international journal of hydrogen energy xxx (xxxx) xxx

Available online at www.sciencedirect.com

ScienceDirect

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

A systematic approach for matching simulated and

experimental polarization curves for a PEM fuel cell

Muhammad Arif, Sherman C.P. Cheung, John Andrews*

School of Engineering, RMIT University, PO Box 71, Bundoora, 3083, VIC, Australia

highlights

Exchange current density and charge transfer coefficient affect activation losses.

Effective water removal enhances PEM fuel cell performance at high currents.

A systematic procedure is presented to match simulated and experimental V-j curves.

Maximum percentage difference between simulation-4 and experimental data is 5%.

The uniqueness of the best-fit input parameter values is discussed.

article info abstract

Article history: Matching simulated and experimental polarization curves is an essential step in the
Received 30 June 2019 modelling of polymer electrolyte membrane (PEM) fuel cells, but the numerical values of
Received in revised form many input parameters like exchange current densities, charge transfer coefficients, pro-
3 November 2019 tonic conduction coefficient and water removal coefficient are hard to be found experi-
Accepted 7 November 2019 mentally. In this paper, the influence of these input parameters on the performance of PEM
Available online xxx fuel cells has been investigated using the ANSYS PEM Fuel Cell Module. The simulation
results show how the exchange current densities and charge transfer coefficients influence
Keywords: the activation losses; membrane resistance and contact resistance between the different
Simulation modelling components of a fuel cell contribute to the ohmic losses; and the coefficient of liquid water
PEM fuel cells removal affects the concentration losses. A systematic procedure to match a simulated
ANSYS PEM Fuel Cell Module polarization curve with an experimental curve is presented and illustrated by application
Polarization curve to an experimental PEM fuel cell with 5 cm2 active area.
Matching simulated and © 2019 Hydrogen Energy Publications LLC. Published by Elsevier Ltd. All rights reserved.
experimental curves

ensuring continuous, reliable supply from renewable energy

Introduction sources are major technical challenges requiring the devel-
opment of a range of energy storage technologies [2]. Great
To reduce global carbon dioxide emissions, dependence on efforts have been made to address these challenges, but
fossil fuels needs to be reduced by utilising renewable energy amongst all electrical energy storage is the most promising
sources for generation of electricity [1]. As a result of their solution [3]. Hydrogen gas produced from renewable energy,
intermittent nature, balancing demand and supply and stored and then reconverted back into electricity in a PEM fuel

* Corresponding author.
E-mail addresses: [email protected] (M. Arif), [email protected] (S.C.P. Cheung), john.andrews@rmit.
edu.au (J. Andrews).
https://doi.org/10.1016/j.ijhydene.2019.11.057
0360-3199/© 2019 Hydrogen Energy Publications LLC. Published by Elsevier Ltd. All rights reserved.

Please cite this article as: Arif M et al., A systematic approach for matching simulated and experimental polarization curves for a PEM
fuel cell, International Journal of Hydrogen Energy, https://doi.org/10.1016/j.ijhydene.2019.11.057
2 international journal of hydrogen energy xxx (xxxx) xxx

cell is an increasingly important option for storing electricity model over-predicted the experimental results due to
[4]. neglecting factors like the dissolution of species gases in the
Fabrication and testing of PEM fuel cell prototypes takes liquid, Knudsen diffusion, clogging of active reaction sites on
considerable time and is costly [5]. Hence, instead of laborious catalyst layers due to the formation of liquid water, etc.
and costly experiments, Computational Fluid Dynamics (CFD) Moein-Jahromi and Kermani [14] used the agglomerate
coupled with a 3D electrochemical and electric field e current model to simulate the cathode catalyst layer of a PEM fuel cell
density modelling have been increasingly used for modelling and compare the results with homogenous cathode catalyst
and performance assessment of PEM fuel cells by running model. It was observed that the homogeneous model over-
numerical simulations [6]. Besides analysing PEM fuel cell predicted the experimental results by around 80% while the
performance, it is also imperative to understand the local agglomerate model has a good agreement.
processes that take place within the cells [7]. Two funda- Xing, Mamlouk and Scott [15] developed a 2D steady state
mental phenomena in PEM fuel cells are water management isothermal model to study the effects of platinum and carbon
and thermal management, which are impossible to study in loading on the performance of the PEM fuel cell. According to
situ within the compact and closed structure of a cell. The best the modelling results, the sharp turn down at high current
way of understanding these processes is thus by modelling densities is due to an increase in resistance to the diffusion of
and simulation [8] with constant reference to experimental oxygen through water film.
data at the aggregate level, that is, local current and mass flow Limjeerajarus and Charoen-Amornkitt [16] investigated
rates. The challenges faced by researchers in investigating the influence of channel flow configurations and number of
these processes solely through experiments have motivated channels on the performance of the PEM fuel cell. The
development of computational models for PEM fuel cells. modelling results reported a reasonable agreement between
Modelling can be further used as a tool to diagnose the factors experimental data and simulation results with a small dif-
limiting the performance of PEM fuel cell and optimise its ference due to exchange current density evaluation.
performance. Macedo-Valencia et al. [17] developed their models to study
The modelling of PEM fuel cells started with one- the performance of five cells PEM fuel stacks. The values used
dimensional models, which was later extended to 2D and for the physical properties of MEA were slightly different from
3D. Validation is an essential step in CFD ebased fuel cell experimental values, which caused a difference in the simu-
simulations, in which the results of the simulation are lation and experimental data.
compared with experimental results [9]. Most researchers Heidary et al. [18] investigated the influence of inline and
have compared the model's simulated polarization curves staggered blockages in the parallel channels configuration and
with the experimental curves and found close agreement but then compared with the performance of parallel channels
with some discrepancies. without blockage. The simulated polarization curve was
Bernardi et al. [10] developed a one-dimensional (in the compared with the experimental polarization curve for the
flow direction) mathematical model for a solid polymer elec- model validation. A reasonable agreement was found between
trolyte fuel cell to find out the factors affecting the cell per- the two polarization curves with some differences. The dif-
formance and explain the transport mechanism of species. ference in the performance was due to some assumptions
The value of exchange current density was used as a fitting made in the model.
variable to get the best fit between the model and experi- Chowdhury, Genc and Toros [19] developed a 3D single
mental results, but exchange current density is a function of phase isothermal model to optimise the channel to land width
temperature and Pt loading. ratio of the PEM fuel cell. The model was validated by
Thomas and John [11] extended computational simulation comparing the model and experimental polarization curves,
of fuel cells to a 2D mathematical model for a solid polymer and excellent agreement found between the both results with
electrolyte fuel cell, by representing the membrane electrode less than 1% difference.
assembly in two dimensions to analyse water management, Havaej et al. [20] developed a 3D two-phase model to
thermal management and hydrogen consumption. The sharp investigate the effects of catalyst loading distribution on the
turn down due to mass transport limitations, was not performance of PEM fuel cell. The simulated polarization
observed in model's polarization curve until the value of cur- curve was compared with experimental polarization curve for
rent density reached 5 A/cm2, because their model did not two different operating conditions. The numerical results
consider the mass transport losses due to the adsorption and obtained from the model presented in the paper was in a good
transport of reactants and oxidants in ionomer film and water. agreement with experimental results under both operating
Jung S and Trung V [12] built a 2D along-the-channel model conditions.
to investigate the influence of different channels configura- Li and Sunde
n [21] studied the influence of GDL deforma-
tion and operating parameters on the performance of PEM fuel tion on the transport phenomena and PEM fuel cell perfor-
cell using two dimensional along the channel model. The mance by developing a 3D, non-isothermal, two-phase
model could not predict the sharp turn down in cell voltage at mathematical model. The accuracy of the mathematical
high current densities due to mass transport limitations, model was assessed by comparing numerical model's polari-
because their model did not include the resistance to the zation curve with the experimental polarization curve form
gasses flowing through porous electrodes. the literature. The results of mathematical model were in a
Later Mazumder and Cole [13] developed a 3D model with reasonable agreement with the experimental results.
governing equations for formation and transport of liquid Carcadea et al. [22] investigated the influence of catalyst
water within PEM fuel cell. The results obtained from the layer composition and properties (like; platinum loadings,

Please cite this article as: Arif M et al., A systematic approach for matching simulated and experimental polarization curves for a PEM
fuel cell, International Journal of Hydrogen Energy, https://doi.org/10.1016/j.ijhydene.2019.11.057
 international journal of hydrogen energy xxx (xxxx) xxx 3

ionomer volume fraction, particle radius and electrochemical curve to an experimental curve has so far not been articulated.
active surface area) and the mass transport resistance due to This is the task taken up in the present paper, with reference
ionomer and liquid water covering catalyst particles, on the to a 5 cm2 PEM fuel cell with a parallel channel configuration
performance of the PEM fuel cell. The model results showed for which experimental polarization curves under various
that lower particle radius and high platinum loadings can conditions have been obtained. A cell with the same di-
result in high cell performance. The results of numerical mensions and component properties has been simulated
model were validated with experimental data and good using the ANSYS PEM Fuel Cell Module. The equations used
agreement was found between the results. for modelling electrochemistry, current and mass transport,
According to Bednarek and Tsotridis [6,9], one of the reason and the formation and transport of liquid water have been
for not obtaining accurate results from simulation is lack of studied in detail.
robust and reliable data for various input parameters like; The values of the model input parameters like exchange
exchange current densities, open cell voltage, charge transfer current density, charge transfer coefficient, membrane pro-
coefficients and etc. In the literature the range of values for tonic conduction coefficient, and liquid removal coefficient
exchange current densities and charge transfer coefficients are systematically varied to investigate their impacts on the
are very broad. There is no clear guide about the range for the different regions of the polarization curve, and hence achieve
values of exchange current densities and charge transfer co- a close fit to the experimental polarization curve. On the basis
efficients. Zalitis et al. [23] had compared a broad range of of this analysis, a detailed step-by-step procedure to match a
exchange current densities values at different conditions ob- simulated and experimental curve is formulated to calibrate
tained from the literature. Similarly, no range of values has the embedded empirical parameters to capture the fuel cell
been found in the open literature for the value of coefficient of characteristics accurately.
liquid removal. It should be noted that the model calibration procedure
Harvey, Pharoah and Karan [24] used a 3D CFD model to used here is simply relying on matching experimental and
compare different approaches used for modelling of catalyst simulated polarization curves. These Vej curves are extracted
layers in the PEM fuel cell. The study pointed out that a large from the model by integrating current density over the entire
number of CFD models used for modelling of fuel cells do not active area. We are assuming that the detailed equations in
include catalyst agglomerate model due to which sharp turn the simulation model that are solved numerically in each
down due to mass transport losses at higher current densities small elemental volume over the 3D space occupied by the cell
is not observed. Same discrepancies were found by Jung S and e which are all based on accepted classical electrostatic and
Trung V [12] and Mazumder and Cole [13] while validating fluid dynamics theory e are accurately representing the
their models against the experimental data. In the agglom- spatially-disaggregated physical and electrochemical pro-
erate catalyst model, the catalyst layer is assumed to be made cesses taking place.
up of many agglomerates, each of consisting of carbon black
particles and platinum catalysts dispersed on its surface.
Ju and Wang [25] found a very good agreement between the The influence of input parameters on a
simulated and experimental polarization curves, but the polarization curve
agreement between simulated and experimental at current
distribution level for the same input parameters values was Model description
not satisfactory. This concluded that by just comparing
simulated and experimental global polarization curves is The ANSYS PEM Fuel Cell Module is an add-on module to the
inadequate for the PEM fuel cell model validation. Lum and ANSYS Fluent CFD software that has the capability of
McGuirk [26] validated their model by comparing numerical modelling electrochemistry, current and mass transport, for-
and experimental local current distribution data. Such data on mation and transport of liquid water, and heat sources in PEM
the spatial distribution of current density was not obtainable fuel cells [28]. The ANSYS PEM Fuel Cell Module has been
in our experimental cell, and indeed is generally not obtain- selected for the present study because it has the capability of
able without a highly specialised experimental set up and test full 3D modelling of all components in fuel cell, so it can
cell. completely represent the geometry and structure of an actual
The experimental data for the local distribution of current experimental fuel cell, and the properties of different com-
density and species gasses are not generally available. A ponents [29]. Moreover, Fluent can solve the basic governing
highly specialised experimental setup, as used by Ref. [27] is equations required to model the PEM fuel cell solely.
required to obtain the experimental data for the local distri- The ANSYS PEM Fuel Cell Module has undergone signifi-
bution of current density and species gasses. So as a result, in cant revisions since first introduced. In the latest versions
the open literature, most of the researchers have compared (ANSYS 17.0 and subsequent), the new PEM Fuel Cell Module
the model's global polarization curves with the experimental included the agglomerate catalyst model, along with several
polarization curves to assess the accuracy of their models as other modifications compared to the old module [28]. As
mentioned by Ref. [21]. mentioned by Ref. [24], unlike the thin-interface and discrete
Various commercial softwares like ANSYS Fluent, Open volume catalyst models, the agglomerate catalyst model is
FOAM, STAR-CD, and COMSOL have been used for modelling capable of modelling the mass transport losses due to the
and simulation of fuel cells over the fifteen years. Despite the dissolution of oxygen in the ionomer.
availability of several simulation packages for PEM fuel cells, a This ANSYS PEM Fuel Cell Module uses equations that are
step-by-step procedure for matching a simulated polarization based on conservation of mass, conservation of charge,

Please cite this article as: Arif M et al., A systematic approach for matching simulated and experimental polarization curves for a PEM
fuel cell, International Journal of Hydrogen Energy, https://doi.org/10.1016/j.ijhydene.2019.11.057
4 international journal of hydrogen energy xxx (xxxx) xxx

species transport, conservation of momentum, and conser-

vation of energy. The principal assumptions that are made in
the Module for current study are:

laminar flow

non-isothermal

steady state

reactants and oxidants obey ideal gases laws

materials of catalyst, GDLs and membrane are isotropic

liquid water formation is included.

ANSYS PEM Fuel Cell Module setup

A 5 cm2 PEM fuel cell was created in the SolidWorks exactly

with the same dimensions as actual test cell. The actual test
cell and the test bench are shown in Fig. 1 [29], and Fig. 2 [29], Fig. 2 e A test rig for URFC in the Sustainable Energy
respectively. The dimensions of the cell components are listed Hydrogen Lab at RMIT Bundoora East campus [29].
in Table 1.
After creating geometry, the next step is to discretise the
geometry into small cells which is called meshing. The Table 1 e Model geometry parameters.
simulation of PEM fuel cell requires a mesh that receives a
Parameter Value Unit Reference
converged solution in less computational time. The edge
GDL anode thickness 0.30 mm From experimental cell
sizing for each edge of bipolar plates and gas flow channels
GDL cathode thickness 0.27 mm From experimental cell
were set 0.5 mm. Similarly, the size of each element on the
Catalyst layer thickness 0.02 mm From experimental cell
length and width of the membrane, GDLs and catalyst layers Bipolar plates thickness 3 mm From experimental cell
were also set to 0.5 mm. Channel width 1 mm From experimental cell
A grid independent analysis has been done to keep balance Channel height 1 mm From experimental cell
between the accurate results and computational time. The Active area 500 mm2 From experimental cell
number of grid layers along the MEA was varied and simulated
for one voltage point to check the difference in the results for
different number of grid layers. After that the mesh was imported into ANSYS FLUENT for
The total number of elements was increased as the number simulation using ANSYS PEM Fuel Cell Module. The material
of grid layer increased along MEA (i.e.: membrane, catalyst properties of membrane, GDL and catalyst layers used in the
layer and GDL). Three different sets of numbers of grid layers simulations are listed in Table 2.
within the membrane-CL-GDL structure e i.e.; 6-4-6, 8-6-8 and The boundary conditions applied were the same as used
10-8-10 respectively were analysed to check the mesh sensi- during experimental cell testing (Table 3). A pressure-velocity
tivity. The total numbers of elements in the model for the 6-4- coupling solution method with SIMPLE scheme were used for
6 membrane-CL-GDL structure was 344318; for the 8-6-8 simulations in this study. A first-order upwind spatial dis-
structure, 364558; and for the 10-8-10 structure, 384798. As it cretization scheme was applied to density, momentum, spe-
can be seen from Fig. 3, the values of current densities for all cies, energy, water content, electrical potential, protonic
grid layers are almost the same, but by increasing the number potential, water content and liquid saturation in the channels.
of grids layers the computational time increases significantly. An accurate convergence criterion is needed for every
So, to reduce the computational time without affecting the simulation to get precise results. Therefore in this study, the
accuracy of the results, the mesh with 6-4-6 number of grid four convergence criteria suggested by Ref. [30] were used for
layers was finalised for all simulations. simulations which include calculating of current at terminals,

Fig. 1 e Experimental PEM fuel cell with parallel channels configuration [29].

Please cite this article as: Arif M et al., A systematic approach for matching simulated and experimental polarization curves for a PEM
fuel cell, International Journal of Hydrogen Energy, https://doi.org/10.1016/j.ijhydene.2019.11.057
 international journal of hydrogen energy xxx (xxxx) xxx 5

calculating the consumption of species, checking the reported

current density and checking the current densities are not
changing with further iterations. The simulation output value
of current at the cathode terminal was compared with the
calculated value of the current from the oxygen consumption.
The solution was considered to be converged if the difference
between the values of current at cathode terminal and
calculated from the oxygen consumption was less than
0.0001. Similarly, the simulations were stopped when the
difference in the reported current densities between iterations
was less than 0.0001 for the same voltage point.

Polarization regions of PEM fuel cell polarization curve

The polarization curve of a PEM fuel cell can be categorised

into three distinct regions based on the causes of potential
loss; activation, ohmic and concentration (Fig. 4) [31].
Fig. 3 e Grid independent analysis of the mesh. At lower current densities, the voltage drops due to the
activation energy required for the electrochemical reaction at
the catalyst surface. The lower this energy is, the more
Table 2 e Material properties of cell MEA components. effective is the catalyst. The initial rapid drop region of the
Physical Value Units Reference polarization curve is called activation polarization region.
parameters Charge transfer coefficient and the exchange current densities
Membrane NAFION 115
are affecting the activation polarization region of the curve.
Dry membrane 1968.5 Kg/m3 Manufacturer data sheet At medium current densities, the cell voltage drops linearly
density with current due to ohmic losses. These losses occur due to
Thermal 2 W/m.K Manufacturer data sheet the electrical resistance in different components of fuel cell
conductivity like; membrane, catalyst layers, GDLs and bipolar plates. The
16
Electrical 1E Simens/m Manufacturer data sheet
linear drop region due to the electrical resistance of different
conductivity
fuel cell components is called the ohmic polarization region.
GDL-anode Elat-elt1400w
Density 377 Kg/m3 Manufacturer data sheet Protonic conduction coefficient and contact resistivity are the
Thermal 2 W/m.K Manufacturer data sheet model input parameters that influence the curve in this
conductivity region.
Electrical 267059 Simens/m Manufacturer data sheet At high current densities, the rate of water generation in-
conductivity creases at cathode electrode and causes flooding. Through
Porosity 0.63 Manufacturer data sheet
water flooding, the pores of the electrode are blocked with
Contact angle 130 degrees calculated
water that minimises the flow of oxidants into the catalyst
GDL-cathode Sintered Titanium
Density 1351.8 Kg/m3 Manufacturer data sheet layer, due to which less current is generated at a given voltage.
Thermal 21.9 W/m.K Manufacturer data sheet This drop in performance causes a sharp turn down in the
conductivity polarization curve, which is called the concentration polari-
Electrical 2380000 Simens/m Manufacturer data sheet zation region. The concentration losses can be reduced by
conductivity effective water removal. Hence the input parameter that im-
Porosity 0.63 Manufacturer data sheet
pacts the concentration losses is coefficient of liquid water.
Contact angle 70 degrees calculated
Catalyst layer
Porosity 0.28 [29]
Contact angle 130 degrees calculated
Catalyst loading 4 mg/cm2 Manufacturer data sheet
anode PtB
Catalyst loading 4 mg/cm2 Manufacturer data sheet
cathode PtB/IrO2

Table 3 e List of operating conditions for test cell.

Operating condition Value
Cell Temperature 53 C
Relative Humidity at anode 66%
Relative Humidity at cathode 34%
Inlet mass flow rate of hydrogen at anode 3 E09 kg/s
Inlet mass flow rate of air at cathode 8.20 E07 kg/s
Fig. 4 e Polarization curve of a PEM fuel cell [31].

Please cite this article as: Arif M et al., A systematic approach for matching simulated and experimental polarization curves for a PEM
fuel cell, International Journal of Hydrogen Energy, https://doi.org/10.1016/j.ijhydene.2019.11.057
6 international journal of hydrogen energy xxx (xxxx) xxx

In the next sub-sections, a 5 cm2 PEM fuel cell with parallel varied to get the best match with experimental results. There
flow channels is simulated using the ANSYS PEM Fuel Cell are four different charge transfer coefficients, which have
Module with varying input parameters, including: exchange different values that can be separately set as an input to get a
current densities, charge transfer coefficients, protonic con- good fit with the experimental polarization curve, especially
duction coefficient and water removal coefficient. Over the in the activation polarization region. This has emerged from
past few years some researchers (e.g. Refs. [32e34], have the work done by Wu [29].
studied the influence of various input parameters on the po- From the Butler-Volmer equation, the charge transfer co-
larization curve of a PEM fuel cell. In the present study, the efficients have an exponential effect on the current density. By
sensitivity analysis of the input parameters such as exchange increasing the value of charge transfer coefficients, the cur-
current densities, charge transfer coefficients, protonic con- rent density increases for the given voltage. As the voltage
duction coefficient and water removal coefficient is performed decreases, the current density rises more by increasing the
to investigate the effect of these parameters on each region of charge transfer coefficient, this because the charge transfer
the polarization curve e i.e.: the activation polarization, coefficient is multiplying with the overpotential value in the
ohmic and mass transport losses regions. The analysis per- Butler-Volmer equation. The values for charge transfer co-
formed is used in developing for the first time a detailed step- efficients were changed in two different ways: first the value
by-step procedure to match a simulated and experimental of one coefficient was varied while keeping other same; and
polarization curves. It is important to note that the influence then values for both coefficients were changed together.
of the values of coefficient of liquid removal on the polariza- To explore the effect of charge transfer coefficients on a
tion curve has not been investigated in the past studies. In polarization curve, simulations for the 5 cm2 test cell
addition, the ANSYS PEM Fuel Cell Module has gone through described above were run first by varying the value of one
vast modification and improvements over time, especially coefficient while keeping the other same; and secondly by
compared to the study performed by Ref. [32]. Hence it is changing the values of both coefficients together.
necessary to check the sensitivity of these input parameters The simulated polarization curve for the first set of runs is
using the updated version of ANSYS PEM Fuel Cell Module for displayed in Fig. 5a. When the value of the charge transfer
developing systematic procedure to match numerical and coefficient of the cathode has increased the values of current
experimental polarization curves. densities increases significantly. While on the other side, by
One input parameter is varied at a time while keeping increasing the values of the anode charge transfer coefficient
other input parameters constant to analyse the individual while keeping the values of charge transfer coefficients for
effect of each input parameter on the simulated polarization cathode constant, the rise in current density values is negli-
curve. The cell operating conditions are the same for all these gible. The change in values of cathodic charge transfer coef-
simulations and are taken from actual experimental results ficient is more effective because the overpotential at the
(see Table 3). anode is almost insignificant compared to the cathode [38],
and charge transfer coefficient is multiplied with the surface
Charge transfer coefficients overpotential in the Butler-Volmer equation. Similar trend
was found in the study conducted by Ref. [32].
The amount of the potential energy applied to an electro- For the second case, it can be seen from the polarization
chemical reaction that enhances the rate of reaction is called curves obtained from simulation results in Fig. 5b that current
charge transfer coefficient [35]. The charge transfer coefficient density increases by increasing the value of charge transfer
is generally a function of operating parameters like tempera- coefficients. Polarization curves with high charge transfer
ture and material properties of the electrode. Tijani, Binti coefficients (0.7, 0.8) are getting closer to a straight line,
Kamarudin and Binti Mazlan [36] and Biaku et al. [35] found especially at high current densities. Meanwhile for the lower
that the value of the charge transfer coefficient increases with charge transfer coefficients (0.4), the curve tends to turn down
an increase in the operating temperature. Soderberg et al. [37] steeply. It can be observed in both cases that by increasing the
investigated the influence of the properties of a porous elec- value of charge transfer coefficients the slope (negative) of the
trode on the value of charge transfer coefficient. It was found curves decreases in all the regions.
that the value of charge transfer coefficient increases with an
increase in electrode porosity while it decreases with increase Exchange current densities
in the thickness of electrode layers.
The volumetric transfer current (R ) terms are source terms Exchange current density is a parameter that estimates the
that are described in the PEM Fuel Cell Module [28] by Butler- ability of an electrode to undergo an electrochemical reaction.
Volmer equation as: A higher value of the exchange current density indicates that
 gan the electrode's surface is highly active for the electrochemical

½A

 eþaan Fhan =RT þ eacat Fhan =RT
an an
R an ¼ zan jan ðTÞ (1) reaction. For a PEM fuel cell, the value of the overpotential at
½Aref
the anode electrode is almost negligible in comparison to the
 gcat  cathode electrode. Since the total current density at both

½C 
 eþaan Fhcat =RT þ eacat Fhcat =RT electrodes must be equal by conservation of charge, the value
cat cat
R cat ¼ zcat jcat ðTÞ (2)
½Cref of exchange current density at the anode must be several or-
All the parameters listed above, excluding Faraday and ders of magnitude larger than the at the cathode [38]. Different
universal gas constant, are input parameters and can be materials have different values of exchange current density.
Platinum has the highest value (103 A/cm2) making it most

Please cite this article as: Arif M et al., A systematic approach for matching simulated and experimental polarization curves for a PEM
fuel cell, International Journal of Hydrogen Energy, https://doi.org/10.1016/j.ijhydene.2019.11.057
 international journal of hydrogen energy xxx (xxxx) xxx 7

Fig. 5 e Comparison of simulated polarization curves for different values of charge transfer coefficient on anode and cathode
electrodes.

favourable as a catalytic material, but it is very expensive. In Butler-Volmer Eqs. (1) and (2), the catalyst with the high value
the literature, a large variability is found in the range of the of exchange current density will produce high current
values of exchange current density even for the same mate- compared to the catalyst with the lower value of exchange
rial. Therefore, most researchers use it as a fitting parameter current density at same voltage drop. As can be seen in Fig. 6,
to match a simulated curve with experimental data. with an increase in the values of exchange current densities
From Butler-Volmer Eqs. (1) and (2), by increasing the value the values of output current densities increase by a significant
of exchange current densities, the value of the current density amount even at low values of overpotential. These results
increases for the same voltage. By increasing the value of the indicate that a better performing catalyst can reduce the
exchange current density, the activation losses decline, and activation losses, which will result in the better performance
high current is generated for the small values of overpotential, of the cell.
showing that the catalytic surface is favourable for rapid
electrochemical reaction to occur. Exchange current densities Protonic conduction coefficient
reflect the importance of catalyst selection in a PEM fuel cell.
The 5 cm2 PEM fuel cell operating under the same condi- The driving force behind the electrochemical reactions taking
tions was simulated for three different values exchange cur- place at anode and cathode catalysts is the potential differ-
rent densities on each electrode while all other input ence between the solid phase (GDL catalyst layers and end
parameters were constant. Fig. 6 shows the polarization plates) and membrane phase (membrane and catalyst layer).
curves obtained from the simulation for three different sets of The ANSYS PEM Fuel Cell Module [28] uses conservation of
values of exchange current densities. As expected from charge equations to solve for the two potentials e electronic
and protonic potentials.

V , ðssol Vfsol Þ þ R sol ¼0 (3)

V , ðsmem Vfmem Þ þ R mem ¼ 0 (4)

The membrane phase conductivity is modelled by

equation:
 
1 1
Ei 303T
smem ¼ bi ð0:514l  0:326Þu e (5)

The polymer electrolyte membrane used in a PEM fuel cell

allows only protons to pass through it. The proton conduc-
tivity mainly depends on the water content in the membrane.
The higher the water content, the higher will be the conduc-
tivity and vice versa. The protonic conduction coefficient (bi Þ is
a numerical constant that must be input to the model used in
ANSYS fluent for generality. It can be used as a fitting variable
to get the best match between the simulated and experi-
mental curves.
Fig. 6 e Comparison of polarization curves for different The test cell defined in section Polarization regions of PEM
values of exchange current densities on anode and cathode fuel cell polarization curve was simulated for three different
electrodes.

Please cite this article as: Arif M et al., A systematic approach for matching simulated and experimental polarization curves for a PEM
fuel cell, International Journal of Hydrogen Energy, https://doi.org/10.1016/j.ijhydene.2019.11.057
8 international journal of hydrogen energy xxx (xxxx) xxx

values of protonic conduction coefficient (bi Þ under the same using the ANSYS PEM Fuel Cell Module for two different cases:
operating conditions while keeping other input parameters without and with contact resistivity of 11.5 E06 U-m2. The
constant. As can be anticipated from Eq. (5), the polarization polarization curves obtained from the simulation results are
curves obtained from the simulation are raised in the ohmic shown in Fig. 8. As expected, including the contact resistance
polarization portion by increasing the value of protonic con- led to a drop in current density in the ohmic region of the
duction coefficient (Fig. 7), thus reflecting the lower mem- polarization curve.
brane resistance and improved performance. Two criteria can
be used to check the valid range for the protonic conduction Coefficient of liquid removal
coefficient. Firstly, the value of membrane conductivity
calculated from the simulation output values should be The rate of water generation at the cathode is directly related
compared with the membrane conductivity value reported by to the current produced. From the ANSYS PEM Fuel Cell
literature. Secondly, the membrane resistance should not be Module [28], the water produced due to oxygen reduction re-
greater than the value of total cell resistance as measured in action is:
an experimental cell.
Mw;H2 O
Sl ¼ Rcat > 0 (6)
2F
Contact resistance
At high current densities, the rate of water generation in-
The total fuel cell resistance is the sum of the membrane creases, which results in the clogging of pores of GDL. The
resistance and contact resistances between different compo- clogging of pores minimises the flow of oxidants to the cata-
nents of the fuel cell, assuming the electrical resistance of the lyst, which limits the performance of fuel cell due to mass
GDLs and endplates are negligible compared to the contact transport losses. Most of the models quantify the generation
resistance and membrane resistance. This assumption is of liquid water as liquid saturation or volume fraction of liquid
made because the value of electrical conductivities for GDLs water “s”. Liquid saturation is a ratio of the volume of water
materials are very high as it can be seen in Table 2. The pri- inside the pores to the volume of pores. The term “s” is a
mary source of the contact resistance in a fuel cell is the critical parameter for modelling of mass transport losses in
resistance at the interface between GDL and bipolar plates. PEM fuel cells. This parameter also works as an indicator of
The contact resistance depends upon the electrical conduc- how much the flow of species is restricted due to liquid water
tivity of components and the clamping pressure. The contact formation. The blockage of GDL pores due to liquid water is
resistance decreases as the clamping pressure increases, but more severe on the cathode side, because water is mainly
very high clamping pressure may damage the MEA. The con- generated at the cathode side.
tact resistance between different components can be reduced The ANSYS PEM Fuel Cell Module [28] models the drop in
by applying optimal clamping pressure. Mishra, Yang and performance of a PEM fuel cell due to clogging of GDLs and
Pitchumani [39] found that cloth based GDLs have lower catalyst pores by liquid water “s” as follows:
values of contact resistance than the paper-based GDLs. ij
Deff ¼ ð1  sÞrs ε1:5 Dij (7)
The test PEM fuel cell described in section Polarization
regions of PEM fuel cell polarization curve was simulated rs
ð1  sÞ is the pore blockage term due to liquid water, rs is

Fig. 7 e Comparison of polarization curves for the test cell

for different values of protonic conduction coefficient of Fig. 8 e Comparison of simulated polarization curves for
membrane. the test for different values of contact resistance.

Please cite this article as: Arif M et al., A systematic approach for matching simulated and experimental polarization curves for a PEM
fuel cell, International Journal of Hydrogen Energy, https://doi.org/10.1016/j.ijhydene.2019.11.057
 international journal of hydrogen energy xxx (xxxx) xxx 9

the exponent of pore blockage, which shows the effect of Fig. 10aeb depicts the liquid water saturation distribution
liquid water on the blockage of pores of electrodes and GDLs. at GDL and channel interface on the cathode side for two
The gas species diffusivity }Dij } term in Eq. (7) is evaluated different values of the coefficient of liquid removal. For the
with respect to individual species based on operating condi- low value of water removal coefficient (5 E15s/m), the liquid
tions by the full multicomponent diffusion method. The gas saturation at GDL and channel interface has a maximum
diffusion coefficient governs the diffusion rate of the gas value of 0.22 which is very high compared to the high value of
species. water removal coefficient (5 E05 s/m) which is 0.0053. It
The liquid water can also cover the active sites of catalysts means that the electrode and GDL pores for the case of the low
and thus reducing the effective surface area of the catalyst value of the coefficient of liquid removal are more clogged
layer for the electrochemical reaction. than the high liquid removal rate case.
Fig. 11aeb presents the distribution of oxygen mass frac-
Rj ¼ ð1  sÞgj R0cat (8) tion at catalyst and membrane interface on the cathode. The
ð1  sÞgj is the term that shows the reduction of catalyst maximum value of the mass fraction of oxygen for the high
effective surface area due to liquid water. value of the coefficient of liquid removal (5 E05 s/m) is 0.16
The liquid water generated at the cathode as a result of the while in the case of the low-value coefficient of liquid removal
electrochemical reaction is driven out of the GDL to channels (5 E15 s/m) it is 0.154. This reduction in the oxygen concen-
by capillary pressure [28]. tration is due to the increase of concentration losses because
of electrodes and GDL pores blockage by liquid water.
  
1
f liq ¼ Qεs:max pc þ rV2 ; 0 (9)
2

The rate of water removal depends on the porosity of GDL, Matching simulated and experimental curve
capillary pressure and the coefficient of liquid removal. By
decreasing the value of the coefficient of liquid removal, more Matching procedure
liquid water will remain inside the pores of GDL and electrode,
which will minimise the supply of oxidants to the catalyst and In the modelling of PEM fuel cells, matching of the simulated
less current will be produced. and experimental curve is an essential step as this provides
The test cell operating at the same conditions was simu- the best-fit values of input parameters to the model that
lated for two different values of liquid removal coefficient cannot usually or readily be found independently by experi-
while all other input parameters were held constant. The ment. By applying the model with these inputs to a number of
simulation results show that the cell with the high value of the cases (e.g. different operating conditions) in which their
coefficient of liquid removal (5 E05 s/m) has performed better values would not be expected to change and checking that the
in the concentration losses region compared to the cell with a simulated curve still matches the modified experimental
low value of the coefficient of liquid removal (5 E15 s/m) as curve.
depicted in Fig. 9. The concentration losses can be reduced by The matching of simulated and experimental polarization
increasing the value of the coefficient of liquid removal and curves is in practice a challenging process because the precise
vice versa. One point that is worth noting is that concentration numerical values of key model input parameters e such as
losses begin at medium current densities and get more with exchange current densities, charge transfer coefficients, spe-
an increase in current densities, as shown in Fig. 9. cific active surface areas, local species concentrations,
condensation rate, and water removal coefficient e are not
available through independent experiments. Hence these
input parameters are used as fitting variables to get a close
match between the simulated and experimental curve, which
is a time-consuming practice. A step-by-step procedure thus
needs to be developed, to allow matching of the simulated and
experimental polarization curves within as short a computing
time as possible.
In this section, each polarization region of experimental
curve-1 (Fig. 12) obtained from the testing of the 5 cm2
parallel-channel URFC operating at specified conditions (that
is, relative humidity and cell temperature) were matched with
the simulated polarization curve based on the findings in
section The influence of input parameters on a polarization
curve. After that to check the uniqueness of the values of
the best-fit input parameters, a second simulation curve was
compared with experimental curve-2 (Fig. 12) obtained from
the testing of the same 5 cm2 parallel-channel URFC but
operating at different conditions (see Table 4) A close

Fig. 9 e Comparison of simulated polarization curves for

different values of water removal coefficient.

Please cite this article as: Arif M et al., A systematic approach for matching simulated and experimental polarization curves for a PEM
fuel cell, International Journal of Hydrogen Energy, https://doi.org/10.1016/j.ijhydene.2019.11.057
10 international journal of hydrogen energy xxx (xxxx) xxx

Fig. 10 e Water saturation distribution on the cathode GDL/channel interface at 0.46 V of; a) water removal
coefficient ¼ 5 E¡05 s/m and b) water removal coefficient ¼ 5 E¡15 s/m.

Fig. 11 e Oxygen mass fraction distribution on the cathode GDL/channel interface at 0.46 V of; a) water removal
coefficient ¼ 5 E¡05 s/m and b) water removal coefficient ¼ 5 E¡15 s/m.

Matching in activation polarization region

The process of the matching simulated curve with experi-

mental started with getting a better fit in the activation po-
larization region. Based on the observation made in the earlier
section, it was found that the activation loss region is influ-
enced mainly by changing the values of exchange current
densities and charge transfer coefficients. So, in the first step,
initial values for ja and jc were set at 50 A/m2 and 5 A/m2
respectively. When the simulation output values of current
densities were plotted against voltage and compared with
experimental polarization curve as displayed in Fig. 13, it was
found that there was no match between two curves in any
region.
In the second step, the values of ja and jc were raised to 100
A/m2 and 10 A/m2 respectively to get a lift in the activation
polarization region. By comparing simulation polarization
curve with experimental as shown in Fig. 14, it is observed
Fig. 12 e Experimental curves for different operating
conditions.

Table 4 e Operating conditions of different experiments.

Experiment Relative Relative Cell
agreement was found between both curves by keeping the Humidity on Humidity on Temperature
entire model input parameters the same as for the getting fit Air Side Hydrogen Side
for the experimental curve-1 while changing the operating 1 34% 66% 53 C
conditions. 2 48% 78% 55 C

Please cite this article as: Arif M et al., A systematic approach for matching simulated and experimental polarization curves for a PEM
fuel cell, International Journal of Hydrogen Energy, https://doi.org/10.1016/j.ijhydene.2019.11.057
 international journal of hydrogen energy xxx (xxxx) xxx 11

Fig. 13 e Comparison of experimental polarization curve

and simulation polarization curve for set-1 values. Fig. 15 e Comparison of experimental polarization curve
and simulation polarization curve for set-3 values.

Matching in ohmic polarization region

As discussed in sections Protonic conduction coefficient and

Contact resistance, the ohmic loss region is influenced by
the values of proton conductivity coefficient and contact
resistance. The value of the proton conductivity coefficient
was set one from the beginning. Two criteria were used to
check the value of the proton conductivity coefficient was
within the appropriate range. Firstly, the membrane conduc-
tivity was calculated using Eq. (5) from the simulation data
and then compared with the value of proton conductivity re-
ported in the literature. According to the literature, the
membrane conductivity for Nafion 115 at room temperature is
reported to be 0.095 S/cm [40]. The calculated value of mem-
brane conductivity based on the simulation data is 0.090 S/cm,
Fig. 14 e Comparison of experimental polarization curve which is closer to the value reported in the literature. Sec-
and simulation polarization curve for set-2 values. ondly, the membrane resistance was calculated and
compared with the value of total cell resistance. The value for
the membrane resistance comes out to be 41 mU while the
total cell resistance is 61 mU, which is higher than the mem-
that there was a relatively a good fit between the two curves in
brane resistance means that the rest of resistance is due to the
the activation polarization region, but the slope of the simu-
contact resistance of PEM Fuel cell components. The value for
lation curve is more than that of experimental.
the contact resistance between components of PEM Fuel cell
Based on the findings in section Charge transfer
was calculated by subtracting membrane resistance from the
coefficients, the value of charge transfer coefficient at the
total cell resistance and was fed in as an input value. Hence
cathode electrode was then tuned to reduce the slope of the
the values for proton conduction coefficient and contact
curve in the activation losses region. As it was observed in
resistance were set based on the experimental data.
section Charge transfer coefficients the charge transfer co-
efficients values for the anode electrode have less influence on
Matching in concentration polarization region
the overall polarization curve, so for the sake of reducing
fitting variable the value for charge transfer coefficient for
As shown in Fig. 15, the simulation curve overestimates the
anode electrode was kept constant. In this step, the value of
experimental curve in the ohmic and concentration polari-
the charge transfer coefficient at the cathode was raised from
zation regions. The value for protonic conduction coefficient
0.60 to 0.65. As can be seen in Fig. 15, both curves are now
and contact resistivity were set based on the value of total cell
closely matched in activation loss regions.
resistance (section Matching in ohmic polarization region). It

Please cite this article as: Arif M et al., A systematic approach for matching simulated and experimental polarization curves for a PEM
fuel cell, International Journal of Hydrogen Energy, https://doi.org/10.1016/j.ijhydene.2019.11.057
12 international journal of hydrogen energy xxx (xxxx) xxx

removal that is why we have only focused on the coefficient of

liquid removal to get fit in the concentration polarization
region.

Evaluating closeness of fit

The closeness of fit between simulated and experimental

curve was evaluated by using different statistical criteria,
namely: percentage difference (between current density
points on simulated and experimental curves at a given cell
voltage), mean squared difference, sum of squared differ-
ences, and coefficient of determination (R2). Initially, per-
centage differences values were calculated for all voltage
points of each simulation, as shown in Table 6. The maximum
acceptable percentage difference value was kept at 5%. The
values of mean squared difference, sum of squared differ-
ences, and coefficient of determination (R2) are shown in Table
7.
Fig. 16 e Comparison of experimental polarization curve If it emerged that the fit obtained by tuning exchange
and simulation polarization curve for set-4 values. current densities, charge transfer coefficients, protonic con-
duction coefficient and was still insufficiently close, it is
important to note that there are also other numerical input
Table 5 e Sets of input parameter values for different parameters in the Module that may also be varied to improve
simulations. the match. These additional parameters include most notably:
Simulation Ja (A/m2) Jc (A/m2) aa ac b W (s/m) ṙ (U-m2) exponent of pore blockage concentration dependence, and
1 50 5 0.6 0.6 1 5E05 11.5 E06
local species concentration reference values.
2 100 10 0.6 0.6 1 5E05 11.5 E06
3 100 10 0.6 0.65 1 5E05 11.5 E06 Checking the uniqueness of input parameters
4 100 10 0.6 0.65 1 5E15 11.5 E06
After getting closer fits in all the regions, the next step was to
make sure that the values of exchange current densities,
is assumed that the simulation curve overestimates the charge transfer coefficients and coefficient of liquid removal
experimental curve because the experimental curve has more that was used for getting closer fit are unique values. Simu-
concentration losses than the simulation curve. The value of lation polarization curve-using the same values of exchange
water removal coefficient was thus reduced from 5 E05 s/m to current densities, charge transfer coefficients, and coefficient
5 E15 s/m to increase the concentration losses in the simu- of liquid removal were therefore compared with experimental
lation curve. Because of that, more liquid water was trapped in curve for the same cell operating under different conditions,
GDL and electrode pores due to which the flow of oxidant was specifically a different cell temperature and relative humidity.
restricted. The ohmic and concentration polarization regions While the difference in temperature between experiment 1
of the simulation and experimental curves are now reason- and 2 is only 2 Co, but Relative humidity difference is much
ably matched, as depicted in Fig. 16. This shows that con- larger at 14% for air and 12% for the hydrogen. Hence this
centration losses start their effects at medium current degree of difference between the cases is more than enough to
densities but get more severe at high current densities. give a clear difference in performance (as can be seen in
The values of input parameters that were used from Fig. 12), and thus a sufficient test of the ability of the model
simulation 1 to simulation 4 are listed in Table 5. with the ‘best-fit values’ of input parameters, and just these
As it can be seen in Eqs. (7) And (8), terms like; rs and gj can changes in operating conditions, to represent both conditions
also influence the concentration polarization regions. But with reasonable accuracy. A similar procedure was also sug-
since our test cell is generating less current due to which the gested by Ref. [32] for checking the uniqueness of the values of
value of liquid saturation is low. In our earlier work through input parameters. This procedure was also followed by Hu
independent runs, it was found that due to low value of liquid et al. [41] to check the accuracy of the model by comparing the
saturation “s”, the variation of values of exponent of pores numerical polarization curve using the same values of ex-
blockage have a little effect on the mass transport losses re- change current densities with the experimental polarization
gion of the polarization curve. Similarly, it was also found that curve at different operating pressure.
due to low value of liquid saturation “s” the reduction in The comparison is shown in Fig. 17 and indicates a close fit
performance of PEMFC due to covering of active sites of in all three regions of polarization curve. This closeness of fit
catalyst surface is negligible. indicates strongly that the values of exchange current den-
Hence the degree of effect of the exponent of pores sities, charge transfer coefficients, and water removal coeffi-
blockage and exponent of reduction in the catalyst effective cient found reflect the properties of the cell itself and hence
area on the mass transport losses region of the polarization apply equally under different cell operating conditions. While
curve were relatively small than the coefficient of liquid this fit does not exclusively prove that the solution set for

Please cite this article as: Arif M et al., A systematic approach for matching simulated and experimental polarization curves for a PEM
fuel cell, International Journal of Hydrogen Energy, https://doi.org/10.1016/j.ijhydene.2019.11.057
 international journal of hydrogen energy xxx (xxxx) xxx 13

Table 6 e Values of percentage errors for different simulated curves.

Voltage (V) Percentage Differences
Simulation-1 (%) Simulation-2 (%) Simulation-3 (%) Simulation-4 (%)
0.993 0 0 0 0
0.915 148 15 3 3
0.874 88 14 5 5
0.836 80 15 0.3 0.3
0.784 52 4 8 5
0.704 17 6 13 1
0.661 7 9 13 4
0.553 3 1 4 0.1
0.460 10 5 3 2

to any particular case. The overall procedure is shown as a

Table 7 e Values of mean squared difference, sum of
flow diagram in Fig. 18.
squared differences and coefficient of determination for
all simulations. The first step is to get a good fit in the activation polariza-
tion region (top box in Fig. 18). As discussed in sections Charge
Simulation Mean Squared Sum of Squared R2
transfer coefficients and Exchange current densities, ex-
difference differences
change current densities and charge transfer coefficients are
1 2.74 E03 2.5 E02 0.81
the main input parameters that influence the curve in the
2 1.30 E06 1.00 E05 0.96
activation polarization region. Initially, starting values for ja,
3 9.26 E04 8.33 E03 0.93
4 1.0 E06 9.50 E06 0.99 jc, aa and ac are chosen and two voltage points in the activation
polarization region are simulated. The value of ja should be
between 101 and 104 times that of jc, as explained in section
these input parameters found is unique, the probability of this Exchange current densities.
not being the case is low. This probability of error in the so- Once the first simulation run is completed, the values of
lution set is reduced further if additional simulations with current density at the anode catalyst and membrane inter-
these inputs of the cell operating under different conditions face, with the cathode catalyst and membrane interface, are
again do not reveal significant differences between the compared. If both values are equal, then keep the same ratio
resulting simulated and experimental curves. of ja to jc else change the ratio. After that, the current densities
in the activation polarization regions obtained from the
Step-by-step procedure simulation and experimental data are compared. If they are
close enough, tuning the values of ja and jc is stopped;if the
In this section, the procedure used to match a simulated and values are not close, then the values of ja and jc are increased
an experimental polarization curve for a PEM fuel cell is or decreased by some specific percentage (e.g. 10%, 20%, etc)
generalised into a step-by-step procedure that can be applied according to the difference between simulated and experi-
mental values. The tuning of exchange current densities can
be stopped when the simulated and experimental curves get
closer in the activation polarization region. In the next step,
the slopes of the simulated and experimental curve should be
compared. The slopes of the two curves can be matched by
changing the value of charge transfer coefficients at the
cathode side. As it was found in section Charge transfer
coefficients, the values of current densities are more sensi-
tive to change of the values of charge transfer coefficients on
the cathode side only. Once the activation regions of both
curves get a reasonably good match, then these parameters
should be kept fixed for the next simulations. It was found in
the present case that ja ¼ 100 A/m2 and jc ¼ 10 A/m2 give a
closer fit in the activation region, but there was still a differ-
ence in the slope of both curves. The value of the charge
transfer coefficient on the cathode side was increased from 0.6
to 0.65 to match the slopes of both curves. Variation of values
of exchange current densities and charge transfer coefficients
can give fit in the activation polarization region.
Ohmic losses region of the polarization curve can be fitted
Fig. 17 e Comparison of experimental polarization with
by varying the value of the protonic conduction coefficient
different operating conditions curve and simulation
(middle box in Fig. 18). The starting value for running the
polarization curve for set-4 values.

Please cite this article as: Arif M et al., A systematic approach for matching simulated and experimental polarization curves for a PEM
fuel cell, International Journal of Hydrogen Energy, https://doi.org/10.1016/j.ijhydene.2019.11.057
14 international journal of hydrogen energy xxx (xxxx) xxx

Fig. 18 e Flow diagram of the step-by-step procedure.

simulation is always the default value of the protonic con- conductivity matches the values of membrane conductivity in
duction coefficient. The value of membrane conductivity is the literature, and the membrane resistance is less than the
calculated using Eq. (5) based on the values of simulation total cell resistance, then the default value of protonic con-
output data. If the value of the simulated membrane duction coefficient can be fixed for the next simulation.

Please cite this article as: Arif M et al., A systematic approach for matching simulated and experimental polarization curves for a PEM
fuel cell, International Journal of Hydrogen Energy, https://doi.org/10.1016/j.ijhydene.2019.11.057
 international journal of hydrogen energy xxx (xxxx) xxx 15

Otherwise the value of protonic conduction coefficient is iteration of all the input parameters is required to obtain the
tuned until it meets the two criteria. The values of contact optimal solution set.
resistance can be calculated by subtracting membrane resis- The closeness of the fit between simulated and experi-
tance from the total cell resistance. In our case, the default mental polarization curves may initially be assessed by visual
value e i.e.: 1 e of protonic conduction coefficient satisfied the inspection, but as the match improves quantitative statistical
criteria mentioned above, so the value of protonic conduction criteria must be used, such as mean squared difference, sum
coefficient was kept constant in all simulations. of squared differences, and coefficient of determination (R2).
The concentration losses in the PEM fuel cell depend on the As shown in Table 6, the maximum percentage difference
amount of liquid water present in the PEM fuel cell (bottom between simulation and experimental values in the example
box in Fig. 18). The liquid water removal rate can vary the presented in this paper for simulation 4 is ± 5%, and this dif-
amount of liquid water present in the fuel cell: the lower the ference was progressively reduced with each simulation
assumed rate, the more water remains in the cell. The water conducted. The fit is said to be perfect if the value of the sum
removal coefficient can be used as a fitting parameter to of squared differences is zero and R2 is one [42]. From Table 7,
match the simulated and experimental curves in the con- the values of the sum of squared differences and R2 for
centration region. The values of the current densities in the simulation-4 are 9.50 E06 and 0.99, respectively, which are
activation and ohmic polarization regions will not be affected very close to the ultimate limit.
significantly by varying the value of the coefficient of the The selected set of input parameters has been calibrated
liquid water. on the basis of their physical significance in the different po-
With the default value of water removal coefficient larization regions of the polarization curve. Driven by the
(5 E05 s/m), the simulated polarization curve overestimated physical meaning of the set of parameters, it poses dominant
the experimental polarization curve. By decreasing the value effects on the polarization curve compared to others. Hence
of water removal coefficient to 5 E15 s/m, more water will clog the values of these parameters have been calibrated within an
the GDL pores, and less oxidant will reach the catalyst for acceptable physical range to achieve a close agreement with
electrochemical reaction. By comparing Figs. 15 and 16, it can the experimental measurements. Without a solid theory de-
clearly be observed that concentration losses start at medium viation or experimental measurement, these parameters
current densities but get more severe at higher current should not be calibrated. So, the focus of this paper is on a
densities. systematic approach to capture the system performance with
As mentioned in section Evaluating closeness of fit, there physical consideration.
are also some other input parameters in the ANSYS PEM Fuel If necessary, the values of the maximum percentage dif-
Cell Module e like: exponent of pore blockage, concentration ference can be further reduced, by tuning other input pa-
dependence, local species concentration reference values e, rameters that have so far been kept constant, including the
that can be varied to get a match between simulated and exponent of pore blockage concentration dependence, and
experimental polarization curves but tuning all of them to get local species concentration reference values.
the fit between the two curves will increase the computational In the present work, this step-by-step procedure has been
time. The average time taken for a solution to be converged in implemented manually, with each set of input parameters fed
this study was 4 hours for each voltage point. Hence tuning into the model, and the resulting polarization curve compared
only the primary input parameters (charge transfer co- with the experimental curve outside of the model itself. It is
efficients, exchange current densities, protonic conduction noteworthy, however, that the entire step-by-step matching
coefficient, and coefficient of liquid removal) can match procedure can be converted into an algorithm and incorpo-
simulated and polarization curve with less computational rated into an expanded simulation model, which then sys-
time. If the closeness of fit is still less than required, additional tematically sought the best fit in terms of say coefficient of
parameters can also be tuned. determination (R2) by tuning all the input parameters. In this
systematic procedure there will be always an opportunity to
expand this procedure to all the input parameters that have
Discussion values that cannot be found independently before the
simulation.
Matching a simulated and experimental polarization curve is In a complex model of a fuel cell such the ANSYS PEM Fuel
a time-consuming practice if a systematic procedure is not Cell Module and other similar softwares, with a multiplicity of
followed. The systematic step-by-step procedure presented in input parameters that cannot be determined directly and
this paper can substantially reduce the research personnel, individually by experiment, it is always possible that the so-
and computer time, required to achieve this match. lution set found that gives a close fit to an experimental curve
The procedure relies on an understanding of how different is not unique. To test for uniqueness, the best course of action
input parameters influence the polarization curve in its three is to run the model for a particular cell with fixed character-
regions to varying degrees. However, tuning one parameter at istics, and hence by expectation a unique set of input pa-
a time can improve the fit in one polarization region, but it rameters that define these characteristics, under several
may worsen it in the other regions. For example: by varying different experimental conditions, to check that the best-fit
value of liquid removal coefficient can fit the concentration set of input parameters does not change significantly. Of
losses region but may miss-match the already fitted values in course, it is always important to look for experiments that can
the ohmic polarization region. Hence overall some degree of independently measure these values, or at least indicate a

Please cite this article as: Arif M et al., A systematic approach for matching simulated and experimental polarization curves for a PEM
fuel cell, International Journal of Hydrogen Energy, https://doi.org/10.1016/j.ijhydene.2019.11.057
16 international journal of hydrogen energy xxx (xxxx) xxx

likely range for these. For example, accurate data on the A step-by-step procedure has been developed to match
voltage e current density relationship at very low currents can simulated and experimental polarization curves and
give ranges for the values of the charge transfer coefficient shown to work effectively for the case of a single PEM fuel
and exchange current density for the cathode via Tafel plots. cell under two sets of operating conditions. The same
But a Tafel plot assumes only the cathode has a significant best-fit values for key input parameters defining the
overpotential, with the anodic overpotential assumed negli- properties of the cell itself were used under both condi-
gible. In practice the latter can be significant [43], so there will tions, and a close match between simulated and experi-
be some error in the cathodic charge transfer coefficient and mental curves was found. This lends support to the
exchange current density thereby estimated. Moreover, the uniqueness of the best-fit solution set. In applying this
Tafel plot gives no information on the corresponding anodic procedure in practice, it may be necessary to undertake
values. further checks of the uniqueness of the solution by, for
Cells with the same MEAs should have the same exchange example, comparing the simulated and experimental po-
current densities, even if the flow channel configurations larization curves for two cells with the same MEA and
varied, or the cells are operated under different operating GDLs (and hence many of the same input parameters) but
conditions. with different flow channel configurations.
It remains important to be aware that, due to the multi- In future, this systematic procedure may be incorporated
plicity of numerical input parameters in a model such as the as an algorithm within an expanded simulation package,
ANSYS Fuel Cell Module, a closer agreement between a which then automatically and systematically seeks the best
simulated and experimental curve does not guarantee that fit, in terms of say coefficient of determination (R2), by tuning
the solution set for these input parameters is unique. But the all the input parameters.
probability of this not being the case is reduced if additional The findings of this study can further contribute to using
simulations of the cell operating under different conditions comprehensive computer simulation models of PEM fuel cells,
continue to show close agreement between the simulation such as the ANSYS PEM Fuel Cell Module, as a tool for opti-
and experiment. As shown in the additional simulation mising the design of PEM fuel cells e for example, the optimal
(Fig. 17), under different operating condition, the model was flow channel configuration, and membrane electrode assem-
able to capture the system characteristics working under bly and GDL characteristics e and diagnosing areas of poor
different operating conditions. performance of a particular cell. This procedure may also be
The step-by-step procedure explained here can be applied extended in the future to modelling fuel cell stacks.
to any PEM fuel cell to match a simulated and experimental
polarization curves and hence calibrate the model. Once the
best fit values for the input parameters are found or the model
Acknowledgements
has been calibrated, under particular set of operating condi-
tions like; operating temperatures, relative humidities and
The authors are grateful to the members of the Reversible Fuel
gasses mass flow rates, this model maybe used to predict the
Cell team in the School of Engineering at RMIT University for
performance of the same cell under different operating
use of data on the performance of a single-cell PEM URFC; and
conditions.
for the general funding support for URFC development
received from the Australian Department of Defence through
the Defence Science Technology Group for the Capability
Conclusion Technology Demonstrator project, “The development and
demonstration of a low signature, rechargeable and portable
This paper has highlighted the input parameters that are
energy supply using reversible hydrogen fuel cells to support
influencing the different polarization regions of the PEM fuel
forward operating bases” (2015e2017; grant number CTD
cell polarization curves. It was found that exchange current
2014-3).
densities and charge transfer coefficients mainly control the
activation losses in the PEM fuel cell. To match the activation Nomenclature
losses region of the simulation and experimental polarization
curve, these two input parameters should first be tuned. The j Reference exchange current density per active
key factors responsible for the ohmic losses are membrane
surface area
resistance and contact resistance. Membrane protonic con-
z Specific active surface area
duction coefficient value can be tuned to increase and
[], []ref Local species concentration, reference value
decrease the value of the membrane conductivity to get the fit
g Concentration dependence exponent
between experimental and simulation polarization curves in
a Charge transfer coefficient
the ohmic polarization region. The water removal coefficient
R Universal gas constant
influences mass transport losses: the higher its value, the less
T Temperature
will be the concentration losses at higher current densities.
F Faraday constant
The value of the water removal coefficient can be adjusted to s Electrical conductivity
get the sharp turn down in the mass transport losses region at f Electric potential
high current densities. It was also observed that the concen- R Volumetric transfer current
tration losses begin at medium current densities and get more
b Protonic conduction coefficient
critical as current densities increases.

Please cite this article as: Arif M et al., A systematic approach for matching simulated and experimental polarization curves for a PEM
fuel cell, International Journal of Hydrogen Energy, https://doi.org/10.1016/j.ijhydene.2019.11.057
 international journal of hydrogen energy xxx (xxxx) xxx 17

u Protonic conduction exponent [9] Kone J-P, Zhang X, Yan Y, Hu G, Ahmadi G. Three-
E Activation energy dimensional multiphase flow computational fluid dynamics
l Water content models for proton exchange membrane fuel cell: a
theoretical development. J Comput Multiph Flows
M Molecular mass
2017;9(1):3e25.
D Gas species diffusivity [10] Bernardi MD, Verbrugge WM. A mathematical model of the
s Liquid saturation solid-polymer-electrolyte fuel cell. J Electrochem Soc
r Exponent of pore blockage 1992;139(9):2477e91.
ε Porosity [11] Thomas FF, John N. Water and thermal management in
f Flux solid-polymer-electrolyte fuel cells’. J Electrochem
Q Liquid removal coefficient 1993;140(5):1218e25.
[12] Jung SY, Trung VN. An along-the-channel model for proton
p Pressure
exchange membrane fuel cells. J Electrochem
r Density 1998;145(4):1149e59.
V Velocity [13] Mazumder S, Cole JV. Rigorous 3-D mathematical modeling
ṙ Contact resistivity of PEM fuel cells e II. Model predictions with liquid water
transport’. J Electrochem Soc 2003;150(11):A1510e7.
Subscripts [14] Moein-Jahromi M, Kermani MJ. Performance prediction of
an Anode PEM fuel cell cathode catalyst layer using agglomerate
cat Cathode model. Int J Hydrogen Energy 2012;37(23):17954e66.
ref Reference [15] Xing L, Mamlouk M, Scott K. A two dimensional agglomerate
sol Solid model for a proton exchange membrane fuel cell. Energy
2013;61:196e210.
mem Membrane
[16] Limjeerajarus N, Charoen-Amornkitt P. Effect of different
w Weight flow field designs and number of channels on
H2 O Water performance of a small PEFC. Int J Hydrogen Energy
eff Effective 2015;40(22):7144e58.
liq Liquid water [17] Macedo-Valencia J, Sierra JM, Figueroa-Ramı́rez SJ, Dı́az SE,
c Capillary Meza M. 3D CFD modeling of a PEM fuel cell stack. Int J
Hydrogen Energy 2016;41(48):23425e33.
[18] Heidary H, Kermani MJ, Prasad AK, Advani SG, Dabir B.
Numerical modelling of in-line and staggered blockages in
Appendix A. Supplementary data parallel flowfield channels of PEM fuel cells’. Int J Hydrogen
Energy 2017;42(4):2265e77.
Supplementary data to this article can be found online at [19] Chowdhury MZ, Genc O, Toros S. Numerical optimization of
https://doi.org/10.1016/j.ijhydene.2019.11.057. channel to land width ratio for PEM fuel cell. Int J Hydrogen
Energy 2018;43(23):10798e809.
[20] Havaej P, Kermani MJ, Abdollahzadeh M, Heidary H,
references Moradi A. A numerical modeling study on the influence of
catalyst loading distribution on the performance of Polymer
Electrolyte Membrane Fuel Cell. Int J Hydrogen Energy
2018;43(21):10031e47.
[1] Strachan N, Foxon TIM, Fujino J. Policy implications from the
n B. Effects of gas diffusion layer deformation on
[21] Li S, Sunde
Low-Carbon Society (LCS) modelling project. Clim Policy
the transport phenomena and performance of PEM fuel cells
2008;8(1):S17e29.
with interdigitated flow fields. Int J Hydrogen Energy
[2] Luo X, Wang J, Dooner M, Clarke J. Overview of current
2018;43(33):16279e92.
development in electrical energy storage technologies and
[22] Carcadea E, Varlam M, Marinoiu A, Raceanu M, Ismail MS,
the application potential in power system operation. Appl
Ingham DB. Influence of catalyst structure on PEM fuel cell
Energy 2015;137:511e36.
performance e a numerical investigation. Int J Hydrogen
[3] Chen H, Cong TN, Yang W, Tan C, Li Y, Ding Y. Progress in
Energy 2019;44(25):12829e41.
electrical energy storage system: a critical review. Prog Nat
[23] Zalitis CM, Sharman J, Wright E, Kucernak AR. Properties of
Sci 2009;19(3):291e312.
the hydrogen oxidation reaction on Pt/C catalysts at
[4] Staffell I, Scamman D, Velazquez Abad A, Balcombe P,
optimised high mass transport conditions and its relevance
Dodds PE, Ekins P, Shah N, Ward KR. The role of hydrogen
to the anode reaction in PEFCs and cathode reactions in
and fuel cells in the global energy system. Energy Environ Sci
electrolysers. Electrochim Acta 2015;176:763e76.
2019;12(2):463e91.
[24] Harvey D, Pharoah JG, Karan K. A comparison of different
[5] Hosseinzadeh E. Modeling and design of hybrid PEM fuel cell
approaches to modelling the PEMFC catalyst layer. J Power
systems for lift trucks. DTU Mechanical Engineering; 2012.
Sources 2008;179(1):209e19.
[6] Bednarek T, Tsotridis G. Issues associated with modelling of
[25] Ju H, Wang C-Y. Experimental validation of a PEM fuel cell
proton exchange membrane fuel cell by computational fluid
model by current distribution data, vol. 151; 2004.
dynamics. J Power Sources 2017;343:550e63.
p. A1954e60. 11.
[7] Iranzo A, Mun ~ oz M, Rosa F, Pino J. Numerical model for the
[26] Lum KW, McGuirk JJ. Three-dimensional model of a
performance prediction of a PEM fuel cell. Model results and
complete polymer electrolyte membrane fuel cell e model
experimental validation. Int J Hydrogen Energy
formulation, validation and parametric studies. J Power
2010;35(20):11533e50.
Sources 2005;143(1):103e24.
[8] Berning T, Djilali N. Three-dimensional computational
[27] Mench MM, Wang CY, Ishikawa M. In: Situ current
analysis of transport phenomena in a PEM fuel cell - a
distribution measurements in polymer electrolyte fuel cells,
parametric study. J Power Sources 2003;124(2):440e52.
vol. 150; 2003. A1052e9. 8.

Please cite this article as: Arif M et al., A systematic approach for matching simulated and experimental polarization curves for a PEM
fuel cell, International Journal of Hydrogen Energy, https://doi.org/10.1016/j.ijhydene.2019.11.057
18 international journal of hydrogen energy xxx (xxxx) xxx

[28] Ansys. ANSYS fluent fuel cell module manual. 2017. [36] Tijani AS, Binti Kamarudin NA, Binti Mazlan FA.
[29] Wu X. The influence of active area and stacking on PEM fuel Investigation of the effect of charge transfer coefficient (CTC)
cell performance: a simulation modelling and experimental on the operating voltage of polymer electrolyte membrane
investigation. Doctor of Philosphy thesis. RMIT University; (PEM) electrolyzer. Int J Hydrogen Energy
2017. 2018;43(19):9119e32.
[30] Arvay A, Ahmed A, Peng XH, Kannan AM. Convergence [37] Soderberg JN, Co AC, Sirk AHC, Birss VI. Impact of porous
criteria establishment for 3D simulation of proton exchange electrode properties on the electrochemical transfer
membrane fuel cell. Int J Hydrogen Energy 2012;37(3):2482e9. coefficient. J Phys Chem B 2006;110(21):10401e10.
[31] Huang X, Zhang Z, Jiang. J ’fuel cell Technology for [38] Larminie James DA. Fuel cell systems explained. 2nd ed.
distributed generation: an overview, vols. 9e13; July 2006. England: John Wiley & Sons Ltd; 2003.
p. 1613e8. [39] Mishra V, Yang F, Pitchumani R. Measurement and
[32] Min CH, He YL, Liu XL, Yin BH, Jiang W, Tao WQ. Parameter prediction of electrical contact resistance between gas
sensitivity examination and discussion of PEM fuel cell diffusion layers and bipolar plate for applications to PEM fuel
simulation model validation: Part II: results of sensitivity cells. J Fuel Cell Sci Technol 2004;1(1):2e9.
analysis and validation of the model. J Power Sources [40] Lufrano F, Staiti P, Minutoli M. Influence of nafion content in
2006;160(1):374e85. electrodes on performance of carbon supercapacitors. J
[33] Vetter R, Schumacher JO. Experimental parameter Electrochem Soc 2004;151(1):A64e8.
uncertainty in proton exchange membrane fuel cell [41] Hu M, Gu A, Wang M, Zhu X, Yu L. Three dimensional, two
modeling. Part II: sensitivity analysis and importance phase flow mathematical model for PEM fuel cell: Part I.
ranking. J Power Sources 2019;439:126529. Model development. Energy Convers Manag
[34] Zhao D, Dou M, Zhou D, Gao F. Study of the modeling 2004;45(11):1861e82.
parameter effects on the polarization characteristics of the [42] Sayad DS. Model evaluation e regression. 2019. viewed 25/05/
PEM fuel cell. Int J Hydrogen Energy 2016;41(47):22316e27. 2019, https://www.saedsayad.com/model_evaluation_r.htm.
[35] Biaku CY, Dale NV, Mann MD, Salehfar H, Peters AJ, Han T. A [43] Doddathimmaiah A, Andrews J. Theory, modelling and
semiempirical study of the temperature dependence of the performance measurement of unitised regenerative fuel
anode charge transfer coefficient of a 6kW PEM electrolyzer. cells. Int J Hydrogen Energy 2009;34(19):8157e70.
Int J Hydrogen Energy 2008;33(16):4247e54.

Please cite this article as: Arif M et al., A systematic approach for matching simulated and experimental polarization curves for a PEM
fuel cell, International Journal of Hydrogen Energy, https://doi.org/10.1016/j.ijhydene.2019.11.057