See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/245105112

Three-dimensional computational analysis of transport phenomena in a PEM

fuel cell

Article in Journal of Power Sources · April 2002

DOI: 10.1016/S0378-7753(01)01057-6

CITATIONS READS

648 1,067

3 authors, including:

Ned Djilali
University of Victoria
314 PUBLICATIONS 13,334 CITATIONS

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Wall Flows friction turbulence and roughness effects View project

PEM fuel cell modeling View project

All content following this page was uploaded by Ned Djilali on 29 September 2017.

The user has requested enhancement of the downloaded file.

 Journal of Power Sources 106 (2002) 284±294

Three-dimensional computational analysis of transport

phenomena in a PEM fuel cell
T. Berninga, D.M. Lub, N. Djilalia,*
a
Institute for Integrated Energy Systems, University of Victoria, Victoria, Canada
b
Trojan Technologies, London, Ont., Canada

Abstract

A comprehensive non-isothermal, three-dimensional computational model of a polymer electrolyte membrane (PEM) fuel cell has been
developed. The model incorporates a complete cell with both the membrane-electrode-assembly (MEA) and the gas distribution ¯ow
channels. With the exception of phase change, the model accounts for all major transport phenomena.
The model is implemented into a computational ¯uid dynamics code, and simulations are presented with an emphasis on the physical insight
and fundamental understanding afforded by the detailed three-dimensional distributions of reactant concentrations, current densities,
temperature and water ¯uxes. The results show that signi®cant temperature gradients exist within the cell, with temperature differences of
several degrees K within the MEA. The three-dimensional nature of the transport is particularly pronounced under the collector plates land area
and has a major impact on the current distribution and predicted limiting current density. # 2002 Elsevier Science B.V. All rights reserved.

Keywords: PEM fuel cells; PEFC; Fuel cell modelling; CFD

1. Introduction The hydrogen rich fuel fed to the anode side diffuses
through the porous gas-diffusion electrode (GDE). At the
Fuel cells convert the chemical energy of hydrogen and catalyst layer, the hydrogen splits into hydrogen protons and
oxygen directly into electricity. Their high ef®ciency and electrons according to:
low emissions have made them a prime candidate for
2H2 ) 4H‡ ‡ 4e (1)
powering the next generation of electric vehicles, and their
modular design and the prospects of micro-scaling them Driven by an electric ®eld, the H‡ ions migrate through the
have gained the attention of cellular phone and laptop polymer electrolyte membrane. The oxygen in the cathode
manufacturers. Their scalability and relative ¯exibility in gas stream diffuses through the towards the catalyst interface
terms of fuel makes them prime candidates for a variety of where it combines with the hydrogen protons and the
stationary applications including distributed residential electrons to form water according to:
power generation. The basic structure and operation prin-
O2 ‡ 4H‡ ‡ 4e ) 2H2 O (2)
ciple of the polymer electrolyte membrane (PEM) fuel cell
considered here are illustrated in Fig. 1. The overall reaction is exothermic and can be written as:
The polymer electrolyte consists of a per¯uorinated poly-
2H2 ‡ O2 ) 2H2 O ‡ electricity ‡ heat (3)
mer backbone with sulfonic acid side chains. When fully
humidi®ed, this material becomes an excellent protonic Several coupled ¯uid ¯ow, heat and mass transport processes
conductor. The membrane, the catalyst (platinum supported occur in a fuel cell in conjunction with the electrochemical
on carbon particle) and the two electrodes (te¯onated porous reaction. These processes have a signi®cant impact on two
carbon paper or cloth) are assembled into a sandwich important operational issues: (i) thermal and water manage-
structure to form a membrane-electrode-assembly (MEA). ment; (ii) mass transport limitations. Water management
The MEA is placed between two graphite bipolar plates with ensures that the PEM remains fully hydrated to maintain
machined groves that provide ¯ow channels for distributing good ionic conductivity and performance. Water content of
the fuel (hydrogen) and oxidant (oxygen from air). the membrane is determined by the balance between water
production and three water transport processes: electro-
*
Corresponding author. Tel.: ‡1-250-7216034; fax: ‡1-250-7216051. osmotic drag of water, associated with proton migration
E-mail address: ndjilali@uvic.ca (N. Djilali). through the membrane from the anode to the cathode side;

0378-7753/02/$ ± see front matter # 2002 Elsevier Science B.V. All rights reserved.
PII: S 0 3 7 8 - 7 7 5 3 ( 0 1 ) 0 1 0 5 7 - 6
 T. Berning et al. / Journal of Power Sources 106 (2002) 284±294 285

be observed to apply the methods of computational ¯uid

dynamics to fuel cell modelling. Gurau et al. [9] published a
fully two-dimensional model of a whole fuel cell, i.e. two
gas-¯ow channels separated by the MEA. Um et al. [18] and
Wang et al. [19] have developed a similar model and
included two phase ¯ow. However, the underlying assump-
tion was isothermal behaviour, which is a serious modelling
limitation as we will see later. The local temperature dis-
tribution has a signi®cant impact on the amount of water that
undergoes phase change, and therefore the isothermal
assumption can lead to results that are not physically
Fig. 1. Schematic of a PEM fuel cell.
representative when phase change is accounted for. Finally
as a result of the architecture of a cell, the transport
phenomena in a fuel cell are inherently three-dimensional,
back diffusion from the cathode; and diffusion of water to/ but no models have yet been published to address this.
from the oxidant/fuel gas streams. Without control, an In this paper we address simultaneously the need to
imbalance between production and removal rates of water account for thermal gradients and multi-dimensional trans-
can occur. This results in either dehydration of the mem- port using a computational ¯uid dynamics based approach
brane, or ¯ooding of the electrodes; both phenomena have a that couples convective transport in the gas-¯ow channels
very detrimental effect on performance and fuel cells have to with transport and electrochemistry in the MEA.
be carefully designed to avoid the occurrence of either
phenomenon. Thermal management is required to remove
the heat produced by the electrochemical reaction (up to 2. Model description
50% of the energy produced during high power density
operation) in order to prevent drying out of the membrane The PEM fuel cell model presented here is a comprehen-
and excessive thermal stresses that may result in rupture of sive three-dimensional, non-isothermal, steady-state model
the membrane. The small temperature differentials between providing a detailed description of the following transport
the fuel cell stack and the operating environment make phenomena:
thermal management a challenging problem in PEMFCs.
Because of the highly reactive environment of a fuel cell it  multi-component flow;
is not possible to perform detailed in situ measurements  convective heat and mass transport in the flow channels;
during operation. Such information has been sought through  diffusion of reactants through porous electrodes;
modelling and simulation in order to improve understanding  electrochemical reactions;
of water and species transport, optimize thermal manage-  migration of H‡ protons through the membrane;
ment and shorten the design and optimization cycles. Mod-  transport of water through the membrane;
elling of fuel cells is challenging, because the processes  transport of electrons through solid matrix;
involve multi-component, multi-phase, and multi-dimen-  conjugate heat transfer.
sional ¯ow, heat and mass transfer with electrochemical
reactions, all occurring in irregular geometries including The equations governing these processes include the full
porous media. Numerous authors have developed fuel cell mass and momentum conservation equations (Navier±
models accounting for various physical processes. The most Stokes equations) governing ¯uid ¯ow, the species conser-
prominent earlier works stem from Bernardi and Verbrugge vation and energy equations and four additional phenom-
[3,4] and Springer et. al. [14], who developed one-dimen- enological equations tailored to account for processes
sional, isothermal models of the MEA. Fuller and Newman speci®c to fuel cells [3]:
[8] published a quasi two-dimensional model based on
concentration theory. The work by Nguyen and White  the Stefan±Maxwell equations for multi-species diffu-
[12], and Yi and Nguyen [21] was two-dimensional in sion;
nature, but the GDEs were omitted, assuming ``ultrathin''  the Nernst±Planck equation for the transport of protons
electrodes. The importance of accounting for temperature through the membrane;
gradients in fuel cells modelling was demonstrated in the  the Butler±Volmer equation for electrochemical kinetics
work of Woehr et. al. [20] and Djilali and Lu [7]. The and;
important issue of water ¯ooding was addressed by Baschuk  the SchloÈgl equation for the transport of liquid water
and Li in a recent one-dimensional model [2]. through the membrane.
Earlier models were primarily analytic and required a
number of simpli®cations due to the limitations of the These equations and appropriate boundary conditions
numerical techniques. More recently, a general trend can were implemented in their three-dimensional form into
286 T. Berning et al. / Journal of Power Sources 106 (2002) 284±294

the commercial computational ¯uid dynamics code CFX-4.3 2.2. Modelling domain and geometry
(AEA Technology); the implementation required the devel-
opment of an extensive suite of user subroutines. Custo- The computational domain that was employed for the
mized iterative procedures were also implemented to ensure simulations is shown in Fig. 2. In addition to transport across
effective coupling between the electrochemistry and the the MEA (y-direction), the formulation allows us to account
¯uid transport processes. for and investigate the effect of non-uniform transfer rates
and species concentration along the ¯ow channels (x-direc-
2.1. Model assumptions tion), as well as the three-dimensional effects in the trans-
verse z-direction due to the geometry (alternating open ¯ow
A complete fuel cell is an extremely complex system channels with land areas). These effects are expected to be
involving both microscale and macroscale geometric fea- particularly important in the regions of the GDEs under the
tures and transport processes. In order to devise a numeri- collector plates and not directly exposed to the ¯ow ®elds.
cally tractable three-dimensional model of a complete cell it We take advantage of the geometric periodicity of the cell in
is necessary to invoke a number of simplifying assumptions. order to reduce the size of the computational domain and
The most important ones are: hence computational cost. Symmetry is therefore assumed in
the middle planes of the ¯ow channel and land areas and
(1) the fuel cell operates under steady-state conditions;
hence only half of each needs to be incorporated in the
(2) all gases are assumed to be compressible ideal gases,
domain. This is a valid assumption as long as no cross-¯ow
fully saturated with water vapour;
takes place between adjacent channels, as in inter-digitated
(3) the flow in the channels is considered laminar;
designs, and as long as the region considered consists of
(4) the membrane is assumed to be fully humidified so
parallel straight channels, as is the case for the bulk of the
that the electronic conductivity is constant and no
collector plates of an actual fuel cell.
diffusive terms have to be considered for the liquid
In order to effectively implement the numerical solution
water flux;
of the various transport equations, three subdomains were
(5) since it was determined in an earlier study [4] that
de®ned within the main computational domain.
cross-over of reactant gases can be neglected, the
membrane is currently considered impermeable for
 The main domain accounts for the flow, heat and mass
the gas phase;
transfer of the reactant gases inside the flow channels and
(6) the water in the pores of the diffusion layer is
the GDEs.
considered separate from the gases in the diffusion
 Subdomain I consists of the MEA only, and accounts for
layers, i.e. no interaction between the gases and the
the heat flux through the solid matrix of the GDEs and the
liquid water exists;
membrane. Hence, the only variable of interest here is the
(7) the product water is assumed to be in the liquid phase;
temperature. Exchange terms between this subdomain
(8) Ohmic heating in the collector plates and in the GDEs
and Domain I account for the heat transfer between the
is neglected due to their high conductivity;
solid phase and the gas phase.
(9) heat transfer inside the membrane is accomplished by
 Subdomain II is used to solve for the flux of liquid water
conduction only, i.e. the enthalpy carried by the net
through the MEA. The flux of the water in the membrane
movement of liquid water is currently neglected;
is coupled to the electrical potential calculated in Domain
(10) the catalyst layer is assumed to be a thin interface
III via the SchloÈgl equation.
where sink and source terms for the reactants and
 Subdomain III consists of the membrane only and is used
enthalpy are specified;
to calculate the electrical potential inside the membrane.
(11) electroneutrality prevails inside the membrane. The
proton concentration in the ionomer is assumed to be
constant and equal to the concentration of the fixed
sulfonic acid groups.
Most of the above assumptions are a ``standard'' feature
of almost all previous modelling studies. In addition, com-
monly used assumptions in previous studies were: (i) one or
two-dimensionality; (ii) de-coupling of MEA and ¯ow
channel transport; (iii) isothermal conditions. In the present
model these limiting assumptions are removed. One of the
other major limitations, namely the assumption of a single
phase and of non-interacting water and gas phases in the
pores of the GDEs will be addressed in future work with the
implementation of a two-phase model with both phases
interacting in the same computational domain. Fig. 2. Three-dimensional computational domain.
 T. Berning et al. / Journal of Power Sources 106 (2002) 284±294 287

The total enthalpy H is determined from the static (ther-

modynamic) enthalpy h via:
Hg ˆ hg ‡ 12 u2g (7)
where the bulk enthalpy is related to the mass fraction y
and the enthalpy of each gas by:
X
hg ˆ ygi hgi (8)
The mass fraction of the different species obeys a transport
equation of the same form as the generic advection-diffusion
equation:
r
rg ug ygi † r
rg Dgi rygi † ˆ Sgi (9)
Fig. 3. Computational mesh for the MEA and gas-flow channel regions.
The ``empty'' blocks on the left corresponds to the bipolar plates, which
where the i represents oxygen at the cathode side and
have been left out for clarity.
hydrogen at the anode side and Dgi is the diffusivity of
the species in the background ¯uid. The source term Sgi is
The main domain and three subdomains are all coupled
determined by solving the Stefan±Maxwell equations,
through appropriate boundary exchange terms and an itera-
which account for the diffusion of multiple species [15]:
tive solution procedure. The computational mesh is shown in X xgi xgj
Fig. 3. Due to the large number of coupled equations that are r
xgi ˆ vi vj † (10)
solved and the overall complexity of the problem, the Dij
number of computational cells was limited to roughly where vi is the diffusion velocity vector of species i, x the
80,000. The calculations presented here have all been molar fraction and Dij the binary diffusivity of any two
obtained on a PC with a 450 MHz Pentium II. The number species.
of iterations required to obtain converged solutions ranged The second species on both sides is water vapour, which is
from 2000 for lower current densities to 20,000 for higher assumed to exist at the saturation pressure, so that the molar
current densities; the latter required about 50 h of CPU time. fraction of water vapour is given by

2.3. Model equations psat

w T†
xgw ˆ (11)
pg
2.3.1. Notation The saturation pressure of water vapour has been approxi-
In the following, the subscript ``g'' denotes properties of mated by [14]
the gas phase, whereas ``l'' stands for the liquid phase and
``s'' for the solid phase. Different species are denoted by the log10 psat 2
w T† ˆ 2:1794‡0:02953T 9:1837T ‡1:4454T
3

subscript ``i'', i.e. the subscript ``gi'' denotes the species ``i'' (12)
in the gas phase. ``w'' is used for water (species), ``sat''
means saturation value. Cathode side and anode side proper- where T is the temperature in K. The mass fraction is related
ties are denoted by the subscripts ``c'' and ``a'', respectively. to the molar fraction by
xgi Mi
ygi ˆ P (13)
2.3.2. Fuel cell channels xgi Mi
In the fuel cell channels, the gas-¯ow ®eld is obtained by
solving the steady-state Navier±Stokes equations, i.e. the where M is the molecular weight of the different species.
continuity equation: The ideal gas assumption leads to
r
rg ug † ˆ 0 (4) pg Mi
rgi ˆ (14)
RT
and momentum equation
and the bulk density becomes:
r
rg ug ug mg rug † ˆ r
p ‡ 23 mg r
ug †
1 X ygi
‡ r
‰mg rug †T Š (5) ˆ (15)
rg rgi
The temperature ®eld is obtained by solving the convective
energy equation The sum of all mass fractions is equal to unity
X
r
rg ug Hg lg rTg † ˆ 0 (6) ygi ˆ 1 (16)
here rg is the gas phase density, u ˆ u; v; w† the ¯uid which determines the mass fraction of the third species on
velocity vector, p the pressure, T the temperature, m is the both sides (nitrogen at the cathode and carbon dioxide at the
molecular viscosity and l is the thermal conductivity. anode).
288 T. Berning et al. / Journal of Power Sources 106 (2002) 284±294

2.3.3. Gas diffusion electrodes whereas the sink term for hydrogen is speci®ed as
In the porous GDEs the Navier±Stokes equations have to
MH2
be corrected for the porosity e of the carbon ®ber paper. SH 2 ˆ ia (26)
Thus, the conservation equation for mass becomes 2F
where F is the Faraday constant (96487 C/mol) and i is the
r
rg eg ug † ˆ 0 (17) local current density. Again, M is the molecular weight of
whereas the momentum equation reduces to Darcy's law: the species i. The product water is assumed to be in liquid
form, and hence the source term can be written as
kp
eg ug ˆ rpg (18) MH2 O
mg SH2 O l† ˆ ic (27)
2F
where kp denotes the hydraulic permeability. It was men- The generation of heat in the cell is due to entropy
tioned earlier that the liquid water pores are de-coupled from changes as well as irreversibilities due to the activation
the gas pores, and Darcy's law is again used for liquid water overpotential Zact [11]:
transport:  
T DSc †
kp q_ c ˆ ‡ Zact ic (28)
el ul ˆ rpl (19) ne F
ml
where T is the temperature, DSc is the entropy change in the
The mass transport equation in porous media becomes chemical reactions, ne is the number of electrons trans-
r
rg eg ug ygi † r
rg Dgi eg rygi † ˆ Sgi (20) ferred and Zact is the activation overpotential. Because both
these contributions are small at the anode side, the source
and the Stefan±Maxwell equations remain the same: term is currently neglected here. The overpotential Zact can
X xgi xgj be estimated a priori. This does not introduce a large error,
r
xgi ˆ vi vj † (21) since the range of the activation overpotential at the cathode
Deff
ij side is well known, depending on the catalyst loading and the
where the binary diffusivities Deff
ij have been corrected for
expected exchange current density.
the porosity. This was done by applying the so-called As can be seen from Eqs. (25)±(28), for an accurate
Bruggemann correction: solution of the reactant gas distribution inside the fuel cell,
it is crucial to obtain the local current density distribution i,
Deff 1:5
ij ˆ Dij  eg (22) which is described by the Butler±Volmer equation [1]:
!    
Finally, the energy equation in the diffusion layer is given c aa F ac F
ref O
by ic ˆ i0 2
exp Z exp Z (29)
cref
O2 RT act RT act
r
rg eg ug Hg lg rTg † ˆ eg b Tg Ts † (23)
and
where the term on the right hand side accounts for the heat !1=2     
transfer from the solid matrix to the gas phase. b is a cH2 aa F ac F
ia ˆ iref exp Zact exp Z
modi®ed heat transfer coef®cient that accounts for the 0
cref
H2 RT RT act
convective heat transfer in W/m2 and the speci®c surface
(30)
area m2 /m3 of the porous medium. Hence, the unit of b is
[W/m3 ]. where c denotes the concentration of the reactants, and aa
In the solid matrix of the gas-diffusion layer, heat transfer and ac are the so-called transfer coef®cients. The reference
is calculated via: exchange current density iref
0 depends on various parameters
such as operating temperature and catalyst loading, and a
r
ls rTs † ˆ eg b Ts Tg † (24)
number of experiments have been conducted to quantify this
dependence empirically [13,16].
2.3.4. Catalyst layer
As mentioned before, the catalyst layer is treated as a thin 2.3.5. Membrane model
interface, where sink and source terms for the reactants are In the membrane, the model accounts for the ¯ux of liquid
implemented. Due to the in®nitesimal thickness, the source water and the protonic ¯ux together with the distribution of
terms are actually implemented in the last grid cell of the electrical potential F are being considered. The transport of
porous medium. At the cathode side, the sink term for liquid water through the membrane is governed by a mod-
oxygen is given by i®ed version of Darcy's law, the SchloÈgl equation [3]:
MO2 kf kp
SO2 ˆ ic (25) ul ˆ zf cf F
rF
rp (31)
4F ml ml
 T. Berning et al. / Journal of Power Sources 106 (2002) 284±294 289

where kf denotes the hydraulic permeability, zf is the ®xed- change in entropy. The standard reversible potential E0 is
charge number in the membrane, cf is the ®xed-charge given by [5]
concentration and ml is the liquid water viscosity. This
equation accounts for two different water transport pro- DG0
E0 ˆ (38)
cesses: the electro-osmotic drag whereby hydrogen protons nF
migrating through the membrane drag water molecules with where DG0 is the change in the standard Gibbs free energy of
them, and pressure driven ¯ux, which is usually directed the reaction. Using standard values for the entropy produc-
from the cathode side to the anode side. Strictly speaking, a tion, Eq. (37) yields [2]
diffusive term has to be accounted for as well, since the back
0
diffusion of water can play an important role for water ET;p ˆ 1:229 0:83  10 3 T 298:15†‡
 
management schemes. However, when the membrane is 5 1
fully humidi®ed, as is assumed here, this term drops.  4:31  10 T ln pH2 ‡ ln pO2 (39)
2
Heat transfer in the membrane is governed by
i2 2.4. Boundary conditions
rlmem
rT ˆ (32)
k
where k is the electric resistance of the membrane material. Boundary conditions have to be applied for all variables of
Note that this implies that the transport of energy by the interest in each computational domain. In order to reduce
liquid water is neglected and the membrane is considered as computational cost, we take advantage of the geometric
a heat-conducting solid. The term on the right hand side of periodicity of the cell. Hence symmetry is assumed in the z-
the equation denotes Ohmic heating inside the membrane. direction, i.e. all gradients in the z-direction are set to zero at
The local current density inside the membrane is obtained the x±y plane boundaries of the domain. With the exception
from Ohm's law: the channel inlets and outlets, a zero ¯ux condition is applied
at all x-boundaries (y±z planes).
iˆ krF (33) The inlet values at the anode and cathode are prescribed
Finally, for the electrical potential in the membrane, it can be for the velocity, temperature and species concentrations
shown that for a fully hydrated membrane it is governed by (Dirichlet boundary conditions). The inlet velocity is a
the Laplace equation [4]: function of the desired average current density iave, the
geometrical area of the membrane AMEA , the channel
r2 F ˆ 0 (34) cross-section area Ach , and the stoichiometric ¯ow ratio z,
according to:
2.3.6. Bipolar plates iave 1 RTin;c 1
Bipolar plates serve to transfer the electrons and separate uin;c ˆ zc AMEA (40)
4F xO2 ;in pc;in Ach
different cells. Since the electrical conductivity of graphite
lgr is high, Ohmic losses are neglected, and the energy and
equation reduces to iave 1 RTin;a 1
uin;a ˆ za AMEA (41)
rlgr
rT ˆ 0 (35) 2F xH2 ;in pa;in Ach
where R is the universal gas constant, Tin is the inlet
2.3.7. Cell potential temperature and pin is the inlet pressure.
The cell potential E is obtained by substracting all over- At the outlets of the gas-¯ow channels, only the pressure
potentials (losses) from the equilibrium potential, i.e. is being prescribed as the desired electrode pressure; for all
other variables, the gradient in the ¯ow direction (x) is
0
E ˆ ET;p Zact ZO Zmem (36) assumed to be zero (Neumann boundary conditions).
0
Since the ¯uid channels are bordered by the collector
where ET;p is the equilibrium potential for a given tempera- plates, no boundary conditions have to be prescribed here
ture and pressure, Zact the activation overpotential at both and conjugate heat transfer, impermeability and no-slip
sides, ZO are the Ohmic losses in the GDE plus contact conditions are implemented implicitly at solid±¯uid inter-
resistances and Zmem is the Ohmic loss in the membrane. faces within the domain. At the outer boundaries of the
0
The equilibrium potential ET;p is a function of pressure bipolar plates (y-direction), boundary conditions need only
and temperature and is determined using the Nernst law [5], to be given for the energy equation. This can be done by
  prescribing the heat ¯ux or the temperature distribution.
0 0 DS 0 RT 1
ET;p ˆ E ‡ T T †‡ ln pH2 ‡ ln pO2 (37) In the present simulations a zero heat ¯ux condition was
nF nF 2
used, @T=@y ˆ 0. This is not an entirely physical condition,
where p denotes the partial pressure of the species, n is the but is adequate for the present simulations in which the focus
number of electrons transferred in the reaction, and DS is the is on model validation and identi®cation of key transport
290 T. Berning et al. / Journal of Power Sources 106 (2002) 284±294

processes. In future work we intend to implement a cooling At the interfaces exposed to the land area, no-¯ux conditions
¯ow channel in the computational domain to remove arbi- are imposed.
trary speci®cation of the heat ¯ux or temperature.
2.4.3. Subdomain III
2.4.1. Subdomain I Finally, for the membrane domain, where the only vari-
As discussed earlier, three subdomains are de®ned for able of interest is the electric potential, a reference potential
numerical ef®ciency. Subdomain I is used to solve for the value of zero is arbitrarily set at the anode side
heat ¯ux through the solid matrix of the porous GDE. The only
variable of interest here is the temperature, and it is assumed Fˆ0 (45)
that all the heat transfer takes place to and from the gas phase
and at the cathode side, the potential distribution at the
inside the porous medium only. Adiabatic boundary condi-
membrane/catalyst interface is computed from [4]
tions are therefore applied at all boundaries of this domain
@T @F 1
ˆ0 (42) ˆ ‰i Fcf vel;mem Š (46)
@n @y keff

2.4.2. Subdomain II 2.5. Modelling parameters

For the liquid water transport through the MEA in Sub-
domain II, the pressure is given at the outer boundaries of the The parameters used for the base case simulations pre-
GDE, i.e. the channel/GDE interface sented here are shown in Table 1. Since the model presented
here was initially developed by extending the formulation of
Pa;l ˆ 3 atm (43)
Bernardi and Verbrugge [3,4] to three-dimensions, many of
and the key parameters de®ned by these authors are still used
here. It is important to note that because this model accounts
Pc;l ˆ 5 atm (44)
for all major transport processes and the modelling domain
comprises all the elements of a complete cell, no parameters
Table 1 needed to be adjusted in order to obtain physical results.
Geometrical, operational, electrode and membrane parameters for the base When comparing our results with experiments published in
case
the literature, however, many of the experimental data such
Property Value as the stoichiometric ¯ow ratio or the exact cell geometry
Channel length, l 0.05 m
and dimensions are unknown, which makes a quantitative
Channel height, h 1.010 3 m comparison dif®cult. The strength of the current model is
Channel width, wc 1.010 3 m clearly to perform parametric studies and explore the impact
Land area width, wl 1.010 3 m of various parameters on the transport mechanisms and on
Electrode thickness, te 0.2610 3 m fuel cell performance.
Membrane thickness, tmem 0.2310 3 m
Inlet fuel and air temperature, Tin 80  C
For the binary diffusivities Dij required in the Stefan±
Air side pressure, pc 5 atm Maxwell equations, experimentally obtained values at atmo-
Fuel side pressure, pa 3 atm spheric pressure p0 were taken and scaled with the tem-
Air stoichiometric flow ratio, zc 3 perature and pressure according to [6]
Fuel stoichiometric flow ratio, za 3
 
Relative humidity of inlet gases, x 100% p T 1:5
Oxygen/nitrogen ratio, c 0.79/0.21 Dij ˆ Dij T0 ; p0 † (47)
Gas phase electrode porosity, eg 0.4 [4]
p0 T0
Electronic conductivity, s 6000 S/m Table 2 lists the reference binary diffusivities.
Effective thermal conductivity, leff 75 W/mK [9]
Heat transfer coefficient between solid 7.0105 W/m3
and gas phase, b
Transfer coefficient, anode side, aa 0.5
Transfer coefficient, cathode side, ac 1 [17] Table 2
Ref. exchange current density, anode, iref 0.6 A/cm2 Binary diffusivities at reference temperatures
0;a
Ref. exchange current density, cathode, iref
0;c 4.410 7 A/cm2 Gas-pair Reference Binary diffusivity,
Entropy change of cathode side reaction, DSPt 326.36 J/(mol K) [11] temperature, T0 (K) Dij (cm2 /s)
Ionic conductivity of the membrane, k 0.06 S/cm
Protonic diffusion coefficient, DH‡ 4.510 9 m2 /s2 [4] DH2 H2 O 307.1 0.915
Fixed-charge concentration, cf 1200 mole/m3 [4] DH2 CO2 298.0 0.646
Fixed-site charge, zf 1 DH2 O CO2 307.5 0.202
Electrokinetic permeability, kf 2.010 20 m2 [4] DO2 H2 O 308.1 0.282
Hydraulic permeability, kp 1.810 18 m2 [4] DO2 N2 293.2 0.220
Thermal conductivity of the membrane, l 0.67 W/mK [10] DH2 O N2 307.5 0.256
 T. Berning et al. / Journal of Power Sources 106 (2002) 284±294 291

3. Results and discussion

3.1. Fuel cell performance

The fuel cell performance is shown in terms of the

polarization in Fig. 4. The results agree well with the
experiments [17] in the low and intermediate current density
region. The discrepancy at high current densities is attrib-
uted to the fact that experimental data at the higher current
densities was insuf®cient, and in fact the experimental curve
in this region is a curve ®t weighted by the lower current
density data. It should also be pointed out that the exact
geometry of the fuel cell used in the experiments is not
known. The ability of the present model to reproduce the
polarization curve is a necessary validation check but by no
means especially informative since one can always obtain
good agreement between experimental results and any
model that somehow captures the logarithmic drop-off in
the low current density region and the Ohmic losses inside
the membrane. The strength of the numerical approach is in
providing detailed insight into the various transport mechan-
isms and their interaction, and in the possibility of perform- Fig. 5. Three-dimensional distribution of reactant gases in the gas-flow
ing parameters sensitivity analyses.

3.2. Reactant gas and temperature distribution The effect of the ``land area'' is even more pronounced at
higher current densities, as shown in Fig. 6. Furthermore, for
One of the major advantages of using a model as detailed a higher current density at otherwise similar conditions, the
as the one presented here is the detailed distribution of the oxygen molar fraction at the catalyst layer decreases due to
reactant gases inside the fuel cell. Such distributions, which diffusion limitations. Note that the stoichiometric ¯ow ratio
cannot be measured in situ, provide valuable information is the same for both cases, i.e. three times the amount of
about the onset of concentration losses and their effect on the oxygen consumed enters the cell. Since the composition of
limiting current density. Fig. 5 shows the reactant gas
distribution in the fuel cell channels and the gas-diffusion
layers. The molar fraction of oxygen decreases noticeable
inside the gas-diffusion layer, and the effect of oxygen
depletion is signi®cant, particularly under the land areas.
A realistic prediction of the limiting current density cannot
be obtained from a two-dimensional model in which in®-
nitely wide channels are assumed and the area that is not
exposed to the gas-¯ow channels is not accounted for.

Fig. 4. Comparison of the experimental and simulated polarization curves. Fig. 6. Molar oxygen fraction at the catalyst layer for two different current
Also shown is the power density curve obtained with the model. densities: 0.2 A/cm2 (upper) and 1.0 A/cm2 (lower).
292 T. Berning et al. / Journal of Power Sources 106 (2002) 284±294

Fig. 8. Fraction of the total current generated under the area exposed to the
gas-flow channels.

gas-¯ow channels increases steadily, as shown in Fig. 8. The

fraction of the current generated under the channel area
increases almost linearly with the current density, the max-
imum being just under 80% at a current density close to the
limiting current density. This is important, since it provides
an indication of how well the catalyst is being used in these
areas. For optimal fuel cell performance, a uniform current
density generation is desirable, and this could only be
achieved with a non-uniform catalyst distribution, possibly
in conjunction with non-homogeneous GDEs.
Fig. 7. Dimensionless current density distribution at the catalyst layer One of the most important features distinguishing the
interface for three different nominal current densities: 0.2 A/cm2 (upper);
present model is the fact that it is non-isothermal. The
0.8 A/cm2 (middle); 1.4 A/cm2 (lower). The dashed lines represent the
boundaries between the flow channel and the land area. temperature distribution inside the fuel cell has important
effects on nearly all transport phenomena, and knowledge of
the magnitude of temperature increases due to irreversibil-
the inlet gases is the same, this is accomplished by an ities might help preventing failure. Fig. 9 shows that the
increase in velocity for a higher current density. At an increase in temperature can exceed several degrees Kelvin
average current density of 1.4 A/cm2 , the molar oxygen near the inlet area, where the local current density is highest.
fraction at the catalyst layer is almost zero throughout the Due to the low electric conductivity of the polymer electro-
interface, indicating that the limiting current density has lyte, the temperature maximum occurs inside the membrane.
been reached. In general, the temperature at the cathode side is slightly
The detailed oxygen distribution at the catalyst layer is of higher than at the anode side; this is due to the reversible and
importance, because it determines the local current density, irreversible entropy production. The detailed temperature
according to Eq. (29). Assuming a constant activation over- distribution is also key for the eventual extension of the
potential throughout the catalyst interface, this results in local current model to include multiphase phenomena. Phase
current densities as shown in Fig. 7. The local current change is not yet accounted for, but the mechanism shall
densities here have been normalized by the nominal current be brie¯y outlined here. In the ®rst place, the temperature
density in each case. It can be seen that for a low nominal rise at the cathode determines, how strong the under-satura-
current density, the distribution is quite uniform with varia- tion of the gas phase in this area is. This in turn gives rise to
tions from 75 to 125% relative to the nominal current density the evaporation of liquid product water, which provides
of 0.2 A/cm2 . This changes for intermediate current densities, cooling and thus offsets the temperature rise. Hence, any
where under the land areas a noticeable decrease takes place implementation of phase change in an isothermal model can
and the minimum current fraction drops below 40% of the not lead to physically representative results.
average (nominal) current density of 0.8 A/cm2 . The max- Finally, the water ¯ux and the potential distribution is
imum, however, remains almost the same at just over 130%. presented in Fig. 10. The ¯ux of liquid water is governed by
This pattern changes further at high current densities, where three mechanisms: electro-osmotic drag from the anode to
the maximum local current density can be as high as three the cathode, back diffusion driven by a concentration gra-
times the average current density near the cathode side inlet. dient from the cathode to the anode and, if a pressure
With an increase in current density, the fraction of the gradient exists, convection which is usually directed from
current density generated under the area exposed to the the cathode to the anode in order to counter-balance the
 T. Berning et al. / Journal of Power Sources 106 (2002) 284±294 293

Fig. 9. Three-dimensional temperature distribution inside the fuel cell at a Fig. 11. Net drag coefficient a for the flux of water through the membrane
nominal current density of 1.2 A/cm2 . for two different values of the electrokinetic permeability of the
membrane.

electro-osmotic drag. Assuming a fully humidi®ed mem-

brane implies no concentration gradient exists, and in any strongest, the electro-osmotic drag dominates convection,
case, the effect of the diffusion on the water ¯ux through the and water ¯ows from the anode to the cathode. Near the
membrane has been found to be small compared to convec- outlet, the current is weaker and the net water ¯ux is directed
tion and electro-osmotic drag. As illustrated in Fig. 10, the towards the anode.
pressure gradient might outweigh the effect of the electro- The net ¯ux of water through the membrane is often
osmotic drag for low current densities, and the net ¯ux of characterized by the parameter a, the net amount of water
water is directed from the cathode to the anode. Humidi®ca- molecules dragged through the membrane per hydrogen
tion schemes have to be therefore devised for the cathode proton. Fig. 11 shows the net water ¯ux for two different
side only. At a current density of 0.4 A/cm2 , the ¯ux of water values of the electrokinetic permeability of the membrane. If
is reversed. Near the inlet area, where the local current is the value cited by Bernardi and Verbrugge is used [4], a
becomes unphysically high. A comparison with the model
published by Yi and Nguyen [21] reveals that reducing the
permeability to 2:0  10 20 m2 as was used in the current
base case yields a more realistic values for a. Still, these
results differ substantially from the experimentally deter-
mined values of a, which range from 0:6 at low current
densities to around 0:3 for intermediate current densities in
the absence of a pressure gradient. The membrane model has
therefore to be improved in order to predict the amount of
water that need to be supplied at the electrodes in order to
prevent drying out of the membrane.

4. Conclusions

A comprehensive three-dimensional computational

model of a PEM fuel cell has been developed. With the
exception of phase change, the model accounts for all major
transport phenomena in the ¯ow channels, electrodes and
electrolyte membrane. Results that are physically consistent
and in good agreement with available experimental data are
obtained. The three-dimensional nature of the distribution of
¯ow velocities, species concentration, mass transfer rates,
electric current and temperature was clearly illustrated by
the simulations. The capabilities of the model for providing
detailed insight into water transport mechanisms and the
Fig. 10. Liquid water flux (vectors) and potential distribution (contours)
onset of mass transport limitations was demonstrated, and its
inside the membrane for three different current densities: 0.2 A/cm2 potential for parametric studies of interest in design and
(upper); 0.4 A/cm2 (middle); 0.6 A/cm2 (lower). prototyping were also illustrated.
294 T. Berning et al. / Journal of Power Sources 106 (2002) 284±294

Though this comprehensive model represents to our [9] V. Gurau, H. Liu, S. Kakac, Two-dimensional model for proton
knowledge the only three-dimensional model developed exchange membrane fuel cells, AIChE J. 44 (11) (1998) 2410±2422.
[10] F.P. Incropera, D.P. DeWitt, Fundamentals of Heat and Mass
to date, a number of important mechanisms and phenomena Transfer, 4th Edition, Wiley, New York, 1996.
need to be incorporated to increase its generality and [11] M.J. Lampinen, M. Fomino, Analysis of free energy and entropy
usefulness. Work is currently in progress to account for changes for half-cell reactions, J. Electrochem. Soc.140 (12) (1993)
(a) the effect of phase change of water, and (b) partial 3537±3546.
[12] T.V. Nguyen, R.E. White, A water and heat management model for
dehydration of the membrane during operation. Finally
proton-exchange-membrane fuel cells, J. Electrochem. Soc. 140 (8)
the electrochemistry in the computational model relies (1993) 2178±2186.
on ®rst order kinetics and empirical data. Further work in [13] A. Parthasarathi, S. Srinivasen, J.A. Appleby, C.R. Martin,
this area is required on both theoretical and experimental Temperature dependance of electrode kinetics of oxygen reduction
fronts. at the Platinum/Nafion interfaceÐA microelectronic investigation, J.
Electrochem. Soc. 139 (9) (1992) 2530±2537.
[14] T.E. Springer, T.A. Zawodzinski, S. Gottesfeld, Polymer electrolyte
fuel cell model, J. Electrochem. Soc.138 (8) (1991) 2334±2342.
References [15] R. Taylor, R. Krishna, Multicomponent Mass Transfer, Wiley, New
York, 1993.
[1] A.J. Bard, L.R. Faulkner, Electrochemical Methods, Wiley, New [16] E.A. Ticianelli, J.G. Berry, S. Srinivasan, Dependence of perfor-
York, 1980. mance of solid polymer electrolyte fuedl cells with low platinum
[2] J.J. Baschuk, X. Li, Modelling of polymer electrolyte membrane fuel loading on morphologic characteristics of the electrodes, J. Electro-
cells with variable degrees of water flooding, J. Power Sources 86 anal. Chem. 251 (1988) 275±295.
(2000) 181±195. [17] E.A. Ticianelli, C.R. Derouin, A. Redondo, S. Srinivasan, Methods to
[3] D.M. Bernardi, M.W. Verbrugge, Mathematical model of a gas advance technology of proton exchange membrane fuel cells, J.
diffusion electrode bonded to a polymer electrolyte, AIChE J. 37 (8) Electrochem. Soc.135 (9) (1988) 2209±2214.
(1992) 1151±1163. [18] S. Um, C.Y. Wang, K.S. Chen, Computational fluid dynamics
[4] D.M. Bernardi, M.W. Verbrugge, A mathematical model of the solid- modeling of proton exchange membrane fuel cells, J. Electrochem.
polymer-electrolyte fuel cell, J. Electrochem. Soc. 139 (9) (1992) Soc. 147 (12) (2000) 4485±4493.
2477±2491. [19] Z.H. Wang, C.Y. Wang, K.S. Chen, Two-phase flow and transport in
[5] J. O'M Bockris, S. Srinivasan, Fuel Cells: Their Electrochemistry, the air cathode of proton exchange membrane fuel cells, J. Power
McGraw-Hill, New York, 1969. Sources 94 (2001) 40±50.
[6] E.L. Cussler, Diffusion±Mass Transfer in Fluid Systems, Cambridge [20] M. Woehr, K. Bolwin, W. Schnurnberger, M. Fischer, W. Neubrand,
University Press, Cambridge, 1984. G. Eigenberger, Dynamic modelling and simulation of a polymer
[7] N. Djilali, D.M. Lu, Influence of heat transfer on gas and water membrane fuel cell includinmg mass transport limitations, Int. J.
transport in fuel cells, Int. J. Thermal Sci., in print, 2002. Hydrogen Energy 23 (3) (1998) 213±218.
[8] T.F. Fuller, J. Newman, Water and thermal managament in solid- [21] J.S. Yi, T.V. Nguyen, An along-the-channel model for proton
polymer-electrolyte fuel cells, J. Electrochem. Soc. 140 (5) (1993) exchange membrane fuel cells, J. Electrochem. Soc. 145 (4) (1998)
1218±1225. 1149±1159.

View publication stats