Journal of Power Sources 279 (2015) 133e137

Contents lists available at ScienceDirect

Journal of Power Sources

journal homepage: www.elsevier.com/locate/jpowsour

Short communication

Common inconsistencies in modeling gas transport in porous

electrodes: The dusty-gas model and the Fick law
A. Bertei*, C. Nicolella
Department of Civil and Industrial Engineering, University of Pisa, Largo Lucio Lazzarino 2, 56126 Pisa, Italy

h i g h l i g h t s

Common inconsistencies in modeling gas transport in SOFC electrodes are discussed.

Assuming uniform pressure in dusty-gas model is inconsistent with stoichiometry.

Use of Bosanquet formula in Fick model does not allow diffusive fluxes to sum up to 0.

Dusty-gas model must be used without isobaric assumption for multi-component mixtures.

Fick model can be used for binary mixtures considering ordinary diffusion only.

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

Article history: The paper shows as two assumptions typically made in modeling gas transport in solid oxide fuel cell
Received 11 November 2014 electrodes, i.e., a) uniform pressure in the dusty-gas model, and b) validity of the Bosanquet formula in
Received in revised form the Fick model, may lead to serious inconsistencies (such as molar fractions that do not sum up to one or
16 December 2014
fluxes that do not obey reaction stoichiometry), thus nullifying the efforts of the mechanistic modeling of
Accepted 2 January 2015
Available online 3 January 2015
transport phenomena. The nature of the inconsistent use of the models is explained with clear examples,
then the correct implementation of the gas transport models is discussed. The study aims to promote a
coherent physically-based modeling of gas transport phenomena in porous electrodes in order to assist
Keywords:
Solid oxide fuel cell (SOFC)
their rational design.
Porous media © 2015 Elsevier B.V. All rights reserved.
Gas transport
Dusty-gas model
Fick law
Bosanquet formula

1. Introduction description of viscous multi-component gas mixtures moving

One of the current trends in solid oxide fuel cell (SOFC) research
is the development of more efficient and stable electrodes and cells.
In electrode-supported designs, a significant contribution to
voltage losses may arise from gas transport resistances through
porous electrodes [1]. The correct modeling of gas transport phe-
nomena is important to predict the partial pressure profiles within
a porous electrode, which allow us to evaluate local reaction rates
[2], the occurrence of detrimental redox conditions [3], concen-
tration overpotentials and limiting current densities [4] or to infer
unknown parameters [5].
Modeling gas transport in SOFC electrodes requires the
* Corresponding author.
E-mail address: antonio.bertei@for.unipi.it (A. Bertei).
within a porous medium in transition regime (i.e., Knudsen number
between 0.01 and 10). Therefore, a gas transport model must
encompass, at least, transport mechanisms such as ordinary
diffusion, Knudsen diffusion and viscous flow. Typically, thermal
and surface diffusion are neglected. Several gas transport models
have been adopted in SOFC modeling: Fick model (FM) [6], Max-
welleStefan model (MSM) [7,8], dusty-gas model (DGM) [7,9], bi-
nary friction model (BFM) [10,11], cylindrical pore interpolation
model (CPIM) [12,13]. By far, FM and DGM are the most frequently
used.
However, unfortunately a large number of good modeling
studies are partly flawed by incompatible assumptions or incorrect
applications of these models, which lead to serious inconsistencies
such as molar fractions that do not sum up to 1 or fluxes that do not
obey reaction stoichiometry. In such a case, the gas transport model
fails to predict partial pressure profiles, thus potentially affecting

http://dx.doi.org/10.1016/j.jpowsour.2015.01.007
0378-7753/© 2015 Elsevier B.V. All rights reserved.
134 A. Bertei, C. Nicolella / Journal of Power Sources 279 (2015) 133e137

not only concentration losses, but also activation and ohmic losses 3. The dusty-gas model (DGM)
because transport and reaction phenomena are intrinsically
coupled [14]. 3.1. DGM equations
The aim of this paper is to summarize the typical inconsistencies
found in the literature in the application of DGM and FM, which are The dusty-gas model is an implicit expression among fluxes,
the assumption of uniform pressure and the use of the Bosanquet pressure and molar fractions as follows [9]:
formula, respectively. Proofs of the inconsistencies are shown and !
the correct implementation of the models is remarked in order to X
n
yj Ni  yi Nj Ni P dyi y BP dP
þ ¼  i 1þ (2)
avoid an incorrect use. The paper is thus concerned with the formal B;eff
Dij
K;eff
Di RT dx RT K;eff
mDi dx
jsi
consistency of the model implementation: the inherent soundness
of the gas transport models, their mutual comparison and the DB;eff
ij
and DK;eff
i
are the effective binary and Knudsen diffusiv-
quantification of how their incorrect use may affect the prediction ities, corrected by porosity f and tortuosity factor t as
of partial pressure profiles are out of the scope of the paper, for
DB;eff
ij
¼ ðf=tÞ$DBij and DK;eff
i
¼ ðf=tÞ$DKi [17]. Binary diffusivities
which the reader is referred to specific studies [1,7,10e13,15].
DBij ¼ DBji can be evaluated with kinetic-theory-based methods (e.g.,
Chapman-Enskog [18]) or correlations (e.g., Fuller et al. [19]) while
2. Mathematical background on gas transport modeling in the Knudsen diffusivity is equal to:
SOFC electrodes sffiffiffiffiffiffiffiffiffi
1 8RT
DKi ¼ dp (3)
The core of a continuum electrode-level or cell-level model 3 pMi
consists in a set of mass balance equations within the porous
electrodes. For simplicity, let us consider a 1D geometry, although where dp is the mean pore size and Mi the molecular weight of
the discussion can be easily extended to 2D and 3D. Let us also species i. In Eq. (2), m represents the dynamic gas viscosity and B the
assume isothermal conditions, ideal gas behavior and homoge- electrode permeability, typically calculated as B ¼ (f/t)$d2p/32.
neous electrode microstructure, because most of the gas transport A large number of experimental studies have satisfactorily
models are valid under these assumptions. validated the DGM in porous media with and without reaction
The molar balance (i.e., continuity equation) for a generic gas [8,20]. The superiority of the DGM has been proven in SOFC ap-
species i is as follows: plications too [7,10,12], although we must agree with Bhattacharyya
and Rengaswamy [21] that the same set of experimental data [22]
f vðPyi Þ vNi has been considered in all these studies. For the sake of
þ ¼ si (1a)
RT vt vx completeness, it must be mentioned that some criticisms have been
raised about the formal derivation of the DGM [11,13,23]. On the
P other hand, comparisons with much sounder models like BFM and
Ni ¼ yi v þ Nidiff (1b)
RT CPIM performed by Wang et al. [12] and Vural et al. [10] show that
there is no significant difference among these models.
where f is the porosity, R the gas constant, T the temperature, P the By summing up Eq. (2) for all the n species we obtain
total pressure, yi the molar fraction, x the coordinate (x ¼ 0: [8,10,12,15]:
electrode-gas chamber/channel interface, x ¼ L: electrode- Pn Ni
electrolyte interface), Ni and Nidiff the total and diffusive molar RT i¼i DK;eff
dP
fluxes, v the molar average velocity and si is the source term, which ¼ i
! (4)
dx Pn
is due to the sum of chemical and electrochemical reactions 1þ BP yi
m i¼1 DK;eff
involving the species i. Typically, a Dirichlet boundary condition is i

set on one side (e.g., at x ¼ 0, P ¼ Pch and yi ¼ ych

i are specified) and a
Neumann boundary condition is set on the other side (e.g., if the
electrochemical reaction is assumed to occur at the electrode-
electrolyte interface only, Ni ¼ ni·JS/F is set at x ¼ L, where ni is
the stoichiometric coefficient and JS the current density per unit
surface).
For n gas species, the problem is solved once evaluated the
dependent variables, which are the velocity and n independent
state variables (for example, given the algebraic constraint
Pn
i¼1 yi ¼ 1, n independent state variables are P and yi for i ¼ 1, …,
ne1). In general, the complete description of the system requires:
a) n molar balances as Eq. (1a), b) the momentum balance (i.e.,
equation of motion), c) the equation of state (e.g., the ideal gas law,
as assumed in this study), and d) constitutive equations (i.e., gas
transport models, as those mentioned in Section 1) to link the
fluxes Ni to the state variables and the velocity field [16]. Note that
some gas transport models as DGM, BFM and CPIM already
comprise the momentum balance, thus the momentum balance
must not be explicitly included in the set of equations.
If a flawed gas transport model is used, an inconsistency is created.
P i¼i
The inconsistency can be ni¼1 yi s1 or one of the fluxes may not
obey the corresponding molar balance (some examples are given in
Sections 3.2 and 4.2).
which shows that in the DGM the pressure gradient is related to
Knudsen diffusion and viscous flow.
3.2. Assumptions inconsistent with the DGM
Once specified the gas transport model as in Eq. (2), the math-
ematical problem is determined since it consists of n molar bal-
ances as Eq. (1a) in n state variables (remember that the DGM
already includes the momentum balance, thus the molar average
velocity is implicitly calculated). Any additional relationship among
P, yi and Ni makes the problem overdetermined, thus leading to an
inconsistency in its solution.
The typical inconsistent assumption made in the DGM appli-
cation is the assumption of uniform pressure [6,24e35], which is an
additional constraint on the state variable P. In such a case, by
considering the definition of DKi in Eq. (3), Eq. (4) becomes [8,15]:
X
n pffiffiffiffiffiffi
Ni Mi ¼ 0 (5)
Eq. (5) is the Graham's law of effusion, which correctly describes
the diffusion behavior in a confined system in the absence of
 A. Bertei, C. Nicolella / Journal of Power Sources 279 (2015) 133e137 135

reaction and pressure gradients [8,13,15]. established despite NN2 ¼ 0, suggesting that considering a con-
However, the Graham's law is not suited to describe a system stantyN2 ¼ ych
N2 [34] is not a reasonable approximation.
wherein a reaction occurs [8,15], such as a porous electrode. Indeed, It is worth comparing the concentration overpotential, calcu-
ey ey
in such a case, Eq. (5) represents a constraint among fluxes which is lated as hconc ¼ RT=ð2FÞ,lnððych ch
H2 =yH2 Þ$ðyH2 O =yH2 O ÞÞ [4], for the case
generally inconsistent with molar balances and, ultimately, with reported in Fig. 1 with the correct and inconsistent DGM imple-
reaction stoichiometry. For example, in an anode fed with H2/H2O mentations. By applying correctly the DGM, the concentration
mixture at steady-state, the molar balances set NH2 þ NH2 O ¼ 0 overpotential is equal to 73.1 mV. On the other hand, when the
throughout the anode thickness, which is obviously different from DGM is applied inconsistently under uniform pressure assumption,
Eq. (5) since MH2 sMH2 O . one of the fluxes will not obey reaction stoichiometry: the calcu-
Assuming uniform pressure (or, equivalently, the Graham's law) is lated concentration overpotential is equal to 73.0, 39.2, 80.2 mV for
intrinsically incompatible with the application of the DGM in a N2, H2O, H2 flux not obeying reaction stoichiometry, respectively.
reacting system such as an SOFC electrode, thus it must be avoided. Thus, significant errors may be produced from an inconsistent use
of the model. Moreover, the relative error in the prediction of hconc
3.3. Implementation of the DGM can be as large as 50% for a binary H2/H2O mixture in similar con-
ditions, and the error generally increases as f/t, dp and ych H2 O in-
There is no inconsistency when the DGM is implemented as it is crease or JS and L decrease.
in Eq. (2), that is, without assuming uniform pressure. Eq. (2) can be
applied n times in order to evaluate the n independent fluxes Ni as a 4. The Fick model (FM)
function of the n state variables chosen to describe the system. A
compact way to represent the fluxes is the matrix form proposed by 4.1. FM equations
Zhu et al. [2]:
The Fick model is an expression among fluxes, pressure and
2
0 1 3
X
n molar fractions, that, under constant pressure assumption, reads as
qij
dyj PB A dP 5
Ni ¼  4P @
þ yj 1 þ for i ¼ 1; :::; n follows [16]:
RT dx K;eff dx
j¼1 mD j
X
n
diff
(6a) Ni ¼ yi Nj þ Ni (7a)
j¼1
!
1 X
n
yk   yi
hij ¼ þ dij þ dij  1 B;eff for i; j ¼ 1; :::; n
K;eff B;eff P dyi
Di ksi Dik Dij Nidiff ¼  Deff (7b)
RT im dx
(6b)
eff
where Dim represents the effective diffusivity of the species i in the
where qij are the elements of the matrix Q ¼ H1, hij the elements of mixture. Note that, unlike the DGM, the FM does not include the
the matrix H and dij is the Kronecker delta. momentum balance. Thus, the FM does not give any information
A typical example of application of the DGM is reported in Fig. 1 about the global motion of the mixture, which can be obtained
for an H2/N2/H2O mixture in an anode. Two interesting conclusions either through the solution of the equation of motion or, in some
can be drawn from the Figure: a) although the system represents a simple cases, by assigning beforehand a relationship among the
case of equimolar counter-diffusion, a pressure gradient is gener- fluxes based on physical considerations [16].
ated due to DKH2 > DKH2 O [1,15], showing that pressure is generally not By summing up Eq. (7a) for all the n species, the basic rela-
uniform inside an electrode; b) a gradient of yN2 (and PN2 ) is tionship among diffusive fluxes is obtained [16]:

X
n
0.6 1.05 Nidiff ¼ 0 (8)
H2 i¼1
eff
0.5 Eq. (8) sets a constraint among the n diffusion coefficients Dim ;
1.04 which are not independent one from another. For example, in a
eff
binary mixture, the application of Eq. (7b) into Eq. (8) gives D1m ¼
0.4 N2 eff
D2m ; which is the condition to be obeyed.
1.03
P [atm]

4.2. Assumptions inconsistent with the FM

yi [-]

0.3

1.02 In SOFC applications, very often researchers want to include

0.2 Knudsen effects within the FM. Typically this is done by using the
Bosanquet formula [36] in order to evaluate the effective diffusivity
1.01 Deff
im
as follows [6,14,37e45]:
0.1 H2O
!1
eff 1 1
0.0 1.00 Dim ¼ B;eff
þ K;eff
(9)
0.0 0.2 0.4 0.6 0.8 1.0
Dim Di
x [mm]
wherein the effective ordinary diffusivity of the species i in the
Fig. 1. Molar fraction and pressure profiles along the anode thickness simulated with multi-component mixture is typically evaluated through the Wilke
the DGM. Simulation conditions: T ¼ 800  C, L ¼ 1 mm, JS ¼ 8000 A m2, Pch ¼ 1 atm, B;eff P B;eff
formula Dim ¼ ð1  yi Þ= njsi ðyj =Dij Þ [1,22], despite the validity
H2 ¼ 0:57, yN2 ¼ 0:40, f ¼ 0.4, t ¼ 1.5, dp ¼ 0.5 mm. Binary diffusivities are calculated
ych ch

by using Fuller et al. method [19] while the gas viscosity with the Herning and Zipperer of the Wilke formula is in principle restricted to a situation in which
method [19]. the species i diffuses in a stagnant multi-component mixture (i.e.,
136 A. Bertei, C. Nicolella / Journal of Power Sources 279 (2015) 133e137

Nj ¼ 0 for j ¼ 1, …, n, j s i) [8,16]. electrodes, showing how an incompatible assumption can lead to

The blind use of the Bosanquet formula (Eq. (9)) is not physically either fluxes inconsistent with molar balances and reaction stoi-
justified because the Bosanquet formula has a very narrow range of chiometry or to molar fractions that do not sum up to 1. In both the
validity. Indeed, Eq. (9) is valid under this set of limit conditions: a) cases, the gas transport model would fail, affecting the whole
binary mixture, b) isobaric conditions, and c) yi / 0 or M1 ¼ M2 [8], electrode or cell model, which may include charge transport and
which are hardly met in SOFC applications. reaction too, due to the inherent coupling of transport and reaction
The use of Eq. (9) in the FM (Eq. (7)) generally leads to the phenomena.
violation of Eq. (8). As an example, let us take a binary mixture. In It was shown that assuming uniform pressure in the dusty-gas
eff eff
order to satisfy Eq. (8), it must be D1m ¼ D2m . Since the effective model results in the Graham's law of effusion, which is incompat-
ordinary diffusivities are equal (i.e., DB;eff
1m ¼ DB;eff B;eff B;eff
2m ¼ D12 ¼ D21 ), ible with any reaction stoichiometry. The correct way to implement
the application of the Bosanquet formula leads to: the dusty-gas model is to leave the total pressure as a dependent
variable to be calculated.
!1 !1
1 1 1 1 On the other hand, it was remarked that the Bosanquet formula
Deff
1m ¼ B;eff
þ K;eff
and Deff
2m ¼ B;eff
þ K;eff
(10) has a narrow range of validity and its use in the Fick model did not
D12 D1 D12 D2
allow diffusive fluxes to sum up to 0. Therefore, it was suggested to
eff eff use the Fick model exclusively for binary mixtures by considering
which is D1m sD2m unless M1 ¼ M2, thus violating the definition of
ordinary diffusion only.
diffusive fluxes in Eq. (8).
The considerations reported in this study are aimed to avoid
Let us take as an example an anode fed with an H2/H2O mixture.
confusion and an incorrect use of gas transport models, with the
At steady-state, reaction stoichiometry sets NH2 þ NH2 O ¼ 0.
goal of improving the coherence of physically-based models for
Assuming that the electrochemical reaction occurs at the electrode-
SOFC applications and to assist the rational design of porous
electrolyte interface and applying the FM as in Eq. (7), the
 . electrodes.
solution of the H2 molar balance gives NH2 ¼ PDeff ey ch
H2 m yH2  yH2

ðRTLÞ ¼ J S =ð2F Þ while the solution of the H2O molar balance gives Acknowledgments
 .
NH2 O ¼ PDeff ey ch
H2 Om yH2 O  yH2 O ðRTLÞ ¼ J S =ð2F Þ, which allow us
Helpful discussion with Prof. Roberto Mauri (University of Pisa)
to calculate the H2 and H2O molar fractions yey ey
H2 and yH2 O at the is gratefully acknowledged.
electrode-electrolyte interface. By using the Bosanquet formula (Eq.
(9)) to calculate the effective diffusivities and considering the same
References
inlet and operating conditions applied in Fig. 1, with the only dif-
ference in the inlet condition ych
H2 ¼1 ych
H2 O ¼ 0:97, in the first case [1] M. Garcìa-Camprubì, A. Sa
nchez-Insa, N. Fueyo, Chem. Eng. Sci. 65 (2010)
ey ey 1668.
we obtain yH2 ¼ 0:923 while in the second case yH2 O ¼ 0:119, that is, [2] H. Zhu, R.J. Kee, V.M. Janardhanan, O. Deutschmann, D.G. Goodwin,
yey ey
H2 ¼ 1  yH2 O ¼ 0:881. These final results are different, indicating
J. Electrochem. Soc. 152 (2005) A2427.
[3] J.P. Neidhardt, M. Henke, W.G. Bessler, ECS Trans. 35 (2011) 1621.
that: a) the application of the FM with the Bosanquet formula to only [4] P. Aguiar, C.S. Adjiman, N.P. Brandon, J. Power Sources 138 (2004) 120.
ne1 molar balances does not circumvent the problem of the misuse [5] M. Garcìa-Camprubì, S. Izquierdo, N. Fueyo, Renew. Sustain. Energy Rev. 33
of the Bosanquet formula, because different results are obtained (2014) 701.
[6] R. Suwanwarangkul, E. Croiset, M.W. Fowler, P.L. Douglas, E. Entchev,
depending on the arbitrary choice of the ne1 species considered, M.A. Douglas, J. Power Sources 122 (2003) 9.
and b) if the FM with the Bosanquet formula is applied simulta- [7] K. Tseronis, I.K. Kookos, C. Theodoropoulos, Chem. Eng. Sci. 63 (2008) 5626.
P [8] R. Krishna, J.A. Wesselingh, Chem. Eng. Sci. 52 (1997) 861.
neously to all the n species, an inconsistency like ni¼1 yi s1 arises.
[9] E.A. Mason, A.P. Malinauskas, Gas Transport in Porous Media: the Dusty-Gas
This clearly demonstrates that the use of the Bosanquet formula in Model, Elsevier, Amsterdam, 1983.
the FM leads to an inconsistency because diffusive fluxes do not sum [10] Y. Vural, L. Ma, D.B. Ingham, M. Pourkashanian, J. Power Sources 195 (2010)
4893.
up to 0.
[11] P.J.A.M. Kerkhof, Chem. Eng. J. 64 (1996) 319.
[12] S. Wang, W.M. Worek, W.J. Minkowycz, Int. J. Heat Mass Transf. 55 (2012)
3933.
4.3. Implementation of the FM [13] J.B. Young, B. Todd, Int. J. Heat Mass Transf. 48 (2005) 5338.
[14] M. Andersson, J. Yuan, B. Sunde n, Appl. Energy 87 (2010) 1461.
We personally discourage to use the FM in multi-component [15] I.K. Kookos, Chem. Eng. Sci. 69 (2012) 571.
[16] R.B. Bird, W.E. Steward, E.N. Lightfoot, Transport Phenomena, Wiley, New
mixtures because it is not suited for n > 2: Krishna and Wesse-
York, 1960.
lingh [8] present a clear experimental example in which the FM fails [17] N. Epstein, Chem. Eng. Sci. 44 (1989) 777.
to predict the diffusion behavior even in a ternary ideal gas mixture [18] S. Chapman, T.G. Cowling, The Mathematical Theory of Non-uniform Gases,
third ed., Cambridge University Press, Cambridge, 1970.
for a situation corresponding to equimolar diffusion. Instead, for
[19] B. Todd, J.B. Young, J. Power Sources 110 (2002) 186.
n > 2 we suggest to use the DGM or, as a second choice, the MSM, [20] J.W. Veldsink, R.M.J. van Damme, G.F. Versteeg, W.P.M. van Swaiij, Chem. Eng.
although the latter neglects Knudsen effects and pressure gradients. J. 57 (1995) 115.
In binary mixtures the FM can be used as a first approximation [21] D. Bhattacharyya, R. Rengaswamy, Ind. Eng. Chem. Res. 48 (2009) 6068.
[22] H. Yakabe, M. Hishinuma, M. Uratani, Y. Matsuzaki, I. Yasuda, J. Power Sources
when Knudsen effects are negligible by considering ordinary 86 (2000) 423.
diffusion only, that is, with Deff eff B;eff
1m ¼ D2m ¼ D12 . On the other hand, if [23] W. Kast, C.-R. Hohenthanner, Int. J. Heat Mass Transf. 43 (2000) 807.
one wants to take into account pressure gradients and Knudsen [24] Y. Jiang, A.V. Virkar, J. Electrochem. Soc. 150 (2003) A942.
[25] E. Hern andez-Pacheco, D. Singh, P.N. Hutton, N. Patel, M.D. Mann, J. Power
effects too, the DGM must be used (an explicit form of the DGM for Sources 138 (2004) 174.
binary mixtures is given by Liu et al. [46]). [26] E. Herna
ndez-Pacheco, M.D. Mann, P.N. Hutton, D. Singh, K.E. Martin, Int. J.
Hydrog. Energy 30 (2005) 1221.
[27] Q. Wang, L. Li, C. Wang, J. Power Sources 186 (2009) 399.
5. Conclusions [28] F.N. Cayan, S.R. Pakalapati, F. Elizalde-Blancas, I. Celik, J. Power Sources 192
(2009) 467.
[29] P. Chinda, W. Wechsatol, S. Chanchaona, P. Brault, Fuel Cells 11 (2011) 184.
This paper summarized the most common inconsistencies [30] S.A. Hajimolana, M.A. Hussain, W.M.A.W. Daud, M. Soroush, A. Shamiri,
found in the literature in modeling gas transport in SOFC Renew. Sustain. Energy Rev. 15 (2011) 1893.
 A. Bertei, C. Nicolella / Journal of Power Sources 279 (2015) 133e137 137

[31] K. Matsuzaki, N. Shikazono, N. Kasagi, J. Power Sources 196 (2011) 3073. [38] M.M. Hussain, X. Li, I. Dincer, Int. J. Energy Res. 29 (2005) 1083.
[32] B. Vo€lker, R.M. McMeeking, J. Power Sources 215 (2012) 199. [39] M. Ni, M.K.H. Leung, D.Y.C. Leung, J. Power Sources 163 (2006) 460.
[33] E. Lay-Grindler, J. Laurencin, G. Delette, J. Aicart, M. Petitjean, L. Dessemond, [40] M.M. Hussain, X. Li, I. Dincer, Int. J. Therm. Sci. 46 (2007) 48.
Int. J. Hydrog. Energy 38 (2013) 6917. [41] M. Cannarozzo, S. Grosso, G. Agnew, A. Del Borghi, P. Costamagna, J. Fuel Cell
[34] K.J. Yoon, S. Lee, H. An, J. Kim, J.-W. Son, J.-H. Lee, H.-J. Je, H.-W. Lee, B.-K. Kim, Sci. Technol. 4 (2007) 99.
Int. J. Hydrog. Energy 39 (2014) 3868. [42] Y. Wang, F. Yoshiba, T. Watanabe, S. Weng, J. Power Sources 170 (2007) 101.
[35] H. Geisler, A. Kromp, A. Weber, E. Ivers-Tiffe e, J. Electrochem. Soc. 161 (2014) [43] T.X. Ho, P. Kosinski, A.C. Hoffmann, A. Vik, Chem. Eng. Sci. 63 (2008) 5356.
778. [44] L. Andreassi, C. Toro, S. Ubertini, J. Fuel Cell Sci. Technol. 6 (2009) 021307.
[36] W.G. Pollard, R.D. Present, Phys. Rev. 73 (1948) 762. [45] Y. Shi, C. Li, N. Cai, J. Power Sources 196 (2011) 5526.
[37] S.H. Chan, K.A. Khor, Z.T. Xia, J. Power Sources 93 (2001) 130. [46] S. Liu, W. Kong, Z. Lin, J. Power Sources 194 (2009) 854.