i n t e r n a t i o n a l j o u r n a l o f h y d r o g e n e n e r g y 3 5 ( 2 0 1 0 ) 5 1 7 3 e5 1 8 6

Available at www.sciencedirect.com

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

Analysis of water transport in a high pressure

PEM electrolyzer

P. Medina, M. Santarelli*
Dipartimento di Energetica, Politecnico di Torino, Corso Duca degli Abruzzi 24, 10129 Torino, Italy

article info abstract

Article history: This paper analyzes, through experimental data and a transport model, the water trans-
Received 10 October 2009 ported through the membrane under different operating conditions in a on a Proton
Received in revised form Exchange Membrane (PEM) electrolyzer operating with a high-pressure gradient across the
22 February 2010 membrane from the cathode (high-pressure) side to the anode (nearly ambient-pressure)
Accepted 27 February 2010 side. The phenomena involved in this movement are described and analyzed, with
Available online 3 April 2010 a focus on the electro-osmotic drag coefficient, neo. We have observed that the behavior of
the hydraulic percolation determines the results obtained for the electro-osmotic drag,
Keywords: while the contribution of the water diffusion is negligible. In general, the cathode pressure
Electrolysis significantly reduces the water transport (a positive effect). Also, operation at lower current
High pressure density reduces the net electro-osmotic drag coefficient, ng; therefore, the best operation
Water drag strategy for obtaining dried hydrogen at the cathode is to impose high cathode pressure
and low current density.
ª 2010 Professor T. Nejat Veziroglu. Published by Elsevier Ltd. All rights reserved.

1. Introduction provided that mechanical support is guaranteed from the

membrane.
Among different types of electrolyzers, the PEM electrolyzer Improving water management in PEM is one of the most
using a Nafion membrane is of increasing interest because of interesting topics for improving the efficiency and extending
its compactness, which is linked to a higher achievable the operation of fuel cells and electrolyzers. It is well known
current density, and because of the very high purity of the that the humidity in fuel cells should be well controlled
produced hydrogen [1e3]. within a specific range at the membrane in order to avoid
Advanced systems have recently been developed in order cell flooding or drying, phenomena which cause a drop in
to produce hydrogen at high pressure with an efficient, clean output or cell failure, respectively. In electrolyzers,
and safe production process. Some PEM electrolyzers operate membrane humidification is provided by the high flow of
with a high-pressure gradient across the membrane from the pumped water. Thus, it has been observed that elevated
cathode (high-pressure) side to the anode (nearly ambient- amounts of water are pushed from anode to cathode. The
pressure) side. hydrogen gas must be separated from this large amount of
This device permits elevated hydrogen supply pressures water before the hydrogen can be stored. This is true for
without the need for mechanical compression. Hydrogen storage as a compressed gas, and it is especially true for solid
molecules are reduced electrochemically at the electrode adsorption phases. This type of separation requires struc-
surface and raise the pressure in the cathode chamber, tures that operate at high pressures. Therefore, controlling

* Corresponding author.
E-mail address: massimo.santarelli@polito.it (M. Santarelli).
0360-3199/$ e see front matter ª 2010 Professor T. Nejat Veziroglu. Published by Elsevier Ltd. All rights reserved.
doi:10.1016/j.ijhydene.2010.02.130
5174 i n t e r n a t i o n a l j o u r n a l o f h y d r o g e n e n e r g y 3 5 ( 2 0 1 0 ) 5 1 7 3 e5 1 8 6

Symbols Gpi water percolation mass flow (g/s)

2 tm membrane thickness (cm)
A membrane active area (cm )
Tcella cell temperature (
C)
Ca water concentration at anode (mol/cm3)
X molar fraction (%)
Cc water concentration at cathode (mol/cm3)
Dw water diffusion coefficient (m2/s) Greek
i current density (A/cm2) r water density (g/cm3)
MH2 O water molar weight (g/mol) kD hydraulic permeability of porous media (m2)
n number of cells of the stack m fluid dynamic viscosity (Pa s)
Gi i-element mass flow (g/s)
Constant
Geo electro-osmotic drag mass flow (g/s)
F Faraday constant (96,485 C/mol)
Gd water diffusion mass flow (g/s)

(i.e., reducing) water transport could represent a useful strategy using a high-pressure gradient across the membrane between
for simplifying a plant’s structure. the cathode and the anode side.
This paper considers the water transport inside a Nafion In this paper, we describe an experimental analysis of
membrane for the case of a high-pressure PEM electrolyzer. water transport from the anode to the cathode chambers
The system is described through the experimental analysis of inside a high-pressure PEM electrolyzer. We used a Design of
the phenomena involved in transport of water molecules Experiments based on Yates’ Technique to process data on
across the perfluorinated membrane. water measured inside the separation vessel after a long-
The main interest of this analysis is to evaluate the mass period test under different conditions. We then describe the
transport from the anode to the cathode chamber, that is very three phenomena involved in water transport across a per-
important for the integration of electrolyzing devices into fluorinated membrane, namely electro-osmotic drag, diffu-
a hydrogen production plant: controlling (i.e., reducing) water sion and percolation. We introduced all physical equations
transport could represent a useful strategy for simplify into a model to obtain the electro-osmotic drag coefficient, neo,
a plant’s structure. which represents the number of water molecules dragged by
It is also interesting to study the phenomena related to a proton during its movement across the membrane.
water transport inside a high-pressure PEM electrolyzer
because the behavior of such a system presents many
differences from that of a fuel cell. These differences are 2. Description of the laboratory equipment
generally because of the elevated pressure gradient across
the membrane and the extremely high humidification of the 2.1. Description of the analyzed stack
anode due to fluent liquid water. Under these conditions,
the humidification ratio of the membrane is at a maximum The stack utilized for the tests is the prototype Giner GS-10
and is closely related to the motion of water molecules. electrolyzer provided by Giner Electrochemical Systems LLC
There are many studies on this topic for fuel cell systems (US). The stack produces hydrogen at a mid-high pressure
[4e8], but very few are available to describe an electrolyzer’s (max 70 bar) and oxygen at low pressure (max 3.5 bar).
behavior. Maximum hydrogen output is 0.1 kg/h (1.1 Nm3/h), while
For PEM fuel cells, Maier and Eigenberger [9] reported in oxygen is produced at one-half of that volumetric rate
2003 that both the electro-osmotic drag and the hydraulic (0.55 Nm3/h). The power load of the stack at maximum
permeability coefficients are enhanced by the water content production rate is 5.6 kW. The electrolyzer is made of 12
of the membrane; moreover, the water content is increased by planar cells with an exchange surface of 160 cm2. The main
the membrane thickness. These conditions are also inter- characteristics of the electrolyzer are detailed in Table 1, and
esting for the case of a PEM electrolyzer, where the water a picture of the electrolyzer is reported in Fig. 1.
content of the membrane is at a maximum due to the high The unit has four fluid connections (water in, water and
water flow at the anode and due to the use of thick oxygen out, hydrogen out and nitrogen in) and two electrical
membranes to support the pressure gradient. Therefore, in poles (anodeecathode). Each pole has two wire connections
these conditions, water transport mechanisms are favored. for optimal current conduction.
In [10], which also referred to PEMFC, the authors reported The cell stack, installed with vertical axis, is contained
that the electro-osmotic drag coefficient increases with the between two terminal heads (end plates, or end covers), the
water content of the membrane but that it remains almost cathodic head (the top head) and the anodic head (the bottom
constant while varying the current density values. Finally, in head). The heads are linked with ties, which compress the
2009, Li et al. [11] considered the water flow in a solid polymer fittings between the cells and keep the internal fluids sealed
electrolyte (SPE) in contact with liquid water, and they inside. Each tie is supplied with tongs capable of compen-
reported that the electro-osmotic drag coefficient increases sating the dilatations of the stack, thus maintaining an
with the operation temperature of a Nafion 115 membrane. accurate pressure on the cell seals. The ties are electrically
The authors obtained a regression model (linear fit) as isolated from the anodic head, which is also isolated from the
a function of T, but they did not provide any data for operation base plate where the stack lays. The connectors are set in the
 i n t e r n a t i o n a l j o u r n a l o f h y d r o g e n e n e r g y 3 5 ( 2 0 1 0 ) 5 1 7 3 e5 1 8 6 5175

Table 1 e Electrolyzer operational characteristics. 2,4

Data Value
2,2
Temperature 50e60
C
Anode pressure 0e3.5 bar 2

Cell voltage(V)
Cathode pressure 0e35 bar (max 70 bar)
Water flow >5.5 l/min 1,8
Water resistivity at 25
C >1 MU cm
Maximum H2 in O2 content 0.5%vol 1,6 7 bar, 58 ºC
Maximum O2 in H2 content 0.5%vol
Maximum H2 production rate 1.1 Nm3/h 1,4 7 bar,42 ºC
Current at max. prod. rate 224 A (at 1.4 A/cm2)
70 bar, 58 ºC
Voltage at max. prod. rate 25 V
1,2
Power at max. prod. rate 5.6 kW 70 bar, 42 ºC
Recommended stack current 80 A
1
(for continuous operation)
0 0,25 0,5 0,75 1 1,25
Voltage at recommended current 22 V (at 6.89 bar)
Current density (A/cm2)
H2 prod. rate at recommended current 0.4 Nm3/h
65

cathodic head, where the inflow and outflow channels are 60

connected.

Specific work (kWh/kg)

During operation, the anode side is kept at nearly ambient 55
pressure, while the cathode side is maintained at pressures up
to 70 bar by a back-pressure valve. The hydrogen is pushed
50
electrochemically across the membrane to a much higher
7 bar, 58 ºC
pressure; the cathode pressure rises only from the gas outflow
45
produced by electrolysis. Thus, the electrolyzer works with an 7 bar, 42 ºC

elevated pressure gradient between electrodes, which are 70 bar, 58 ºC

40
therefore designed to support the accompanying mechanical
70 bar, 42 ºC
stresses and to ensure gas-tight seals that block any
dangerous leakage from the cathode to the anode. 35
0 20 40 60 80 100
In Fig. 2, the main characteristics (cell polarization and
Hydrogen production rate (g/h)
specific work) of the electrolyzer under steady-state operation
conditions are represented. It is remarkable that the increased Fig. 2 e Cell polarization curves and specific work function
power required to elevate the hydrogen pressure tenfold is of hydrogen production rate in different operating
completely offset by elevating the stack temperature by 16
C. conditions.
This observation demonstrates the important role that

temperature increases can have against the negative effects of

back pressure (and hydrogen concentration) at the cathode
electrode surfaces.

2.2. Description of the test bench

A schematic flow of the test bench is shown in Fig. 3, and

a picture is shown in Fig. 4. The function of the test bench is to
supply the electrical power and the process fluids in order
to ensure operation of the electrolyzer stack. It allows one to
control several physical parameters, and it permits
measurement of many output data. Moreover, all of the
working conditions must be measured in order to avoid
damage to the stack (e.g., an overload).
The test bench is made up of four main subsystems. First,
the water supply system sends a demineralized water flow
into the stack for the electrochemical reaction, for cooling and
for removing the oxygen bubbles produced at the anode.
Second, the hydrogen circuit delivers the high-pressure gas to
the separation vessels where water is removed from the
hydrogen. Third, the nitrogen supply system renders inert any
Fig. 1 e High-pressure PEM electrolyzer. flammable mixtures inside the ducts and purges the system
5176 i n t e r n a t i o n a l j o u r n a l o f h y d r o g e n e n e r g y 3 5 ( 2 0 1 0 ) 5 1 7 3 e5 1 8 6

Fig. 3 e Test bench layout (EM represents the electrolyzer).

before activation. Finally, there is the electrical power supply

that is regulated from an AC/DC voltage regulator driven from 3. Experimental activity
the control panel. The regulation adjusts the source voltage to
the characteristic curve of the electrolyzer for any current To gather information about membrane transport in PEM
selected. electrolyzers, we conducted several test sessions with the
The control system is based on a programmable logic laboratory-scale PEM electrolyzer. All data collected from
controller (PLC), which acquires all the information these tests were also used as input for an algorithm
provided by sensors and transducers and then controls the designed to estimate the net electro-osmotic drag coeffi-
actuators to impose the desired values on the controlled cient, neo.
parameters. The control system is interfaced with
a computer. Finally, the data acquisition system records all 3.1. Description of the test
of the operational parameters of the process in a file for
subsequent analysis. The acquired data can be shared with The total amount of water that is recovered at the cathode
remote workstations. side has been measured under different operational condi-
For this system, the measured quantities are electrical data tions. The operation variables for these experiments were
(current, voltage), temperature data (water into and out of the cathode pressure, stack temperature and current density.
stack, hydrogen), pressure data (hydrogen, oxygen and Thirty-minute tests have been used so that measurement
nitrogen, differential pressure between the anode inlet and errors can be estimated. At these time intervals, all of the
the outlet), levels in the water separators and other data such water inside the cathode separation vessel was weighed using
as water conductivity and water flow. a precision device.
 i n t e r n a t i o n a l j o u r n a l o f h y d r o g e n e n e r g y 3 5 ( 2 0 1 0 ) 5 1 7 3 e5 1 8 6 5177

Fig. 4 e Picture of the test bench and the control unit.

Before the start of any test, steady-state operation was factors deserve particular attention and to find factors that
achieved, and all of the parameters remained in the desired have a negligible effect. Also, the interaction among the
range. We observed that 15 min was sufficient to stabilize the factors can be assessed.
test bench and the electrolyzer. Once the test was finished, all Once the k input factors have been chosen, it is necessary
of the water collected in the cathode-side atmospheric pres- to determine their range of variation (the experimental
sure vessel was measured. This water comes from another domain). If the experiments are performed only at the lower
pressurized container where liquid water is divided from bound and the upper bound of their possible range (i.e., only
hydrogen gas by means of decantation and where one valve two levels of the factor are considered), then 2k experiments
opens to let water drop through a lamination valve in order to are required. Each combination of different factor levels is
reach atmospheric pressure. The methodology for correctly called a “treatment”; if the treatment is replicated, then an
measuring all of the water transported from the anode to analysis of variance can be performed.
the cathode was to add the change in liquid level inside the
pressurized vessel (WS in Fig. 3) to the water collected in the
atmospheric vessel (HC in Fig. 3) and weighed in the bascule.
Level measurements were displayed on the control board.
Therefore, in addition to the uncertainty related to the
instrumentation (i.e., in the water level sensor in the WS), the
main source of error is linked to the variability in the collec-
tion of water from the HC due to water adhesion on the
surface of the vessel and on the tubes connecting the WS with
the HC.
Tests were programmed using a Design of Experiments
procedure based on Yates’ Technique, a visual representation
of which is presented in Fig. 5.
In the analyzed problem, the water transported through
the membrane of the high cathode pressure electrolyzer stack
is influenced by several independent variables (defined as
factors). For the analysis of their reciprocal influences, an
experimental design methodology is essential. In this work,
a full factorial Design of Experiments (or DoE) has been Fig. 5 e Graphic representation of the Design of
employed; an output variable could be correlated to the input Experiments used for the test session (current density,
variables (factors) to point out the effect that the input vari- factor A; cathode pressure, factor B; stack temperature,
able has on the system. Thus, it is possible to find which factor C).
5178 i n t e r n a t i o n a l j o u r n a l o f h y d r o g e n e n e r g y 3 5 ( 2 0 1 0 ) 5 1 7 3 e5 1 8 6

To analyze the effects of the main factors, we used the of water molecules dragged by proton flow via the driving
Yates’ Technique [12,13]. Concerning the adopted symbolism, force of the potential gradient across the perfluorinated
each factor is labeled with a capital letter (i.e., for Fig. 5, the membrane. The ng value is not linked to a particular driving
current density is factor A; the cathode pressure is factor B; force (or transport mechanism), because it is the result of the
and the stack temperature is factor C). For expressing the concurrent transport mechanism and is only useful for eval-
effects of a factor, a lowercase letter is used when the factor is uating the results of the transport phenomena at different
at the upper level and the letter is omitted when it is at its conditions.
lower level. Thus, it is possible to distinguish main effects (a, b, To obtain ng, water weight was converted to a flow in moles
c) from interaction effects (ab, ac, bc, abc). This notation has per second by using the experiment duration of half an hour.
been used in Fig. 6 (consider the notation used in its headings). Then, by means of Faraday’s equation, current densities were
First, we decided to develop two parallel analyses at two translated directly to ion flow densities in moles per second.
current levels and to include the anode water flow as the All of the values, including ng, have been represented for all
fourth factor. However, it immediately became evident that cases in Table 2.
the water flow did not have a significant effect on water
transport across the membrane, so we dropped the water flow 3.3. Experimental data analysis
factor from the analysis.
The factors have been set at two extreme levels in order to The analysis was developed in order to provide a significant
emphasize the effects of the operating conditions. Current physical characterization of the electrolyzer, and the subse-
density was set at 0.25 and 1 A/cm2, pressure was set at 7 and quent figures are designed to give an idea about the signifi-
70 bar and temperature was set at 40 and 55
. This resulted in cance of each factor’s influence on water transport across the
a total of 23 ¼ 8 different conditions. membrane. Fig. 6 displays each factor’s percentage contribu-
tion to the system’s total output variation. From these data, it
3.2. Experimental data is evident that pressure has a greater effect on water transport
than all of the other factors combined. This result is also
Data logged during the test sessions are summarized in confirmed by the modifications in the ng coefficient, which is
Table 2, the three operating factors involved in the analysis significantly higher for the case of pressure modification than
and their values. for the case of temperature modification (Table 2).
The total amount of water measured after the test was Fig. 7 displays the single effect on the global transport
used to determine the global transport coefficient, ng. This coefficient produced by current density, cathode pressure and
value represents the moles of water transported from anode stack temperature, respectively. These graphs represent the
to cathode for each proton traversing the membrane. variation achieved in the global transport coefficient when
fixing two factors and varying one. The data show that
molH2 O cathode pressure has the greatest effect on water transport
ng ¼ (1)
molHþ across the membrane. In particular, the increase of the
This value does not coincide with the net electro-osmotic drag cathode pressure causes a significant reduction of ng, which is
coefficient, neo, discussed later, which represents the number almost halved as the cathode pressure increases from 7 to
70 bar.
Fig. 8 shows the interaction effects (i.e., the trend of the ng
coefficient while varying one factor, using two different levels
of another factor, and keeping the third factor constant).
Fig. 8a displays the extent of the cathode pressure’s effect
on water transport, and the effect is almost double that for
any temperature condition. Fig. 8b confirms the significant
effect of current density on water transport, which produces
lower variations than pressure.
As seen in Fig. 9a, increasing the pressure produces
different changes in the ng coefficient depending on the level
of the current. At high current density, water transport
decreases much more slowly than at low current density,
Fig. 6 e Percentage effect of factors on the membrane water where it decreases quickly. Also, current density has the
transport (a [ current density at upper bound, other two opposite effect on water transport at extreme pressure levels.
factors at lower bound; b [ cathode pressure at upper These effects are likely due to modification of the lattice of the
bound, other two factors at lower bound; c [ stack membrane at different current densities and at different
temperature, at upper bound, other two factors at lower pressure levels. In particular, the water clusters are modified,
bound; ab [ current density and cathode pressure at upper and higher current densities provide larger clusters. Fig. 9a
bound, temperature at lower bound; ac [ current density shows the variation of ng produced by current density at
and temperature at upper bound, cathode pressure at different pressures. At intermediate pressure (around 40 bar),
lower bound; bc [ cathode pressure and temperature at the gap occurred when varying system current input is about
upper bound, current density at lower bound; abc [ all the 0.5 moles of water, i.e., a 25% reduction. Fig. 9b displays the
three factors at upper bound). negligible effect of temperature, which is only perceptible at
 i n t e r n a t i o n a l j o u r n a l o f h y d r o g e n e n e r g y 3 5 ( 2 0 1 0 ) 5 1 7 3 e5 1 8 6 5179

Table 2 e Experimental results.

Current Stack Cathode Mol/s water  103 Mol/s protons  103 Global transport ng
(A) temperature pressure
(
C) (bar) First replic. Second replic. First replic. Second replic.

40 40 7 0.87 0.86 0.41 2.10 2.08

160 40 7 3.05 3.06 1.66 1.84 1.85
40 40 70 0.24 0.24 0.41 0.59 0.58
160 40 70 2.36 2.42 1.66 1.42 1.46
40 55 7 0.86 0.90 0.41 2.09 2.17
160 55 7 3.18 3.29 1.66 1.92 1.98
40 55 70 0.19 0.18 0.41 0.45 0.44
160 55 70 2.36 2.37 1.66 1.42 1.43

very high pressure. The slope of both lines in Fig. 9b should parallel to its axis. In Fig. 11a and b, the lines are almost
represent the trend at an intermediate current (0.6 A/cm2) parallel and tend to increase with current and to decrease
level and can indicate how cathode pressure can reduce the with pressure. It is remarkable to note the elevated ng range
quantity of water transported through the membrane. achieved when varying the cathode pressure. It is also
noticeable that temperature, despite its almost negligible
effect, has an opposite result on water transport, depending
3.4. Regression model on the levels of the other factors.
Fig. 11c demonstrates that pressure variation has a greater
After collection of the data recorded in the test session, the effect on water transport at low current densities. Also,
DoE methodology allows for the generation of a regression current density has opposite effects on the ng coefficient
model for ng based on the experimental data. This model depending on the cathode pressure. Current density has
could simulate the empirical data obtained from the PEM a positive effect at high pressure and a slightly negative effect
electrolyzer. The algorithm produces several models that fit at low pressure. It has no effect on water transport when
(with different approximations) the experimental data. The operating at approximately 20 bar. Also, pressure variation
best model contains seven “regressors,” or linear term: has greater effects on water transport at high current densi-
ties and has no effect on water transport when operating at
ng ¼ 2:27  0:70i  0:02P þ 0:02Pi þ 0:003T þ 0:005iT  0:0002PT
approximately 0.4 A/cm2. These observations confirmed the
(2)
results shown in Fig. 9a.
The analysis of the residual displays the model’s precision in One possible explanation for the observed results is that
fitting the electrolyzer’s response. The histogram reveals that the hydraulic forces are great enough to reduce up to 25% of
most deviations remain below the 0.02 range and that the the water collected at the cathode separation vessel. This
normal probability of the residual has a symmetric trend situation is more evident at low current density, when fewer
(Fig. 10). water molecules are dragged and pressure produces big
Fig. 11 represents the trend in the number of water mole- changes (Fig. 9a). The distance between the lines at high
cules dragged per proton when varying current, cathode pressure is almost five times that seen at low pressure.
pressure and stack temperature, respectively. The other Current has an intermediate effect, producing different vari-
factors are kept at the middle value of the range studied. ations in ng depending on pressure; this phenomenon can be
First, it is evident how the regression model represents the explained by the water paths through the membrane. At
physical behavior of the phenomena described above. We can higher pressures in the cathode, the clusters are narrower,
observe that temperature has a minor effect in comparison and they allow less water transport inside them. When
with the other two factors, as the iso-ng lines are almost current density is increased at high pressure, the lattice of the

Fig. 7 e Single effect of each parameter included in the analysis.

5180 i n t e r n a t i o n a l j o u r n a l o f h y d r o g e n e n e r g y 3 5 ( 2 0 1 0 ) 5 1 7 3 e5 1 8 6

Fig. 8 e Interaction plot between temperature at extreme pressure and current density rates.

membrane is modified, and larger water cluster paths are Therefore, these results demonstrate that the sensitivity to
produced. At low pressure, these paths are already much temperature is higher than the sensitivity to pressure.
wider, thus allowing higher water transport per ion. A modification of 1
C has a greater effect than a modification
of 1 bar.
From Fig. 12c, we can observe that current density across
3.5. Sensitivity analysis the membrane increases water transport when operating at
elevated pressures, while at approximately 25 bar, it does not
A sensitivity analysis has been performed that determines have any effect. Below 25 bar, the effect is always negative.
conditions when each factor has a predominant effect and in Sensitivity analysis confirms the interaction between current
what magnitude. Basically, it is the first derivative of the and pressure (Fig. 9a), demonstrating that current can have
regression model, and sensitivity curves are the first-grade opposite effects depending on the pressure.
linear expressions, which have been represented in a 2D
contour plot (Fig. 12).
Fig. 12a also represents the effect of temperature on water 4. Description of water transport
transport, as evaluated at different values of cathode pressure phenomena inside a perfluorinated membrane
and current density. Temperature changes always produce
a negative variation in the ng coefficient, which increases at There are many analytical and empirical studies regarding
lower current densities. The pressure range does not seem to water transport inside Nafion membranes.
have any effect: this may be because of the great diffusion From these studies, water transport inside a perfluorinated
from the cathode (higher H2O concentration due to higher membrane is caused by three different phenomena: electro-
pressure) to the anode when temperature rises. osmotic drag transport, diffusion transport, and hydraulic
Pressure does not have a big influence on the ng coefficient pressure transport.
in comparison with temperature, as seen in Fig. 12b. The range of current densities inside the stack is similar to
Remarkably, some of its values are below zero in certain cases, those in common fuel cells (up to 1.4 A/cm2). Thus, the proton
meaning that any increase in cathode pressure produces flux in the cells occurs at an equivalent rate. The main
a decrease in global water transported through the difference with fuel cell phenomena comes from the diffusion
membrane. This transport is caused by the back-pressure flow, which is basically driven from the cathode (high water
forces. concentration due to high pressure) to the anode (fluent liquid

Fig. 9 e Interaction plot between cathode pressure at extreme current density and temperature rates.
 i n t e r n a t i o n a l j o u r n a l o f h y d r o g e n e n e r g y 3 5 ( 2 0 1 0 ) 5 1 7 3 e5 1 8 6 5181

Fig. 10 e Histogram and normal probability plot of residual.

water). In normal operation, we do not account for the 4.1. Electro-osmotic drag transport
Schroeder paradox presented in [14], and we assume that
water is always in the liquid phase due to its high partial Few models have been developed to determine the net electro-
pressure (the membrane is impregnated by water circulating osmotic drag inside the PEM membranes used in electrolyzers.
at the anode) and due to the low temperatures (60
C) inside All of the expressions found in the literature [15,16] that give an
the stack. average value for water molecules dragged by a single
Another resistance that water molecules find on their way hydrogen ion are the result of fitting experimental measure-
from the anode to the cathode is the pressure gradient. This ments in fuel cell applications. In these systems, the reactant
gradient can rise up to 70 bar in less than a quarter of a milli- fluxes are humidified with water vapor to avoid membrane
meter in membrane thickness, which represents a increment drying. In the electrolyzer, the humidification rate of the
of 2.75  105 kPa/cm. Back pressure brings down the net water membrane (i.e., the ratio between SO 3 groups and water
flow due to hydraulic losses. molecules inside the membrane) is much higher due to the

Fig. 11 e Simulated ng values obtained varying current, temperature and pressure.

5182 i n t e r n a t i o n a l j o u r n a l o f h y d r o g e n e n e r g y 3 5 ( 2 0 1 0 ) 5 1 7 3 e5 1 8 6

Fig. 12 e Water transport sensitivity to current density, cathode pressure and stack temperature.

presence of fluent water at the anode side, which attains the is the main factor for correcting the diffusion water coefficient,
maximum water absorption level. As the electro-osmotic drag has been considered:
depends mainly on membrane humidification, high values can
be expected in the electrolyzer case, and these values can be Deff ¼ 31:5 Dw (5)
much higher than those in fuel cells. To obtain the water drag The dimensionless value, 3, is calculated as follows [19]:
inside the membrane, the following expression [17] is used:
1
3¼ (6)
i 1  Fe
Geo ¼ neo   A  n  MH2 O (3)
F
where Fe is the void fraction of the electrode. Because we did
The positive flux is from anode to cathode, and it is the only not have specific data about the porosity of the membrane
possible direction in the case of electrolysis reactions. The used in the electrolyzer, we selected the widely reported
only unknown value in Eq. (3) is the electro-osmotic transport literature value of 3 ¼ 0.3 [20].
coefficient, neo. This parameter represents the number of To calculate the water diffusion coefficient through the
water molecules dragged, due to polarity ionic forces, by each Nafion membrane, many expressions obtained from experi-
proton when crossing the membrane. mental campaigns are available in literature. Because we
The aim here was to use experimental data to determine expected maximum humidification of the membrane due to
the neo under different operating conditions. This determi- the presence of fluent water, we decided to use the expression
nation was accomplished by using an analytical model of the from [21], which only takes into consideration the cell
different transport mechanisms. temperature:
 1:823
Tcell
4.2. Diffusion transport Dw ¼ 0:256  104 (7)
273:15

Diffusion is caused by the presence of different water The concentration gradient between the electrodes is the
concentrations at both sides of the electrolytic membrane. driving force in diffusion through the membrane; therefore,
Fluent water from the main pump flows along the anode the following expressions [22] were considered:
chambers. This flux contacts the Gas Diffusion Layer (GDL) of rm;dry rm;dry
the cells, assuming that the anode electrode is completely Ca ¼ la ; Cc ¼ lc (8)
Mm;dry ðEWÞ Mm;dry
flooded with water. Because the cathode side is initially dry,
one can expect a large concentration gradient from anode to where la and lc represent water content on both sides of the
cathode when the electrolyzer starts operation. As operation membrane. The water content has been related [15,16] with
continues, the cathode chamber is filled with water crossing water activity at the electrode surface by using different
the membrane, and the higher pressure on the cathode side experimental expressions, depending on the activity range:
leads to a higher water concentration than that at the anode
l ¼ 0:043 þ 17:81a  39:85a2 þ 36a3 ; 0 < a  1 (9)
and thus inverts the concentration gradient. This gradient is
the driving force of the diffusion transport flux, according to
l ¼ 14 þ 1:4ða  1Þ3 ; 0 < a  3 (10)
the following expression [18]:
Water activity is then obtained after an evaluation of the
ðCa  Cc Þ
Gd ¼ Deff  A  n  MH2 O (4) mole fraction of water in every element of the electrode [15]:
tm
where Deff represents the effective diffusion coefficient for P
a ¼ xw (11)
Psat
water transport inside the GDL and the Nafion membrane.
Written in this form, a positive flux is from the anode to the On the anode side, the local pressure decreases linearly
cathode. The Bruggeman equation, suitable for randomized- along the channel length, as confirmed by pressure
fiber porous materials, in which the porosity of the membrane measurements at the inlet and the outlet of the electrolyzer
 i n t e r n a t i o n a l j o u r n a l o f h y d r o g e n e n e r g y 3 5 ( 2 0 1 0 ) 5 1 7 3 e5 1 8 6 5183

Table 3 e Connection between assumed porosity and

permeability in the Nafion membrane.
Porosity Permeability
(m2)

0.05 1053
0.4 1019
0.6 1013
0.66 1011
0.7 1010
0.3 7.1410e20

anode. Cathode pressure is constant along the channels, so

the model neglects any small variation of this parameter
during the algorithm’s progress. Fig. 14 e 1D model diagram illustrates equations used in
The molar water content should be nearly unity at the the algorithm.
anode side because the molar production rate of oxygen is low
compared to the amount of water driven by the pump. At the
cathode side, water and hydrogen are present in comparable
The dynamic viscosity of water is only influenced by
quantities, so the mole fraction of water may be significantly
temperature, according to [24]:
lower than unity:

n_ w n_ w m ¼ 0:6612ðT  229Þ1:562 (14)

xw ¼ ¼ (12)
n_ w þ n_ H2 n_ w þ ði  A=2  FÞ

4.3. Hydraulic transport phenomena 5. Determination of the electro-osmotic drag

coefficient, neo
The two transport mechanisms discussed so far are the
significant ones for an atmospheric electrolyzer, where the 5.1. Water balance across the membrane
anode and cathode chambers operate at similar pressures.
In the high-pressure cathode device, hydrogen can be Net water transport across the membrane, ignoring evapora-
produced at up to 70 bar, while the anode is kept at approxi- tive flow, is expressed as the sum of the phenomena already
mately 1 bar. This great pressure opposes the motion of water described:
that is essentially pulled across the membrane by electro-
osmotic drag. Gnet ¼ Geo þ Gd þ Gpi (15)
Water percolation caused from a pressure gradient is Every term of this expression is primarily influenced by
governed by Darcy’s law: current density, humidification ratio or pressure gradient.
kD Except for the humidification ratio, all these factors are
Gpi ¼ rA Vp (13) controlled in the experimental session in order to validate the
m
hypothesis made using the analytical model.
Written in this form, the positive flux is from the cathode to
The aim was to develop a one-dimensional model where
the anode. To determine the permeability coefficient (the
the spatial variable coincides with the anode and the cathode
Darcy constant, KD) of the Nafion membrane, we assumed that
channels (the main transport direction of water inside the
it was directly related to material porosity, e, as described in
anodic half-cell and of the produced hydrogen along the
the last section. The relation between both parameters is
cathode half-cell), assuming that the water flux at the anode is
expressed in Table 3, as defined in [23], and the KD value
(in grey) has been interpolated.

Table 4 e Parameters, independent and dependent

variables of the model.
Model parameters

A 160 cm2 Ncell 12

r 975 g/l tm 0.254 mm
EW 1100 g/mol 3p 0.12
Mm,dry 2 g/cm3 F 96,485 C/mol

Independent variables Dependent variables

(input) (output)

i Pan Pcat ng Can Ccat neo

Fig. 13 e Water mass flow mechanism across the
GH2 O Tan,in Tan,out Tcat Gd Gpi Gnet
perfluorinated membrane.
5184 i n t e r n a t i o n a l j o u r n a l o f h y d r o g e n e n e r g y 3 5 ( 2 0 1 0 ) 5 1 7 3 e5 1 8 6

along the cell diameter. At the anode side, there is a fluent GAC;j
yH2 O;j ¼ yH2 O;j1 þ (22)
water flow that arrives at the electrode in great excess and Gtot;C;j
that is slightly consumed and enriched with the oxygen
where the subscript cons and prod mean “consumed” and
produced along the channel from the inlet to the outlet port.
“produced” by the electrochemical reaction, respectively. The
At the cathode side, there is not a fluent mass flow (Fig. 13)
subscript AC indicates transport from the anode to the
that arrives at the electrode. Rather, there is hydrogen
cathode. Also, the following conditions are imposed:
generated that is taken away from the outlet manifold
together with the water transported from anode. yH2 O;j þ yO2 ;j jA ¼ 1
(23)
The cell is divided along the channel direction using yH2 O;j þ yH2 ;j jC ¼ 1
discrete elements, j, perpendicular to the channel direction
and parallel to the membrane normal vector. The model and, for the first element:
essentially calculates the water concentration variation along yH2 O;0 jA ¼ 1
the anode and the cathode at the electrode-membrane inter- (24)
yH2 O;0 jC ¼ 1
face. The mole fraction of water is modified via oxygen and
hydrogen generation at both sides of the membrane. The In this approximation, several assumptions have been
longest is the path of flux along the channel axis, the highest is made:
the molar fraction of products, and the lowest is the one of
water. An illustrated diagram is shown in Fig. 14. - Water evaporation is neglected.
Assuming that product generation is homogeneous on the - Ionic flux on the membrane is homogeneous and constant.
membrane surface, oxygen and hydrogen production at every - Water flux along the membrane has been assumed as
discrete element of the anode and the cathode are as follows: a Poiseuille flow across a perpendicular porous surface.
- Gradients in characteristic parameters (nd, Deff and KD) have
I been neglected.
GO2 ;j ¼ (16)
4Fm
5.2. Results and determination of the electro-osmotic
I drag coefficient, neo
GH2 ;j ¼ (17)
2Fm
The algorithm makes use of all the expressions previously
The water flux consumed is described as follows:
described in order to evaluate the proportion of water dragged
I electrochemically by the net flux to the cathode. Depending
GH2 O;j ¼ (18) on the different operating conditions, diffusion or pressure
2Fm
gradients can push water molecules in the same or the
where m represents the number of discrete elements into
opposite direction from the drag.
which the path inside the cell is divided. The evolution of the
The values of the model parameters, together with the
mole fractions inside the working fluid can then be described:
independent variables (input) and the dependent variables
Anode:
(output) of the model are summarized in Table 4.
The results of the model for the different operating
GH2 O;cons;j GAC;j
yH2 O;j ¼ yH2 O;j1   (19) conditions are shown in Table 5, where the average values of
Gtot;A;j Gtot;A;j
ng between the two experimental replications have been taken
into account:
GO2 ;prod;j
yO2 ;j ¼ yO2 ;j1 þ (20) Figs. 15 and 16 show the value of neo when the input factors
Gtot;A;j
change.
Cathode: The behavior of neo depends on the balance of water mass
transport between the anode and cathode. Given the net mass
GH2 ;prod;j of water transported, the behavior of the hydraulic percola-
yH2 ;j ¼ yH2 ;j1 þ (21)
Gtot;C;j tion determines the results obtained by the electro-osmotic

Table 5 e Values obtained for the neo coefficient from the designed algorithm
T P i Average ng Gpe  104 Gd  106 Geo  104 Gw,an,cath  104 neo
[
C] [bar] [A/cm2) [molH2O/molHþ] [mol/s] [mol/s] [mol/s] [mol/s] [molH2O/molHþ]

40 7 0.25 2.10 1.44 11.1 9.9 8.35 2.39

40 7 1.00 1.84 1.53 9.11 32 31 1.94
40 70 0.25 0.58 16 93.3 19 2.43 4.70
40 70 1.00 1.44 17 106 42 24 2.55
55 7 0.25 2.13 1.87 5.61 11 8.82 2.59
55 7 1.00 1.94 1.87 5.57 34 32 2.05
55 70 0.25 0.45 20 47 22 1.88 5.33
55 70 1.00 1.44 21 61.3 45 24 2.74
 i n t e r n a t i o n a l j o u r n a l o f h y d r o g e n e n e r g y 3 5 ( 2 0 1 0 ) 5 1 7 3 e5 1 8 6 5185

Therefore, it seems that the pressure has the effect of

increasing the neo coefficient in every situation while the
current density tends to reduce it; the temperature does not
have a remarkable effect, but it does increases the value of the
coefficient.

6. Conclusions

Here, we have analyzed the mechanisms of water transport

into the membrane of a high-pressure PEM electrolyzer
operating with a pressure gradient from the cathode to the
anode. The aim was to determine trends in the net electro-
osmotic drag coefficient, ng, under different operation condi-
tions. The purpose of this was to find the optimal working
conditions for reducing the amount of water transported from
Fig. 15 e neo values obtained when varying pressure and the anode to the cathode, thus simplifying a plant’s
current density at 40
C. complexity by reducing the amount of water that must be
separated from the produced hydrogen. In general, the
cathode pressure significantly reduces the water transport
(a positive effect). Also, operation at lower current density
drag. In fact, the contribution of the water diffusion is negli-
reduces the net electro-osmotic drag coefficient, ng; therefore,
gible. The balances show the following:
the best operation strategy for obtaining dried hydrogen at the
cathode is to impose high cathode pressure and low current
- neo decreases with current density at every pressure and
density.
temperature.
Experimental tests were performed with Design of Exper-
- neo increases with pressure at every current density and
iments procedures that considered current density, stack
temperature.
temperature and cathode pressure as input factors. The
- neo slightly increases with temperature at every pressure
results showed that pressure has a greater effect on water
and current density.
transport than all of the other factors. It seems that
back-pressure forces inside the membrane are the stronger
As pressure inside the membrane rises, a substance’s
driving force for water motion. This effect can be linked to
concentration also rises. As molecules are closer to each other,
a modification of the membrane’s lattice. Modification of the
their movements cause greater interactions, thus increasing the
water clusters where the ions move can greatly vary the
amount of water molecules dragged per proton. The opposite
proportion of water and hydrogen arriving at the cathode
effect of current density is explained with a similar reasoning:
chamber. In this sense, current density also has different
when ion flux density increases, each hydrogen ion is only able
effects depending on the cathode pressure, which is linked to
to carry fewer water molecules because water concentration
the water pathways available inside the perfluorinated
does not depend on current density. Temperature’s positive
membrane. Temperature does not seem to have a remarkable
effect is not very easy to understand, but it may be caused via
effect, except for the case of low pressure and high current,
lowering the electrochemical forces between the water mole-
where it has a positive effect.
cules and making them easier for ions to catch them.
The other aim is to determine the trend of the electro-
osmotic drag coefficient, neo. Our analysis shows that per-
forming electrolysis at elevated pressures and temperatures
and at low current densities achieves greater water transport
rates between the electrodes by electro-osmosis. The value of
neo depends on the water mass transport balance between
anode and cathode. Given the net mass of water transported,
the behavior of the hydraulic percolation determines the
results obtained for the electro-osmotic drag. In fact, the
contribution of the water diffusion is negligible.

Acknowledgements

This work was carried out as a part of a large project finan-

cially supported by the Ministry of Instruction, Universities
and Research of Italy. Technical support was provided by the
Fig. 16 e neo values obtained when varying pressure and Department of Energy of Politecnico di Torino, Italy. The
current density at 55
C. authors are also grateful to Giner Electrochemical Systems
5186 i n t e r n a t i o n a l j o u r n a l o f h y d r o g e n e n e r g y 3 5 ( 2 0 1 0 ) 5 1 7 3 e5 1 8 6

LLC. for providing the S-10 electrolyzer prototype for carrying [11] Li X, Qu S, Yu H, Hou M, Shao Z, Yi B. Membrane water-flow
out the tests. rate in electrolyzer cells with a solid polymer electrolyte
(SPE). Journal of Power Sources 2009;190:534e7.
[12] Walpole RE, Myers RH. Probability and statistics for
engineers and scientists. New Jersey: Prentice-Hall
references
International Inc.; 1993.
[13] Beck JV, Arnold KJ. Parameter estimation in engineering and
science. New York: John Wiley & Sons; 1977.
[1] Grigoriev SA, Porembsky VI, Fateev VN. Pure hydrogen [14] Majsztrik PW, Satterfield MB, Bocarsly AB, Benziger JB. Water
production by PEM electrolysis for hydrogen energy. sorption, desorption and transport in Nafion membranes.
International Journal of Hydrogen Energy 2006;31:171e5. Journal of Membrane Science 2007;301:93e106.
[2] Marangio F, Santarelli M, Calı̀ M. Theoretical model and [15] Springer TE, Zawodzinski TA, Gottesfeld S. Polymer
experimental analysis of a high pressure PEM water electrolyte fuel cell model. Journal of Electrochemical Society
electrolyser for hydrogen production. International Journal 1991;138:2334e41.
of Hydrogen Energy 2009;Vol. 34:1143e58. [16] Shimpalee S, Van Zee JW. Numerical studies on rib &
[3] Santarelli M, Medina P, Calı̀ M. Fitting regression model and channel dimension of flow-field on PEMFC performance.
experimental validation for a high-pressure PEM International Journal of Hydrogen Energy 2007;32:842e56.
electrolyzer. International Journal of Hydrogen Energy 2009; [17] Gorgun H. Dynamic modelling of a PEM electrolyzer.
34:2519e30. International Journal of Hydrogen Energy 2006;31:29e38.
[4] Zawodzinski TA, et al. A comparative study of water uptake [18] Santarelli M, Torchio MF, Cochis P. Parameters estimation of
by and transport through ionomeric fuel cell membranes. a PEM fuel cell polarization curve and analysis of their
Journal of the Electrochemical Society 1993;140:1981e5. behavior with temperature. Journal of Power Sources 2006;
[5] Amphlett JC, Baumert RM, Mann RF, Peppley BA, Roberge PR. 159:824e35.
Performance modeling of the Ballard Mark IV solid polymer [19] Marr C, Li X. An engineering model of proton exchange
electrolyte fuel cell. Journal of the Electrochemical Society membrane fuel cell performance. ARI 1998;50:190e200.
1995;142:1e15. [20] Amphlett JC, Baumert RM, Mann RF, Peppley BA, Roberge PR.
[6] Lee WK, Van Zee JW, Shimpalee S, Dutta S. Effect of humidity Performance modelling of the Ballard Mark IV solid polymer
on PEM fuel cell performance. Part I e experiments. Part II e electrolyte fuel cell. Journal of Electrochemical Society 1995;
numerical simulation. In: Proceedings of ASME IMECE, 142:1e15.
Nashville, TN, HTD 364-1; 1999. p. 359e74. [21] Lin G, He W, Nguyen TV. Modeling liquid water effects in
[7] Weber AZ, Newman J. Transport in polymer-electrolyte the gas diffusion and catalyst layers of the cathode of a PEM
membranes. I. Physical model. Journal of the Electrochemical fuel cell. Journal of Electrochemical Society 2004;151:
Society 2003;150:1008e15. A1999e2006.
[8] Calı̀ M, Giaretto V, Santarelli M. k-Deformed kinetics [22] Shimpalee S, Van Zee JW. Numerical studies on rib &
underlying water uptake in Nafion115 membrane. channel dimension of flow-field on PEMFC performance.
Solid State Ionics 2009;180:76e81. International Journal of Hydrogen Energy 2007;32:842e56.
[9] Meier F, Eigenberger G. Transport parameters for thee [23] Oosthuitzen PH, Sun L, McAuley KB. The effect of channel-
modelling of water transport in ionomer membranes for to-channel gas crossover on the pressure and temperature
PEM-fuel cells. Electrochimica Acta 2004;49:1731e42. distribution in PEM fuel cell flow plates. Applied Thermal
[10] Ge S, Yi B, Ming P. Experimental determination of electro- Engineering 2005;25:1083e96.
osmotic drag coefficient in Nafion membrane for fuel [24] Birgersson E, Noponen M, Vynnycky M. Analysis of a two-
cells. Journal of the Electrochemical Society 2006;153: phase non-isothermal model for a PEFC. Journal of
A1443e50. Electrochemical Society 2005;152:A1021e34.