tessellations: Concepts and Applications of Voronoi Diagrams, Vol. 501, Wiley.com, 2009.
[85] P.H. Winterfeld, L.E. Scriven, H.T. Davis, Percolation and conductivity of random two-dimensional composites, J. Phys. C: Solid State Phys. 14 (17)
(1981) 2361–2376.
[86] G.R. Jerauld, J.C. Hatfield, L.E. Scriven, H.T. Davis, Percolation and conduction on Voronoi and triangular networks: a case study in topological disorder,
J. Phys. C: Solid State Phys. 17 (9) (1984) 1519–1529.
[87] G.R. Jerauld, L.E. Scriven, H.T. Davis, Percolation and conduction on the 3D Voronoi and regular networks: a second case study in topological disorder,
J. Phys. C: Solid State Phys. 17 (19) (1984) 3429–3439.
[88] N.A. Vrettos, H. Imakoma, M. Okazaki, Characterization of porous media by means of the voronoi-delaunay tessellation, Chem. Eng. Process. 25 (1)
(1989) 35–45.
[89] N.A. Vrettos, H. Imakoma, M. Okazaki, Effective medium approximation of 3-D Voronoi networks, J. Appl. Phys. 67 (7) (1990) 3249–3253.
[90] C.H. Arns, M.A. Knackstedt, K. Mecke, 3D structural analysis: sensitivity of Minkowski functionals, J. Microsc. 240 (3) (2010) 181–196.
[91] M. Newman, X. Yin, Lattice Boltzmann simulation of non-Darcy flow in stochastically generated 2D porous media geometries, SPE J. 18 (1) (2013)
12–26.
[92] Y. Yong, X. Lou, S. Li, C. Yang, X. Yin, Direct simulation of the influence of the pore structure on the diffusion process in porous media, Comput. Math.
Appl. 67 (2) (2014) 412–423.
[93] M. Wu, F. Xiao, R.M. Johnson-Paben, S.T. Retterer, X. Yin, K.B. Neeves, Single-and two-phase flow in microfluidic porous media analogs based on
Voronoi tessellation, Lab Chip 12 (2) (2012) 253–261.
[94] W. Xu, J.T. Ok, F. Xiao, K.B. Neeves, X. Yin, Effect of pore geometry and interfacial tension on water-oil displacement efficiency in oil-wet microfluidic
porous media analogs, Phys. Fluids 26 (2014) 093102.
[95] P.X. Ma, R. Zhang, Synthetic nano-scale fibrous extracellular matrix, J. Biomed. Mater. Res. 46 (1) (1999) 60–72.
[96] C. Gambini, B. Abou, A. Ponton, A. Cornelissen, Micro-and macrorheology of jellyfish extracellular matrix, Biophys. J. 102 (1) (2012) 1–9.
[97] J. Banhart, Manufacture, characterisation and application of cellular metals and metal foams, Prog. Mater. Sci. 46 (6) (2001) 559–632.
[98] S. Chen, G.D. Doolen, Lattice Boltzmann method for fluid flows, Annu. Rev. Fluid Mech. 30 (1) (1998) 329–364.
[99] D. Yu, R. Mei, L.-S. Luo, W. Shyy, Viscous flow computations with the method of lattice Boltzmann equation, Prog. Aerosp. Sci. 39 (5) (2003) 329–367.
[100] C.K. Aidun, J.R. Clausen, Lattice-Boltzmann method for complex flows, Annu. Rev. Fluid Mech. 42 (2010) 439–472.
[101] A.J.C. Ladd, R. Verberg, Lattice-Boltzmann simulations of particle–fluid suspensions, J. Stat. Phys. 104 (5–6) (2001) 1191–1251.
[102] R. Mei, W. Shyy, D. Yu, L.-S. Luo, Lattice Boltzmann method for 3-D flows with curved boundary, J. Comput. Phys. 161 (2) (2000) 680–699.
[103] U. Frisch, D. d’Humières, B. Hasslacher, P. Lallemand, Y. Pomeau, J.P. Rivet, Lattice gas hydrodynamics in two and three dimensions, Complex Syst. 1
(1987) 649–707.
[104] R. Mei, L.-S. Luo, W. Shyy, An accurate curved boundary treatment in the lattice Boltzmann method, J. Comput. Phys. 155 (2) (1999) 307–330.
[105] T. Zhang, B. Shi, Z. Guo, Z. Chai, J. Lu, General bounce-back scheme for concentration boundary condition in the lattice-Boltzmann method, Phys. Rev.
E 85 (1) (2012) 016701.
[106] M. Bouzidi, D. d’Humieres, P. Lallemand, L.S. Luo, Lattice Boltzmann equation on a two-dimensional rectangular grid, J. Comput. Phys. 172 (2001)
704–717.
[107] I. Ginzburg, D. d’Humières, Multireflection boundary conditions for lattice Boltzmann models, Phys. Rev. E 68 (6) (2003) 066614.
[108] J. Latt, B. Chopard, O. Malaspinas, M. Deville, A. Michler, Straight velocity boundaries in the lattice Boltzmann method, Phys. Rev. E 77 (5) (2008)
056703.
[109] P.L. Bhatnagar, E.P. Gross, M. Krook, A model for collision processes in gases. I. Small amplitude processes in charged and neutral one-component
systems, Phys. Rev. 94 (3) (1954) 511–523.
[110] D. d’Humières, Multiple–relaxation–time lattice Boltzmann models in three dimensions, Philos. Trans. R. Soc. Lond. Ser. A: Math. Phys. Eng. Sci. 360
(1792) (2002) 437–451.
[111] I. Ginzburg, Equilibrium-type and link-type lattice Boltzmann models for generic advection and anisotropic-dispersion equation, Adv. Water Resour.
28 (11) (2005) 1171–1195.
[112] C. Pan, L.-S. Luo, C.T. Miller, An evaluation of lattice Boltzmann schemes for porous medium flow simulation, Comput. & Fluids 35 (8) (2006) 898–909.
[113] K. Mattila, J. Hyväluoma, J. Timonen, T. Rossi, Comparison of implementations of the lattice-Boltzmann method, Comput. Math. Appl. 55 (7) (2008)
1514–1524.
[114] L. Axner, J. Bernsdorf, T. Zeiser, P. Lammers, J. Linxweiler, A.G. Hoekstra, Performance evaluation of a parallel sparse lattice Boltzmann solver,
J. Comput. Phys. 227 (10) (2008) 4895–4911.
[115] C. Pan, J.F. Prins, C.T. Miller, A high-performance lattice Boltzmann implementation to model flow in porous media, Comput. Phys. Comm. 158 (2)
(2004) 89–105.
[116] M.A. Gallivan, D.R. Noble, J.G. Georgiadis, R.O. Buckius, An evaluation of the bounce-back boundary condition for lattice Boltzmann simulations,
Internat. J. Numer. Methods Fluids 25 (3) (1997) 249–263.
[117] D. Legland, K. Kiêu, M.-F. Devaux, Computation of Minkowski measures on 2D and 3D binary images, Image Anal. Stereol. 26 (2007) 83–92.
[118] C.H. Arns, M.A. Knachstedt, Virtual permeametry on microtomographic images, J. Pet. Sci. Eng. 45 (1) (2004) 41–46.
[119] J.J.L. Higdon, G.D. Ford, Permeability of three-dimensional models of fibrous porous media, J. Fluid Mech. 308 (1996) 341–361.
[120] G.P. Muldowney, J.J.L. Higdon, A spectral boundary element approach to three-dimensional Stokes flow, J. Fluid Mech. 298 (1995) 167–192.

Computers and Mathematics With Applications: Feng Xiao, Xiaolong Yin

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Computers and Mathematics With Applications: Feng Xiao, Xiaolong Yin

Uploaded by

Copyright:

Available Formats

Computers and Mathematics with Applications 72 (2016) 328–348

Contents lists available at ScienceDirect

Computers and Mathematics with Applications

Geometry models of porous media based on Voronoi

article info abstract

2. Porous media geometry

2.1. Two-dimensional (2D) geometry models

2.2. Three-dimensional (3D) geometry models

Fig. 4. Anisotropic granular and tubular geometries.

Fig. 5. Granular and tubular geometries with vugs.

3. Parallelized lattice Boltzmann method

Fig. 6. A synthetic fracture system and a fractured granular geometry.

Fig. 8. Parallel computing efficiency on Mira (left) and Titan (right).

4. Validation in simple geometries

5. Random geometries—lattice resolution and system size

6.1. Homogeneous granular and tubular geometries

6.2. Effect of heterogeneity

Fig. 13. Permeability of heterogeneous granular and tubular geometries.

Geometry φ k s ks2 Comments

G7 24.8% 5.18E−01 9.97E−02 5.15E−03 Base granular geometry

B-x 5.1E−02 1.1E−05

26.5% x 5.64E−01 1.02E−01 5.81E−03

28.4% x 6.74E−01 1.03E−01 7.22E−03

30.3% x 9.14E−01 1.06E−01 1.02E−02

43.0% x 7.23E+00 1.22E−01 1.07E−01

Fig. 14. Variants of a single fracture in a granular geometry.

Fig. 15. Permeability of fractured granular geometries.

G7 0.477 0.103 −0.035 −0.607 0.198 0.473

6.3. Fibrous geometries

SC 93.2% 4.15E+00 2.18E−02 1.11E+01 2.41E−01

Fig. 19. Dimensionless permeability of Voronoi fibrous geometries.

You might also like