Journal of Power Sources 326 (2016) 417e427

Contents lists available at ScienceDirect

Journal of Power Sources

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

Proton exchange membrane fuel cell model for aging predictions:

Simulated equivalent active surface area loss and comparisons with
durability tests

rard a, *, M. Quinaud a, J. d’Arbigny d, Y. Bultel b, c
C. Robin a, b, c, d, M. Ge
a
Univ. Grenoble Alpes, CEA, LITEN, F-38054, Grenoble, France
b
Univ. Grenoble Alpes, LEPMI, F-38000, Grenoble, France
c
CNRS, LEPMI, F-38000, Grenoble, France
d
Zodiac Aerospace, F-78370, Plaisir, France

h i g h l i g h t s

Prediction by simulation of PEM fuel cell lifetime.

Multiscale modelling of catalyst dissolution and global loss of performance.

Reconstruction of active surface area loss by model inversion.

Prediction and validation of current density evolution during aging.

Validation by two 2000-h experimental stack aging tests.

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

Article history: The prediction of Proton Exchange Membrane Fuel Cell (PEMFC) lifetime is one of the major challenges to
Received 17 February 2016 optimize both material properties and dynamic control of the fuel cell system. In this study, by a mul-
Received in revised form tiscale modeling approach, a mechanistic catalyst dissolution model is coupled to a dynamic PEMFC cell
3 July 2016
model to predict the performance loss of the PEMFC. Results are compared to two 2000-h experimental
Accepted 5 July 2016
aging tests. More precisely, an original approach is introduced to estimate the loss of an equivalent active
Available online 15 July 2016
surface area during an aging test. Indeed, when the computed Electrochemical Catalyst Surface Area
profile is fitted on the experimental measures from Cyclic Voltammetry, the computed performance loss
Keywords:
PEM fuel cell
of the PEMFC is underestimated. To be able to predict the performance loss measured by polarization
Durability test curves during the aging test, an equivalent active surface area is obtained by a model inversion. This
Modeling methodology enables to successfully find back the experimental cell voltage decay during time. The
ECSA loss model parameters are fitted from the polarization curves so that they include the global degradation.
Lifetime prediction Moreover, the model captures the aging heterogeneities along the surface of the cell observed experi-
Degradation mentally. Finally, a second 2000-h durability test in dynamic operating conditions validates the
approach.
© 2016 Elsevier B.V. All rights reserved.

1. Introduction
Fuel cells are a promising technology to address energy issues in
various fields. For transportation, Proton Exchange Membrane Fuel
Cells (PEMFC) offer many advantages compared to classic thermal
engines, such as a better efficiency, a good dynamic response, a
* Corresponding author.

rard).
E-mail address: mathias.gerard@cea.fr (M. Ge
cleaner and less noisy energy conversion [1]. However, two main
barriers are still remaining for a widespread commercialization of
PEMFC: high components’ costs and moderate durability [2]. Fuel
cells aging concerns for transportation may be one of the most
challenging issue to be dealt with. Indeed, the strong dynamics of
the power profiles together with constraining operating conditions
are most of the time incompatible with a high lifetime of PEMFC.
Due to the variety of materials used inside a fuel cell, various
degradation mechanisms can occur [3e5]. These mechanisms can
be reversible (temporary performance loss) or irreversible. Among

http://dx.doi.org/10.1016/j.jpowsour.2016.07.018
0378-7753/© 2016 Elsevier B.V. All rights reserved.
418 C. Robin et al. / Journal of Power Sources 326 (2016) 417e427
Nomenclature
Acronyms
BoL Beginning of Life
ECSA Electrochemical Catalyst Surface Area
EoT End of Test
GDL Gas Diffusion Layer
MEA Membrane Electrode Assembly
PEMFC Proton Exchange Membrane Fuel Cell
STEM Scanning Transmission Electron Microscopy
Symbols
Cj concentration of the species j (mol m3)
dnk variation of the platinum particles number of the
population k ()
EGT Gibbs-Thomson energy (J mol1)
F Faraday constant (C mol1)
i local current density (A m2)
Icell fuel cell current (A)
k direct reaction constant (mol m2 s1)
Mj molar mass for the species j (kg mol1)
mj mass for the species j (kg)
n number of exchanged electrons ()
P0 standard pressure (Pa)
Pj partial pressure for the species j (Pa)
r radius (m)
R gas constant ()
Rohm specific ohmic resistance (ohm.m2)
S electrode surface (m2)
T temperature (K)
the latter, the polymer membrane degradation is likely to be the
one mostly responsible for the failure of PEMFC [6]. Mechanical and
thermal membrane degradations could be modeled [7], however
they have little effects on the fuel cell performances evolution
during operation, except at the very end of the cell lifetime when its
permeation current increases dramatically. It is a very sudden
phenomenon which could be correlated also with the Membrane-
Electrode Assembly (MEA) initial fabrication defects and the bipo-
lar plate mechanical constraints. Therefore, the mechanical and
chemical membrane degradations with the estimation of a sudden
failure are beyond the scope of this paper. Another critical degra-
dation mechanism in PEMFC concerns the catalyst layers and
especially the active surface area loss during fuel cell operation. For
instance, the irreversible losses can be due to carbon support
corrosion, catalyst dissolution/redeposition and catalyst layer
micro-porous structure modification [8]. Conversely to the mem-
brane degradation, these mechanisms affect directly the fuel cell
performances with time but not directly a fuel cell failure. Degra-
dation of Gas Diffusion Layers (GDL) and bipolar plates can also be
observed but they seem to be of less value regarding fuel cells
durability [7].
The understanding of degradation mechanisms by modeling has
been conducted intensively with very different approaches [7].
However, the prediction of the fuel cell lifetime under dynamic
cycles is less described even if some studies focus on it. In partic-
ular, Fowler et al. [9] have introduced a parameter in the semi-
empirical electrochemical response of their model to take into ac-
count an Electrochemical Surface Area (ECSA) loss. Soltani et al. [10]
have used some durability tests to build empirical laws giving the
evolution of global parameters (potential, temperature, etc.) with
time. However, this approach is not able to give any physical
t time (s)
Ucell cell voltage (V)
Urev reversible cell voltage (V)
v surface chemical velocity (mol m2 s1)
vdiss dissolution rate of the platinum particle (mol m2 s1)
Greek letters
a reaction transfer coefficient ()
bk semi-empirical electrochemical reaction coefficients
(V or V K1 or U.m2 or m)
DG variation of free Gibbs energy (J)
D Gs variation of free enthalpy required for a platinum atom
extraction (J)
DGelec variation of free enthalpy required for a platinum atom
oxidation (J)
DGdes variation of free enthalpy required for leaching of Pt2þ
(J)
DNk ðtÞ cumulated variation of the particles number in the
population k ()
Dc local potential at the inner layer (V)
g surface tension (J m2)
h over potential (V)
rj density for the species j (kg m3)
s protonic conductivity (S m1)
tdegr degradation rate of the Electrochemical Surface Area
()
Subscripts
a anode
c cathode
m membrane
information neither on the degradations mechanisms that occur
nor on their locations at the cell surface. As an original modeling
approach, Jouin et al. [11e12], have used neural networks to predict
the fuel cell lifetime after a learning phase during the first hundred
hours of a test. Conversely to these approaches, Kulikovsky [13] has
built a purely theoretical model to study catalyst and GDL degra-
dation effects on the fuel cell performance and especially on the
current density distribution. Pohl et al. [14] have used differential
equations to take into account the membrane degradation and the
evolutions of the current density distribution with time. The au-
thors couple this approach with the dual time theory in order to
dissociate the time scales proper to the different degradation
mechanisms. Finally, Dhanushkodi et al. [15] have modeled a
simplified carbon corrosion mechanism which have been success-
fully validated on several Accelerated Stress Tests. In all these
studies however, the simulated data concerning aging phenomena
are scarce. No results are given on the location of the various
degradations on the cell surface.
The objectives of this work are to build a PEM fuel cell predictive
model able to simulate the fuel cell aging with time and to explain
the behavioral evolutions with local degradation mechanisms. A
2000-h durability test is used in this study to help building the
degradation module and to validate it.
2. Experimental part
2.1. Description of the 2000-h durability test with steady operating
conditions
A stack equipped with 30 large-area cells has been operated
during 2000 h with steady operating conditions. The 220 cm2 MEAs
 C. Robin et al. / Journal of Power Sources 326 (2016) 417e427
are in-house from the CEA-LITEN and their properties are listed in
Table 1. The bipolar plates are in steel SS316L with a serpentine flow
field. An in-situ device provided by the Sþþ company [16] [17], is
inserted in the middle of the stack to measure current density and
temperature distributions at the cell surface.
Humidified pure hydrogen and air are fed to the anode and
cathode respectively. Before the test, the stack is conditioned dur-
ing 24 h on bench with highly humidified gases (80% Relative
Humidity). The 30-cell stack has then been operated at a steady
current density of 0.4 A cm2 while keeping constant the operating
temperature (80  C) and the Relative Humidity (Air/H2 ¼ 30%/50%),
as well as the gases stoichiometry (Air/H2 ¼ 2/1.5) and the gases
inlet pressure (1.5 bar).
Voltage-current polarization curves (VeI curves) are recorded
every 200 h during the durability test with the same operating
conditions. They are obtained by measuring the voltage with a
current stepwise increment of 0.0045 A cm2 (between 0 and
0.05 A cm2) and 0.05 A cm2 (between 0.05 and 2 A cm2). Each
current plateau is held during 2 min to get a more stable measure.
Cyclic Voltammetries (CV) are recorded as well after each VeI
curve in order to assess the remaining ECSA at the cathode. The
measurement is conducted for cells n 1, 7, 15, 20, 25 and 30 with a
potentiostat using a 200 mV s1 voltage sweep rate. During the
measurement, nitrogen is fed to the cathode and the operating
conditions are changed to 60  C and 50%/50% RH to prevent the
membranes from drying since no water is produced in the fuel cell
in the meantime.
Linear Sweep Voltammetries (LSV) and Electrochemical
Impedance Spectroscopies (EIS) have also been conducted every
200 h during the durability test for several cells of the stack.
Resulting measures are not shown in this paper for sake of clarity
since no clear evolutions of the permeation current nor of the
impedance have been reported.
The evolution of the 30 cells’ voltages during the durability test
is shown in Fig. 1. The presence of both reversible and irreversible
degradations can be observed on the graph. Indeed, a voltage re-
covery occurs after each characterization phase every 200 h, which
highlights the reversible phenomena. This recovery is mainly due
to the cathode wash-out with N2 and OCV voltage. Meanwhile, an
overall voltage decrease can be noted between the Beginning of Life
(BoL) and the End of Test (EoT), sign of irreversible degradations.
The behavior of the cell n 1 and especially the voltage drop
occurring from approximately 1400 h of test seems isolated
compared to the other cells. It could be attributed to the formation
of a pinhole on the cell n 1 membrane, which could come from an
initial defect of the MEA.
Fig. 2 presents both the polarization curves and the percentage
of ECSA variation during this test. The performances evolution
shown in Fig. 2a enables to assess the total degradation occurring
during the 2000-h durability test. The non-linear evolution of the
performances can be pointed out. An average degradation rate of
14 mV h1 from 200 h to the end of the test can be obtained at
0.5 A cm2 with the test operating conditions, while an average
degradation rate of 250 mV h1 from the BoL to 200 h is observed.
Therefore, a higher loss occurs at the BoL (0e200hr period)
whereas the loss is much smaller for the rest of the test. This
behavior could be related to the evolution of the ECSA obtained by
Table 1
MEAs’ properties for the 2000-h durability test with steady operating conditions.
Anode Cathode
Catalyst nature Pt Pt3Co
Catalyst loading (mgPt/cm2) 0.1 0.4
Thickness (mm) 5 10
419
the CV and shown in Fig. 2b, which follows the same trend. The
surface loss intervenes mainly during the first 600 h of the test,
then stabilizes with a much smaller degradation rate.
This first durability test is used to fit the aging module of the
model developed in this work.
2.2. Description of the 2000-h durability test with dynamic
operating conditions
A second 30-cell stack with the same components’ properties as
the first one has been tested, using dynamic operating conditions.
More precisely, all these conditions remain identical to the first test
(cf. Table 1) except the fuel cell current profile and the cathode
Relative Humidity. The current profile is adapted from the New
European Driving Cycle (NEDC), which is a simplified cycle adapted
by the Motor Vehicle Emissions Group. NEDC considers the driving
cycle as a combination between an urban cycle (800 s) and an
extra-urban cycle (400 s) with constant speeds, accelerations and
decelerations. However, it suffers some criticism for not consid-
ering the real accelerations of a light-duty vehicle. The minimum
and maximum currents are set respectively to 0.1 A cm2 and
0.8 A cm2 in this work. The cathode RH is set to 50% during the
major part of the cycle but is set to decrease to 30% during the high
power phase, in order to reproduce the probable behavior of an on-
board fuel cell system. After approximately 1400 h, this second
durability test had to be stopped to remove the cell n 15, whose
voltage collapsed due to a severe pinhole. The stack was then
reassembled with the original 29 cells and operated again to reach
the 2000 h of test (more than 6000 cycles). The performances loss
between 200 h and the EoT gave an average degradation rate of
25 mV h1 at 0.5 A cm2, slightly higher than the first test’s one.
This second durability test is used to confirm that the model’s
ECSA aging module with parameters fitted on the first test is able to
predict the experimental performances loss, even with a different
operating profile.
2.3. Description of the post-mortem analyses
Two post-mortem devices are exploited in this study to acquire
more data on the ECSA degradation locations, a mini-cell and a
microscope.
For the mini-cell tests, 2 cm2 circular samples are cut in several
aged MEA in 3 locations: the air inlet, the cell center and the air
outlet. The samples are inserted in the mini-cell device for
measuring their ECSA using the Cyclic Voltammetry. To ensure that
all samples are in the same hydration state before the measure-
ment, a 1-h conditioning phase with a steady current density of
0.2 A cm2 at ambient operating conditions is applied. Then the CVs
are measured using the same protocol as described for the mea-
surements in stack.
The cathode catalyst layer’s particles of several aged MEA have
been observed using a commercial field emission gun scanning
electron microscope (LEO 1530 Gemini) for the same samples used
for the mini-cell tests.
3. Modeling
3.1. 2Dþ1D fuel cell model
The fuel cell model used in this study is fully described in a
previous work [18]. It will be referred to as “PSþþ model” in this
Membrane
paper. It is based on the resolution of Ordinary Differential Equa-
e tions (ODE) through a Bond-Graph representation. The complex in-
e plane serpentine flow field of the bipolar plates is modeled in 2D.
25
The through-plane species transports are modeled in 1D (no in-
420 C. Robin et al. / Journal of Power Sources 326 (2016) 417e427

Fig. 1. Evolution of the 30 cells’ voltages during the 2000-h durability test with steady operating conditions (cf. Table 1).

to b9 coefficients are added to compensate some assumptions of the

model, such as the absence of thickness for the active layer
(interface).
8
>
>
>
> Ucell ¼ Urev þ h  Rohm i
>
>
>
> !1 !1
>
> 0
>
> PO*
2
PH*
>
>
>
2 2
>
> B P0 P0 C
>
> RT B ! C
>
> with Urev ¼ U 0 þ logB C
>
> nF @ PH* A
>
< 2O
P0
i¼0
>
> ! !
>
>
>
> PO* 2 *
PH
>
> and h ¼ b1 þ b2 T þ b3 TlnðiÞ þ b4 Tln þ b5 Tln 2
>
>
>
> P0 P0
>
> " ! #
>
>
>
> *
PH i i
>
> þ b6 Tln 2O
þ b7 þ b8 þ b9 i
>
>
>
> P 0 s c s a
:

(1)

Fig. 2. a) Performances evolution through VeI curves and b) average remaining ECSA
obtained from the Cyclic Voltammetries for the 2000-h durability test.
plane transport in the GDL and catalyst layer) using Maxwell-Stefan
equations for the gases and adding the formalism from Young et al.
[19]. The through-plane liquid water transport is modeled using
Darcy equations in the GDL. Mass balance and enthalpy balance are
solved in each discretized layer. The water transport in the mem-
brane is modeled by solving the electro-osmosis and diffusion
mechanisms equations. Thermal aspects are considered in the
PSþþ model by solving enthalpy balances in every physical layer. In
the active layer, which is considered only as an interface in the
model, the electrochemical response is calculated from a semi-
empirical relation derived from the Butler-Volmer expression of
the exchanged current density. The cell voltage is then given by the
system of equation (1), as a function of the local conditions of the
fuel cell which are dynamically solved by the model and of bk co-
efficients which need to be adjusted on experimental data. The b6
where Ucell is the cell voltage (V), Urev is the reversible cell voltage
from the thermodynamic equilibrium (V), h is the overpotential (V)
with h  0, Rohm is the specific ohmic resistance of the cell
(ohm.m2), U0 is the standard voltage (V), R is the gas constant
(J mol1 K1), T is the temperature (K), n is the number of
exchanged electrons (), F is the Faraday constant (C mol1), Pk* is
the partial pressure of the species k at the active layer interface (Pa),
P 0 is the standard pressure (Pa), i is the local current density
(A m2), sk is the protonic conductivity of the active layer at the
electrode k (S m1) and bk are semi-empirical coefficients (V or
V K1 or U m2 or m).
A degradation module is then added to the PSþþ model by a
coupling with a nano/micro scale model of the active layer,
following the approach described in the next sub-section.
3.2. Nano/micro scale model description and coupling method
A nano/micro scale model of PEM fuel cells has been built at CEA
Grenoble for a decade. This model, previously described in
Refs. [20], will be referred to as “EDMOND model” in this paper.
Detailed descriptions of this model can be found in the literature
[20e23]. Its main features are the accurate modeling of the elec-
trochemical double layer using non-equilibrium thermodynamics,
 C. Robin et al. / Journal of Power Sources 326 (2016) 417e427 421

along with the inclusion of several degradation mechanisms Table 2

(platinum dissolution, carbon corrosion). Reactions constants values used in the Edmond model for the Pt dissolution
mechanism.
In particular, the platinum dissolution through equation (2) is
taken into account in the EDMOND model based on [24]: Parameter Value Unit

þ
Pt/Pt 2 þ 2e (2)
The dissolution rate is known to depend on the radius of the
particle. Indeed, small particles are less stable than big ones. In this
paper, the Pt dissolution is assumed to be a 3-step process at the
atomistic scale: i/Pt sintering (removal of a Pt atom from the crystal
lattice and positioning of this adatom on a reactive site). An adatom
can be considered as a Pt atom adsorbed at the surface of the Pt
crystal: it shares one single bond with the Pt crystal (DGs is asso-
ciated to this step); ii/electrochemical reaction: Pt oxidation into
Pt2þ (DGelec is associated to this step); iii/desorption of Pt2þ (DGdes
is associated to this step). Thus, the total variation of free Gibbs
energy is given by (3):
DG ¼ DGs þ DGelec þ DGdes (3)
with DGs the free enthalpy required for a platinum atom extraction
calculated by DFT; DGelec ¼ 2aF Dc is the free enthalpy corre-
sponding to the oxidation of the Pt atom (with a the transfer co-
efficient and Dc the local potential in the inner layer calculated by
the EDMOND model) and DGdes ¼ b EGT ðrPt Þ with 0 < b < 1 the
free enthalpy required for leaching of the Pt 2þ (where EGT ¼ gM2r
Pt =rPt
is the Gibbs Thomson energy that depends on the radius of the
platinum particle). This term accounts for the nanoparticle radius
dependency on the dissolution rate. This means that the smallest
particles will dissolve the fastest.
The dissolution rate of the platinum particles (vdiss in
mol:m2 :s1 ) is given by equation (4). In this work, only Pt disso-
lution is considered. Therefore, the contribution of the reverse re-
action, which corresponds to Pt precipitation, is not taken into
account in this equation:
vdiss ¼ keRT ð2aF DcþbEGT DGs Þ
1
(4)
k 2:2*106 mol:m2 :s1
DGs 75000 J
g 2:4 J:m2
a 0:5 ðÞ
MPt 0:1951 kg:mol1
rPt 21470 kg:m3
initial number of particles. Therefore, the degradation rates of the
two populations exposed to the same local conditions are different.
Also, the local electrode potential at the nanoscale, which is one of
the inputs of each 4-D table, is approximated by (Erev þ h) at the
microscale, Erev and h being respectively the reversible potential
and the overpotential calculated in the PSþþ model according to
equation (1). This means that the potential loss by ionic transport
through the electrochemical diffusion layer is neglected. A
dimensionless effective surface area degradation rate, tdegr ðtÞ, is
then deduced from the catalyst degradation model.
The loss of Pt mass due to dissolution should be represented as
the decrease of the radius of the particles (the number of the par-
ticles in each population being kept constant). For sake of
simplicity, the loss of Pt mass will be described in this paper as the
decrease of the number of particles in each population (their radius
being kept constant).
Even if the first representation is closer to reality, the second one
can be seen as a translation of the variation of the radius into a
variation of the number of particles. However, this second repre-
sentation is interesting since the simulated results correspond
directly to the post-mortem characterization by Transmission
Electron Microscopy.
To translate the radius decrease into a decrease of the particles
number along with the definition of the catalyst degradation rate,
the variation of particles number dnk of the population k at the
time t is given by the equation (6):

with k the rate constant (mol:m2 :s1 ).

A catalyst degradation rate dr (m s1) is then dynamically
dnk 4prk2 rPt drk
dt ¼ (6)
calculated, following the equation (5): dt mk dt

dr M where drk is the variation of the average particles’ radius for the
¼ vdiss Pt (5) population k given by the lookup table, rk is the initial average
dt rPt
particles’ radius for the population k, rPt the platinum density and
where r is the average particles’ radius (m), MPt the platinum molar mk the mass of a particle of the population k. The cumulated vari-
mass (kg mol1), rPt the platinum density (kg m3). The values used ation of particles number in the population k at t, DNk ðtÞ, is given by
for the platinum dissolution are given in Table 2. the equation (7):
In order to import an ECSA degradation mechanism in the PSþþ
model from the Edmond one, an indirect coupling procedure is Zt
conducted. The approach is similar to the one described in DNk ðtÞ ¼ dnk dt (7)
Refs. [20], but has been slightly improved since then. As a pre- 0
liminary step, a 5-D table is built by performing thousands of
simulations with the EDMOND model to calculate the particles’ Therefore, a global degradation tdegr is computed by the
radius degradation rate for each set of values related to 5 input equation (8):
parameters: the local potential, the initial average particles’ radius, P
the temperature, and the oxygen and vapor molar fractions in the k Sk
tdegr ðtÞ ¼ P (8)
cathode active layer. These parameters have been selected thanks k Sk;ini
to a sensitivity study because they affect the catalyst degradation.
As a second step, two 4-D tables created from the 5-D one are where Sk ðtÞ ¼ Sk;ini ðNk;ini þ DNk Þ is the surface of the population k at
integrated in the PSþþ model’s active layer part, with the 4 input t with Nk;ini being the initial number of particles in the population k
parameters being dynamically calculated by the model (local and Sk;ini being the initial surface of the population k.
electrode potential, local temperature, local oxygen molar fraction, Finally, this degradation rate is included in the expression of the
local vapor molar fraction). Each 4-D table is associated to a specific computed cell voltage to update the local current density i with the
particle population, with a given initial average radius and a given evolving ECSA, as shown in equation (9).
422 C. Robin et al. / Journal of Power Sources 326 (2016) 417e427
" !
i PO* 2
Ucell ¼ Urev  Rohm i þ b1 þ b2 T þ b3 Tln
tdegr
!# " *
!
*
PH PH
þ b5 Tln 2
þ b6 Tln 2O
þ b7
P0 P0
# - The initial number of particles in each population, N1;ini and
i i
þ b8 þ b9
sa *tdegr tdegr
Therefore, the PSþþ model is now endowed with an ECSA
degradation module. Some of its parameters still need to be
adjusted. This operation along with first computational results are
given in the next sub-section.
4. Algorithm for the aging module’s parameters fitting
This section aims to propose and compare two algorithms to fit
the parameters of the catalyst layer aging module. The first one
relies on CV measurements that are usually considered for the
measurement of the ECSA. The performance losses are then
computed from the ECSA losses. The second one relies on the global
experimental performance losses from the polarization curves. It
requires inverting the performance model to get access to an
equivalent active surface area degradation that actually integrates
various degradation mechanisms, including reversible ones.
For both approaches, a major assumption is that for modeling
the performance losses of the stack with aging, only catalyst
degradation through Pt dissolution can be considered as a degra-
dation mechanism. The carbon support corrosion is not taken into
account because neither startups nor shutdowns of the fuel cell are
modeled or tested in this study. These operating conditions remain
the main causes of the carbon support corrosion [25]. Neither the
degradations of the ionomer nor of the membrane nor of the GDL
nor of the bipolar plates are taken into account as well. However,
the loss of the equivalent surface area, which modifies locally the
current density of the electrochemical response, impacts also the
ohmic transport of the active layer (both ionic and electronic).
These assumptions are supported by the experimental results
from the two 2000-h durability tests:
- A little internal resistance variation has been measured along
the aging tests by EIS. The minor evolution of the internal
resistance is indirectly taken into account by the modification of
the local current density and the modification of the ohmic
losses in the active layer;
- The permeation current remains constant for most of the cells
(measured by LSV);
- Post-mortem tests on the GDL have not shown any loss of hy-
drophobia (observed by water droplets deposition on the
surface);
- No trace of bipolar plates corrosion has been noticed during
post-mortem observations.
The model parameters are first fitted using experimental data
from the stationary durability test, then the degradation module
obtained from the second algorithm is validated using data from
the dynamic durability test.
4.1. Fitting procedure of the aging module using the CV
measurements
This first algorithm for the fitting procedure aims to be able to
find back the experimental ECSA loss (cf. Fig. 2b) by simulating a
! 2000-h durability test with the PSþþ model using the same
operating conditions. As previously mentioned, there are two pa-
þ b4 Tln
P0 rameters to be fitted for the ECSA calculation:
i - The initial average particles’ radius for each population, r1;ini and
sc *tdegr r2; ini .
N2;ini .
(9) Besides, the kinetic constants involved in Pt dissolution have
been fitted so that the simulated ECSA decay is in good agreement
with the experimental data (previous work to be published).
The values for these parameters after the fitting procedure are
listed in Table 3. A good agreement between the measured ESCA in
the present work and the simulated one is found (Fig. 3a).
The PSþþ model with the fitted aging module is then computed
to simulate the 2000-h durability test. The average cell voltage
evolution monitored during the test is compared with the simu-
lation results in Fig. 3b. A large discrepancy can be observed be-
tween the two curves. The computed cell voltage decrease is much
lower than the experimental one, despite the computed ECSA loss
being virtually identical to the ECSA loss measured from the Cyclic
Voltammetry and shown in Fig. 3a. Interestingly, the voltage decay
predicted from the model seems to match all the upper points of
the experimental curve, i.e. the cell voltage level obtained just after
the characterization phase. This could correspond to the irrevers-
ible losses only, whereas the experimental data during operation
integrate both reversible and irreversible losses. This result is
neither surprising nor inconsistent, owing to the CV measurement
procedure that implies to rinse the cathode with nitrogen before
the ESCA measurement (see section 2). This latter operation may
actually lead to the elimination of the majority of the reversible
catalyst layer aging phenomena occurring inside the fuel cell. The
lower experimental performances are certainly related to the lower
catalyst utilization rate induced either by the catalyst flooding or by
the membrane drying or by the platinum oxidation.
Therefore, keeping in mind the willingness to build a simple
equivalent model of catalyst degradation through Pt dissolution to
model the stack performance decay, fitting the parameters of the
degradation module with the ECSA profile measured from the CV is
not relevant. To simulate the global catalyst losses
(irreversible þ reversible phenomena), the global experimental
performance losses from the polarization curves become our
reference data. This leads to the second algorithm proposed in the
next sub-section.
4.2. Fitting procedure of the aging module using the experimental
performance losses from the polarization curves
Since the first fitting procedure is only able to account for the
irreversible catalyst layer losses, a second one is proposed to ac-
count for both irreversible and reversible losses. This second pro-
cedure consists in fitting the aging parameters based on the
polarization curves. Even though the aging module has been built
considering only the catalyst degradation, the second fitting pro-
cedure considers the equivalent loss of active surface area by the
different irreversible (mainly Pt dissolution) and reversible (such as
Table 3
Resulting values of the ECSA module’s parameters after the fitting procedure with
the experimental ECSA loss (CV).
Parameter r1;ini r2; ini N1;ini N2;ini
Value 0.95 nm 4 nm 1.95  1017 (2/3) 9.73  1016 (1/3)
 C. Robin et al. / Journal of Power Sources 326 (2016) 417e427 423

 
i i
b3; degr ðtÞ T lnðiÞ þ b7;degr ðtÞ þ b8;degr ðtÞ þ b9;degr ðtÞ i
sc sa
" !
i i i
¼ b3;ini T ln þ b7;ini þ b8;ini
tdegr ðtÞ sc *tdegr ðtÞ sa *tdegr ðtÞ
#
i
þ b9;ini
tdegr ðtÞ
(10)

where b3; degr ðtÞ, b7; degr ðtÞ, b8; degr ðtÞ and b9; degr ðtÞ are the values
for b3, b7, b8 and b9 coefficients that are fitted every 200 h to match
the experimental performance losses and where b3; ini , b7;ini , b8;ini
and b9;ini are the b3, b7, b8 and b9 coefficients that are fitted at the
beginning of life.
The resulting evolution of the ECSA degradation rate tdegr is
plotted in Fig. 4a and compared to the evolution obtained from the
Cyclic Voltammetry. It can be noticed that the equivalent loss of
active surface extrapolated from the b3, b7, b8 and b9 coefficients is
around 40%, which is much larger than the one inferred from the CV
measurements that is around 13%. We propose that it can be seen as
an equivalent loss of active surface area that would be represen-
tative of both reversible and irreversible losses in the simple
equivalent catalyst degradation module proposed here. Therefore,
Fig. 3. (a): ECSA loss obtained experimentally from the CV measurements (crosses) the equivalent loss of active surface that integrates all the degra-
and fitted ECSA profile from the model (dash line); (b): Comparison of the average cell dation mechanisms is about three times larger than the ECSA losses
voltage evolution between the stationary test (dark) and the model (light). that integrate only irreversible degradations.
The extrapolated loss of active surface can now be used as the
new reference profile to fit the parameters from the degradation
Pt oxidation [25] and ionomer reversible degradations) degradation
mechanisms.
To simulate the performance losses with our aging model, the bk
coefficients of the electrochemical response (cf. equation (1)) have
to be fitted from the polarization curves recorded during the aging
test (see Fig. 2a). Indeed, these coefficients may be extracted from
each 200-h polarization curve recorded during the test.
However, the expressions of the bk coefficients depend only on
constants and therefore should not vary with aging [9,26]. The
ECSA degradation rate tdegr is assumed to be only involved in the
terms depending on the surface area and hence on the current
density (cf. equation (7)). Therefore, the new strategy is to include
the effect of the variation of tdegr in the b3,degr(t), b7,degr(t), b8,degr(t)
and b9,degr(t) coefficients that are therefore varying with aging.
Then tdegr can be calculated and finally the equivalent active sur-
face area can be deduced.
More precisely, the first step of this second algorithm is to
determine all the values for bk,ini coefficients (b1,ini to b9,ini) at the
BoL for the stationary 2000-h durability test (see Appendix A) by
fitting the computed electrochemical response with the initial
polarization curves measured in various operating conditions.
The second step consists in determining new values for b3,degr(t),
b7,degr(t), b8,degr(t) and b9,degr(t) every 200 h (while keeping con-
stant the other bk coefficients) by fitting the electrochemical
response with the polarization curves recorded during the char-
acterization phases. Along the process, the comparison between
the simulated performances with the new values for b3,degr(t),
b7,degr(t), b8,degr(t) and b9,degr(t) and the experimental polarization
curves is carried out to ensure the validity of the approach.
Besides, it is possible to go further with this analysis by using the
values obtained for the b3,degr(t), b7,degr(t), b8,degr(t) and b9,degr(t)
Fig. 4. a) ECSA loss obtained experimentally (CV) and numerically (bi coefficients)
coefficients every 200 h. Their evolutions can actually be directly
with the corresponding computed profiles after fitting the ECSA degradation module’s
exploited to extrapolate an equivalent active surface area degra- parameters; b) Simulated average cell voltage evolution after the different fitting
dation rate tdegr , by mathematically solving the equation (10). methods and comparison to the experimental profile.
424 C. Robin et al. / Journal of Power Sources 326 (2016) 417e427

Table 4 5. Model validation and additional results

Resulting values of the ECSA module’s parameters after the fitting procedure with
the extrapolated ECSA loss.
5.1. Confrontation with the in-situ measuring device (Sþþ)
Parameter r1;ini r2; ini N1;ini N2;ini

Value 1 nm 2 nm 1.95  1017 (2/3) 9.73  1016 (1/3) Thanks to the Sþþ electronic card which has been inserted in
the middle of the 30-cell stack ([16], [18]), it is possible to monitor
the evolution of the current density distribution during the 2000-h
stationary durability test (Fig. 5). The resulting color maps (Fig. 5a)
module of the PSþþ model. These new parameters’ values are
can be compared to the current density distributions obtained after
listed in Table 4.
running the PSþþ model in the test conditions (Fig. 5b).
To check our assumption that the performance losses due to
Since the stack is operated in dry conditions, the BoL current
aging can be described as losses due to catalyst degradation by Pt
density is higher near the air outlet (bottom right hand corner of
dissolution only, the 2000-h stationary durability test is simulated
the color maps) where the membrane is better humidified. Indeed,
again with the model (including the degradation module whose
the protonic conductivity close to the air inlet is too low because of
parameters are fitted according to the second algorithm) in the
the low humidified gases, resulting in a lower current located in
same operating conditions as for the experimental one. The
this area compared with the air outlet [18] [28], [29]. Moreover, the
computed loss of equivalent active surface area is consistent with
experimental and the computed evolutions of the current density
the data extrapolated from the variations of the b degr coefficients
with aging show similar trends. In particular, the BoL heterogene-
(cf. Fig. 4a). The resulting computed average cell voltage decay is
ities seem to be amplified by the long-time operation with steady
plotted in Fig. 4b and compared to the experimental one.
operating conditions. Indeed, the relative current density distri-
The simulated fuel cell aging behavior matches the experi-
bution obtained by subtracting the EoT map from the BoL one (cf.
mental profile with very good agreement, which means that our
Fig. 5) indicates a slight shift of the current density from the air inlet
assumption is relevant. Introducing the equivalent loss of active
to the air outlet during the 2000-h durability test.
surface area seems therefore sufficient enough to simulate the
Therefore, the equivalent active surface degradation module
performance decay due to the complex aging phenomena occurring
included in the PSþþ model enables to find back the evolution of
during a 2000-h fuel cell operation with steady operating condi-
the current density distributions with aging, which further
tions. For our stationary test, this assumption is also supported by
strengthen the model validity.
post-mortem analyses that have shown that the losses occurring
during the test are mainly due to the catalyst layer aging, no major
membrane degradation being observed [27]. 5.2. Confrontation with a local ECSA assessment using a post-
The values for the parameters that are now used in the PSþþ mortem mini-cell device
model are listed in Table 4 and the validity of the aging module is
further examined in the next part. As described in the modeling part, the PSþþ model is well

Fig. 5. Current density distributions measured by the Sþþ device (upper line) and predicted by the PSþþ model (lower line). The maps of the right column are obtained after
subtracting the EoT map from the BoL one.
 C. Robin et al. / Journal of Power Sources 326 (2016) 417e427 425

Fig. 6. Distribution of the ECSA loss (a) and cathode Relative Humidity (b) computed by the PSþþ model and corresponding to the 2000-h durability test with steady operating
conditions.

adapted to study the distribution of the local conditions on the cell
surface. In particular, the map of the aged equivalent surface area
corresponding to the voltage decrease shown in Fig. 4b can be
obtained. The distribution observed in Fig. 6a indicates a larger loss
of equivalent surface area near the air outlet, even if the degrada-
tion seems to be relatively homogeneous. After examining all the
local conditions thanks to the PSþþ model, it seems that in the
areas of the cell surface where the RH is higher, the loss of equiv-
alent surface area is also stronger (Fig. 6b). This could be explained
by the higher catalyst oxidation in well-humidified zones.
To confirm the heterogeneities in the loss of equivalent surface
area predicted by the PSþþ model, post-mortem analyses are
conducted. After the end of the 2000-h stationary durability test,
several MEAs are selected to be further examined using a mini-cell
experiment. The coulometry curves obtained for the 3 regions of
the cell n 10 are plotted in Fig. 7b and compared to the ones of a
fresh MEA sample (Fig. 7a). Two peaks can be observed on the fresh
MEA coulometry curves, which is typical for a fresh Pt3Co catalyst.
For the aged cell, the second peak of the air outlet sample is much
more pronounced than for the other samples. This would mean that
the air outlet is more prone to catalyst structural modifications such
as facets reorganizations, which could lead to a more severe ECSA
loss in this region. It is important to highlight here that these trends
obtained for the cell n 10 are reproducible and have been observed

Fig. 7. (a): local CV result for a fresh MEA; (b): local CV result for an aged MEA (cell n 10 of the 30-cell stack); (c): cathode catalyst layer’s particles size distribution resulting from
the STEM observations.
426 C. Robin et al. / Journal of Power Sources 326 (2016) 417e427

conditions is performed (see the Experimental part).

Fig. 8. Application of the equivalent ECSA loss approach developed thanks to the first
test during the 0.4 A cm2 plateaus of the current profile.
The average cell voltage decay corresponding to this second test
is simulated with the PSþþ model, with the same parameters for
the degradation module than the ones set thanks to the first test
(see Table 4). The results are shown in Fig. 8 and compared with the
experimental cell voltage decay. The comparison is done during the
cycle plateaus of 0.4 A cm2 for sake of clarity and to be consistent
with the current density of the first durability test. The equivalent
loss of active surface area enables to get a computed electro-
chemical response which is in a relatively good agreement with the
experimental data. The computed performance decay is however
slightly lower than the experimental one. This could be explained
by degradation mechanisms that could be very different in dynamic
conditions compared to stationary ones. Refinement of the equiv-
alent degradation model to better fit to aging in dynamic conditions
is needed to increase the predictability of the model.

on the other aged cells from the 30-cell stack.

Scanning Transmission Electron Microscopy (STEM) has also
been used in this study to complete the mini-cell results. The mi-
croscopy pictures are not shown in this paper. As it can be shown in
Fig. 7c, the count of the particles and their diameters in several
pictures indicate that the air outlet region contains bigger particles
than the air inlet. This confirms that the ECSA loss at the air outlet is
more severe than for the air inlet.
The mini-cell test results and the STEM observations therefore
validate the ECSA degradation location predictions by the PSþþ
model. It is important to highlight here that the shift of the current
density is not expected to be neither a cause nor a consequence of
the location of the loss of ECSA.
5.3. Confrontation with the 2000-h durability test in dynamic
operating conditions
Since the degradation module of the PSþþ model has been built
and adjusted thanks to the 2000-h durability test in steady con-
ditions, it is important to validate the model with another dura-
bility test. To do so, a 2000-h durability test in dynamic operating
6. Conclusion
In this work, an innovative numerical method has been intro-
duced to model an equivalent loss of active surface area during PEM
fuel cells long-time operation. This method leads to simulated
performance losses that are similar to the experimental ones. To do
so, an experimental 2000-h durability test has been performed in
steady operating conditions. The Cyclic Voltammetries measured
regularly during the test are first used to fit the parameters of a
physical ECSA aging module included in a 2Dþ1D fuel cell model.
This first approach revealed an underestimation of the cell voltage.
Interestingly, it was shown to simulate the irreversible degrada-
tions correctly. This is probably due to the CV measurement pro-
tocol, which does not enable to observe the reversible degradation
effects on the ECSA. To consider realistic degradations while
keeping a simple equivalent degradation module whose equations
correspond to catalyst degradation only, a procedure is built to fit
the parameters of the degradation module on an equivalent loss of
active surface area profile, which is extrapolated from the decrease
of the experimental performances with aging. The results show
that the computed cell voltage evolution using this method is in

Table 5
Values of the bk coefficients determined by fitting the performances evolution for the first durability test (steady operating conditions)

0 h (BoT) 200 h 400 h 600 h 800 h 1000 h 1200 h 1400 h 1600 h 1800 h 2000 h

b1 0.704 Coefficient kept constant.

b2 1.75 105 Coefficient kept constant.
b3 3.44 105 4.73 105 4.94 105 5.10 105 5.22 105 5.03 105 4.84 105 5.60 105 5.30 105 5.47 105 5.26 105
b4 1.23 104 Coefficient kept constant.
b5 1.32 105 Coefficient kept constant.
b6 2.70 107 Coefficient kept constant.
b7 3.30 105 6.78 105 8.36 105 6.55 105 8.75 105 5.58 105 8.36 105 5.93 105 9.05 105 9.46 105 6.42 105
b8 1.10 105 4.47 106 9.52 106 1.84 105 4.02 107 6.78 106 1.25 105 2.65 105 2.88 106 3.78 109 4.40 106
b9 9.62 106 1.07 106 4.54 107 7.42 107 1.98 108 6.51 106 7.49 1011 1.15 109 4.62 108 8.07 1011 6.96 106

Table 6
Values of the bk coefficients determined by fitting the performances evolution for the second durability test (dynamic operating conditions)

0 h (BoT) 200 h 400 h 600 h 860 h 1040 h 1270 h 1430 h 1710 h 1970 h

b1 0.739 Coefficient kept constant.

b2 4.34 105 Coefficient kept constant.
b3 2.94 105 3.95 105 4.11 105 4.39 105 4.53 105 4.72 105 4.77 105 4.63 105 4.40 105 4.17 105
b4 1.37 104 Coefficient kept constant.
b5 2.06 106 Coefficient kept constant.
b6 4.67 107 Coefficient kept constant.
b7 2.83 105 4.67 105 5.87 105 9.20 105 8.71 105 9.05 105 9.35 105 8.29 105 8.77 105 2.52 105
b8 8.29 106 3.38 105 2.44 105 8.06 108 1.58 106 4.61 106 5.68 108 1.43 105 3.73 106 2.19 105
b9 1.34 105 2.13 106 1.53 106 5.21 107 1.73 106 1.05 106 1.11 106 9.98 107 5.42 106 1.60 105
 C. Robin et al. / Journal of Power Sources 326 (2016) 417e427
very good agreement with the experimental one, which validates
the assumption that the performance losses due to the complex
degradation phenomena occurring in the stack can be simulated by
a simple equivalent degradation module. The approach is then
further supported by exploiting other data from the first durability
test (current density distributions evolution from the Sþþ in-situ
device, ECSA loss main location from a post-mortem local charac-
terization using a mini-cell device and STEM observations). The
aging model is finally validated by comparing its simulated results
to the experimental results from a second 2000-h durability test in
dynamic conditions.
This study aims at modeling the experimental performances
losses of a PEMFC stack. Interestingly, an equivalent catalyst
degradation module whose parameters are fitted according to the
methodology proposed here was shown to lead to similar perfor-
mance losses. However, to predict the fuel cell lifetime, more
degradation mechanisms should be added to the fuel cell model,
with a special need for a membrane degradation module. Indeed, as
illustrated in this study, the Electrochemical Catalyst Surface Area
degradation seems to be the main aging mechanism decreasing the
cell performances, but it lowers with time and is not responsible for
the cell failure. Conversely, the membrane degradation effect on the
fuel cell voltage evolution remains weak, except when the cell is
very close to failure which entails a dramatic permeation current
increase and cell voltage collapse. Dedicated works should address
the membrane failure modeling in order to improve numerical
lifetime predictions. In particular, a statistical study could be per-
formed to try establishing a link between the membrane failure and
a possible initial defect on the surface of the MEA. Other fuel cell
operating profiles could also be examined, like start/stop cycles
which can be numerous for a fuel cell use in transportation. In
particular, their effects on the ECSA loss should be assessed by
modeling.
Acknowledgement
The authors would like to thank Zodiac Aerospace for their
financial support. The authors are also grateful to Irina Profalitova
and Marion Scohy for their help concerning the mini-cell device.
The authors finally thank Laure Gue
taz for the STEM observations.
Appendix A. bk coefficients’ values
The bk coefficients of the PSþþ model’s electrochemical
response have been fitted on the experimental polarization curves
using a Matlab® linear optimization script. The convergence crite-
rion has been set to 20 mV of gap between the experimental per-
formances and the computed ones. The values of the bk coefficients
for the first 2000-h durability test (steady operating conditions) are
listed in Table 5. The values of the bk coefficients for the second
2000-h durability test (dynamic operating conditions) are listed in
Table 6.
References
[1] V. Ramani, Fuel cells, Electrochem. Soc. Interface 15 (1) (2006) 41e44.
[2] Fuel Cell Technical Team Roadmap, U.S. DRIVE Partnership, 2013.
[3] F.A. de Bruijn, V. a. T. Dam, G.J.M. Janssen, Review: durability and degradation
issues of PEM fuel cell components, Fuel Cells 8 (1) (2008) 3e22.
[4] N. Yousfi-Steiner, Ph. Mocoteguy, D. Candusso, D. Hissel, A review on PEM
voltage degradation associated with water management: impacts, influent
factors and characterization, J. Power Sources 183 (2008) 260e274.
427
[5] J. Wu, X.Z. Yuan, J.J. Martin, H. Wang, J. Zhang, J. Shen, S. Wu, W. Merida,
A review of PEM fuel cell durability: degradation mechanisms and mitigation
strategies, J. Power Sources 184 (1) (2008) 104e119.
[6] L. Dubau, L. Castanheira, M. Chatenet, F. Maillard, Ja ~© Dillet, G. Maranzana,
S. Abbou, O. Lottin, G.D. Moor, A.E. Kaddouri, C. Bas, L. Flandin, E. Rossinot,
N. CaquA©,~ Carbon corrosion induced by membrane failure: the weak link of
{PEMFC} long-term performance, Int. J. Hydrogen Energy 39 (36) (2014)
21902e21914.
[7] T. Jahnke, G. Futter, A. Latz, T. Malkow, G. Papakonstantinou, G. Tsotridis,
P. Schott, M. Gerard, M. Quinaud, M. Quiroga, A.A. Franco, K. Malek, F. Calle-
Vallejo, R.F. de Morais, T. Kerber, P. Sautet, D. Loffreda, S. Strahl, M. Serra,
P. Polverino, C. Pianese, M. Mayur, W.G. Bessler, C. Kompis, Performance and
degradation of proton exchange membrane fuel cells: state of the art in
modeling from atomistic to system scale, J. Power Sources 304 (2016)
207e233.
[8] A.V. Virkar, Y. Zhou, Mechanism of catalyst degradation in proton exchange
membrane fuel cells, J. power sources 154 (2007) B540eB547.
[9] M.W. Fowler, W.F. Mann, J.C. Amphlett, B.A. Peppley, P.R. Roberge, Incorpo-
ration of voltage degradation into a generalised steady state electrochemical
model for a PEM fuel cell, J. Power Sources 106 (2002) 274e283.
[10] M. Soltani, S.M.T. Bathaee, Development of an empirical dynamic model for a
Nexa PEM fuel cell power module, Energy Convers. Manag. 51 (2010)
2492e2500.
[11] M. Jouin, R. Gouriveau, D. Hissel, M.-C. Pera, N. Zerhouni, Prognostics of PEM
fuel cells under a combined heat and power profile, IFAC 48 (3) (2015) 26e31.
[12] M. Jouin, R. Gouriveau, D. Hissel, M.-C. Pera, N. Zerhouni, Prognostics of PEM
fuel cell in a particle filtering framework, Int. J. Hydrogen Energy 39 (2014)
481e494.
[13] A.A. Kulikovsky, The effect of noneuniform aging of a polymer electrolyte fuel
cell on the polarization curve: a modeling study, Electrochimica Acta 123
(2014) 542e550.
[14] E. Pohl, M. Maximini, A. Bauschulte, J. vom Schloss, R.T.E. Hermanns, Degra-
dation modeling of high temperature proton exchange membrane fuel cells
using dual time scale simulation, J. Power Sources 275 (2015) 777e784.
[15] S.R. Dhanushkodi, S. Kundu, M.W. Fowler, M.D. Pritzker, Use of mechanistic
carbon corrosion model to predict performance loss in Polymer Electrolyte
Membrane fuel cells, J. Power Sources 267 (2014) 171e181.
[16] L. Jabbour, C. Robin, F. Nandjou, R. Vincent, F. Micoud, J.-P. Poirot-Crouvezier,
J. d Arbigny, M. Gerard, Feasibility of in-plane GDL structuration: impact on
current density distribution in large-area proton exchange membrane fuel
cells, J. Power Sources 299 (2015) 380e390.
[17] Sþþ Simulation Services, 2015. www.splusplus.com.
[18] C. Robin, M. Gerard, J. d Arbigny, P. Schott, L. Jabbour, Y. Bultel, Development
and experimental validation of a PEM Fuel Cell 2D-model to study hetero-
geneities effects along large-area cell surface, Int. J. Hydrogen Energy 40
(2015) 10211e10230.
[19] J.B. Young, B. Todd, Modelling of multi-component gas flows in capillaries and
porous solids, Int. J. Heat Mass Transf. 48 (25e26) (2005) 5338e5353.
[20] C. Robin, M. Gerard, A.A. Franco, P. Schott, Multi-scale coupling between two
dynamical models for PEMFC aging prediction, Int. J. Hydrogen Energy 38 (11)
(2013) 4675e4688.
[21] A.A. Franco, Chapter 11-Modelling and analysis of degradation phenomena in
polymer electrolyte membrane fuel cells, in: C. Hartnig, C. Roth (Eds.), Poly-
mer Electrolyte Membrane and Direct Methanol Fuel Cell Technology, vol. 1,
Woodhead Publishing, 2012, pp. 291e367.
[22] A.A. Franco, P. Schott, C. Jallut, B. Maschke, A multi-scale dynamic mechanistic
model for the transient analysis of PEFCs, Fuel Cells 2 (2007) 99e117.
[23] A.A. Franco, M. Gerard, Multiscale model of carbon corrosion in a PEFC:
coupling with electrocatalysis and impact on performance degradation,
J. Electrochem. Soc. 155 (4) (2008) B367eB384.
[24] E.F. Holby, W. Sheng, Y. Shao-Horn, D. Morgan, Pt nanoparticle stability in
PEM fuel cells: influence of particle size distribution and crossover oxygen,
Energy Environ. Sci. 2 (2009) 865e871.
[25] G. Maranzana, A. Lamibrac, J. Dillet, S. Abbou, S. Didierjean, O. Lottin, Startup
(and shutdown) model for polymer electrolyte membrane fuel cells,
J. Electrochem. Soc. 162 (7) (2015) F694eF706.
[26] J.C. Amphlett, R.M. Baumert, R.F. Mann, B.A. Peppley, P.R. Roberge, T.J. Harris,
Performance modeling of the Ballard Mark IV solid polymer electrolyte fuel
cell, J. Electrochem. Soc. 142 (1) (Jan. 1995) 9e15.
[27] C. Robin, De
veloppement d’un mode le pre

dictif de dure

e de vie d’une pile
PEMFC pour une application ae
ronautique: e
tude des interactions entre le
cœur de pile et les conditions d’ope
ration du systeme, Universite de Grenoble,
2015.
[28] I. Alaefour, G. Karimi, K. Jiao, X. Li, Measurement of current distribution in a
proton exchange membrane fuel cell with various flow arrangements: a
parametric study, Appl. Energy 93 (2012) 80e89.
[29] M. Schulze, E. Gulzow, S. Schonbauer, T. Knori, R. Reissner, Segmented cells as
tool for development of fuel cells and error prevention/prediagnostic in fuel
cell stacks, J. Power Sources 173 (2007) 19e27.