Journal of Power Sources 269 (2014) 203e211

Contents lists available at ScienceDirect

Journal of Power Sources

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

Modeling an alkaline electrolysis cell through reduced-order and loss-

estimate approaches
Jaroslaw Milewski a, Giulio Guandalini b, Stefano Campanari b, *
a
Institute of Heat Engineering, Warsaw University of Technology, 21/25 Nowiejska Street, 00-665 Warsaw, Poland
b
Politecnico di Milano, Department of Energy, Via Lambruschini 4, 20156 Milano, Italy

h i g h l i g h t s

Modeling of alkaline electrolysis cells.

Setup of a reduced order equivalent circuit model.

Setup of a loss estimation model.

Comparison with experimental data on real alkaline electrolyzer.

Model accuracy within 3%, adequate for use in complex system analysis.

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

Article history: The paper presents two approaches to the mathematical modeling of an Alkaline Electrolyzer Cell. The
Received 12 December 2013 presented models were compared and validated against available experimental results taken from a
Received in revised form laboratory test and against literature data. The first modeling approach is based on the analysis of
13 June 2014
estimated losses due to the different phenomena occurring inside the electrolytic cell, and requires
Accepted 18 June 2014
Available online 7 July 2014
careful calibration of several specific parameters (e.g. those related to the electrochemical behavior of the
electrodes) some of which could be hard to define. An alternative approach is based on a reduced-order
equivalent circuit, resulting in only two fitting parameters (electrodes specific resistance and parasitic
Keywords:
Alkaline electrolysis
losses) and calculation of the internal electric resistance of the electrolyte. Both models yield satisfactory
Modeling results with an average error limited below 3% vs. the considered experimental data and show the
Experimental analysis capability to describe with sufficient accuracy the different operating conditions of the electrolyzer; the
Equivalent circuit reduced-order model could be preferred thanks to its simplicity for implementation within plant
Loss calculation simulation tools dealing with complex systems, such as electrolyzers coupled with storage facilities and
Electrolyzer intermittent renewable energy sources.
© 2014 Elsevier B.V. All rights reserved.

1. Introduction which are suitable for use with hydrogen [12] which is especially
required for the low temperature types. In addition to this,
Hydrogen-based technologies promise ultra high efficiency in hydrogen is considered a promising way for setting up large scale
power generation (up to 65% for Higher Heating Value (HHV) as the energy storage facilities suitable for improving the electric grid
reference based on fuel cells [1]) and virtually without environ- stability against the power production oscillations of intermittent
mental impact [2]. Fuel cells are predicted to be efficient sources of renewable plants like solar and wind power plants. Thus, efficient
combined heat and power in small scale applications [3,4] and are and large scale hydrogen generation is very desirable technology
the primary choice for future hydrogen-based mobility [5]. There [13e15]. Water electrolysis is a well-known process and is currently
are several types of fuel cells, of which the most important nowa- adopted in many applications in order to produce hydrogen with
days are Polymer Exchange Membrane Fuel Cells [6,7], Solid Oxide high purity. The opportunity to connect this technology directly
Fuel Cells [8,9] and Molten Carbonate Fuel Cells [10,11]), all of with renewable energy sources has, for instance, led to many
studies developing island-oriented storage systems that couple an
electrolyzer, local hydrogen storage and a fuel cell to obtain a fully
* Corresponding author.
autonomous and controlled power plant based on solar or wind
E-mail address: stefano.campanari@polimi.it (S. Campanari). power. Several studies are also addressing hydrogen production

http://dx.doi.org/10.1016/j.jpowsour.2014.06.138
0378-7753/© 2014 Elsevier B.V. All rights reserved.
204 J. Milewski et al. / Journal of Power Sources 269 (2014) 203e211

from fluctuating renewable sources for direct injection into natural

gas infrastructures (the so-called power-to-gas concept), where
fast-response, high pressure electrolysis is the technology required
for hydrogen generation [16].
Various technologies for water electrolysis are currently avail-
able or under development. Two electrodes, on which the dissoci-
ation reaction of water takes place, and an interposed electrolyte,
which allows the transport of distinctive ionic species, are common
features of all of them. On the other hand, the nature of the elec-
trolyte brings with it different conceptual operating principles and
designs. The most commonly used electrolyte materials are:

liquid solutions of alkaline hydroxides (usually KOH or NaOH)

Fig. 1. A “zero-gap” design of alkaline electrolyzer.
[14]

polymer exchange membranes (PEM) [17,18]

solid oxides [19,20]

Alkaline and PEM electrolysers usually work in a range of

temperatures between 60 and 80  C. Solid oxide electrolytes need
much higher temperatures to be active (600e900  C); this is a limit
referring to the prospect of intermittent operation, at least if a fast
thermal transient capability is required, but it increases the fraction
of energy required by the dissociation process that can be supplied
as heat instead of electricity [21].
While Alkaline and PEM technologies are commercial solutions,
solid oxides technology is still at a developmental stage. More
precisely, alkaline electrolyzers are a mature technology in indus-
trial applications and a wide number of large units are in operation,
while PEMs presently have significant but limited applications in
terms of production capacity and still present some issues in terms
of lifetime. Nevertheless, much of the research effort is focused on
PEMs, looking to avoid the use of a corrosive electrolyte and
possibly to an easier increase in operating pressure and a less
problematic intermittent operation. The alkaline process is studied
with the aim of improving performance in terms of operating
pressure, temperature and efficiency, with advantages in terms of
proven feasibility on large scales [14].
In this work we focus only on the models related to alkaline
technology, due to the prospect of modeling high power electro-
lyzers working on an intermittent regime with fast load ramps and
easy startup/shutdown procedures that cannot be followed by high
temperature devices.
In addition to the type of electrolyte, other design parameters
that distinguish different electrolyzer technologies are the position
and electrical connection scheme of the electrodes.
An alkaline electrolyzer can be traditionally designed with a gap Fig. 2. Monopolar (a) and bipolar (b) design of an alkaline electrolysis cell [42].
between the electrodes and the interposed separation membrane,
which is necessary to avoid the risk of a dangerous mix of hydrogen
and oxygen, which are produced on opposite electrodes. In this so- A power conditioning system, a water purification system
lution the electrolyte and the products circulate between the elec- (usually based on reverse osmosis) and a downstream purification
trodes. An alternative and more modern design is based on “zero-gap” section (required in the case of very high purity requirements)
cells (see Fig. 1), where the electrodes are pressed on the membrane, complete the balance of the plant, as shown in the next section.
allowing an increase in efficiency ([22,23]). Electrolyte and gases Depending on the operating pressure of the electrolyzer, a
permeate the electrodes and are collected on the other side. compressor can be required to obtain a pressurized hydrogen flow
As far as the electrical scheme is concerned, cells can be con- suitable for storage or other downstream utilization. High pressure
nected in parallel leading to “monopolar” electrolyzers; this choice electrolysis is studied to decrease the compressor consumption; for
is simpler to realize and more robust, but needs more complex instance in [24], the operation of a 25 bar 100 kW electrolyzer for
external connections and circuits, bringing about higher losses. The more than 5000 h is reported. The responses of the main system
alternative is a series connection that makes it possible to manu- parameters during operation are described; at partial load, the
facture the electrodes of two adjacent cells on the two sides of the normally increasing hydrogen percentage in the produced oxygen
same plate (“bipolar” electrolyzers d Fig. 2). The resulting is stabilized by a special circulation system.
component (stack) is more compact; on the other hand parasitic Efforts are being made to produce highly pressurized electro-
currents have more possible paths and it becomes impossible to set lyzers to increase overall energy efficiency by eliminating me-
up modular maintenance of single cells without disassembling the chanical compression, despite practical considerations such as
whole stack. corrosion, hydrogen embrittlement, operational complexity,
 J. Milewski et al. / Journal of Power Sources 269 (2014) 203e211 205

In Ref. [31] experimental work involving the direct-current

electrolysis of highly concentrated potassium hydroxide solutions
at high temperatures (up to 400  C) and under various pressures is
described. The advantages of operating an alkaline electrolysis cell
at high temperatures include increasing the ionic conductivity of
the electrolyte and enhancing the rates of electrochemical reactions
at the electrode surfaces. Cell operation with increasing steam
partial pressure over the solution is also shown to enhance cell
performance. The best cell performance was achieved using a
cobalt-plated nickel anode at a temperature of 400  C and a steam
partial pressure of 87 bar.
Recent progress in alkaline water electrolysis for hydrogen
production and applications is presented in Ref. [32]. The paper
examined the fundamentals of water electrolysis, compared the
performance of various water electrolysis designs and introduced
several emergent water electrolysis technologies.
Fig. 3. Main components of the electrolyzer experimented at Politecnico di Mila-
A new multi-physics approach for modeling and design of
nodDepartment of Energy. alkaline electrolyzers is given in Ref. [25]. The positive impact of
temperature on the power consumed by the electrolyzer, the rate of
hydrogen production and energy efficiency is indicated. According
dynamic response and costs being less favorable for pressurized to this model, increasing the operating pressure induces a reduc-
electrolyzers. A first study, developed by Roy [22], shows a com- tion in the power consumed by the electrolyzer and the rate of
parison of total energy consumption at various operating pressures hydrogen production by a factor depending on relative pressure
(up to 700 bar), which finds that atmospheric electrolyzers can be and temperature.
more efficient. On the other hand, in the work by Hammoudi et al. The electrical behavior of a commercial advanced alkaline water
[25] pressurized systems of up to 70 bar are simulated, finding total electrolyzer is investigated in Ref. [33]. The paper contains a model
energy consumptions (including both electrolysis and compres- based on the thermodynamic, activation, double-layer and ohmic
sion). The obtained results are very similar for both the pressurized effects and phenomena that take place in a real alkaline electrolysis
and the equivalent atmospheric system in conditions near the process. The parameters of the model are experimentally deter-
optimal temperature, while at lower temperatures the pressurized mined from both static and dynamic operating tests on the elec-
system shows better performances. A third work by Onda [26] es- trolyzer. The model was implemented in a MATLAB-Simulink
timates the power required to produce high-pressure hydrogen to environment and validated in three different operating conditions.
be about 5% less than that required for atmospheric electrolysis. According to the literature, the available models are not gener-
Nevertheless, in any case the influence of the balance of the plant is alized and are typically non-predictive, so they do not allow one to
very relevant and requires a detailed case-by-case evaluation; shape the design and detailed prediction of electrolyzer behavior.
generally speaking, it can be said that presently the pressurized In this work a classical mathematical model proposed in the liter-
technology results show that it is particularly suitable for medium- ature is compared with an alternative one, based on a reduced-
pressure (up to 25e100 bar) and small production volumes (e.g. up order equivalent electric circuit and a lumped model description
to 100 Nm3H2 =h), although it requires improvement in terms of of the phenomena involved, as described below.
reliability and durability.
A mathematical model for an advanced alkaline electrolyzer is 2. Experimental setup
presented in [27]. The model is based on a combination of funda-
mental thermodynamics, heat transfer theory, and empirical elec- Experimental investigation was performed to measure the
trochemical relationships. The number of required parameters has performance of an Alkaline Electrolyzer Cell (AEC) stack built for
been reduced to a minimum to make the model suitable for use in hydrogen and oxygen generation and installed at the Laboratory of
integrated hydrogen energy system simulations. Micro-cogeneration (LMC) at Politecnico di Milano [4].
In Ref. [28] a critical review of the current data on the specific The AEC was manufactured by H2Nitidor Srl for SAPIO SpA and
conductivity of aqueous potassium hydroxide (KOH) solutions is experimented on for a trial perioddsee Fig. 4 for details. The AEC
presented. Based on this review, the empirical correlation was works at 30 bar and has a planar area of 500 cm2 and a maximum
developed over a molarity range of 0e12 at temperatures of hydrogen production of 6.6 Nm3/h, the anode and cathode (bipolar
0e100  C. design) are based on a solid Ni structure with thickness of 0.3 mm,
The increase in electrolysis temperature was investigated in and the electrolyte is a potassium hydroxide and water mixture
Ref. [29] and it was found that efficiency, defined by a higher (usually 8 M KOH), operating with a total gap between the elec-
heating value of the produced hydrogen (HHV) divided by the total trodes of about 14 mm. A diaphragm is placed between the two
energy of the heat and the electricity required to produce the opposing separator plates and the hydrogen and oxygen are
hydrogen, above 75% (1000  C) can be obtained (with a share be- generated in the opposing channels. The cathodic and anodic cur-
tween heat and electricity of about 20%e80%) d compared with an rent collectors are made by a Ni mesh. The cell was tested in the
efficiency level of 47e82% with conventional alkaline solution experimental facility shown in Fig. 3, where the stack is held in a
electrolysis at lower temperatures. supporting frame and it was possible to set up and control each
Thermal performance of a commercial alkaline water electro- operating parameter. The surface liquid indicator enables mea-
lyzerdexperimental study and mathematical modeling is pre- surement and control of the water directed to the electrolyte.
sented in Ref. [30]. In this paper a study of the thermal performance A proprietary control system continuously monitors through
of a commercial alkaline water electrolyzer designed for a rated thermocouples (PT-100 type) the temperatures in stack inlet/outlet
hydrogen production of 1 Nm3H2 /h at an overall power consumption flows as well as in various positions, allowing one to easily view and
of 4.90 kWh/Nm3H2 is obtained. control the operating condition and performances of the unit. A
206 J. Milewski et al. / Journal of Power Sources 269 (2014) 203e211

Fig. 4. Simplified scheme of the electrolyzer components and of the instrumentation.

nominal DC current of 144 A is generated by a power system based electrodes, can be conglomerated for simplicity in a single value of
on a rectifier and a 400/180 V transformer, with nominal power rother.
45 kW, and directly applied at the electrodes. Within the first term, the ionic conductivity of the alkaline so-
For the purpose of this work, experimentation consisted in lution as a function of electrolyte temperature and composition is
collecting voltageecurrent curves during operation of the unit, relatively well known [28]. Fig. 6 presents the expected conduc-
which were later used for model validation. The accuracy of the tivity variation according to literature [28], which can be expressed
voltage and current measurements was within ±1% of the reading. by the following equation:

3. Model definition

s ¼ 2:041$M  0:0028$M2 þ 0:001043$M3 þ 0:005332$M$T

Water electrolysis consists in the molecular decomposition by
means of an externally applied direct electric current, through the M
þ 207:2  0:0000003$M2 $T 2
anodic and cathodic reactions: T
(4)
1
2$OH
ðlÞ /H2 OðlÞ þ O2ðgÞ þ 2$e (1)
2 where: M dmolarity, mol l1; T dtemperature, K.
Eq. (4) is plotted in Fig. 6; it should be noted that this correlation
2$H2 OðlÞ þ 2$e /2$OH
ðlÞ þ H2ðgÞ (2) is not valid above a molarity value of 12. To extend the range of
function utilization we derived a new correlation, reproducing the
as shown in Fig. 5. same results at lower molarity:
In this paper, two different modeling approaches are discussed.
A common feature is the description of ionic conductivity of the cell
structure, which can be split among the contributions of the layers
that constitute the cell itself: the electrolyte, which is the primary
ion conductor, sandwiched by anode and cathode layers that in-
fluence ionic conductivity as well as due to the triple boundary
phase processes occurring at the electrolyte interface. The material
used, porosity and design of the electrodes affect cell voltage
significantly; as a consequence, the total ionic resistance of the cell
is a function of many parameters. Additionally, electrodes can be
built as multi-layers by themselves. Thus, the total ionic resistance
of the AEC can be estimated by the following relationship:

dKOH
r¼ þ rother (3)
sKOH

where the first term is related to the electrolyte:s is the ionic

conductivity and d the thickness of the electrolyte layer. All other
effects, depending on thicknesses and conductivities of the Fig. 5. Operation principle of alkaline water electrolysis [27].
 J. Milewski et al. / Journal of Power Sources 269 (2014) 203e211 207

constant and G Gibbs free energy. Reversible potential at a given

rev ) can be calculated ac-
temperature and standard pressure (Et;p0
cording to the correlation proposed by Ref. [25]

rev
Et;p0 ¼ 1:50342  9:956$104 T þ 2:5$107 T 2 (8)

The last term of Eq. (7) expresses the impact of pressure on the
ideal voltage, depending on the fact that gas products are in contact
with electrolyte vapors next to the electrodes. The partial pressure
pw of the gaseous electrolyte solution in contact with gas products
can be calculated from the partial pressure of pure water pw at the
same temperature [34]. A dependence of pw on both temperature
and molarity is considered, while pure water vapor pressure and
molarity are only functions of temperature due to volumetric ef-
fects. Other correlations for reversible potential at standard pres-
sure and more complex corrections are proposed in the literature
Fig. 6. Temperature and composition dependencies of ionic conductivity for KOH (see for instance [33]), yielding results that are only slightly
solution.
different from those of Eq. (7) and of the described pressure model.

3.1.2. Activation losses

132:1 The second contribution in Eq. (6) is given by the activation
s¼
 (5)
overpotential, which includes the effects of reaction kinetics. The
exp 0:01592$ðM  12:27Þ2 þ 0:3819$Mþ9:406
R$T rate of reaction depends on temperature, the nature of the elec-
trodes and on the composition of the electrolyte solution next to
Eq. (5) is obtained based on the same experimental data as for the electrodes; it can be in general expressed by Arrhenius' law and
Eq. (4), but it is structured with an exponential (Arrhenius type) re-written in the well-known simplified formulation of the Butler-
form instead of a polynomial interpolation; it allows predicting Volmer equation, valid for redox reactions in electrochemical cells:
physically consistent results for molarities greater than 12.
h a   a i
a c
i ¼ i0 exp nFEaact  exp  nFEcact (9)
3.1. Loss-estimate approach for modeling of AEC RT RT

where i0 is the current circulating at equilibrium from and to the

A complete description of the phenomena involved in alkaline
electrodes, named “exchange current density”, and aa/c are “charge-
electrolysis would require a multi-physics approach including
transfer coefficients” which describe the share of the energy barrier
different domains. The model proposed by Hammoudi in his work
between the electrodes. An accurate description of the theory
[25] can be taken as a reference because of its completeness and
related to this model can be found in Ref. [35]. For sufficiently larger
high level of detail. Alternatively, the classical simplified approach
values of the overpotential, the second term becomes negligible
to electrochemical devices modeling describes the behavior of the
and the Butler-Volmer equation can be approximated by the Tafel
system by means of a voltageecurrent polarization curve. The curve
law [25]:
is obtained as a sum of contributions related to various physical
phenomena, and the fundamental equations included in the model

a=c RT i
usually describe the thermodynamics of the system, the kinetics of Eact ¼ 2:3026 log (10)
nFaa=c i0
the electrochemical reaction and the losses in the electrical
equivalent circuit. This approach can be applied to both fuel cell and Exchange current densitydi0dis strongly dependent on the
electrolyzer modeling, but this work is focused on alkaline tech- nature of the electrodes (materials, roughness, geometry) and
nology and, therefore, the model is adapted in order to describe the therefore is considered as a fitting parameter of the model. The
particular characteristics of this device. Generally speaking, the charge-transfer coefficients aa/c depend linearly on temperature
polarization curve is described by the equation: [36] and the following correlations can be adopted to consider the
impact of temperature:
E ¼ Erev þ Eact ðiÞ þ Eohm ðiÞ (6)
aa ¼ 0:0675 þ 0:00095$T
where the three voltage contributions are respectively the ideal (11)
ac ¼ 0:1175 þ 0:00095$T
reversible voltage, the overvoltage due to electrochemical kinetics
and the one related with electrical losses. which are valid for Ni-based electrodes [25].

3.1.1. Standard potential 3.1.3. Ohmic losses

Reversible potential can be calculated from the thermodynamics Ohm's law can be used to model electrical losses (Eohm)
of the water dissociation reaction, considering an ideal electrolysis
cell at standard temperature and pressure conditions. It can be Eohm ¼ r$i (12)
formulated as:
where the area specific resistance r is calculated by Eq. (3).
 
p
rev
nFEt;p ¼ DGt;p ¼ rev
nFEt;p0 þ RT ln ðp  pw Þ1:5 w (7)
pw 3.1.4. Other losses
Two other explicit main contributions to cell losses can be
where the temperature and pressure influences are separated in considered: the resistance of the electrolyte itself and the addi-
two terms. n is the number of electric charges involved in each tional resistance given by the presence of bubbles, both in the bulk
reaction, equal to 2 in the case of water electrolysis, F is Faraday's electrolyte and as a curtain over the electrodes. In addition to this,
208 J. Milewski et al. / Journal of Power Sources 269 (2014) 203e211

all other resistances (i.e. due to the electrodes and the gas separa- 3.2. Reduced-order approach for modeling of AEC
tion membranes) and parasitic currents dispersed through the el-
ements of the cells without contributions to the reaction are With the reduced-order approach, the main processes occurring
comprised in the additional term rother. during electrolyzer operation can be described by the flows of ions
The gas bubbles produced on the surface of the electrodes form (OH) and electrons through an adequate equivalent electric circuit
an unstable layer that induce a local increase in the electrical of the electrolysis cell, as indicated in Fig. 8. Making reference to the
resistance due to the lower conductivity of the gas with respect to figure, the total current i3 is imposed on the electrolyzer by an
the electrolyte. Many approaches are proposed in the literature in external power supply, where the current source (DC) has its own
order to quantify this effect, mainly proposing correlations be- resistivity (rDC); then the current is divided into two circuits, the
tween the additional resistance of the bubble layer and the pure first one indicated as i1 is the total electron flow transported by
electrolyte one [37]. In particular we adopt here the correlation alkaline ions (OH), whereas i2 indicates electrons which pass
proposed by Bruggeman [37]: through the electrolyte in the same direction as the alkaline ions
due to the presence of a small but significant electronic (electric)
sε conductivity.
¼ ð1  εÞ1:5 (13)
s0 The current i2 could be neglected for low water utilization fac-
tors, which would simplify the model but result in an underesti-
where s0 and sε are respectively the electrical conductivity in mation of the voltage yield EOCV ¼ Eideal, where EOCV is an open
bubble-free electrolyte and in presence of bubbles. ε is the void circuit voltage (no electric current applied). Based on the available
fraction in the electrolyte that can be estimated as experimental data, the value of i2 can be very easily found by a
fitting procedure. During the calculation the value of i2 can be kept

 at a constant value. Applying Ohm's and Kirchhoff's laws to the
2 i 0:3
ε¼ q ¼ 0:0153 (14) electric circuit shown in Fig. 8, it is possible to write the following
3 ilim
set of equations:
according to [38]. A suggested value for the limiting current 8
< EAEC ¼ r2 $i2
density ilim is 300 kA m2 [39], that gives a satisfactory correlation
i ¼ i1 þ i2 (15)
for the experimental points reported by Ref. [38]. Note that the : 3
Eideal þ i1 $r1 ¼ r2 $i2
original Bruggeman theory, derived for effective conductivity in
multicomponent systems, has already been applied to liquid The total current which can be used for water decomposition is
electrolytes with internal void fraction [37], similarly to our case, correlated to the amount of demineralized water delivered to the
as validated by other literature works [38]. This is a rough electrolyte as primary reactant, which must be subject to elec-
empirical estimation of a complex phenomenon, but is quite ac- trolysis. Assuming the delivered water is utilized completely, the
curate for a lumped model. value of maximum current that must be applied dimax is defined by
An example of the share of the various overpotentials is shown the following relationship:
in Fig. 7.
Concentration losses due to consumption of reactants and mass imax $A ¼ 2$F$n_ H2 O (16)
flow rate limitation towards the reactive sites, which usually play
an important role in fuel cell analysis, do not have here a significant where F is Faraday's constant; A is the cell area and n_ H2 O is the water
impact on performances. In fact, the whole cell is filled with the molar flow at electrolyte inlet; all currents are expressed in terms of
electrolyte solution, which contains liquid water with a very large current densities (A cm2) referred to the total cell area.
excess with respect to the quantity consumed by the reaction. Hence, the water utilization factor can be correlated with the
Therefore mass transport limitations can be neglected. The only total current applied to the electrolyzer through the following
negative dilution effect could come from the formation of gaseous relation:
products, but it is taken into account by the part of the model which
deals with the presence of bubbles, which reduces the effective i3 ¼ hw $ðimax þ i2 Þ (17)
electrode surface and the useful section for ion conduction, as
described above. where hw is the water utilization factor.
By solving both Eq. (15) and Eq. (17), an equation for electrolyzer
voltage is obtained:

3.0
Ohmic overpotenƟal
2.5 AcƟvaƟon overpotenƟal
Erev
2.0
Voltage [V]

1.5

1.0

0.5

0.0
0.0 0.1 0.2 0.3 0.4 0.5
Current density [A/cm2]

Fig. 7. Example of overvoltages share (T ¼ 70  C, p ¼ 30 bar) obtained by the loss-

estimate model with fitting parameters in 1. Fig. 8. Equivalent electric circuit of an electrolyzer.
 J. Milewski et al. / Journal of Power Sources 269 (2014) 203e211 209

Eideal þ hw $imax $r1

EAEC ¼ r1 þ EDC (18)
r2 $ð1  hw Þ þ 1

Eq. (18) is the main relationship of the proposed model. AEC

voltage is described by a few factors which have clear physical
explanations: Eidealdideal voltage defined by the maximum
work required for an isothermal process of water hydrolysis (see
next chapter); hwdwater utilization factor defined by current
working conditions of the electrolyzer; imaxdmaximum current
density defined by the quantity of delivered water; r1darea
specific internal ionic resistance (ASIIR) defined by the overall
'permeability' of the electrolyte þ electrodes structure to alka-
line ions; and r2darea specific internal electronic resistance
(ASIER) defined by electric conductivity of the electrolyte,
intended as the KOH solution permeating the electrolysis cell
components. The compact expression of AEC voltage makes it
possible to clarify how the development of the specific compo- Fig. 9. Influence of pressure and temperature on the value of reversible electrolysis
nents of the electrolysis cell (e.g. R&D on new electrolyte com- voltage calculated following Eq. (20).
positions, new catalyst layer, etc.) should be related to
optimizing the resistance factors of Eq. (18), yielding additional Thus, partial pressures of hydrogen, oxygen and water vapor
information with respect to the classical approach focused on should be put into Eq. (20). Hydrogen and oxygen pressure values
optimizing the complex loss effects which contribute to the are quite obvious, since they correspond to the pressure of the gases
current densityevoltage curve. In other words, the reduced- during electrolyzer operationdif there is no pressure lossesd-
order equivalent circuit approach may facilitate a parallel and produced and stored gases, which is equal to the operating pressure
easier interpretation of cell behavior. of the electrolyte (varying by adequate humidity of the gases). For
water vapor, the partial pressure can be defined in terms of
hydrogen/oxygen purity. Here we assume that the water content in
3.2.1. Ideal voltage for the main chemical reaction of water the hydrogen and oxygen flows corresponds to the saturation line,
decomposition i.e. 100% of relative humidity for a given temperature of the elec-
The overall process occurring in the electrolysis cell includes a trolyte, according to the following equation [40]:
change of phase of the reactant (liquid water) in addition to split-
ting the water molecules into hydrogen and oxygen. The process pH2 O ¼ 1:508$108 $ðT  273:15Þ4
can be summarized as follows, where E [V] is the voltage applied to
 2:033$106 $ðT  273:15Þ3
obtain electrolysis and I [C mol1] is the current consumed, sup-
plying power to sustain water splitting: þ 2:425$104 $ðT  273:15Þ2  0:01105$ðT  273:15Þ
þ 0:2118
1
H2 OðlÞ þ E$I/H2ðgÞ þ O2ðgÞ (19)
2
The actual process of an AEC does not involve separate intro- 3.2.2. Maximum current densitydimax
duction of heat and electricity. All energy required is supplied In the proposed model, imax can be an assumption or a result
through the electrical current, and the minimum voltage required depending on the considered approach, namely a “design-point” or
to sustain the overall process can be calculated as: an “off-design” modeling of the electrolyzer. During “design-point”
modeling, the cell area can be changed depending on the nominal
1

R$T R$T pH2 $p2O2 Q inlet water flow and according to the optimization of other design
Eideal ¼ ln K þ ln þ (20) parameters (electrolyte type, distance between electrodes, etc.),
2$F 2$F pH O$p12 2$F
2 ref including the maximum current density imax. The value of imax is
then the result of a technical-economic optimization: the lower the
where: Rduniversal gas constant, FdFaraday’s constant [C mol1], value the larger the cell area of the electrolyzer for the same water
pdpartial pressure [Pa], refdreference, Kdchemical equilibrium utilization factor, with higher investment costs. An example could
constant (calculated with outlet concentrations), Qdwater evapo- be imax ¼ 4 A cm2, a reasonable value based on the researchers'
ration heat, Q ¼ f(T) [J mol1] [40]: own calculations and data taken from previous experiments. Since
the value of maximum current density is assumed after the opti-
Q ¼ 5$106 $ðT 273:15Þ4 þ0:0024$ðT 273:15Þ3 mization process, instead of a voltageecurrent density curve
0:4867$ðT 273:15Þ2 11:927$ðT 273:15Þþ44714 (21) (E ¼ f(i)) the voltage-water utilization factor curve (E ¼ f(hw)) can be
assumed as an electrolyzer characteristic curve.
It should be noted that the Nernst equation is referred to The typical “off-design” behavior of the cell is instead commonly
the partial pressures of reactants in the gaseous state (e.g. at defined by a voltageecurrent density curve (E ¼ f(i)), and in this
the outlet), although water is delivered as a liquid. Accordingly, case the area of the cell is a fixed parameter, while reactant flow
an additional term regarding the energy required for water parameters can be changed. Hence factor imax has to be calculated
evaporation (Q) must be included in the equation. Expression using Eq. (16), depending on the actual water flow delivered to the
for calculating Eideal is quite complex as can be seen in Fig. 9; electrolyzer and the assumed active area of the electrolyzer. This
the contribution given by evaporation heat is evident and causes approach makes it possible to compare various cells with different
the step in the ideal voltage curve. Assuming standard condi- areas. Actually, imax is assumed as the maximum current that
tions (p ¼ 1 atm, T ¼ 25  C), the reversible potential Eideal is should be consumed by the electrolyzer to convert all make-up
1.229 V. water into hydrogen and oxygen.
210 J. Milewski et al. / Journal of Power Sources 269 (2014) 203e211

3.2.3. Ionic and electric resistances These parameters are related with electrochemical behavior
Total ionic resistance of the cell r1 is calculated using Eq. (3) and (Tafel equation) and with resistive effects due to bubbles. A com-
rother is in the end a fitting parameter of the model. parison between the models and the experimental data is shown in
Existence of r2 (electronic resistance of the cell) means that Fig. 10.
some electric current flows through the cell itself without partici- After the fitting procedure, both models give similar errors with
pating in water decomposition and without being visible exter- average deviation of below 3% with respect to the experimental
nally. Thus, the water utilization factor defined by Eq. (17) is not the data, a reasonably low value which is also not too far from mea-
real value of water decomposed due to r2. The real value of water surement accuracy in typical applications.
utilization factor can be defined by the following function: The relevant difference in the results given by the two described
approaches is the behavior at very low current density values; in
1  hw Fig. 10 the curves are extended in the low current field in order to
hw;real ¼ hw  EAEC $ (22)
r2 $imax show this effect. The reduced-order model maintains its almost
linear trend whereas the loss-estimate model shows a drop in
In practice both values are very close to each other and the
voltage due to the non-linear description of the electrochemical
discussion is largely theoretical; nevertheless it should be noted
kinetics. As a consequence, the predicted “open circuit” voltage of
that there is a limiting value of hw for which there is no real water
the cell results are very different. In common operation of elec-
utilization.
trolysis alkaline units this issue is not very important, because the
lower power limit for continuous operation is set at around 20%, i.e.
4. Model calibration and validation
above the region where the two models differ. Nevertheless, further
experimental investigations at low current densities could be
Both presented models were calibrated based on the same set of
useful in terms of validating the proposed models and studying an
experimental data, obtained by the AEC unit described above in
extension of the operating range of the units.
Section 2. The main parameters of the model are listed in Table 1.
The obtained experiment results were also compared against
While volumetric (molar) flow rate for hydrogen is double than
other literature data for the same AEC operating temperature (see
for oxygen, output flows from the AEC unit are controlled so that
Fig. 11). Although the polarization curves have the same tendency,
total pressure is more or less the same for oxygen and hydrogen
the comparison showed that the data obtained during experiments
after the system is pressurized and the separation tanks (see Fig. 4)
shifted slightly above (z0.11 V) other literature data, evidencing
are filled in up to the rated output pressure. All the fitting param-
that the experimented unit was affected by higher electrical losses
eters were selected based on feasible ranges obtained from the
than the average literature references. This mismatch can be also a
literature, and calibrated to best fit the experimental results (for
result of the relatively large gap (14 mm) between the electrodes of
instance, electrolyte resistance of 300 cm2 S1 is compatible with
the electrolyzer under consideration. Nevertheless, this could be
conductivity values given in Ref. [41] for aqueous solutions). It
the effect of an intrinsic different internal design of the cells. Due to
turned out for both approaches the possibility to correctly predict
an interruption of the experimental investigation for maintenance
the cell behavior, validating their physical sense. The reduced-order
on the AEC unit, this issue was not investigated further, but future
model contains two fitting parameters:
planned experimental investigations will focus on this aspect as
well.

area specific internal electric resistance of the electrolyte r2
As a concluding remark, it can be said that from a general point

electrodes losses rother (which added to the electrolyte ionic
of view, both of the discussed models can reproduce rather accu-
resistance yields r1, see Eq. (3)).
rately the voltageecurrent density curve of an alkaline electrolyzer.
Thus, the reduced-order approach is an efficient alternative to the
On the other hand, the loss-estimate model contains the
more established loss-estimate approach currently proposed in
following five fitting parameters:
literature, and may be applied with the advantage of requiring a
lower number of parameters (mainly easily measurable), whereas

exchange current i0
the parameters of the electrochemical description in the loss-

anode and cathode charge-transfer coefficient aa and ac
estimate model are based on correlations that are strongly

limiting current density in void fraction estimation ilim
dependent on the nature of electrode materials and on local elec-

coefficient for bubble coverage of electrodes (see Eq. (14))
trochemical and fluid-dynamic conditions. On the other hand, with

Table 1
Main parameter of the two proposed models.

Parameter Value

RO model LE model

Temperature,  C 70 70
Electrolyte thickness, cm 0.66 0.66
Electrolyte specific flow, ml cm2 h 1.5 e
Hydrogen pressure, bar 25 25
Oxygen pressure, bar 23 e
Other (e.g. electrodes) resistance, cm2 S1 0.96a e
Electrolyte electric resistance, cm2 S1 300a e
Exchange current density, i0, A cm2 e 3.15 104a
Anode charge-transfer coefficient, aa e 0.39a
Cathode charge-transfer coefficient, ac e 0.44a
Limiting current density, ilim, A cm2 e 30a
Bubbles coverage coefficient e 0.0153a
Inlet electrolyte molarity, M (molKOH/l) 8 8
Fig. 10. Results of the models validation. Average error on level of 2.5% for reduced-
a
Fitting parameter or literature correlation. order model and of 1.5% for loss-estimate model.
 J. Milewski et al. / Journal of Power Sources 269 (2014) 203e211 211

061/10). We acknowledge and are grateful to SAPIO and H2Nitidor

for the support in the experimental investigation activity.

References

[1] EG&G.TechnicalServices, Fuel Cell Handbook, Tech. rep, seventh ed., DOE,
2004.
[2] W. Budzianowski, Rynek Energii 95 (4) (2011) 127e133.
[3] J. Kupecki, K. Badyda, Sofc-based micro-chp system as an example of efficient
power generation unit, Archives Thermodyn. 32 (3) (2011) 33e43.
[4] S. Campanari, G. Valenti, E. Macchi, G. Lozza, N. Ravida , Appl. Therm. Eng.
Available online 8 November 2013, ISSN: 1359e4311, doi:10.1016/j.
applthermaleng.2013.10.067.
[5] McKinsey, The Role of Battery Electric Vehicles, Plug-in Hybrids and Fuel Cell
Electric Vehicles, Tech. rep., 2010 www.zeroemissionvehicles.eu
[6] P. Costamagna, S. Srinivasan, J. Power Sources 102 (1e2) (2001) 253e269.
[7] H. Zhang, P. Shen, Chem. Rev. 112 (5) (2012) 2780e2832.
[8] S. Singhal, K. Kendall, High Temperature Solid Oxide Fuel Cells: Fundamentals,
Fig. 11. Comparison of polarization curves taken from available literature [22,27,33] Design and Applications, Elsevier, 2003.
(operating temperature at z70  C) against data obtained during experiments. [9] K. Huang, J.B. Goodenough, Solid Oxide Fuel Cell Technology, Woodhead
Publishing - CRC Press, 2009.
respect to the geometric design purpose, both proposed ap- [10] J.-H. Wee, Appl. Energy 88 (12) (2011) 4252e4263.
[11] A. Kulkarni, S. Giddey, J. Solid State Electrochem. 16 (10) (2012) 3123e3146.
proaches are limited since they neglect the real distribution of the
[12] L. Blum, R. Deja, R. Peters, D. Stolten, Int. J. Hydrogen Energy 36 (17) (2011)
reactant and products between the cell layers; thus, any optimi- 11056e11067.
zation of the cell geometry would be based only on its electrical [13] F. Díaz-Gonza lez, A. Sumper, O. Gomis-Bellmunt, R. Villafa fila-Robles, Renew.
effects, neglecting analysis of relevant internal mass transfer Sustain. Energy Rev. 16 (4) (2012) 2154e2171.
[14] L. Aline, Hydrogen Technology, Springer, 2008.
phenomena. [15] W. Kreuter, H. Hofmann, Int. J. Hydrogen Energy 23 (8) (1998) 661e666.
[16] G. Guandalini, S. Campanari, M.C. Romano, Comparison of gas turbines and
5. Conclusions power-to-gas plants for improved wind park energy dispatchability, in: Proc.
Of ASME Turbo Expo 2014, 2014, pp. GT2014e26838.
[17] M. Carmo, D. Fritz, J. Mergel, D. Stolten, Int. J. Hydrogen Energy 38 (12) (2013)
This work presents two approaches to the modeling of Alkaline 4901e4934.
Electrolyzers: the first based on a classic loss-estimate, requiring [18] V. Di Noto, S. Lavina, G. Giffin, E. Negro, B. Scrosati, Electrochim. Acta 57 (1)
(2011) 4e13.
calibration of specific parameters related to the electrochemical [19] M. Laguna-Bercero, J. Power Sources 203 (2012) 4e16.
and thermophysical behavior of the cell (e.g. activation losses, [20] M. Ni, M. Leung, D. Leung, Int. J. Hydrogen Energy 33 (9) (2008) 2337e2354.
bubble coverage effects, transport phenomena), the second based [21] J. Stempien, Q. Sun, S. Chan, Energy 55 (2013) 647e657.
[22] A. Roy, S. Watson, D. Infield, Int. J. Hydrogen Energy 31 (14) (2006)
on a reduced-order equivalent circuit approach, decoupled from 1964e1979.
the fundamental physical phenomena which occur in the electro- [23] A. Ursua, L.M. Gandia, P. Sanchis, Proc. IEEE 100 (2) (2012) 410e426 iD: 1.
chemical cell. [24] C. Schug, Int. J. Hydrogen Energy 23 (12) (1998) 1113e1120.
[25] M. Hammoudi, C. Henao, K. Agbossou, Y. Dube , M. Doumbia, Int. J. Hydrogen
Based on the results of experiments on a real electrolyzer car-
Energy 37 (19) (2012) 13895e13913.
ried out at Politecnico di Milano, both models show a capacity to [26] K. Onda, T. Kyakuno, K. Hattori, K. Ito, J. Power Sources 132 (1e2) (2004)
describe the general behavior of an electrolysis device and fitting 64e70.
on a specific device with sufficiently accurate results (average error [27] O. Ulleberg, Int. J. Hydrogen Energy 28 (1) (2003) 21e33.
[28] R. Gilliam, J. Graydon, D. Kirk, S. Thorpe, Int. J. Hydrogen Energy 32 (3) (2007)
below 3% after fitting of model parameters), which are quite similar 359e364.
for the two approaches. They differ just in the very low current [29] Y. Shin, W. Park, J. Chang, J. Park, Int. J. Hydrogen Energy 32 (10e11) (2007)
density zone, which is generally not interesting for practical ap- 1486e1491.
[30] P. Dieguez, A. Ursúa, P. Sanchis, C. Sopena, E. Guelbenzu, L. Gandía, Int. J.
plications due to safety and control issues. In both cases the number Hydrogen Energy 33 (24) (2008) 7338e7354.
of required parameters was reduced to the minimum and the [31] J.C. Ganley, Int. J. Hydrogen Energy 34 (9) (2009) 3604e3611.
models are suitable for use in complex hydrogen energy system [32] K. Zeng, D. Zhang, Prog. Energy Combust. Sci. 36 (3) (2010) 307e326.
[33] A. Ursúa, P. Sanchis, Int. J. Hydrogen Energy 37 (24) (2012) 18598e18614.
simulations. Therefore the evaluation of an equivalent electrical [34] R.L. LeRoy, C.T. Bowen, D.J. LeRoy, J. Electrochem. Soc. 127 (9) (1980)
circuit or reduced-order lumped description of the phenomena 1954e1962.
involved makes for an interesting alternative to the classical [35] A.J. Bard, L.R. Faulkner, Electrochemical Methods: Fundamentals and Appli-
cations, second ed., Wiley, New York, 2001.
approach. Further investigations should focus on the application of
[36] B.E. Conway, D.J. MacKinnon, B.V. Tilak, Transactions Faraday Soc. 66 (1970)
this approach to unsteady modeling of electrolyzers and coupling 1203e1226.
with intermittent energy sources. [37] G. Kreysa, M. Kuhn, J. Appl. Electrochem. 15 (4) (1985) 517e526.
[38] H. Vogt, R.J. Balzer, Electrochim. Acta 50 (10) (2005) 2073e2079.
[39] R. Piontelli, B. Mazza, P. Pedeferri, A. Tognoni, Electrochim. Metall. 2 (1967)
Acknowledgment 385.
[40] J.H. Keenan, F.G. Keyes, Thermodynamic and Transport Properties of Steam,
Work co-financed by Didactic Development Program for Faculty Wiley and Sons, 1959.
[41] D.R. Lide, CRC Handbook of Chemistry and Physics, ninetyth ed., 2009.
of Power and Aeronautical Engineering from the European Social [42] C.A.C. Sequeira, D.M.F. Santos, Cie ^ncia Tecnologia Dos Materiais 22 (3/4)
Fund Operational Programme Human Capital (POKL.04.01.01-00- (2010) 76e86.