Applied Energy 382 (2025) 125199

Contents lists available at ScienceDirect

Applied Energy
journal homepage: www.elsevier.com/locate/apenergy

Dynamic mass and energy balance model of a 50 kW proton exchange

membrane electrolyzer system
Georgios Sakas a ,∗, Philipp Rentschler b , Antti Kosonen a , Peter Holtappels b ,
Vesa Ruuskanen a , Pertti Kauranen a , Jero Ahola a , Roland Dittmeyer b
a
Lappeenranta–Lahti University of Technology LUT, P.O. Box 20, FI-53851, Lappeenranta, Finland
b Karlsruhe Institute of Technology, Karlsruhe, Germany

ARTICLE INFO ABSTRACT

Keywords: This paper presents a parameter-adjustable dynamic mass and energy balance process model for a 50 kW proton
Industrial-scale exchange membrane (PEM) electrolyzer plant. Energy and mass balances are derived for the electrolyzer stack,
Green hydrogen heat exchangers, and gas–liquid separation vessels. These balances, along with semiempirical submodels, are
Water electrolysis
integrated and solved within MATLAB system blocks connected through a Simulink environment. The model is
PEM
validated using experimental data obtained from a comparable industrial plant with similar pressure, power,
Modeling
Simulation
and system design parameters. The PEM plant is capable of operating within a pressure range of 5–55 bar. The
electrochemical and thermal behaviors, along with hydrogen production, are compared between the process
model and the actual PEM plant to assess the accuracy of the simulations. The results demonstrate a satisfactory
agreement between the model predictions and experimental data. In nominal operation, 68.5% of the total
power supplied to the stack is converted into hydrogen, while the remaining power is dissipated as heat due
to overvoltages. This excess heat is primarily transferred through the heat exchangers to the secondary leg.

1. Introduction advantage, PEM electrolyzers use expensive catalysts like platinum and
iridium, which could limit their scalability for larger production [2].
Hydrogen is primarily used in heavy industries for crude oil re- This paper will focus on commercial-size PEM electrolyzer plant
fining, ammonia production, and methanol synthesis, accounting for modeling, including the stack and essential peripheral unit operations.
93% of the total hydrogen consumption. Presently, the main method So far, the literature has focused on characterizing the electrochem-
of hydrogen production involves the use of fossil fuels. However, the ical and thermal behavior at the cell/stack level [3]. Additionally,
emergence of green hydrogen (GH2) offers a promising alternative there is a scarcity of literature on the simulation and modeling of
to mitigate environmental impacts. GH2 is derived from renewable PEM water electrolyzer systems that provide detailed characterization
energy sources (RES) and is labeled as the fuel of the future. Its of commercial-size systems that include all the important peripheral
distinction lies in being both clean and easily storable, making it a components, such as gas–liquid separation vessels, heat exchangers,
compelling solution for sustainable energy needs. While the current use
and pumps. To fill this research gap, in this work, a commercial 50 k W
of hydrogen is centered around heavy industries, GH2 is designed to
PEM water electrolyzer system is modeled and validated with exper-
extend its application beyond this sector. Its intended use spans various
imental data from an analogous operating system, aiming to provide
applications, including aviation, maritime transport, and energy [1].
a numerical tool to investigate the dynamic performance and energy
Low-temperature electrolyzers, specifically the proton exchange
efficiency of such systems.
membrane (PEM) and the alkaline water electrolyzer (AWE), stand out
Examining the published research work on the simulation and
as two technologically mature options for industrial-scale GH2 produc-
tion. These electrolyzers use the water electrolysis reaction to produce modeling of PEM water electrolyzer systems, Yigit and Selamet [4]
high-purity GH2. Among the two, the AWE is more mature and widely performed a simulation for a PEM system with the target of under-
adopted than the PEM, owing to its advantages of low-cost catalysts standing how the stack and the overall system perform and experience
and a prolonged lifespan. However, the AWE faces limitations in terms inefficiencies under varying operating conditions. The simulation relied
of achievable current density compared with the PEM. Despite this on mathematical models solved by using MATLAB/Simulink software.

∗ Corresponding author.
E-mail address: [email protected] (G. Sakas).

https://doi.org/10.1016/j.apenergy.2024.125199
Received 26 April 2024; Received in revised form 15 October 2024; Accepted 19 December 2024
Available online 1 January 2025
0306-2619/© 2024 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
G. Sakas et al. Applied Energy 382 (2025) 125199

Nomenclature 𝛥 Difference
𝛿 Thickness
Acronyms 𝜂 Efficiency
𝜆 Latent heat
AC Alternating current
𝜇 Viscosity
AWE Alkaline water electrolyzer
𝜌 Density
DC Direct current
𝛴 Sum
GH2 Green hydrogen
𝜎 Stefan–Boltzmann constant
HE Heat exchanger
𝜏 Time constant
IGBT Insulated gate bipolar transistors
Subscript
ODE Ordinary differential equation
PEM Proton exchange membrane A Stack A
PSU Power supply unit act Activation
PtX Power to x amb Ambient
RES Renewable energy sources an Anode
SEC Specific energy consumption B Stack B
Variables c Cell
cat Cathode
𝛥𝐻 Higher heating value
cd Cold
𝑚̇ Mass flow rate
cn Concentration
𝑛̇ Molar flow rate
con Consumed
𝑄̇ Heat flow
cv Convection
𝐴 Surface area
cw Cooling water
𝐶 Thermal capacitance
ef Effective
𝐷 Diameter
eo Electroosmotic
𝐹 Faraday constant
F Faraday
𝐻 Height
gen Generated
ℎ Heat transfer coefficient
h Hot
𝐼 Current
hc Hydrogen crossover
𝑖 Current density
in Inlet
𝐾 Darcy coefficient
init Initial
𝑘 Thermal conductivity
liq Liquid
𝐿 Length
m Membrane
𝑀 Molar mass
mp Makeup water
𝑁 Number
net Equilibrium
𝑛 Coefficient
oc Oxygen crossover
𝑃 Pressure
ohm Ohmic
𝑝 Model parameter
out Outlet
𝑞 Heat transfer rate
pe Pressure effect
𝑅 Gas constant
pp Pump
𝑟 Model parameter
rd Radiation
𝑠 Model parameter
rev Reversible
𝑇 Temperature
s Stack
𝑡 Model parameter
s,H2 Hydrogen gas separation
𝑡 Time
s,O2 Oxygen gas separation
𝑈 Voltage
s,l Liquid separation
𝑉 Volume
sec Secondary
𝑧 Moles of electrons transferred
SP Set-point
Cp Specific heat
sur Surface
Nu Nusselt number
tn Thermoneutral
Pr Prandtl number
v Vapor
Re Reynolds number
vap Vaporization
w Water

The stack design in the simulation was considered to comprise five cells
with an active area of 100 cm2 . The study presented several findings: the overall stack efficiency. (3) Raising the operational temperature,
(1) Notably, the anodic activation losses and the ohmic losses caused conversely, lowered the polarization curve of the stack, resulting in an
by ionic resistance were considerably greater than the electronic and enhanced overall stack efficiency. (4) Alterations in temperature had
cathodic activation losses. (2) Elevating the operational pressure led a more pronounced impact on cell voltage than pressure changes. (5)
to an increase in the stack voltage, coinciding with a reduction in Lastly, the research compared the losses of the stack with the losses

2
G. Sakas et al. Applied Energy 382 (2025) 125199

in the power supply, pump, fan, controller, and sensors. In all cases Table 1
of current densities, the stack losses surpassed the losses of the other Technical characteristics of the experimental 50 k W plant. The voltage
range is estimated during operation at a constant temperature of 60 ◦ C.
components. However, at lower current densities, the difference in
Parameter Value Unit
stack and other component losses was less significant compared with
higher current densities. Number of stacks 2 –
Number of cells per stack 40 –
Expanding the research field of modeling PEM water electrolyzer
Stack length 98 cm
systems, Espinosa et al. [5] developed a model for a high-pressure Stack height 25 cm
46 kW PEM water electrolyzer system, which they validated using Stack width 35 cm
a real-scale demonstration. The entire model was built in Matlab- Active cell area 875 cm2
Simulink, and several experiments were conducted at different temper- Maximum operating pressure 55 bar
◦
Maximum operating temperature 90 C
atures and loads to align the modeled electrochemical behavior with
Operating cell block current range 0–300 A
that of the real system. Their analysis showed that high temperature Maximum allowable cell block current 750 A
and low pressure operating conditions improve electrolyzer efficiency. Cell block voltage range 1.27–2.16 V
Continuing along the same lines, Ni et al. [6] constructed a model for a Total stack voltage range 50.8–86.4 V
PEM electrolyzer plant to investigate its energy and exergy efficiency. Separation tank volume 15.2 L
The model was validated by using polarization curve data extracted
from Ioroi et al. [7]. The modeled plant included a single PEM stack,
a heat exchanger, and an oxygen separation unit. The findings of the • The study presents a dynamic mass and energy balance model
study suggest that the energy efficiency of the plant can be enhanced methodology to simulate a 50 kW PEM electrolyzer plant process.
by using a thinner PEM electrolyte and increasing the temperature of
• The overall process model incorporates several sub-models for
the operating process.
all unit operations within the PEM system, including stack, gas–
Hancke et al. [8] analyzed the results from experimental testing of a
liquid separation vessel, and heat exchanger sub-models. In ad-
12 kW PEM water electrolyzer system. The research focused on investi-
dition, the energy and mass balance for each unit operation are
gating the effect of pressure on electrolyzer performance and discussing
derived and presented in this work.
the trade-off between stack and system efficiency. The case studies con-
• Measurements from a commercial 50 kW PEM electrolyzer plant
sidered scenarios where downstream mechanical compression loss is
were collected.
reduced due to high-pressure stack operation. The study confirmed that
• The measurements were used to build, configure and verify the
high-pressure PEM stacks have diminished performance, particularly
dynamic process model. Specifically, multiple measured polar-
above 30 bar, with a significant drop occurring over 100 bar. It also
revealed that high-pressure PEM water electrolyzer systems consume ization curves at various temperatures were used to fit the op-
more energy compared to state-of-the-art electrolyzer systems with erational polarization curve to the model. Additionally, verifica-
mechanical compression. However, the business case remains open, and tion was performed at multiple dynamic levels, including stack
further techno-economic optimization studies are needed to determine temperature, plant hydrogen production, and H2 inO2 impurities.
the trade-off between systems that combine low-pressure stack oper- These parameters were compared between the operational plant
ation with high mechanical compression loss and those that combine and the model to ensure the accuracy of the dynamic simulation.
high-pressure stack operation with reduced mechanical compression • The main contribution of this research is the utilization of the
loss. verified 50 kW commercial-scale PEM water electrolyzer process
Pfennig et al. [9] developed an initial model of megawatt-size model to investigate the performance and energy efficiency of
PEM water electrolyzer systems. Since the authors did not have access such systems. Mainly, the stack heat balance and SEC are ana-
to industrial measurements, the model was built using parameters lyzed, and the energy flow across the plant is illustrated using a
from the literature and operational points from published commercial Sankey diagram.
PEM stacks. The study aimed to introduce a preliminary approach
The remainder of the publication is structured as follows: Sec-
to modeling such systems. However, the model only investigated the
tion 2 describes the experimental PEM plant process used to gather the
performance of the stack, excluding the peripheral components. The
necessary data. Section 3 presents the dynamic process model and out-
authors also noted that there are hardly any corresponding system
lines the methodology. Section 4 elaborates on the model verification
models of this size in the literature that dynamically simulate the
and discusses the results of the study. Section 5 contains conclusions,
performance of commercial-scale PEM water electrolyzer systems.
providing a summary of the article and highlighting the main findings.
1.1. Aim of the study
2. Plant process of proton exchange membrane electrolysis
Summarizing prior research on PEM electrolyzer plant modeling,
it is evident that there is a scarcity of research articles addressing The PEM electrolyzer plant under investigation is manufactured in
commercial- or industrial-scale PEM water electrolyzers as a complete 2013. In Fig. 1, a simplified version of the process diagram for the
plant process, including all essential peripheral unit operations. The experimental setup is presented. The system consists of a total of four
majority of modeling articles in the field of PEM electrolyzers tend PEM stacks, each with a capacity of 25 kW and 100 kW in total. For
to concentrate on characterizing the electrochemical and thermal be- short-term peak loads, the system can be operated at 300 kW for a
havior of the stack, rather than considering the system at a holistic maximum of 15 min, whereby the maximum current per cell block is
level [3]. Moreover, these studies typically focus on small-scale and limited to 750 A (0.857 A cm−2 ) and the maximum cell block voltage
single-cell applications within the watt range. Furthermore, verifica- to 120 V DC. In this study, half of the system, i.e., two stacks, were
tion is typically carried out using a single polarization curve at a operated. As a result, the nominal power supply for the process is
fixed temperature. Similar observations were reported from Espinosa- 50 kW. The technical characteristics of the experimental setup are
Lopez et al. [5] literature review, indicating that while research on shown in Table 1.
PEMWE is extensive, there is limited work on modeling commercial-
scale PEM plant systems that also include all the important peripheral 2.1. Process diagram description
components, not only the cell or stack.
The novel contributions of our study concerning the PEM elec- As shown in Fig. 1, the PEM electrolyzer plant typically employs
trolyzer plant modeling are summarized as follows: alternating current (AC), initially directed to the power supply unit.

3
G. Sakas et al. Applied Energy 382 (2025) 125199

Fig. 1. Plant diagram of the experimental setup. In the cathode and anode circulation lines, the cathode line carries a two-phase flow of hydrogen gas mixed with water and a
small fraction of oxygen gas (impurities), while the anode line carries a two-phase flow of oxygen gas and water with a small fraction of hydrogen gas (impurities).

The AC electric current is converted into direct current (DC) with a Before re-entering the stack, the liquid undergoes a cooling process
transformer and rectifiers. The three common rectifier topologies for to achieve the desired temperature. This cooling takes place in heat
industrial-scale water electrolysis are 6-pulse and 12-pulse thyristor exchangers and is crucial for regulating the thermal behavior of the
rectifiers and buck rectifiers based on insulated gate bipolar transistors stack. The heat generated by overpotential losses within the stack can
(IGBT) [10,11]. In continuation, the DC current, along with AC rip- accumulate, requiring extraction from both the stack and associated
ple [12], is supplied to both stack A and stack B, converting the water unit operations in the circulation loop.
into hydrogen molecules at the catalyst surface of the cathode electrode
and into oxygen molecules at the catalyst surface of the anode electrode 3. PEM plant modeling
by the water splitting reaction [9,13]. Each stack is constructed of 40
cells connected electrically in series. The electrolyzer can be operated In this section, the methodology for stack- and system-level mod-
up to a maximum pressure of 55 bar. The pressure is regulated by the eling is presented and explained in detail. The stack-level model con-
gas outlets, and there is the same pressure at the anode and the cathode, sists primarily of electrochemical, thermal, and mass transfer mod-
which reduces the mechanical load on the membrane. els. The system-level modeling incorporates the main unit operations
The produced hydrogen and oxygen gases exit the stacks as a two- dynamically connected in MATLAB Simulink. These unit operations
phase flow of water and gas, flowing into their respective hydrogen include: (1) stack; (2) hydrogen–liquid separation tank; (3) oxygen–
and oxygen separation tanks. The vertical gas–liquid separation tanks liquid separation tank; and (4) heat exchangers. Furthermore, the
of 15.2 L volume employ gravity mechanisms to separate the gas phase model methodology for each unit is presented and discussed.
from the liquid phase [14]. The gas remains in the upper tank area, and
the liquid settles in the lower tank area. The separated hydrogen gas is
3.1. Stack hydrogen production
typically stored in an outer storage vessel until use, while the oxygen
gas is released into the atmosphere. However, the use or storage of the
The total hydrogen production rate in a water electrolysis stack,
produced hydrogen and oxygen gases is beyond the scope of this study,
whether for the AWE or the PEM, is directly proportional to the
and the gases are vented into the atmosphere.
supplied DC current and can be estimated as follows [16]:
The liquid from the lower section of the separation tanks circulates ( )
𝑖 ⋅ 𝐴ef
back to the stacks with the assistance of centrifugal pumps. Centrifugal 𝑛̇ H2 = 𝜂F ⋅ c ⋅ 𝑁c , (1)
𝑧⋅𝐹
pumps serve two main purposes. First, they regulate the mass flow rate
necessary to maintain circulation in the closed loop. Secondly, they where 𝑖c represents the current density, 𝐴ef stands for the effec-
ensure the system efficiency by facilitating the removal of produced tive/active cell area, z denotes the number of moles of electrons
gases from the stack. Additionally, in connection with heat exchangers, transferred in the reaction (which is two for hydrogen), 𝐹 represents
they control the cooling of the stack and the peripheral units [15]. the Faraday constant, 𝑁c is the total number of cells electrically

4
G. Sakas et al. Applied Energy 382 (2025) 125199

connected in series in the stack, and 𝜂F is the Faraday/current efficiency 3.2.2. Activation overvoltage
of the stack. The activation overvoltage 𝑈act is caused by the reaction kinetics
Current efficiency 𝜂F losses are primarily caused by shunt currents. at the anode and cathode electrode interfaces [24,25]. It can be esti-
The manifold ducts within the stack can introduce additional resistance mated using Eq. (7), where 𝑡1 , 𝑡2 , and 𝑡3 are fitting model parameters
paths, through which a portion of the supplied current may flow [17]. associated with temperature. Similarly, 𝑠 is a model parameter that
These shunt currents lead to corrosion of the manifold ducts and result represents the Tafel slope coefficient, and 𝐼 is the current supplied to
in the loss of current through secondary electrochemical reactions. the stack [20]:
Shunt currents are commonly observed in industrial-scale AWE stacks ( ( ) ) (( ) ( ) )
𝐼 𝑡 𝑡 𝐼
manufactured as a bipolar configuration [18] owing to the geometric 𝑈act = 𝑠 ⋅ log 𝑡 ⋅ + 1 = 𝑠 ⋅ log 𝑡1 + 2 + 3 ⋅ +1 .
design and the use of circulating conductive electrolyte. On the other 𝐴ef 𝑇 𝑇2 𝐴ef
hand, PEM water electrolyzer stacks circulate highly resistive water (7)
and typically do not exhibit a high level of shunt currents, and thus,
an assumption is often made that the current efficiency is 100%. 3.2.3. Ohmic overvoltage
Nevertheless, in this study, the current efficiency is not assumed to be The ohmic overvoltage, denoted as 𝑈ohm , arises from the electrical
100%. This approach allows us to examine the validity of the commonly resistances present within the electrolyzer cell. It signifies the voltage
used assumption. drop between the anode and cathode electrodes. This drop in voltage is
a consequence of the resistance of the solid material to electric current
3.2. Electrochemical model
and the resistance of the liquid to the transport of ion flux. It can be
The electrochemical reaction in water electrolyzer stacks occurs estimated using Eq. (8), where 𝑟1 and 𝑟2 are fitting model parameters
relatively fast compared with other physical phenomena. Therefore, related to temperature [20]:
( ) ( )
a static characteristic model is sufficient for predicting system-level 𝐼 𝐼
𝑈ohm = 𝑟 ⋅ = 𝑟1 + 𝑟2 ⋅ 𝑇 ⋅ . (8)
performance, allowing us to ignore the details of electrochemical pro- 𝐴ef 𝐴ef
cesses [19]. The overall cell voltage 𝑈c can be expressed as the sum
of the reversible voltage 𝑈rev and additional overvoltages, namely the
activation 𝑈act , ohmic 𝑈ohm , and concentration overvoltages 𝑈cn [20]: 3.3. Stack thermal model—Energy balance
𝑈c = 𝑈rev + 𝑈act + 𝑈ohm + 𝑈cn . (2)
The temperature of the electrolyzer stack can be estimated using
the lumped-capacitance method. This method offers a straightforward
Concentration overvoltages primarily occur at high current densities approach to dynamically estimate the stack temperature over time. For
because of mass transfer limitations. In our case, the current density is an electrolyzer, this method is justified, because the heat convection
relatively low, 0.34 A cm−2 , at the nominal operating point, allowing us from the stack to the environment is slower than the heat conduction
to disregard concentration overvoltages. within the body [26]. Therefore, the ordinary differential equation
(ODE) for the stack temperature is formulated as follows:
3.2.1. Reversible voltage d𝑇
The reversible voltage represents the minimum electrochemical po- 𝐶⋅ = 𝑄̇ gen − 𝑄̇ liq − 𝑄̇ amb − 𝑄̇ vap , (9)
d𝑡
tential required for the water splitting reaction to occur. It depends on
thermodynamic properties, such as pressure and temperature, and can where 𝑄̇ gen represents the heat generated by overpotentials, 𝑄̇ liq indi-
be estimated using the Nernst equation as follows [12]: cates the cooling caused by the colder circulating water, 𝑄̇ amb denotes
the heat loss to the ambient, and 𝑄̇ vap represents the latent heat of
⎛ (𝑃 − 𝑃 ) ⋅ (𝑃 − 𝑃 )1∕2 ⎞
𝑅⋅𝑇 vaporization.
⋅ ln ⎜ ⎟.
0 cat v,w an v,w
𝑈rev = 𝑈rev + (3)
2⋅𝐹 ⎜ 𝑃
3∕2 ⎟ The heat loss from the stack to the ambient is estimated as convec-
⎝ sd ⎠
tive plus radiative heat transfer [27]:
⎛ (𝑃 − 𝑃 )3∕2 ⎞ 𝑄̇ amb = (ℎcv + ℎrd ) ⋅ 𝐴sur ⋅ (𝑇s − 𝑇amb ), (10)
𝑅⋅𝑇
⋅ ln ⎜ ⎟.
0 v,w
𝑈rev = 𝑈rev + (4)
2⋅𝐹 ⎜ 𝑃
3∕2 ⎟ where ℎcv is the convective heat transfer coefficient, and ℎrd is the
⎝ sd ⎠
radiative heat transfer coefficient.
0 under stan-
The first term represents the reversible potential 𝑈rev The thermal capacitance of the entire stack is denoted by 𝐶. It
dard conditions, and the second term expresses the deviation from can be determined either theoretically or experimentally. In theoret-
standard conditions due to variations in pressure and temperature. ical estimation, the thermal capacitance is a function of the density,
𝑅 is the universal gas constant, 𝑇 is the temperature at which the heat capacity, and volume of each element that collectively forms
electrochemical reaction occurs, 𝑃 is the process pressure, 𝑃v,w is the an electrolyzer cell [15,27]. In experimental estimation, the thermal
water vapor saturation pressure, and 𝑃sd is the pressure at standard capacitance can be derived from the cooling-down data of the stack.
conditions. It should be noted that the water vapor pressure is assumed The theoretical model of Newton’s law of cooling can be fitted to the
to be equal to the water vapor saturation pressure due to the assump- cooling-down data to estimate the thermal capacitance of the stack. For
tion of thermodynamic equilibrium between the gas and liquid phase this study, the thermal capacitance is determined experimentally, and
of water. However, because of liquid injection, the water activity of the the equations for Newton’s law of cooling are as follows:
catalyst layer cannot be ignored, and this pressure may not actually be −𝑡
equal to the saturated water vapor pressure. Nevertheless, for the scope 𝑇s (𝑡) = 𝑇amb + (𝑇s (0) − 𝑇amb ) ⋅ exp( ), (11)
𝜏
of this study, the saturation water vapor pressure is used. The standard
𝐶
reversible potential can be empirically estimated as [21,22]: 𝜏= , (12)
ℎ ⋅ 𝐴sur
0
𝑈rev = 1.5184 − 1.5421 ⋅ 10−3 ⋅ 𝑇 + 9.526 ⋅ 10−5 ⋅ 𝑇 ⋅ ln 𝑇 + 9.84 ⋅ 10−8 ⋅ 𝑇 2 . (5)
ℎ = ℎcv + ℎrd , (13)

Similarly, the saturated water vapor pressure is empirically calcu- where 𝑇s is the temperature of the stack, 𝑇amb is the ambient temper-
lated as follows [23]: ature, and 𝜏 represents a time constant. In Eq. (13), 𝐴sur stands for
( ) the surface area of the stack, and ℎ denotes the overall heat transfer
7699.68
𝑃v,w = exp 81.6179 − − 10.9 ⋅ ln 𝑇 + 9.5891 ⋅ 10−3 ⋅ 𝑇 . (6)
𝑇 coefficient, which is the sum of the convective and radiative heat
transfer coefficients.

5
G. Sakas et al. Applied Energy 382 (2025) 125199

Fig. 2. Determining stack thermal capacitance by fitting experimental cooling data to

the theoretical model based on Newton’s law of cooling.
Fig. 3. Overview of mass transfer mechanisms and cell mass balance.

The fitting process is performed in MATLAB using the fminsearch

function [28,29]. The results of fitting Newton’s law of cooling (Eq. stack is located. For simplification, the surrounding area temperature
(11)) to the cooling-down experimental data are shown in Fig. 2. It is assumed to be equal to the ambient temperature. The stack surface
is noteworthy that during the experiment the pumps were closed and temperature is considered equal to the overall stack temperature, ac-
there was no cooling by the circulating water. While the thermal ca- cording to the lumped-capacitance method. This method assumes that
pacitance exhibits some variation with stack temperature, the changes the temperature remains uniform throughout the body.
are minimal, allowing us to assume a constant value to simplify the 3
ℎrd = 4 ⋅ 𝜖 ⋅ 𝜎 ⋅ 𝑇mean . (19)
calculations. In our scenario, the mean value of the thermal capacitance
function as shown in Fig. 2 is selected as a constant for stack thermal Concluding, the cooling effect resulting from water vaporization can
capacitance, set at 404 k J ◦ C−1 . be estimated using Eq. (20), where 𝑚̇ vw,out represents the total mass
The heat generated by overvoltages across the entire stack is calcu- flow rate of water vapor at the stack outlet, and 𝜆H2 O denotes the latent
lated as [20]: heat of vaporization (2308 k J k g−1 , 80 ◦ C) [31].
𝑄̇ gen = 𝑁c ⋅ (𝑈c − 𝑈tn ) ⋅ 𝐼 , (14) 𝑄̇ vap = 𝑚̇ vw,out ⋅ 𝜆H2 O . (20)
where 𝑁c is the number of cells connected in series, and 𝑈tn is the
thermoneutral voltage: 3.4. Mass transfer model
𝛥𝐻
𝑈tn = , (15) In water electrolysis, mass transfer occurs between the cathodic and
𝑧⋅𝐹
where 𝛥𝐻 is the higher heating value (HHV) heat of reaction. anodic cell components, involving the transfer of both water and gas.
The circulating water, flowing through the heat exchangers before Three methods of water transfer have been identified in the literature:
re-entering the stack, plays an important role in cooling down the stack. (1) diffusion; (2) electroosmotic drag; and (3) water transfer due to
This cooling effect is estimated as follows: the pressure gradient across the cell [4,32]. Gas transfer, also known
as gas crossover, results from two main mechanisms: (1) convection
𝑄̇ liq = 𝑚̇ w,in ⋅ Cpw ⋅ (𝑇s − 𝑇w,in ), (16) and (2) diffusion of the product gases through the membrane. The
convective mechanism is further divided into two phenomena leading
where 𝑚̇ w,in is the total mass flow rate of water at the stack inlet, Cpw
to gas crossover, namely cell differential pressure and electroosmotic
is the specific heat of water (4.18 k J k g−1 ◦ C), and 𝑇w,in represents the
drag [33].
water temperature at the inlet.
The overall cell mass balance and the mass transfer mechanisms
According to [27,30], the convective heat transfer can be modeled
considered in this model are illustrated in Fig. 3, where 𝑚̇ H2 represents
as a low-velocity forced convection scenario. The Nusselt number for
the mass flow rate of hydrogen production, 𝑚̇ O2 is the mass flow rate of
forced convective flow through a rectangular body can be estimated
oxygen production, 𝑚̇ hc is the mass flow rate of hydrogen crossing the
empirically, as shown in Eq. (17). For a slow-moving air flow, the
anode cell compartment, 𝑚̇ oc is the mass flow rate of oxygen crossing
Reynolds number Re is equal to 5000, and the Prandtl number Pr is
the cathode cell compartment, 𝑚̇ pe indicates water transfer from the
0.7. The convective heat transfer coefficient is calculated as shown in cathode to the anode due to the pressure gradient, 𝑚̇ eo represents water
Eq. (18), where 𝑘 is the thermal conductivity of the air at 20 ◦ C, and transfer from the anode to the cathode due to electroosmotic drag, and
𝐿s is the total length of the stack. 𝑚̇ con denotes the water consumption required for the water-splitting
Nu = 0.66 ⋅ Re0.675 ⋅ Pr1∕3 , (17) reaction.
It should be noted that, unlike standard PEM electrolyzer cells,
𝑁𝑢 ⋅ 𝑘 which receive water only on the anode half-cell for the electrochemical
ℎcv = . (18)
𝐿s reaction of water splitting, our experimental setup introduced wa-
ter into both half-cell compartments to improve thermal distribution
The radiative heat transfer coefficient is estimated as shown in within the cells and the entire stack and create a uniform temperature
Eq. (19), where 𝜖 is the emissivity of stainless steel, 𝜎 is the Stefan– profile. Additionally, the hydrogen gas introduced into the anode half-
Boltzmann constant, and 𝑇mean is the mean temperature between the cell represents recirculated impurities, and the same applies to the
temperatures of the stack surface and the surrounding area where the oxygen gas introduced into the cathode half-cell.

6
G. Sakas et al. Applied Energy 382 (2025) 125199

Based on Lin et al. [34], the uniformity of the flow field can
significantly impact the performance of a PEM electrolyzer. This is why
it is generally recommended to design electrolyzers to operate with
uniform voltage, flow, pressure, and temperature profiles across the
cells and stack.

3.4.1. Water transfer

In the process of PEM electrolysis, protons (H+ ions) move from
the anode cell compartment to the cathode through the membrane,
and in this transfer, they can carry water molecules with them. This
phenomenon is referred to as electroosmotic drag and is recognized
as the primary mechanism for water transport in PEM cells [32]. The
dragged water molar flow is estimated as:
𝐼
𝑛̇ eo = 𝑛eo ⋅ , (21)
𝐹
where 𝑛eo is a water electroosmotic drag coefficient, and it can be
empirically calculated as [4]:
𝑛eo = 0.016 ⋅ 𝑇 − 2.89556. (22) Fig. 4. Comparison of experimental H2 inO2 data with the modeled H2 inO2 .

The molar flow of water from the cathode to the anode is also driven Table 2
Experimental values of H2 inO2 impurities at various pressures and
by the pressure effect or the pressure gradient present between the half-
current supplies.
cells. Based on Darcy’s law, the estimation of this molar flow is derived
Pressure (bar) Current supply (A) H2 inO2 (mol%)
as [32,35]:
𝐾darcy ⋅ 𝐴m ⋅ 𝜌w ⋅ 𝛥𝑃 5 30 0.959
𝑛̇ pe = , (23) 5 300 0.676
𝛿m ⋅ 𝜇w ⋅ Mw
10 30 0.952
where 𝐾darcy represents the membrane permeability to water (1.58 ⋅ 10 300 0.676
10−18 m2 ), 𝜇w is the water viscosity (1.1 ⋅ 10−3 Pa s), 𝛥𝑃 is the pressure
difference between the cathode and anode half-cells, 𝜌w is the density
of water [36], Mw is the molar mass of water, 𝐴m is the membrane
area, and 𝛿m (0.018 cm) represents the membrane thickness.
The total mass flow rate at the inlet is estimated as shown in Eq.
3.4.2. Gas transfer (26), where 𝑚̇ w,cat,in is the mass flow rate of water in the inlet of the
The gas transfer process is simulated with the assistance of exper- cathode side, 𝑚̇ H2 ,cat,in is the mass flow rate of the dissolved hydrogen
imental data of H2 inO2 impurities from the PEM plant. The recorded gas in the inlet of the cathode side, 𝑚̇ O2 ,cat,in is the mass flow rate of
H2 inO2 values during nominal and minimum current supply are fitted the dissolved oxygen gas in the inlet of the cathode side, 𝑚̇ w,an,in is the
to Eq. (24). The fitting process is executed with the help of the fmin- mass flow rate of water in the inlet of the anode side, 𝑚̇ H2 ,an,in is the
search function in MATLAB [28,29]. The impurities are represented as mass flow rate of the dissolved hydrogen gas in the inlet of the anode
an exponential decay function dependent on current supply, a behavior side, and 𝑚̇ O2 ,an,in is the mass flow rate of the dissolved oxygen gas in
documented in the literature [37]. the inlet of the anode side.
∑
𝑓 (𝐼) = 𝑝1 ⋅ exp(𝑝2 ⋅ 𝐼). (24) 𝑚̇ in = 𝑚̇ w,cat,in + 𝑚̇ H2 ,cat,in + 𝑚̇ O2 ,cat,in + 𝑚̇ w,an,in + 𝑚̇ H2 ,an,in + 𝑚̇ O2 ,an,in .
(26)
The experimental values of H2 inO2 used for the fitting are presented
in Table 2. It is also evident that there is no substantial difference
between the 10 bar case and the 5 bar case. The resulting model The total mass flow rate generated within the cell is calculated
parameters, for the 5 bar case, are 0.9889 for 𝑝1 and −0.0013 for 𝑝2 . as illustrated in Eq. (27). This value results from the product of the
Additionally, the accuracy of the model is compared with experimen- hydrogen and oxygen production within the cathode and anode cell
tal data, as illustrated in Fig. 4. Based on percentage deviation, the components, respectively. Thus, the total mass flow rate at the outlet
accuracy stands at 92.36%. The fit is slightly off because the model is estimated according to Eq. (28):
assumes the mixing of H2 inO2 occurs instantaneously. This assumption ∑
𝑚̇ gen = 𝑚̇ H2 + 𝑚̇ O2 , (27)
results in a faster response of the impurities in the model compared to
the slower response observed in the real operational plant, as shown
in Fig. 4. However, as the precision of the mass and gas transfer is not ∑
𝑚̇ out = 𝑚̇ w,cat,out +𝑚̇ H2 ,cat,out +𝑚̇ O2 ,cat,out +𝑚̇ w,an,out +𝑚̇ H2 ,an,out +𝑚̇ O2 ,an,out . (28)
the primary focus of this study, the accuracy achieved is considered
satisfactory.

3.5. Cell mass balance 3.6. System-level model methodology

The mass balance equation for the modeled PEM cell is expressed The system-level process model for the PEM plant is developed
by Eq. (25). This equation states that the rate of mass accumulation within the MATLAB Simulink environment, with the corresponding
within the cell is equal to the sum of the total mass flow rate at the
diagram illustrated in Fig. 5. A total of eight MATLAB Simulink block
inlet (𝑚̇ in ) and the total mass flow rate of the produced hydrogen and
models have been developed: two for the vertical gas–water separation
oxygen (𝑚̇ gen ), subtracted by the total mass flow rate at the outlet (𝑚̇ out )
vessels, two for PEM stacks A and B, and four for each heat exchanger.
and the consumed water (𝑚̇ con ).
The pumps are not explicitly modeled because the mass flow rate after
𝑑𝑚 ∑ ∑ ∑
= 𝑚̇ in + 𝑚̇ gen − 𝑚̇ out − 𝑚̇ con . (25) the pump is considered a model parameter, provided by the user. The
𝑑𝑡

7
G. Sakas et al. Applied Energy 382 (2025) 125199

Fig. 5. MATLAB Simulink block diagram and organization of the models. The diagram depicts a system where blue-colored lines signify oxygen as the primary gas, while red-colored
lines indicate hydrogen. In the Simulink model, individual signal lines are dedicated to hydrogen, oxygen, and water. However, for diagrammatic simplicity, these signals are
combined into a single line, with distinct colors denoting their respective primary gases. Each block represents a MATLAB model, with the added bullet points indicating the
primary methodologies used to code the specific unit operations.

∑
MATLAB system block methodology [38] is employed for the process 𝑚̇ in = 𝑚̇ H2 ,A,in + 𝑚̇ H2 ,B,in + 𝑚̇ O2 ,A,in + 𝑚̇ O2 ,B,in + 𝑚̇ w,A,in + 𝑚̇ w,B,in + 𝑚̇ mp,in .
model. Signals are sent to each block at every time step; the blocks then
(30)
perform calculations based on the signal inputs and estimate certain
outputs, which are subsequently sent as inputs to the next block. Each
input/output signal contains four major thermodynamic properties: The total outlet mass flow rate at the downer of the vessel is
mass flow rate, temperature, pressure, and density. These signals can expressed in Eq. (31), while the total outlet mass flow rate at the
represent hydrogen, oxygen, or water flow. upward of the vessel is presented in Eq. (32). These mass balances
are solved using separation efficiencies, where 𝜂s,l denotes the sepa-
3.7. Gas–water separation vessels ration efficiency of water, 𝜂s,H2 represents the separation efficiency of
hydrogen gas, and 𝜂s,O2 signifies the separation efficiency of oxygen
The PEM plant incorporates two vertical vessels dedicated to gas– gas. These separation efficiencies serve as model parameters, aiding in
water separation: one designed for hydrogen–water separation and the either fitting the model with experimental data of H2 inO2 and O2 inH2
other for oxygen–water separation. Each vessel is modeled individually. impurities or establishing assumptions regarding the amount of oxygen
Each model contains its specific mass and energy balances, along with and hydrogen dissolved in the water. Another approach to this is to
dimensioning parameters. model the dissolution of hydrogen and oxygen in water to determine
the amount of dissolved gas molecules leaving the vessel through the
3.7.1. Mass balance downer and upward outlets [17]. However, these detailed models are
The mass balance equation for the vertical gas–water separation beyond the scope of this study.
vessels is presented in Eq. (29). This equation states that the rate of ∑
𝑚̇ out,down = (𝑚̇ w,A,in + 𝑚̇ w,B,in + 𝑚̇ mp,in ) ⋅ 𝜂s,l
mass accumulation within the vessel equals the total mass flow rate at
the inlet minus that at the outlet. The outlet, in turn, is divided into + (𝑚̇ H2 ,A,in + 𝑚̇ H2 ,B,in ) ⋅ (1 − 𝜂s,H2 ) + (𝑚̇ O2 ,A,in + 𝑚̇ O2 ,B,in ) ⋅ (1 − 𝜂s,O2 ),
two ports: the downer vessel outlet and the upward vessel outlet. (31)
𝑑𝑚 ∑ ∑ ∑
= 𝑚̇ in − ( 𝑚̇ out,down + 𝑚̇ out,up ). (29) ∑
𝑑𝑡
𝑚̇ out,up = (𝑚̇ w,A,in + 𝑚̇ w,B,in + 𝑚̇ mp,in ) ⋅ (1 − 𝜂s,l )

The total mass flow rate at the inlet is shown in Eq. (30), where + (𝑚̇ H2 ,A,in + 𝑚̇ H2 ,B,in ) ⋅ 𝜂s,H2 + (𝑚̇ O2 ,A,in + 𝑚̇ O2 ,B,in ) ⋅ 𝜂s,O2 .
𝑚̇ H2 ,A,in is the inlet mass flow rate of hydrogen coming from stack (32)
A, 𝑚̇ H2 ,B,in is the inlet mass flow rate of hydrogen coming from stack
B, 𝑚̇ O2 ,A,in and 𝑚̇ O2 ,B,in stand for the inlet mass flow rates of oxygen
coming from stacks A and B, respectively, 𝑚̇ w,A,in and 𝑚̇ w,B,in are the 3.7.2. Vessel thermal model—Energy balance
inlet mass flow rates of water coming from stacks A and B, respectively, The lumped capacitance method is used to transiently estimate the
and finally, 𝑚̇ mp,in represents the inlet mass flow rate of makeup water. thermal behavior of the vessels, as shown in Eq. (33). In this equation,

8
G. Sakas et al. Applied Energy 382 (2025) 125199

the thermal capacitance of the vessel is expressed as 𝐶v . Additionally,

𝑇out = 𝑇 − 𝛥T2 , (43)
𝑄̇ in represents the total incoming heat resulting from the combined
flows of water, hydrogen, and oxygen at the inlet, while 𝑄̇ out signifies where 𝛥Tcd is the absolute temperature difference between the process
the total heat exiting the system with the outflowing water, hydrogen, temperature 𝑇 and the cooling water temperature 𝑇cw , 𝑆 is the specific
and oxygen. Sensible heat transfer equations are used to estimate both energy required to cool the process temperature, 𝛥T1 is the temperature
the total incoming and outflowing heat. Furthermore, 𝑄̇ amb represents difference between the process temperature and set-point temperature,
the heat dissipated to the ambient environment from the vessel and 𝑄sec is the amount of heat transferred from the primary leg to the
is calculated as free convective heat transfer of cylinder. Finally, 𝑄̇ mp secondary leg within the heat exchanger, and 𝑇out is temperature of
denotes the cooling of the vessel caused by the colder makeup water the primary liquid (hot side) at the outlet of the heat exchanger.
inlet flow.
d𝑇
𝐶v ⋅ = 𝑄̇ in − 𝑄̇ out − 𝑄̇ amb − 𝑄̇ mp . (33) 4. Results
d𝑡

As outlined in Section 1, one the primary objectives of this study

The lumped capacitance method operates under the assumption
is to develop a verified dynamic mass and energy balance model
of uniform temperature within the system. Consequently, the vessel
for a PEM plant. This model will serve as a foundation for future
temperature equals the temperature of the outflowing streams. Hence,
investigations, analyzing the potential of PEM plants connected to RES,
𝑄outlet is consistently zero because no higher temperature flow exits the
either as standalone units focused on green hydrogen production or
vessel.
as components within PtX systems. To verify the model, experiments
The thermal capacitance of the vessel is determined by considering
were conducted using a real PEM plant situated in Karlsruhe Institute
the total volume occupied by water within the vessel, and it can be
of Technology (KIT), Germany. Throughout these experiments, opera-
calculated as follows:
tional transient data were recorded, including stack power, total stack
𝐶v = 𝜌w ⋅ Cpw ⋅ 𝑉L , (34) voltage, total stack current supply, liquid temperatures at the anode and
cathode inlets and outlets of the stack, temperatures of individual cells,
where 𝜌w is the density of water, Cpw is the specific heat capacity of mass flow rates of all four pumps, hydrogen production rate, process
water, and 𝑉L is the total volume of the vessel that is occupied by water. pressures at six different locations, H2 inO2 impurities, and voltage
across each cell. The experimental data were sampled at one-second
3.8. Heat exchangers time steps.

The heat transfer between the hot and cold tubes was assumed to 4.1. Model verification
occur in parallel, and the heat exchanger was modeled as an ideal
component. In this setup, the heat transfer phenomena can be ex- The verification and comparison of the model was performed at
plained using the principles of the first law of thermodynamics [39]. three separate levels, namely electrochemical model, stack tempera-
Consequently, the energy balance and the rate of heat transfer can be ture, and hydrogen production. To model the electrochemical behavior,
characterized as: the real polarization curve was compared with the modeled one, as
𝑞 = 𝑚̇ h ⋅ Cpw ⋅ (𝑇h,in − 𝑇h,out ) = 𝑚̇ cd ⋅ Cpw ⋅ (𝑇cd,in − 𝑇cd,out ), (35) shown in Fig. 6. The polarization curve of the actual PEM plant was
obtained at three different temperatures: 52.2 ◦ C, 56 ◦ C, and 58.7 ◦ C.
where 𝑚̇ h represents the mass flow rate of the hot water from the The MATLAB fminsearch function was employed to fit the modeled
plant process circulation, and 𝑚̇ cd denotes the mass flow rate of the polarization curve. The polarization curve fitting process followed the
cold cooling water. 𝑇h,in , 𝑇h,out , 𝑇cd,in , and 𝑇cd,out correspond to the steps outlined by Ulleberg et al. [20]. At first, each individual measured
temperatures of the hot water at the inlet, hot water at the outlet, cold polarization curve underwent an initial fitting without considering tem-
water at the inlet, and cold water at the outlet, respectively. perature dependence. This involved fitting Eq. (4) and the first sections
of Eqs. (7) and (8) while estimating the model parameters 𝑟 and 𝑡 for
3.8.1. Closed-loop temperature control logic each case. Subsequently, linear and quadratic curve fitting was applied
During the simulation, the temperature of the stack is controlled by to parameters 𝑟 and 𝑡, respectively, to derive preliminary estimations
adjusting the process temperature using an analytical solution within of 𝑟i and 𝑡i , as demonstrated in the second parts of Eqs. (7) and (8).
the heat exchanger sub-model. The temperature set-point and the cool- Finally, these preliminary estimations were validated to ensure that
ing water temperature are used as initial model parameters. The an- they follow the temperature and pressure dependence behavior, and
alytical solution estimates the temperature of the liquid at the outlet an overall curve fitting of Eq. (2) was performed. The resulting values
of the heat exchanger based on the given temperature set-point and is of the electrochemical model parameters are presented in Table 3. For
determined as follows: model replication, note that the fitting was performed at the stack level,
and the use of the provided model parameter values estimates the stack
𝛥Tcd = |𝑇 − 𝑇cw |, (36)
voltage and overvoltages.
In addition to verifying the electrochemical model, the dynamic
𝑆 = 𝛥Tcd ⋅ Cpw , (37)
thermal behavior of the modeled stack was compared with that of the
real PEM plant, as shown in Fig. 7. The thermal model demonstrated
𝛥T1 = 𝑇 − 𝑇SP , (38) satisfactory performance in predicting the actual dynamic stack tem-
perature behavior, with a percentage deviation accuracy of 98.32%,
𝑄sec = 𝛥T1 ⋅ Cpw ⋅ 𝑚̇ h , (39) average relative error of 1.75%, and root mean square error (RMSE)
of 1.2075 ◦ C. These results allowed an analysis of the energy balance
𝑄sec within the stack, as illustrated in Fig. 8. The analysis was performed
𝑚̇ cd = , (40)
𝑆 for both stacks A and B, producing identical results because the stacks
are identical, and thus, only stack A is demonstrated and analyzed. The
𝑄sec,new = 𝑆 ⋅ 𝑚̇ cd , (41)
stack temperature was maintained constant at 60 ◦ C, and the simulation
𝑄sec,new was run until reaching a steady state for each current supply ranging
𝛥T2 = , (42) from 0 A to 300 A. The only heat source within the stack arises from
𝑚̇ h ⋅ Cpw

9
G. Sakas et al. Applied Energy 382 (2025) 125199

Table 3
Model parameters used in the model.
Sub-model Variable Value Unit Equation
Electrochemical 𝑟1 3.57 Ω cm2 Eq. (8)
Electrochemical 𝑟2 0.46 Ω cm2 ◦ C−1 Eq. (8)
Electrochemical 𝑡1 198.77 cm2 A−1 Eq. (7)
Electrochemical 𝑡2 −41319.46 cm2 ◦ C A−1 Eq. (7)
Electrochemical 𝑡3 2168180.07 cm2 ◦ C2 A−1 Eq. (7)
Electrochemical 𝑠 7.09 V Eq. (7)
Gas transfer 𝑝1 0.9889 – Eq.(24)
Gas transfer 𝑝2 −0.0013 – Eq.(24)
Vessel mass balance 𝜂s,l 1 – Eq. (31) and Eq. (32)
Vessel mass balance 𝜂s,O2 0.99 – Eq. (31) and Eq. (32)
Vessel mass balance 𝜂s,H2 0.99 – Eq. (31) and Eq. (32)

Fig. 7. Verification of the thermal model. The measured current supply, recorded at
Fig. 6. Comparison of the modeled and measured polarization curves. one-second intervals, was used as the input for the process model. The resulting stack
temperatures from the model are compared with the values measured from the PEM
plant. The stack temperature is recorded under the same current supply conditions
presented in Fig. 4.
the heat generated by overvoltages (𝑄̇ gen ), which increase gradually
with current supply up to 8.2 k W at nominal load. The lowest means
of cooling is the vaporization of water (𝑄̇ vap ), which increased with
current supply from 0.002 k W to 0.07 k W. The impact of vaporization
decreases with an increase in pressure; in particular, at pressures
above atmospheric pressure it becomes less significant. Heat loss to
the ambient environment (𝑄̇ amb ) was constant 0.5 k W across the whole
current supply range, ranking second in magnitude. Lastly, cooling by
circulating water increased with current supply and was estimated to
be 3.6 k W for the anodic side (𝑄̇ liq,an ) and 4.1 k W for the cathodic side
(𝑄̇ liq,cat ) at nominal load. The amount of heat transferred from the stack
to the cooling water is lower for the anodic side compared with the
cathodic side. This is due to the fact that the overall mass flow rate
of water entering the anodic side is lower compared with the cathodic
side. The pump mass flow rates for the case are shown in Table 4.
The final model verification focused on hydrogen production. As
shown in Fig. 9, the modeled hydrogen production closely matched the
measurements from the PEM plant. A minimal deviation was observed,
primarily due to the submodel of H2 inO2 impurities. As shown in
Fig. 4, the model tends to overestimate the amount of hydrogen gas Fig. 8. Energy balance analysis of stack A. Here, 𝑄̇ net represents the equilibrium of
lost as impurities to the oxygen line. Nevertheless, despite this slight the energy balance between the energy entering and leaving the system, which must
inaccuracy, the process model can predict the measured hydrogen always be zero.
production with a percentage deviation accuracy of 98.64%, average
relative error of 1.36%, and RMSE of 0.15 Nm3 ∕h.
parameters used for simulation and data retrieval are given in Table 4.
4.2. Analysis of stack and heat exchangers Fig. 10 shows the SEC of stack A. Both stacks were operated with
identical loads, and no significant differences were observed. Hence,
The SECs of stacks A and B were analyzed across various stack only stack A was analyzed for simplification.
temperatures within a current supply range. The results were obtained It is evident that as the stack temperature increases, the SEC de-
by running the process model until a steady state was reached for creases. This trend comes primarily from the decrease in activation
each current supply, ranging from 30 A to 300 A. The initial design overvoltage with a temperature rise due to favorable reaction kinetics,

10
G. Sakas et al. Applied Energy 382 (2025) 125199

Table 4
Initial design parameters used for the simulations.
Parameter Variable Value Unit
Pump 1A mass flow rate 𝑚̇ pp,1A 1163 k g h−1
Pump 2B mass flow rate 𝑚̇ pp,2B 1255 k g h−1
Pump 3A mass flow rate 𝑚̇ pp,3A 1329 k g h−1
Pump 4B mass flow rate 𝑚̇ pp,4B 1191 k g h−1
◦
Initial stack temperature 𝑇s,init 25 C
◦
Ambient temperature 𝑇amb 20 C
◦
Makeup water temperature 𝑇mp 25 C
Vessel diameter 𝐷vessel 0.1862 m
Vessel height 𝐻vessel 0.5585 m
Set point of heat exchangers 𝑇SP 60 ◦C

Cooling water temperature 𝑇cw 20 ◦C

Fig. 9. Comparison of the modeled hydrogen production with that measured from
the PEM plant. The hydrogen production is recorded under the same current supply
conditions presented in Fig. 4.

Fig. 11. Heat transfer to the secondary leg through the heat exchangers.

more heat extraction is required from the primary circulation as the

stack approaches its nominal load. It is also noteworthy that until 70 A
of current supply, no heat is transferred to the secondary leg. This is
due to the same reason as the upside-down U-shape observed in Fig. 10.
Fig. 10. Analysis of stack A SEC at various temperatures. At 50 ◦ C, 60 ◦ C, and 70 ◦ C, Essentially, the overvoltages are not sufficiently high to elevate the
and in lower current supply ranges, note that the stack temperature did not reach the temperature to the set point, which was 60 ◦ C for the simulation case.
set point temperature because of insufficient heat from the stack overvoltages. Furthermore, it is observed that, in general, a substantial amount
of heat is transferred during the heat exchange process. Specifically,
during nominal load, heat exchanger 1A transferred 3.3 k W of heat, heat
thereby enhancing the energy efficiency of the stack [40]. On the exchanger 2B 3.6 k W, heat exchanger 3A 4.1 k W, and heat exchanger 4B
other hand, the ohmic overvoltage increases with a temperature rise, 3.7 k W. Broadly, the amount of heat transferred by each heat exchanger
although with a lesser impact compared with the decrease in activation can be controlled by adjusting the pump mass flow rate. Currently,
overvoltage. As a result, the overall polarization curve decreases with for industrial-scale water electrolyzers, this waste heat is not generally
a temperature increase. This phenomenon is also illustrated in Fig. 6. utilized; however, it could serve as valuable heat for nearby residential
Fig. 10 also illustrates that at lower current densities and higher areas.
temperatures, the SEC shows an upside-down U-shape, indicating a
loss of energy efficiency during partial load operation. This occurs 4.3. Power distribution across the plant
because the stack overvoltages (heat source) are not sufficient to raise
the temperature at 50 ◦ C, 60 ◦ C, and 70 ◦ C, respectively, during a low The distribution of the power supplied to stacks A and B is shown
current supply. In such cases, reducing the mass flow rate of the in Fig. 12. In the diagram, yellow indicates electrical power, purple
incoming circulating water or installing external heating may enhance represents chemical power, and red denotes thermal power. It can be
the energy efficiency at partial loads. However, implementing such observed that out of the total of 52 kW supplied to the stack, 35.6 kW
an installation should be optimized and evaluated to determine its is converted into hydrogen, with only 0.04 kW lost to gas impurities,
economic viability. This outcome might be an important considera- leaving the remainder as end-product hydrogen, ready for storage and
tion during RES connected to PEM plants. In these cases, electrolyzer utilization. On the other hand, 16.4 kW of the total supplied power
plants might operate at partial loads for extended periods due to the is lost to overvoltages and is dissipated as heat. Heat exchangers 1A,
intermittency of renewable electricity production [41]. 2B, 3A, and 4B extract a total of 14.7 kW of heat, while 1.15 kW is
The heat exchangers were analyzed to determine the precise amount lost to the ambient environment from the stacks and the gas–liquid
of waste heat transferred from the primary circulation leg to the separation vessels. Additionally, 0.15 kW is lost due to the latent heat
secondary. As shown in Fig. 11, the heat transferred increases with of vaporization, and 0.4 kW is cooled through the introduction of cold
an increase in current owing to the rise in overvoltages. Consequently, makeup water.

11
G. Sakas et al. Applied Energy 382 (2025) 125199

Fig. 12. Sankey diagram illustrating the distribution of power supplied to stacks A and B across the plant.

In closer analysis and in combination with previous research [15], production between the real and modeled cases matched with the
several factors can influence the hydrogen production (end-line) and percentage deviation accuracy of 98.64%.
energy efficiency of a commercial or industrial-scale electrolyzer plant: The stack SEC analysis in steady-state conditions for various tem-
(1) stack overvoltages, (2) stack shunt currents, (3) hydrogen impurities peratures showed that a higher stack temperature can enhance the SEC
lost on the oxygen side, and (4) hydrogen burned in deoxidizer units. of the stack. Additionally, the study showed that within low current
Our study shows that, unlike alkaline water electrolyzers [15], PEM supply ranges the heat generated by overpotentials was not sufficient
electrolyzers do not exhibit any shunt current losses. Additionally, the to increase the temperature of the stack to the set point value. Thus, in
amount of hydrogen lost to the oxygen side and burned in deoxidizer partial load operation, the stack can lose efficiency because of colder
units is a very small fraction of the total generated. Therefore, as shown conditions.
in Fig. 12 and supported by previous studies, the energy efficiency The stack heat balance analysis in steady-state conditions showed
of the PEM electrolyzer stack and plant is significantly influenced by that the latent heat of vaporization and heat dissipation to the ambient
the overvoltage losses. Thus, to improve efficiency, the primary focus environment played a minor role in the cooling of the stack. On the
should be on reducing the polarization curve (UI) of the stack [42]. other hand, the circulating water to the anode and cathode carried
Fig. 12 examines the electrical power supplied to the stack and away most of the generated heat, which was primarily transferred to
its utilization. Specifically, it addresses how much of the supplied the secondary leg through the heat exchangers. In addition, a Sankey
current to the stack is lost to overvoltages, how much is converted diagram showed the distribution of power supplied to the stacks across
into hydrogen, and generally how the current is distributed, converted the plant. Out of the total of 52 kW supplied to stacks A and B, 35.6 kW
and utilized throughout the plant process. It should be noted that this became hydrogen gas, while the remaining 16.4 kW was dissipated as
analysis excludes the energy consumed to power up the four pumps and heat caused by overvoltages.
the water purification system. In conclusion, the paper presented a validated dynamic process
model to predict the electrochemical, thermal, and hydrogen pro-
5. Conclusions duction behavior of a 50 kW PEM plant, incorporating all necessary
peripheral components of the system. Future studies that integrate the
model with RES as a standalone unit or as part of a PtX scheme can
In this work, a parameter-adjustable dynamic mass and energy provide valuable insights.
balance process model of a 50 kW PEM electrolyzer plant was de-
veloped using MATLAB system blocks within the Simulink environ-
ment and validated with experimental data from an analogous PEM CRediT authorship contribution statement
plant. The process model comprises models for the stack, gas–liquid
separation vessel, and heat exchangers. The stack model integrates Georgios Sakas: Writing – review & editing, Writing – original
electrochemical, thermal, and mass transfer submodels, while the gas– draft, Visualization, Validation, Software, Resources, Project admin-
liquid separation vessel model incorporates mass balance and thermal istration, Methodology, Investigation, Formal analysis, Data curation,
submodels. The heat exchangers are modeled as ideal components, and Conceptualization. Philipp Rentschler: Writing – review & editing,
heat transfer phenomena are estimated using the principles of the first Resources, Formal analysis, Data curation, Conceptualization. Antti
law of thermodynamics. Kosonen: Writing – review & editing, Supervision, Project admin-
The verification and comparison was performed at three different istration, Funding acquisition, Conceptualization. Peter Holtappels:
levels, namely electrochemical, thermal, and hydrogen production. Writing – review & editing, Supervision, Funding acquisition, Formal
Fitting of the electrochemical model was used to match the real polar- analysis, Conceptualization. Vesa Ruuskanen: Writing – review & edit-
ization curve with the modeled one. The polarization curve fitting was ing, Supervision, Formal analysis, Conceptualization. Pertti Kauranen:
performed at various temperatures, and the resulting model parameters Writing – review & editing, Supervision, Formal analysis, Conceptual-
were provided. In addition, the thermal behavior of the modeled stack ization. Jero Ahola: Writing – review & editing, Supervision, Funding
showed excellent agreement with the real case, with a percentage acquisition, Formal analysis, Conceptualization. Roland Dittmeyer:
deviation accuracy of 98.32%. Similarly, the comparison of hydrogen Writing – review & editing, Supervision.

12
G. Sakas et al. Applied Energy 382 (2025) 125199

Declaration of Generative AI and AI-assisted technologies in the [13] Ibáñez-Rioja A, Sakas G, Järvinen L, Puranen P. In: Sguarezi Filho AJ, Jaco-
writing process mini RV, Capovilla CE, Casella IRS, editors. Smart grids—renewable energy,
power electronics, signal processing and communication systems applications.
Green energy and technology, Cham: Springer International Publishing; 2024, p.
During the preparation of this work the authors used Grammarly, 37–72. http://dx.doi.org/10.1007/978-3-031-37909-3_2.
QuillBot and ChatGPT in order to improve the grammar and readability [14] Huusari L. Analysis of phase separator design criteria using computational fluid
of the text. After using these tools/services, the authors reviewed and dynamics [Master’s thesis], Lappeenranta, Finland: LUT University; 2015.
edited the content as needed and take full responsibility for the content [15] Sakas G, Ibáñez-Rioja A, Ruuskanen V, Kosonen A, Ahola J, Bergmann O. Int J
Hydrog Energy 2022;47(7):4328–45. http://dx.doi.org/10.1016/j.ijhydene.2021.
of the publication. 11.126.
[16] Koponen J. Energy efficient hydrogen production by water electrolysis (Ph.D.
Declaration of competing interest thesis), Lappeenranta, Finland: LUT University; 2020.
[17] Sakas G, Ibáñez-Rioja A, Pöyhönen S, Järvinen L, Kosonen A, Ruuskanen V,
et al. Appl Energy 2024;359:122732. http://dx.doi.org/10.1016/j.apenergy.2024.
The authors declare that they have no known competing finan-
122732.
cial interests or personal relationships that could have appeared to [18] Hine F. Electrochemical engineering. John Wiley & Sons, Inc; 2018.
influence the work reported in this paper. [19] Hu S, Guo B, Ding S, Yang F, Dang J, Liu B, et al. Appl Energy 2022;327:120099.
http://dx.doi.org/10.1016/j.apenergy.2022.120099.
Acknowledgments [20] Ulleberg O. Int J Hydrog Energy 2003;28(1):21–33. http://dx.doi.org/10.1016/
S0360-3199(02)00033-2.
[21] García-Valverde R, Espinosa N, Urbina A. Int J Hydrog Energy 2012;37(2):1927–
Business Finland is acknowledged for the main financial support of 38. http://dx.doi.org/10.1016/j.ijhydene.2011.09.027.
the Bio-CCU project. Dr. Hanna Niemelä is also thanked for her assis- [22] LeRoy RL, Bowen CT, LeRoy DJ. J Electrochem Soc 1980;127(9):1954–62.
tance in language revision, and improving the grammar and readability http://dx.doi.org/10.1149/1.2130044.
[23] Gambou F, Guilbert D, Zasadzinski M, Rafaralahy H. Energies 2022;15(9):3452.
of the text.
http://dx.doi.org/10.3390/en15093452.
[24] Ahmadi P, Dincer I, Rosen MA. Int J Hydrog Energy 2013;38(4):1795–805.
Data availability http://dx.doi.org/10.1016/j.ijhydene.2012.11.025.
[25] Sun M, Zhang Y, Liu L, Nian X, Zhang H, Duan L. Appl Energy 2025;377:124636.
The data that has been used is confidential. http://dx.doi.org/10.1016/j.apenergy.2024.124636.
[26] Incropera FP, DeWitt DP. Fundamentals of heat and mass transfer. 2006.
[27] Tiktak WJ. Heat management of PEM electrolysis: A study on the potential of
excess heat from mediumto largescale PEM electrolyisis and the performance
References
analysis of a dedicated cooling system. TU Deft; 2019.
[28] Lagarias JC, Reeds JA, Wright MH, Wright PE. SIAM J Optim 1998;9(1):112–47.
[1] Green hydrogen for sustainable industrial development: A policy toolkit for de- http://dx.doi.org/10.1137/S1052623496303470.
veloping countries, Tech. rep., International Renewable Energy Agency (IRENA) [29] MATLAB fminsearch, https://se.mathworks.com/help/matlab/ref/fminsearch.
and the German Institute of Development and Sustainability (IDOS). html.
[2] Buttler A, Spliethoff H. Renew Sustain Energy Rev 2018;82:2440–54. http: [30] Mills A. In: Always learning, second ed.. Pearson; 2014.
//dx.doi.org/10.1016/j.rser.2017.09.003. [31] Dieguez P, Ursua A, Sanchis P, Sopena C, Guelbenzu E, Gandia L. Int J Hy-
[3] Olivier P, Bourasseau C, Bouamama PB. Renew Sustain Energy Rev drog Energy 2008;33(24):7338–54. http://dx.doi.org/10.1016/j.ijhydene.2008.
2017;78:280–300. http://dx.doi.org/10.1016/j.rser.2017.03.099. 09.051.
[4] Yigit T, Selamet OF. Int J Hydrog Energy 2016;41(32):13901–14. http://dx.doi. [32] Awasthi A, Scott K, Basu S. Int J Hydrog Energy 2011;36(22):14779–86. http:
org/10.1016/j.ijhydene.2016.06.022. //dx.doi.org/10.1016/j.ijhydene.2011.03.045.
[5] Espinosa-López M, Darras C, Poggi P, Glises R, Baucour P, Rakotondrainibe A, [33] Trinke P, Haug P, Brauns J, Bensmann B, Hanke-Rauschenbach R, Turek T. J
et al. Renew Energy 2018;119:160–73. http://dx.doi.org/10.1016/j.renene.2017. Electrochem Soc 2018;165(7):F502–13. http://dx.doi.org/10.1149/2.0541807jes.
11.081. [34] Lin R, Lu Y, Xu J, Huo J, Cai X. Appl Energy 2022;326:120011. http://dx.doi.
[6] Ni M, Leung MK, Leung DY. Energy Convers Manage 2008;49(10):2748–56. org/10.1016/j.apenergy.2022.120011.
http://dx.doi.org/10.1016/j.enconman.2008.03.018. [35] Marangio F, Santarelli M, Cali M. Int J Hydrog Energy 2009;34(3):1143–58.
[7] Ioroi T, Yasuda K, Siroma Z, Fujiwara N, Miyazaki Y. J Power Sources http://dx.doi.org/10.1016/j.ijhydene.2008.11.083.
2002;112(2):583–7. http://dx.doi.org/10.1016/S0378-7753(02)00466-4. [36] Kell GS. J Chem Eng Data 1975;20(1):97–105. http://dx.doi.org/10.1021/
[8] Hancke R, Bujlo P, Holm T, Ulleberg Ø. J Power Sources 2024;601:234271. je60064a005.
http://dx.doi.org/10.1016/j.jpowsour.2024.234271. [37] Haug P, Koj M, Turek T. Int J Hydrog Energy 2017;42(15):9406–18. http:
[9] Pfennig M, Schiffer B, Clees T. Int J Hydrog Energy 2024. http://dx.doi.org/10. //dx.doi.org/10.1016/j.ijhydene.2016.12.111.
1016/j.ijhydene.2024.04.335, S0360319924016471. [38] MATLAB system block - MATLAB & simulink - mathworks nordic, https://se.
[10] Solanki J, Wallmeier P, Böcker J, Averberg A, Fröhleke N. IET Power Electron mathworks.com/help/simulink/ug/what-is-matlab-system-block.html.
2015;8(6):1068–80. http://dx.doi.org/10.1049/iet-pel.2014.0533. [39] Vepsäläinen A, Pitkänen J, Hyppänen T. Fundamentals of heat transfer.
[11] Hysa G, Ruuskanen V, Kosonen A, Niemelä M, Aarniovuori L, Guilbert D, et al. Lappeenranta University of Technology; 2012.
J Power Sources 2023;574:233108. http://dx.doi.org/10.1016/j.jpowsour.2023. [40] Hourng LW, Tsai TT, Lin MY. IOP Conf Ser: Earth Environ Sci 2017;93:012035.
233108. http://dx.doi.org/10.1088/1755-1315/93/1/012035.
[12] Järvinen L, Puranen P, Kosonen A, Ruuskanen V, Ahola J, Kauranen P, et al. Int J [41] Sakas G, Ibáñez-Rioja A, Pöyhönen S, Kosonen A, Ruuskanen V, Kauranen P,
Hydrog Energy 2022;47(75):31985–2003. http://dx.doi.org/10.1016/j.ijhydene. et al. Renew Energy 2024;225:120266. http://dx.doi.org/10.1016/j.renene.2024.
2022.07.085. 120266.
[42] Carmo M, Fritz DL, Mergel J, Stolten D. Int J Hydrog Energy 2013;38(12):4901–
34. http://dx.doi.org/10.1016/j.ijhydene.2013.01.151.

13