ARTICLE IN PRESS

Energy 32 (2007) 269–280

www.elsevier.com/locate/energy

Keynote Paper

Computational modelling of polymer electrolyte membrane (PEM) fuel

cells: Challenges and opportunities
N. Djilali
University of Victoria, Institute for Integrated Energy Systems & Department of Mechanical, Engineering, Victoria, BC, Canada V8W 3P6
Received 7 November 2005

Abstract

Fuel cells are still undergoing intense development, and the combination of new and optimized materials, improved product
development, novel architectures, more efficient transport processes, and design optimization and integration are expected to lead to
major gains in performance, efficiency, reliability, manufacturability and cost-effectiveness. Computational fuel cell engineering (CFCE)
tools that allow systematic simulation, design and optimization of fuel cell systems would facilitate the integration of such advances,
allow less heavy reliance on hardware prototyping, and reduce development cycles.
CFCE requires the robust integration of models representing a variety of complex multi-physics transport processes characterized by a
broad spectrum of length and time scales. These processes include a fascinating, but not always well understood, array of phenomena
involving fluidic, ionic, electronic and thermal transport in concert with electrochemical reactions. In this paper, we report on some
progress in both fundamental modelling of these phenomena, as well as in the development of integrated, computational fluid dynamics
(CFD) based models for polymer electrolyte membrane (PEM) fuel cells. A new rational model for coupled protonic and water transport
in PEMS, as well direct numerical simulations of two-phase flow in porous gas diffusion electrodes are discussed. Illustrative applications
of CFD-based simulations are presented for conventional fuel cells and novel micro-structured fuel cells. The paper concludes with a
perspective on some of the remaining theoretical, experimental and numerical challenges to achieve truly functional CFCE tools.
r 2006 Elsevier Ltd. All rights reserved.

Keywords: Fuel cell; Transport phenomena; Two phase flow; Modelling; CFD; Polymer electrolyte membranes; Porous electrodes; Computational design;
Micro-fuel cell; Review

1. Introduction as supplied with a fuel. When fuelled by hydrogen, the only

Fuel cells and hydrogen technologies offer a promising
pathway towards sustainable energy systems. They offer
the prospect of low to zero emission power generation for
subWatt to MWatt applications in transportation, heating,
manufacturing, and communication. However, many en-
gineering and scientific obstacles (as well as economic and
socio-political ones) need to be surmounted before wide
spread adoption of these technologies becomes possible.
The potential of fuel cells for clean and efficient energy
conversion have long been recognized. Fuel cells are
adaptable, have a high theoretical efficiency, and, unlike
batteries, they can produce electricity continuously as long
Tel.: +1 250 721 6034; fax: +1 250 721 6323.
E-mail address: ndjilali@uvic.ca.
by-products of a fuel cell are water and heat. In addition to
the important question surrounding the development of a
hydrogen fuelling infrastructure discussed elsewhere [1], the
main obstacles to the deployment and commercialization
of polymer electrolyte membrane fuel cells (PEMFCs) are
cost, durability, and performance [2]. In this paper we will
focus on the role of computational tools in addressing
some of the technical challenges in the development of
better performance and affordable PEMFC technology,
the lead contender for replacing the ubiquitous battery and
internal combustion engine.
The overall operating principle of a PEMFC is relatively
simple. Hydrogen is fed to the cell and oxidized at the
anode, while oxygen, usually carried in an air feed stream,
is reduced at the cathode. The protons produced at the
anode are conducted through the polymer electrolyte

0360-5442/$ - see front matter r 2006 Elsevier Ltd. All rights reserved.
doi:10.1016/j.energy.2006.08.007
 ARTICLE IN PRESS
270 N. Djilali / Energy 32 (2007) 269–280

Nomenclature S_ source term

s saturation
c molar concentration T Temperature
I, i current; current density u velocity
k permeability e porosity
F Faraday’s constant F potential
Np protonic flux m viscosity
Nl water flux r density
P pressure s conductivity; interfacial tension

membrane (PEM) to the cathode. Since the membrane is
not electronically conductive, the electrons released at the
anode have to flow along an external circuit, thereby
generating useful electric power, and eventually recombine
with the protons on the cathode side to produce water.
These reactions and pathways are illustrated schematically
in Fig. 1.
Broadly speaking, improved performance in PEMFC
can be achieved by enhancing the rates of transport and/or
reaction, and by increasing the surface/volume ratio of
interfacial areas where reactions/transport take place. This
can be achieved by, e.g., engineering the transport proper-
ties or processes in specific areas, or by introducing new
- +
V
Membrane
Electrode
Assembly (MEA)
Electrode
materials or designs. A particularly promising avenue is the
use of micro-structured features [3] as will be discussed
later. In any case, it is first necessary to identify the
prevailing limitations and their location, and to understand
their origin. In-situ measurements to obtain the relevant
information are exceedingly difficult due to the confined
and electrochemically active environment of fuel cells, the
complexity of the couplings, and the often transient nature
of the processes. Rational models based on the funda-
mental physico-chemical processes together with experi-
mental observation are thus an essential tool to gain a
more complete understanding of the operation of a fuel
cell. Coupling these models with modern computational
fluid dynamics (CFD) and multi-disciplinary optimization
methods would allow the creation of powerful computa-
tional fuel cell engineering (CFCE) tools that could lead to
dramatic reductions in lead times and development costs,
and spur innovative design.
While much progress has been made in recent years in
CFD based modelling of fuel cells, truly functional and
predictive capabilities remain a challenge due to, amongst
others factors:

Anode Cathode
 lack of sufficiently general models for ionic and water
transport in polymer membranes;
 deficiencies in models for two phase transport in porous
gas diffusion electrodes and gas flow channels;
O2 +4e-
 inadequate macroscopic modelling and resolution of
2H2
catalyst layers;
4e- +4H+ +4H+
2H2O
 limited data for material and transport parameters and
lack of in situ data for validation; and
 existence of a wide range of characteristic length and
time scales.

An excellent overview of the potential gains to be made

by theoretical modelling was recently provided in [4], and
many notable advances in CFD modelling of PEM fuel
Proton Exchange
cells have been made by several groups over the last few
Membrane (PEM)
years [5–9], and a thorough review was provided in 2004 in
Catalyst Layer [10]. Some representative contributions made since that
Gas Diffusion Layer(GDL) review include parallelized three-dimensional CFD model-
ling [11–13]; parametric computational studies in conjunc-
Gas Channel Current Collector tion with experimental validation [14,15]; and develop-
ments addressing the important aspects of two-phase
Fig. 1. Schematic of a PEM fuel cell. transport [16,17] and catalyst layer modelling [18,19].
 ARTICLE IN PRESS
N. Djilali / Energy 32 (2007) 269–280
A review of challenges and modelling issues in both low
and high temperature fuel cells, with particular attention to
heat management, appeared recently [20]. In this paper, we
provide an overview focused primarily on recent progress
made at the University of Victoria in addressing some of
the key transport phenomena issues and in the develop-
ment of CFD based modelling strategies for PEMFCs. The
first part of the paper focuses on fundamental modelling of
specific transport phenomena encountered in component
layers of the membrane electrode assembly (MEA). In the
second part, we discuss the integration of some of these
sub-models into multi-dimensional CFD codes and illus-
trate their application in plate and frame unit cells as well
as in novel micro-structured designs. The paper closes with
a perspective on some of the remaining challenges.
2. Transport phenomena in PEMFCS
The cathode side of the MEA, shown schematically in
Fig. 2, provides a good illustration of the complex interplay
between various transport phenomena. The MEA consists
of a proton exchange membrane sandwiched between
catalyst and gas diffusion layers (GDL), with the latter two
components essentially forming the electrode. This elec-
trode is a buffer zone that facilitates a number of processes:
 The conduction of electrons between the current
collectors and the reaction sites.
 The conduction of protons between the membrane and
catalyst layer.
 The transport of the reactants to the catalyst layer.
Proton Carbon
Electrically
Conducting Supported
Conductive
Media Catalyst
Fibers
e-
HO2
O2
H+
PEM Catalyst GDL
Layer
Fig. 2. Transport of reactants, charged species and products in a PEMFC
cathode.
271
 The transport of heat away from the reaction site
primarily through the solid matrix.
 Condensation/evaporation and transport of liquid water
and vapour.
Optimal tailoring of an effective electrode requires
balancing of often conflicting requirements. On the
cathode side for instance, the reaction is prone to mass
transport limitations at high current densities, when
depletion of oxygen in the air stream becomes significant.
This can be further compounded by flooding of the
electrode pores by the product water and blocking of
pathways for oxygen transport to the reaction sites.
Effective removal of excess water is thus essential, but
must be balanced against the need to prevent drying of the
adjacent proton conducting media (membrane), which not
only causes deterioration in protonic conductivity (and
thus higher ohmic losses and lower efficiency), but also
impacts durability.
Computational design of fuel cells with effective reactant
distribution, optimized structures and proper heat and
water management, requires understanding of the various
transport processes in each of the components, the
development of physically representative models for these
processes, and, finally, proper coupling of the models.
2.1. Polymer electrolyte membranes
Polymer electrolyte membranes (PEM) provide the key
functions of conducting protons from anode to cathode,
and of preventing reactant crossover. These membranes
consist of microphase-separated structures, comprising
hydrophobic polymer chains (PTFE/Teflon) and hydro-
philic sulfonic acid groups. A thorough review of transport
phenomena in PEM was recently provided in [21]. The
protonic conductivity characteristics of a PEM are strongly
dependent on the membrane hydration given in terms of l
(number of sorbed waters per sulfonate head). Excess
protons in bulk water are primarily in the form of
hydronium ions (H3O+), and higher complexes such as
‘‘Eigen’’ ion (H9O4)+ may also participate. A phenomenon
closely linked to membrane conductivity is electro-osmotic
drag (EOD), the process whereby a given number of water
molecules associated with a proton are dragged as the
proton migrates from anode to cathode. This EOD
mechanism plays a major role in determining overall water
balance.
The breadth of PEM transport models proposed to date
ranges from molecular dynamics simulations and statistical
mechanics models, to empirical macroscopic models as
illustrated in Fig. 3. Our focus is on the development of
macroscopic models required for computational simula-
tions of complete fuel cells. The most widely used model to
date for determining membrane conductivity in large scale
fuel cell models is due to Springer, Zawodzinski and
Gottesfeld [22]. This ‘‘diffusion’’ model has limited
predictive abilities as it relies on empirical correlations
 ARTICLE IN PRESS
272 N. Djilali / Energy 32 (2007) 269–280
Fig. 3. Classification of transport models for proton exchange mem-
branes.
obtained for Nafion membranes only [23,24], and in
particular does not perform well under low humidification
conditions [23].
A more fundamental and general approach to macro-
scopic modelling transport in membranes requires micro-
scopic consideration of the dominant interactions between
the aqueous system and the polymer and how they affect
the conductivity and transport properties. A general model
that considers such interactions based on the fundamentals
of multi-component diffusion was recently derived. This
model proceeds from a generalization of the Stefan–Max-
well relations to a free solution in a membrane pore
structure [25]. Specializing this formulation to polymer
electrolyte membranes (specifically perfluorosulfonic acid
membranes) and expressing the electrochemical potential
gradient in terms of composition, activity, Gibb’s free
energy, and electrical potential yields a general expression
that can be simplified through an order of magnitude
analysis [26]. Considering a limiting case scenario with a
pressure drop of order (5  105 N/m2) and a potential drop
of order (0.3 V), shows that the dominant driving terms are
the potential and mole fraction gradients, rFm and rX.
The resulting simplification and expressing X in terms of
membrane water content l yields the new Binary Friction
Membrane Model (BFM2) [26]:
" # " #
fDpp Dpw  rFm 
Np
¼ c . (1)
Nw fDwp Dww rl
The BFM2 provides a general constitutive relation
coupling the protonic flux Np, which is proportional the
ionic current im, to the water flux Nw; c is the molar
concentration of water; f ¼ F/(RT), and Fm and l are the
membrane potential and water content. The coupling
between the protonic and water transport is through the
diagonal transport parameters Dpw and Dwp which are
considered to be functions of water content l and
temperature T. A feature of the model is that all
parameters can be obtained using a simplified version of
the model obtained in the limit of conditions representing
AC impedance conductivity measurements. The BFM2 has
a number of features that make it promising. The model:
 relies on rationally derived transport equations based on
the physics of multi-component transport in the
membrane;
 is not restricted by the assumption of equimolar counter
diffusion;
 accounts for the effect of temperature on the sorption
isotherm;
 is not restricted to Nafion, but applies to the entire
family of perfluoro-sulfonated membrane.
Fig. 4 illustrates the enhanced capabilities of the BFM2
model in capturing the variation of the conductivity for all
water contents. Fig. 5 illustrates the predicted water
crossover coefficient WC ¼ N l ½I=ð2F Þ. In the second
part of this paper, the BFM2 is coupled to a fuel cell model
and we will see that the resulting overpotentials and overall
performance predicted by the BFM2 can, for some
operating conditions, differ significantly from those ob-
tained using the classical Springer model.
2.2. Gas diffusion electrodes
The gas diffusion layers (GDLs) consist of an aniso-
tropic fibrous structure, either in the form of a ‘‘paper’’ or a
‘‘woven cloth’’, that allows distribution of reactant gases
through the porous matrix and collection of current
through the fibres (see Fig. 6). At higher current densities,
or when local water and heat management is inadequate,
water condensation takes place leading to potential
‘‘flooding’’ within the GDL (and catalyst layer) as
discussed earlier. In order to remedy this, GDLs are
commonly treated with PTFE (Teflon) to impart hydro-
phobicity, and force water droplets to agglomerate at the
free surface of the GDL. However, the hydrophilic Nafion
in the catalyst layer will absorb and retain liquid water.
7
6
nd
ou
Conductivity σ (S/m)
5 erb
pp
U
d
4 un
r bo
we
Lo
3
r
ge
Sp
rin Thampan
2
1
0 2 4 6 8 10 12
λ (mol H2O / molSO3-)
Fig. 4. Calculated conductivity using BFM2 (solid) and models of
Springer et al. [22] and Thampan et al. [24] against experimental data
[27] for E-form Nafion 117 (dashed bounds).
 ARTICLE IN PRESS
N. Djilali / Energy 32 (2007) 269–280
Fig. 5. Comparison of water crossover (+ve from anode to cathode) for
various cathode (la) and anode side (lc) water content boundary
conditions obtained with (a) Springer et al. membrane model; (b) BFM2.
Thus, liquid water produced is transported from a
saturated catalyst layer to the free surface of the GDL.
A major stumbling block in the reliability of two-phase
models that attempt to account for liquid water transport
in GDLs is the necessity to rely on empirical correlations
and experimental observations obtained in soil and sand
samples—media with a structure that differs significantly
from fibrous or woven GDLs.
The phase distribution in a porous media is potentially
the result of viscous, capillary, and gravitational forces.
When modelling the diffusion layer of a PEMFC, the
second (liquid) phase is taken to consist of a single
component, with transfer of water only across the phase
boundary through evaporation/condensation. For porous
media in which the void space is occupied by two-phases,
the bulk porosity e is divided between the liquid e1 and gas
eg volume fractions, and the liquid saturation is defined as
s1 ¼ 1 =. Depending on saturation, and also to some
273
Fig. 6. Microscope image depicting the random fibre structure of a GDL
formed of Toray carbon paper.
degree on wettability, two-phase flow can exist in three
possible regimes [28].
In PEMFC electrodes, the capillary number (the ratio of
viscous forces to interfacial tension forces) is generally
small and liquid transport is dominated by capillary
diffusion. Extending classical relationships for the hydro-
dynamics of a single-phase in porous media to each phase,
the conservation equations for the gas and liquid phases
take the form [28]:
qð1  sl Þrg
þ r  ðrg ug Þ ¼ S_ g ,
qt
qsl rl
þ r  ðrl ul Þ ¼ S_ l , ð2Þ
qt
where ug and ul are the gas and liquid phase superficial
velocities, ð1  sl Þ and sl  in the left-hand sides represents
the volume of each phase, and S_ g and S_ l are volumetric
sources of gas and liquid.
Similarly, extending the single-phase momentum equa-
tion to a two-phase system yields
kg
ug ¼  rPg ,
mg
kl
ul ¼  rPl , ð3Þ
ml
where rP is the pressure gradient, ðkg Þ and ðkl Þ are phase
specific permeabilities, obtained by correcting the bulk
permeability (k) for the effect of the reduced area open to
each phase due to the presence of the other phase. With five
variables ðsl ; ug ; ul ; Pg and Pl Þ and only four equations, an
additional relation is required to close the system. This can
be obtained by considering the key phenomenon of
capillarity associated with interfacial tension. The micro-
scopic capillary pressure is directly proportional to the
interfacial tension (s) and inversely proportional to the
radius curvature of the interface (r), i.e. Pc / s=r.
 ARTICLE IN PRESS
274 N. Djilali / Energy 32 (2007) 269–280

Low Capillary Pressure with the capillary diffusivity defined as

Low Non-wetting Pressure  
kl dPc
r1
Dðsl Þ ¼ . (6)
ml dsl

Pnw = Pw+PC
Pc σ/r
r1 > r2
unw
Pnw, 2 > Pnw, 1
r2
High Capillary Pressure
High Non-wetting Pressure
Wetting Phase (Gas in GDL)
Non-wetting Phase (Liquid Water in GDL)
Fig. 7. Schematic of capillary diffusion, in which the liquid is the non-
wetting phase.
A schematic illustration of capillary motion in a pore is
shown in Fig. 7. At the end of the pore where the liquid
radius is smaller (lower local saturation), the capillary
pressure is greater than at the end with the larger liquid
radius (greater local saturation). Because the liquid
pressure is the sum of the capillary pressure and the gas
pressure ðPc ¼ Pl  Pg Þ, the hydrodynamic pressure of the
liquid is greater at the end of the pore with the smaller
radius. Therefore, the bulk motion of the liquid is toward
the end with the greater radius (and local saturation). The
transport of liquid water from low to high saturation, as
shown in Fig. 7, is counter-intuitive, and illustrates the
tendency for water, in hydrophobic media such as those
used for gas diffusion layers, to move to ever-increasing
pore diameters according to the capillary pressure’s inverse
proportionality to the liquid radius. This analysis will be
instructive in the subsequent interpretation of visualiza-
tions and simulations of flow between the fibres that make
up the structure of a GDL.
Introducing the definition of macroscopic capillary
Eqs. (4) are the basis of the most general multi-fluid
formulation used so far in complete fuel cell models [29],
and a hierarchy of simplifications, ranging from the
mixture model [30] to simple porosity correction schemes
have been proposed [28]. One of the central problems in all
models of two-phase transport in GDLs is the determina-
tion of a constitutive relation for capillary pressure as a
function of saturation, the expected general form of which
is illustrated in Fig. 8. The prescription of the constitutive
relation has so far relied on the so-called Leverett J-
function, specifically in the form proposed by Udell [31]:
 
Pc k 1=2
JðsÞ ¼ , (7)
s 
with
JðsÞ ¼ 1:417ð1  sÞ  2:120ð1  sÞ2 þ 1:263ð1  sÞ3 . (8)
This is then used to compute the capillary pressure in
conjunction with the contact angle y
s cosðyÞ
Pc ¼ JðsÞ (9)
ðk=Þ1=2
and in turn to determine the capillary diffusivity.
There are a number of issues with Udell’s relation [28],
not the least of which is that the experimental data upon
which it is based was obtained in porous samples of soils,
which have a radically different structure from fibrous
media used in GDLs.
Another key issue in the application of two-phase
transport models is the determination of the phase specific
permeabilities. These must be corrected for the volume
fractions occupied by different phases. One of the
Capillary Pressure

pressure into the momentum equations yields:

kg
ug ¼  rPg ,
mg
kl kl
ul ¼  rPg  rPc . ð4Þ
ml ml
The last term in Eq. (4b) represents capillary diffusion.
Assuming the capillary pressure gradient depends solely on
the saturation gradient, the liquid phase momentum
equation can be rewritten as
Sim Saturation
kl
ul ¼  rPg  Dðsl Þrsl , (5)
ml Fig. 8. General form of the capillary pressure curve.
 ARTICLE IN PRESS
N. Djilali / Energy 32 (2007) 269–280 275

approaches we have recently used to determine the

functional dependence of relative permeability and capil-
lary pressure on saturation is based on stochastic pore
network model simulations illustrated in Fig. 9. These
simulations consist of building a network of pores and
throats, with distributions of pore radii corresponding to
the porous media of interest, and then solving the
momentum transport equations to account for capillary
dominated drainage and cluster formation as a function of
invading phase saturation. Statistical analysis of the
stochastic realizations allows determination of, e.g., the
variation of the relative phase permeability and of capillary
pressure as a function of saturation.
A more fundamental approach consists of resolving the
two phase transport by performing direct numerical
simulations that explicitly resolve the flow, including
dynamic interface tracking, between the fibres. The
equivalent 2D reconstruction of a fibre array correspond-
ing to a typical gas diffusion layer is shown in Fig. 10. The
simulation in this systems start with a medium in contact
with a liquid water reservoir at the bottom at a higher
pressure. A volume of fluid (VOF) technique in conjunc-
Fig. 10. Computational grid for 2D reconstruction of gas diffusion layer.
tion with sub-models for surface tension and wall adhesion
is used to track the evolution of the two phase flow,
including the interface. The instantaneous pressure field In recent studies of diffusion layers and two-phase fuel
obtained from such simulations, Fig. 11, shows clearly the cell models [32,33], it was proposed that liquid water in a
two-phase interface. 2D, as well as 3D simulations provide GDL evolves in an ‘‘inverted tree’’ structure; i.e. a high
fundamental understanding of the two-phase transport, number of small capillaries disperse evenly within the GDL
and parametric simulations that directly account for and eventually converge into larger capillaries and
detailed features, such as degree of hydrophobicity, can eventually result in one very large capillary that breaks
yield data for determining capillary functions that are more through the surface. Analysis of the VOF simulations
representative of GDLs than Eq. (8). together with recent quantitative visualization experiments
performed with laser induced fluorescence [34] suggest
fundamentally different transport dynamics.
For a hydrophobically treated GDL, assuming well
dispersed PTFE coating (i.e. uniform contact angle), the
transport within the GDL is a process of pressure buildup
and breakthrough, due to the fact that capillary forces
provide the main resistance to the progression of the
interface in the fibrous hydrophobic structure. As water is
produced in the catalyst layer, the pressure in the liquid
increases, and the fluid surface slowly passes through the
constrictions of the cross-sections generated by intersecting
fibers (see Fig. 12). When the water has passed halfway
through each constriction, the curvature of the surface
decreases due to the expanding radius of the surface. Thus,
the liquid pressure at the interface drops and induces
hydrodynamic flow in the direction of the expanding
surface. Subsequently, the liquid expands rapidly above
this cross-section until the fluid interface contacts the next
set of fiber intersections. The process then begins again.
The fluid will preferentially pass through the fiber cross-
sections featuring the greatest spacing, as this reduces
capillary pressure resistance. A schematic of the overall
flow mechanism inferred from the analysis of VOF
Fig. 9. Example of invading phase (water) saturation pattern obtained simulations and quantitative visualization [34] is illustrated
using capillary pore network simulations. in Fig. 13, showing a process dominated by fingering and
 ARTICLE IN PRESS
276 N. Djilali / Energy 32 (2007) 269–280

Fig. 11. Two phase flow in GDL: instantaneous pressure field liquid phase (gas phase in blue).

Catalyst Layer Gas Diffusion Layer

Fig. 12. Liquid interface in the cross-section of four intersecting GDL

fibers.

channeling, and featuring numerous ‘‘dead ends’’ where

water transport recedes when an adjacent breakthrough
channel forms.

3. Integration into CFD-based fuel cell models

One of the most challenging aspects of computational
modelling of PEMFCs is the multi-physics nature of the
transport processes, and the coupling between these
processes as illustrated in Fig. 14. Each circle in the figure
represents a transport equation governing a particular
process, and the arrows pointing outward represent the
Fig. 13. Schematic of the mechanism for liquid water transport in PEM
fuel cell gas diffusion layers.
effects of a given transport process on other equations. The
transport equations for gas species, water in liquid and
solid (membrane) phases, and energy are closely coupled
with all other conservation equations. Among all the
 ARTICLE IN PRESS
N. Djilali / Energy 32 (2007) 269–280 277

processes, the transport of water, which exists in the system
in the forms of vapour, liquid, and absorbed in the
polymer, is central to all other transport processes.
Unsaturated water in the vapour phase affects local relative
humidity in the vicinity of the membrane and thus water
transport across the membrane and protonic conductivity,
which in turn impacts the electric potential field. The
transport of liquid water in the porous media affects mass
transport of gas species, while liquid water in the gas
channels changes the pressure field and may alter the
overall flow distribution in a unit cell. A key aspect of any
simulations involving two-phase transport and phase
change is the proper resolution of heat transfer and
irreversibilities [35].
Fig. 15 illustrates the predicted distribution of the rate of
phase change and of liquid water saturation in the cathode
of a PEMFC using a 3D coupled implementation of the
multi-fluid model presented earlier. In this figure, the top
and bottom of the vertical axis correspond to the catalyst
layer/ GDL and gas flow channel/GDL interfaces,
respectively, and the planes correspond to successive
locations from the inlet to the outlet of the fuel cell
section. Positive phase change values correspond to
evaporation, which can be found under the land areas
due to the increased pressure drop. The pressure drop, an
artifact of the increased resistance to gas transport below
the land area, reduces the vapour pressure below the
saturation point. Negative values, indicating condensation,
are most prevalent at the catalyst layer/GDL interface
because of the oxygen consumption and the production of
water vapour. The corresponding distribution of liquid
water saturation in the cathode’s GDL, shown on the right-
hand side, indicates maximum saturation levels of about
10% under the land area at the end of the channel.
Transport of water across the membrane links, as noted
earlier, the anode and cathode sides of a fuel cell.
Polarization curves obtained using a complete fuel cell
model that incorporates the binary friction membrane
model (BFM2) discussed in Section 2.1 are presented in
Fig. 16. A particularly important issue here is the coupling
Springer, RH = 75/ 25
Springer, RH = 25/ 75
0.8 BFM2, RH = 75/ 25 0.8
BFM2, RH = 25/ 75
cell voltage [V]

0.6 0.6
Temperature
(Energy) Latent heat
i2R, iη
0.4 0.4
convection Reaction
mass source rates
Gas species Φ source
Mass & Water Electrical
flux 0.2
Momentum potential 0.2
Δp Membrane σ

s source λ B.C. EOD

convection Water content
Saturation 0
s λ 0 0.2 0.4 0.6 0.8 1 1.2 1.4
cell current [A /cm2]

Fig. 16. Polarization curve predicted using Springer’s membrane model

Fig. 14. Coupling of Transport Phenomena in a PEMFC. and BFM2.

Fig. 15. Rate of phase change [kg/(m3 s)] (left) and liquid water saturation (right) inside the cathodic gas diffusion layer at a current density of 0.8 A/cm2.
(Gas channel opening is located between 0.0005 and 0.0015 m of the cell width, and the land areas between 0–0.0005 and 0.0015–0.002 m.).
 ARTICLE IN PRESS
278 N. Djilali / Energy 32 (2007) 269–280

of water transport within the membrane to water transport

in the electrodes in such a way that mass conservation is
ensured. An empirical relationship between water content
and water activity (membrane sorption isotherm) has to be
enforced to this end. Fig. 16 compares the effects of
different humidification regimes on the anode and cathode
sides (25% and 75% relative humidity). The Springer
model predicts similar polarization curves for both
regimes, while the BFM2 predicts a substantially larger
loss for the RH ¼ 25/75 regime (dry anode). An analysis of
the source of this loss reveals that it is primarily due to
larger ohmic losses predicted through the membrane as a
result of drying on the anode side. In this case, there is no
water to hydrate the anode catalyst layer, unlike the case of
the dry cathode (RH ¼ 75/25), where the water production
from the oxygen reduction and the cathode-to-anode back- Fig. 17. Schematic of a micro-structured air-breathing PEMFC.
diffusion helps to maintain a more uniform membrane
saturation. This highlights the crucial importance of
membrane modelling, particularly for fuel cell systems that
rely on passive water management, such as air breathing
cells considered next.
Exploitation of microscale transport processes in con-
junction with macro-manufacturing makes it possible to
conceive of very high power density fuel cells [3], which
also have the potential of being significantly cheaper than
planar plate and frame fuel cells. Micro-fabrication of flow
fields, current collectors, and electrical interconnects have
been used in a number of designs, but these have generally
relied on the traditional planar MEA architecture. Non-
planar designs can achieve much higher active area to
volume ratios, and hence higher volumetric power densi-
ties, as was demonstrated in our waved cell topology [36].
For applications requiring low power, operation in passive
air-breathing mode offers several advantages, including
system simplification resulting from elimination of mani-
folds and ancillaries such as fans and increased volumetric
power density. Mass transfer limitations are however an Fig. 18. Mass fraction distribution of oxygen in three adjacent electrodes.
issue in such designs [37]. The ambient oxygen mass fraction in the convective air flow (top) is 0.23.
Computational modelling has been particularly valuable
in analyzing a concept that exploits the microscale
transport principles in conjunction with an air-breathing The CFD-based tools were adapted to this problem [37]
non-planar patented design illustrated in Fig. 17. The to assess the functionality of the design, determine the
micro-structured electrodes in this design consist of a nano- severity of potential transport limitations, and to identify
porous material featuring mean pore diameters on the design parameters to alleviate these. In addition to the
order of 100 nm, i.e. two orders of magnitude smaller than coupled transport processes discussed earlier, two specific
in typical GDLs (Fig. 6). With comparable levels of process that have to be accounted for in this case are
porosity, the nano-porous GDL can increase the active natural convection in the ambient air, and Knudsen
area of the deposited catalyst by exploiting the surface to diffusion though the GDL.
volume ratio [3]. The higher functionality of such a GDL Fig. 18 illustrates the distribution of oxygen for a system
can be impeded by [37]: consisting of three adjacent electrodes subjected to non-
uniform convective heat and mass transport currents at the
 the inability of nano-scale porous media to effectively top. The catalyst layer is located along the left hand
transport reactant and product gases to the entire active vertical axis for each electrode. A minimum oxygen mass
area; and fraction of 0.089, a 61% drop from the ambient, is
 the limited ability of natural convection to passively obtained in the central cathode. Though the reduction in
regulate temperature and supply adequate amounts of oxygen concentration is significant, the fuel cell is still far
oxygen to the cathode. from being oxygen starved, indicating that the GDL
 ARTICLE IN PRESS
N. Djilali / Energy 32 (2007) 269–280
parameters and dimensions considered will not result in
insurmountable mass transport limitations.
4. Concluding perspective
Fuel cells are at the vanguard of emerging clean energy
technologies, but must surmount a number of key
challenges in order to successfully compete against incum-
bents such as IC engines and batteries. The availability of
reliable and functional computational tools could help
achieve dramatic reductions in development cycles and
costs, and could spur innovative design required to
surmount some of the cost and performance challenges.
In this paper we have reviewed the array of transport
phenomena encountered in PEM fuel cells, and presented
some of the progress made in our laboratory in addressing
some of the fundamental issues that still limit the
capabilities of computational fuel cell engineering (CFCE).
A new model for coupled protonic and water transport
in polymer membranes was presented. This general model
is based on rational and physical consideration, removes
several of the limitations of the empirical models used to
date, and should provide a good foundation for modelling
all purfluoro-sulfonic acid membranes. Approaches com-
bining pore network modelling, direct numerical simula-
tion of two phase flow in reconstructed gas diffusion layers,
and quantitative visualization have been described, and the
sample results presented revealing fundamental mechan-
isms of liquid water transport in hydrophobic GDLs. The
important issue of integration and coupling was also
discussed in the context of CFD-based fuel cell models, and
we have shown that used judiciously, current models and
computational tools can be used in both conventional
designs and in novel micro-structured fuel cells to provide
detailed information and to analyze design features.
Numerous outstanding problems need to be resolved
before truly functional computational fuel cell engineering
capabilities are achieved. The broader applicability of the
multi-fluid model comes at the cost of solving for an
additional set of field equations and the required coupling
of the phases. This makes the numerical solution less stable
and more computationally intensive. Alternative models
based on various levels of simplifications should be better
assessed, and their range and regimes of validity fully
established in order to allow multiple levels of modelling
and practical simulations, particularly for complete cells
and stacks. In any case a critical issue for all two-phase
models is the determination of appropriate transport
parameters for the various electrode materials (including
the catalyst layer), and in particular the relative perme-
ability and the capillary pressure function.
One of the inherent advantages of the Binary Friction
Membrane Model is that it can be reduced to represent
condition under which AC impedance measurements are
conducted, thus allowing determination of some of the
transport parameters using conductivity data. Experiments
and fundamental micro/nanoscale simulations, such as
279
molecular dynamic simulations, are required to determine
the fundamental properties of specific PFSA membranes,
including the diffusion and interaction coefficients, and the
sorption isotherms. Further this model needs to be
extended to liquid water equilibrated membranes and to
account for transients.
PEMFC catalyst layers consist of a complex composite
structure of interconnected, microporous agglomerates
surrounded by macropores. Charged species, heat, liquid
water, and gas all flow through this layer. Modelling and
experimental characterization of this layer is one of the
most difficult problems in PEMFCs. Multi-scale modelling
strategies are required to resolve the catalyst structure and
guide the formulation of improved macroscopic models
suitable for use in a CFCE framework.
The functionality and usefulness of computational fuel
cell models in a design environment will require the
effective and robust integration of the various submodels
in a CFCE framework that will allow not only design
analysis, but also systematic optimization. This will
necessitate the development of novel algorithms that take
into account the specific nature of the couplings, the large
range of scales encountered in PEMFCs fuel cells, and the
multi-variable nature of the optimization problems.
Acknowledgements
The work presented in this paper is the result of
collaboration with a number of former and current
graduate students, research associates and colleagues. The
author would like to particularly acknowledge the con-
tributions of: Torsten Berning, Brian Carnes, Jeff Fimrite,
Shawn Litster, Ged McLean, Bojan Markicevic, Henning
Struchtrup, and Jay Sui.
Financial support from the following is also gratefully
acknowledged: the MITACS Network of Centres of
Excellence; the Natural Sciences and Engineering Research
Council (NSERC) of Canada; the Canada Research Chairs
program; Angstrom Power Inc.; and Ballard Power
Systems.
References
[1] Sperling D, Cannon JS, (editors). The hydrogen energy transition.
elsevier, 2004.
[2] Stone C, Morrison AE. From curiosity to ‘power to change the
world’. Solid State Ionics 2002;152:1–13.
[3] McLean G, Djilali N, Whale M, Niet T. Application of micro-scale
techniques to fuel cell systems design. In: Proceedings of the 10th
Canadian Hydrogen Conference Quebec City, 2000; p.349–358.
[4] Eikerling M, Kornyshev A, Kulikovsky A. Can theory help to
improve fuel cells? The Fuel Cell Review. 2005; January: 15–24.
[5] Liu HT, Zhou TH. Fuel cell performance augmentation: mass
transfer enhancement. J Enhanc Heat Transf 2003;10:257–74.
[6] Berning T, Lu DM, Djilali N. Three-dimensional computational
analysis of transport phenomena in a PEM fuel cell. J Power Sources
2003;106:284–94.
 ARTICLE IN PRESS
280 N. Djilali / Energy 32 (2007) 269–280
[7] Mazumder S, Cole JV. Rigorous 3-D mathematical modeling of PEM
fuel cells—I. Model predictions without liquid water transport. J
Electrochem Soc 2003;150(11):A1503–9.
[8] Meng H, Wang CY. Large-scale simulation of polymer electrolyte
fuel cells by parallel computing. Chem Eng Sci 2004;59:3331–43.
[9] Nguyen PT, Berning T, Djilali N. Computational model of a PEM
fuel cell with serpentine gas flow channels. J Power Sources
2004;130:149–57.
[10] Weber AZ, Newman J. Modeling transport in polymer-electrolyte
fuel cells. Chem Rev 2004;104:4679–726.
[11] Shimpalee S, Greenway S, Spuckler S, Van Zee JW. Predicting water
and current distributions in a commercial-size PEMFC. J Power
Sources 2004;135(1–2):79–87.
[12] Meng H, Wang CY. Large-scale simulation of polymer electro-
lyte fuel cells by parallel computing. Chem Eng Sci 2004;59:
A3331–43.
[13] Sivertsen BR, Djilali N. CFD based modelling of proton exchange
membrane fuel cells. J Power Sources 2005;141:65–78.
[14] Wang L, Liu HT. Performance studies of PEM fuel cells with
interdigitated flow fields. J Power Sources 2004;134(2):185–96.
[15] Lum KW, McGuirk JJ. Three-dimensional model of a complete
polymer electrolyte membrane fuel cell—model formulation, vali-
dation and parametric studies. J Power Sources 2005;143(1-2):
103–24.
[16] Birgersson E, Noponen M, Vynnycky M. Analysis of a two-phase
non-isothermal model for a PEFC. J Electrochem Soc 2005;152(5):
A1021–34.
[17] Senn SM, Poulikakos D. Multiphase transport phenomena in the
diffusion zone of a PEM fuel cell. J Heat Transfer 2005;127(11):
1245–59.
[18] Wang QP, Song DT, Navessin T, Holdcroft S, Liu ZS. A
mathematical model and optimization of the cathode catalyst
layer structure in PEM fuel cells. Electrochim Acta 2004;50(2–3):
725–30.
[19] Sun W, Peppley BA, Karan K. An improved two-dimensional
agglomerate cathode model to study the influence of catalyst layer
structural parameters. Electrochim Acta 2005;50(16–17):3359–74.
[20] Faghri A, Guo Z. Challenges and opportunities of thermal manage-
ment issues related to fuel cell technology and modeling. Int J Heat
Mass Transf 2005;48(19–20).
[21] Kreuer KD, Paddison SJ, Spohr E, Schuster M. Transport in proton
conductors for fuel-cell applications: simulations, elementary reac-
tions, and phenomenology. Chem Rev 2004;104:4637–78.
[22] Springer TE, Zawodzinski TA, Gottesfeld S. Polymer electrolyte fuel-
cell model. J Electrochem Soc 1991;138:2334–42.
[23] Fimrite J, Struchtrup H, Djilali N. Transport phenomena in polymer
electrolyte membranes: I modelling framework. J Electrochem Soc
2005;152:A1804–14.
[24] Thampan T, Malhotra S, Tang H, Datta RJ. Modeling of conductive
transport in proton-exchange membranes for fuel cells. Electrochem
Soc 2000;147:3242–50.
[25] Kerkhof PJAM. A modified Maxwell–Stefan model for transport
through inert membranes: the binary friction model. Chem Eng J
1996;64(3):319–43.
[26] Fimrite J, Carnes B, Struchtrup H, Djilali N. Transport phenomena
in polymer electrolyte membranes: II binary friction membrane
model. J Electrochem Soc 2005;152:A1815–23.
[27] Sone Y, Ekdunge P, Simonsson D. Proton conductivity of Nafion 117
as measured by a four-electrode AC impedance method. J Electro-
chem Soc 1996;143:1254–9.
[28] Litster S, Djilali N. Two-phase transport in porous gas diffusion
electrodes. In: Faghri M, Sundén B, editors. Chapter 5 Transport
Phenomena in Fuel Cells, WIT Press, 2005 175–213.
[29] Berning T, Djilali N. A 3D multi-phase, multicomponent model of
the cathode and anode of a PEM fuel cell. J Electrochem Soc
2003;150:A1589–98.
[30] Wang ZH, Wang CY, Chen KS. Two-phase flow and transport in the
air cathode of proton exchange membrane fuel cells. J Power Sources
2001;94:40–50.
[31] Udell KS. Heat-transfer in porous-media considering phase-change
and capillarity—the heat pipe effect. Int J Heat Mass Transf
1985;28:485–95.
[32] Nam H, Kaviany M. Effective diffusivity and water-saturation
distribution in single- and two layer PEMFC diffusion medium. Int J
Heat Mass Transf 2003;46:4595–611.
[33] Pasaogullari U, Wang CY. Liquid water transport in gas diffusion
layer of polymer electrolyte fuel cells. J Electochem Soc
2004;151:A399–406.
[34] Litster S, Sinton D, Djilali N. Ex-situ visualization of liquid water
transport in PEM fuel cell gas diffusion layers. J Power Sources
2006;154(1):95–105.
[35] Kjelstrup S, Rosjorde A. Local and total entropy production and
heat and water fluxes in a one-dimensional polymer electrolyte fuel
cell. J Phys Chem B 2005;109(18):9020–33.
[36] Mérida WR, McLean G, Djilali N. Non-planar architecture for
proton exchange membrane fuel cells. J Power Sources 2001;102:
178–85.
[37] Litster S, Pharoah JG, McLean G, Djilali N. Computational analysis
of heat and mass transfer in a micro-structured PEMFC cathode. J
Power Sources 2006;156(2):334–44.