International Journal of Heat and Mass Transfer 48 (2005) 4410–4427

www.elsevier.com/locate/ijhmt

Constructal PEM fuel cell stack design

a,*
J.V.C. Vargas , J.C. Ordonez b, A. Bejan c

a
Departamento de Engenharia Mecânica, Universidade Federal do Paraná, C.P. 19011, Curitiba, Paraná 81531-990, Brazil
b
Department of Mechanical Engineering and Center for Advanced Power Systems, Florida State University,
Tallahassee, FL 32310-6046, USA
c
Department of Mechanical Engineering and Materials Science, Duke University, Durham, NC 27708-0300, USA

Received 8 March 2005; received in revised form 12 May 2005

Available online 14 July 2005

Abstract

This paper describes a structured procedure to optimize the internal structure (relative sizes, spacings), single cells
thickness, and external shape (aspect ratios) of a polymer electrolyte membrane fuel cell (PEMFC) stack so that net
power is maximized. The constructal design starts from the smallest (elemental) level of a fuel cell stack (the single
PEMFC), which is modeled as a unidirectional flow system, proceeding to the pressure drops experienced in the headers
and gas channels of the single cells in the stack. The polarization curve, total and net power, and efficiencies are
obtained as functions of temperature, pressure, geometry and operating parameters. The optimization is subjected to
fixed stack total volume. There are two levels of optimization: (i) the internal structure, which accounts for the relative
thicknesses of two reaction and diffusion layers and the membrane space, together with the single cells thickness, and
(ii) the external shape, which accounts for the external aspect ratios of the PEMFC stack. The flow components are
distributed optimally through the available volume so that the PEMFC stack net power is maximized. Numerical
results show that the optimized single cells internal structure and stack external shape are ‘‘robust’’ with respect to
changes in stoichiometric ratios, membrane water content, and total stack volume. The optimized internal structure
and single cells thickness, and the stack external shape are results of an optimal balance between electrical power output
and pumping power required to supply fuel and oxidant to the fuel cell through the stack headers and single-cell gas
channels. It is shown that the twice maximized stack net power increases monotonically with total volume raised to the
power 3/4, similarly to metabolic rate and body size in animal design.

2005 Elsevier Ltd. All rights reserved.

Keywords: Constructal; Fuel cells; PEMFC structure; Fuel cell thickness; Shape

1. Introduction

The world currently relies almost entirely on fossil

*
Corresponding author. Tel.: +55 41 361 3307; fax: +55 41
fuels to supply its energy needs. In the near future, it
361 3129. is expected that the worldÕs fossil fuel reserves will not
E-mail addresses: jvargas@demec.ufpr.br (J.V.C. Vargas), meet the growing energy demand. To reduce dependence
ordonez@eng.fsu.edu (J.C. Ordonez), dalford@duke.edu on oil, the worldÕs most developed nations have been
(A. Bejan). investing in cutting edge research to develop sustainable

0017-9310/$ - see front matter
2005 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ijheatmasstransfer.2005.05.009
 J.V.C. Vargas et al. / International Journal of Heat and Mass Transfer 48 (2005) 4410–4427 4411

Nomenclature

A area, m2 PEMFC polymer electrolyte membrane fuel cell

Ac total gas channel cross-sectional area, m2 Pr Prandtl number, lcp/k
Ach headers cross-sectional area, m2 q tortuosity
As unit fuel cell cross-sectional area, m2 Q reaction quotient
e
A dimensionless area Q_ heat transfer rate, W
B dimensionless constant Qe dimensionless heat transfer rate
c specific heat, kJ kg
1 K
1 r pore radius, m
cp specific heat at constant pressure, kJ kg
1 K
1 R ideal gas constant, kJ kg
1 K
1
cv specific heat at constant volume, R universal gas constant, 8.314 kJ kmol
1 K
1
kJ kg
1 K
1 ReDh Reynolds number based on Dh, uDhq/l
C constant S dimensionless conversion factor
CV control volume t time, s
D Knudsen diffusion coefficient, m2 s
1 T temperature, K
E dimensionless conversion factor, Eq. (32) u mean velocity, m s
1
Dh gas channel hydraulic diameter, m ~
u dimensionless mean velocity
f friction factor U global wall heat transfer coefficient,
F Faraday constant, 96,500 C eq
1 W m
2 K
1
h heat transfer coefficient, W m
2 K
1 V electrical potential, V
~
h dimensionless heat transfer coefficient VT total volume, m3
Hi(Ti) molar enthalpy of formation at a tempera- Ve dimensionless electrical potential
ture Ti of reactants and products, kJ kmol
1 Ve T dimensionless total volume
of compound i W electrical work, J
He i ðhi Þ dimensionless molar enthalpy of formation We dimensionless required pumping power in
at a dimensionless temperature hi of reac- the headers, Eq. (29)
tants and products e net
W dimensionless PEMFC stack net power, Eq.
io,a, io,c exchange current densities, A m
2 (30)
iLim,a, iLim,c limiting current densities, A m
2 e net;s
W dimensionless single PEMFC net power
I current, A ep
W dimensionless required single PEMFC
eI dimensionless current pumping power
j mass flux, kg s
1 m
2 es
W dimensionless single PEMFC total electrical
k thermal conductivity, W m
1 K
1 power
K permeability, m2 x axial direction, Fig. 2
Kbc bending-contraction coefficient y2,4,6 size constraints
Kbe bending-expansion coefficient [] molar concentration of a substance, mol l
1
~k dimensionless thermal conductivity
L length, m Greek symbols
Lc, Lt gas channels internal dimensions as shown aa, ac anode and cathode charge transfer coeffi-
in Fig. 1, m cients
Lx, Ly, Lz stack length, width and height, respec- b electrical resistance, X
tively, m c ratio of specific heats
m_ mass flow rate, kg s
1 d gas channel aspect ratio
M molecular weight, kg kmol
1 DG molar Gibbs free energy change,
n equivalent electron per mole of reactant, kJ kmol
1 H2, Eq. (9)
eq mol
1 DGe dimensionless Gibbs free energy change
n_ molar flow rate, kmol s
1 DH molar enthalpy change, kJ kmol
1 H2
nc number of parallel ducts in gas channel DHe dimensionless enthalpy change
ns number of single cells in the stack DS molar entropy change, kJ kmol
1 K
1
N dimensionless global wall heat transfer coef- f stoichiometric ratio
ficient ga, gc anode and cathode charge transfer overpo-
p pressure, N m
2 tentials, V
~
ps perimeter of cross-section, m gd,a, gd,c anode and cathode mass diffusion overpo-
P dimensionless pressure tentials, V
4412 J.V.C. Vargas et al. / International Journal of Heat and Mass Transfer 48 (2005) 4410–4427

gi ideal efficiency, Eq. (31) H2O water produced in a single PEMFC

gI first law efficiency, Eq. (32) i irreversible
gII second law efficiency, Eq. (33) i, a irreversible at the anode
gnet PEMFC stack net efficiency, Eq. (34) i, c irreversible at the cathode
ga , ~gc
~ dimensionless anode and cathode charge in inlet
transfer overpotentials Lim limiting condition
gd;a , ~gd;c
~ dimensionless anode and cathode mass dif- (l) liquid phase
fusion overpotentials m maximum with respect to fuel cell internal
gohm
~ dimensionless fuel cell total ohmic potential structure and cell thickness
loss, Eq. (18) mm maximum with respect to PEMFC internal
h dimensionless temperature structure, cell thickness and external shape
k ionomer water content ohm ohmic
l viscosity, Pa s opt optimal value
mi reaction coefficients out outlet
n dimensionless length ox oxidant
q density, kg m
3 O2 oxygen
q
~ dimensionless density pm polymer electrolyte membrane
r electrical conductivity, X
1 m
1 ref reference level
s dimensionless time s single PEMFC
/ porosity s, a anode solid side
w dimensionless mass flow rate s, c cathode solid side
v header cross-section vertical side
Subscripts w wall
a anode wa water
(aq) aqueous solution wet wetted surface
c cathode 0 initial condition
e reversible 1, . . ., 7 control volumes, Fig. 2
f fuel ij interaction between CVi and CVj
(g) gaseous phase (i, j = 1, . . . ,7)
h header cross-section horizontal side 1 ambient
hf,in input fuel header
hf,out output fuel header Superscript
hox,in input oxidant header  standard conditions [gases at 1 atm, 25 C,
hox,out output oxidant header species in solution at 1 M, where M is
H+ hydrogen cation the molarity = (moles solute)/(liters solu-
H2 hydrogen tion)]

alternative energy sources such as fusion, and to employ the precious material load in the catalyst layers, mem-
hydrogen produced from diverse sources. The energy brane improvement, and on the multiple internal physi-
engineering community is expected to play an important cal and electrochemical aspects of a single PEMFC.
role in this first international attempt to limit atmo- During the last two decades, such efforts have reduced
spheric pollution (in particular, to reduce the emission the energy cost of PEMFC from US$4500/kW to
of greenhouse gases). To this end, polymer electrolyte US$1200/kW, which is still far from the energy cost
membrane fuel cells (PEMFC) offer an attractive and (US$20/kW) of internal combustion engines [1,2]. At
environmentally friendly on-site alternative power gen- the PEMFC stack level, suitable materials have been
eration technology, which is highly efficient, and has sta- developed for making small and light weight compo-
tionary and mobile applications [1–3]. nents [4], high power density electrodes to reduce the
The commercial development of PEMFC still has total numbers and sizes of cells and bipolar plates
serious technical and economic hurdles to overcome. A [4,5]. Furthermore, light weight metal conductive ele-
practical fuel cell system requires the integration of the ments together with nonconductive elements fabricated
stack with fuel processing, heat exchange, power condi- from engineering thermoplastics were used to yield light
tioning, and control systems. Considerable research has weight and low cost PEMFC stacks [6]. Alternative
been conducted on new and advanced materials to lower materials and design concepts were presented by Kumar
 J.V.C. Vargas et al. / International Journal of Heat and Mass Transfer 48 (2005) 4410–4427 4413

and Reddy [7] for the bipolar/end plate of PEMFC ration of the stack in terms of number of cells and cell
stacks, which is one of the most important and costliest surface area, but did not consider headers and gas chan-
stack components and accounts for more than 80% of nels pumping power losses. No transient model was
the total weight of the stack. Therefore, one direction found in the literature that addresses the spatial temper-
to be pursued to make PEMFC systems competitive eco- ature and pressure gradients in a PEMFC stack, pres-
nomically with other existing power and vehicular sys- sure drops in the headers and single cells gas channels
tems is system optimization at the stack level. and their effect on performance.
The PEMFC stack is a combination of several flow This study proposes a new integrative framework for
systems, i.e., electrical, chemical, fluid and heat flows. the energy-based design of PEMFC stacks. The ap-
The optimization of flow-system architecture is common proach is cross-disciplinary and pursues simultaneously
in engineering and nature. Constructal theory [8] is the (i) the local optimization of components and processes
thought that geometry (flow architecture) is generated with (ii) the optimal global integration and configuration
by the pursuit of global performance subject to global of the system. The constructal approach chosen in this
constraints, in flow systems the geometry of which is free paper was introduced in previous studies for a single
to vary. According to constructal theory, the optimiza- alkaline fuel cell [27] and for a single PEMFC [28],
tion of flow architecture starts at the smallest (elemental) which divides the single fuel cell into several control vol-
scale, in this case, the single PEMFC. Irreversibilities umes that correspond to the most representative parts of
due to pressure drops, charge transfer, electrical resis- the flow system. For the PEMFC stack, the single cell
tances, and mass diffusion are minimized together for model is extended for predicting unsteady state opera-
maximum global performance at the stack level. The tion, and the stack input and output headers pressure
technique proposed in this paper traces the general drops are accounted for. The model is represented by
direction of flow geometry optimization subject to glo- a system of ordinary differential equations with respect
bal constraints (e.g., volume), with the objective of max- to time, the solution of which consists of the tempera-
imizing the PEMFC stack net power or the net power tures and pressures of each control volume, and the
density. This approach is general because it may be used polarization and net power curves for the PEMFC
in conjunction with other methods of improvement dis- stack. The model is simple enough to insure small com-
cussed in the previous paragraph. putational time requirements, so that it is possible to
The technical literature shows several previous simulate the flow in a very large number of competing
PEMFC analytical and numerical models. At the single flow configurations.
PEMFC level, models considering one or two-dimen-
sional geometry, isothermal and steady state operation
[9–15], and three-dimensional steady state [16] were 2. Mathematical model
developed. The two- and three-dimensional models are
not suitable for the optimization of flow geometry, A PEMFC stack is shown schematically in Fig. 1.
because for a PEMFC stack they would require the solv- The stack is fed by two input headers, one carries fuel
ing of partial differential equations for flow simulation and the other oxidant, which are delivered to each single
in a very large number of flow configurations. At the PEM fuel cell in the stack. After the fuel and oxidant
PEMFC stack level, Amphlett et al. [17] introduced a streams sweep the single PEMFC gas channels in the
transient mathematical model for predicting the re- stack, two output headers collect the used gases and
sponse and performance of the system for applications the water that is produced by the reactions in the fuel
where the operating conditions change with time, con- cells. Electrical power is produced by the stack, and a
sidering heat losses, changes in stack temperature, reac- fraction is used to pump the fuel and oxidant across
tant gas concentrations, and other internal phenomena, the PEMFC stack, through the input and output head-
but no consideration was given to the pressure drops ers, and through very narrow gas channels in each single
experienced by the fuel and oxidant in the feeding head- fuel cell. For simplicity, the model is based on the
ers and gas channels. Experimental results were pre- assumption that the fuel stream is pure hydrogen, and
sented by Hamelin et al. [18] for the time response of a that the oxidant is pure oxygen.
stationary PEMFC stack under fast load commutations, A single PEMFC is divided into seven control vol-
and compared to the Amphlett et al. [17] model with umes that interact energetically with one another. In
good agreement. Several other experimental studies have the present model, it is assumed the existence of a cool-
been conducted to access the dynamic behavior of ing channels system that irrigates the bipolar plates, and
PEMFC stacks in search for performance improvement that the cooling fluid (e.g., water, air) is kept at an aver-
in the transient mode [19–23]. Recent studies presented age specified temperature, T1, between single cells and
PEMFC stack steady state mathematical models to around the stack. In this way, the electrical power deliv-
predict performance [24–26]: one study [26] presented a ered by the entire stack results from the analysis of one
genetic algorithm technique for finding the best configu- single cell, i.e., the electrical power produced by the
4414 J.V.C. Vargas et al. / International Journal of Heat and Mass Transfer 48 (2005) 4410–4427

model. The instantaneous electrical potential and power

(O2)in
(H2)in
of the fuel cell are obtained as functions of each com-
partment temperature, by subtracting from the revers-
ible potential the losses due to surface overpotentials
(poor electrocatalysis), slow diffusion and internal ohmic
losses through the cell (resistance of individual cell com-
ponents, including electrolyte membrane, bipolar plates,
interconnects and any other cell components through
Lx
which electrons flow). These are functions of the total
cell current (I), which is directly related to the external
load (or the cell voltage), and time. In sum, the total cell
current and time are considered the independent vari-
ables. The following analysis is therefore for unsteady
state fuel cell operation.
The hydrogen mass flow rate required by a single
Ls Lz PEMFC in the stack for the current (I) dictated by the
external load is
Ly
I
(O2 + H2O)out m_ H2 ¼ n_ H2 M H2 ¼ M H2 ð1Þ
nF
(H2)out
where n_ i is the molar flow rate for species i, Mi the
molecular weight of species i, n the number of moles
of electrons formed in the reaction and F the Faraday
Fig. 1. The configuration of a PEMFC stack. constant, 96,500 C eq
1. Accordingly, the oxygen mass
flow rate needed for a single PEMFC in the stack is
m_ O2 ¼ n_ H2 M O2 .
stack is the electrical power produced by a single cell Nondimensional variables are defined based on the
multiplied by the number of cells in the stack, which is geometric and operating parameters of the system. Pres-
a variable in the PEMFC stack constructal design proce- sures are referenced to ambient conditions: Pi = pi/p1,
dure. However, the required pumping power depends on and temperatures referenced to the specified coolant
the flow structure of the entire PEMFC stack. (e.g., water, air) average temperature: hi = Ti/T1.
The mathematical model for a single cell operating at The total volume of the fuel cell stack, V T ¼ Lx Ly Lz ,
steady state was introduced in a previous study by Var- is finite and fixed. This is a realistic design constraint,
gas et al. [28]. In this paper, the mathematical model is which accounts for the finiteness of the available space
extended for the unsteady state to predict the fuel cell and the general push for doing the most with limited
stack start up time, and for possible future control and resources. The maximization of performance for the spec-
real time operation purposes. For that, the model equa- ified volume means the maximization of performance
tions are rederived with the inclusion of the thermal density. The fixed length scale V 1=3
T is used to nondimen-
inertia terms for each fuel cell compartment. sionalize all the lengths of the fuel cell stack geometry.
Fig. 2 shows schematically the division of a single Also, for presenting dimensionless results and recog-
PEMFC into seven control volumes. In addition, there nizing the fuel cell net power varies with size VT, a dimen-
are two bipolar plates (interconnects) that have the func- sionless fuel cell total volume is defined. The
tion of allowing the electrons produced by the electro- dimensionless lengths and total volume are written as
chemical oxidation reaction at the anode to flow to the Lj VT
external circuit or to an adjacent cell. The control vol- nj ¼ ; Ve T ¼ ð2Þ
V 1=3
T
V T;ref
umes (CV) are the fuel channel (CV1), the anode diffu-
sion backing layer (CV2), the anode reaction layer where the subscript j indicates a particular dimension of
(CV3), the polymer electrolyte membrane (CV4), the the fuel cell stack geometry, Figs. 1 and 2, and VT,ref is a
cathode reaction layer (CV5), the cathode diffusion reference volume. Additional dimensionless variables
backing layer (CV6) and the oxidant channel (CV7). are defined as
The model consists of the complete conservation
equations for each control volume, and the equations m_ i U wi V 2=3
T e i ¼ Ai ;
accounting for electrochemical reactions. The reversible wi ¼ ; Ni ¼ ; A
m_ ref m_ ref cp;f V 2=3
T
electrical potential and power of the fuel cell are then
computed (based on the reactions) as functions of the cp;i t
ci ¼ ; s¼ ð3Þ
temperature and pressure fields determined by the cv;i tref
 J.V.C. Vargas et al. / International Journal of Heat and Mass Transfer 48 (2005) 4410–4427 4415

Tf , m· f , pf , Af coolant (T ∞ , p∞ ) Tox , m· ox ,pox , Aox

O2 O2

Bipolar plate
(interconnect)

H2 H2
Perfluorosulfonic
acid membrane Lz

H+ H+
H2 O H2 O
- +
CV1 CV2 CV3 CV4 CV5 CV6 CV7

m· H2O

Lb L1 L2 L3 L4 L5 L6 L7 Lb

0 x
Ls

Lt
Lc
Ly Anode Electrolyte Cathode
I I

Fig. 2. The internal structure of a single PEMFC.

where subscript i indicates a substance or a location in (2 6 j 6 6) and Vj = ncLcLjLz (j = 1, 7), where nc is the
the fuel cell, m_ ref ¼ p1 V T =ðRf T 1 tref Þ is a reference mass integer part of Ly/(Lt + Lc), i.e., the number of parallel
flow rate, and Rf is the ideal gas constant of the fuel, N is ducts in each gas channel (fuel and oxidant).
the dimensionless global wall heat transfer coefficient, A e The average fuel pressure (pf in headers and CV1)
is the dimensionless area, c is the ratio of specific heats, and oxidant pressure (pox in headers and CV7) are
cp,i and cv,i are the specific heats at constant pressure and assumed known and constant during fuel cell operation.
volume of substance i, respectively, and tref is a specified The stoichiometric ratio for an electrode reaction is
reference time. defined as the provided reactant (moles/s) divided by
In a single fuel cell shown in Fig. 2, the wall heat trans- the reactant needed for the electrochemical reaction of
fer area of one control volume is Awi ¼ ~ps Li ð2 6 i 6 6Þ interest. In the present model, stoichiometric ratios
and Awi ffi ~ps Li þ Ly Lz (i = 1, 7; assuming that Lt  Lc in greater than 1 are prescribed on the fuel side (f1) and
Fig. 2), where ~ps ¼ 2ðLy þ Lz Þ is the perimeter of the fuel oxidant side (f7). The mass and energy balances for
cell cross-section. The control volumes are Vj = LyLzLj incompressible flow yield the temperature in CV1,
4416 J.V.C. Vargas et al. / International Journal of Heat and Mass Transfer 48 (2005) 4410–4427

dh1 h e e 12 þ Q
e 1ohm
i h c
1;0 f average pressures in CV2 and CV6 are estimated as
¼ Q w1 þ wf ðhf
h1 Þ þ Q
ds P f nc n1 nc nz P i ¼ 12 ðP i;in þ P i;out Þ, i = 2, 6.
ð4Þ The energy balance delivers the CV2 temperature,
" #
where h1,0 is a specified initial condition, wf ¼ f1 wH2 , dh2 Qe2 cs;a
e wi ¼ N i A
e wi ð1
hi Þ, where subscript i refers to one of ¼ h1
h2 þ ð7Þ
Q ds wH2 q ~s;a ð1
/2 Þn2 ny nz
the control volumes, Q e 12 ¼ ~h1 A e s ð1
/2 Þðh2
h1 Þ,
h1 ¼ h1 V 2=3
~
T =ð m_ c
ref p;f Þ, / i are the porosities, and where cs,a = cp,f/cs,a, cs,a is the specific heat of the anode
e s ¼ Ly Lz =V 2=3 is the dimensionless cross-sectional area
A solid material, and q ~i is the dimensionless density de-
T
of the fuel cell. The dimensionless heat transfer rates fined by q ~i ¼ qi Rf T 1 =p1 .
for all the compartments are Q e i ¼ Q_ i =ðm_ ref cp;f T 1 Þ, In the anode reaction layer (CV3), the electrical cur-
where i accounts for any of the heat transfer interactions rent is generated by the electrochemical oxidation
that are present in the model. The ohmic heating is reaction,
e iohm ¼ I 2 bi =ðm_ ref cp;f T 1 Þ, where subscript i refers to a
Q
control volume (1–7), and b (X) is the electrical H2ðgÞ ! 2Hþ
ðaqÞ þ 2e


ð8Þ
resistance.
Next is the anode backing diffusion layer (CV2), where a perfluorosulfonic acid membrane (e.g., Nafion
where reactions are absent. Both electrodes in a fuel cell 117 by DuPont) has been assumed in the electrolyte.
are porous, such that a large surface area can be ob- CV3 is divided into two compartments, the solid and
tained to provide good contact between the electrode the ionomer that form the anode reaction layer. How-
and the electrolyte (ionomer). Although the porous med- ever, in the thermal analysis only the solid is taken into
ium consists of a solid side and a fluid side, the mass of account, because the mass of ionomer in CV3 is negligi-
fluid in CV2 is negligible relative to the mass of solid, ble in comparison with the mass of solid. The dimen-
therefore only the solid is taken into account in the sionless net heat transfer in CV3 is given by
e2 ¼ Qe 3 ¼
Q e 23 þ Q
e w3 þ Q e 34 þ Q e 3ohm . The heat transfer rate
energy balance. The net heat transfer rates are Q

Q e 12 þ Q
e w2 þ Q e 23 þ Q e 2ohm , where Q e 23 ¼
e k s;a ð1
/2 Þ  between CV3 and CV4 (the polymer electrolyte mem-
e s ðh2
h3 Þ=½ðn2 þ n3 Þ=2. The dimensionless thermal brane) is dominated by conduction, therefore Q e 34 ¼
A

ð1
/3 Þðh3
h4 Þ A k s;a e
e s 2e k p =ðn4 e
k s;a þ n3 e
k p Þ.
conductivity is defined by e k i ¼ k i V 1=3
T =ðm _ ref cp;f Þ.
The contact areas in the porous anode and cathode The mass and energy balances for CV3, together with
are estimated by assuming dual-porosity electrodes. the anode reaction equation deliver the relations
The pores are approximated as parallel tubes with an n_ H2 ¼ m_ H2 =M H2 , n_ Hþ ¼ 2n_ H2 , m_ Hþ ¼ 2n_ H2 M Hþ and
average diameter of the same order as the square root dh3 h e e3
e 3 þ DG
i cs;a
of the porous medium permeability, K1/2. Therefore, ¼ Q3
DH ð9Þ
ds ~s;a ð1
/3 Þn3 ny nz
q
the contact (wetted) area for each porous control vol-
ume is Aj;wet ¼ 4/j Lj K
1=2 j As , where Kj are the where ðD H e 3; DG e 3 Þ ¼ n_ H ðDH 3 ; DG3 Þ=ðm_ ref cp;f T 1 Þ.
2
permeabilities. The dimensionless enthalpy of formation is defined
Diffusion is assumed to be the dominant transport by H e i ¼ n_ i H i =ðm_ ref cp;f T 1 Þ, where the subscript i
mechanism across the diffusion and catalyst layers. For refers to a substance or a control volume. The enthalpy
Knudsen flow, the fuel and oxidant mass fluxes in the change due to the anode reaction is given by DH 3 ¼
P P
porous layers are given by [29] products ½mi H i ðT i Þ
reactants ½mi H i ðT i Þ and We3 =
DG3,
DH3 is the CV3 reaction enthalpy change (kJ/kmol H2); mi
ji ¼
½Dðqout
qin Þ=Li ; i ¼ 2; 6 ð5Þ are the stoichiometric coefficients; Hi(Ti) is the molar en-
thalpy (kJ/kmol) of formation at a temperature Ti of reac-
where D ¼ Bfr½8RT =ðpMÞ1=2 /q g is the Knudsen diffu- tants and products of compound i; DG3 is the CV3
sion coefficient, q the density, R the universal gas con- reaction Gibbs free energy change (kJ/kmol H2) and
stant, / the porosity, q the tortuosity [30,31], and B is We3 is the maximum (reversible) electrical work generated
a correction coefficient. By using Eq. (5) and the ideal due to the reaction in CV3 (kJ/kmol H2).
gas model for H2 and O2, the pressures of the hydrogen The molar enthalpies of formation are obtained from
and oxygen that enter the catalyst layers are calculated tabulated values [32,33] at T2 for H2(g) and T3 for Hþ ðaqÞ ,
from and at 1 atm, because DH is independent of pressure.
ji Ri T 1 Li h i The reaction Gibbs free energy change DG is a function
P i;out ¼ P i;in
; i ¼ 2; 6 ð6Þ of temperature, pressure and concentrations,
Di p 1
DG ¼ DG þ RT ln Q ð10Þ
where j2 ¼ m_ H2 =A3;wet and j6 ¼ m_ O2 =A5;wet , and A3,wet
  
and A5,wet are the wetted areas in the porous catalyst where DG = DH
TDS is the standard Gibbs free
layers. Note also that P2,in = Pf and P6,in = Pox. The energy (kJ/kmol H2); DH is the standard enthalpy
 J.V.C. Vargas et al. / International Journal of Heat and Mass Transfer 48 (2005) 4410–4427 4417

change (kJ/kmol); DS is the standard entropy change catalyst layer morphology, temperature and pressure).
(kJ/kmol K
1) [gases at 1 atm, 25 C, species in solution The potential loss due to mass diffusion is [35]
at 1 M, where M is the molarity = (moles solute)/(liters

solution)], and Q is the reaction quotient. The reaction RT 3 I
gd;a ¼ ln 1
ð14Þ
quotient Q has the same mathematical form as the reac- nF A3;wet iLim;a
tion equilibrium constant: the difference is that the terms
that appear in Q are instantaneous pressures and con- In Eq. (14) the limiting current density at the anode
centrations rather than equilibrium values. Therefore, (iLim,a) occurs at high values of the surface overpotential,
in the present reaction [Eq. (8)] the resulting expression when the gas is completely depleted in the very thin layer
for Q3 is Q3 ¼ ½Hþ 2 þ of active catalyst situated at the interface with the gas
ðaqÞ  =p H2 , where ½HðaqÞ  is the molar
concentration of the acid solution, (moles/1), and diffuser, i.e., P2,out = 0. Therefore, Eq. (6) is rearranged
pH2 = p2,out, i.e., the partial pressure of H2 in atmo- to obtain iLim,a = Pfp1D2nF/(MH2L2Rfh2T1).
spheres at the CV2 outlet. Recall that pure liquids or sol- The dimensionless potentials are defined based on a
ids do not appear in the calculation of Q3; neither does reference voltage, Ve i ¼ V i =V ref and ~ g ¼ gi =V ref , where
the solvent in a dilute solution. subscript i accounts for all the potentials that are present
In a polymer electrolyte membrane, the water content in the fuel cell. The resulting electrical potential at the
(k) is described as the ratio of the number of water mol- anode is Ve i;a ¼ Ve e;a
~
ga
j~ gd;a j, where we have taken
ecules to the number of charge sites, i.e., the number of the absolute value of ~ gd;a , because ~ gd;a < 0 (cathodic
ions, SO
þ overpotential).
3 H . Zawodzinski et al. [34] measured water
content for the Nafion 117 membrane, and found that, Next is the analysis of the polymer electrolyte mem-
for equilibrium with saturated water vapor, k = 14 at brane (CV4), which interacts with CV3, CV5 and the
30 C. Springer et al. [9] found that water vapor and ambient. In the cathode reaction layer (CV5), the reac-
liquid water (in equilibrium with each other) equilibrate tion is
separately to different membrane water contents, 1
namely, k = 16.8 at 80 C and k = 22 at 100 C. Usually, O2ðgÞ þ 2e
þ 2Hþ
ðaqÞ ! H2 OðlÞ ð15Þ
2
the anode water content in the anode is different than in
the cathode [28]; therefore for assumed values of ka (an- Eqs. (8) and (15) and the conservation of mass in CV4
ode water content) and kc (cathode water content), and require 2n_ H2 ¼ n_ Hþout ¼ n_ Hþin ¼ 2n_ O2 . In conclusion, n_ O2 ¼
by assuming a linear variation of the water content n_ H2 , where n_ O2 ¼ 2m_ O2 =M O2 . Accordingly, the required
along the membrane thickness, the average water con- oxidant mass flow rate is m_ O2 ¼ m_ H2 M O2 =ð2M H2 Þ. The
tent in the membrane is defined as net heat transfer in CV4 is obtained from Q e 4 ¼
Q
e 34 þ
ka þ kc e e
Q w4 þ Q 45 þ Q 4ohm e and e
Q 45 ¼
ð1
/5 Þðh4
h5 Þ
k¼ ð11Þ e s 2e
2 A k s;c e
k pm =ðn4 e
k s;c þ n5 ek pm Þ. Next, the CV4 temperature
Eq. (11) allows the calculation of ½Hþ is obtained from
ðaqÞ  as a function of  
k, namely ½HþðaqÞ  ffi qwa =ðkM H2 O Þ for a dilute water solu- dh4 e4 þ H e ðh3 Þ þ
H e ðh4 Þ þ
cpm
tion. The present model also assumes that the volume ¼ Q HðaqÞ HðaqÞ ð16Þ
ds ~pm n4 ny nz
q
fraction of ionomer in the catalyst layer is approxi-
mately equal to the porosity either in CV3 or CV5. where cpm = cp,f/cpm, and cpm is the polymer electrolyte
The reversible electrical potential at the anode is gi- membrane specific heat.
ven by the Nernst equation [33], The internal ohmic losses are usually dominated by
the low electrolyte conductivity. However in order to
RT 3 study of the effect of varying the thickness of the react-
V e;a ¼ V e;a
ln Q3 ð12Þ
nF ing layers, the model also accounts for the potential loss
where Ve,a = DG3/(
nF) and V e;a ¼ DG3 =ð
nF Þ. At the due to the electrical resistance posed by the ionomer
anode there are two mechanisms for potential losses: within the porous reaction layers present in CV3 and
(i) charge transfer, and (ii) mass diffusion. The potential CV5. The ionic conductivity, r (X
1 m
1), of Nafion
loss due to charge transfer is obtained implicitly from 117 as a function of temperature is given by the empiri-
the Butler–Volmer equation for a given current I [35,36] cal formula [9]
2 3 

ð1
aa Þga F aa ga F 1 1
I
ri ðhÞ ¼ exp 1268
ð0.5139ki
0.326Þ;
6 RT 3 7 303 hi T 1
¼ io;a 4e
e RT 3 5 ð13Þ
A3;wet
i ¼ 3; 4; 5 ð17Þ
where ga is the potential loss at the anode, aa is the Based on the electrical conductivities and geometry
anode charge transfer coefficient, and io,a is the anode of each compartment, the electrical resistances, b (X)
exchange current density (a function of catalyst type, e s V 1=3 ri ð1
/i Þ, i = 1, 2, 6, 7, and
are given by bi ¼ ni = A T
4418 J.V.C. Vargas et al. / International Journal of Heat and Mass Transfer 48 (2005) 4410–4427

e s V 1=3 ri /i , i = 3, 4, 5 (/4 = 1). Therefore, the

bi ¼ ni = A The potential loss due to mass diffusion is calculated
T
total dimensionless ohmic loss in the space from CV1 based on Eq. (14) by using gd,c, T5, A5,wet and iLim,c, in
to CV7 is place of gd,a, T3, A3,wet and iLim,a, respectively. The limit-
X
7 ing current density at the cathode (iLim,c), is calculated
I
gohm ¼
~ bi ð18Þ using the same reasoning as for CV3, based on Eq. (6),
V ref i¼1 such that iLim,c = 2Poxp1D6nF/(MO2L6Roxh6T1). Final-
The conductivities of the catalyst layers are given by ly, the dimensionless electrical potential at the cathode
r3/3 and r5/5 [cf. Eq. (17)], and agree qualitatively with is Ve i;c ¼ Ve e;c
~gc
j~ gd;c j.
previously measured catalyst layers ionic conductivities The mass balance for CV6 yields m_ O2;out ¼ m_ O2;in ¼ m_ O2
[37], i.e., the ionic conductivity increases with increasing and n_ H2 O ¼ n_ H2 O;out ¼ n_ H2 O;in ¼ n_ O2 . The dimensionless
net heat transfer rate in CV6 results from Q e 6 ¼
Q e 56 þ
Nafion content, which increases as / increases in the _Qw6 þ Q e 67 þ Q e 6ohm , with Q e 67 ¼ ~ e
present model. Additionally, as a quantitative example, h7 A s ð1
/6 Þðh7
h6 Þ,
the product r/ leads to a resistivity of 40.29 X cm, for ~7 ¼ h7 V 2=3 =ðm_ ref cp;f Þ. The dimensionless temperature
h T
k = 14, / = 0.2 and T = 80 C, which agrees with a pre- for CV6 is given by
viously reported catalyst layer electronic resistivity of  
dh6 e 6 þ wO cp;ox ðh7
h6 Þ þ H e ðh5 Þ e ðh6 Þ
40 X cm [38]. The conductivities of the diffusive layers, ¼ Q H2 O
H H2 O
ds 2
cp;f
r2 and r6, are the carbon-phase conductivities [12]. Fi-
cs;c
nally, the conductivities of CV1 and CV7, r1 and r7,  ð20Þ
are the electrical conductivities of the bipolar plates ~s;c ð1
/6 Þn6 ny nz
q
material. The void fractions of CV1 and CV7 are ac- The dimensionless net heat transfer rate in CV7 is
counted for, i.e., /1 = /7 = nc/(nt + nc), computed for Qe 7 ¼
Q e 67 þ Q e w7 þ Qe 7ohm . The balances for mass and
any particular fuel cell internal geometry according to energy in the oxidant channel (CV7), the assumptions
Fig. 2. of non-mixing incompressible flow, and the assumption
The analysis in the cathode reaction layer (CV5) is that the space is filled mainly with dry oxygen, yield
analogous to what we presented for the anode reaction m_ H2 O ¼ m_ H2 O;in ¼ m_ H2 O;out ¼ n_ O2 M H2 O and
layer (CV3). This time the reaction equation is Eq.  
(15). CV5 is also divided into two compartments, fluid dh7 e 7 þ wox cp;ox ðhox
h7 Þ þ H e ðh6 Þ e ðh7 Þ
¼ Q H2 O
H H2 O
and solid, but in the thermal analysis only the solid is ds cp;f
taken into account. The dimensionless net heat transfer
e 5 ¼
Q e 45 þ Q e w5 þ Q e 56 þ Q
e 5ohm , Rox h7;0 cox
in CV5 is given by Q  ð21Þ
e ~ e
with Q 56 ¼
k s;c ð1
/6 Þ A s ðh5
h6 Þ=½ðn5 þ n6 Þ=2. Rf P ox nc n7 nc nz
The mass and energy balances in CV5, together with where h7,0 is a specified initial condition, and wox ¼
Eq. (15) deliver n_ Hþ ¼ 2n_ O2 , n_ H2 O;out ¼ n_ O2 , and f7 wO2 .
in

dh5 h e e5
e 5 þ DG
i cs;c
¼ Q5
DH ð19Þ
ds ~s;c ð1
/5 Þn5 ny nz
q
3. Constructal design
where ðD H e 5; DG e 5 Þ ¼ n_ O ðDH 5 ; DG5 Þ=ðm_ ref cp;f T 1 Þ. The
2
enthalpy
P change during
P cathode reaction is DH 5 ¼ The model presented in Section 2 allows the calcula-
products ½mi H i ðT i Þ
reactants ½mi H i ðT i Þ, while We5 = tion of the electrical power output of a single PEMFC

DG5. Similarly to the anode analysis, the molar enthal- when its geometry and operating parameters change.
pies of formation, Hi(Ti), are obtained from tabulated The model accounts for internal temperature and pres-
values at T6 for O2(g), T4 for Hþ ðaqÞ and T5 for H2O(l) sure gradients and potential losses. An important step
at 1 atm. The change in the Gibbs free energy DG5 in constructal design is the identification of realistic
for the reaction of Eq. (15) is calculated by using Eq. design constraints. As it was stated before Eq. (2), the
(10). The CV5 reaction quotient is therefore Q5 ¼ total PEMFC stack volume, VT, is fixed in the optimiza-
2 1=2
1
f½Hþ ðaqÞ  p O2 g , where p O2 ¼ p 6;out . The reversible electri- tion process. In dimensionless terms, the volume con-
cal potential at the cathode results from Eq. (12) after straint reads as nxnynz = 1. According to Fig. 1, in the
using Ve,c, V e;c , DG5, DG5 , T5 and Q5 in place of Ve,a, PEMFC stack, the number of single cells is given by
V e;a , DG3, DG3 , T3 and Q3. ns = nx/ns, where ns ¼ Ls =V T1=3 , and Ls is the total length
The analysis for the cathode reaction layer (CV5) fol- (or thickness) of a single cell, as it is shown in Fig. 2.
lows the same path as for the anode reaction layer (CV3). Proceeding the analysis, it is necessary to account for
The potential losses are due to charge transfer and mass the pressure drops in the gas channels of the single cells,
diffusion. The potential loss due to charge transfer is and also in the input and output headers, as shown sche-
obtained through Eq. (13), using A5,wet, io,c, ac, gc and matically in Fig. 3. The calculation of the total pressure
T5, in place of A3,wet, io,a, aa, ga and T3, respectively. drops in the fuel cell stack will lead to the required total
 J.V.C. Vargas et al. / International Journal of Heat and Mass Transfer 48 (2005) 4410–4427 4419

pumping power to supply the stack with fuel and oxi- and bending-expansion coefficients, respectively. Accord-
dant at any current level. ing to mass conservation, the flow dimensionless
The stoichiometric ratios in the fuel and oxidant mean velocities in the gas channels are ~ u1 ¼ Ch1 ðwf

channels, f1 and f7, are assumed known as specified e c1 P f Þ, ~
wH2 =2Þ=ð A e c7 P ox Þ,
u7 ¼ Rox Ch7 ðwox
wO2 =2Þ=ðRf A
operating parameters. Therefore, the inlet mass flow 1=2 2=3 e
where C ¼ ðRf T 1 Þ m_ ref =ðp1 V T Þ, and A ci ¼ nc Lc Li =
rates are wf ¼ f1 wH2 and wox ¼ f7 wO2 . V 2=3
T , i = 1, 7, is the dimensionless total duct cross-sec-
The pressure drops in the gas channels, from the in- tion area in the fuel and oxidant channels, respectively.
put headers to the output headers, are given by the pres- The model also requires the evaluation of the friction
sure drops in a single cell only, because they are in factor and heat transfer coefficients in the gas channels.
parallel (Fig. 3). The resulting dimensionless expression For the laminar regime (ReDh < 2300) we used the corre-
for rectangular shaped gas channels, accounting for the lations [39]
flow directional change (90 at the cell inlet and outlet),
contraction (inlet) and expansion (outlet), together with fi ReDh;i ¼ 24ð1
1.3553di þ 1.9467d2i
1.7012d3i
the ideal gas model, is the following: þ 0.9564d4i
0.2537d5i Þ ð23Þ

 
nz nz K bc K be P j Rf 2 hi Dh;i
DP i ¼ fi þ þ þ ~u ð22Þ ¼ 7.541ð1
2.610di þ 4.970d2i
5.119d3i
ni nc 2 2 hi Rj i ki
þ 2.702d4i
0.548d5i Þ ð24Þ
where i = 1, 7 and j = f, ox, respectively. Here u~i ¼
ð~
ui;in þ ~ui;out Þ=2 is the gas dimensionless mean velocity where di = Lc/Li, for Lc 6 Li and di = Li/Lc, for Lc > Li;
in the channel, defined as ~u ¼ u=ðRf T 1 Þ1=2 , f is the fric- Dh,i = 2LcLi/(Lc + Li), ReDh;i ¼ ui Dh;i qi =li and i = 1, 7.
tion factor, Kbc and Kbe are the bending-contraction The correlations used for the turbulent regime were [40]

headers cross section

Lh

m· ox ,in m· f ,in
Lv

m· f

i=1 Ls
m· ox

m· f
2
m· ox

m· f
m· ox 3

m· f

Lx

m· ox

m· f
· ns −1
m ox

m· f
ns
m· ox

Lz
m· ox ,out + m· wa m· f ,out

Fig. 3. Flow structure sketch of headers and gas channels in a PEMFC stack.
4420 J.V.C. Vargas et al. / International Journal of Heat and Mass Transfer 48 (2005) 4410–4427

fi ¼ 0.079Re
1=4
Dh ð2300 < ReDh;i < 2  104 Þ to ns, and j = H2 or O2, with w0,k = wf,in or wox,in, for
ð25Þ the input headers, and for the output headers, wi,k =
fi ¼ 0.046Re
1=5
Dh ð2  104 < ReDh;i < 106 Þ wi
1,k + (fg
1)wj, for i = 1 to ns, g = 1 or 7, and
j = f, ox, accordingly, with w0,k = 0. The water mass
hi Dh;i ðfi =2ÞðReDh;i
103 ÞPri flow rate at each single cell level ‘‘i’’ in the output oxi-
¼
ki 1 þ 12.7ðfi =2Þ1=2 ðPr2=3
1Þ dant header is wwa;i ¼ wwa;i
1 þ wH2 O , for i = 1 to ns, with
wwa,0 = 0. Eqs. (22)–(27) deliver the pressure drops from
ð2300 < ReDh;i < 5  106 Þ ð26Þ the input to the output headers, DP1 and DP7, and the
where Pr is the gas Prandtl number, lcp/k. Next, the pressure drops in the input and output headers at each
pressure drops in the input and output headers are cal- cell level, DPi,k and DPwa,i, for each tested geometry dur-
culated according to the schematic diagram presented ing the constructal process.
in Fig. 3. In the input headers, the inlet fuel and oxidant According to Fig. 2, the architecture of the single fuel
mass flow rates, wf,in = nswf and wox,in = nswox, supply cell is determined
P completely by the internal structure
each single cell level with fuel and oxidant mass flow ðLs ¼ 2Lb þ 7i¼1 Li Þ. The fuel cell stack external shape
rates, wf and wox, respectively. In the output headers, is given by (Lx, Ly, Lz) for a constrained total volume,
each single cell delivers depleted fuel and oxidant mass VT. The optimization objective is to determine the entire
flow rates, and the water mass flow rate produced by system architecture: the optimal volume allocation, such
the electrochemical reaction, (f1
1)wf, (f7
1)wox, that the PEMFC stack total net power is maximized.
and wH2 O , respectively. Therefore, the fuel, oxidant and The determination of the total stack net power starts
water mass flow rates in the headers vary according to with the single PEMFC polarization curve, i.e., the fuel
each single cell level, and so do the flow mean velocities. cell total potential as a function of current, and the total
As a result, the fuel and oxidant pressure drops in each net power (available for utilization) of a single fuel cell
segment of the headers adjacent to a single cell level ‘‘i’’, in the stack are given by
and the water flow pressure drop in the oxidant output Ve i ¼ Ve i;a þ Ve i;c
~
gohm ; e net;s ¼ W
W es
W
ep ð28Þ
header are calculated by

 where We s ¼ Ve i eI , and We p ¼ wf S f h1 DP 1 =P 1 þ
ns ns P j Rf 2
DP i;k ¼ fi;k þ ~u wox S ox h7 DP 7 =P 7 , with S j ¼ m_ ref T 1 Rj =ðV ref I ref Þ; j ¼
nv nh hi.k Rj i;k

 ð27Þ f; ox.
ns ns 2 Additional power is required to pump fuel, oxidant,
DP wa;i ¼ fi þ q~ ~u
nv nh wa wa;i and produced water in and out the headers shown in
where k = hf,in, hox,in, hf,out, hox,out stand for input fuel (f) Fig. 3. The dimensionless pumping power required for
and oxidant (ox) headers, output fuel and oxidant head- the gases at the fuel cell level i, and for the produced
ers, respectively. For Eq. (27), in the input headers the water out of the oxidant output header are given by
temperatures are the known inlet fuel and oxidant tem-  
peratures, and in the output headers, the temperatures e i;k ¼ S j whDP ; j ¼ f; ox;
W
P i;k
are the temperatures of the fuel and oxidant exiting ð29Þ
the single cell, calculated by the model, i.e., h1 and h7, e S f
W wa;i ¼ ½wDP wa;i
respectively. Here ~u is the dimensionless mean velocity, q
~wa
f is the friction factor calculated by Eq. (23), with
di = Lh/Lv, for Lh 6 Lv and di = Lv/Lh, for Lh > Lv; The total PEMFC stack power available for utiliza-
Dh,i = 2LhLv/(Lh + Lv), or by Eq. (25), for the laminar tion is therefore obtained from
or turbulent flow regimes, respectively. According to XX
ns X
ns
e net ¼ ns ð W
W es
W
e pÞ
e i;k

W e wa;i
W ð30Þ
mass conservation, the dimensionless gas mean veloci-
k i¼1 i¼1
ties in the headers at each single cell level ‘‘i’’ in the stack
are given by ~ui;k ¼ C½Rhw=ð/P Þi;k =ðRf A e ch Þ, and for where k = hf,in, hox,in, hf,out, hox,out. The objective func-
the water in the oxidant output header, ~uwa;i ¼ tion defined by Eq. (30) depends on the internal struc-
Cwwa;i ½~ e ch , where k = hf,in, hox,in, hf,out,
qwa ð1
/i;hox;out Þ A ture and thickness of each single cell, and on the
hox,out represent the input fuel and oxidant headers, out- external shape of the PEMFC stack. The mathematical
put fuel and oxidant headers; A e ch ¼ Lh Lv =V 2=3 , is the model allows the computation of the total net power
T
dimensionless cross-section area in the fuel and oxidant of the PEMFC stack, W e net . This is possible to achieve
headers; /i,k = 1 for all headers, except in the oxidant as soon as the physical values (Table
P 1) and a set of geo-
output header, when /i,k < 1 since oxidant and water metric internal (1 ¼ 2nb =ns þ 7i¼1 ni =ns and ns) and
flow in the header. external (ny/nx and nz/nx) parameters are chosen for
Finally, the gases dimensionless mass flow rates at the overall system. The maximum (theoretical) PEMFC
each single cell level ‘‘i’’ are wi,k = wi
1,k
wj, for i = 1 stack efficiency is given by
 J.V.C. Vargas et al. / International Journal of Heat and Mass Transfer 48 (2005) 4410–4427 4421

Table 1 3. Compute nx, ny, nz.

Physical properties used as reference case in the numerical 4. For Ae s ¼ ny nz , vary the thickness of a single fuel cell,
optimization of the PEMFC stack ns within a pre-specified range.
B = 0.156 [43] tref = 10
3 s 5. For each ns, find the optimal internal structure that
cp,f = 14.95 kJ kg
1 K
1 Tf, Tox, T1 = 298.15 K delivers a maximum value for W e net .
cpm = 4.18 kJ kg
1 K
1 TLim = 373.15 K 6. The result of the process of step 5 is ns,opt for W e net;m ,
cp,ox = 0.91875 kJ kg
1 K
1 Uwi = 50 W m
2 K
1, i = 1,7 which is the maximum value of W e net found for all
cv,f = 10.8 kJ kg
1 K
1 Vref = 1 V tested ns in step 5.
cv,ox = 0.659375 kJ kg
1 K
1 VT,ref = 10
5 m3
7. Compute the resulting number of single cells in the
i0,a, i0,c = 10 A m
2 aa, ac = 0.5 nx n
Iref = 1 A l1 = 10
5 Pa s
stack: ns;opt ¼ ns;opt , increment nyx ¼ nnxz and return to step
n
kf = 0.2 W m
1 K
1 l7 = 2.4 · 10
5 Pa s 2 until the entire nyx ¼ nnxz pre-specified range is covered.
kox = 0.033 W m
1 K
1 nc/ny = 0.005 8. The result of the outer loop (2–7) is the optimized
kpm = 0.21 W m
1 K
1 nh = nv = 0.022 configuration of the PEMFC stack for maximum
ks,a, ks,c = 71.6 W m
1 K
1 nt/ny = 0.05
net power density that is possible to be obtained in
Kbc, Kbe = 0.9 qwa = 1000 kg m
3
K2, K6 = 4 · 10
14 m2 r1, r7 = 1.388 · 106 X
1 m
1
a constrained total volume, VT, given by Eq. (30):
K3, K5 = 4 · 10
16 m2 r2, r6 = 4000 X
1 m
1
 
ny nz ns
pf, p1 = 0.1 MPa /hox;out ¼ 0.5 ¼ ; )W e net;mm
nx nx ns opt
pox = 0.12 MPa /2, /6 = 0.4
q = 1.5 /3, /5 = 0.2
Rf = 4.157 kJ kg
1 K
1
The optimization of internal structure mentioned in
Rox = 0.2598 kJ kg
1 K
1 step 5 of the optimization algorithm is executed follow-
ing the procedure introduced by Vargas et al. [28] for a
single PEMFC. First, the anode and cathode are as-
sumed to have the same thickness. The thicknesses of
e 3 þ DG
DG e5
gi ¼ ð31Þ the diffusion and reaction layers of the cathode and
e e5
DH 3 þ DH anode are varied simultaneously subject to fixed ratios
The actual first-law efficiency of the fuel cell stack is of cathode thickness (y2) and anode thickness (y6) to to-
tal single cell length, i.e., n2/ns + n3/ns = y2, n5/ns + n6/
es
W ns = y6, where y4 is the ratio of membrane thickness to
gI ¼ E ð32Þ
e 3 þ DH
DH e5 total single cell length. The ratio of overall thickness
to total length of the single fuel cell is also fixed,
where E ¼ V ref I ref =ðm_ ref cp;f T 1 Þ. The second-law effi-
y2 + y4 + y6 = 0.8. Under the simplifying hypotheses as-
ciency is defined as the ratio of the actual electrical
sumed, the internal structure optimization problem is re-
power to the reversible electrical power,
duced to one degree of freedom, i.e., the ratio n3/ns = n5/
Wes ns. The end result is the optimized configuration of the
gII ¼ E ð33Þ
e
DG3 þ DG e5 single fuel cell electrodes (n2, n3, n5, n6)opt, for which the
PEMFC stack net power, given by Eq. (30) is maximum.
The net efficiency of the PEMFC stack is
The physical properties assumed in this study were
We net suggested by previous studies and obtained from hand-
gnet ¼ E ð34Þ
e e 5Þ
ns ðD H 3 þ D H books [9,11,12,35,41], and are listed in Table 1. Eqs.
(4), (6), (7), (9), (16), (19)–(21), and the specified initial
In previous studies [27,28] we found the optimal dis- conditions form a system of seven ordinary differential
tribution of the compartments shown in Fig. 2 for fuel equations and two algebraic equations. The unknowns
cell maximum power under a volume constraint, for are hi and Pi, i.e., the temperatures in the seven control
an alkaline fuel cell, and for a single PEMFC where volumes, and the gas pressures in CV2 and CV6. Once
the external shape of the cell was allowed to vary. In this the temperatures and pressures are known, the electrical
study, the constructal procedure seeks the PEMFC stack potentials, electrical power, and PEMFC stack net
optimal internal structure, length (or thickness) of the power can be calculated for any current level.
single fuel cell, and external shape (ny/nx and nz/nx) Two numerical methods are used. The first method
based on the general configuration presented in Fig. 1, calculates the transient behavior of the system, starting
and according to the following algorithm: from a set of initial conditions, then the solution is
marched in time (and checked for accuracy) until a stea-
1. Fix total PEMFC stack volume, Ve T , and assume a dy state is achieved at any current level. The equations
n
fuel cell square section, i.e., nyx ¼ nnxz . are integrated in time explicitly using an adaptive time
ny
2. Select the lowest value for nx
¼ nnxz within a pre-speci- step, fourth to fifth order Runge–Kutta method [42].
fied range of variation. The time step is adjusted automatically according to
4422 J.V.C. Vargas et al. / International Journal of Heat and Mass Transfer 48 (2005) 4410–4427

the local truncation error, which is kept below a speci-

( ) ( )
fied tolerance of 10
6. The second method is for the
( )
steady state solution. The time derivatives are dropped ( )
from Eqs. (4), (7), (9), (16), (19)–(21). Pressures are re-
lated to temperatures via Eq. (6). The system reduces
to seven nonlinear algebraic equations, in which the un-
knowns are the temperatures of the seven control vol-
umes. This system is solved using a quasi-Newton
method [42]. Convergence was achieved when the Euclid-
ean norm of the residual of the system was less than 10
6.

4. Results and discussion

In the procedure described by the algorithm pre- a

sented in Section 3, the PEMFC stack net power is cal-
culated by starting from open circuit ðeI ¼ 0Þ and ( ) ( )
proceeding in pre-specified increments DeI (e.g., 2, 5, ( )
10) until the net power is zero, the membrane limiting ( )
operating temperature is achieved, TLim, or the limiting
current level is reached. This procedure is illustrated in
Fig. 4, which shows simulation results for one selected
PEMFC stack external shape (ny/nx = nz/nx = 0.3),
which has been optimized according to step 5 of the
algorithm of Section 3, i.e., an optimal internal structure
and single cell thickness (ns = 0.0044) was found for that
particular external shape. The resulting electrical power,
and polarization curves of a single cell, and the PEMFC
stack net power curves are shown in Fig. 4a. According
to the model, the actual open circuit voltage is equal to b
the reversible cell potential, because it has been assumed
that no losses result from species crossover from one ( ) ( )
electrode through the electrolyte, and from internal cur- ( )
rents. The change in the Gibbs free energy of reaction ( )
decreases as the temperature increases. Therefore,
according to Eq. (12) the reversible electrical potential
decreases as the temperature increases, which happens
when the current increases.
In Fig. 4a, the simulation stopped when the temper-
ature at any compartment of the single fuel cell reached
the limiting membrane operating temperature, i.e.,
hi 6 hLim, and at this point W e net > 0 for the selected con-
figuration under analysis. The other ending criterion is
We net ¼ 0, i.e., up to the point where the electrical power
c
produced by the PEMFC stack matched the required
pumping power to supply fuel and oxidant to the stack Fig. 4. (a) Example of single PEMFC polarization, total
at any specified stoichiometric ratios f1 = f7. Under such electrical power and PEMFC stack net power output curves;
selected operating conditions and geometry (ny/nx = nz/ (b) the ideal, first law, second law and net efficiencies for the
nx = 0.3), the PEMFC stack did not reach the concen- PEMFC stack considered in (a), and (c) the behavior of
tration polarization region as Ve e;a or Ve e;c would ap- temperature versus flow length for several current levels, in a
single PEMFC for the stack considered in (a) and (b).
proach zero. The net power curve exhibits a maximum
at eI  250, which is central in the evaluation of global
performance, by balancing total electrical power pro- procedure described by the algorithm presented in
duced with required pumping power to supply fuel Section 3.
and oxidant to the PEMFC stack. The maximum net Fig. 4b shows that the ideal efficiency (gi) decreases
power is the quantity maximized during the constructal as the current increases. This effect is due to the
 J.V.C. Vargas et al. / International Journal of Heat and Mass Transfer 48 (2005) 4410–4427 4423

temperature increase in the anode and cathode reaction

layers, which is captured by the present model. The first ( ) ( )
law efficiency is equal to the ideal efficiency at open cir- ( )
cuit (eI ¼ 0), and decreases monotonically as the current ( )
increases. The second law efficiency is equal to 1 (revers-
ible operation, no losses) at open circuit ðeI ¼ 0Þ, and
also decreases monotonically as the current increases.
The net efficiency behavior shows the effect of increasing
current and therefore pressure drops in the single cells
gas channels and fuel and oxidant input and output
headers, i.e., the pumping power increases and the net
power decreases, eventually reaching a zero limit when
We net ¼ 0.
The fuel flow rate increases as the current increases,
Eq. (1), therefore more heat is generated by the reactions
at the anode and cathode, increasing the temperature. a
The temperatures in the single fuel cell compartments
shown in Fig. 2 increase as the current increases, as ( ) ( )
shown in Fig. 4c, because more heat is generated by ( )
the electrochemical reaction and by Joule effect (ohmic ( )
heating). The higher the current, the more accentuated
are the temperature spatial gradients between the single
fuel cell compartments, even for the single fuel cell con-
sidered in Fig. 4, with the selected high geometric aspect
ratio ny, nz  ns,opt (ny/ns = nz/ns = 152), i.e., with a
small thickness compared to width and height. There-
fore, the PEMFC stack net power and polarization
curves produced by the present model take into account
all the internal spatial temperature gradients, which af-
fect the polarization curve and the total electrical power
output. Most importantly, the spatial temperature distri-
bution results of Fig. 4c show that, although a cooling
b
system is considered across the bipolar plates with an
externally controlled and monitored average tempera- Fig. 5. (a) The transient response of the internal compartments
ture T1, the actual single cell internal temperature is temperatures in a single PEMFC for the stack considered in
substantially higher than that. This finding should be Fig. 4 and (b) the transient response of the PEMFC stack net
considered in fuel cell design. power for the stack considered in Fig. 4.
Fig. 5 shows the start-up transient—the evolution of
the temperatures and net power until steady state is
reached at a high current level (eI ¼ 250) required by (ka, kc) = (16, 18), and Ve T ¼ 1.16. The graphical results
the load. In Fig. 5a, there is a considerable temperature follow the steps of the algorithm presented in Section
spatial gradient between the internal compartments of 3. Fig. 6a shows the results of steps 4, 5 and 6, which
the single fuel cell, and it affects fuel cell performance. identify ns,opt for several PEMFC stack external shapes,
Hydrogen is about ten times more thermally conductive ny/nx = nz/nx. For each value of ns, a corresponding fuel
than oxygen, therefore the fuel gas channel temperature cell optimal internal structure is found, but this is not
is higher than the oxidant gas channel temperature, i.e., shown in Fig. 6 for the sake of brevity. However, the
with a better thermal contact with the other internal bulk of results of this study for the optimized internal
compartments. The effect on PEMFC stack net power structure corroborate the results discussed and docu-
is documented in Fig. 5b. The net power increases mented in the constructal optimization of a single
monotonically as the internal cell temperatures increase, PEMFC [28], i.e., n3/ns = n5/ns ffi 0.01. The optimal
eventually reaching a plateau at steady state operation. single cell thickness ns,opt results from the trade-off
A transient model is therefore mandatory if one is to between two effects: activation polarization losses and
evaluate fuel cell performance for simulations of appli- ohmic losses. As ns increases, the electrode wetted areas
cations with variable load. increase, thus the activation losses decrease. On the
Fig. 6 illustrates the constructal procedure for one set other hand, the ohmic losses increase as thickness
of external parameters, i.e., Table 1, f1 = f7 = 2, increases. Another interesting phenomenon is that the
4424 J.V.C. Vargas et al. / International Journal of Heat and Mass Transfer 48 (2005) 4410–4427

ment the final steps of the optimization algorithm pre-

sented in Section 3, pinpointing the optimal single cell
( ) ( ) thickness and stack external shape for maximum net
power output. The maximum is sharp, therefore impor-
tant in fuel cell design. The existence of a PEMFC stack
net power maximum with respect to stack external shape
is explained by analyzing two extremes: (i) small ny/
nx = nz/nx implies that nx is large, W e s is small due to
large flow resistances in the x-direction, the number of
single cells ns is high, therefore the resulting total stack
electrical power is still high, but the headers pumping
power increase as nx increases, therefore W e net;m ! 0,
and (ii) large ny/nx = nz/nx implies that nx is small, W es
is large due to small flow resistances in the x-direction
a and also large wetted areas at the electrodes, the headers
pumping power decrease as nx decreases, but W e p is also
large due to a large swept length nz and small hydraulic
( ) ( )

( ) ( )

Fig. 6. (a) The internal structure and single cell thickness

optimization according to Fig. 2 and the dependence on
external shape of the PEMFC stack, and (b) the existence of
a maximum PEMFC stack net power and the corresponding a
optimized single cell thickness with respect to external shape of
the stack.
( ) ( )

ionomer electrical conductivity increases as tempera-

ture increases at higher current levels, and with water
content increase as well, therefore reducing ohmic loss
)

at higher temperatures, which is represented by Eq.

(17). The results of those trade-offs observed during sin-
gle cell thickness variation in the optimization process is
shown in Fig. 6a, which shows net power maxima for ( )
)

five PEMFC stack external geometric aspect ratios.

Note that the single cell thickness is optimized for the
maximum PEMFC stack net power, i.e., the optimiza-
tion of the component shape for maximum performance
of the global system. The maxima are sharp, stressing
the importance of these optima found in future fuel cell b
design. Fig. 7. (a) The external shape optimization and the dependence
The results of Fig. 6a are summarized in Fig. 6b, on stoichiometric ratio for (ka, kc) = (16, 18) and Ve T ¼ 1.16,
which also shows the existence of a second maximum and (b) the results of the external shape optimization with
for the PEMFC stack net power, this time with respect respect to the stoichiometric ratios, f1 = f7, for (ka, kc) = (16, 18)
to external geometric aspect ratios. The curves docu- and Ve T ¼ 1.16.
 J.V.C. Vargas et al. / International Journal of Heat and Mass Transfer 48 (2005) 4410–4427 4425

diameters Dh, therefore W e net;m ! 0 in this extreme as channels increase, and therefore pressure drops and
well. In conclusion, there must exist an intermediate pumping power increase. The maxima are sharp, mainly
and optimal ny/nx = nz/nx geometric configuration such for lower stoichiometric ratios. Fig. 7b shows the twice
that W e net;m is maximum, which balances the trade-off maximized stack net power and the corresponding opti-
between electrical power output and pumping power mal external shapes as functions of the stoichiometric
to supply fuel and oxidant to the PEMFC stack, accord- ratios. The stack net power W e net;mm decreases monoton-
ing to Eq. (30). ically as the stoichiometric ratios increase. The reason
In a subsequent phase of the study, a sensitivity anal- for that is that as the stoichiometric ratios increase the
ysis was conducted to find the dependence of the optimi- pumping power increases as well. Most important is
zation results on the variation of three operating and the observation that the optimal external geometry is ro-
design parameters, i.e., f1 = f7, (ka, kc), and Ve T . The bust with respect to changes in the stoichiometric ratios,
PEMFC stack net power is plotted in Fig. 7a as a func- i.e., for 1.5 6 f1 = f7 6 3, the optimal external shape
tion of ny/nx = nz/nx for three different stoichiometric varied only slightly, i.e., 0.05 6 (ny/nx = nz/nx)opt 6 0.08.
ratios regimes, f1 = f7. For all stoichiometric ratios, a Next, the effect of the variation of membrane water
twice maximized stack net power, W e net;mm , is observed, content, k = (ka + kc)/2, based on specified pairs (ka, kc)
determining the optimal external PEMFC stack struc- was investigated. The results of Fig. 8a show stack net
ture, represented by (ny/nx = nz/nx)opt. As the stoichiom- power maxima for three different combinations of
etric ratios increase, the net power decreases because the (ka, kc). As the water content increases, the twice maxi-
fuel and oxidant mass flow rates in the headers and gas mized net power also increases, which is explained by

( ) ( )

a
a

( ) ( )
)

)
(

( ) )
(
)

b
b
Fig. 8. (a) The external shape optimization and the dependence
on average membrane water content for f1 = f7 = 2 and Fig. 9. (a) The external shape optimization and the dependence
Ve T ¼ 1.16, and (b) the results of the external shape optimiza- on total stack volume for (ka, kc) = (16, 18) and f1 = f7 = 2, and
tion with respect to average membrane water content, k, (b) the results of the external shape optimization with respect to
f1 = f7 = 2 and Ve T ¼ 1.16. total stack volume, Ve T , for (ka, kc) = (16, 18) and f1 = f7 = 2.
4426 J.V.C. Vargas et al. / International Journal of Heat and Mass Transfer 48 (2005) 4410–4427

higher membrane electrical conductivities and smaller stack external shape was shown to be ‘‘robust’’ with
ohmic losses. Fig. 8b summarizes the optimization re- respect to the analyzed parameters, i.e., 1.5 6 f1 = f7 6 3,
sults showing the twice maximized net power and the 12 6 k 6 21, and 0.1 6 Ve T 6 1.16, and the twice maxi-
corresponding optimized stack external shapes. The mized stack net power increases monotonically with total
optimized external shape is also robust with respect to volume, i.e., W e net;mm increases approximately as Ve T3=4 ,
the variation of membrane water content. similarly to metabolic rate and body size in animals [8].
The investigation of the variation of the volume con- This is an important finding for the purpose of ‘‘scaling
straint is conducted in Fig. 9, where Ve T was varied over up’’ or ‘‘scaling down’’ PEMFC stack design.
the range 0.1–1.16, for stoichiometric ratios f1 = f7 = 2. Fundamentally, it was shown that electrical and fluid
This variation is indicated by the three curves of Fig. 9a. flow trade-offs exist, and that from them results the sin-
All curves exhibit PEMFC stack maxima with respect to gle cell internal structure and total thickness, and the
the external shape aspect ratios. The fuel cell stack net PEMFC stack external shape—the relative sizes and
power increases as Ve T increases. Additional results were spacings—of flow systems, in accordance with other fea-
produced to cover the entire range 0.1 6 Ve T 6 1.16, tures of constructal design [8,27,28]. In practice, such
which allowed the twice maximized stack net power trade-offs must be pursued based on models that corre-
and the optimized external shape aspect ratios to be spond to real applications. The maxima found are sharp,
plotted in Fig. 9b as functions of Ve T . The twice maxi- stressing their importance for practical design, and
mized stack net power increases monotonically with therefore must be identified accurately in the quest for
total volume, i.e., We net;mm increases approximately as increasing the stack net power efficiency, approaching
3=4
e
V T . The optimized external shape aspect ratios are the actual PEMFC first-law efficiency level. The con-
robust with respect to the fuel cell stack total vol- structal design results reported in this study demonstrate
ume, staying approximately within the range 0.03 6 clearly that in a PEMFC stack, gas supply causes pres-
(ny/nx = nz/nx)opt 6 0.07. sure drops that induce considerable power consumption.
Those facts need to be taken into account by modeling
or experimentally analyzing the entire PEMFC stack,
5. Conclusions in practical fuel cell design.

In this paper, the constructal optimization of a

PEMFC stack was conducted for maximizing the stack Acknowledgements
net power output. The procedure started with the con-
struction of a mathematical model for fluid flow, mass The authors acknowledge with gratitude the support
and heat transfer in a PEMFC stack, which takes into of NASA under the Hydrogen Research for Florida
account spatial temperature and pressure gradients Universities Program, the Center for Advanced Power
in a single PEMFC (Fig. 2), pressure drops in the Systems at Florida State University, the Brazilian Na-
headers and all gas channels in the entire PEMFC stack tional Council for Scientific and Technological Develop-
(Fig. 3). The single PEMFC internal structure has an ment, CNPq, and the State of Parana Araucaria
optimal allocation, and total thickness being such that Foundation, Curitiba, Brazil.
wetted area in the reaction layers and electrical resis-
tance are optimally balanced for maximum electrical
power, and maximum global stack net power. Addition-
References
ally, a three-dimensional flow space with the dimen-
sions Lz and Ly in the plane perpendicular to Lx was [1] National Energy Technology Laboratory, Fuel Cell Hand-
considered and the total volume was fixed (Fig. 1). book, sixth edition, EG&G Technical Services, Inc.,
The new degrees of freedom, i.e., the aspect ratios Ly/ Science Applications International Corporation, DOE/
Lx and Lz/Lx, allowed for the optimization of the NETL-2002/1179, Morgantown, West Virginia, 2002.
PEMFC stack external shape, in addition to the single [2] M.M. Mench, C. Wang, S.T. Thynell, An introduction to
PEMFC internal structure and total thickness. As a re- fuel cells and related transport phenomena, Int. J. Transp.
sult, an external shape was found such that electrical Phenom. 3 (2001) 151–176.
and pumping power are optimally balanced for maxi- [3] K. Kordesch, G. Simander, Fuel Cells and their Applica-
mum PEMFC stack net power. Dimensionless optimiza- tions, VCH Publishers, New York, 1996, ISBN 3-527-
28579-2.
tion results were presented graphically for the sake of
[4] H. Chang, P. Koschany, C. Lim, J. Kim, Materials and
generality. processes for light weight and high power density PEM fuel
A parametric analysis investigated the effect of stoi- cells, J. New Mater. Electrochem. Syst. 3 (1) (2000) 55–59.
chiometric ratio, membrane water content, and total [5] E. Middelman, W. Kout, B. Vogelaar, J. Lenssen, E. de
PEMFC stack volume on the optima found. For the Waal, Bipolar plates for PEM fuel cells, J. Power Sources
set of parameters given by Table 1, the optimal PEMFC 118 (1–2) (2003) 44–46.
 J.V.C. Vargas et al. / International Journal of Heat and Mass Transfer 48 (2005) 4410–4427 4427

[6] O.J. Murphy, A. Cisar, E. Clarke, Low-cost light weight [24] Y.J. Zhang, M.G. Ouyang, Q.C. Lu, J.X. Luo, X.H. Li, A
high power density PEM fuel cell stack, Electrochim. Acta model predicting performance of proton exchange mem-
43 (24) (1998) 3829–3840. brane fuel cell stack thermal systems, Appl. Therm. Eng.
[7] A. Kumar, R.G. Reddy, Materials and design development 24 (4) (2004) 501–513.
for bipolar/end plates in fuel cells, J. Power Sources 129 (1) [25] J.J. Baschuk, X.G. Li, Modelling of polymer electrolyte
(2004) 62–67. membrane fuel cell stacks based on a hydraulic network
[8] A. Bejan, Shape and Structure, from Engineering to approach, Int. J. Energy Res. 28 (8) (2004) 697–724.
Nature, Cambridge University Press, Cambridge, UK, [26] I. Mohamed, N. Jenkins, Proton exchange membrane
2000. (PEM) fuel cell stack configuration using genetic algo-
[9] T.E. Springer, T.A. Zawodzinski, S. Gottesfeld, Polymer rithms, J. Power Sources 131 (1–2) (2004) 142–146.
electrolyte fuel cell model, J. Electrochem. Soc. 138 (8) [27] J.V.C. Vargas, A. Bejan, Thermodynamic optimization of
(1991) 2334–2341. internal structure in a fuel cell, Int. J. Energy Res. 28 (4)
[10] S.F. Baxter, V.S. Battaglia, R.E. White, Methanol fuel cell (2004) 319–339.
model: anode, J. Electrochem. Soc. 146 (1999) 437–447. [28] J.V.C. Vargas, J.C. Ordonez, A. Bejan, Constructal flow
[11] V. Gurau, F. Barbir, H. Liu, An analytical solution of a structure for a PEM fuel cell, Int. J. Heat Mass Transfer 47
half-cell model for PEM fuel cells, J. Electrochem. Soc. 147 (19–20) (2004) 4177–4193.
(2000) 2468–2477. [29] R.B. Bird, W.E. Stewart, E.N. Lightfoot, Transport
[12] A.A. Kulikovsky, J. Divisek, A.A. Kornyshev, Two- Phenomena, second ed., Wiley, New York, 2002.
dimensional simulation of direct methanol fuel cells—a [30] J.S. Newman, Electrochemical Systems, second ed., Pre-
new (embedded) type of current collector, J. Electrochem. ntice Hall, Englewood Cliffs, NJ, 1991, pp. 255, 299, 461.
Soc. 147 (2000) 953–959. [31] J.A. Wesselingh, P. Vonk, G. Kraaijeveld, Exploring the
[13] T. Thampan, S. Malhotra, H. Tang, R. Datta, Modeling Maxwell–Stefan description of ion-exchange, Chem. Eng.
of conductive transport in proton-exchange membranes for J. Biochem. Eng. J. 57 (1995) 75–89.
fuel cells, J. Electrochem. Soc. 147 (2000) 3242–3250. [32] M.J. Moran, R. Shapiro, Fundamentals of Engineering
[14] L. Pisani, G. Murgia, M. Valentini, B. DÕAguanno, A Thermodynamics, third ed., Wiley, New York, 1993.
working model of polymer electrolyte fuel cells—compar- [33] W.L. Masterton, C.N. Hurley, Chemistry Principles and
isons between theory and experiments, J. Electrochem. Reactions, third ed., Saunders College Publishing,
Soc. 149 (7) (2002) A898–A904. Orlando, FL, 1997.
[15] S.M. Senn, D. Poulikakos, Tree network channels as fluid [34] T.A. Zawodzinski, M. Neeman, L.O. Sillerud, S. Gottes-
distributors constructing double-staircase polymer electro- feld, Determination of water diffusion-coefficients in per-
lyte fuel cells, J. Appl. Phys. 96 (1) (2004) 842–852. fluorosulfonate ionomeric membranes, J. Phys. Chem. 95
[16] T. Zhou, H. Liu, A general three-dimensional model for (15) (1991) 6040–6044.
proton exchange membrane fuel cells, Int. J. Transp. [35] J.OÕM. Bockris, D.M. Drazic, Electro-chemical Science,
Phenom. 3 (2001) 177–198. Taylor and Francis, London, 1972.
[17] J.C. Amphlett, R.F. Mann, B.A. Peppley, P.R. Roberge, [36] A.J. Bard, L.R. Faulkner, Electrochemical Methods—
A. Rodrigues, A model predicting transient responses of Fundamentals and Applications, second ed., Wiley, New
proton exchange membrane fuel cells, J. Power Sources 61 York, 2001.
(1–2) (1996) 183–188. [37] G. Li, P.G. Pickup, Ionic conductivity of PEMFC
[18] J. Hamelin, K. Agbossou, A. Laperriere, F. Laurencelle, electrodes, J. Electrochem. Soc. 150 (11) (2003) C745–
T.K. Bose, Dynamic behaviour of a PEM fuel cell stack C752.
for stationary applications, Int. J. Hydrogen Energy 26 (6) [38] A.P. Saab, F.H. Garzon, T.A. Zawodzinski, Determina-
(2001) 625–629. tion of ionic and electronic resistivities in carbon/polyelec-
[19] S.O. Morner, S.A. Klein, Experimental evaluation of the trolyte fuel-cell composite electrodes, J. Electrochem. Soc.
dynamic behavior of an air-breathing fuel cell stack, J. 149 (12) (2002) A1541–A1546.
Solar Energy Eng.—Trans. ASME 123 (3) (2001) 225–231. [39] R.K. Shah, A.L. London, Laminar flow forced convection
[20] H.I. Lee, C.H. Lee, T.Y. Oh, S.G. Choi, I.W. Park, K.K. in ducts. Advances in Heat Transfer, Academic Press, New
Baek, Development of 1 kW class polymer electrolyte York, 1978. (Suppl. 1).
membrane fuel cell power generation system, J. Power [40] A. Bejan, Convection Heat Transfer, third ed., Wiley, New
Sources 107 (1) (2002) 110–119. York, 2004, Chapter 8.
[21] T. Van Nguyen, M.W. Knobbe, A liquid water manage- [41] D.R. Lide (Ed.), CRC Handbook of Chemistry and
ment strategy for PEM fuel cell stacks, J. Power Sources Physics, 83rd ed., CRC Press, Boca Raton, FL, 2002–2003.
114 (1) (2003) 70–79. [42] D. Kincaid, W. Cheney, Numerical Analysis Mathematics
[22] M.W. Knobbe, W. He, P.Y. Chong, T.V. Nguyen, Active of Scientific Computing, first ed., Wadsworth, Belmont,
gas management for PEM fuel cell stacks, J. Power Sources CA, 1991.
138 (1–2) (2004) 94–100. [43] M.R. Tarasevich, A. Sadkowski, E. Yeager, in: B.E.
[23] D. Thirumalai, R.E. White, Steady-state operation of a Convay, J.OÕM. Bockris, E. Yeager, S.U.M. Khan, R.E.
compressor for a proton exchange membrane fuel cell White (Eds.), Comprehensive Treatise of Electrochemistry,
system, J. Appl. Electrochem. 30 (5) (2000) 551–559. 7, Plenum, New York, 1983, pp. 310–398.