Electrochimica Acta 147 (2014) 294–309

Contents lists available at ScienceDirect

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

Analysis of non-isothermal effects on polymer electrolyte fuel cell

electrode assemblies
M. Bhaiya a , A. Putz b , M. Secanell a,∗
a
Energy Systems Design Laboratory, Department of Mechanical Engineering, University of Alberta, Edmonton, AB, Canada
b
Automotive Fuel Cell Cooperation, Burnaby, British Columbia V5J 5J8, Canada

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

Article history: A non-isothermal, single phase membrane electrode assembly (MEA) mathematical model accounting
Received 30 June 2014 for most applicable heat sources, viz., reversible, irreversible, ohmic heating, phase change, heat of sorp-
Received in revised form 8 September 2014 tion/desorption, is presented. The mathematical model fully couples a thermal transport equation with
Accepted 14 September 2014
an MEA model and allows the study of non-isothermal effects, such as thermal osmosis through the
Available online 19 September 2014
membrane, local relative humidity variations in the catalyst layers and water sorption into the mem-
brane. A detailed breakdown of various heat sources in the MEA at different current densities is provided
Keywords:
and the impact of various thermal effects previously neglected in the literature such as thermal-osmosis,
Polymer electrolyte fuel cell
Membrane electrode assembly
reversible heat distribution, and heat of sorption are studied. Results show that sorption heat cannot
Non-isothermal be neglected as it contributes up to 10% of the total heat under normal operating conditions. Reversible
Open-source heat distribution can significantly affect the temperature distribution shifting the hottest location of the
Macro-homogeneous model cell from anode and cathode. Analyzing the water transport across the membrane, results show that
Sorption heat thermal-osmosis contributes up to 25% of the water flux inside the membrane at moderate and high
Reversible heat current densities.
Thermal osmosis © 2014 Elsevier Ltd. All rights reserved.
Finite elements

1. Introduction Experimental results have recently shown that the tempera-

Thermal management can significantly influence polymer elec-
trolyte fuel cell (PEFC) operation [1,2]. At high current density, the
temperature inside the fuel cell can be significantly higher than
at the end plates. In automotive applications, where high power
densities are achieved, the temperature variations in the cell might
be of the order of several degrees. Temperature influences mass,
charge and reaction rates. Reaction rates and species transport rates
improve at high temperatures. However, high temperatures also
result in: i) a reduction of the open cell potential; ii) a reduction
in local relative humidity (RH), due to the increased saturation
pressure of water, resulting in a significant decrease in membrane
hydration levels and thereby proton conduction; and iii) increased
fuel crossover through the membrane. Local hot spots might also
cause pin holes and degeneration of the membrane significantly
reducing fuel cell performance and durability.
∗ Corresponding author. University of Alberta Mechanical Engineering University
of Alberta 4-9 Mechanical Engineering Building Edmonton, Canada T6G2G8,
Tel.: +0017804926961; fax: +001780 49202200
E-mail address: secanell@ualberta.ca (M. Secanell).
tures are not constant inside the membrane electrode assembly
(MEA) of a PEMFC [3,4]. A temperature difference of 5 ◦ C or more
between the membrane-catalyst layer interface and the gas chan-
nel was observed by Vie and Kjelstrup [3], thereby showing the
limitations of the isothermal modeling assumption commonly used
in the literature [2]. Nguyen and White [5] and Fuller and Newmann
[6] first analyzed heat transfer in fuel cell MEAs. Assuming local
thermal equilibrium between the different phases in the electrode,
a single heat transfer equation was proposed. Then, an overall heat
transfer coefficient for each layer was assumed in order to predict
the heat transport inside the PEFC. In these preliminary studies,
several heat sources, such as reversible and irreversible losses asso-
ciated with the electrochemical reactions, were neglected. There
are numerous heat generation mechanisms inside the PEMFC, viz.,
reversible and irreversible heat generation due to the electrochem-
ical reactions, ohmic heating due to electron and ion transport, and
heat released/absorbed due to phase change of water. In recent
years, non-isothermal models have therefore been extended to
account for each one of these mechanisms as shown in Table 1.
Table 1 shows that, even though non-isothermal fuel cell model
in the literature account for many of the heat sources, the heat
released due to water sorption has been neglected in the major-
ity of models. Furthermore, a complete MEA model that couples

http://dx.doi.org/10.1016/j.electacta.2014.09.051
0013-4686/© 2014 Elsevier Ltd. All rights reserved.
 M. Bhaiya et al. / Electrochimica Acta 147 (2014) 294–309 295

simulation of the MEA. Ostrovskii and Gostev [9] has shown that the
Nomenclature enthalpy of sorption decreased from 68 kJ/mol (at   0) to nearly
45 kJ/mol (at   5), but values below the latent heat of water vapor-
S Change in entropy per mole of fuel, J/(mol · K) ization are not reached. On considering the enthalpy of sorption the
T̂ Temperature, ◦ C same as the latent heat of vaporization, the heat term correspond-
 Sorbed water content ing to sorption/desorption of water is observed to be of similar
F Universal Faraday’s constant, A · s/mol magnitude as the irreversible heat in a two-dimensional thermal
H+ Proton model [10]. The model however does not solve for transport of other
k Thermal conductivity, W/(cm · K) species (except temperature) and reaction kinetics.
ptot Total pressure, Pa The reversible heat release in the fuel cell is usually accounted
Qrev Reversible heat released per mole of fuel, J/mol for in most non-isothermal models. The total entropy change asso-
T Temperature, K ciated to the overall reaction producing liquid water per unit mole
xH2 O Water mole fraction of fuel (H2 ), S, is well known in the literature [11]. At T = 353 K, it
ACL Anode Catalyst Layer results in a reversible heat release of 55.99 kJ/mol-H2 inside the fuel
CCL Cathode Catalyst Layer cell. However, this heat is not evenly distributed amongst the anode
CGDL Cathode Gas Diffusion Layer and cathode reactions. Efforts to determine the entropy changes
CMPL Cathode Microporous Layer in a single half cell reaction are marred with difficulties because
FEM Finite Element Method of a lack of reliable knowledge of the entropy values of hydro-
GDL Gas Diffusion Layer nium ions (H3 O+ ) and electrons (e− ). To date, most non-isothermal
HOR Hydrogen Oxidation Reaction models in the literature assume that almost all of the reversible
MEA Membrane Electrode Assembly heat is produced inside the cathode electrode. This might not be
ML Membrane Layer the case. Ramousse et al. [12] reported a wide discrepancy in the
MPL Microporous Layer literature data for the hydrogen oxidation reaction (HOR), from -
ORR Oxygen Reduction Reaction 133.2 J/(mol-K) (Exothermic) to 84.7 J/(mol-K) (Endothermic). The
PEMFC Polymer Electrolyte Membrane Fuel Cell HOR half cell reaction is reported to be slightly exothermic [13],
RH Relative Humidity athermic [14] and slightly endothermic [15]. On studying proton
solvation in water, Ramousse et al. [12] calculated that the entropy
change is -133.2 J/(mol-K) (highly exothermic). These values are
all heat generation terms with an accurate membrane water trans- opposite to the normally modelled values in the literature. Kjel-
port model has not yet been developed. The development of such strup et al. [16] performed experiments on a hydrogen-hydrogen
a model would allow researchers to better understand the relative cell under the Soret equilibrium conditions (no current is being
importance of each heat generation term and the impact of thermal generated in the cell). They reported that at 340 K, the entropy
management on water transport in the cell. change for the HOR is −66 ± 5 J/(mol-K). However, it is also con-
Even though several non-isothermal models have been pro- cluded that as current is being generated in the cell (moving away
posed in the literature, the analysis of the relative importance of from the Soret equilibrium), there can be a considerable enthalpy
each heat generation term in the MEA has only seldom been ana- transport along with water and protons shifting the reversible heat
lyzed. Ju et al. [7] estimated the distribution of major heat source production towards the cathode side. Hence there is ambiguity in
terms, viz., reversible heat, irreversible heat, and ohmic heating. the literature for the reversible heat production distribution inside
They found the distribution of major heat sources to be roughly the cell. Therefore, the impact of the reversible heat distribution
35%, 45%, and 17% respectively at cell voltage of 0.6 V under 75%/0% should be studied.
anode/cathode RH conditions. However, water sorption effects Thermal and water management must be studied simulta-
were neglected and the reversible heat term was formulated for liq- neously since they severely affect one another. It has been
uid water product even though the model was single phase, thereby shown that water sorption/desorption will release/absorb energy.
over-predicting the heat generated in the cell. For a single phase Temperature gradients across the membrane also affect water
model to be consistent, a heat sink due to complete evaporation of management across the membrane. Water in sorbed phase is trans-
water should be included in the model, which is ignored in their ported under temperature gradients inside the polymer electrolyte
work. membrane [17,18]. This so-called thermal osmosis effect can be
Ramousse et al. [8] introduced the heat source/sink term cor- explained by the second law of thermodynamics, since water
responding to sorption/desorption of water into the electrolyte moves from the cold side (ordered state inside the hydrophillic
taking place at the catalyst layers in a one-dimensional heat transfer membranes) to the hot side in order to increase the entropy.

Table 1
Summary of non-isothermal models in the literature with respect to various heat source terms and other limitations.

Ref. Irrev. Rev. Ohm. Sorp. Other remarks

√ √ √
[7] × Reversible heat formulated for liquid water product in single phase
√ √ √ √
[8] 1D model with independent heat transfer analysis
√ √ √ √
[10] 1D heat transfer only
√ √
[53,54] × × -
√ √
[55] × × Global heat source terms instead of location-specific
√ √ √
[56] × 1D model, fully hydrated membrane (undermined ohmic heating)
√
[57] × × × Interface CL
√ √
[58] × × -
√ √ √
[59–63] × -
√ √ √
[13,64] × Interface CL
√ √
[65] × × -
√ √ √
[66] × Cathode electrode only
√ √ √
[67] × Two equation thermal model
296 M. Bhaiya et al. / Electrochimica Acta 147 (2014) 294–309

The water flux is found to be linearly dependent on the tem- Table 2

Layers in which solution variables are solved for in the MEA model.
perature difference across the membrane. Kim and Mench [19]
directly measured the thermo-osmotic fluxes on non-reinforced Variable AGDL AMPL ACL ML CCL CMPL CGDL
Nafion® membranes and reinforced Gore-Select® and Flemion® xO2
√ √ √
√ √ √ √ √ √
membranes. They observed that the water flux is from the cold to xH2 O
√ √ √ √ √ √
the hot side. It was observed that with a reduction in the equivalent s
√ √ √
weight, the thermo-osmotic flux increased significantly and was m
√ √ √

comparable to water flux due to back diffusion. Lower equivalent √ √ √ √ √ √ √
T
weight leads to better connected pathways hence increased water
flux. It is shown that the thermo-osmotic water flow becomes more
significant at high current densities and during the cell shutdown including the thermal transport equation which is discussed in the
(forcing a significant amount of water from the anode to the cath- next section, are given by,
ode, hence blocking pores for the next start-up of the cell) [20].  
Thermo-osmotic diffusion coefficients values for different mem- eff
−∇ · ctot DO ∇ xO2 = SO2 ,
brane materials have been reported in the literature [18,19,21]. 2 ,N2

Thermal osmosis effects in sorbed water transport inside the  

eff
ionomer in a non-isothermal membrane electrode assembly (MEA) −∇ · ctot DH ∇ xH2 O = SH2 O ,
2 O,N2 or H2
model have thus far not been considered in the literature.  
Based on the literature review, it is observed that a detailed non- −∇ ·
eff
m ∇ m = SH + ,
isothermal MEA model that accounts for all non-isothermal effects, (1)
and couples thermal and water management does not exist. A  
one-dimensional model utilizing non-equilibrium thermodynam- ∇ · seff ∇ s = Se − ,
ics for heterogeneous system allowed for consistent inclusion of all  
thermal effects [22]. However, this approach does not result in an eff
dry eff D
eff

explicit governing equation for determining temperature profiles −∇ · nd m ∇ m + D ∇ + T ∇T = S ,
F EW  MH2 O
inside the cell. This article presents a comprehensive single-phase
non-isothermal MEA model solving for transport of various species
where some equations are only solved in specific domains as shown
besides temperature. It assumes local thermal equilibrium and
in Table 2 and where the hydrogen mole fraction is obtained as,
accounts for anisotropic heat transport, all applicable heat sources,
xH2 = 1 − xH2 O . Equation (1), with the source terms described in
viz., reversible, irreversible, ohmic heating, phase change, heat of
the equations (3) to (7), represents the conservation equations for
sorption/desorption, and non-isothermal effects such as thermal
various transporting species (except temperature) considered in
osmosis. The mathematical model also accounts for more accu-
this model. It is noteworthy that an equal rate of sorpton/desorption
rate multi-step reaction kinetics for the oxyen reduction reaction
for the anode and cathode catalyst layers ensure that the mass of
(ORR) [23] and hydrogen oxidation reaction (HOR) [24]. The model
sorbed water is conserved. ctot , is the total concentration of the
is implemented in a modular fashion into a fuel cell simulation
gaseous mixture, expressed in terms of total pressure of the gaseous
framework, rigorously accounting for non-isothermal effects on
mixture, ptot ,
transport properties and reaction kinetics, and will be made avail-
able open-source to the PEMFC research community [25]. ptot
ctot = . (2)
In this article, the MEA model, with a detailed insight into the RT
derivation of layer-specific thermal transport equations, is first dis-
As discussed in references [26,27], only the relevant equations
cussed in sections 2 and 3. The results of the non-isothermal model
are solved in each layer. The source terms corresponding to each
are then compared to experimental data and isothermal model pre-
equation are given by
dictions, under different operating conditions in section 5. Utilizing
⎧
the model, new physical insight is provided in the article by study- ⎨ −j in CCL,
ing the impact of various effects such as thermal osmosis, reversible 4F
SO2 = (3)
heat distribution, and heat of sorption. A detailed breakdown of ⎩
0 otherwise,
various heat sources at different current densities is also provided.
The impact of non-isothermal effects on the cell performance and ⎧ j dry
water management are assessed. ⎪
⎪ − k (eq − ) in CCL,
⎪
⎨ 2F EW
SH2 O = dry (4)
⎪ −k (eq − ) in ACL,
⎪
⎪ EW
2. Membrane Electrode Assembly model ⎩
0 otherwise,
The MEA model is detailed in Bhaiya [26]. It is an across- ⎧
the-channel unit cell expanded upon the single-phase, isothermal ⎨ −j in CCL,
model by Secanell et al. [27,35,36]. The model solves for six vari- SH + = j in ACL, (5)
ables, viz., oxygen mole fraction (xO2 ), water mole fraction (xH2 O ), ⎩
0 otherwise,
electrolyte potential (m ), solid phase potential (s ), sorbed water ⎧
content (), and temperature (T). Gas and charged species trans- ⎨ −j in CCL,
port are solved using Fick’s and Ohm’s law respectively. Sorbed Se− = j in ACL, (6)
water transport inside the ionomer phase accounts for electro- ⎩
0 otherwise,
osmotic drag, diffusion under concentration gradients, thermal
osmosis, and equal rate of sorption/desorption of sorbed water
⎧
⎨ k dry (eq − ) in CLs,
inside the anode and cathode catalyst layers [28,26]. Multi-step EW
S = (7)
reaction kinetics models for source/sink terms corresponding to ⎩0 otherwise.
ORR and HOR are considered [23]. The governing equations, not
 M. Bhaiya et al. / Electrochimica Acta 147 (2014) 294–309
3. Thermal Transport Equation
In order to develop an implementation of the thermal math-
ematical model for the MEA, the following assumptions will be
considered:
1. Steady state model.
2. Ideal gas mixtures.
3. Soret and Dufour effects are neglected.
4. Very low Brinkmann number flow, hence heat production due
to viscous dissipation is negligible [29].
5. Pressure gradients are negligible, i.e., the system is isobaric.
6. Two-phase flow and condensation are not considered in the
model, therefore the gas phase is allowed to be in super-
saturated form (higher than 100% RH values)
7. Gas and solid phases are in thermal equilibrium because porous
layers, such as GDL, MPL, and CL, have very high interstitial sur-
face area between the fluid (gas) phase and the solid phase,
and therefore, there should be enough convective heat transfer
between the two phases.
In order to satisfy the assumptions made in the model, the fuel
cells are studied only under high temperature, i.e., 80 ◦ C and above,
and low relative humidity, i.e., 70% and below, operating conditions.
Weber and Hickner showed that at 80 ◦ C, even under fully humidi-
fied conditions, the amount of water removed from the cell by the
liquid phase is less than 20% [30].
The transport equation for a representative elementary volume
of the porous media under the assumptions above is [26,31],
∂[(1 − )s ĥs + g ĥg ]
 Since the convective velocity is zero, the nitrogen gas transport
+ ∇ · g ĥg vg = ∇ · keff ∇ T
∂t
297
humidified air primarily consisting of oxygen, nitrogen and water
is supplied. Therefore, in the cathode side porous diffusion media,
four species, viz., oxygen, water vapour, nitrogen and electrons
are transported. Moreover no electrochemical reactions take place
inside the layers, thus there are no additional heat and work terms,
i.e., Sheat = 0 and Ẇelectrical = 0. The thermal transport equation in
expanded form is therefore given by,
∇ · keff ∇ T − ∇ · H O2 JO2 − ∇ · H H2 O JH2 O − ∇ · H N2 JN2
− ∇ · H e− Je− = 0. (11)
If the gas mixture velocity is negligible (non-convective system),
the mixture velocity and molar flux are zero, and the species molar
flux and inter-species diffusion molar flux are equivalent such that,
Ntot = NN2 + NO2 + NH2 O = 0, (12)
Ji = Ni . (13)
The second, third, fourth and fifth terms in equation (11) can be
re-written as,
∇ · H i Ji = Ji · ∇ H i + H i ∇ · Ji , (14)
where for a non-convective, non-reactive system, mass conserva-
tion dictates that the last term must be zero.
Then, equation (11) can be further re-arranged as,
∇ · keff ∇ T − JO2 · ∇ H O2 − JH2 O · ∇ H H2 O − JN2 · ∇ H N2
− Je− · ∇ H e− = 0. (15)
is not solved for in the model. Instead, the following relation is used,
 
−∇ · H i Ji + Sheat − Ẇelectrical . (8)
where  is the layer porosity, keff is the effective thermal conductiv-
ity, , ĥ and v are density, specific enthalpy and velocity respectively
with subscripts s and g representing the solid phase and gas mixture
respectively, Ji and H i are the effective inter-species diffusion molar
flux and molar enthalpy of specie i where i can be any of O2 , N2 , H2 ,
H2 O, e− , H+ as applicable, Sheat and Ẇelectrical are volumetric rates
of heat production and electrical work done by the system respec-
tively. The terms in the equation, from left to right, represent the
energy accumulation term, which is zero for steady-state problems,
convective enthalpy transport, Fourier heat conduction, enthalpy
transport due to inter-species diffusion, other heat sources, and
electrical work done by the fuel cell. Therefore, there are three
modes of heat transport, namely, conduction, convection and inter-
species diffusion. An order of analysis shows that [26],
O(convection):O(diffusion):O(conduction)∼10−6 :10−2 :100 . (9)
The transient and convective heat transport terms are therefore
neglected in equation (8) resulting in
JN2 = −JO2 − JH2 O . (16)
Using the latter equation, equation (11) can be further re-
arranged as,
∇ · keff ∇ T − JO2 · (∇ H O2 − ∇ H N2 ) − JH2 O · (∇ H H2 O − ∇ H N2 )
− Je− · ∇ H e− = 0. (17)
The diffusive gas molar fluxes are still not the independent vari-
ables of the problem, therefore they need to be substituted by
appropriate expressions that depend only on the molar faction of
the gases, electrolyte and solid phase potentials, sorbed water and
temperature as discussed in section 2. For the gaseous species, diff-
usive transport is assumed to be governed by Fick’s law such that,
under isobaric conditions,
Ji = ctot Di,j ∇ xi . (18)
The species diffusive molar flux can be replaced by the expres-
sion above in equation (11) resulting in,
∇ · keff ∇ T − ctot DO2 ,N2 ∇ xO2 · (∇ H O2 − ∇ H N2 )
  
− ctot DH2 O,N2 ∇ xH2 O · (∇ H H2 O − ∇ H N2 ) − Je− · ∇ H e− = 0, (19)
∇ · keff ∇ T − ∇ · H i Ji + Sheat − Ẇelectrical = 0, (10)
where terms H i , molar enthalpy of species i, are a function of tem-
where different species transport, and heat sources vary with var- perature and are given by expression provided by Sonntag et al.
ious layers of the PEMFC. Hence the thermal transport equation [32]. Applying differentiation rules and re-arranging, the following
is derived separately for various PEMFC layers and described in expression is obtained,
subsequent sections.


 ∂H O2 ∂H N2
∇ · keff ∇ T + DOeff ,N ctot − ∇ T · ∇ xO2
3.1. Thermal transport equation in porous diffusion media
2 2 ∂T ∂T


eff ∂H H2 O ∂H N2
This section deals with the porous diffusion media inside the + DH ctot − ∇ T · ∇ xH2 O − Je− · ∇ H e− = 0. (20)
2 O,N2 ∂T ∂T
PEMFC, viz., GDL and MPL. Considering the cathode electrode,
298 M. Bhaiya et al. / Electrochimica Acta 147 (2014) 294–309

The last term in the equation still contains the diffusive flux where H sorption is molar enthalpy change (heat released) corre-
of electrons in the electronically conductive phase which is not sponding to sorption of water vapour into the electrolyte and H H2O
a primary variable. Electron movement inside the porous layer is is molar enthalpy of the water vapour. Based on the literature, the
governed by Ohm’s law, value of H sorption is taken to be constant at 45 kJ/mol in this work
eff [12]. Since the flux of sorbed water in the membrane is described
s
Je − = + ∇ s . (21) in the MEA model as,
F
eff eff
Further, in an adiabatic system, the molar enthalpy of an elec- m dry eff D
J = nd ∇ m + D ∇ + T ∇T (29)
tron can be simplified using electrochemical potentials [26] such F EW  MH2 O
that,
and there is no sorption/desorption of water, the thermal transport
∇ H e− = −F ∇ s . (22) equation inside the membrane layer can be finally expressed as a
function of the primary variables of the problem as [26],
According to the last two equations, the diffusive enthalpy trans-
port due to electron movement term simplifies to irreversible
heating source (Ohmic or Joule heating),

 eff
nd m ∂H  dry eff ∂H 
∇ · keff ∇ T + ∇ T · ∇ m + D ∇T · ∇

 eff
F ∂T EW  ∂T
−∇ · Je− H e− = −Je− · ∇ H e− = s (∇ s · ∇ s ) . (23)
1 eff ∂H  eff
+ D ∇ T · ∇ T + m (∇ m · ∇ m ) = 0. (30)
Replacing the last equation in equation (20), a governing equa- MH2 O T ∂T
tion for heat transport in the cathode GDL or MPL is obtained
that is only a function of the primary variables of the problem,
i.e., 3.3. Thermal transport equation in the catalyst layers



 ∂H O2 ∂H N2 In the catalyst layers, all phases co-exist, i.e. gas, electron and
∇ · keff ∇ T + DOeff ,N ctot − ∇ T · ∇ xO2 + DHeff O,N ctot proton conducting. The electrochemical reactions also take place.
2 2 ∂T ∂T 2 2
Therefore, in order to derive the governing equations, all the terms
 are considered including, Sheat and Ẇelectrical in equation (10). The
∂H H2 O ∂H N2
× − ∇ T · ∇ xH2 O + seff (∇ s · ∇ s ) = 0. (24) heat generation corresponding to the electrochemical reactions
∂T ∂T includes reversible and irreversible terms, however it will be seen
that the reversible term is automatically included due to the elec-
Following a similar process for GDL and MPL in the anode, a
trical work term. In order to derive a governing equation that is
governing equation can be obtained that is a function only of the
only a function of the primary variables, first the irreversible heat
primary variables. In the anode, humidified hydrogen (binary mix-
source and electrical work terms will be discussed in detail, and
ture of hydrogen and water vapour gases) is supplied to the gas
then the inter-species diffusion flux terms will be expanded fol-
channels. Assuming the gas mixture is infinitely dilute and consists
lowing a similar approach to that in sections 3.1 and 3.2.
primarily of hydrogen gas, water vapour transport can be approxi-
Inside the catalyst layers, energy is lost due to the activation
mated as a diffuse species moving in hydrogen according to Fick’s
polarization associated with the electrochemical reactions. It is
law. The thermal transport equation in the anode GDL or MPL is
irreversibly generated as waste heat given as [7],
derived as,

 Sirrev = j = j Eeq − Vcell , (31)

 ∂H H2 O ∂H H2
∇ · keff ∇ T + DHeff O,H ctot − ∇ T · ∇ xH2 O where j is the current density, is the overpotential, Eeq is the
2 2 ∂T ∂T
equilibrium cell potential, and Vcell is the actual cell voltage. This
eff
+ s (∇ s · ∇ s ) = 0. (25) overall irreversible heat release can be broken down into localized
irreversible heat generation corresponding to the ORR and HOR in
the respective catalyst layers,

3.2. Thermal transport equation in the membrane Sirrev,ORR = −j = −j (s − m − EORR ) , (32)

Sirrev,HOR = j = j (s − m − EHOR ) , (33)

Two species are basically transported inside the membrane,
viz., sorbed water () and protons (H+ ). Therefore, the membrane where EORR and EHOR are the equilibrium potentials for the ORR and
governing equation should only be a function of three primary vari- HOR respectively, determined using the Nernst equation [23,33].
ables: sorbed water, electrolyte potential, and temperature. There The overpotential, , for a cathodic reaction is negative, hence a
is neither production/consumption of these species nor any elec- negative sign is used in the formulation above.
trochemical reaction taking place inside the layer, therefore Sheat = 0 The electrical work term is given as
and Ẇelectrical = 0 in equation (10). The thermal transport equation j j

in expanded form is therefore given by, −Ẇelectrical = GORR = H ORR − TS ORR , (34)
nF nF




∇ · keff ∇ T − ∇ · H  J − ∇ · H H + JH + = 0. (26) where GORR , H ORR and, S ORR are molar Gibbs free energy
change, enthalpy change and entropy change respectively for the
Similar to the discussion in section 3.1, the diffusive enthalpy
ORR producing n moles of electrons per mole of fuel (H2 ). The value
transport due to proton movement term is given by,
of n is 2 in this case. H ORR can be determined as,

 eff
−∇ · JH + H H + = m (∇ m · ∇ m ) . (27) 1
H ORR = H H2 O(l) − HO − 2H H + − 2H e− . (35)
2 2(g)
The membrane is assumed to be in equilibrium with water
vapour. Also the molar enthalpy of sorbed water is given by, Species transported inside the cathode catalyst layer are oxy-
gen, water vapour, nitrogen, electrons, protons and sorbed water.
H  = H H2O − H sorption , (28) The catalyst layers are non-convective but reactive systems, so the
 M. Bhaiya et al. / Electrochimica Acta 147 (2014) 294–309 299

last term in the equation (14) is not zero. For the CCL, on utiliz-
ing the source terms given in the equations (3)-(7) and the latent
heat of vaporization for expressing enthalpy of water vapour in
terms of liquid water enthalpy, the last term can be expanded
as,

j
1  j
H i ∇ · Ji = − H O2 − H H2 O(l) + 2H e− + 2H H + + H
2F 2 (g) 2F lv
k dry

− eq −  (H H2 O − H  ), (36)
EW

where H lv is the molar latent heat of vaporization of water [34],

H lv = 2500.3 − 2.25T̂ − 0.022T̂ 1.5 + 3.18 × 10−4 T̂ 2.5

− 2.86 × 10−5 T̂ 3 , (37)

where T̂ is the temperature in degree-Celsius.

It can be clearly seen that the enthalpy terms corresponding to
the ORR cancel each other out when equation (36) is added to the
electrical work term. The remaining term in the electrical work (last
term in the equation (34)) represents the reversible heat generated
in the reaction. A factor, fORR , is used to consider the fraction of
reversible heat released in the ORR with respect to reversible heat
released in the overall reaction,
Fig. 1. Description of the computational domain.
j
 j

Srev,ORR = −TS ORR = −TfORR S overall , (38)
2F 2F
3.4. Summary
where S overall is the overall entropy change per mole of fuel (H2 ).
The last two terms in equation (36) represents the heat released In summary, the governing equations of the thermal model are a
or absorbed due to water evaporation, and sorption/desorption of function of all primary variables of the MEA model, namely, oxygen
water vapour inside the ionomer respectively. Expanding the sec- and water vapour mole fraction, electronic and electrolyte poten-
ond last term in the equation (14) as discussed in the previous two tials, sorbed water, and solid phase temperature. In the cathode and
sections, the thermal transport equation inside the CCL can finally anode GDL and MPL, equations (24) and (25) are solved respec-
be expressed as [26], tively. In the cathode and anode CL, equations (39) and (40) are
solved respectively. Finally, in the membrane, equation (30) is used.


 eff ∂H O2 ∂H N2
∇ · k ∇T eff
+ DO ,N ctot − ∇ T · ∇ xO2 4. Boundary conditions, numerical solution and input
2 2 ∂T ∂T
 parameters
eff
eff ∂H H2 O ∂H N2 nd m ∂H 
+ DH ctot − ∇ T ·∇ xH2 O + ∇ T ·∇ m Fig. 1 describes the two-dimensional seven-layered computa-
2 O,N2 ∂T ∂T F ∂T
tional domain considered in this work. Contact resistances between
dry eff ∂H  1 eff ∂H  the layers are neglected, i.e., the solution is assumed to be con-
+ D ∇ T · ∇ + D ∇ T · ∇ T −j (s −m −EORR )
EW ∂T MH2 O T ∂T tinuous across the complete domain. The geometric dimensions
of the computational domain are given in Table 3. There are five
j j eff eff
− TfORR S overall − H + m (∇ m · ∇ m ) + s (∇ s · ∇ s ) types of boundaries, as shown in the Fig. 1. Symmetric (zero
2F 2F lv flux) boundary conditions are applied at the top and bottom
k dry
 boundaries.
+ eq −  H sorption = 0. (39)
EW The boundary conditions at AGDL-Gas channel boundary are
given by,

Similarly, a separate equation can be derived for the ACL [26],


−n · Deff ∇ xO2 = 0,

 xH2 O ◦
= xH ,

 ∂H H2 O ∂H H2 2 O,a
∇ · k ∇T eff eff
+ DH O,H ctot − ∇ T · ∇ xH2 O  
2 2 ∂T ∂T −n ·
eff
s ∇ s = 0,
eff
nd m ∂H  dry eff ∂H  1 eff ∂H    (41)
+ ∇ T ·∇ + D ∇ T ·∇ + D ∇ T ·∇ T −n ·
eff
m ∇ m = 0,
F ∂T EW  ∂T MH2 O T ∂T

j eff
 
+ j (s − m − EHOR ) − T (1 − fORR )S overall + m (∇ m · ∇ m ) 
eff
dry eff D
eff
2F −n · nd m ∇ m + D ∇ + T ∇T = 0,
F EW  MH2 O
k dry

eff
+ s (∇ s · ∇ s ) + eq −  H sorption = 0 (40)

EW −n · keff ∇ T = 0,
300 M. Bhaiya et al. / Electrochimica Acta 147 (2014) 294–309

Table 3 interface until the variable is solved. For example, all electrolyte
Geometric dimensions of the computational domain
potential boundaries are applied at MPL-CL interfaces. At the gas
Parameters Value, (cm) channel/GDL interfaces, the water molar fraction is specified that
lGDL 2.5 × 10−2 results in the given relative humidity for the operating tempera-
lMPL 5 × 10−3 ture and pressure of the cell. The relative humidity is calculated
lACL 3.33 × 10−4 locally using the ratio of the water partial pressure and the satura-
lCCL 1 × 10−3 tion pressure. If the temperature of the gas at the interface increases
lML 2.5 × 10−3
due to non-isothermal effects, the relative humidity in the channel
lRib 5 × 10−2
lChannel 5 × 10−2 is slightly reduced, e.g., at 2100 mA/cm2 it is 47.3% instead of 50%.
The system of governing equations is non-linear, therefore a
Newton’s iterative method is employed to solve the problem where,
◦
where n is the unit surface normal, and, xH , is determined based
2O
starting with an initial guess, a solution update is obtained at each
on the operating conditions [35]. iteration by solving a linearized form of the governing equations.
The boundary conditions at AGDL-Bipolar plate boundary are The solution variables are then updated accordingly, and the pro-
given by, cess is repeated. In the case of openFCST, the L2 norm of the residual

 corresponding to all equations is computed at every iteration step
−n · Deff ∇ xO2 = 0, and the simulation terminates when the L2 norm has reached the

 tolerance value of 10−8 .
−n · Deff ∇ xH2 O = 0, The linearized form of the governing equations in this case
is obtained analytically using calculus of variations, see refer-
s = 0, ence [35]. The linearized system of equations is discretized using
  (42) Galerkin weighted residuals method with continuous second order
eff
−n · m ∇ m = 0, Lagrange finite elements for test and solution approximation func-
tions. The global system matrix and the right hand side vectors are
 
eff
dry eff D
eff assembled using Gaussian numerical quadrature. A direct solver,

−n · nd m ∇ m + D ∇ + T ∇T = 0, namely UMFPACK [37], is used to solve the linear system since the
F EW  MH2 O
global system matrix is non-symmetric.
T = Tcell , The governing equations are solved in a discretized domain
using an unstructured quadrilateral mesh. Once a converged solu-
where Tcell is the cell operating temperature applied at the bipolar tion is achieved, the mesh is adaptively refined to two more levels
plates. of refiniment utilizing the adaptive error estimator by Kelly et al.
The boundary conditions at CGDL-Gas channel boundary are [38], provided in the deal.II [39,40], in order to achieve a grid-
given by, independent solution. The last solution has 125,718 degrees of
xO2 ◦
= xO , freedom.
2 ,c
Parameters for electron, proton, gas, and sorbed water transport
xH2 O ◦
= xH ,
2 O,c
are detailed in Bhaiya [26], adapted from [27,41]. The electro-
  osmotic drag coefficient, nd , is considered to be one based on the
eff
−n · s ∇ s = 0, data by Ren and Gottesfeld [42]. The water diffusion coefficient, D ,
  is obtained using the functions proposed by Motupally et al. [43].
−n ·
eff
m ∇ m = 0, (43) Correlations for thermo-osmotic diffusion coefficients are used
from Kim and Mench [19]. The GDL effective thermal conductivity
  in the through-plane direction is given by [44],
eff eff
 dry eff D
−n · nd m ∇ m + D ∇ + T ∇T = 0,

F EW  MH2 O kthrough = M −7.166 × 10−6 T̂ 3 +2.24 × 10−3 T̂ 2 −0.237T̂ +20.1
eff


−n · keff ∇ T = 0,
× W/(m − K), (45)
where ◦
xO
is determined based on the operating conditions [35].
2
The boundary conditions at CGDL-Bipolar plate boundary are where T̂ is the temperature in degree-Celsius and M is the heat
given by, barrier resistance coefficient. Fitting to the data by Zamel et al. [44],


−n · Deff ∇ xO2 = 0,

 M = −1.495 × 10−11 T̂ 5 + 2.601 × 10−9 T̂ 4 − 6.116 × 10−8 T̂ 3
−n · Deff ∇ xH2 O = 0,
− 9.829 × 10−6 T̂ 2 + 8.754 × 10−4 T̂ + 0.0664. (46)

s = Vcell ,
  (44)
eff The GDL effective thermal conductivity in the in-plane direction
−n · m ∇ m = 0,
is given by [45],
 
eff eff
 dry eff D kin−plane = −7.166 × 10−6 T̂ 3 + 2.24 × 10−3 T̂ 2 − 0.237 T̂
eff
−n · nd m ∇ m + D ∇ + T ∇T = 0,
F EW  MH2 O
+ 20.1 W/(m − K). (47)
T = Tcell ,

where Vcell is the operating cell voltage.

Note that if the solution variable is not solved in a given inter- The thermal conductivities for the various layers considered in
face, the boundary conditions is translated inwards into the next this work are given in Table 4.
 M. Bhaiya et al. / Electrochimica Acta 147 (2014) 294–309 301

Table 4
Effective thermal conductivity, W/(cm-K)

Layers Value

GDL through-plane Equation (45)

GDL in-plane Equation (47)
MPL through-plane 0.003 [52]
MPL in-plane 0.0387 [51]
CL 0.0027 [68]
Membrane 0.0013 [68]

Table 5
Operating conditions

Parameters Value

Anode relative humidity 50% and 70%

Anode pressure 1 atm and 2 atm
Cathode relative humidity 50% and 70%
Cathode pressure 1 atm and 2 atm
Cell temperature 353.15 K

Fig. 3. Polarization curve for the 50% RH and 2 atm case where solid red line repre-
5. Results and Discussion sents the non-isothermal case, solid blue line represents the isothermal case, and dark
green triangle points represents the experimental data [41]. The point at which the
5.1. Non-isothermal effects on performance under different model predicts non-physical RH values (more than 100%) is denoted by × symbol,
and the polarization curves are dashed line afterwards.
operating conditions

The cell performance under different operating conditions is

evaluated with the isothermal and non-isothermal mathematical
models developed. Simulations are run for two different operating
RH conditions, viz., 50% and 70% RH, and two different operating
pressures, viz., 1 atm and 2 atm. Operating conditions parameters
are given in Table 5. Non-isothermal and isothermal simulations
are run employing exactly the same parameters such as transport
properties, kinetics and layer composition. In the case of isother-
mal simulations, temperature is not being solved for and the cell
temperature is maintained at 80 ◦ C. The same temperature value
is applied as a Dirichlet BC on the bipolar plate boundaries in the
non-isothermal simulation.
Figs. 2, 3, 4 and 5 show the polarization curves (i-V curve) for the
various operating conditions discussed before. Model predictions
are also compared against experimental data. The experimental
data was published previously by our research group in collabo-
ration with the National Research Council Canada - Institute for
Fuel Cell Innovation (NRC-IFCI) [41]. The MEA parameters such as
Fig. 4. Polarization curve for the 70% RH and 1 atm case where solid red line repre-
sents the non-isothermal case, solid blue line represents the isothermal case, and dark
green triangle points represents the experimental data [41]. The point at which the
model predicts non-physical RH values (more than 100%) is denoted by × symbol,
and the polarization curves are dashed line afterwards.

computational geometry, transport properties and kinetics param-

eters, are in line with the tested PEMFC configuration. Simulations
are performed by varying the cell voltage from 1.0 V to 0.05 V. The
current density is computed as an integral in the CCL (see reference
[46] for details). The isothermal and non-isothermal polarization
curves change from solid to dashed lines at the point where RH
values in the cell become non-physical (greater than 100%). The RH
values are determined locally as,

pH2 O ptot × xH2 O

RH = = , (48)
psat × 101325 psat × 101325

where psat is the saturation pressure of water (in atm) determined

locally as a function of temperature [28],

Fig. 2. Polarization curve for the 50% RH and 1 atm case where solid red line repre-
sents the non-isothermal case, solid blue line represents the isothermal case, and dark
green triangle points represents the experimental data [41]. The point at which the
log10 (psat ) = −2.1794 + 0.02953 (T − 273.15) − 9.1837
model predicts non-physical RH values (more than 100%) is denoted by × symbol,
× 10−5 (T − 273.15)2 + 1.4454 × 10−7 (T − 273.15)3 . (49)
and the polarization curves are dashed line afterwards.
302 M. Bhaiya et al. / Electrochimica Acta 147 (2014) 294–309
Fig. 5. Polarization curve for the 70% RH and 2 atm case where solid red line repre-
sents the non-isothermal case, solid blue line represents the isothermal case, and dark
green triangle points represents the experimental data [41]. The point at which the
model predicts non-physical RH values (more than 100%) is denoted by × symbol,
and the polarization curves are dashed line afterwards.
At higher operating pressure, relative humidity values above
100% are predicted at smaller current densities than for 1 atm
conditions. The reason the current density at which two-phase
flow starts changes with gas pressure is due to slower diffusion
in the electrode at higher gas pressures. It is well known from the
kinetic theory of gases that the product of pressure and diffusion
coefficient is constant [31]. Therefore, as the pressure increases
water vapor transport in the electrode is reduced leading to an
increase of water vapour mole fraction and hence RH values inside
the CCL.
In the 50% RH and 1 atm case, the non-isothermal model pre-
dictions significantly under-perform the isothermal results and are
closer to the experimental data. Differences of upto 150 mA/cm2 are
observed at 0.1 V. The non-isothermal model predicts high proton
transport losses due to drying of the ionomer in the catalyst layer.
The drop in electroyte potential then results in the observed per-
formance drop. Ionomer drying is not significant in other cases due
to high RH values which lead to a fully humidified ionomer in both
simulations, i.e., isothermal and non-isothermal. At high current
densities, oxygen gas transport is improved in the case of the non-
isothermal simulations due to higher temperatures inside the MEA.
At high humidity, this effect can be observed by a slight increase in
the cell limiting current.
In the higher humidity and higher pressure cases, negligible dif-
ferences are observed between the isothermal and non-isothermal
model predictions. These results suggest that the use of a non-
isothermal model is required at low relative humidity conditions
and that, at high humidity conditions, both models might lead to
similar polarization curves at low and medium current densities.
At higher operating temperatures, the differences between iso-
thermal and non-isothermal models are likely to increase since
membrane dryout would become more frequent. At high current
densities, a two-phase flow model is required in order to assess
the main differences between the two models, however based on
the different current density at which both models reach 100%
relative humidity, it is expected that the non-isothermal model
would provide higher current density predictions in this region.
This is the subject of our future research. This work is limited to
studying the current densities at which the relative humidity any-
where inside the electrode is less than 100%. Only the breakdown
of heat losses and water flux inside the membrane are extended
beyond this value for completeness even though two-phase flow
Fig. 6. Temperature distribution plots in the MEA under 50% RH and 1 atm condi-
tions at: (a) 1000 mA/cm2 , and (b) 2100 mA/cm2 .
is likely to influence the results in this case. Simulation predic-
tions match quite well with the experimental data in the kinetics
regime (low current densities). No difference is observed between
the isothermal and non-isothermal results in this regime. However
in 1 atm cases, both the isothermal and non-isothermal model pre-
dictions significantly deviate from the experimental data as current
densities increase. This deviation is not observed in the 2 atm cases.
The disagreement at moderate current densities, i.e., when the rel-
ative humidity is below 100%, might be due to several reasons such
as multi-component and convective mass transport effects, Knud-
sen effects in the catalyst layer and micro-porous layer and, the
lack of consideration of micro-scale effects, such as agglomeration
of catalyst in the catalyst layer. At high current densities, i.e., when
the relative humidity is above 100%, the single-phase modelling
assumption is not valid and the discrepancy is increased further
due to the presence of liquid water inside the pores. It is notewor-
thy that the purpose of this article is not to match the experimental
data, but to analyze the impact of various non-isothermal effects on
model predictions. The 50% RH and 1 atm case is considered for fur-
ther analysis, as the non-physical RH values are observed at the high
current density regime only (more than 2000 mA/cm2 ), and hence
the single phase predictions are largely valid. Subsequent sections
discuss the impact of various non-isothermal effects on cell perfor-
mance, temperature and heat distribution, and water management
inside the MEA at 50% RH and 1 atm.
5.2. Temperature distribution
The temperature distribution in the MEA at two current den-
sities, i.e., 1000 mA/cm2 and 2100 mA/cm2 , is shown in Fig. 6 for
the 50% RH and 1 atm case. The highest temperatures in the MEA
are observed in the regions where electrochemical reactions take
place, i.e., in the catalyst layers. Fig. 7 shows the temperature
Fig. 7. Temperature distribution plots in ACL + ML + CCL under 50% RH and 1 atm
conditions at: (a) 1000 mA/cm2 , and (b) 2100 mA/cm2 .
 M. Bhaiya et al. / Electrochimica Acta 147 (2014) 294–309 303

Fig. 8. Plot of maximum temperature in the MEA versus the current under different Fig. 9. Plot of magnitudes of individual heat terms in the MEA versus the current,
operating conditions. for the 50% RH and 1 atm case.

is now removed in the gas phase remains inside the cell. Third, in
distribution in the anode, membrane and cathode catalyst layers. the case of Thomas et al. [48], supply gases were heated to 80 ◦ C
At 1000 mA/cm2 , the CCL region under the rib is the hottest due to before feeding into the cell. This will result in a convective heat
water sorption in the membrane as will be discussed in detail in flux into the channel boundary, in contrast to the no heat flux
section 5.4. At 2100 mA/cm2 however, the hottest region switches assumption used in our model. To study the feasibility of second
to the region under the channel. This is due to the fact that current hypothesis, the proposed model was solved without accounting
production is limited to the under-the-channel region at high for water vaporization, i.e., assuming all water leaves in liquid form
current densities. It is also observed that the CCL (right side) is without affecting transport. In this case, at 1000 mA/cm2 , the tem-
generally hotter than the ACL (left side), since there are a number of perature rise inside the cell increased to 2.8 ◦ C. This result highlights
significant heat sources in the CCL, such as, irreversible, reversible that thermal and water management are closely interconnected.
and protonic ohmic heating. Experimental studies at low relative humidity would be beneficial
Fig. 8 illustrates the maximum temperature in the MEA at dif- as they would allow researchers to study thermal and water man-
ferent current densities at the four different operating conditions agement independently of two-phase flow, an area that is still not
under study. The maximum temperature increases by up to 13 ◦ C at well understood in fuel cells.
high current densities. Based on these results, it is difficult to justify
the isothermal assumption generally used in the PEMFC modelling 5.3. Heat source distribution
literature. The lowest temperature in all three cases is 353.15 K
(80 ◦ C). This temperature is observed at the Dirichlet boundaries In order to understand the heat distribution inside the cell and
(bipolar plates). Higher temperatures are observed when the cell its main contributors, the heat source/sink distribution in the MEA
operates at 2 atm since higher current densities are achieved; how- is studied. The magnitudes of the individual heat terms in the MEA
ever, the temperature rise is steeper in the 1 atm cases because of are plotted in Fig. 9, for the 50% RH and 1 atm case, in the whole
the higher overpotential losses, i.e., higher irreversible heat pro- range of current densities. The fraction of the total reversible heat
duction. released in the cathode during the ORR, fORR , is considered to be
Temperature measurements inside the PEMFC have been 1. Irreversible heating due to the overpotential in the CCL dom-
reported by Vie and Kjelstrup [3], Zhang et al. [47] and Thomas inates the bulk of heat release in the cell, followed by protonic
et al. [48] for a fuel cell operating at 64 ◦ C, 50 ◦ C and 60 ◦ C under ohmic heating and reversible heat generation. The reversible heat
fully humidified conditions, respectively. At 1000 mA/cm2 , they released by to ORR is mostly counterbalanced (roughly 80 %) by the
report a cell temperature rise of approximately 5 ◦ C, 5 ◦ C and 3.5 ◦ C,
respectively. Due to the single-phase assumption in the proposed
model, fuel cells under fully humidified conditions cannot be stud-
ied, therefore only a qualitative assessment can be performed. At
1000 mA/cm2 , the proposed model predicts a temperature rise of
about 2 ◦ C under any of the operating conditions under study,
thereby under predicting the maximum temperature recorded
experimentally. The under prediction might be due to several rea-
sons. First, the geometry and fuel cell materials are different in this
study and the experimental studies discussed above. Second, at the
operating conditions under study, the proposed model predicts the
complete vaporization of water produced during the reaction. This
acts as a considerable heat sink in the CCL, as it almost counter-
balances the reversible heat released during the reaction, see Figs. 9
and 10. In fully humidified cells, complete evaporation of water will
not take place inside the CCL as the gas phase is already near sat-
uration. Finally, due to the fully humidified conditions, additional
condensation of water will take place in the GDL as the tempera-
ture in the GDL is lower, resulting in heat release in the GDL due Fig. 10. Plot of magnitudes of individual heat terms in the MEA versus the current,
to phase change and hence higher temperatures as the energy that upto 1200 mA/cm2 , for the 50% RH and 1 atm case.
304 M. Bhaiya et al. / Electrochimica Acta 147 (2014) 294–309
Fig. 11. Temperature (T) distribution plots in the ACL + ML + CCL, under 50% RH
and 1 atm operating conditions at 1000 mA/cm2, (a) with heat of sorption, and (b)
without heat of sorption.
energy needed to completely evaporate the water produced dur-
ing the ORR (single phase model assumption). Fig. 10 zooms in the
low to medium current densities (up to 1200 mA/cm2 ) of the heat
distribution plot. Up to 1000 mA/cm2 , the heat released/absorbed
due to sorption/desorption of water in the catalyst layers is
higher than the protonic ohmic heat source. Even at high current
densities as shown in the Fig. 9, the heat of sorption/desorption
can reach up to 10 % of the total heat released/absorbed in the cell.
Ohmic heating due to proton current overtakes the heat of sorp-
tion/desorption at 1000 mA/cm2 , and increases significantly (up to
30 %) at higher current densities. Therefore, the effect of the heat
of sorption/desorption is not negligible and should be included in
any non-isothermal model. On the other hand, irreversible heating
in the ACL and electronic ohmic heating in the MEA are generally
negligible.
Based on the heat source distribution discussed above, it is
easy to explain the observed temperature distribution reversal in
the catalyst layer in Fig. 7. At low current density, the reaction
is uniformly distributed in the CCL. The only heat source that is
not uniformly distributed is the heat absorbed due to desorption
of water in the CCL which is largest under the channel where
there is less water in the vapour phase, hence resulting in higher
temperatures under the land area. At high current densities, the
electrochemical reaction is concentrated in the area under the
channel and the heat generated due to irreversible heating is very
high, therefore the highest temperature is observed in that region
of the CCL.
5.4. Effect of heat of sorption
In order to assess the importance of water sorption in determin-
ing the temperature distribution, a simulation is performed using
the same parameters but with heat of sorption turned OFF. The
comparison between temperature distribution in the ACL + ML +
CCL, for 50% RH and 1 atm case at 1000 mA/cm2 and for 70% RH and 1
atm at 1200 mA/cm2 are plotted in Figs. 11 and 12 respectively with
and without soprtion. It is observed that the ACL is colder and CCL is
hotter when the heat of sorption is not included in the model. This
is observed because equal amounts of heat are released/absorbed
in the ACL/CCL, leading to cooling of CCL and heating of ACL. In
the case of 50% RH, Fig. 11(b) shows that the temperature distri-
bution is uniform in the in-plane direction for the case without
heat of sorption. This is not observed when the heat of sorption
is included thereby showing that the heat of sorption creates con-
siderable anisotropy in the heat distribution. In the 70% RH case
however, a uniform temperature distribution in the in-plane direc-
tion is observed when heat of sorption is considered (Fig. 11(a)).
Fig. 12. Temperature (T) distribution plots in the ACL + ML + CCL, under 70% RH
and 1 atm operating conditions at 1200 mA/cm2, (a) with heat of sorption, and (b)
without heat of sorption.
The 70% RH case is evaluated at a higher current density, i.e., when
irreversible heat generation is higher, causing the temperature in
the under-the-channel region to be higher when the heat of sorp-
tion is not included. Cooling in the under-the-rib region due to
heat of sorption causes the temperature to drop in this area and
become uniform in the in-plane direction. The sorption/desorption
heat term therefore has a significant impact on determining the
temperature distribution and therefore, it should be considered for
proper assessment of the thermal management in the PEMFC.
5.5. Effect of reversible heat distribution
As discussed in the introduction, the reversible heat distribution
between the ORR and HOR is still ambiguous. The overall reaction
of hydrogen and oxygen producing water is exothermic. In order to
assess the impact of the reversible heat distribution, a simulation
is run employing the factor of heat released in the ORR, fORR , to be
1.0, 0.5, 0.0 and -0.385 (the latter is based on the data by Ramousse
et al. [12]). Fig. 13 compares the temperature distribution inside
the ACL + ML + CCL for different values of fORR , at 1000 mA/cm2 and
210 mA/cm2 . The ACL is observed to be hotter than the CCL in the
case of exothermic HOR reaction. A gradual shift in temperature
distribution can be observed in Fig. 13, with temperature profiles
completely reversed in the in-plane and through-plane direction
at 1000 mA/cm2 . The heat of sorption alongside the reversible heat
release makes the under-the-channel region of the ACL the hottest.
Fig. 14 plots the variation of reversible, irreversible and water
vaporization heat terms in the CLs with current density for the case
with fORR = −0.385. The irreversible heating in the CCL is observed
to be the dominating heat source. However it is counterbalanced by
the heat sinks corresponding to the ORR and complete vaporization
of water such that the net heat source due to these three effects in
the CCL is significantly smaller than the reversible heat produced
due to the HOR in the ACL.
Fig. 15 compares the performance under 50% RH and 1
atm conditions for fORR = 1.0 and fORR = −0.385. The temperature
distributions are different, however only a slight reduction in per-
formance is predicted in the case of an endothermic ORR reaction.
Even though the effect on the overall performance is small, the
reversible heat distribution significantly influences the tempera-
ture distribution in the CCM. Experimental evidence by Vie and
Kjelstrup [3] under fully humidified conditions suggests that the
anode catalyst layer is hotter than the cathode, while Thomas et al.
[48] suggest otherwise. This results suggest that experiments at low
relative humidity and high temperature could be used to clarify the
reversible heat distribution in the cell.
 M. Bhaiya et al. / Electrochimica Acta 147 (2014) 294–309
Fig. 13. 50% RH and 1 atm case temperature distribution in the ACL + ML + CCL, at 1000 mA/cm2 , for: (a) fORR = 1.0, (b) fORR = 0.5, (c) fORR = 0.0, and (d) fORR = −0.385; and at
2100 mA/cm2 , for: (e) fORR = 1.0, (f) fORR = 0.5, (g) fORR = 0.0, and (h) fORR = −0.385.
Fig. 14. Variation of reversible, irreversible and water vaporization heat terms in
the CLs with current for fORR = −0.385.
5.6. Non-isothermal effects on water management
In the 50% RH and 1 atm case, the lower performance of the
non-isothermal model is attributed to proton transport limitations.
Proton conductivity in the electrolyte (Nafion® ) increases with
membrane water content () and temperature (T). In a vapour-
equilibriated membrane,  values in the CLs are determined
based on the local RH values. The production rate of water vapour
increases with current density, leading to higher xH2 O values
(hence higher RH). This can be observed by the increasing  values
in the CCL as the current density increases, as shown in the Fig. 16.
Water vapour saturation pressure is constant for the isothermal
model (due to constant temperature). At a current density of
2100 mA/cm2 , this leads to non-physical RH values (more than
305
Fig. 15. Polarization curve for the 50% RH and 1 atm case, where solid blue line rep-
resents the case of fORR = 1.0, and dotted red line represents the case of fORR = −0.385.
100 %) in the CCL for the isothermal model. These results in a nearly
constant high  value in the CCL in Fig. 16(b). In the non-isothermal
model, RH values decrease with temperature (saturation pressure).
The membrane is therefore not fully humidified as observed in
Fig. 16(d). The increase in protonic ohmic losses are therefore
corroborated by this observation.
Fig. 17 compares the membrane water content distribution in
the ACL + ML + CCL at 1000 mA/cm2 and 2100 mA/cm2 , for the
isothermal and non-isothermal cases. In both cases, a large gra-
dient in water content is observed between anode and cathode in
agreement with the neutron imaging results for a fully humdified
MEA operating at 80 ◦ C and 1000 mA/cm2 in reference [30]. It can
be seen that all the layers (ACL + ML + CCL) have much lower water
306 M. Bhaiya et al. / Electrochimica Acta 147 (2014) 294–309

Fig. 17. 50% RH and 1 atm case sorbed water content () plots in the ACL + ML +
CCL, for the isothermal case at: (a) 1000 mA/cm2 , and (c) 2100 mA/cm2 ; and for the
non-isothermal case at: (b) 1000 mA/cm2 , and (d) 2100 mA/cm2 .
Fig. 16. 50% RH and 1 atm case sorbed water content () plots in the CCL, for
the isothermal case at: (a) 1000 mA/cm2 , and (b) 2100 mA/cm2 ; and for the non-
isothermal case at: (c) 1000 mA/cm2 , and (d) 2100 mA/cm2 .

content in the non-isothermal case due to high temperature values.

Temperatures in the ACL increase due to heat released by sorption
of water. RH values in the ACL are therefore decreased leading to
even lower  values as observed in the Figs. 17(b) and 17(d). Water
in the sorbed form travels from the ACL to the CCL due to electro-
osmotic drag, counterbalanced by the back diffusion. However due
to temperature gradients, thermal osmosis is observed from the low
temperature region (ACL) to the high temperature region (CCL).
The non-isothermal model also affects the water transport in
the vapour phase. Fig. 18 shows the relative humidity values in
the cathode electode (CCL + CMPL + CGDL) for the isothermal and
non-isothermal models at different current densities. No significant
differences are observed for the low current case. However as the
current density increases, water vapour under-the-rib area is very
different for the two cases. Fig. 18(d) shows higher relative humid-
ity values under-the-rib for the non-isothermal case. Non-physical
RH values (more than 100 %) can be observed for the isothermal
case. It is interesting to note that the isothermal model predicts
liquid water condensation in the CCL, while in the case of non-
isothermal model, liquid water condensation will take place near
the rib in the CGDL. It also asserts the need for a two-phase model
to be solved alongside the non-isothermal model.
The water flux across the membrane is also evaluated. The
parameter, ˇ, is estimated as,

Water membrane flux × F

ˇ= , (50)
i
where the water membrane flux is measured from the ACL to the
CCL, i is the current density, and F is the Faraday’s constant. It
effectively represents the number of moles of water transported
Fig. 18. 50% RH and 1 atm case relative humidity plots in the cathode electrode (CCL
+ CMPL + CGDL), for the isothermal case at: (a) 1000 mA/cm2 , and (b) 2100 mA/cm2 ;
and for the non-isothermal case at: (c) 1000 mA/cm2 , and (d) 2100 mA/cm2 . It can
be seen in the case (b) that isothermal model predicts non-physical RH conditions
at current values lower than the single-phase non-isothermal model.
 M. Bhaiya et al. / Electrochimica Acta 147 (2014) 294–309 307

Fig. 19. ˇ versus the current in 50% RH and 1 atm case, where dashed blue line represents the isothermal model, solid red line represents the non-isothermal model without
thermal osmosis, and dash-dot green line represents the non-isothermal model with thermal osmosis.

Fig. 20. ˇ versus the current in 70% RH and 1 atm case, where dashed blue line represents the isothermal model, solid red line represents the non-isothermal model without
thermal osmosis, and dash-dot green line represents the non-isothermal model with thermal osmosis.

per one mole of electron produced in the reaction. Figs. 19 and 20
plots the ˇ as the current density increases, for 50% and 70% RH,
1 atm cases. Thermal osmosis is not usually accounted for in MEA
models. In order to assess the impact of thermal osmosis on the non-
isothermal model, simulations are performed with the isothermal
model, the non-isothermal model and the non-isothermal model
with the thermal osmosis term switched OFF. For all three cases, ˇ
increases as the current density is increased from low to medium
current densities. In the isothermal model, ˇ is the smallest due to
higher water accumulation in the CCL thereby increasing back dif-
fusion. For the isothermal case, the water flux also starts dropping
at medium current density, as back diffusion counteracts electro-
osmotic drag. It can be seen that the isothermal model predicts zero
ˇ value at around 1500 mA/cm2 for 70% RH case. However as cur-
rent density is increased further, electro-osmotic drag continues
to increase while back diffusion has reached its maximum value
thereby resulting in a rapid rise in water flux. The non-isothermal
models on the other hand predict a steady rise in water flux from the
ACL to the CCL. This is due to the fact that the CCL is drier (due to the
increased temperatures), hence back diffusion does not completely
counterbalance the electro-osmotic drag.
Comparing the two non-isothermal models, it is clear that the
introduction of thermal osmosis in the mathematical model results
in a significant increase in water flux predictions from anode to
cathode. In the 50% RH case, the water flux increases by as much as
15 % due to thermal osmosis at medium to high current densities. In
the 70% RH case, thermo-osmosis causes the water flux to increase
by 10% at low to medium current densities, but the effect is more
pronounced at high current densities, i.e., more than 2000 mA/cm2 ,
where water flux increases by as much as 30%. It is noteworthy that
temperature differences between the ACL and the CCL are expected
to be higher with the introduction of a two-phase model. This will
further enhance the water flux due to thermal osmosis. Thermal
osmosis effects are therefore significant and cannot be neglected
when analyzing water management in an MEA.
Water flux across the membrane at higher temperatures are
not usually reported. Yan et al. [49] reported water fluxes at
500 mA/cm2 under different channel humidity conditions. In order
to qualitatively validate the proposed water flux model, the MEA
model was used to simulate water transport under the follow-
ing conditions: 30%/70%, 50%/70% and 70%/70% cathode and anode
relative humidity values, respectively. To better compare to the
model, the membrane properties were changed to those used in
the study, i.e., Nafion® 117, by modifying the thickness and using
the proton conductivity, water diffusion coefficient and electro-
osmotic drag coefficient from Springer et al. [28] and sorption
isotherm from Hinatsu et al. [50]. The ˇ values reported by Yan
et al. [49] were 0.21, 0.12 and -0.01 while the numerical values
obtained from our model are 0.57, 0.51 and 0.43. Even though the
numerical predictions for the ˇ are much higher, the same trend
with decreasing cathode humidity is observed. Given the low sto-
ichiometries, which challenge the assumption of constant reactant
concentration along the channel used to formulate an across the
channel model, and that the GDL and CL properties are not reported
by Yan et al. and hence, it could not be modelled accurately, only
qualitative agreement could be expected.
6. Conclusions
A single-phase, non-isothermal model of a polymer electrolyte
membrane electrode assembly has been developed in the open-
source software, openFCST [25]. The model takes into account all
the relevant non-isothermal effects such as anisotropic heat trans-
port, irreversible, ohmic, phase change, reversible, sorption heating
effects and thermal osmosis. The model was shown to reproduce
the performance curves observed experimentally at a variety of
operating pressures and temperatures. At high oxygen concentra-
tions the predictions are in excellent agreement. At low oxygen
pressures, results are in agreement at low and moderate current
densities but differ at high current density most likely due to mass
308 M. Bhaiya et al. / Electrochimica Acta 147 (2014) 294–309
transport losses corresponding to catalyst layer micro-structure
and liquid water presence in the pores.
Under low humidity conditions, the model could account
for drying out of ionomer, predicting severe under performance
compared to the isothermal model results. In the case of high
humidity operations, negligible differences are observed between
the isothermal and non-isothermal model predictions. However, at
high current densities, temperatures inside the cell are predicted to
be 10-12 ◦ C higher than the plate temperature; therefore, it is likely
that once a two-phase flow model is included there will be signif-
icant differences between isothermal and non-isothermal model
predictions.
Using the non-isothermal model, the contribution of each heat
source term in the cell was analyzed. Results show that at high cur-
rent densities, irreversible heating inside the CCL, protonic ohmic
heating and heat of sorption account for roughly 60%, 25% and 10%
of the heat generated inside the cell. The heat of sorption is signifi-
cant and, at low and medium current densities, higher than protonic
ohmic heating. The temperature distribution inside the catalyst lay-
ers and membrane is found to be significantly influenced by the
heat of sorption.
The impact of reversible heat distribution in the individual half-
cell reactions of the ORR and HOR is studied. If the majority of
the reversible heat is released in the cathode, reversible heating
is almost counterbalanced by the the heat needed for evaporation
of water produced during the ORR. Significant shifts in the temper-
ature distribution in the cell are observed when the reversible heat
production is assumed to take place in the anode side. Temperature
profiles are observed to be completely reversed when the HOR is
considered exothermic, based on the data by Ramousse et al. [12].
The cell performance dropped further in this case due to increased
protonic ohmic losses.
The non-isothermal model predicted significant temperature
effect on the water transport in both phases, viz., gas and sorbed
forms. A heat pipe effect is observed with water moving out of the
CCL in gas phase due to higher saturation pressures, and condens-
ing in the under-the-rib portion of the CGDL. Due to ionomer dry
out, the non-isothermal model predicts significant higher water
flux from the ACL to CCL in sorbed phase. Thermal osmosis results
in an increase of sorbed water flux by up to 15% and 30%, under low
and high humidity conditions respectively.
The focus of this article has been on developing a detailed
non-isothermal model capable of accounting for all relevant heat
gains/losses in a fuel cell and on providing a detailed breakdown
of all the major losses at different current densities. Further, the
effect of thermal gradients on water transport has also been illus-
trated by the inclusion of a thermal-osmosis term. An accurate
thermal model is a prerequisite for the development of any detailed
two-phase flow model. The results provided regarding temperature
distribution have been obtained only for high temperature and low
relative humidity cases in order to satisfy the single-phase model
assumptions. The presence of liquid water in fully humidified cells
will likely modify the result especially at high current densities.
Future work will include the development of a two-phase flow
model in openFCST.
Acknowledgements
The authors would like to thank the Natural Sciences and Engi-
neering Research Council of Canada (NSERC) and the Automotive
Fuel Cell Cooperation (AFCC) for their financial support.
References
[1] S. Kandlikar, Z. Lu, Thermal management issues in a PEMFC stack - A brief review
of current status, Applied Thermal Engineering 29 (7) (2009) 1276–1280.
[2] A. Weber, J. Newman, Modeling transport in polymer-electrolyte fuel cells,
Chemical Reviews 104 (10) (2004) 4679–4726.
[3] P. Vie, S. Kjelstrup, Thermal conductivities from temperature profiles in the
polymer electrolyte fuel cell, Electrochimica Acta 49 (7) (2004) 1069–1077.
[4] S. He, M. Mench, S. Tadigadapa, Thin film temperature sensor for real-time mea-
surement of electrolyte temperature in a polymer electrolyte fuel cell, Sensors
and Actuators, A: Physical 125 (2) (2006) 170–177.
[5] T.V. Nguyen, R.E. White, Water and heat management model for proton-
exchange-membrane fuel cells, Journal of the Electrochemical Society 140 (8)
(1993) 2178–2186.
[6] T.F. Fuller, J. Newman, Water and thermal management in solid-polymer-
electrolyte fuel cells, Journal of the Electrochemical Society 140 (5) (1993)
1218–1225.
[7] H. Ju, H. Meng, C.-Y. Wang, A single-phase, non-isothermal model for PEM fuel
cells, International Journal of Heat and Mass Transfer 48 (7) (2005) 1303–1315.
[8] J. Ramousse, J. Deseure, O. Lottin, S. Didierjean, D. Maillet, Modelling of heat,
mass and charge transfer in a PEMFC single cell, Journal of Power Sources 145
(2) (2005) 416–427.
[9] V. Ostrovskii, B. Gostev, Heat effects and rates and molecular mechanisms of
water sorption by perfluorinated polymer materials bearing functional groups,
Journal of Thermal Analysis 46 (2) (1996) 397–416.
[10] J. Pharoah, O. Burheim, On the temperature distribution in polymer electrolyte
fuel cells, Journal of Power Sources 195 (16) (2010) 5235–5245.
[11] G. Lewis, M. Randall, International Critical Tables, Vol. 7, McGraw Hill, New
York, 1930.
[12] J. Ramousse, O. Lottin, S. Didierjean, D. Maillet, Heat sources in proton exchange
membrane (PEM) fuel cells, Journal of Power Sources 192 (2) (2009) 435–441.
[13] A. Weber, J. Newman, Coupled thermal and water management in polymer
electrolyte fuel cells, Journal of the Electrochemical Society 153 (12) (2006)
A2205–A2214.
[14] O. Burheim, S. Kjelstrup, J. Pharoah, P. Vie, S. Mller-Holst, Calculation of
reversible electrode heats in the proton exchange membrane fuel cell from
calorimetric measurements, Electrochimica Acta 56 (9) (2011) 3248–3257.
[15] M.J. Lampinen, M. Fomino, Analysis of free energy and entropy changes for
half-cell reactions, Journal of the Electrochemical Society 140 (12) (1993)
3537–3546.
[16] S. Kjelstrup, P. Vie, L. Akyalcin, P. Zefaniya, J. Pharoah, O. Burheim, The Seebeck
coefficient and the Peltier effect in a polymer electrolyte membrane cell with
two hydrogen electrodes, Electrochimica Acta 99 (2013) 166–175.
[17] M. Tasaka, T. Mizuta, O. Sekiguchi, Mass transfer through polymer membranes
due to a temperature gradient, Journal of Membrane Science 54 (1-2) (1990)
191–204.
[18] M. Tasaka, T. Hirai, R. Kiyono, Y. Aki, Solvent transport across cation-exchange
membranes under a temperature difference and under an osmotic pressure
difference, Journal of Membrane Science 71 (1-2) (1992) 151–159.
[19] S. Kim, M. Mench, Investigation of temperature-driven water transport in poly-
mer electrolyte fuel cell: Thermo-osmosis in membranes, Journal of Membrane
Science 328 (1-2) (2009) 113–120.
[20] S. Kim, M. Mench, Investigation of temperature-driven water transport in
polymer electrolyte fuel cell: Phase-change-induced flow, Journal of the Elec-
trochemical Society 156 (3) (2009) B353–B362.
[21] J. Villaluenga, B. Seoane, V. Barragn, C. Ruiz-Bauz, Thermo-osmosis of mixtures
of water and methanol through a Nafion membrane, Journal of Membrane
Science 274 (1-2) (2006) 116–122.
[22] S. Kjelstrup, A. Røsjorde, Local and total entropy production and heat and water
fluxes in a one-dimensional polymer electrolyte fuel cell, Journal of Physical
Chemistry B 109 (18) (2005) 9020–9033.
[23] M. Moore, A. Putz, M. Secanell, Investigation of the ORR using the double-trap
intrinsic kinetic model, Journal of the Electrochemical Society 160 (6) (2013)
F670–F681.
[24] J.X. Wang, T.E. Springer, R.R. Adzic, Dual-Pathway K inetic Equation for the
Hydrogen Oxidation Reaction on Pt Electrodes, Journal of the Electrochemical
Society 153 (9) (2006) A1732–A1740.
[25] http://www.openfcst.mece.ualberta.ca
[26] M. Bhaiya, An open-source two-phase non-isothermal mathematical model of
a polymer electrolyte membrane fuel cell, Master’s thesis, University of Alberta
(2014).
[27] M. Secanell, R. Songprakorp, A. Suleman, N. Djilali, Multi-objective optimization
of a polymer electrolyte fuel cell membrane electrode assembly, Energy and
Environmental Science 1 (3) (2008) 378–388.
[28] T. Springer, T. Zawodzinski, S. Gottesfeld, Polymer electrolyte fuel cell model,
Journal of the Electrochemical Society 138 (8) (1991) 2334–2342.
[29] S. Litster, Djilali, Transport Phenomena in Fuel Cells, 1st Edition, WIT Press,
Ashurst, U.K., 2006, Ch. 5.
[30] A.Z. Weber, M.A. Hickner, Modeling and high-resolution-imaging studies of
water-content profiles in a polymer-electrolyte-fuel-cell membrane-electrode
assembly, Electrochimica Acta 53 (26) (2008) 7668–7674.
[31] R.B. Bird, W.E. Stewart, E.N. Lightfoot, Transport Phenomena, Revised 2nd Edi-
tion, 2nd Edition, John Wiley & Sons, Inc., 2006.
[32] R. Sonntag, C. Borgnakke, G. Van Wylen, Fundamentals of Thermodynamics,
Wiley, 2003.
[33] A.J. Bard, L.R. Faulkner, Electrochemical Methods: Fundamentals and Applica-
tions, 2nd Edition, John Wiley & Sons, Inc., 2001.
[34] C. Popiel, J. Wojtkowiak, Simple formulas for thermophysical properties of liq-
uid water for heat transfer calculations (from 0 ◦ C to 150 ◦ C), Heat Transfer
Engineering 19 (3) (1998) 87–101.
 M. Bhaiya et al. / Electrochimica Acta 147 (2014) 294–309
[35] M. Secanell, Computational modeling and optimization of proton exchange
membrane fuel cells, Ph.D. thesis, University of Victoria (November 2007).
[36] M. Secanell, K. Karan, A. Suleman, N. Djilali, Optimal design of ultralow-
platinum PEMFC anode electrodes, Journal of the Electrochemical Society 155
(2) (2008) B125–B134.
[37] T.A. Davis, Algorithm 832: UMFPACK, an unsymmetric-pattern multifrontal
method, ACM Transactions on Mathematical Software 30 (2) (2004) 196–199.
[38] D. Kelly, J. de, S.R. Gago, O. Zienkiewicz, I. Babuska, A posteriori error analysis
and adaptive processes in the finite element method: Part I - error analysis.,
International Journal for Numerical Methods in Engineering 19 (11) (1983)
1593–1619.
[39] W. Bangerth, R. Hartmann, G. Kanschat, deal.II Differential Equations Anal-
ysis Library, Technical Reference. http://www.dealii.org
[40] W. Bangerth, R. Hartmann, G. Kanschat, deal.II - a General Purpose Object
Oriented F inite Element Library, ACM Trans. Math. Softw. 33 (4) (2007), 24/1-
24/27.
[41] P. Dobson, C. Lei, T. Navessin, M. Secanell, Characterization of the PEM fuel cell
catalyst layer microstructure by nonlinear least-squares parameter estimation,
Journal of the Electrochemical Society 159 (5) (2012) B514–B523.
[42] X. Ren, S. Gottesfeld, Electro-osmotic Drag of Water in Poly(perfluorosulfonic
acid) Membranes, Journal of the Electrochemical Society 148 (1) (2001)
A87–A93.
[43] S. Motupally, A.J. Becker, J.W. Weidner, Diffusion of water in Nafion 115 mem-
branes, Journal of the Electrochemical Society 147 (9) (2000) 3171–3177.
[44] N. Zamel, E. Litovsky, X. Li, J. Kleiman, Measurement of the through-plane ther-
mal conductivity of carbon paper diffusion media for the temperature range
from -50 to +120 ◦ c, International Journal of Hydrogen Energy 36 (19) (2011)
12618–12625.
[45] N. Zamel, E. Litovsky, S. Shakhshir, X. Li, J. Kleiman, Measurement of in-plane
thermal conductivity of carbon paper diffusion media in the temperature range
of -20 ◦ C to +120 ◦ C, Applied Energy 88 (9) (2011) 3042–3050.
[46] M. Secanell, B. Carnes, A. Suleman, N. Djilali, Numerical optimization of pro-
ton exchange membrane fuel cell cathodes, Electrochimica Acta 52 (7) (2007)
2668–2682.
[47] G. Zhang, L. Guo, L. Ma, H. Liu, Simultaneous measurement of current and
temperature distributions in a proton exchange membrane fuel cell, Journal
of Power Sources 195 (11) (2010) 3597–3604.
[48] A. Thomas, G. Maranzana, S. Didierjean, J. Dillet, O. Lottin, Heat fluxes and elec-
trodes temperature in a proton exchange membrane fuel cell, Mechanics &
Industry 13 (04) (2012) 255–260.
[49] Q. Yan, H. Toghiani, J. Wu, Investigation of water transport through membrane
in a PEM fuel cell by water balance experiments, Journal of Power Sources 158
(1) (2006) 316–325.
[50] J.T. Hinatsu, M. Mizuhata, H. Takenaka, Water uptake of perfluorosulfonic acid
membranes from liquid water and water vapor, Journal of the Electrochemical
Society 141 (6) (1994) 1493–1498.
309
[51] P. Teertstra, G. Karimi, X. Li, Measurement of in-plane effective thermal con-
ductivity in PEM fuel cell diffusion media, Electrochimica Acta 56 (3) (2011)
1670–1675.
[52] G. Karimi, X. Li, P. Teertstra, Measurement of through-plane effective thermal
conductivity and contact resistance in PEM fuel cell diffusion media, Elec-
trochimica Acta 55 (5) (2010) 1619–1625.
[53] D. Bevers, M. Whr, K. Yasuda, K. Oguro, Simulation of a polymer electrolyte fuel
cell electrode, Journal of Applied Electrochemistry 27 (11) (1997) 1254–1264.
[54] B. Sivertsen, N. Djilali, CFD-based modelling of proton exchange membrane fuel
cells, Journal of Power Sources 141 (1) (2005) 65–78.
[55] S. Shimpalee, S. Dutta, Numerical prediction of temperature distribution in PEM
fuel cells, Numerical Heat Transfer; Part A: Applications 38 (2) (2000) 111–128.
[56] A. Rowe, X. Li, Mathematical modeling of proton exchange membrane fuel cells,
Journal of Power Sources 102 (1-2) (2001) 82–96.
[57] N. Djilali, D. Lu, Influence of heat transfer on gas and water transport in fuel
cells, International Journal of Thermal Sciences 41 (1) (2002) 29–40.
[58] L. Wang, A. Husar, T. Zhou, H. Liu, A parametric study of PEM fuel cell
performances, International Journal of Hydrogen Energy 28 (11) (2003)
1263–1272.
[59] S. Mazumder, J. Cole, Rigorous 3-D mathematical modeling of PEM fuel cells I.
Model predictions without liquid water transport, Journal of the Electrochem-
ical Society 150 (11) (2003) A1503–A1509.
[60] Y. Wang, C.-Y. Wang, A nonisothermal, two-phase model for polymer elec-
trolyte fuel cells, Journal of the Electrochemical Society 153 (6) (2006)
A1193–A1200.
[61] A. Shah, G.-S. Kim, P. Sui, D. Harvey, Transient non-isothermal model of a poly-
mer electrolyte fuel cell, Journal of Power Sources 163 (2) (2007) 793–806.
[62] F. Jiang, W. Fang, C.-Y. Wang, Non-isothermal cold start of polymer electrolyte
fuel cells, Electrochimica Acta 53 (2) (2007) 610–621.
[63] C. Bapat, S. Thynell, Effect of anisotropic thermal conductivity of the GDL and
current collector rib width on two-phase transport in a PEM fuel cell, Journal
of Power Sources 179 (1) (2008) 240–251.
[64] E. Birgersson, M. Noponen, M. Vynnycky, Analysis of a two-phase non-
isothermal model for a PEFC, Journal of the Electrochemical Society 152 (5)
(2005) A1021–A1034.
[65] L. Matamoros, D. Brüggemann, Simulation of the water and heat management
in proton exchange membrane fuel cells, Journal of Power Sources 161 (1)
(2006) 203–213.
[66] N. Zamel, X. Li, Non-isothermal multi-phase modeling of PEM fuel cell cathode,
International Journal of Energy Research 34 (7) (2010) 568–584.
[67] T. Berning, D. Lu, N. Djilali, Three-dimensional computational analysis of trans-
port phenomena in a PEM fuel cell, Journal of Power Sources 106 (1-2) (2002)
284–294.
[68] M. Khandelwal, M. Mench, Direct measurement of through-plane thermal con-
ductivity and contact resistance in fuel cell materials, Journal of Power Sources
161 (2) (2006) 1106–1115.