Journal of Power Sources xxx (2005) xxx–xxx

Capillary pressure and hydrophilic porosity in gas diffusion

layers for polymer electrolyte fuel cells
Jeffrey T. Gostick a , Michael W. Fowler a,∗ , Marios A. Ioannidis a ,
Mark D. Pritzker a , Y.M. Volfkovich b , A. Sakars b
a Department of Chemical Engineering, University of Waterloo, Waterloo, Ont., Canada N2L 3G1
b Porotech Ltd., Vaughan, Ont., Canada L4L 5Y9

Received 27 April 2005; received in revised form 20 May 2005; accepted 23 May 2005

Abstract

Capillary pressure versus saturation curves for drainage of a wetting phase were measured for several gas diffusion layers that are commonly
used in polymer electrolyte membrane fuel cells. The technique employed can measure capillary pressure curves for both the total pore network
and the pore network consisting of only hydrophilic pores. This enables the determination of capillary pressure curves directly relevant to
the study of gas diffusion layer flooding. The overall distributions compared well with mercury intrusion data. It was found that the pore
size distribution for the hydrophilic pores were similar in shape to the overall distribution for standard substrate materials. Materials with a
microporous layer did not follow this trend and the microporous layer was found to be completely hydrophobic. Due to their similarity, the
overall and hydrophilic capillary pressure curves for all materials could be correlated using a single Leverett J-function. The results were
described by several standard capillary models, the parameters of which can be further used to predict the relative permeability of the phases.
© 2005 Elsevier B.V. All rights reserved.

Keywords: Hydrophilic; Hydrophobic; Porosity; Capillary pressure; Cathode; Water management

1. Introduction the latter effects of liquid water, GDLs are typically coated
with a non-wetting polymer such as polytetrafluoroethylene
Increasing the maximum power density is a key objec- (PTFE) to create hydrophobic surfaces and pores throughout
tive of polymer electrolyte membrane fuel cell (PEMFC) the GDL that are thought to remain free of water and facilitate
research. Improvements in catalyst activity, accessibility and transport of oxygen to the catalyst layer.
utilization have led to a dramatic increase in reaction rates [1]. Pore scale phenomena associated with the movement of
The corresponding increase in water generation, however, has liquid water and the interplay between the vapor and liquid
tended to negate these improvements due to flooding of the phases have recently received considerable attention. Numer-
cathode with liquid water at higher current densities. The ous PEMFC models have been proposed [2–20] and recent
presence of excessive liquid water in the cathode can debil- reviews are available [21,22]. However, analysis of these
itate the cell in two ways. On a microscopic scale, liquid models has made evident the scarcity of constitutive rela-
water covers catalyst particles, thereby increasing the mass tions that describe the distribution and capillary flow of liquid
transfer resistance and reducing oxygen access to the catalyst water in GDL materials. To date, very limited experimental
sites. Macroscopically, liquid water may fill the pore network information has been reported on the dependence of capillary
within the gas diffusion layer (GDL) support and decrease the pressure (pc ) on water saturation (sw ) or on the dependence
effective diffusivity of oxygen through the layer. To counter of water relative permeability (krw ) on saturation in different
GDL materials. This hinders progress in model validation.
∗ Corresponding author. Tel.: +1 888 4567x3415; fax: +1 519 746 4979. Several studies focusing on the morphological properties
E-mail address: mfowler@uwaterloo.ca (M.W. Fowler). of the porous GDLs have reported Hg–Air capillary pres-

0378-7753/$ – see front matter © 2005 Elsevier B.V. All rights reserved.
doi:10.1016/j.jpowsour.2005.05.086

POWER-7121; No. of Pages 13

2 J.T. Gostick et al. / Journal of Power Sources xxx (2005) xxx–xxx

[24] used mercury intrusion porosimetry (MIP) to examine

Nomenclature the consequences of various microporous layer (MPL) man-
ufacturing methods on the morphology of the MPL. Jordan
A area (m2 ) et al. [25] and Passalacqua et al. [26] experimented with dif-
BC Brooks–Corey model, given by Eq. (10) ferent types of carbon black in the MPL. Antolini et al. [27]
D diameter of fibers in GDL model used MIP to study the effect of PTFE addition on the total
J Leverett J-function, defined by Eq. (16) pore volume of MPL. Other works have focused specifically
k permeability (m2 ) on the GDL properties [28–30], but were aimed at qualita-
kr relative permeability tively explaining PEMFC performance on the basis of GDL
m VG model fitting parameter morphological features. In none of these studies has a sys-
md mass of dry sample (kg) tematic investigation of the capillary behavior in GDLs been
ms mass of saturated sample (kg) conducted. All of the research has been carried out using
n VG model fitting parameter mercury intrusion porosimetry (MIP), which cannot distin-
pc capillary pressure (N m−2 ) guish between the hydrophilic and hydrophobic components
pcb breakthrough pressure used to fit BC and VG of GDL porosity.
model (N m−2 ) In the absence of GDL specific pc –sw data, various
RMS root mean squared
 error: assumptions and approximations have been employed. One


n
(pobserved −pc
predicted 2
) common, but poorly justified approach [2–8], has been to use
RMS = n1 c
pc observed
a polynomial fit, originally obtained by Udell [31], of capil-
i=1
s saturation, defined by Eq. (1) lary pressure data measured by Leverett for water imbibition
S spacing between fibers in GDL model in water-wet unconsolidated sand packs [32]. Acknowledg-
VB bulk volume (m3 ) ing the lack of applicable data for GDLs, some researchers
Vp pore volume (m3 ) have arbitrarily assumed a linear dependence between capil-
VG van Genuchten model, given by Eq. (9) lary pressure and saturation [10,11,14]. A few attempts have
been made to model pc –swp relationships specifically for
Greek letters GDLs. Divisek et al. [17] have fitted an empirical relation-
ε porosity ship to data generated from a model developed by Ustohal
θ contact angle (rad) et al. [33]. Pisani et al. [18] and Natarajan and Nguyen
λ BC model fitting parameter [12,13] adjusted parameters of a capillary pressure model to
ρ density (kg m−3 ) match observed polarization behavior. Only Weber et al. [16]
σ surface tension (N m−1 ) expressly attempted to empirically fit mercury porosimetry
ϕ ratio of substrate pore volume to total data for use in their models. These various approaches for cal-
pore volume culating pc –sw curves have produced a wide range of results,
leading to predictions of capillary pressure that range over
Superscripts several orders of magnitude.
BA SGL10BA The heterogeneous nature of GDL materials requires
BB SGL10BB that attention be given to the coexistence of hydrophilic
eff effective and hydrophobic pores spaces within these media. Few
S free substrate of SGL10BB, not intruded researchers have addressed this point [10,16,18] and only
by MPL Weber et al. [16] have attempted to model it. The latter authors
have assumed that hydrophilic and hydrophobic pore spaces
Subscripts have separate size distributions, but that each has the same
B bulk shape as the overall distribution, while the proportion of each
G gas pore type is dictated by the amount of PTFE in the GDL and
Hi hydrophilic its expected effect on hydrophobicity. Although this approach
nwp non-wetting phase is more realistic, the assumption that the hydrophilic and
r residual hydrophobic pore size distributions have shapes similar to
T total the overall pore size distribution has not been tested against
wp wetting phase experimental data. Furthermore, to implement this approach,
information about the relative proportions of hydrophilic and
hydrophobic porosity is required. In this respect, only slightly
more information is available since few studies have inves-
sure curves or pore size distributions, but none have been tigated the amount of hydrophilic porosity in PTFE-treated
concerned with investigating the dependence of capillary GDLs [28,34,35].
pressure on saturation for the water–air system, as required The objective of the present work is to determine the pore
for modeling water transport. Lee et al. [23] and Kong et al. size distribution and void fractions associated with the overall
 J.T. Gostick et al. / Journal of Power Sources xxx (2005) xxx–xxx 3

and hydrophilic porosity of various commercially available iment, the non-wetting phase (mercury) saturation of an ini-
gas diffusion layers. Such a breakdown of the overall porosity tially evacuated sample is measured as the capillary pressure,
distribution of a GDL into its hydrophobic and hydrophilic pc,Hg–Air , is gradually increased, providing the capillary pres-
components has not been previously reported. Here, the sure curve for drainage of a strongly wetting phase [40]. Pore
dependence of capillary pressure on wetting phase satura- size information in terms of an equivalent cylindrical capil-
tion along a desaturation path (drainage) is directly measured lary size may be extracted from capillary pressure data using
using the method of “standard porosimetry” (MSP) [36] with the Young–Laplace equation:
octane–air and water–air as the working fluid pairs. Measure-
2σHg–Air cos θHg–Air
ments with octane, which strongly wets both hydrophobic and r= (1)
hydrophilic surfaces, are compared to MIP measurements. pc,Hg–Air
Measurements with water enable the determination of the where r is the pore radius, σ Hg–Air the surface tension of
pore size distribution of hydrophilic porosity of interest in the mercury–air interface and θ Hg–Air the contact angle of
fuel cell modeling. The effect of temperature is not investi- mercury on the solid surface of the pore wall. In general,
gated in the present study despite the documented dependence inferring the pore size distribution from the results of a
of capillary pressure curves on temperature [37,38]. Conse- MIP experiment is fraught with difficulties stemming from
quently, the capillary pressure data obtained are not directly pore accessibility limitations [41]. Larger pores shielded by
applicable to operating fuel cells. Nonetheless, these data smaller ones are only intruded by mercury at the capillary
constitute a first step towards understanding the capillary pressure corresponding to entry of mercury in the smaller
behavior of water–air–GDL systems. Fitting of the capillary pores and their volume is incorrectly attributed to smaller
pressure data using the Brooks–Corey and van Genuchten pores. In chemically heterogeneous materials such as a GDL,
models is also discussed. Finally, in light of the MSP results, the assumption that mercury exhibits a single, well-defined
capillary pressure data specific to the hydrophobic MPL are contact angle on all solid surfaces is also challenged. Par-
obtained. tial coverage of the carbon fibers in a GDL by PTFE creates
non-uniformity in θ Hg–Air , meaning that two pores of similar
size, but with different wettability, are intruded at different
2. Materials and methods pressures. Uncertainty regarding the value of θ Hg–Air thus
confounds the conversion of capillary pressure measured with
2.1. GDL materials one fluid (e.g. Hg–Air) to equivalent capillary pressure for
another fluid pair (e.g., water–air), on the basis of the follow-
A representative range of GDL materials was selected for ing equation:
investigation in this work, including cloth, felt and paper with
σ2−Air cos θ2−Air
and without a MPL. The materials used were SGL 10BA plain pc,2 = pc,1 (2)
paper and SGL 10BB paper with microporous layer (SGL σ1−Air cos θ1−Air
Carbon Group, Short Hills, NJ), Toray TGPH-090 plain paper in which σ 1–Air and σ 2–Air are the surface tensions of the
(Toray Corp., Tokyo, Japan), E-TEK Cloth ‘A’ (E-TEK Inc., two liquids, θ 1–Air and θ 2–Air the contact of angles of the two
Somerset, NJ), as well as Lyflex 484C and 352C felts (Lydall, liquids on the material and pc,1 the experimentally measured
Manchester, CT). The relevant properties of each material are capillary pressure.
listed in Table 1. To obtain an estimate for the contact angles to be used in
Eq. (2), the contact angle of each liquid on the surface of each
2.2. Mercury intrusion porosimetry GDL was measured by image analysis using a video contact
angle system (AST Products 2500XE, Billerica, MA). The
Mercury intrusion porosimetry (MIP) is widely used to repeatability of these measurements was ±1◦ . The observed
measure the distribution of accessible pore volume by entry contact angles were then corrected for the effects of surface
pressure for pores between 1 nm and 1 mm. In an MIP exper- roughness and porosity using the Cassie–Baxter equation for

Table 1
Physical properties of GDL materials testeda
Material Type ε k (m2 × 1012 ) ρB (g mL−1 ) Thickness (␮m) PTFE (wt.%)
SGL10BA Paper 0.88 18.0 [29] 0.22 380 5
SGL10BB Paper w/MPL 0.84 0.33 [29] 0.30 420 5 (in paper)
Toray 090 Paper 0.78 8.3 0.44 190 0
E-Tek Cloth ‘A’ Cloth – 6.3b 0.33 350 0
Lyflex 484C Felt 0.83 – 0.30 417 3.1
Lyflex 352C Felt 0.74 – 0.47 268 6.2
a Given by manufacturer except where noted.
b Calculated by the method of Happel [39], with fiber diameter of 5 ␮m determined by fitting model to Toray 090, and porosity taken as 0.75.
4 J.T. Gostick et al. / Journal of Power Sources xxx (2005) xxx–xxx

Table 2
Observed and estimated contact angles of mercury on GDL materials
Material Observed θ [◦ ] Effective θ [◦ ]
SGL10BA 148 108
SGL10BB/MPLa 148/153 108/119
Toray 090 146 112
E-Tek Cloth ‘A’ 147 114
Lyflex 484C 147 110
Lyflex 352C 148 114
Graphite 124 [44] –
PTFE 150 [43] –
a For a drop deposited on MPL side. A porosity value of 0.72 was used in

Eq. (3), based on calculations detailed in Section 3.2.6.

S
S D
f2 = = S
(6)
S+D D +1
Fig. 1. Schematic diagram of fiber structure used to determine f1 and f2 in
Eqs. (5) and (6). The results of this analysis are given in Table 2, where the esti-
mated effective contact angles are also compared to reported
porous surfaces [42] modified for surface roughness [43]: contact angles of mercury on smooth graphite and PTFE
cos θObs = bf1 cos θEff − f2 (3) [43,44].

where θ Obs is the measured contact angle, b the ratio of actual 2.3. Method of standard porosimetry
area of contact between the drop and the solid portion of the
surface to the projected area (b = π/2 for cylinders) and f1 The method of standard porosimetry (MSP) offers two
and f2 the fractions of the GDL surface occupied by fiber and distinct advantages over MIP [36]. Since mercury is a non-
void, respectively. The value of θ Eff thus obtained is a rough wetting fluid to both carbon and PTFE surfaces, an MIP
estimate of the contact angle of the liquid on the chemically measurement cannot distinguish between hydrophilic and
heterogeneous fibers. hydrophobic porosity. MSP can be performed with any wet-
To estimate the values of f1 and f2 for the GDLs, the fiber ting fluid as the working liquid. Using water as the sat-
network model of Nam and Kaviany was used [10]. In this urating fluid permits one to determine the distribution of
model the GDL is composed of a stack of interwoven screens hydrophilic pores only. The distribution of the overall poros-
(Fig. 1). Spacing between each parallel fiber is equal in both ity can be obtained by using a strongly wetting fluid such
directions and this is also the spacing between the layers. For as octane, which fully wets both graphite and PTFE sur-
simplicity the fibers are allowed to intersect. The unit cell faces uniformly. The second advantage of this technique is
in this model has dimensions of S + D in all three directions, that the pc –swp data are obtained directly for the fluid–solid
giving a volume of (S + D)3 . To calculate porosity, the fiber system of interest. Therefore, the results are not subject
volume is subtracted from the volume of the unit cell and the to the uncertainties in contact angle introduced by the use
following relationship is obtained: of Eq. (2). To date, application of the MSP technique
VUnit Cell − 2VFiber + VIntersection has been limited to the characterization of cermet materi-
ε= als in solid oxide fuel cells [45] and Nafion membranes
VUnit Cell
[46].
(S + D)3 − 2 π4 D2 (S + D) + 23 D3 The method of standard porosimetry is based on the prin-
= ciple of capillary equilibrium. When two partially saturated
(S + D)3
 S 3  S 2 S S porous bodies are in contact, the system moves toward an
D +3 D +3 D + 1 − π2 D − π
+ 2
equilibrium state where the capillary pressures of the liq-
=  S 3  S 2 S
2 3
uid in both bodies are equal. MSP exploits this phenomenon
D +3 D +3 D +1
(4) by placing the unknown sample in capillary contact with a
standard sample having a known pc –swp curve. A schematic
Eq. (4) may be solved for S/D given ε, thus enabling the diagram of the steps required to obtain the pc –swp curve of
calculation of the fractions f1 and f2 from the following an unknown sample is shown in Fig. 2. When the system is
expressions: deemed to be in capillary equilibrium, the mass of each sam-
ple is determined. Knowledge of the mass of each sample
D 1
f1 = = S
(5) permits the determination of its wetting phase saturation swp ,
S+D D +1 defined as the ratio of the volume of liquid in the sample to
 J.T. Gostick et al. / Journal of Power Sources xxx (2005) xxx–xxx 5

Fig. 2. Procedure for obtaining GDL capillary pressure curves using MSP.

the total void volume of the sample, i.e., to a pressure of 10 mTorr. The chamber was then flooded with
ms − md the working liquid. When octane was the working fluid, this
swp = (7) procedure was sufficient to fully saturate the sample, since
ρL εVB
octane spontaneously imbibes into all pores due to its high
where ms is the mass of the saturated sample, md the mass affinity for both graphite and PTFE surfaces. When water
of the dry sample, ρL the density of the liquid, ε the porosity was the working fluid, an extra step was required to ensure
of the pore network being measured and Vb the sample bulk that all hydrophilic pores had been filled. After removing the
volume. The capillary pressure, pc , in the standard sample sample from the vacuum chamber, it was placed in water at
corresponding to its saturation can then be determined from 80 ◦ C for several hours. The elevated temperature reduced the
its known capillary pressure curve (Fig. 2a). At equilibrium contact angle of water on PTFE, allowing deeper intrusion
this is also equal to the capillary pressure in the unknown into the material to reach any inaccessible hydrophilic pores.
sample (Fig. 2b). This represents a direct measurement of The sample was returned to room temperature before testing.
the capillary pressure in the unknown sample with the fluid A Quantachrome Poremaster (Boynton Beach, FL) was
of interest. This process can then be repeated over a range used for MIP testing. Triply distilled ACS grade mercury
of saturation values to yield the capillary pressure curve of (99.99% purity) was used to obtain the pore size distribution
the unknown sample (Fig. 2c). The saturation is changed by of the samples. Each sample with mass of about 0.20 mg was
allowing slow evaporation of the working liquid while the cut into about 20 small 20 mm by 5 mm rectangular tabs to
standard and sample are in capillary contact. The capillary fit into the penetrometer cell.
pressure curve thus obtained corresponds to drainage of the
wetting phase.
One drawback of MSP is its inability to measure residual 3. Results
saturations. When the wetting phase only moderately wets
the porous material (such as water in the hydrophilic pores 3.1. Total porosity
of a GDL) there is the possibility that rivulets of liquid will
detach and become disconnected. There is no way to differ- The overall porosity of the GDL materials (with one
entiate the mass of these water rivulets from the mass of the exception) was provided by the manufacturers, whereas the
connected liquid, since the evaporation of the connected and hydrophilic porosity was not. Table 3 compares the poros-
disconnected liquid occurs simultaneously during the desat- ity values given by the manufacturers with those obtained by
uration phase of the experiment. As a result, the measured MIP and MSP in this study. There is good agreement between
capillary pressure is associated with the total saturation in the manufacturers’ listed porosities and those determined
the sample. experimentally by MIP and MSP with octane. The porosities
measured in this study differ from the manufacturers’ values
2.4. Experimental procedure by not more than 0.04, with a maximum deviation of 5.1%.
The fact that octane reports correct values for total porosity
The MSP tests were run on a Porotech Automated Stan-
indicates that it fully wets all surfaces and the assumption
dard Porosimeter (Vaughan, ON), which is a unique apparatus
for the automated execution of the MSP procedure. Each
Table 3
GDL sample tested was a 20 mm diameter circular section
Total and hydrophilic porosities
cut from the supplied material. The standards fitted onto the
Material Total porosity, εT Hydrophilic
top and bottom of each sample were fabricated porous disks
porosity, εHi
of proprietary composition, perfectly wettable by both octane Manufacturer MIP MSP-octane MSP-water
and water. The working fluids used were ACS grade octane SGL10BA 0.88 0.91 0.86 0.63
(99.99%) and deionized water (>18.0 M). SGL10BB 0.84 0.87 0.81 0.45
One of the most important steps when using MSP was to Toray 090 0.78 0.79 0.74 0.63
ensure that the samples were fully saturated at the start of the E-Tek Cloth ‘A’ – 0.72 0.79 0.74
measurement. To accomplish this, the sample to be tested was Lyflex 484C 0.83 0.86 0.84 0.81
Lyflex 352C 0.78 0.74 0.82 0.36
placed in a sealed chamber that was subsequently evacuated
6 J.T. Gostick et al. / Journal of Power Sources xxx (2005) xxx–xxx

of zero contact angle is valid. Conversely, the fact that the Inspection of the MSP results for the two Lyflex samples
hydrophilic porosity is less than the total porosity in materi- and SGL 10BB indicates the presence of bimodal pore size
als without PTFE (Toray 090 and ETEK Cloth ‘A’) indicates distributions in these materials. These results are understood
that water does not perfectly wet the virgin carbon fibers. by considering that Lyflex GDLs are heavily laden with car-
bon powder, which introduces structural heterogeneity in the
3.2. Porosity distributions form of a uniformly distributed microporosity, whereas the
SGL 10BB sample has a microporous layer, which is both
In this section, the measured capillary pressure data are chemically and structurally dissimilar to the GDL substrate.
analyzed in terms of commonly used pc –swp models. In
the following, the capillary pressure is defined in terms of 3.2.2. Correlation of capillary pressure data
the difference between the wetting and non-wetting phase One of the main goals of the present investigation is to
pressures as: provide relevant data for pc –swp relationships in GDLs. Sev-
pc ≡ pnwp − pwp (8) eral models are commonly used to fit pc –swp curves. One is
the van Genuchten (VG) model [49]:
   
3.2.1. Capillary pressure curves: total pore volume pc n −m
The pc –swp curves for the overall pore network were swp = 1 + ; pc > 0 (9)
pcb
obtained using MIP and MSP. Octane was used as the working
fluid for the MSP measurements. MIP data were converted to where pcb , m and n are fitting parameters. In physical terms,
equivalent octane–water capillary pressure using Eq. (2) and pcb corresponds to a characteristic capillary pressure (break-
the values of contact angle given in Table 2. The agreement through pressure) associated with the first formation of a
between the capillary pressure curves for the six GDL sam- sample-spanning cluster of pores invaded by the non-wetting
ples obtained by the two methods is quite good, as evident phase [41]. Another common model is the Brooks–Corey
from Fig. 3. Using Eq. (1), the mean pore radius of Toray (BC) model [50]:
090 was calculated from the MIP data as 9 ␮m. On the same  
material, Park et al. [47] have measured a mean pore radius of pc −λ
swp = ; pc > pcb (10)
11 ␮m, whereas Mathias et al. [48] have reported a value of pcb
12 ␮m for the slightly thinner Toray 060 without PFTE, both In both models, swp is identified with an effective saturation
results obtained by capillary flow porometry. The agreement defined as:
with the literature values for Toray 090 and with MSP results
swpt − swpr
for all samples (see Fig. 3) lends support to the effective mer- swp = (11)
cury contact angles used in the interpretation of MIP data. 1 − swpr

Fig. 3. Comparison of capillary pressure curves for the six GDL samples obtained using MIP and MSP with octane for (a) SGL 10BA (b) SGL 10BB (c) Toray
090 (d) E-Tek Cloth ‘A’ (e) Lyflex 352 and (f) Lyflex 484 ( : MSP (octane); —: MIP).
 J.T. Gostick et al. / Journal of Power Sources xxx (2005) xxx–xxx 7

Table 4
Model fitting parameters for total pc –swp curves
Material Brooks–Corey van Genuchten

pcb (bar) λ RMS pcb (bar) m n RMS

SGL10BA 0.0562 1.29 0.25 0.0802 0.6489 2.848 0.34
Toray 090 0.1020 1.59 0.24 0.1502 0.7262 3.652 0.17
E-Tek Cloth A 0.0733 1.08 1.4 0.0990 0.5929 2.457 0.72

where swpr is the residual saturation and swpt the total satura- bimodal pore size distribution. Consequently, capillary pres-
tion. sure curves were fit according to the BC and VG models
When applicable, these models offer the possibility of esti- only for the remaining three samples. The fitting parame-
mating the dependence of wetting and non-wetting phase ters so obtained are given in Table 4. In the case of the BC
relative permeabilities on saturation. For example, on the model, data points below pc = pcb (swp greater than ca. 0.8)
basis of the van Genuchten model, the relative permeabili- were excluded since this model is not intended to describe
ties may be estimated as follows: the behavior in these regions.
A comparison of the fitted curves and the experimental
m 1/2
krwp = swp
1/2
(1 − (1 − swp
1/m
) ) (12) data for these samples is presented in Fig. 4. As expected,
the BC model does not agree well with the experimental
2m
krnwp = (1 − swp )1/3 (1 − swp
1/m
) (13) data at high wetting phase saturations. The VG model is the
better choice overall, as it captures the smooth rise in cap-
On the basis of the Brooks–Corey model, the equations for illary pressure at high wetting phase saturations. This rise
relative permeability are: is due to the fact that a large proportion of the pore vol-
ume is accessible directly from the external GDL surfaces,
krwp = swp
3+2/λ
(14)
allowing these pores to be penetrated by the fluid at pres-
krnwp = (1 − swp )2 (1 − swp
3+2/λ
) (15) sures below the breakthrough pressure of the medium. In
hydrogeology and reservoir engineering applications, where
A drawback of both these models is that they can only be the capillary behavior of a very large volume of materials is
applied to unimodal pc –swp curves, and thus are not appro- of interest, this part of an experimental capillary pressure
priate for the SGL 10BB and the Lyflex felt GDLs (Fig. 3). curve is ignored since it is not representative of an infi-
Carbon powder is incorporated in these GDLs, leading to a nite medium. However, this region is an intrinsic aspect of

Fig. 4. Comparison of model fit and experimental capillary pressure curves determined by MSP with octane for (a) E-TEK Cloth ‘A’ (b) Toray 090 (c) SGL
10BA. Solid lines are VG model and dashed lines are BC model.
8 J.T. Gostick et al. / Journal of Power Sources xxx (2005) xxx–xxx

Fig. 5. Comparison of octane–air and water–air capillary pressure curves obtained using the MSP technique for (a) SGL 10BA (b) SGL 10BB (c) Toray 090
(d) E-Tek Cloth ‘A’ (e) Lyflex 352 and (f) Lyflex 484 (䊉: hydrophilic pore network (water); : total pore network (octane)).

GDL behavior, particularly at the GDL–gas channel inter- that the relative proportion of each pore type is known. How-
face, and so should not be ignored. Considerable error asso- ever, this generalization cannot be made in all cases, and the
ciated with the fitting of the VG model is mostly observed Lyflex felts and SGL 10BB are exceptions. Although the dis-
in the high swp region. If only wetting phase saturations tributions for the felt materials are actually similar in shape,
below 95% are considered, the error is less than 10% for all some differences appear in certain regions of the curve. This
samples. suggests that the hydrophobicity is not distributed evenly over
the entire pore size range, likely due to the presence of carbon
3.2.3. Capillary pressure curves: hydrophilic pore powder. The plots for SGL 10BB present a more interesting
volume result. They illustrate the similarities of the 10BA and the
The capillary pressure curves associated with the drainage 10BB samples, which are based on the same substrate mate-
of water from the network of hydrophilic pores were obtained rial. The overall pc –snwp curves of these two materials diverge
using MSP with water as the working fluid. Fig. 5 shows the at higher capillary pressures, because the effect of the micro-
pc –snwp curves for the hydrophilic pore network for each porous layer in the 10BB sample is superimposed on that of
GDL under study. The capillary pressure curves for drainage the substrate material which is common to both GDLs. It can
of octane from the entire pore network are also shown for be seen in Fig. 5 that the hydrophilic portions of 10BA and
comparison. In order to convert the capillary pressure data 10BB are virtually identical in shape. This result suggests that
obtained with octane to the equivalent water–air capillary capillary pressure curves for the individual components, i.e.,
pressure it must be recalled that in the MSP test the capillary substrate and MPL, can be combined to generate the overall
pressures are measured in the perfectly wettable standard, capillary pressure curve. This possibility is investigated in
which is assumed to be in capillary equilibrium with the Section 3.2.6.
GDL. This means that conversion between capillary pres-
sures obtained with octane and water is done by means of 3.2.4. Correlation for hydrophilic pore distribution
Eq. (2) with both contact angles taken as that in the standard, The purpose of wet-proofing the GDL is to render a por-
which is zero. tion of the pores hydrophobic in order to ensure open pores
In most cases, the shapes of the hydrophilic pc –snwp curves for gas flow particularly when flooding occurs. This treat-
are quite similar to their respective overall pc –snwp curves. ment confines liquid water to a subset of the overall pore
This supports the common assumption that the hydrophilic network, which is of great interest when modeling liquid
and hydrophobic pore size distributions have similar shapes. water behavior in GDLs. Fitting of the Brooks–Corey and
In cases where this assumption is true, a single overall pore van Genuchten models to each of the unimodal hydrophilic
size distribution can be used to describe both pore types given distributions yields the parameters in Table 5.
 J.T. Gostick et al. / Journal of Power Sources xxx (2005) xxx–xxx 9

Table 5
Model fitting parameters for unimodal hydrophilic pc –swp curves
Material Brooks–Corey van Genuchten

pcb (bar) λ RMS pcb (bar) m n RMS

SGL10BA 0.0636 1.61 0.38 0.0918 0.740 3.843 0.32
SGL10BB 0.387 1.17 0.33 0.0580 0.627 2.683 0.33
Toray 090 0.120 4.06 1.8 0.1457 0.8667 7.499 0.26
E-Tek Cloth A 0.0565 0.950 1.9 0.0781 0.5407 2.177 0.62

3.2.5. Leverett J-function correlation Jc was introduced into the BC and VG models in place of pcb
In order to directly compare pc –swp data from different and denotes the critical curvature of the J-function.
materials it is necessary to account for the effects of different The success of the J-function correlation is particularly
material and fluid properties. Leverett [32] has shown that encouraging given the diversity of the materials studied. SGL
capillary pressure data of materials with similar structure can 10BA and Toray 090 have similar structures composed of
be correlated in terms of the J-function, defined as follows: randomly oriented carbon fibers, with the Toray material con-
  taining no PTFE and SGL 10BA containing 5% PTFE. This
pc,Air–Water k 1/2 indicates that the Leverett function effectively accounts for
J(snwp ) = (16)
σ ε the differences in porosity and permeability caused by the use
of PTFE. The success with which this correlation describes
Since most GDLs have a similar carbon fiber structure, pre-
the behavior of the E-Tek cloth is also noteworthy. The woven
sumably they have similar pore structures so that the Leverett
substrate is obviously very different from the randomly ori-
correlation should be able to provide a general description of
ented SGL and Toray papers, yet it appears that this difference
the pc –swp data. It is not necessary to include the contact angle
does not affect the capillary behavior at a microscopic level.
in this correlation since the samples are all made of similar
material and the capillary pressures have already been con-
3.2.6. Microporous layer
verted to an air–water basis. The present goal is to determine
It is interesting that the contribution of the MPL to the pore
whether the capillary pressure data for all the GDL samples
size distribution of the SGL 10BB sample appears to be distin-
can be correlated by a single J–snwp plot.
guishable from that of the substrate. This can be clearly seen
To convert the pc –swp data to the J(snwp ) form, the total
by a comparison of the pore size histograms for SGL 10BA
porosities determined by MSP with octane were used (see
and 10BB compared in Fig. 7a and b. In the range between 10
Table 3). The permeabilities used to determine J are listed in
and 100 nm, the 10BB sample shows additional pore volume
Table 1, while the surface tension values for octane and water
due to the microporous layer. In Fig. 7c the capillary pressure
were taken as 0.02175 and 0.0725 N m−1 , respectively. Any
curves obtained by MSP with water for both 10BA and 10BB
curves that showed multimodal distributions were excluded.
are compared. It is apparent that the addition of the hydropho-
The plot in Fig. 6 shows that the data for the different mate-
bic microporous layer to one side of the 10BA material (i.e.
rials fall reasonably well along a single curve, which can be
to create 10BB) does not alter the pc –swp behavior of the
fitted by VG- or BC-type expressions. The fitting parameters
substrate appreciably. Based on this information it should be
obtained for each model are listed in Table 6. The parameter
possible to determine the pc –swp curve for the MPL. How-
ever, this is not a trivial matter because the combined effects
of the added MPL material, additional pore volume in the
MPL and reduced pore volume in the substrate due to MPL
penetration, causes SGL 10BB to have a larger bulk volume
and lower overall porosity than SGL 10BA. Instead, the pore
distribution of the MPL can be found by assuming that two
unimodal distributions, one for the substrate and one for the
MPL, can be added together. Thus, the saturation swp BB for the

10BB GDL can be obtained from the following:

swp
BB
(pc ) = ϕswp
BA
+ (1 − ϕ)swp
MPL
(17)

Table 6
Model fitting parameters for Leverett J(s) curve
Brooks–Corey van Genuchten

Fig. 6. Comparison of experimental J(s)–swp curves for the overall and Jc λ RMS Jc m n RMS
hydrophilic pore networks with Brooks–Corey and van Genuchten models J(s) 0.6274 2.528 0.44 0.6982 0.7114 3.465 0.51
for all unimodal GDLs tested.
10 J.T. Gostick et al. / Journal of Power Sources xxx (2005) xxx–xxx

Fig. 7. Comparison of porosimetry data for SGL 10BA and 10BB. (a) Pore volume vs. pore size histogram of SGL10BB and (b) pore volume vs. pore size
histogram of SGL 10BA. Both are based on MSP with octane. (c) pc –swp curves of hydrophilic pore network in SGL 10BA and 10BB showing similarity of
substrate.

where ϕ is the proportion of the total pore volume in the substrate can be taken as the porosity of the bare paper since
portion of SGL 10BB not intruded by the MPL material, swp BA the porosity distribution of the substrate is not altered by the
the saturation of SGL 10BA, and swp BB the saturation of SGL addition of the MPL (see Fig. 7a), i.e.:
10BB. Values of ϕ and swp can be determined by minimizing
MPL
VpS
the difference between the left and right hand sides of Eq. = εBA
T (19)
(17) using least squares. This regression yields the VG model VBS
parameters for the MPL layer as well as ϕ (Table 7). As shown Eqs. (18) and (19) can be combined to eliminate VPS , yielding:
in Fig. 8, this approach provides an excellent fit for 10BB and
Eq. (17) expresses the pc –swp dependence for this GDL very VBS ϕεBB
= T
(20)
well. VBBB εBA
T
Estimates for the penetration depth of the MPL into the
substrate and the porosity of the MPL can be found by inter- T and εT from this study
Inserting numerical values for ϕ, εBB BA

preting the ϕ factor: into Eq. (20) gives a substrate bulk volume fraction of 0.754.

VpS VpS
ϕ= = (18)
VpBB VBBB εBB
T

where VPS is the pore volume of the free substrate, VPBB the
total pore volume, VBBB the bulk volume and εBBT the overall
porosity, all pertaining to the combined GDL. In the follow-
ing analysis, it is assumed that the portion of the substrate
penetrated by the MPL has the same porosity as the MPL.
The pore volume of the substrate per unit bulk volume of

Table 7
Model fitting parameters for MPL pc –swp curve
Material van Genuchten Additional

pcb (bar) m n ϕ RMS Fig. 8. Comparison of Eq. (17) with experimental pc –swp data for SGL
10BB. The curves for the individual MPL and substrate contributions are
MPL 0.7773 0.5477 2.211 0.7938 0.32
also shown, scaled by ϕ.
 J.T. Gostick et al. / Journal of Power Sources xxx (2005) xxx–xxx 11

Fig. 9. SEM image of GDL cross-section obtained by freeze fracturing in liquid nitrogen. (a) SGL 10BB showing MPL thickness and (b) SGL 10BA for
comparison.

The bulk volume, VBS , of the substrate per unit area A normal This value for εMPL
T confirms the expected result that the MPL
to the flow direction through the GDL can be found from has lower porosity than the substrate.
knowledge of the GDL thickness, i.e.:
VBS VBS BB ϕεBB
= BB t = BA t
T BB
(21) 4. Conclusions
A VB εT
where the SGL 10BB thickness tBB is reported by the man- This study has demonstrated the use of an alternative
ufacturer to be 420 ␮m (Table 1). From knowledge of the method of porosimetry capable of measuring the distribu-
substrate bulk volume fraction, the free substrate thick- tion of the total and hydrophilic pore volume by pore size
ness is determined to be 317 ␮m and the MPL thickness to in wet-proofed GDL materials. The total porosity distribu-
be 103 ␮m. Assuming the quoted SGL 10BA thickness of tion obtained with this method was found to compare very
380 ␮m is the same as that of SGL 10BB prior to the MPL favorably with that obtained from the conventional MIP tech-
incorporation, the penetration depth of the MPL into the sub- nique. The experimental pc –swp curves of the overall and
strate is estimated to be 63 ␮m. This estimate agrees with hydrophilic networks for GDL samples exhibiting unimodal
visual inspection of the MPL (see Fig. 9). Finally, this infor- behavior could be well described by the van Genuchten
mation allows a value of 0.72 to be determined for the MPL model over the entire range of saturations. The shapes of
porosity εMPL
T from the following expression: both hydrophilic and overall pc –swp curves of these GDLs
were also similar and could be normalized using the Lev-
t BB erett function. The entire data set for these samples could be
εMPL
T = (1 − ϕ)εBB
T (22)
t MPL correlated using the van Genuchten model to provide a gen-
12 J.T. Gostick et al. / Journal of Power Sources xxx (2005) xxx–xxx

eral equation from which the pc –swp curves for other GDL [11] G. Lin, W. He, T. Van Nguyen, Modeling liquid water effects in the
materials could be predicted. However, materials with car- gas diffusion and catalyst layers of the cathode of a PEM fuel cell,
J. Electrochem. Soc. 151 (2004) A1999–A2006.
bon powder or a highly hydrophobic layer such as an MPL,
[12] D. Natarajan, T. Van Nguyen, Three-dimensional effects of liquid
did not exhibit this behavior. Separate analysis of the microp- water flooding in the cathode of a PEM fuel cell, J. Power Sources
orous layer of one GDL sample showed that its behavior could 115 (2003) 66–80.
be independently described by the van Genuchten model. [13] D. Natarajan, T. Van Nguyen, A two-dimensional, two-phase, multi-
Then, this could be superimposed with the behavior of the component, transient model for the cathode of a proton exchange
membrane fuel cell using conventional gas distributors, J. Elec-
standard GDL substrate to predict the pc –swp curve of the
trochem. Soc. 148 (2001) A1324–A1335.
combined substrate–MPL GDL. This analysis allowed sev- [14] W. He, J.S. Yi, T. Van Nguyen, Two-phase flow model of the cathode
eral important parameters of the GDL structure such as the of PEM fuel cells using interdigitated flow fields, AICHE J. 46
MPL porosity and its penetration depth into the substrate to (2000) 2053–2064.
be estimated. The van Genuchten model that was used to [15] L. You, H. Liu, A two-phase flow and transport model for the cathode
of PEM fuel cells, Int. J. Heat Mass Transfer 45 (2002) 2277–
describe the pc –swp curves also allows the relative perme-
2287.
abilities to be estimated. This work has proposed alternative [16] A.Z. Weber, R.M. Darling, J. Newman, Modeling two-phase
relationships for the modeling of liquid transport in the GDL behaviour in PEFCs, J. Electrochem. Soc. 151 (2004) A1715–
of PEMFCs. It is anticipated that this relationship will lead A1727.
to more accurate prediction of behavior in the concentration [17] J. Divisek, J. Fuhrmann, K. Gartner, R. Jung, Performance model-
ing of a direct methanol fuel cell, J. Electrochem. Soc. 150 (2003)
overpotential region.
811–825.
[18] L. Pisani, G. Murgia, M. Valentini, B. D’Aguanno, A working model
of polymer electrolyte fuel cells comparisons between theory and
Acknowledgments experiments, J. Electrochem. Soc. 149 (2002) 898–904.
[19] H.K. Hsuen, Performance equations for cathodes in polymer elec-
The authors acknowledge the financial support of the Nat- trolyte fuel cells with non-uniform water flooding in gas diffusers,
J. Power Sources 137 (2004) 183–195.
ural Sciences and Engineering Research Council of Canada [20] M. Grujicic, K.M. Chittajallu, Optimization of the cathode geometry
(NSERC) during the course of this project. They also wish in polymer electrolyte membrane (PEM) fuel cells, Chem. Eng. Sci.
to thank Porotech Inc. (Vaughan, ON) for kindly providing 59 (2004) 5883–5895.
access and support in the operation of their equipment and [21] A.Z. Weber, J. Newman, Modeling transport in polymer-electrolyte
SGL Carbon for donating GDL sample materials. fuel cells, Chem. Rev. 104 (2004) 4679–4726.
[22] C.Y. Wang, Fundamental models for fuel cell engineering, Chem.
Rev. 104 (2004) 4727–4765.
[23] H.K. Lee, J.H. Park, D.Y. Kim, T.H. Lee, A study on the character-
References istics of the diffusion layer thickness and porosity of the PEMFC,
J. Power Sources 131 (2004) 200–206.
[1] S. Gottesfeld, Polymer Electrolyte Fuel Cells, Wiley–VCH, New [24] C.S. Kong, D.-Y. Kim, H.-K. Lee, Y.-G. Shul, T.-H. Lee, Influence
York, 1997, pp. 195–301. of pore-size distribution of diffusion layer on mass-transport prob-
[2] U. Pasaogullari, C.Y. Wang, Liquid water transport in gas diffusion lems of proton exchange membrane fuel cells, J. Power Sources 108
layer of polymer electrolyte fuel cells, J. Electrochem. Soc. 151 (2002) 185–191.
(2004) A399–A406. [25] L.R. Jordan, A.K. Shukla, T. Behrsing, N.R. Avery, B.C. Muddle, M.
[3] U. Pasaogullari, C.Y. Wang, Two-phase transport and the role of Forsyth, Effect of diffusion-layer morphology on the performance of
micro-porous layer in polymer electrolyte fuel cells, Electrochim. polymer electrolyte fuel cells operating at atmospheric pressure, J.
Acta 49 (2004) 4359–4369. Appl. Electrochem. 30 (2000) 641–646.
[4] Z.H. Wang, C.Y. Wang, K.S. Chen, Two-phase flow and transport in [26] E. Passalacqua, G. Squadrito, F. Lufrano, A. Patti, L. Giorgi, Effects
the air cathode of proton exchange membrane fuel cells, J. Power of the diffusion layer characteristics on the performance of poly-
Sources 94 (2001) 40–50. mer electrolyte fuel cell electrodes, J. Appl. Electrochem. 31 (2001)
[5] T. Berning, N. Djilali, A 3D, multiphase multicomponent model of 449–454.
the cathode and anode of a PEM fuel cell, J. Electrochem. Soc. 150 [27] E. Antolini, A. Pozio, L. Giorgi, E. Passalacqua, Morphological
(2003) A1589–A1598. characteristics of carbon/polytetrafluoroethylene films deposited on
[6] M. Noponen, E. Birgersson, J. Ihonen, M. Vynnycky, A. Lundblad, porous carbon support, J. Mater. Sci. 33 (1998) 1837–1843.
G. Lindbergh, A two-phase non-isothermal PEFC model: theory and [28] M.V. Williams, E. Begg, L. Bonville, H.R. Kunz, J.M. Fenton, Char-
validation, Fuel Cells 4 (2004) 365–377. acterization of gas diffusion layers for PEMFC, J. Electrochem. Soc.
[7] M. Hu, A. Gu, M. Wang, X. Zhu, L. Yu, Three dimensional, two 151 (2004) 1173–1180.
phase flow mathematical model for PEM fuel cell: Part I. Model [29] J. Ihonen, M. Mikkola, G. Lindbergh, Flooding of gas diffusion
development, Energy Conv. Manage. 45 (2004) 1861–1882. backing in PEFCs: physical and electrochemical characterization, J.
[8] S. Mazumder, J.V. Cole, Rigorous 3-D mathematical modeling of Electrochem. Soc. 151 (2004) 1152–1161.
PEM fuel cells II. Model predictions with liquid water transport, J. [30] M. Prasanna, H.Y. Ha, E.A. Cho, S.A. Hong, I.H. Oh, Influence of
Electrochem. Soc. 150 (2003) 1510–1517. cathode gas diffusion media on the performance of the PEMFCs, J.
[9] N.P. Siegel, M.W. Ellis, D.J. Nelson, M.R. Von Spakovsky, A two- Power Sources 131 (2004) 147–154.
dimensional computational model of a PEMFC with liquid water [31] K.S. Udell, Heat transfer in porous media considering phase change
transport, J. Power Sources 128 (2004) 173–184. and capillarity - The heat pipe effect, Int. J. Heat Mass Transfer 28
[10] J.H. Nam, M. Kaviany, Effective diffusivity and water-saturation dis- (1985) 485–495.
tribution in single- and two-layer PEMFC diffusion medium, Int. J. [32] M.C. Leverett, Capillary behavior in porous solids, AIME Trans. 142
Heat Mass Transfer 46 (2003) 4595–4611. (1941) 152–168.
 J.T. Gostick et al. / Journal of Power Sources xxx (2005) xxx–xxx 13

[33] P. Ustohal, F. Stauffer, T. Dracos, Measurement and modeling of [42] A.B.D. Cassie, S. Baxter, Wettability of porous surfaces, Faraday
hydraulic characteristics of unsaturated porous media with mixed Soc. Trans. 40 (1944) 546–551.
wettability, J. Contam. Hydrol. 33 (1998) 5–37. [43] A.W. Adamson, Physical Chemistry of Surfaces, John Wiley and
[34] D. Bevers, R. Rogers, M. Von Bradke, Examination of the influ- Sons, New York, 1982.
ence of PTFE coating on the properties of carbon paper in [44] T.E. Hutchinson, Direct measurement of microscopic contact angle,
polymer electrolyte fuel cells, J. Power Sources 63 (1996) 193– Thin Solid Films 8 (1971) R25–R29.
201. [45] J. Divisek, R. Wilkenhoener, Y. Volfkovich, Structure investigations
[35] H. Dohle, H. Schmitz, T. Bewer, J. Mergel, D. Stolten, Development of SOFC anode cermets. Part I: porosity investigations, J. Appl.
of a compact 500 W class direct methanol fuel cell stack, J. Power Electrochem. 29 (1999) 153–163.
Sources 106 (2002) 313–322. [46] J. Divisek, M. Eikerling, V. Mazin, H. Schmitz, U. Stimming, Y.M.
[36] Y.M. Volfkovich, V.S. Bagotzky, V.E. Sosenkin, I.A. Blinov, The Volfkovich, Study of capillary porous structure and sorption prop-
standard contact porosimetry, Colloids Surf. A 187–188 (2001) erties of Nafion proton-exchange membranes swollen in water, J.
349–365. Electrochem. Soc. 145 (1998) 2677–2683.
[37] S.A. Grant, Extension of a temperature effects model for capil- [47] G.G. Park, Y.J. Sohn, T.H. Yang, Y.G. Yoon, W.Y. Lee, C.S. Kim,
lary pressure saturation relations, Water Resour. Res. 39 (2003) 11– Effect of PTFE contents in the gas diffusion media on the perfor-
110. mance of PEMFC, J. Power Sources 131 (2004) 182–187.
[38] H.Y. She, B.E. Sleep, The effect of temperature on capillary pre- [48] M.F. Mathias, J. Roth, J. Fleming, W. Lehnert, Diffusion media
ssure–saturation relationships for air–water and perchloroethylene– materials and characterization, in: W. Vielstich, H.A. Gasteiger, A.
water systems, Water Resour. Res. 34 (1998) 2587–2597. Lamm (Eds.), Handbook of Fuel Cells—Fundamentals, Technology
[39] G.W. Jackson, D.F. James, Permeability of fibrous porous media, and Applications, Part 1, vol. 3, John Wiley and Sons, New York,
Can. J. Chem. Eng. 64 (1986) 364–374. 2003, pp. 517–237.
[40] F.A.L. Dullien, Porous Media: Fluid Transport and Pore Structure, [49] M.T. van Genuchten, Closed-form equation for predicting the
Academic Press, New York, 1992. hydraulic conductivity of unsaturated soils, Soil Sci. Soc. Am. J.
[41] M.A. Ioannidis, I. Chatzis, Network modelling of pore structure and 44 (1980) 892–898.
transport properties of porous media, Chem. Eng. Sci. 48 (1993) [50] J. Bear, Dynamics of Fluids in Porous Media, Dover Publications,
951–972. New York, 1972.