Electrochimica Acta 53 (2008) 7630–7637

Contents lists available at ScienceDirect

Electrochimica Acta
journal homepage: www.elsevier.com/locate/electacta

Numerical and microfluidic pore networks: Towards designs for directed

water transport in GDLs
A. Bazylak, V. Berejnov, B. Markicevic, D. Sinton, N. Djilali ∗
Institute for Integrated Energy Systems and Department of Mechanical Engineering, University of Victoria, Victoria, BC, Canada V8W 3P6

a r t i c l e i n f o a b s t r a c t

Article history: Based on a pore network representation of porous media, numerical and experimental methods are
Received 10 January 2008 employed to explore the design of gas diffusion media in order to control and direct water transport.
Received in revised form 28 March 2008 Randomized pore networks with structured biasing in both diagonal and radial directions show a marked
Accepted 29 March 2008
influence on liquid water transport through the two-dimensional media. Radial biasing is shown to have
Available online 11 April 2008
the beneficial effect of decreasing the network saturation, which is a desirable quality in fuel cell gas dif-
fusion layers. Similar flow patterns are obtained from both numerical simulations and experiments, and
Keywords:
it is found that relatively small radial biasing can yield a 43% decrease in average saturation compared to
Gas diffusion layer
Porous media
pore networks without a prescribed radial gradient.
Two-phase flow © 2008 Elsevier Ltd. All rights reserved.
Microfluidic
Polymer electrolyte membrane fuel cell
Biased networks

1. Introduction Since the early development of pore network modelling [10–13],

In polymer electrolyte membrane (PEM) fuel cells, the gas dif-
fusion layer (GDL) provides pathways for gaseous fuel to reach the
catalyst sites, for electrons generated at the catalyst sites to reach
the current collector, and very importantly, for excess liquid water
to exit the system. Conventional GDLs are ineffective for water
removal and organized water transport due in part to its arbitrary
structure [1], and excess water can reduces transport of reactant
gases [2] as well as exacerbate degradation [3]. Significant improve-
ments in the PEM fuel cell are contingent on advancements in water
management, particularly in the GDL. Recently, Litster et al. [4] and
Bazylak et al. [5,6] investigated liquid water transport through PEM
fuel cell GDLs at the microscale level by employing fluorescence
microscopy. New methods for enhanced water transport have been
demonstrated via the active removal of water through the use of
transport plates [7] and electro-osmotic pumping [8,9]. One poten-
tial avenue for improving water management is the development
of GDL materials engineered for improved water management at
the microscale.
∗ Corresponding author at: Institute for Integrated Energy Systems and Depart-
ment of Mechanical Engineering, University of Victoria, PO Box 3055, Victoria, BC,
Canada V8W 3P6. Tel.: +1 250 721 6034; fax: +1 250 721 6323.
E-mail address: ndjilali@uvic.ca (N. Djilali).
models with spatially homogeneous pores and throats have been
employed to study the displacement processes in idealized porous
media [14]. Over the past two decades, network models have
been applied to a multitude of areas ranging from oil recov-
ery [15–20], ground water hydrology [21], textile characterization
[22], etc. Blunt [23] provides a broad review of recent pore net-
work model developments. The pore network modelling approach
has been very recently extended to the gas diffusion layer by
Gostick et al. [24], Markicevic et al. [25] and Sinha and Wang
[26]. Marckicevic et al. [25] used network simulations to inves-
tigate and establish the functional dependence of the relative
permeability and capillary pressure on liquid water saturation,
two of the key parameters required for two-phase PEM fuel cell
models [2]. Recently, Chapuis et al. [27] employed pore net-
works composed of randomly distributed disks to simulate the
anisotropic arrangement of GDL fibers, and they found good
agreement between their numerical and experimental flow pat-
terns.
Isichenko [28] introduced correlation as a useful modelling tech-
nique for porous media, whereby the liquid occupation probability
of pores and throats are considered. The method of correlation
[29,30] has been used to study the large-scale effects of two-phase
flow in porous media [31–34]. These correlated or long-range cor-
related porous media differ from those employed in this work.
Here, pore network obstacles are arranged in a biased manner with
respect to the entire occupied volume. To distinguish our networks

0013-4686/$ – see front matter © 2008 Elsevier Ltd. All rights reserved.
doi:10.1016/j.electacta.2008.03.078
 A. Bazylak et al. / Electrochimica Acta 53 (2008) 7630–7637
from the correlated networks, we refer to our networks as biased
networks.
Experimental validation has been provided for theoretical pore
network models [35–45]. Recently, Berejnov et al. [46] presented
a lab-on-chip approach to study two-phase transport in porous
media, employing microfluidic rapid prototyping fabrication meth-
ods in conjunction with fluorescence microscopy to study water
transport in pore networks. Experimental network results have also
been used in other porous media applications to guide and validate
theoretical modelling [47,48] or as inputs to existing models [49].
Work to date in the area of pore network and experimental
modelling of porous media has focused on characterizing existing
media. Our goal is to demonstrate a new approach to design-
ing porous networks for directing water transport by applying
long-scale correlations (biasing) to an initially homogeneously ran-
dom 2D porous network. This approach enables us to qualitatively
demonstrate the potential use of biasing for directing water trans-
port, which is of specific interest for PEMFC gas diffusion layers.
This novel technique is demonstrated with both numerical and
experimental networks that clearly illustrate the effect of biasing
on directing water transport.
2. Methodology
Pore network modelling provided a method to evaluate the liq-
uid water transport through the gas diffusion layer in a PEM fuel
cell, as indicated in Fig. 1. In this work, pore networks with pore size
characteristics of the gas diffusion layer [50] were designed with a
software package developed in-house, and the invasion percolation
with trapping algorithm developed by Markicevic et al. [25] was
used to evaluate the drainage process. Once a photomask design
was chosen, the photomask was generated and printed with a high
resolution printer. The photomask was employed during the pho-
tolithographic microfabrication process, where a microfluidic pore
network was generated. The spread of the water into the microflu-
idic network was monitored using fluorescence microscopy, and
flow patterns and time series evolution of liquid water transport
through the network were recorded. The pore network design
process is illustrated in Fig. 2. The experimental breakthrough pat-
tern were then compared to pore network simulations based on
Wilkinson and Willemsen’s algorithm for invasion percolation with
trapping [14,25].
Fig. 1. Schematic showing pore network modelling results with respect to PEMFC geometry: (a) diagram of a PEMFC, (b) sample numerical pore network breakthrough
pattern of a simulated gas diffusion layer, and (c) scanning electron microscope image of Toray TGP-H-060 GDL (length bar represents 200 ␮m).
7631
2.1. Pore network modelling
Pore network models provide an alternative to continuum phase
models to elucidate transport phenomena and determine trans-
port parameters in porous media. Continuum modelling of porous
media becomes impractical when the porous media has a com-
plex three-dimensional heterogeneous and disordered geometry,
such as the gas diffusion layer shown in Fig. 1(c). Numerically track-
ing phase interfaces and calculating surface tension forces becomes
computationally expensive. In contrast, pore network modelling is
an intermediate approach where the disordered porous media is
represented by an equivalent network of pores and throats, and the
throats are assigned varying hydraulic conductivities.
The invasion percolation algorithm with trapping [14] was used
to investigate deterministic realizations of pore network filling. The
two-dimensional model accounts for capillary dominated drainage
and cluster formation in a regular square capillary network of size
nL × nL pores. The porous media is represented by a matrix of
pores and connecting throats. The invading phase (water) enters
the network along the top side (inlet), and advances through the
network in a series of quasi-static steps by occupying available
throats having the largest throat radius. The fluid exits through
the opposite side (outlet) to the entrance, and the invasion pro-
cess is stopped once breakthrough occurs, where breakthrough to
the state when the first throat at the outlet is filled with the invad-
ing fluid. In [25], this model was applied to determine multiphase
parameter such as relative permeability and capillary pressure and
their dependence on the invading phase saturation. Since on any
given network, each simulation is deterministic, the analysis in [25]
required averaging of simulations performed with several stochas-
tically generated networks having the same global properties. It
should be noted that the two-dimensional pore networks consid-
ered here have limitations as pointed out by Chatzis and Dullien
[51], as well as by Chapuis et al. [27] in their very recent GDL
study. In particular 2D and 3D simulations yield different percola-
tion thresholds for a given coordination number. Notwithstanding
these limitations, the 2D network systems are expected to be quali-
tatively representative and allow acceptable simulation turnaround
times to explore the concept of biasing for the purpose of control-
ling water transport. The deterministic nature of the pore network
invasion process provides is a useful feature allowing compari-
son with experimental results. Fig. 1 shows a sample numerical
7632 A. Bazylak et al. / Electrochimica Acta 53 (2008) 7630–7637

breakthrough pattern in relation to a fuel cell schematic. For fur-

ther details on the pore network model, readers are directed to
Markicevic et al. [25].

2.2. Microfluidic pore networks

Using Matlab 7.0, a pore network was designed, and the

coordinates were then imported into Autocad 2007, where the
photomasks were drafted. All photomasks were then printed by
CAD/Art Services, Inc. (California, USA), with a maximum resolution
of 20,000 dpi. The microfabrication process begun with pattern-
ing a photoresist resin layer on a silicon substrate. A thin layer of
SU8-25 photoresist (25 ␮m thick) was distributed over the silicon
substrate through the use of a spin coater. A 25 ␮m height was
chosen for the experimental pore network fabrication due to the
manufacturing constraints in the laboratory. The photoresist layer

Fig. 2. Flow diagram illustrating the pore network design and fabrication procedure.
Fig. 3. Schematic illustrating various methods to construct a microfluidic pore net-
work: (a) randomly perturbing a regular lattice of square obstacles, (b) sequentially
placing square obstacles in random locations, and (c) sequentially placing square
obstacles of varying sizes in random locations.
 A. Bazylak et al. / Electrochimica Acta 53 (2008) 7630–7637 7633

was then soft baked to evaporate solvent and condense the photore-
sist. The photoresist layer was exposed to a near-UV light source
(350–400 nm), which allowed for photoresist cross-linkage. The
substrates were baked to complete the photoresist cross-linking
reaction. Any SU8-25 photoresist substrate that was not exposed
to UV light was removed with chemical etching. The pore network
was then fabricated by curing a polymer on the patterned template,
removing the polymer from the template and sealing the template
to a planar surface. Through this rapid prototyping technique, sev-
eral microfluidic pore networks were generated and tested within
a short period of time. The reader is directed to recent work by
Berejnov et al. [46] for specific microfabrication details. Typically,
in PEM fuel cells the capillary number, Ca , is on the order of 10−8
for two-phase flow in the GDL. In this work, the employed flow
rates can be characterized by Ca on the order of 10−7 to 10−8 , which
reasonably represents the flow conditions in a GDL of an operating
PEMFC.
3. Pore network design
There are several possible approaches to biasing a network.
In the present study, two biasing functions were applied to the
pore (throat) distribution: a random function and a determinis-
tic function. From an engineering point of view, introducing the
deterministic function allows networks that are intrinsically ran-
dom to be imparted with new properties for the porous medium.
The random function defines the local random properties of the
network, while the deterministic function defines the deviation of
these properties on the scale of the entire network. Generating a
network using the combination of these functions yields a random
network with a biased pattern. If the dominance of the random
and deterministic functions are comparable, then the preferential
pathway for water cannot be determined a priori, but rather the
pattern needs to be determined according to the network laws. This
approach of combining deterministic and random biasing functions

Fig. 4. Perturbation procedure for generating photomask: (a) regular square lattice, (b) translation of regular lattice in both the horizontal and vertical directions, (c)
uniform distribution of horizontal and vertical translations, (d) resulting arrangement of random pore network, (e) throat width distribution, and (f) hydraulic throat radius
distribution.
7634 A. Bazylak et al. / Electrochimica Acta 53 (2008) 7630–7637

also provides a systematic way of characterizing and modelling a

network/porous medium under realistic conditions. For instance
when considering a purely random network under some deforma-
tion, such as expansion or compression, the characteristics of the
medium may be related to a deterministic function of a similar
nature.

3.1. Pore network structure

A random pore network can be generated in a number of ways.

Starting with a blank two-dimensional domain, as shown in Fig. 3,
square obstacles can be arranged randomly to fill the domain, while
overlapping is prohibited. The spaces that are not covered by these
square obstacles will constitute the throats and pores. In Fig. 3(a),
the network is generated by randomly perturbing a regular lattice
of square obstacles. In Fig. 3(b), the domain is filled by adding one
square obstacle at a time, until the maximum number of obstacles
is reached. Fig. 3(c) shows a randomly arranged network, which is
generated by first filling the network with square obstacles (obsta-
cle 1). Once the maximum number of obstacles is placed in the
domain, the remaining space is filled as tightly as possible with-
out over lapping with obstacles that are half the size as obstacle 1
(obstacle 2). Again, once the maximum number of obstacles of type
2 are placed, smaller type obstacles can be used to fill the domain.
As shown in Fig. 3, the benefit of utilizing method (a) is that a higher
packing density can be obtained, while maintaining a constant con-
nectivity of 4. In method (b), the porosity of the media is very high
and difficult to control. In method (c), the obstacle size variation
provides a high packing density, but at the cost of varying connec-
tivity and network complexity. All pore networks in this work were
generated with the first method (Fig. 3(a)), whereby a regular lat-
tice of square obstacles is randomly perturbed to generate a random
network.

3.2. Randomization

The pore network generation procedure begins with a regular

lattice of square obstacles. Fig. 4(a) shows a schematic of the regu-
lar lattice, where w is the throat width, and L is the obstacle square
side length. A random translational perturbation is then applied
to each obstacle, where X refers to the horizontal translation,
and Y refers to the vertical translation of the original obstacle.
The uniform random distribution of translational perturbations is
shown in Fig. 4(c). Fig. 4(d) is the resulting schematic of a portion
of the random pore network. The resulting throat width distribu-
tion is shown in Fig. 4(e). For a uniform channel height of 25 ␮m, the
hydraulic throat radii distribution is shown in Fig. 4(f). All hydraulic
throat radii employed in this work are characteristic of Toray TGP-
H-060 gas diffusion layers [50]. The hydraulic radius is calculated
with the following formula: rh = wt/(w + t), where w is the throat
width, and t is the throat thickness.

3.2.1. Isotropic vs. diagonal biasing

To generate an isotropic random pore network, the transla-
tion of the regular lattice in the horizontal direction X and the
vertical direction Y must be independent. It is also possible to
generate a random pore network with a diagonal bias by coupling
the translation of the regular lattice in the horizontal and vertical
directions. To implement a diagonal bias of bias = 45◦ , for each
square obstacle, Y = X. The ratio of vertical to horizontal obsta-
cle translations can be adjusted to implement the desired diagonal
bias for 0◦ < bias < 90◦ , as shown in
Fig. 5. Procedure for generating photomask for a pore network with a prescribed
Y radial gradient: (a) regular square lattice, (b) radial translation of individual square
tan bias = (1) obstacles, (c) network with radial gradient, and (d) random perturbations applied
X to the network with a radial gradient.
 A. Bazylak et al. / Electrochimica Acta 53 (2008) 7630–7637 7635

3.2.2. Radial gradients square obstacle from the central coordinate axis is labeled r. The
The procedure to generate a pore network with a radial gradient square obstacle is then translated in the radial direction to a new
is illustrated in Fig. 5. The first step is to begin with a regular lat- radial distance, R, from the origin, as shown in Fig. 5(b). The new
tice of square obstacles, as shown in Fig. 5(a). The distance of each radial distance, R, is transformed by the following expression:

Fig. 6. Comparison of numerical and experimental random pore network flow patterns at breakthrough: (a) isotropic perturbations, (b) diagonally biased perturbations, (c)
radial gradient, and (d) checkerboard pattern with multiple radial gradient patches. Each numerical and experimental result comparison is accompanied by a photomask
schematic with a reduced number of obstacles for clarity (along left column).
7636 A. Bazylak et al. / Electrochimica Acta 53 (2008) 7630–7637
˛r 2
R→r+
rmax
where, ˛ is the coefficient that is employed to vary the radial gradi-
ent of the network. The radial gradient, G, is the increase in throat
width per obstacle, beginning from the central coordinate axis.
After a regular network (with radial distance, r) is transformed
into a new network (with radial distance, R), the radial gradient,
G, was numerically measured. Numerical tests yield the following
expression between G, ˛, rmax , and N:
4rmax
G= ײ
N2
Fig. 5(c) shows the resulting pore network design when a radial
gradient has been applied to the regular lattice of square obstacles.
The final step is to apply a random translation with independent
horizontal (X) and vertical (Y ) perturbations to each obsta-
cle. An example of the resulting pore network pattern is shown
in Fig. 5(d). To quantify the effect of an applied radial gradient on a
random network, the radial gradient is normalized with respect to
the average throat width of a network without a radial gradient.
The network saturation is calculated to further quantify the
effect of liquid transport in a network with a radial gradient. The
saturation of the pore network is defined as the number of filled
throats divided by the total number of throats in the system, as
described in
Nfilled
s=
Ntotal
4. Results and discussion
Numerical and experimental pore networks with identical net-
work designs were invaded with liquid water. Comparison of
the flow patterns and associated characteristics of the numeri-
cal and experimental networks, which have nominally identical
microstructures, confirms the applicability of the pore network
rules and dominant mechanisms in the processes considered.
Fig. 6 shows the numerically and experimentally obtained pat-
terns at breakthrough for 48 × 48 pore networks with (a) isotropic
perturbations, (b) diagonally biased perturbations, (c) a radial gra-
dient, and (d) multiple radial gradients. For each network design
in Fig. 6 shown from left to right are: a scaled down version of
the photomask, the numerical breakthrough and the experimen-
tal breakthrough patterns. By comparing the percolation patterns
shown in Fig. 6(a) and (b), the effect of a diagonal bias on the pore
network compared to the isotropically random pore network is
evident, even though the difference cannot be detected when com-
paring their respective photomasks alone. Fig. 6(c) illustrates the
effect of a radial gradient on the liquid water transport. With a pos-
itive radial gradient, larger throat sizes are located around the outer
edges of the pore network. As expected in a hydrophobic medium,
the liquid penetrates the throats with the lowest threshold capillary
pressure, which are the largest throats. With an applied radial gra-
dient, the liquid saturation is zero at the center of the network and
increases with radial distance from the center. With this radial gra-
dient design, liquid water percolates along the outer left and right
sides of the network, until breakthrough is achieved. The radial gra-
dient effectively provides directed water transport along the sides
of the media. The percolation thresholds for 2D and 3D systems
are expected to be different in general, and though the results pre-
sented here should be interpreted with caution in the context of
an actual GDL, they are expected to provide the right trends and
be qualitatively useful for guiding the design of novel media for
directed water transport.
A larger pore network with several patches with independent
(2)
radial gradients is shown in Fig. 6(d). The individual patches have
been highlighted with gray and black to indicate their size and loca-
tion. The percolation breakthrough pattern associated with that
checkerboard pattern illustrates the preferential pathways created
from the application of radial gradients.
For all the trials shown in Fig. 6, the experimental results agree
well with the numerical results, especially when considering that
the dynamic effects and contact angle hysteresis that exist in the
microfluidic pore networks were not considered in the model.
The agreement between the experimental and numerical results
(3) is especially highlighted by the trials shown in Fig. 6(c) and (d),
where the percolation pathways leading to breakthrough occur in
the same place. The small deviations at the small (nodal)-scale level
between experiments and simulations are expected and inherent
to the fabrication technique which results in micro-feature vari-
ations between individual microchip channels. It is interesting to
note that when comparing the numerical and experimental flow
patterns of deterministic network designs, the overall water-air
flow pattern is of more importance than the exact saturation val-
ues. For example, the biased networks shown in Fig. 6 exhibit very
distinct flow structures that correspond well between the numer-
ical and experimental results, while their specific saturation value
could correspond to any number of arbitrary flow patterns.
Of interest from a GDL design perspective is the relationship
between the degree of geometric perturbation of the network and
its influence on water transport. In an operating fuel cell, it is
(4) desirable to optimize the access of reactants to the catalyst layer.
Reactant transport takes place via a combination of convection and
diffusion. From the view point of convective transport, no defi-
nite relationship between relative permeability and saturation has
yet been established for GDL media, but it is generally accepted
that the functional relationship is of the form krg = (1 − s)n , i.e.
higher relative permeabilities are obtained at lower saturations.
The dependence of saturation on increasing radial gradient was
studied for a 100 × 100 pore network, as shown in Fig. 7. We
note that the patterns obtained with the larger network are iden-
tical to those in Fig. 6, which were performed with a 48 × 48
pore network. Fig. 7 shows that as the normalized radial gradient
increases, the saturation of the network decreases. This is attributed
to the increasing localization and channeling of the invading phase
clearly shown by the results. The saturation of individual deter-
Fig. 7. Dependence of saturation on increasing radial gradient for a 100 × 100
numerical pore network. The radial gradient is normalized by the average throat
width (30 ␮m). Filled squares represent single realizations, where empty squares
represent an average value over 100 stochastic realizations with the standard devia-
tion marked with error bars. The shaded region of the plot shows the expected range
of saturation values.
 A. Bazylak et al. / Electrochimica Acta 53 (2008) 7630–7637
ministic simulations is plotted and shown with solid black squares.
Since pore network modelling is stochastic by nature, 100 ran-
dom simulations were performed for each of four points (shown
with non-filled black squares), with the average value and standard
deviation marked with error bars. The largest standard devia-
tion corresponds to pore networks without a radial gradient, and
decreases with increasing radial gradient. With a radial gradient
of 0.016/obstacle (i.e. the throat width increases by 1.6%/obstacle
in the radial direction), the average saturation decreased by 43%
compared to the average pore network without a radial gradient.
5. Conclusions
Numerical and experimental pore networks were employed to
explore the design of gas diffusion media for directed water trans-
port through the use of pore network modelling and microfluidic
fabrication techniques. The concept of biasing was shown to be
effective from the 2D numerical simulations and planar experi-
mental networks. These 2D results can guide more representative
and complex 3D network simulations that need to be performed
in the future for a full quantification of the patterns and transport
processes.
Randomized pore networks with diagonal and radial biasing
show a marked influence on liquid water transport through the
two-dimensional media. Unidirectional water transport is achieved
through the use of diagonally biased networks. Radial biasing is
shown to have the beneficial effect of decreasing the network sat-
uration, which is a desirable quality in fuel cell gas diffusion layers.
It was found that with a radial gradient of 1.6% per obstacle, the
average saturation decreased by 43% compared to the average pore
network without a radial gradient. Both the numerical and exper-
imental pore network liquid breakthrough patterns were in good
agreement. The results from this paper indicate that porous media
can be designed to have these directed water transport features,
which can be applied to PEM GDL designs.
Acknowledgments
The authors are grateful for the financial support of the Natural
Sciences and Engineering Research Council (NSERC) of Canada, the
Canada Research Chairs Program, and the National Research Council
(NRC). The authors would also like to acknowledge the support of
the Canadian Foundation for Innovation (CFI).
References
[1] K. Jiao, B. Zhou, Journal of Power Sources 169 (2007) 296.
[2] T. Berning, N. Djilali, Journal of the Electrochemical Society 150 (12) (2003)
A1589.
[3] A. Taniguchi, T. Akita, K. Yasuda, Y. Miyazaki, Journal of Power Sources 130
(2004) 42.
[4] S. Litster, D. Sinton, N. Djilali, Journal of Power Sources 154 (2006) 95.
[5] A. Bazylak, D. Sinton, Z.-S. Liu, N. Djilali, Journal of Power Sources 163 (2007)
784.
[6] A. Bazylak, D. Sinton, N. Djilali, Journal of Power Sources 176 (1) (2007) 240.
[7] A. Weber, R. Darling, Journal of Power Sources 168 (2007) 191.
[8] C. Buie, J. Posner, T. Fabian, S.-W. Cha, D. Kim, F. Prinz, J. Eaton, J. Santiago, Journal
of Power Sources 161 (2006) 191.
7637
[9] S. Litster, C. Buie, T. Fabian, J. Eaton, J. Santiago, Journal of the Electrochemical
Society 154 (10) (2007) B1049.
[10] I. Fatt, The network model of porous media I. Capillary pressure characteristics,
Petroleum Transactions, the American Institute of Mining, Metallurgical, and
Petroleum Engineers (AIME) 207 (1956) 144.
[11] I. Fatt, The American Institute of Mining, Metallurgical, and Petroleum Engi-
neers (AIME) 207 (1956) 160.
[12] I. Fatt, The American Institute of Mining, Metallurgical, and Petroleum Engi-
neers (AIME) 207 (1956) 164.
[13] S. Broadbent, J. Hammersley, Proceedings of the Cambridge Philosophical Soci-
ety 53 (1957) 629.
[14] D. Wilkinson, J. Willemsen, Journal of Physics A: Mathematical and General 16
(1983) 3365.
[15] R. Chandler, J. Koplik, K. Lerman, J. Willemsen, Journal of Fluid Mechanics 119
(1982) 249.
[16] A. Payatakes, Annual Review of Fluid Mechanics 14 (1982) 365.
[17] J. Koplik, T. Lasseter, Society of Petroleum Engineers Journal 25 (1) (1985) 89.
[18] A.B. Dixit, S. McDougall, K. Sorbie, J. Buckley, SPE Reservoir Evaluation and
Engineering 2 (1) (1999) 25.
[19] M. Valavanides, A. Payatakes, Advances in Water Resources 24 (2001) 385.
[20] L. Paterson, Physical Review E 66 (2002) 1.
[21] M. Bernadiner, Transport in Porous Media 30 (1998) 251.
[22] K. Thompson, American Institute of Chemical Engineers Journal 48 (7) (2002)
1369.
[23] M. Blunt, Current Opinion in Colloid & Interface Science 6 (2001) 197.
[24] J. Gostick, M. Ioannidis, M. Fowler, M. Pritzker, Journal of Power Sources 173 (1)
(2007) 277.
[25] B. Markicevic, A. Bazylak, N. Djilali, Journal of Power Sources 171 (2) (2007) 706.
[26] P. Sinha, C.-Y. Wang, Electrochimica Acta 52 (2007) 7936.
[27] O. Chapuis, M. Prat, M. Quintard, E. Chane-Kane, O. Guillot, N. Mayer, Journal of
Power Sources 178 (2008) 258.
[28] M.B. Isichenko, Review of Modern Physics 64 (4) (1992) 961.
[29] M. Sahimi, Journal De Physique I France 4 (9) (1994) 1263.
[30] S. Cordero, I. Kornhauser, A. Domı́nguez, C. Felipe, F. Esparza, R.H. López, A.M.
Vidales, J.L. Riccardo, G. Zgrablich, Particle & Particle Systems Characteristics 21
(2004) 101.
[31] A. Vidales, R. Faccio, J. Riccardo, E. Miranda, G. Zgrablich, Physica A 218 (1995)
19.
[32] M. Sahimi, S. Mukhopadhyay, Physical Review E 54 (4) (1996) 3870.
[33] M. Knackstedt, A. Sheppard, M. Sahimi, Advances in Water Resources 24 (2001)
257.
[34] R. López, A. Vidales, A.D. Ortiz, G. Zgrablich, Colloids and Surfaces A: Physico-
chemical and Engineering Aspects 300 (2007) 122.
[35] R. Lenormand, C. Zarcone, Physical Review Letters 54 (20) (1985) 2226.
[36] R. Lenormand, E. Touboul, C. Zarcone, Journal of Fluid Mechanics 189 (1988)
165.
[37] R. Lenormand, C. Zarcone, Transport in Porous Media 4 (1989) 599.
[38] A. Birovljev, L. Furuberg, F. Feder, T. Jossang, A. Maloy, A. Aharony, Physical
Review Letters 67 (5) (1991) 584.
[39] O. Vizika, D. Avraam, A. Payatakes, Journal of Colloid and Interface Science 165
(1994) 386.
[40] B. Xu, Y. Yortsos, Physical Review E 57 (1) (1998) 739.
[41] C. Tsakiroglou, M. Theodoropoulou, V. Karoutsos, D. Papanicolaou, V. Sygouni,
Journal of Colloid Interface Science 267 (2003) 217.
[42] E. Kossel, M. Weber, R. Kimmich, Solid State Nuclear Magnetic Resonance 25
(2004) 28.
[43] C. Perrin, P. Tardy, K. Sorbie, J. Crawshaw, Journal of Colloid and Interface Science
295 (2006) 542.
[44] M. van Dijke, K. Sorbie, A. Danesh, SPE Journal 9 (1) (2004) 57.
[45] M. van Dijke, K. Sorbie, M. Sohrabi, A. Danesh, Journal of Petroleum Science and
Engineering 52 (2006) 71.
[46] V. Berejnov, N. Djilali, D. Sinton, Lab Chip 8 (2008) 689.
[47] R. Lenormand, C. Zarcone, Proceedings of the 59th Annual Technical Conference
and Exhibition of SPE AIME, SPE, Houston, Texas, 1984, p. 13264.
[48] U. Oxaal, M. Murat, F. Bojer, A. Aharony, J. Feder, T. Jossang, Nature 329 (1987)
32.
[49] M. Theodoropoulou, V. Sygouni, V. Karoutsos, C. Tsakiroglou, International Jour-
nal of Multiphase Flow 31 (2005) 1155.
[50] M. Mathias, J. Roth, J. Fleming, W. Lehnert, Handbook of Fuel Cells, John Wiley
& Sons, Ltd., New York, 2003, pp. 1–21.
[51] I. Chatzis, F. Dullien, Canadian Journal of Petroleum Technology (January–
March, 1977) 97.