International Journal of Heat and Mass Transfer 189 (2022) 122705

Contents lists available at ScienceDirect

International Journal of Heat and Mass Transfer

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

Validation methodology for PEM fuel cell three-dimensional

simulation
Biao Xie a, Meng Ni b, Guobin Zhang c, Xia Sheng d, Houwen Tang e, Yifan Xu e,
Guizhen Zhai e, Kui Jiao a,∗
a
State Key Laboratory of Engines, Tianjin University, 135 Yaguan Rd, Tianjin, 300350 China
b
Department of Building and Real Estate, Research Institute for Sustainable Urban Development (RISUD) and Research Institute for Smart Energy (RISE),
Hong Kong Polytechnic University, Hung Hom, Kowloon, Hong Kong, China
c
MOE Key Laboratory of Thermo-Fluid Science and Engineering, Xi’an Jiaotong University, 28 Xianning West Road, Xi’an, Shanxi, 710049 China
d
New Energy Development Inst, China FAW Group Co. Ltd, 1 Xinhongqi Street, 130013 Changchun, China
e
Shanghai H-Rise New Energy Technology Co. Ltd, 168 Xinsiping Road, Shanghai, 200241 China

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

Article history: For modeling and simulation of proton exchange membrane (PEM) fuel cell, validation has been an essen-
Received 10 December 2021 tial and challenging task. This study implements a comprehensive validation including both overall cell
Revised 19 January 2022
performance and local distribution characteristics under different operating conditions with experimental
Accepted 13 February 2022
data from two public sources. Polarization curve, cell ohmic resistance, current density distribution and
Available online 7 March 2022
temperature distribution are all involved. A “three dimensional + one dimensional” (“3D+1D”) model is
Keywords: adopted which simplifies part of cell components in order to boost the calculation efficiency. The val-
PEM fuel cell idation methodology is clarified by listing those undetermined model parameters and analyzing their
Validation “accessibility” as well as correlations with the three kinds of voltage losses (activation, ohmic and mass
Three-dimensional simulation transfer). It is found that the control regions of ohmic voltage loss and concentration voltage loss overlap
Current density distribution among a wide current density range, which may lead to misjudgment in the validation process. The de-
Channel liquid water
tails of parameter adjustment are also shared. Simulation results of the two validation tests both obtain
decent agreement with the experiments and reflect consistent variation trends as the condition changes.
The liquid water in gas channel is proved to have a double effect on cell performance and should be
taken into careful consideration especially under low humidification and high current density working
conditions.
© 2022 Elsevier Ltd. All rights reserved.

1. Introduction transport phenomena occurring in the fuel cell as working con-

Proton exchange membrane (PEM) fuel cell is a device that con-
verts chemical energy into electricity through electrochemical re-
action and has been widely considered as a promising choice for
automobile power source [1]. Recently, Toyota has released its new
version of fuel cell vehicle (FCV), MIRAI second generation [2],
which confirms the potential and illustrates that the technology
development of PEM fuel cell has advanced to a new stage. As the
power density of PEM fuel cell stack has jumped to higher level
(4–5 kW L−1 ) and keeps rising, the current density at rated oper-
ating condition may attain 3 A cm−2 or higher, which means much
quicker reaction consumption, water and heat generation, and so
could bring unforeseeable problems. Consequently, it is of great
significance to gather a deep understanding about the underlying
∗
Corresponding author.
E-mail address: kjiao@tju.edu.cn (K. Jiao).
dition evolves. Numerical simulation usually functions as an effi-
cient tool in the pre-development stage [3]. Simulation tools with
high efficiency, decent validity and good applicability are urgently
needed to ensure the effectiveness of preliminary evaluation for
designing and optimization work.
As for the macroscale PEM fuel cell models, it is typical to
make trade-off between complexity and efficiency, as the scale
of simulation domain varies from one-dimensional (1D) to three-
dimensional (3D) [4] along with different assumptions and simpli-
fications. Consequently, validation process is indispensable to en-
sure the validity of utilizing these models, qualitatively or quan-
titatively. 3D modeling and simulation benefits from the capabil-
ity of reproducing the geometry structure of bipolar plate (BP) and
is preferred by studies with designing purpose [5,6]. However, the
validation requirements of 3D simulation tools concomitantly be-
come higher because it presents more comprehensive information,
some of which also enters the detectable scope of laboratory. Usu-

https://doi.org/10.1016/j.ijheatmasstransfer.2022.122705
0017-9310/© 2022 Elsevier Ltd. All rights reserved.
B. Xie, M. Ni, G. Zhang et al. International Journal of Heat and Mass Transfer 189 (2022) 122705

Nomenclature MPL microporous layer

m mass
A specific surface area (m−1 ) mix gas and liquid mixture
C gas molar concentration (mol m−3 )/specific heat ca- mw membrane water
pacity (J mol−1 K−1 ) n number of computing node
D diffusivity (m2 s−1 ) O2 oxygen
EW equivalent weight of ionomer (kg mol−1 ) pt platinum
F faraday’s constant (C mol−1 ) ref reference
H henry’s constant (Pa m3 mol−1 ) sat saturation state
h latent heat of water (J mol−1 ) T temperature
i exchange current density (A m−2 ) ∗ correction
j electrochemical reaction rate in current form
(A m−3 )
K intrinsic permeability (m2 ) ally, polarization curve is used for verification as the most com-
k relative permeability/thermal conductivity mon characterization method. However, evidence suggests that one
(W m−1 K−1 ) polarization curve is not nearly enough to prove the reliability of
M molar mass (kg mol−1 ) a model with a host of given parameters, i.e., different parame-
m specific mass (kg m−2 ) ter combinations can draw similar results [7,8]. Actually, multi-
P pressure (Pa) conditional comparison has been gradually accepted by more and
R universal gas constant (J mol−1 K−1 )/transport resis- more numerical modeling researchers [9,10,11], e.g., adopting two
tance (s m−1 ) or more polarization curves at different operating conditions or cell
RH relative humidity configurations to guarantee the quality of validation procedure. An-
S source term (kg m−3 s−1 , W m−3 …) other feasible way to obtain more credible results is to compare
s liquid water saturation the cell ohmic loss as well as output voltage at different work-
T temperature (K) ing current density [12]. Even so, the validation indexes mentioned
u
 velocity vector (m s−1 ) above are all lumped or global features of PEM fuel cell, which
V output voltage (V) contain very limited information of local characteristics.
Yi gas species mass fraction In order to conduct more comprehensive and convincing vali-
dation, comparing the spatial distribution of key parameters, e.g.,
Greek letters current density, is another important means. Actually, methods of
α transfer coefficient spatially resolved measurement with segmented bipolar plate (BP)
δ thickness (m) [13] or specially designed membrane electrode assembly (MEA)
ε porosity or phase fraction [14] have been proposed by prior work such as printed circuit
ζpt/c platinum weight percentage of Pt/carbon catalyst board method [15], resistance array method [16], Hall sensor
ζim/c mass ratio of ionomer to carbon method [17], et al. And there has been quite a few valuable data
η overpotential (V) which are available for simulation researchers. However, relatively
θ correction coefficient scant studies that have implemented distribution validation are re-
κ charge conductivity (S m-1 ) ported on PEM fuel cell modeling and simulation. Ju and Wang
λ membrane water content [18] performed early comparison of current distribution between
μ dynamic viscosity (kg m−1 s−1 ) simulation results and measured data. They used 9 average val-
ρ density (kg m−3 ) ues of segmental current density along the channel. Li et al [19].
σ surface tension coefficient (N m−1 ) underlined the in-plane difference of current distribution between
ϕ charge potential (V) channel and land. They found that the position of current density
peak would change as operating conditions varied, indicating the
Subscripts and superscripts importance of understanding local characteristics. Hakenjos et al
0 standard state [20]. measured the current density and the high frequency resis-
a anode tance (HFR) of a 45 segmented flow field plate (3 rows of 15).
act active They validated their model at three different air flow rates and
ave average only the 50 sccm case showed a distinct gap between simulation
BP bipolar plate and measurement. There is no doubt that higher spatial resolu-
CL catalyst layer tion of segment directly improves the accuracy of measurement.
c cathode/carbon support Nevertheless, it also adds to the difficulty and complicity, not only
ECSA electrochemical active surface area for experiment setup but also for model validation. Carnes et al
eff effective [21]. increased the number of segments to 100 on a 50 cm2 single
ele electronic cell with 10 × 10 arrangement. They compared their model results
GDL gas diffusion layer with experimental data under different inlet humidity and working
g gas temperature. The validation results showed acceptable trend with
i gas species less than 30% local error. Similarly, Zhang et al [22]. compared two
im ionomer sets of experimental data from different sources (including Carnes
im/c ionomer/carbon et al [21].) with their simulation results of two cell domain, re-
ion ionic spectively. Fink et al [23]. conducted current distribution valida-
local local characteristic tion using a simulation domain of metallic BP and considered the
lw liquid water effect of manufacturing assembly and welding spots. Tardy et al
MEM membrane [24]. validated the liquid water heterogeneities of a large area PEM
fuel cell by extracting liquid water thickness from neutron imaging

2
B. Xie, M. Ni, G. Zhang et al. International Journal of Heat and Mass Transfer 189 (2022) 122705

measurement. It is also worth mentioning that both Fink et al. and

Tardy et al. made some simplification on the 3D transport in MEA
to facilitate the computation process. This is kind of in common
with this study and will be clarified in next section. Hao et al [25].
gave a comprehensive validation of current, temperature and liquid
water distribution under different working conditions with differ-
ent cathode platinum (Pt) loadings. The experimental data origi-
nated from a U.S. Department of Energy (DOE) project [26] and
the measuring points was set by 4 × 25 arrangement. The cur-
rent and temperature are simultaneously measured on anode side
and cathode side, respectively. The completeness of Hao’s work is
marvelous and stands out from most of peer studies. However, the
ointment is that only part of the cases showed good agreement
on the comparison of current density and temperature contours.
Distinct difference between experiment and simulation could still
be seen under low cathode inlet humidity and low Pt loadings. Of
course, there should still be some unknown factors, either for the
understanding of transport mechanism or the uncertainty of exper-
iment operation.
Validation plays an essential part in simulation studies and
could be challenging and time-consuming. To the authors’ knowl-
edge, only a tiny quantity of PEM fuel cell modeling studies has
implemented model validation with adequate data and in-depth
analysis. And the details of model validation are seldom men-
tioned. In this study, a comprehensive validation will be carried
out including both overall (polarization curve and ohmic loss) and
local characteristics (current and temperature distribution) based
on two different experimental data sources with a “3D+1D” PEM Fig. 1. Flow field layouts of the two computational domains for model validation.
fuel cell model. Different inlet relative humidity, operating temper- (a) Active area: 50 cm2 ; Anode: triple serpentine flow field; Cathode: quadruple
ature and flow arrangement are all involved, which is also rare in serpentine flow field [27]. (b) Active area: 40 cm2 ; Anode: 11 zig-zag gas channels
and 12 straight coolant channels; Cathode: 22 zig-zag gas channels and 12 straight
the open literature. The validation methodology will be clarified
coolant channels [25].
by listing those undermined model parameters and elucidating the
adjustment strategy. At last, the effect of liquid water in gas chan-
nel will be discussed based on the simulation results.

2. Implementation method

2.1. Computational domain

Fig. 1 shows the flow field layouts of two computational do-

mains for this study. In the first experiment, current density dis-
tribution was measured with the Hall-Effect sensor by separating
the active area into 10 segments as depicted in Fig. 1a [27]. Co-
flow arrangement was adopted with two kinds of serpentine flow
fields. For the second data source in Fig. 1b [25], current distri-
bution and temperature distribution were monitored at the same
time at anode BP surface and cathode BP surface with 4 × 25
measuring points, respectively. The zig-zag gas channels and the
straight coolant channels are reproduced. For simplicity, the part
of bipolar plate around the curving manifold is removed. All the
geometry dimension is kept consistent with the reference as much
as possible, using directly available or estimated parameters. Fig. 2. Framework of the “3D+1D” PEM fuel cell model.

2.2. “3D+1D” PEM fuel cell model

Model Assumptions
• In-plane transport in microporous layer (MPL), catalyst layer
(CL) and membrane (MEM) is neglected;
• GDL and MPL are seen as homogeneous porous media;
• Catalyst layer is composed of evenly dispersed platinum-
carbon-ionomer agglomerates [28];
• Ideal gas; Laminar flow; Steady state.
The “3D+1D” modeling method adopted in this study has been
clarified with details in our previous work [29] and the model
framework is exhibited in Fig. 2. The 3D part contains bipolar plate,
gas channel (GCH), coolant channel (not schematized here) and gas
diffusion layer (GDL), where 3D mesh is also drawn. The rest com-
ponents (MPL, CL and MEM) are mathematically embedded into
the model code with 1D description. The two parts are bridged
through two extra layers of mesh connected to anode and cath-
ode separately, in which the 1D solutions are obtained and data
are exchanged between the 3D computational fluid dynamics (CFD)
model and the 1D sub-model. For every computing node set at
the component interface (MPL-CLs, CL-MEMs), equations are built
among three adjacent nodes (numbered 0, 1, 2) and two compo-
nents (numbered 0, 1). The involved conservation equations are
listed as below.

3
B. Xie, M. Ni, G. Zhang et al. International Journal of Heat and Mass Transfer 189 (2022) 122705

3D domain, Leveret-J function [31] is adopted to obtatin the diffusion coef-

ficient of channel liquid water and also to calculate liquid water
Mass ∇ · (ρg u
 g ) = Sm (1)
saturation from liquid pressure in porous electrodes.
 ε 0 . 5
Pc = σ cos θ J (s ) (15)
Momentum ∇ · (ρg u g u g ) K
 2

= −∇ Pg + ∇ · μmix (∇ u g + ∇ u
 Tg ) − μmix ∇ u
 g + Su (2)
3 1.42(1 − s ) − 2.12(1 − s )2 + 1.26(1 − s )3 θ < 90◦
J (s ) =
  1.42s − 2.12s2 + 1.26s3 θ > 90◦
Species ∇ · (ρg u gYi ) = ∇ · ρg Di,eff ∇ Yi + Si (3)
(16)
  The electrochemical reaction rate is calculated by the Butler-
Energy ∇ · ρgCpg,eff u g T = ∇ · (keff ∇ T ) + ST (4)
Volmer equation and the cathode part is revised with agglomerate
Liquid water saturation in flow channel correction [28],
 0 . 5
 2F α η   
∇ · (ρlw u lw s ) = ∇ · (ρlw Dlw ∇ s )+Slw (5) RT CH2 2 F ( 1 − αa ) η a
0,a Apt θT,a
a a
ja = iref eff
exp − exp −
Liquid pressure in porous electrodes HH2 CHref2 RT RT
  (17)
K klw
0=∇· ρlw ∇ Plw + Slw (6)
μlw
 −1
Electronic potential RT CO2 COref2 R
  jc =  4F αc ηc   4F (1−αc )ηc  + local
0=∇· κele,eff ∇ φele +Sele (7) HO2 i A θT,c exp
ref eff
0,c pt
− exp −
RT
4 F Aim
RT

1D domain, (18)


1
Cin+1 − Cin 
1 where j(A m−3 ),
i(A m−2 ),θT , R (JA(m−1 ), H (Pa m3mol−1 K−1 ),
Species 0 = Dni,eff + 0.5δ n n
Si (8) mol ), F (C mol ), α , η(V) and Rlocal (s m ) are the electro-
−1 −1 −1

n=0
δn n=0 chemical reaction rate, the exchange current density, the specific
area, the temperature correction coefficient, the universal gas con-

1
T n+1 − T n 
1
stant, the Henry’s coefficient, the Faraday constant, the transfer co-
Temperature 0 = kni,eff + 0.5δ n STn (9)
n=0
δn n=0
efficient, the overpotential and the local gas transport resistance.
Sm (kg m−3 s−1 ), Su (N m−3 ), Si (kg m−3 s−1 ), ST (W m−3 ), Slw (kg m−3
Liquid pressure in porous electrodes s−1 ), Sele (A m−3 ) and Sion (A m−3 ) are corresponding source terms
of above conservation equations. The subscripts “g”, “lw”, “mix”,

1
ρlw K n knlw Plw
n+1
− Plw
n 1
0= + 0.5δ n Slw
n
(10) “i”, “ele”, “ion”, “eff”, “im”, “pt” and “mw” represent gas, liquid wa-
n=0
μ lw δ n
n=0 ter, mixture of gas and liquid, gas species (hydrogen, oxygen and
water vapor), electric, ionic, effective value, ionomer, platinum and
Membrane water content in ionomer
membrane water, respectively. The geometry parameters and oper-

1
ρim Dmw λn+1 − λn  1
ating conditions are given in Table 1.
0= + 0.5δ n Smw
n
(11)
n=0
EW δ n
n=0 2.3. Validation methodology
Electronic potential
In order to build a PEM fuel cell model, there are usually a

1
φele
n+1
− φele
n 1
multitude of input parameters and correlations regarding transport
0= κ n
+ 0.5δ n Sele
n
(12)
n=0
ele,eff
δ n
n=0
properties or electrochemical mechanisms. Due to the nature of
unobservability or unpredictability for most of microscale transport
Ionic potential phenomena, undetermined coefficients and correction parameters
are usually necessary. These parameters are attached with physi-

1
φion
n+1
− φion
n 1
0= κion
n
+ 0.5δ n Sion
n
(13) cal meaning to describe those factors which surely exist but re-
n=0
,eff
δ n
n=0 main unclear. On the one hand, a model needs adequate degrees
of freedom to be capable of adapting to the variation of practical
where ε , ρ (kg m−3 ), u  (m s−1 ), s, P (Pa), μ(kg m−1 s−1 ), Y , Deff (m2
conditions. On the other hand, those parameters or correlations are
−1
s ), Cp (J mol −1 K ), T (K), keff (W m−1 K−1 ), K(m2 ), klw , κeff (S
−1
often introduced or derived based on relevant research findings or
m−1 ), φ (V), Cin (mol m−3 ), δ (m), EW (kg mol−1 ) and λ are the poros-
convincing data, and thereby, should follow their valid range or ap-
ity, the density, the superficial velocity vector, the liquid water sat-
plication scope. Apart from the complexity of model itself, reason-
uration, the pressure, the viscosity, the species mass fraction, the
able and subtle adjustment of the undetermined parameters func-
effective gas diffusivity, the specific heat capacity, the temperature,
tions as the most important part to accomplish the validation pro-
the effective heat conductivity, the intrinsic permeability, the liquid
cess.
phase relative permeability, the charge conductivity, the charge po-
Normally speaking, the core of model validation is to cap-
tential, the gas molar concentration, the component thickness, the
ture the change rule of all the (ideally) or the currently-dominant
equivalent ionomer weight and the membrane water content, re-
voltage losses (activation, ohmic and mass transfer or concentra-
spectively. The u  lw in the convection term of liquid water satura-
tion) by recognizing and adjusting those associate parameters. The
tion equation in gas channel is derived from the gas velocity with
most difficult task is to coordinate all the undetermined parame-
the correction of viscosity ratio and local saturation values [30].
ters within reasonable range, i.e., without violating general knowl-
u
 lw μ g  s 3 edge or physical laws, at the same time yielding results with cor-
= (14)
u
g μlw 1 − s rect trends Table 2. has listed the undetermined parameters of

4
B. Xie, M. Ni, G. Zhang et al. International Journal of Heat and Mass Transfer 189 (2022) 122705

Table 1
Geometry parameters and operating conditions.

Parameter Validation 1 [27] (anode/cathode) Validation 2 [25] (anode/cathode)

Gas channel width (mm) 0.8 0.4

Gas channel height (mm) 0.5 0.7
Rib width (mm) 0.8 1.5/0.5
Coolant channel width (mm) NA 0.7
Coolant channel height (mm) NA 0.15
GDL thickness (μm) 180 160
MPL thickness (μm) 20 30
CL thickness (μm) 5/10 6/16
Membrane thickness (μm) 10 18
Platinum loading (mg cm−2 ) 0.1/0.4 0.1/0.3
Inlet relative humidity (%) 100/50 (case1) 0/100
25/25 (case2)
Stochiometric ratio 1.5/2.0 1.5/2.0
Back pressure (atm) 1.35/1.35 (case1) 1.0/1.5
1.0/1.0 (case2)
Operating temperature (K) 353.15 313 (case1)
333 (case2)
Coolant inlet temperature (K) NA 307.65 (case1)
327.65 (case2)

Table 2
Undetermined parameters of the “3D+1D” model in this study.

Parameter Accessibility Typical value or magnitude Voltage loss relevance

Act. Ohm. Mass.
Exchange current density (A cm−2 ) 1 An 100 –103 –
Ca 10−5 –10−2
Electrochemical active surface area (ECSA, m2 kg−1 ) 2 50–80 –
Temperature correction coefficient 0 10−2 –100 –
Transfer coefficient 1 An 0.5–0.7 –
Ca 0.3–0.5
Concentration correction exponent 0 An 0.5–2.0 ++
Ca 1.0–3.0
Open circuit voltage (V) 5 0.9–0.97
− 1
Intrinsic electric conductivity (S m ) BP 3 BP 105 –106 BP –
GDL 3 GDL 103 –104 GDL –
MPL 2 MPL 102 –103 MPL –
CL 2 CL 102 –103 CL –
Intrinsic ionic conductivity (S m−1 ) MEM 3 MEM 100 –101 –
CL 2 CL 10−1 –100
Interfacial electric resistance (m cm2 ) 3 10–50 ++
Contact angle for liquid water (°) GCH 3 GCH 90–110 GCH + GCH –
GDL 3 GDL 110–130 GDL + GDL –
MPL 2 MPL 120–140 MPL + MPL –
CL 2 CL 90–100 CL ++ CL –
Intrinsic permeability (m2 ) GCH 2 GCH 10−8 –10−9 + –
GDL 2 GDL 10−10 –10−12
MPL 1 MPL 10−11 –10−12
CL 1 CL 10−11 –10−13
Relative permeability exponent 0 3.0–5.0 – ++
Porosity GDL 3 GDL 0.6–0.7 GDL ++ GDL –
MPL 2 MPL 0.5–0.6 MPL + MPL –
CL 2 CL 0.3–0.5 CL + CL –
CL ionomer volume fraction 2 0.2–0.3 – +
CL agglomerate radius (nm) 3 25–50 +
Oxygen diffusivity in ionomer (m2 s−1 ) 3 10−10 –10−11 –
Gas diffusivity (m2 s−1 ) 3 10−5 –10−6 –
Gas diffusivity correction exponent 0 1.5–3.5 ++
Electric/Ionic conductivity correction exponent 0 1.5–2.0 ++
Intrinsic thermal conductivity (W m−1 K−1 ) 3 100 –102 +
Water phase change rate (s − 1 ) 0 Evap 100 –103 Abos – Evap -
Cond 100 –103 Deso ++ Cond +
Abos 100
Deso 100

The accessibility is a rough evaluation of the adjustment room or range for undetermined parameters. Smaller number means larger tolerance for
adjustment. Specifically,
0: Nearly unmeasurable, strongly related to the given condition without explicit pattern to follow. Deeper investigation and explanation are needed
beyond the scope of macroscale model;
1: Difficult to be measured, multiple steps are needed with assumptions and large uncertainty may exist;
2: Available for ex-situ measurement with data processing or extrapolation. Limitation and restriction could be strict;
3: Available for ex-situ measurement with decent accuracy;
4: Available for in-situ measurement with data processing or extrapolation; Deviation or error should be considered;
5: Available for in-situ measurement with decent accuracy.
The “+” and “-” represents positive or negative correlation and double symbol means high relevance. The abbreviation “an”, “ca”, “evap”, “cond”,
“abso” and “deso” represents anode, cathode, evaporation and condensation of free state water, absorption and desorption of dissolved state water
(membrane water). Blank space indicates that there is no direct correlation or the mechanism hasn’t been clearly understood and not included in
this study.
5
B. Xie, M. Ni, G. Zhang et al. International Journal of Heat and Mass Transfer 189 (2022) 122705

Fig. 3. Comparison of polarization curves between simulation and experiment

[27] under higher inlet humidification, anode 100%/cathode 50% relative humidity
(RH 100%/50%) and lower inlet humidification, anode 25%/cathode 25% relative hu-
midity (RH 25%/25%) in validation test 1.

the “3D+1D” model in this work, which are also shared by other
macroscale PEM fuel cell models (may not be completely same be-
cause of different model framework or assumptions). Their typical
ranges for usage are also given. And the accessibility is lumped in-
dex that varies from 0 to 5 and represents multiple meanings, in-
cluding whether the parameter can be obtained, accurately mea-
sured or roughly, in-situ or ex-situ, i.e., how much room it has
for adjustment and whether the mechanism is clearly understood.
Larger value of accessibility means the parameter can be measured
or accessed with less uncertainty or deviation, i.e., the parameter
could be adjusted in a more limited range. It should be noted that
though smaller accessibility represents more freedom for param-
eter adjustment, the setting of values could not be more cautious
and must obey physical laws and existing credible theories. The re-
lationship between these parameters and the three kinds of volt-
age losses is also annotated to help choose suitable parameters ac- Fig. 4. Comparison of current density distribution between simulation and experi-
cording to the practical demand, e.g., adjusting only one kind of ment [27] from two cases in validation test 1. (a) 1.25 A cm−2 under higher inlet
voltage loss or two kinds at the same time, slightly or sharply. humidification. (b) 0.81 A cm−2 under lower inlet humidification.
It should be noted that Table 2 is concluded based on authors’
knowledge and public data and may have deviations. Neverthe-
less, it provides a more intuitional guide of adjusting strategy. In cause the slope of the curve nearly maintains the same, i.e., the
Table 3, values and expressions of key model parameters used for obvious downward trend originated from oxygen shortage doesn’t
the two validation tests in this study are given. appear. And the reason should be related to the 50% relative hu-
midity of cathode inlet and the co-flow arrangement. Under rela-
tively higher current density, the area near the cathode outlet is
3. Results and discussion
susceptible to oxygen shortage. On the other hand, the membrane
hydration becomes better as current density increases near the in-
3.1. Validation 1
let region, alleviating the drying effect of unsaturated air. Conse-
quently, the combined influence of ohmic loss and concentration
As shown in Fig. 3, the simulated polarization curves give good
loss keeps the slope almost unchanged. This demonstrates that
agreement with experimental data under both higher and lower
the ohmic control region and mass transfer control region over-
inlet humidification. And Fig. 4 depicts the comparison of current
lap among a relatively wide current density range, as highlighted
density distribution of two cases between simulation and experi-
in Fig. 3. It also proves the importance of raising the validation re-
ment, at 1.25 A cm−2 and 0.81 A cm−2 under different conditions
quirement with other characteristics like cell ohmic resistance and
respectively. The consistency of the variation trend is also quite
current density distribution.
well. It can be seen that the current density decreases from in-
let to outlet in the high humidity case (Fig. 4a) and the tendency
reverses for the low humidity case (Fig. 4b). This indicates that 3.2. Validation 2
the former is at the stage that mass transfer voltage loss begins to
dominate, and the latter is in the ohmic control region. However, if To further test the validation methodology, we choose two cases
only judging from the polarization curve, view can be easily drawn of different operating temperatures (40 °C and 60 °C) from Hao
that the 1.25 A cm−2 case is still in the ohmic control region be- et al.’s work [25]. The results are shown as in Fig. 5 and 6, together

6
B. Xie, M. Ni, G. Zhang et al. International Journal of Heat and Mass Transfer 189 (2022) 122705

Table 3
Values and expressions of key model parameters used for validation.

Parameter Validation 1 (anode/cathode) Validation 2 (anode/cathode)

Exchange current density (A cm−2 ) 1.2 × 103 /1.2 × 10−2 1.2 × 103 /1.2 × 10−2
Electrochemical active surface area (ECSA, m2 kg−1 ) 7.0 × 104 7.0 × 104
 1 1

Temperature correction coefficient θT,a = exp −2800 −
T 353.15
 1 1

θT,c = exp −140 0 0 −
T 353.15
Transfer coefficient 0.5/0.5 0.5/0.5
Concentration correction exponent 0.5/1.0 0.5/1.0
Open circuit voltage (V) 0.94 (case1) 0.97
0.97 (case2)
− 1
Intrinsic electric conductivity (S m ) BP 100,000 BP 100,000
GDL 40,000 (IP) GDL 80,000 (IP)
5000 (TP) 10,000 (TP)
MPL 3000 MPL 5000
CL
 3000 1 1
CL
 5000
Intrinsic ionic conductivity (S m−1 ) κion = (0.5319λ − 0.326 ) exp −300 −
T 303.15
Interfacial electric resistance (m cm2 ) 50 12
Contact angle for liquid water (°) GCH 110 GCH 110
GDL 120 GDL 120
MPL 130 MPL 130
CL 95/95 CL 97/95
−8
Intrinsic permeability (m2 ) GCH 1.0 × 10 GCH 1.0 × 10−8
GDL 2.0 × 10−12 GDL 2.0 × 10−12
MPL 1.0 × 10−12 MPL 1.0 × 10−12
CL 1.0 × 10−13 CL 1.0 × 10−13
MEM 2.0 × 10−20 MEM 2.0 × 10−20
Relative permeability exponent 3.0 3.0
Porosity GDL 0.65 GDL 0.625
MPL
  0.5
 MPL 0.5
m 1 1 1
CL solid phase fraction εpt/c = pt + −1
δCL ρpt ζpt/c  ρc
ζim/c mpt 1
 M ρim

CL ionomer volume fraction εim = − 1 1 + lw λ
δCL ρim ζpt/c ρlw EW
CL porosity εCL = 1 − εpt/c − εim
CL agglomerate radius (nm) 25 25
Oxygen diffusivity in ionomer (m2 s−1 ) DO2 ,im = 1.14698e − 10λ0.708
 T 1.5  101325 
Gas diffusivity (m2 s−1 ) DH2 = 1.005 × 10−4
 333T.15 1.5  101325
P

DO2 = 2.652 × 10−5
333.15 P
 T 1.5  101325 
DaH2 O = 1.005 × 10−4
 333T.15 1.5  101325
P

DcH2 O = 2.982 × 10−5
333.15 P
Gas diffusivity correction exponent 3.5 2.0
Electric conductivity correction exponent 1.5 1.5
Ionic conductivity correction exponent 1.7 1.5
Water phase change rate (s−1 ) Evap 100 Evap 100
Cond 100 Cond 100
Abos 1.3 Abos 1.3
Deso 1.3 Deso 1.3

with output voltage and cell resistance. The 100 measuring points bution, temperature distribution, output voltage and cell ohmic re-
are picturized into partitioned contour and the simulated current sistance are validated at the same time and the agreement is also
density on the BP surface is also treated into 100 segments for the good. It should be noted that most of the undetermined model in-
convenience of comparison. The working current density is fixed at put parameters are kept the same as listed in Table 3, which means
1.5 A cm−2 . The overall consistency of the temperature distribution the model complexity stands well with the two validation tests
and the 25 average values (see Fig. 5b, 5c and 6b, 6c) along the and the applicability of the current methodology is good.
air flow direction is fairly well but part of the current density dis-
tribution still manifests relatively obvious deviation, especially for
the inlet region. The cause could originate from the unknown gap
3.3. Effect of liquid water in gas channel
between the real inlet condition and the ideal state of simulation
setup to some extent. However, for both experiment and simula-
The proper consideration of liquid water in gas channel has
tion, it can be seen from Fig. 5a and 6a that the current density
been a challenging and tough task for PEM fuel cell models [32],
peak at 60 °C becomes closer to the cathode inlet and the current
which also accounts for the deviation between model prediction
density near the outlet decreases distinctly compared with the 40
and experimental measurement to a large extent. The space and
°C case. This is mainly because that the inlet oxygen partial pres-
time scales of those two-phase models which consider phase in-
sure is lower at higher temperature for both fully humidified air,
teraction, dynamic contact angle or wall adhesion, are too small for
which means the effect of mass transfer voltage loss increases. For
macroscale models and the demand of computing capability could
these two different operating temperatures, current density distri-
be unaffordable even for a small cell domain. Instead, a continuous

7
B. Xie, M. Ni, G. Zhang et al. International Journal of Heat and Mass Transfer 189 (2022) 122705

Fig. 6. Comparison of (a) current density distribution, (b) temperature distribution

Fig. 5. Comparison of (a) current density distribution, (b) temperature distribution
and (c) 25 average values along the air flow direction of the 40 °C operating temper-
ature case between simulation and experiment [25] in validation test 2. The current
density distribution and temperature distribution are simultaneously measured on
the end plane of anode and cathode bipolar plates, respectively.
and (c) 25 average values along the air flow direction of the 60 °C operating tem-
perature case between simulation and experiment [25] in validation test 2.
liquid water saturation is usually solved to consider the existence
of liquid water as written in Eq. (5).
Fig. 7 shows the liquid water saturation contours in cathode
gas channel of the above 4 validation cases. It can be seen that
the liquid water saturation gains higher values around corners and

Fig. 7. Liquid water saturation contours in cathode gas channel of the 4 validation cases. (a) 1.25 A cm−2 case under 100%/50% inlet humification; (b) 0.81 A cm−2 case
under 25%/25% inlet humidification; (c) 1.5 A cm−2 case at 40 °C operating temperature; (c) 1.5 A cm−2 case at 60 °C operating temperature.

8
B. Xie, M. Ni, G. Zhang et al.
Fig. 8. Comparison of whether the liquid water saturation in gas channel is consid-
ered. (a) Polarization curve and ohmic loss under the 100%/50% inlet humification
in validation test 1; (b) Current density distribution of the 1.25 A cm−2 case; (c) 25
average current density values of the 60 °C operating temperature case in validation
test 2.
near the outlet region, where gas flow is obstructed or oxygen is
in short supply. And it has a close correlation with cell working
conditions. There is nearly no water accumulation in gas channel
for the low inlet relative humidity and low current density case
(see Fig. 7b). In Fig. 7c and 7d, the liquid water saturation distri-
bution pattern is very similar for the two cases. The reason is that
the inlet air is kept saturated and the current density is the same
Fig. 8. compares the difference of solving the liquid water satura-
9
International Journal of Heat and Mass Transfer 189 (2022) 122705
tion equation in gas channel or not for the two validation tests. In-
terestingly, the two polarization curves under higher inlet humidi-
fication in Fig. 8a are almost the same while the ohmic loss shows
apparent discrepancy, which means the liquid water in gas chan-
nel has a double effect on cell ohmic resistance and gas transport
resistance at the same time. On the one hand, the liquid water in
channel hinders the water removal from the electrode and adds to
the gas blockage. On the other hand, the retained water improves
the ionic conductivity to a certain degree under not fully humid-
ified condition. The variation trend of current density distribution
for the 1.25 A cm−2 case turns opposite in Fig. 8b also proves the
influence of channel liquid water and again the importance of vali-
dating current density distribution. For the 60 °C case of validation
2 (see Fig. 8c), it can be seen that the cell output voltage becomes
higher even the ohmic resistance increases and the current den-
sity distribution gets more uniform without the existence of liquid
water in channel. This indicates that its influence on gas transport
overcomes the ohmic part. These results illustrate that it is of great
significance to consider the liquid water in gas channel. In addition
to the negative effect on mass transfer loss, which is also focused
by most of preceding studies, the positive effect on ionic ohmic
loss should also be brought to the forefront, especially under low
humidification and high current density conditions.
4. Conclusion
In this study, a comprehensive validation process for PEM fuel
cell three-dimensional simulation is accomplished and the valida-
tion methodology is investigated and analyzed. Polarization curves,
ohmic loss, current density distribution and temperature distribu-
tion under different inlet humidification and operating tempera-
ture are compared with experimental data from two studies and
decent agreement are achieved simultaneously. Both co-flow and
counter-flow arrangement are included. The core part of model
validation, that recognizing, adjusting and coordinating the unde-
termined parameters, is clarified by listing their typical range, ac-
cessibility and relationship with three kinds of voltage losses. By
comparing the current density distribution, the ohmic voltage loss
and the concentration voltage loss manifest a combined influence
region across a relatively wide current density range, which is easy
to mislead the validation process and strongly proves the necessity
of multi-index comparison. The validation details are also shared
by giving the setting of key parameters. Additionally, the liquid
water saturation in gas channel is found to have a double effect
on cell ohmic loss and concentration loss, which should be care-
fully considered under low humidification and high current den-
sity conditions. The validation work in this study may still have
doubtful points on current model assumptions, e.g., evenly dis-
tributed contact resistance and inadequate spatial discretization for
one-dimensional domain. And the proposed methodology contains
some subjective and empirical understandings such as the defini-
tion of “accessibility” and the adjustment of individual correction
coefficients. Nevertheless, one thing is for sure that model valida-
tion should be as comprehensive as possible and the demand of
improving the model complexity would inevitably increase in or-
der to keep pace. With a mutual promotion, the misleading and
the uncertainty caused by parameter combination could be grad-
ually reduced. The results and analyses for validation methodol-
ogy in this study are expected to provide some instructions for
researchers and engineers in the field of PEM fuel cell design, to
facilitate the application of three-dimensional modeling and simu-
lation.
Declaration of Competing Interest
The authors declare that they have no conflict of interest.
B. Xie, M. Ni, G. Zhang et al.
CRediT authorship contribution statement
Biao Xie: Methodology, Software, Data curation, Writing – orig-
inal draft. Meng Ni: Methodology, Writing – review & editing.
Guobin Zhang: Methodology, Software. Xia Sheng: Project admin-
istration. Houwen Tang: Supervision, Funding acquisition. Yifan
Xu: Investigation. Guizhen Zhai: Investigation. Kui Jiao: Concep-
tualization, Formal analysis, Writing – review & editing.
Acknowledgements
This work is supported by the National Natural Science Foun-
dation of China (grant No. 51920105010) and the National Natu-
ral Science Foundation of Tianjin (China) for Distinguished Young
Scholars (Grant No. 18JCJQJC46700), and the joint supervision
scheme of The Hong Kong Polytechnic University (Project number:
P0035138; Work Program: SB3N).
References
[1] K. Jiao, J. Xuan, Q. Du, et al., Designing the next generation of proton-exchange
membrane fuel cells, Nature 595 (7867) (2021) 361–369.
[2] Full Specs of Toyota MIRAI 2021, Toyota Motor Sales, U.S.A., Inc, 2020 https:
//www.toyota.com/MIRAI/features/mileage_estimates/30 02/30 03.
[3] K. Jiao, B. Wang, Q. Du, et al., Water and Thermal Management of Proton Ex-
change Membrane Fuel Cells, Elsevier, 2021.
[4] H. Wu, A review of recent development: transport and performance modeling
of PEM fuel cells, Appl. Energy 165 (2016) 81–106.
[5] X. Zhang, X. Ma, S. Shuai, et al., Effect of micro-porous layer on PEM fuel cells
performance: considering the spatially variable properties, Int. J. Heat Mass
Transf. 178 (2021) 121592.
[6] A. Ebrahimzadeh, I. Khazaee, A. Fasihfar, Experimental and numerical investi-
gation of obstacle effect on the performance of PEM fuel cell, Int. J. Heat Mass
Transf. 141 (2019) 891–904.
[7] W. Tao, C. Min, X. Liu, et al., Parameter sensitivity examination and discussion
of PEM fuel cell simulation model validation: part I. Current status of modeling
research and model development, J. Power Sources 160 (1) (2006) 359–373.
[8] Y. Jiang, Z. Yang, K. Jiao, et al., Sensitivity analysis of uncertain parameters
based on an improved proton exchange membrane fuel cell analytical model,
Energy Convers. Manage. 164 (2018) 639–654.
[9] H. Liu, P. Li, K Wang, Optimization of PEM fuel cell flow channel dimensions—-
Mathematic modeling analysis and experimental verification, Int. J. Hydrogen
Energy 38 (23) (2013) 9835–9846.
[10] L. Hao, K. Moriyama, W. Gu, et al., Modeling and experimental validation of
Pt loading and electrode composition effects in PEM Fuel Cells, J. Electrochem.
Soc. 162 (8) (2015) F854–F867.
[11] T. Tsukamoto, T. Aoki, H. Kanesaka, et al., Three-dimensional numerical simu-
lation of full-scale proton exchange membrane fuel cells at high current den-
sities, J. Power Sources 488 (2021) 229412.
10
International Journal of Heat and Mass Transfer 189 (2022) 122705
[12] G. Zhang, L. Fan, J. Sun, et al., A 3D model of PEMFC considering detailed mul-
tiphase flow and anisotropic transport properties, Int. J. Heat Mass Transf. 115
(2017) 714–724.
[13] L. Pérez, L. Brandão, J. Sousa, et al., Segmented polymer electrolyte membrane
fuel cells—A review, Renew. Sustain. Energy Rev. 15 (1) (2011) 169–185.
[14] J. Stumper, S. Campbell, D. Wilkinson, et al., In-situ methods for the deter-
mination of current distributions in PEM fuel cells, Electrochim. Acta 43 (24)
(1998) 3773–3783.
[15] M. Schulze, E. Gülzow, S. Schönbauer, et al., Segmented cells as tool for de-
velopment of fuel cells and error prevention/prediagnostic in fuel cell stacks,
Energy 173 (2007) 19–27.
[16] Z. Liu, Z. Mao, B. Wu, et al., Current density distribution in PEFC, J. Power
Sources 141 (2) (2005) 205–210.
[17] M. Santis, S. Freunberger, M. Papra, et al., Experimental investigation of cou-
pling phenomena in polymer electrolyte fuel cell stacks, J. Power Sources 161
(2) (2006) 1076–1083.
[18] H. Ju, C.-.Y. Wang, Experimental validation of a PEM fuel cell model by current
distribution data, J. Electrochem Soc. 151 (11) (2004) A1954–A1960.
[19] J. Li, C.-.Y. Wang, A. Su, Prediction and experimental validation of in-plane cur-
rent distribution between channel and land in a PEFC, J. Electrochem. Soc. 155
(1) (2008) B64–B69.
[20] A. Hakenjos, C. Hebling, Spatially resolved measurement of PEM fuel cells, J.
Power Sources 145 (2) (2005) 307–311.
[21] B. Carnes, D. Spernjak, G. Luo, et al., Validation of a two-phase multidimen-
sional polymer electrolyte membrane fuel cell computational model using cur-
rent distribution measurements, J Power Sources 236 (2013) 126–137.
[22] G. Zhang, J. Wu, Y. Wang, et al., Investigation of current density spatial dis-
tribution in PEM fuel cells using a comprehensively validated multi-phase
non-isothermal model, Int. J. Heat Mass Transf. 150 (2020) 119294.
[23] C. Fink, N. Fouquet, Three-dimensional simulation of polymer electrolyte mem-
brane fuel cells with experimental validation, Electrochim. Acta 56 (28) (2011)
10820–10831.
[24] E. Tardy, F. Courtois, M. Chandesris, et al., Investigation of liquid water hetero-
geneities in large area PEM fuel cells using a pseudo-3D multiphysics model,
Int. J. Heat Mass Transf. 145 (2019) 118720.
[25] L. Hao, K. Moriyama, W. Gu, et al., Three dimensional computations and exper-
imental comparisons for a large-scale proton exchange membrane fuel cell, J.
Electrochem. Soc. 163 (7) (2016) F744–F751.
[26] Micro- and Macro-Scale PEMFC Transport, http://pemfcdata.org.
[27] V. Lilavivata, S. Shimpaleea, J. Zee, et al., Current distribution mapping for
PEMFCs, Electrochim. Acta 174 (2015) 1253–1260.
[28] B. Xie, G. Zhang, J. Xuan, K. Jiao, Three-dimensional multi-phase model of PEM
fuel cell coupled with improved agglomerate sub-model of catalyst layer, En-
ergy Convers. Manage. 199 (2019) 112051.
[29] B. Xie, G. Zhang, Y. Jiang, et al., 3D+1D” modeling approach toward large-s-
cale PEM fuel cell simulation and partitioned optimization study on flow field,
eTransportation 6 (2020) 10 0 090.
[30] G. Zhang, L. Wu, Z. Qin, et al., A comprehensive three-dimensional model cou-
pling channel multi-phase flow and electrochemical reactions in proton ex-
change membrane fuel cell, Adv. Appl. Energy 2 (2021) 10 0 033.
[31] M. Leverett, Capillary behavior in porous solids, Trans. AIME 142 (01) (1941)
152–169.
[32] R. Anderson, L. Zhang, Y. Ding, et al., A critical review of two-phase flow in gas
flow channels of proton exchange membrane fuel cells, J. Power Sources 195
(15) (2010) 4531–4553.