You are on page 1of 16

Energy Conversion and Management xxx (xxxx) 116622

Contents lists available at ScienceDirect

Energy Conversion and Management


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

F
Experimental and theoretical evaluation of a 60 kW PEM electrolysis system
for flexible dynamic operation

OO
Elena Crespi a, Giulio Guandalini a, Luca Mastropasqua b, Stefano Campanari a, Jacob Brouwer b
a Group of Energy Conversion Systems, Department of Energy, Politecnico di Milano - Via Lambruschini 4A, Milan 20156, Italy
b National Fuel Cell Research Center, University of California, Irvine, CA 92697-3550, United States

PR
ARTICLE INFO ABSTRACT

Keywords: Electrolysis systems based on low temperature PEM cells have fast ramp rate and load-following capability that
PEM electrolysis system makes them suitable to provide services to the power grid, whose reliability is hindered by the increasing deploy-
Dynamic model ment of dynamic and non-controllable renewable energy sources. Flexible operation of the complete electrolysis
Dynamic data
unit is influenced by system design and auxiliaries: this work provides an insight on the dynamics of integrated
Flexibility
components, aiming at the enhancement of the flexible operation of an industrial-scale PEM electrolysis system.
Part-load operation
A numerical dynamic model of a complete electrolysis unit is implemented with the software Simulink, combin-
D
ing customized dynamic models of the main balance of plant components. Mass and energy balances are solved
during variable load operation, while PI–type and on–off controllers are included for system operation control.
The electrolysis stack performance depends on semi-empirical polarization curves, reflecting voltage dependance
on temperature, pressure, and current density. The solved transients of the device provide original indications for
TE

increasing the dynamic performance of the system.


A 60 kWe commercial unit is tested to validate the model, assessing its nominal and partial load performance.
The experimental data confirm the capability of the system to work at partial load without relevant deviations in
the stack temperature and pressure. Despite the lower specific consumption of the stack at partial load, the sys-
tem shows an increase in the average net specific consumptions when the load decreases toward its minimum
(from 67 kWhe/kgH2 above 1 A/cm2 to 140 kWhe/kgH2 at 0.3 A/cm2).
The simulation results and the experimental data are in good agreement, confirming the prediction capacities
EC

of the model with a good accuracy. Among the different investigated operation strategies, the strongest reduction
in average net specific consumption at partial load is given by the control of water flowrate to the stack (110
kWhe/kgH2 at 0.3 A/cm2) or of the hydrogen dryer regeneration (70 kWhe/kgH2 at 0.3 A/cm2). These are the sug-
gested points to be investigated for further improvements of fast-changing load systems.
RR

Nomenclature C heat capacity [kJ/K]


F faraday constant − 96,485 [C/mol]
Acronyms H enthalpy [kJ]
h specific enthalpy [kJ/kg]
PEM Proton Exchange Membrane or Polymer Elec- i current density [A/cm2]
trolyte Membrane m mass [kg]
CO

RES Renewable Energy Sources n number of electron moles per each hydrogen
RMSE Root Mean Square Error mole
PSA Pressure Swing Adsorption P power [kW]
p pressure [bar]
Symbols Q heat [kJ]
R molar gas constant – 8.314 [J/mol/K]
activity T temperature [K]
A area [cm2] V voltage [V]

https://doi.org/10.1016/j.enconman.2022.116622
Received 11 June 2022; Received in revised form 15 November 2022; Accepted 21 December 2022
0196-8904/© 20XX

Note: Low-resolution images were used to create this PDF. The original images will be used in the final composition.
E. Crespi et al. Energy Conversion and Management xxx (xxxx) 116622

Table 1 for instance, the possibility of grid congestions. For these reasons, con-
Proton onsite (now Nel ASA) C10 electrolyzer: nominal operating para- trollable (dispatchable) power generation and utilization devices able
meters [32]. to provide ancillary grid services are increasingly needed to guarantee a
Hydrogen reliable and resilient renewable grid [2,3].
At the same time, a higher penetration of RES for electric power
Net generation rate 10 Nm3/h (21.6 kg/24 h) generation coupled with electrification of both stationary and trans-
Delivery pressure 31 bar portation demands have become insufficient to meet the stringent
Rated specific power consumption 6.2 kWhe/Nm3H2 (68.9 kWhe/kgH2)
greenhouse gas reduction targets in many jurisdictions. Optimal solu-
Water tions to decarbonize sectors such as transport, building heating, and en-
DI water required 9 l/h ergy-intensive industry are still to be developed and, in many instances,

F
Circulation rate (at anode) 37.9 l/min
discovered.
Temperature 5 °C to 65 °C
Water resistivity > 1 MΩ·cm
In this framework, hydrogen has a key role to play because it can
help decarbonizing and ensuring reliability and resilience in the electric

OO
Electrical specifications
Breaker rating 480 VAC 3-phase, 100 kVA power sector and provide features, such as high temperature heat, re-
System power 60 kWAC ducing gas, or chemical feedstock in other hard-to-electrify sectors [4].
Electrolysis stack 140 V / 410 A On the one hand, the deployment of electrolysis systems enables in-
Cooling system creasing the penetration of renewable in the power sector, acting as
Chiller system Air-cooled chiller peak shaving and long-term storage [5]. The installation of an electroly-
Chiller capacity 55.7 kWth sis system in the same location as that of a renewable power plant al-
Max refrigeration 33.4 kWth lows stabilizing the electricity injection into the grid, absorbing gener-

PR
ated electricity to locally balance demand and supply. In this case, the
electrolysis system acts as a renewable energy storage system, convert-
Gibbs free energy variation ing electricity into hydrogen, that can be stored for a later use. Addi-
tionally, it is possible to use the flexibility of the electrolysis system to
provide ancillary services (e.g., voltage and frequency control or ramp-
ing services) to the grid, as discussed by the authors in previous work
1. Introduction [6]. On the other hand, renewable hydrogen can be used to decarbonize
the industrial sector, as well as the transport sector or the heating and
D
In recent years, most of the industrialized countries have set strin- cooling sector, following an integration and optimization strategy
gent targets for the reduction of greenhouse gas emissions [1], which named ‘sector coupling’ [7,8]. For example, ∼45 % of the hydrogen
are only achievable with a massive growth in the installation of renew- produced today is used in the oil refinery sector and ∼ 51 % takes part
in chemical production, mostly for the production of clean fuels and
TE

able energy sources (RES). These units are mainly photovoltaic plants
and wind farms, whose power generation is based on the availability of other products such as ammonia (representing ∼ 75 % of the H2 con-
sun irradiation and wind, that are intrinsically variable and sumption in the chemical sector) and methanol (∼25 %) [9].
non–controllable. When these units are connected to the power grid, Among the different electrolysis technologies available [10], this
the unpredictable fluctuations of their generated power must be coun- work focuses on industrial-size electrolysis systems based on low tem-
terbalanced by the other dispatchable power plants. Additionally, the perature Polymer Electrolyte Membrane (PEM) cells. These systems,
EC

modularity of these systems allows the diffusion of several small power thanks to their fast ramp rates and load–following capability, have been
plants, connected to the distribution grid. This changes the grid opera- proposed as flexible resources for the power grid [11,12]. For flexible
tion strategy, from a centralized generation scheme with one- application, PEM systems are preferred to high-temperature electrolyz-
directional energy flows (from the transmission to the distribution grid ers, that can reach higher electrical efficiency but have much longer
to the load) to a more complex operation with bi-directional energy start-up time (hours vs minutes). Additionally, PEM electrolyzers are
flows. This operation exacerbates grid management issues, increasing, preferred to low temperature Alkaline Electrolysers because of their
RR
CO

Fig. 1. Schematic of the electrolysis system layout.

2
E. Crespi et al. Energy Conversion and Management xxx (xxxx) 116622

Table 2 moisturizing process is instead required because the inlet water, fed to
Plant control system: controlled variables, manipulated variables and con- the anode, is only partially dissociated. A portion of the remaining wa-
troller type. ter flows out of the anode compartment with the generated oxygen,
Controlled variable Manipulated variable Controller while a portion is dragged through the membrane to the cathode by
type both diffusion and electro-osmotic drag [17].
The balance of plant always includes pumps for water circulation
Cathode backpressure (P2) H2 flowrate leaving the H2 PI
separator
(mainly to supply water to the stack), heat exchangers to control the
Anode backpressure (P1) O2 flowrate leaving the O2 PI
stack temperature, and liquid–vapor separators (to separate the gener-
separator ated H2 and O2 from water vapor). A deionization system is required to
Stack temperature (T2) Flowrate of coolant in feed water PI reach very high purity of the feed water (e.g., conductivity < 1 μS/cm)

F
cooler (F3) to ensure a proper non-conductive behavior inside the cell. Addition-
Water content in H2 separator Water purged from H2 separator ON/OFF ally, when the final applications require very high hydrogen purity
Water content in O2 separator Water refill to O2 separator ON/OFF (e.g., in fuel cell vehicles), a purification system is also required to re-

OO
Water content in deionized Water refill to deionized water ON/OFF move trace amounts of oxygen and moisture. Otherwise, the hydrogen
water tank tank purity downstream of the separator in a PEM electrolysis system is typi-
cally higher than 99.5 % (reaching also 99.9 % in some cases) [17]. The
faster response time (milliseconds vs seconds), possibility of reaching oxygen is removed through catalytic reduction, using a small amount of
lower minimum loads, and ability to electrochemically pressurize the hydrogen, while the moisture is generally removed through Pressure
hydrogen. A promising solution is given by Anion Exchange Membrane Swing Adsorption (PSA) units, using activated carbon filters or silica
(AEM) cells, which is currently characterized by a lower readiness level gel.

PR
for large scale commercialization [13].
The main element of an electrolysis system is the electrolysis cell 2. Review of existing models
stack. The PEM electrolysis stack adopts a solid electrolyte, consisting
of a polymeric membrane, usually Nafion® or other perfluorosulphonic According to recent reviews of PEM electrolysis models [18,19],
acid polymer. The membrane separates the electrodes and allows the PEM electrolyzers were commercialized in the early 2000 s, but model-
transfer of protons (H+ ions) from the anode to the cathode. The cur- ling of the electrolyzer dynamic behavior gained importance only be-
rently available membranes are thin, flexible, and impermeable, result- ginning in 2009, given the increasing use of these systems with inter-
ing in low resistance cells and compact systems that can work at mittent power sources.
D
medium–high absolute pressure and at high differential pressure. The Among the existing models, most focus on a single cell or a short-
main drawback of this solution is the need of precious metal electrocat- stack - connecting few cells together – to study the cell performance and
alysts, such as platinum, Pt and/or iridium, Ir, for activating semi- how it is influenced by different operating and design parameters. Ad-
reactions on the electrodes. In commercial units, adopting the so–called ditionally, some models have been realized to understand the phenom-
TE

differential pressure technology, the anode is fed with water at ambient ena happening in specific components of the cells, or to study cell
pressure or a slightly higher pressure, while the cathode side produces degradation, for diagnostic purposes or for control purposes.
hydrogen that is pressurized up to 25–35 bar (operation at higher pres- The core of the cell (or stack) model is always the electrochemical
sure is also being tested [14,15]). This solution allows limiting the out- model, relating the input electrical power to the output hydrogen flow.
put hydrogen contamination with oxygen, reducing the minimum load, The electrochemical models are mostly steady-state models, since the
and limiting the need of the oxygen removal processes [16]. A de-
EC

electrochemical response of the cell is very fast compared to the time


RR
CO

Fig. 2. Experimental data from C10 operation over 37 h with different load and operating conditions: stack current and voltage (above), stack temperature and cath-
ode and anode backpressure (below).

3
E. Crespi et al. Energy Conversion and Management xxx (xxxx) 116622

cluding not only the electrolyzer stack but also the water pump, the
cooling fan, the hydrogen storage tank, the water tank, the power sup-
ply, and the electronic control unit. However, only the cell performance
curves are compared to experimental data for validation. In [30], a dy-
namic model of a PEM electrolysis system is realized with a Bond Graph
approach, including models of converter, separator vessels, recircula-
tion circuits, a dryer and valves. The model is validated and is able to
describe accurately the behavior of a semi-industrial 25 kWe PEM elec-
trolyzer. Also in [31], a model of PEM electrolyzer based on the Bound
Graph approach is developed, including sub models of stack, converter,

F
gas–water separators, cooling and recirculation circuits, hydrogen
dryer, and venting system. The values of the system-specific parameters
are identified through comparison with the experimental data from a

OO
laboratory scale electrolyzer. No complete model of large integrated
PEM electrolysis systems, including all BoP and sufficiently representa-
tive of an industrial scale has been found in the literature to-date.
With the goal of assessing and improving the dynamic behavior of
PEM electrolysis systems, this work focuses on the development and
validation of a dynamic physical simulation model for complete PEM
electrolysis system. Compared to available literature models, the model

PR
developed in this study includes the entire system in the dynamic analy-
sis, considering large scale completely integrated systems (i.e., at com-
mercial level). The reference unit in this work is a commercial system of
industrial size (60 kW or higher, thanks to modularity), but the present
work can be applied to other systems and to scale-up processes. In prac-
tice, the capability of analyzing the entire system, including dynamic
models of physical phenomena for the stack and all the balance of plant
components, allows for discussing the required improvements for MW-
D
scale units, that are now required to meet the ambitious ‘green’ hydro-
gen production targets. The validation with experimental data collected
Fig. 3. Experimental data from C10 operation: stack current–voltage polariza- from testing of the commercial unit represents a further added value of
tion curve as a function of temperature at 30 bar cathode backpressure (above) this work.
TE

and as a function of pressure at 50 °C (below).


2.1. Outline and scope
response of the entire system. In this case, a mathematical description
of the current–voltage cell characteristic is typically considered, Through experimental testing and dynamic numerical modelling,
through polarization curves derived with analytical (e.g., [20]–[24]), this work evaluates the flexibility of an industrial-size PEM electrolysis
EC

semi-empirical or empirical approaches (e.g., [25,26]). Models includ- system. Additionally, it develops and applies a dynamic physical model
ing the electrochemical dynamics (e.g., [27,28]), related to internal to understand and improve the contribution that these systems can give
storage phenomena such as double layer capacitance, are used to study to grid stability. The study is based on a commercial 60 kWe unit, the
the coupling of the electrolysis stacks with converters. They describe Proton Onsite (now Nel ASA) C10 electrolyzer, featuring a state-of-the-
the cells through an equivalent electrical circuit including resistive, ca- art PEM stack and system layout, which is applied also to larger sizes
pacitive, and eventually inertial elements. thanks to modularity. The goal is to identify, through simulations, the
Some cell (or stack) models, in order to improve the accuracy of the main factors limiting the system efficiency and to improve the control
RR

cell efficiency calculation, include dynamic thermal models and fluidic strategies of the commercial electrolyzer. The improved control strate-
and mass transfer models (e.g., [20,21,23,24]). The thermal models gies, applied to the real system, would allow decreasing the specific sys-
generally consider the stack as a lumped thermal capacitance, to model tem consumption, thus decreasing the cost of operating the system and
the effect of the stack operating point on the stack temperature and effi- increasing its economic competitiveness. The model is also able to sup-
ciency. Only a few models are spatially resolved distributed parameter port the study of larger systems.
models, describing the temperature and species concentration distribu- A detailed model of the integrated unit, including the BoP compo-
CO

tions in the cell. Fluidic and mass transfer models compute partial pres- nents and the control system, is introduced (see section 3). The refer-
sures and species concentrations. They generally include water perme- ence commercial unit is then operated to collect experimental data dur-
ation through the membrane, while only a few models include gas cross ing steady-state and dynamic operation following different load pro-
permeation. files. The hydrogen production setpoint is changed over time, and time
Although many stack models are lumped parameter models, suit- profiles of different physical values (temperature, pressure and flowrate
able for system modelling, very few models consider the complete Bal- in different points, as well as stack and auxiliaries power consumption)
ance of Plant (BoP). Some models include the electrical consumption of are observed and analyzed (see section 4.1). The model is tested
the converter, or a part of the thermal architecture necessary to better through the comparison of experimental and simulated values of these
describe the stack and system thermal behavior, but models including physical quantities, when the same hydrogen generation setpoint and
all the components for thermal management (e.g., pumps for coolant initial conditions are considered (see section 4.2). The validated model
circulation, fans, chillers), the water supply system, the gas–water sepa- is finally applied to the simulation of different operation strategies,
rators, and systems for gas purity control and regulations (e.g., dryers, aiming to improve the plant operation at partial load, thus improving
condensers) are very rare. The ones that include a significant fraction of the system flexibility with a view to coupling such systems with renew-
the BoP components with detailed models are described in [29–31]. In able power plants or providing ancillary services to high renewable
[29] a model of a high pressure PEM electrolyisis system is realized, in- penetration electric grid networks (see section 4.3).

4
E. Crespi et al. Energy Conversion and Management xxx (xxxx) 116622

F
OO
PR
D
TE
EC

Fig. 4. Experimental data from C10 system warm-up and operation at variable load (ranging from ∼ 15 % to 100 % of nominal load) over time: (a) hydrogen mass
flow rate at stack outlet and system outlet stack, and stack current, (b) pressure in oxygen and hydrogen separators, (c) stack temperature, water temperature at
stack inlet, coolant temperature and opening angle of the valve to control the coolant flow rate, (d) gross electric power consumption and net power consumption
excluding and including the chiller consumption.
RR

2.2. Experimental setup

The electrolysis system analyzed in this work is based on the design


of a commercial unit, the Proton Onsite electrolyzer model C10 [32],
available for testing at the University of California, Irvine’s National
Fuel Cell Research Center laboratories. The system is a differential pres-
CO

sure PEM electrolyzer, whose stack is comprised of 65 cells, each with


an active area equal to 214 cm2. It has nominal net power demand of 62
kWe and a stack nominal current equal to 410 A. The main characteris-
tics of the unit are reported in Table 1, while the system layout and the
available measurement probes are shown in Fig. 1.
The electrolyzer stack is supplied with water at the anode side. Oxy-
gen/water mixture and hydrogen/water mixture leave the anode and
cathode compartments, respectively, and are collected in dedicated sep-
arators where the gas separates from the liquid water by condensation
Fig. 5. Experimental data from C10 system operation: cathode backpressure, during cooling and by gravity. Indeed, since an excess of water is sup-
stack current density, on/off state of water purge from the hydrogen separator plied to guarantee an adequate control of the cell temperature and hu-
and on/off state for the pressurization of the PSA bed. midity, a consistent amount of water leaves the cells with the gas. The
system is designed to provide the hydrogen flow rate required by the fi-
nal user, hence the power supplied to the stack is regulated to keep the
pressure in the separators constant. In practice, the pressure influences

5
E. Crespi et al. Energy Conversion and Management xxx (xxxx) 116622

process lasting 17 min, and vented. Then, the venting valve is closed,
allowing to the hydrogen drawn from the operating column to pressur-
ize the regenerated column. After 1 min the two columns are switched
and the process is repeated.
While the commercial unit is designed for steady-state operation at
nominal power, before testing the system, a pressure regulation valve
(‘H2 control valve’ in Fig. 1) is installed at the outlet of the PSA in order
to allow partial load operation. Decreasing the opening area of the
valve, the flow rate of hydrogen leaving the system is limited and, as
discussed before, the electrolyzer control system acts on the power sup-

F
plied to the stack in order to keep the pressure in the H2-H2O separator
constant.

OO
3. Methods for dynamic modelling

This section describes the dynamic model of an industrial-scale elec-


trolysis system, realized with the software Simulink and reflecting the
layout and operation mode of the Proton Onsite (now Nel ASA) C10
commercial system. The model is general and can be applied to any unit
of the same technology (i.e., polymer electrolyte membrane elec-

PR
trolyzer – PEMEL). For each plant component, dynamic mass and en-
ergy balances are solved to simulate variable load operation, as detailed
below. Then, the single component models are combined to build the
Fig. 6. Experimental data from C10 warm up operating conditions at partial
integrated system model, implementing PI–type and on–off type con-
load: (a) stack current over time, (b) Stack temperature over time.
trollers. Internal Simulink ODE solver is used (ode15s stiff/NDF), with a
variable step in order to take into account the different behavior during
the withdrawal of gaseous oxygen or hydrogen, both saturated with wa- fast transients and steady operation of the system.
ter, by means of a valve on outlet piping. While the pressure in the oxy- Regarding the thermodynamic properties, ideal liquid, ideal pure
D
gen separator (P1) is atmospheric, according to differential pressure gas and ideal gas mixtures are considered. Enthalpy is computed with
technology, the hydrogen separator pressure (P2) is maintained at the polynomial equations from NASA [33] (Eq. (1)), where - are nu-
higher values, up to 30 bar (nominal pressure), thus minimizing oxygen merical coefficients available on NASA thermodynamic libraries for
contamination of the hydrogen stream. 1130 solid, liquid and gaseous pure chemical species.
TE

The DI water that accumulates in the oxygen separator is used to


supply the stack, after being cooled in a dedicated heat exchanger
(‘feedwater cooler’ in Fig. 1). The water temperature at the stack inlet
(1)
(T3) is regulated to control the electrolysis stack temperature (approxi-
mated by the temperature of the water in the O2/H2O separator, T2).
The regulation is performed by varying the flow rate of coolant sent to
EC

the heat exchanger (F3). To avoid running out of water in the oxygen When a gaseous mixture of different species is present (such as moist
separator (only a fraction of the water sent to the stack anode returns to air and moist hydrogen), the enthalpy of the mixture is computed from
the separator, due to water consumption for hydrogen generation and the enthalpy of each species in the mixture ( ) and their molar fraction
water migration to the cathode), a high-purity water refill is present. ( ), assuming ideal gas mixture behavior (Eq. (2)). For liquid water,
This water make-up is taken from a buffer tank (‘deionized water tank’ the density is assumed to be constant and enthalpy variation with pres-
shown in Fig. 1) and filtered in a demineralization system. The buffer sure is taken into account as follows.
RR

tank receives the water recovered from the hydrogen separator, whose
pressure is previously decreased to allow separation of the dissolved hy- (2)
drogen, and fresh deionized water. Keeping a high purity in the reacting
water loop (e.g., conductivity < 1 μS/cm) is required to ensure the reli-
Water in liquid phase and in vapor phase are treated as two different
ability and lifetime of the electrolyzer. The water levels in the separa-
species, and properties are selected according to the partial pressure in
tors as well as in the deionizer buffer tank are also managed by the con-
each specific point. The maximum molar fraction of water in vapor
CO

trol system, including on–off controls for the water purge in the hydro-
phase coincides with the water molar fraction at saturation, which is a
gen separator and for the water refill elsewhere.
function of the mixture temperature and pressure. Saturation pressure
The oxygen leaving the separator is vented to the atmosphere with-
of water is computed as a function of the temperature, according to a
out any further treatment, unless additional systems for oxygen recov-
polynomial equation (Eqs. (3), (4)) derived in [34] and valid for tem-
ery are foreseen. Conversely, the hydrogen leaving the separator is
perature values ranging from water triple point ( = 273.16 K) to the
cooled in a dedicated heat exchanger, decreasing the hydrogen dew
critical point ( 647.096 K, = 220.64 bar), suitable for this ap-
point as much as possible. The condensed water is separated, and the
plication.
hydrogen enters a Pressure Swing Adsorption (PSA) unit, where it is ad-
ditionally dried and purified, removing also the small traces of oxygen
that permeated to the hydrogen side. The PSA is comprised of two par-
(3)
allel columns, operating alternatively in drying mode and regenera-
tion/pressurization mode. In the column operating in drying mode, the
water is absorbed in a desiccant drying bed. The two columns are con- (4)
nected by means of an orifice through which a fraction of the dry hydro-
gen leaving the column in operation is driven to the other column. This
constant flow is used to regenerate the second column in a regeneration

6
E. Crespi et al. Energy Conversion and Management xxx (xxxx) 116622

F
OO
Fig. 7. Experimental data from C10 system operation at variable load (ranging from ∼ 15 % to 100 % of nominal load) as a function of the stack current density:
(a) gross power, net power excluding the chiller and net power including the chiller, (b) gross (stack outlet) and net (system outlet) hydrogen flow rate, (c) gross

PR
specific consumption, (d) net specific consumption excluding the chiller.

3.1. PEM electrolyzer stack 36] through the electro-osmotic drag coefficient , which depends
upon the degree of hydration of the membrane (λ = 22) [37], as in Eq.
The electrolyzer stack is comprised of 65 cells electrically connected (7).
in series to increase the voltage. Thus, knowing the number of cells
( ), the stack voltage ( ) can be computed from the cell voltage (7)
( ), as in Eq.(5), while the current is the same in all the cells.
D
(5) The diffusion driven transport is proportional to the difference in
concentration at the two sides of the membrane, and is modelled as in
All the cells are considered identical, and the reactant and coolant [23]. The concentrations on the two sides of the membrane are ex-
TE

flows are supposed to be equally distributed among the cells. The model pressed according to the Fick’s law of diffusion [23], where the O2-H2O
of a single representative unit cell is realized with a lumped-volume ap- and H2-H2O effective binary diffusion coefficients are computed
proach, allowing the simulation of larger full-scale stack effects, as re- through Eq. (8) and Eq. (9), respectively. The porosity coefficient is
quired by system simulation, without the need for simulation stack de- assumed equal to 0.3, the percolation threshold is equal to 0.11, the
sign, flow manifolds, mass flow, temperature or pressure differences empirical coefficient is 0.785 and the binary diffusion coefficient of a
amongst the cells, while including sufficient attention to the perfor- mixture of two substances A and B is estimated with Eq. (10) (where the
EC

mance details of a representative unit cell. The model receives as input: dimensionless coefficients a and b are equal to 3.64 · 10-4 and 2.334, re-
flow rate, temperature, pressure, and composition of the inlet water spectively [23,36]).
(supplied to the cell anode compartments), the electric power supplied
to the stack and anodic and cathodic backpressure. It computes flow
(8)
rate, composition and temperature of the oxygen-water and hydrogen-
water mixtures leaving the stack, as well as the stack operating temper-
RR

ature. Side effects due to stack layout and connections are calibrated on (9)
system data, if required.
The overall cell model includes three sub-models: fluid dynamic
model, thermodynamic model, and electrochemical model. (10)
The fluid dynamic model subsystem solves lumped mass conserva-
tion equations over the electrolyzer stack for each chemical species
flowing through the anodic and the cathodic channels, to determine
CO

The pressure driven transport is proportional to the difference in


flow rate and composition of the anodic and cathodic streams leaving pressure on the two sides of the membrane, as in Eq. (11), where
the stacks. Hydrogen production at cathode, oxygen production at an- (membrane permeability to water) is assumed equal to 1.58·10-18 m2,
ode and water consumption at anode are computed from the stack cur- and (viscosity of water) is equal to 1.1·10-2 g/cm/s [23,35].
rent density (Eq. (6)).

(11)

(6) Since the volume of the cell channels is negligible with respect to
the volume of the other system components (i.e., O2-H2O and H2-H2O
separation tanks), the effects of gas accumulation on the cell channels
are neglected. Additionally, pressure drops in the cell channels are as-
Water net transport from the anode to the cathode is included as the
sumed negligible.
sum of three different transport mechanisms: electro-osmotic drag, dif-
The thermodynamic model subsystem determines the temperature
fusion driven transport and pressure driven transport. The elec-
of the cell and the temperature of the anodic and cathodic streams leav-
tro–osmotic drag is proportional to the transported hydrogen ions [35,

7
E. Crespi et al. Energy Conversion and Management xxx (xxxx) 116622

Table 3 metric [40]–[42], and therefore that the charge transfer coefficient
Stack polarization curves model parameters. is equal to 0.5.
Parameter Ref

Anodic exchange current density 1.7 · A/cm2 Calibration


@55 °C 10-6 (16)
Pre-exponential factor for 2.16 · A/cm2 Calibration
(with Tref = 55 °C) 106

Activation energy for anode 76,000 J/mol [16]


Additionally, the model neglects the cathode activation overvoltage
Electrode (anode/cathode) 1 mm [44]

F
because the hydrogen evolution reaction at the cathode is significantly
thickness
Electrode (anode/cathode) 7.5 mΩ·cm [44] faster than the kinetics of the oxygen evolution reaction at the anode
resistivity [22,43]. Regarding the exchange current density ( ), the values re-

OO
Equivalent Membrane thickness 322 μm Calibration ported in the literature for the anode side varies over several orders or
Membrane humidification level λ 22 molH2O/molSO3 [35,39] magnitude [16], ranging from 10-13 A/cm2 to 10-3 A/cm. The model in-
Anode charge transfer coefficient 0.5 – [40–42] cludes the exchange current density dependance upon cell temperature
Limiting current density 6 A/cm2 [46]
with Eq. (17) [42], where the activation energy for the anode (
Faraday’s efficiency 99 % [47]
is equal to 76,000 J/mol and the pre-exponential factor ( ) is cali-
brated on the experimental data (see Section ‘Model validation’).
ing the electrolyzer stack, solving the dynamic overall energy balance
over the stack (Eq. (12)).

PR
(17)

(12)
The ohmic losses are modelled through the ohmic overpotential
( ) as in Eq. (18), where the electric resistance of the current col-
Temperature dynamics are taken into account through the heat ca-
lector and the electrodes to the flow of electrons is expressed by
pacity of the stack ( ), that is lumped in the cell and stack compo-
Eq. (19) and the resistance of the membrane to the flow of protons
nents (mainly comprising the mass of the bipolar plates). Constant heat
( ) is expressed by Eq. (20).
transfer coefficients between the gas–water streams and the stack are
D
considered and convective heat losses to the environment ( ) are im- (18)
plemented by means of specific correlations for heat transfer coefficient
in natural convection [38]. Additionally, it is assumed that the temper- (19)
ature of the outlet streams is equal to the temperature of the stack (ther-
TE

mal equilibrium between the flowing fluid and the stack constituent). (20)
The electrochemical model computes the cell performance through
semi-empirical polarization curves, reflecting voltage dependance upon The model assumes titanium-supported Pt and Ir electrocatalysts for
temperature, pressure and current density [22,39]. The implemented the cathode and the anode, respectively. Indeed, titanium is currently
equation (Eq.(13)) computes the cell voltage as the sum of the ideal the most effective solution for PEM electrolyzer electrode supports
voltage ( ) and the overpotentials ( :
EC

since it is a good electrical conductor, it provides mechanical support


and is chemically stable in acidic environments. The electrode thickness
(13) ( ) is assumed equal to 1 mm and the electrode resistivity ( )
equals 7.5 mΩ·cm (intermediate values from the literature, [44]). The
proton conductivity of the membrane is computed with Eq. (21),
The ideal voltage or open circuit voltage, that is the minimum cell where the value of λ, representing the number of water molecules per
voltage required for the reaction to take place in ideal conditions, is de- sulfonic acid site, SO3H-, is assumed equal to 22 because the membrane
RR

fined by the Nernst equation (Eq. (14)). is exposed to liquid water [35,39]. The thickness of the membrane
( ), which a sensitivity analysis shows as the main term influencing
the ohmic loss and therefore the cell polarization curve, is calibrated
based on experimental data (see Section ‘Model validation’).
(14)
CO

(21)

The concentration losses are modelled through the concentration


For hydrogen production operating conditions, voltage losses (over-
overpotential ( ), considering only the anode side, where the dom-
potentials) affect the cell performance due to sluggish electrochemical
inant contribution is present [22,45] (Eq. (22)).
kinetics of charge transport processes (activation losses), ohmic losses,
mass transport (concentration losses), and parasitic losses.
Activation losses ( ) are modelled via Eq. (15). (22)

(15) The limiting current density is assumed equal to 6 A/cm2 [46].


The parasitic losses are modelled through the Faradaic efficiency ex-
pression [47], by Eq. (23). A constant value for the Faradaic efficiency
The equation is a simplification of the Butler—Volmer equation, ex-
equal to 99 % is assumed. This choice is supported by the low pressure
pressed by Eq. (16), where the subscript X is either anode or cathode.
(maximum pressure expected is 30 bar) and by the high loss of hydro-
The simplified equation is obtained assuming that the processes in-
volved in electron transfer at the electrode–electrolyte interface is sym-

8
E. Crespi et al. Energy Conversion and Management xxx (xxxx) 116622

Regarding the hydrogen-water separator model, the volume of the


separator lumps the volume of the separator itself and the volume of the
hydrogen dryer components (approximated by the volume of a PSA col-
umn). Indeed, a stand-alone dynamic model for the PSA column is not
realized.

3.3. Deionized water tank

The deionized water tank model receives as input temperature, pres-


sure and flow rate of the inlet water flows as well as the required flow

F
rate of the outlet flows, resulting from the other components balances.
Dynamic mass and energy balances are hence solved to compute the
water content and temperature in the tank. The model assumes a uni-

OO
form temperature in the tank (perfect mixing) and convective heat
losses to the environment. The heat accumulation term lumps the heat
capacity of the accumulated water, while the heat capacity of the tank
surfaces is neglected.

3.4. Hydrogen dryer

PR
Simplified models of the hydrogen cooler and water separator, and
of the PSA are used to simulate the hydrogen drying process, while their
Fig. 8. Comparison of experimental and simulated cell polarization curves for dynamic behavior (i.e., mass accumulation) is lumped into the separa-
different stack temperature conditions (above) and error on the cell voltage (be- tor model.
low). A stationary model of the hydrogen cooler computes the amount of
heat removed to cool down the humid hydrogen flow to a given temper-
gen in the PSA, that results much higher than the hydrogen loss due to ature and the amount of water that condenses due to the lowering of
short-circuit current and crossover. temperature. The water separator model computes the flow rate and
D
composition of the hydrogen leaving the component, assuming that all
the condensed water is properly separated.
(23)
The PSA column model computes the amount of water that is re-
TE

moved from the hydrogen flow to reach the 99.9995 % purity target,
3.2. Oxygen-water and hydrogen-water separators and the flow rate of dry hydrogen leaving the column. The loss of dry
hydrogen through the orifice (or the valve) connecting the two columns
The separators are tanks where the O2-H2O (or H2-H2O) stream from is included in the model through a simplified model of the orifice ex-
the electrolyzer anode (or cathode) outlet accumulates and cools to al- pressed by Eq. (26) [48], where: is the mass flow rate of hydro-
low liquid water to condense and separate from the oxygen (or hydro- gen lost through the orifice, is the orifice coefficient, is the
EC

gen) by gravity. The gas (oxygen or hydrogen saturated with water) is gravity acceleration, is the density of hydrogen upstream the valve
then removed from the top of the tank, while the liquid water is with- and is the pressure difference across the valve.
drawn from the bottom. An additional inlet stream is present in the O2-
H2O separator, to refill the separator with fresh water, while the H2- (26)
H2O separator receives part of the water coming from the hydrogen
dryer. The pressure difference is given by the difference between the
pressure in the column in operation, assumed equal to the pressure in
RR

The separator model input variables are flow rate, temperature,


pressure, and composition of the inlet flows. The model implements dy- the H2-H2O separator (negligible pressure drops in the hydrogen cooler
namic mass and energy balances (Eqs. (24)–(25)) to compute the and in the operating PSA column are assumed), and the pressure in the
amount of liquid and vapor in the tank and their temperature, pressure, other column, whose value depends on whether the column is in regen-
and composition, as well as flow rate, temperature, pressure and com- eration or pressurization mode. When the column is in regeneration
position for the outlet flows. Mass balances are solved assuming that va- mode, its pressure is assumed equal to the atmospheric pressure (negli-
por–liquid equilibrium is always satisfied in the tank: the vapor phase is gible pressure drops in the regenerating PSA column are assumed). In
CO

oxygen (hydrogen) with water in saturated conditions, calculated as a this case, the value of the term assumed constant in the operat-
function of the temperature and the pressure in the separator. It is then ing range of interest, is computed from the mass flow rate of hydrogen
assumed, and verified a posteriori, that the solubility of oxygen (hydro- passing through the orifice at nominal conditions. An intermediate
gen) in the liquid is negligible, leaving the liquid phase as pure water. phase (pressurization mode) is designed to increase pressure over time
Energy balances are solved to determine the separator temperature, as- from ambient pressure (regeneration mode) to the nominal pressure
suming convective heat losses into the environment and a uniform tem- (operation mode), according to Eq (27). Since in the real system an ad-
perature in the tank (perfect mixing). The heat accumulation term takes ditional valve connecting the two columns is opened, for the value of
into account the heat capacity of the accumulated water only, neglect- is calibrated such that the column reaches the nominal pressure
ing the heat capacity of the separator walls. in 1 min at nominal operating conditions (as from a calibration on the
real system).
(24)
(27)

(25)

9
E. Crespi et al. Energy Conversion and Management xxx (xxxx) 116622

F
OO
PR
D
TE
EC
RR
CO

Fig. 9. Comparison of experimental data and simulation results for system warm-up and variable load operation: (a) current setpoint, and cell voltage, (b) stack
temperature (c) temperature of water at stack inlet, (d) pressure in hydrogen separator, (e) pressure in oxygen separator and (f) hydrogen flow rate at plant outlet.

3.5. Feed water cooler uniform and considers constant values for the heat capacities as well as
for the heat transfer coefficients. It neglects heat transfer by conduction
The feed water cooler is a counter-current plate-type heat ex- along the flow direction (due to the relatively small thermal gradients
changer. The unit is modelled as a sequence of identical sub-units, and the necessity to avoid more complex iterations, which would im-
given the modular structure of the plates and assuming that the fluids pact the simulation speed), the thermal resistance of the plate and heat
equally distribute among the channels. Each sub-unit includes a single losses to the environment. Temperature dynamics are related to the
plate and half of the adjacent cold and hot channels with control vol- heat capacity of the heat exchanger materials, while fluid mass accumu-
umes identified through a discretization along the direction of the chan- lation in the channels is neglected. Pressure is assumed to vary linearly
nel (1D-model). For each control volume, the model solves mass and with the volumetric flow rates.
energy balances to compute the heat transferred, the plate temperature
distribution and the temperature of the outlet stream. The model as-
sumes that, in each control volume, the temperature of each plate is

10
E. Crespi et al. Energy Conversion and Management xxx (xxxx) 116622

eter and volumetric flow rate (Eq. (30)). Incompressible fluids are as-
sumed, since air accumulation is concentrated in the air supply mani-
fold volume, while hydrogen accumulation is concentrated in the hy-
drogen dryer volume.

(28)
(29)

(30)

F
3.8. Control system

OO
The control system implemented in the dynamic model includes
both PI controllers and on–off controllers. PI controllers are based on a
proportional gain ( ) and an integral gain ( ), as in Eq. (31), and are
applied in the system when it is required an accurate tracking of a set-
point.

PR
(31)

On-off controllers are instead used when it is sufficient to guarantee


that a variable remains within a given range of values. Details are re-
ported in Table 2, showing the controlled variable, the corresponding
manipulated variable and the controller type.
D
4. Results

4.1. System testing and characterization


TE

The C10 electrolyzer has been tested in order to assess its perfor-
mance and to collect data for calibration and validation of the dynamic
model.
A first dataset allows analyzing the performance of the PEM stack at
different imposed temperature setpoints. The data (see Fig. 2) includes
stack voltage, current, temperature and cathode and anode backpres-
EC

sures, recorded every 15 s over about 37 h of operation at variable load,


including 6 start-ups. The pressure at the cathode side is always close to
the nominal value of 30 barg (between 28 barg and 32 barg), since this
setpoint cannot be varied and the required pressure is reached very
quickly after system startup. The pressure at the anode side is always
between 1.0 barg and 2.7 barg.
RR

Fig. 10. Auxiliaries consumption and net and gross specific energy consump-
tion as a function of the current density, obtained at nominal temperature and
From these data, stack polarization curves can be depicted: the de-
pressure operating conditions, with fixed vs regulated water flow rate to the pendance of the cell voltage on the temperature is shown in Fig. 3
stack. (above), showing data measured with a cathode backpressure of
30 barg and temperature ranging from about 45 °C to about 60 °C. A
non–negligible improvement in the stack performance at increasing
3.6. Pump
temperature is observed. Conversely, plotting the data obtained with a
constant temperature (50 °C) and cathode backpressure between 29
CO

The flow conditions at pump outlet (pressure and temperature), as


barg and 32 barg (Fig. 3, below) no significant variation in the cell volt-
well as the electric consumption, are calculated according to a steady-
age is observed.
state model from flow rate, temperature and pressure of the inlet flow,
A second dataset has been collected while testing the plant under
assuming constant isentropic and mechanical efficiency.
variable load operation, imposing stepwise changes to the setpoint of
the system outlet hydrogen flowrate and covering the entire operation
3.7. Pipes
range (from 15 % to 100 % of the nominal load). System start–up from
standby conditions has also been tested. Data have been collected with
Pipelines connecting the plant components are included in the all the installed meters (see Fig. 1) and recorded every second (except
model, considering calculation of pressure drops and transport delay. for the power consumption data that are recorded every 15 s), allowing
Included pressure drops are distributed, which are a function of the vol- the characterization of the plant performance over the entire range of
umetric flow rate (Eq. (26), where λ is assumed to be constant) and con- possible loads.
centrated (e.g. in pipes elbow joints, taken into account though a cor- An example of the recorded data, including a warm-up and stepwise
rective coefficient (Eq. (29))) ones. Transport delay, representing changes in the load setpoint (which ranges from the minimum to the
the time that the fluid takes to go from one component to the next one, maximum value) is shown in Fig. 4. The system is started-up from cold
is simulated through a time delay that depends upon pipe length, diam- room temperature conditions (the initial temperature of the water at

11
E. Crespi et al. Energy Conversion and Management xxx (xxxx) 116622

Fig. 6 compares the time profile of the current and of the stack tem-
perature for three warm-up operating conditions, characterized by a
different setpoint for the hydrogen flow rate at plant outlet. The data
show that, whatever is the setpoint, the current increases linearly for
the first two minutes, when no hydrogen is withdrawn from the plant,
to allow the pressure in the H2-H2O to build up to reach the setpoint.
Then, the current varies to meet the generation requirements. The stack
temperature is shown in Fig. 6b. The temperature values obtained in
the first 2 min are not significant because they depend on the previous
operation state of the plant. Indeed, the control of the coolant flow rate

F
is quite slow and, when the initial value of the coolant flowrate is not
zero (warm-up 1 and warm-up 3), almost 2 min are required for the
coolant flow rate to reach the setpoint. In this period, the stack temper-

OO
ature decreases. When the coolant flowrate is at the setpoint (equal to
zero), the rate of variation of the temperature increases with the cur-
rent. This is expected because the higher is the current, the greater is
the heat generation in the cells, responsible for the temperature in-
crease.
It can be observed that, once the warm-up phase is concluded, the
system is able to keep the hydrogen pressure and the stack temperature

PR
close to their nominal values, also when the load varies over the full
range of operating conditions. Fig. 7a shows gross and net power con-
sumption as a function of the stack current density. As expected, the
gross power increases with the current density. The net power, includ-
ing the consumption of the plant auxiliaries (mainly the feedwater
pump, the pump for water refilling and the fans) and the loss related to
the AC/DC power converter ( = 93 %), increases with the cur-
rent density with the same slope as that of the gross power, showing
D
that the internal power consumption is constant at any load. Fig. 7b
shows the gross and net hydrogen flow rates, that are the flowrate of
hydrogen at the stack outlet and at the system outlet, respectively. The
Fig. 11. Variable load operation with variable water flowrate fed to the stack: gross hydrogen flowrate increases in proportion to the current density.
TE

dynamics of water flow rate (above), water utilization factor and fraction of dis- On the contrary, a cloud of points is obtained for the net hydrogen
solved oxygen (below). flowrate because, with the adopted control strategy, peaks in the cur-
rent density are required to keep the hydrogen flowrate constant at the
stack inlet is 23 °C and the initial stack temperature is 25 °C – see Fig. system outlet when the water is purged from the H2-H2O separator. To
4c), and the stack reaches the nominal temperature (55 °C) after better characterize the system performance, each value of net hydrogen
EC

13 min. The flow rate of hydrogen generated by the stack (proportional flowrate has to be associated with the concurrent average of the current
to the current density) increases linearly from the minimum to the nom- density, weighted upon conditions in which the net hydrogen flow rate
inal value in 3 min (Fig. 4a). In contrast, the hydrogen flow rate at plant is constant. The gross specific consumption (Eq. (30)) and the net spe-
outlet is null for 2.5 min, to allow for initial pressurization of the hydro- cific consumption (Eq. (31)) are shown in Fig. 7c and Fig. 7d, respec-
gen product as it increases to the nominal value. The hydrogen pres- tively. The gross specific consumption increases with the current den-
sure, that is equal to 1 atm when the plant is started-up, increases to sity from an average value of 45 kWhe/kgH2 at 0.25 A/cm2 to 53
meet the setpoint (33 bar) in<3 min (Fig. 4b). Once the system is pres- kWhe/kgH2 at 1.6 A/cm2. The net specific consumption reflects the dis-
RR

surized, the current density decreases to the value that allows the pres- persion of the net hydrogen flowrate. However, it increases exponen-
sure in the H2-H2O separator to remain constant. This value depends tially when the current density decreases below 0.6 A/cm2. This trend is
upon the opening of the valve that determines the flow rate of H2 leav- explained by the bigger weight of the nearly constant auxiliary con-
ing the plant, manually controlled. Fig. 4a shows that, to keep the H2- sumption and losses and the constant internal use of hydrogen to regen-
H2O separator pressure constant with a constant flowrate of hydrogen erate the PSA column at lower load operating conditions.
leaving the system (i.e., leaving the H2-H2O separator), the flowrate of
CO

generated hydrogen (i.e., the H2 flowrate entering the H2-H2O separa-


tor) fluctuates over time. Continuous changes of the current density (32)
(and, proportionally, of the rate of hydrogen generation) are required
to avoid large pressure dynamics in the H2-H2O separator, caused by
the intermittent purge of water from the separator and by the pressur- (33)
ization phase of the PSA column. Changes in the current density, which
reacts to the pressure variation related to these events, are shown in
Fig. 5. Regarding the stack temperature, it is important to highlight that 4.2. Model validation
it is measured indirectly through the temperature of the water in the
O2-H2O separator. The temperature oscillation over time is related to The model validation considers a first validation step of the stack
control delays, since it is controlled by varying the temperature of the polarization curve model, and a second step that evaluates the dynamic
water at stack inlet, in turn regulated by controlling the flow rate of model of the entire electrolysis system.
coolant to the heat exchanger. The fluctuations in the temperature re- Regarding the stack polarization curve model, a preliminary sensi-
sults are larger when the current is higher, since more heat has to be re- tivity analysis showed that the values of the anodic exchange current
moved from the stack. density ( ) and of the membrane thickness ( are the two para-

12
E. Crespi et al. Energy Conversion and Management xxx (xxxx) 116622

results of ad hoc simulations. These simulations are performed by set-


ting, for each simulated instant of time, stack current density, coolant
temperature at feedwater cooler inlet, and environmental temperature
equal to the corresponding experimental values. Additionally, the ini-
tial values measured during the plant test are set as initial simulation
values for: temperature of the stack, temperature in the H2-H2O separa-
tor and in the O2-H2O separator, temperature of the water tank, temper-
ature of the coolant, and pressure in the H2-H2O separator and in the
O2-H2O separator (cathode and anode backpressure). The setpoint for
stack temperature and cathode and anode backpressure are also im-

F
posed. The coolant flowrate to the feedwater cooler is null during
warm-up, to allow a faster temperature ramp up, consistent with exper-
imental observations.

OO
Fig. 9 shows the comparison between the experimental data and
the simulated values, for the main parameters, representative of the
system performance. The RMSE on the stack voltage comparison
is < 4 % of the setpoint value, and mainly related to the error asso-
ciated with the stack temperature, which is better controlled in the
simulation. The thermal inertia of the system components is also ac-
curately modelled: during warm-up, experimental and simulated val-

PR
ues of both stack temperature and water temperature at stack inlet
increase with the same slope. The discrepancies obtained in the first
two minutes are explained by the coolant flowrate, which is always
null in the warm-up simulations, while it takes about 2 min to be-
come null in the real plant due to control system delays. Anode and
cathode backpressure are also well modelled: the RMSE are compara-
ble to the experimental error and experimental and simulation results
show the same profile during start-up. Finally, the hydrogen flowrate
D
at plant outlet reflects the different control strategies measured ex-
Fig. 12. Mass flow rate of hydrogen at system outlet over time, obtained at perimentally. In the experimental data, where the cathode backpres-
nominal temperature and pressure, constant current density (1.8, 1.2 and 0.6 sure is kept constant by varying the current density, the flowrate is
constant. Conversely in the simulations, where the current profile is
TE

A/cm2), with base vs modified strategy for PSA beds regeneration.


imposed and the cathode backpressure is kept constant by varying
the backpressure valve opening, the flowrate fluctuates over time. On
meters that mostly influence the results. Thus, the values of these two
average, neglecting the hydrogen flowrate peaks representing the PSA
parameters are calibrated, while the values of the other parameters re-
column pressurization, the error in the prediction of the hydrogen
quired in the polarization curve model are assumed from the literature.
flowrate at plant outlet is very small.
All parameter values used in the cell model are reported in Table 3.
EC

The same kind of comparison between experimental and simulated


As regard the calibration procedure, it consists in comparing the
values has been conducted with 6 other experimental datasets, obtain-
simulated polarization curve (computed through Eq. (13)) to the exper-
ing similar results in terms of errors. The model validation has therefore
imental data, obtained with a constant temperature setpoint equal to
shown the that the system model is able to reproduce the dynamic be-
55 °C and an anodic backpressure equal to 30 bar. The calibration pro-
havior of the real system, allowing improvements of the system in dif-
cedure assigns to the anodic exchange current density at 55 °C and to
ferent conditions or with different control strategies. The application of
the membrane thickness the values that minimize the summed square
the model to optimize the average net efficiency at partial load is the
RR

of residuals for the voltage. The value obtained for the anode exchange
object of the next Section of this work.
current density is compatible with the values reported in the literature
(spanning from 10-13 A/cm2 to 10-3 A/cm2 [13]). Additionally, the
4.3. Control strategies improvement through dynamic simulations
value obtained for the equivalent membrane thickness (i.e., the thick-
ness giving the overall proton resistance of the cell) is compatible with
typical membrane thickness (e.g., 180 μm for Nafion 117). Finally, de- The analysis of the experimental data has shown that the system is
fined the value of the anodic exchange current density at 55 °C, the not optimized to operate at partial load, since the net specific consump-
CO

value of the pre-exponential factor ( is computed with Eq. (17), tion increases exponentially when the current decreases below 0.6
with the activation energy for the anode (Eact,anode) equal to 76,000 J/ A/cm2 (see Fig. 7). The most probable reasons for poor part-load perfor-
mol. mance are that the hydrogen withdrawal to regenerate the PSA bed and
After the parameters calibration phase, the polarization curve the electrical consumption of the system auxiliaries, whose main contri-
model is tested against available experimental data (previously shown bution is the consumption of the supply water pump working at con-
in Fig. 2 and Fig. 3) at different temperature and current density operat- stant rotational speed, remaining the same for any operating power.
ing conditions, as shown in Fig. 8. Simulation results reflect the experi- Simulations are performed to evaluate the benefit of adopting a better
mental data at any current density (ranging from 0.01 to 1.95 A/cm2) control of the feedwater pump and of the PSA regeneration process.
and any temperature (in the 45 °C–60 °C range). The model slightly un- To characterize the system performance at every load, the simula-
derestimates the voltage at low temperature and slightly overestimates tions consider current density setpoints ranging from 0.2 A/cm2 to 1.9
it at high temperature; however, the maximum error is always below A/cm2. For each setpoint, the system operation is simulated for 2 h, and
0.0451 V (2.71 %). The Root Mean Square Error (RMSE) is equal to the average specific consumption values are computed. The instanta-
0.007 V. neous performance values are indeed not representative of complete
The system dynamic model is validated through comparison of ex- system performance since a cyclic fluctuation of the instantaneous per-
perimental data (including measures from all the installed meters) and

13
E. Crespi et al. Energy Conversion and Management xxx (xxxx) 116622

current stepwise, while the water flowrate has a maximum variation


equal to 10 % each second (see Fig. 11). The water utilization factor
and the oxygen concentration in the water at anode outlet show a maxi-
mum variation from the setpoint equal to 1.3 %pt. As regards the stack
temperature control, by fixing the area of the heat exchanger, a correct
control of the temperature is obtained by decreasing the coolant
flowrate when the water flowrate fed to the stack decreases.

4.3.2. PSA column regeneration control strategy improvements


Two different strategies to control the PSA column regeneration are

F
compared. The base case simulations consider that the PSA columns are
switched every 18 min: 17 min are required for the column regenera-
tion and 1 min for the pressurization. During the regeneration phase, a

OO
constant flowrate of hydrogen is withdrawn from the outlet of the col-
umn in operation to regenerate the other column. In the modified strat-
egy, the PSA beds are switched only when the bed in operation is satu-
rated with water. The amount of hydrogen that can be regenerated be-
fore saturating the column is computed from the nominal case, for
which the plant is designed. This means that, when the system operates
at partial load the time between two switches is longer than 18 min but,

PR
since the flowrate of hydrogen withdrawn to regenerate the column is
the same as in the base case, only 17 min are required for the regenera-
tion. After that, the withdrawal of hydrogen is stopped, allowing the net
hydrogen generation to increase.
Fig. 12 compares the flowrate of hydrogen at system outlet for the
base case (top) and with the modified PSA control strategy (bottom).
While in the base case the drop in the hydrogen flowrate due to the
PSA column regeneration is always every 18 min, with the modified
D
control strategy it happens every 28 min at 1.2 A/cm2 and every
58 min with 0.6 A/cm2. The higher hydrogen flowrate obtained in the
extra time between two switches allows decreasing the average inter-
nal hydrogen consumption (i.e., hydrogen vented for drying opera-
TE

tions). This is shown in Fig. 13. In both cases, the 10 % of the gener-
ated hydrogen is internally consumed on average at 1.8 A/cm2. De-
creasing the load, with the base control strategy the weight of the inter-
nal consumption increases, reaching as high as 80 % of the generated
hydrogen at 0.2 A/cm2, while with the modified control strategy it re-
EC

mains constant at 10 %. As a consequence, the average net specific


consumption at partial load improves substantially with the modified
strategy. At the minimum current density (0.2 A/cm2), the net average
specific consumption decreases from above 250 kWhe/kgH2 to 79
kWhe/kgH2 (-68 %), very close to the ideal average specific consump-
tion obtained assuming that no hydrogen losses are present (72
Fig. 13. Hydrogen loss (mass flow rate and percentage of generated hydrogen)
kWhe/kgH2). The use of this control strategy is therefore recommended
RR

and net and gross specific energy consumptions as a function of the current den-
to increase the performance of the plant at partial load, as required to
sity, obtained at nominal temperature and pressure, with base vs modified strat-
egy for PSA beds regeneration. increase the profitability of the flexible system.

5. Summary and conclusions


formance is expected as a consequence of the on–off control of the wa-
ter purge from the H2-H2O separator and of the PSA columns switches. Through experimental testing and dynamic numerical modelling,
CO

this work evaluates the flexibility of an industrial-size PEM electrolysis


4.3.1. Feedwater pump control strategy improvement system, to understand and solve the issues related with intermittent and
Two different strategies to determine the flowrate of water fed to part load operation. In fact, the contribution that these systems can give
the stack are simulated. In the base case, the water flowrate is constant to grid stability with high renewable use is strongly related with its dy-
at any load, while the modified control strategy assumes decreasing the namic performance, predicted by means of a specific simulation tool
water flowrate in proportion with the load, to keep a constant water uti- here developed. Real system testing allowed verifying that the conven-
lization factor. Fig. 10 shows how the modified control strategy allows tional system is able to operate with a variable load, confirming the
the auxiliary power consumption to decrease at partial load. At plant flexibility. However, a drop in performance at partial load is ob-
0.2 A/cm2 it represents 36 % of the total power consumption in the served: the net specific consumption is about 67 kWhe/kgH2 at nominal
base case, and only 9 % of the total power consumption with the modi- load (>1 A/cm2), and increases sharply at lower current operating con-
fied control strategy. The net specific consumption at the minimum ditions (e.g., 80 kWhe/kgH2 at 0.6 A/cm2, 140 kWhe/kgH2 at 0.3
simulated load (0.25 A/cm2) decreases from 191 kWhe/kgH2 to A/cm2). The reason for this poor part-load performance was deter-
145 kWhe/kgH2 (–24 %), whereas the minimum value obtained in the mined to be primarily associated with the current plant control logic,
ideal case (null consumption of the feedwater pump) is 125 kWhe/kgH2. the power consumption of the auxiliaries (80 % of which are associated
To verify that the modified operation strategy doesn’t affect the to the water recirculation pump) and the consumption of hydrogen to
flexibility of the system, simulations have been performed varying the

14
E. Crespi et al. Energy Conversion and Management xxx (xxxx) 116622

regenerate the dryer, leaving room for proposing alternative solutions transition,” 2018.
[9] IEA - International Energy Agency, “Hydrogen,” Paris, 2021. [Online]. Available:
and optimization. https://www.iea.org/reports/hydrogen.
The developed dynamic model is able to reproduce the real plant [10] IRENA - International Renewable Energy Agency, “Green Hydrogen Cost
dynamic operation with errors below 4 % of the main parameters. The Reduction: Scaling up Electrolysers to Meet the 1.5°C Climate Goal,” Abu Dhabi,
2020. [Online]. Available: /-/media/Files/IRENA/Agency/Publication/2020/Dec/
possible reduction of the net specific consumption at part-load opera-
IRENA_Green_hydrogen_cost_2020.pdf?rev=
tion is hence quantified to investigate alternative control strategies. 4ce868aa69b54674a789f990e85a3f00.
When the water flowrate is varied to keep constant the water utilization [11] European Commission, “Fit for 55”: delivering the EU’s 2030 Climate Target on the
factor, the specific consumption at 0.3 A/cm2 (∼15 % of nominal load) way to climate neutrality EN. 2021.
[12] “EU project ‘QualyGridS.’” https://www.qualygrids.eu/ (accessed Feb. 13,
decreases to 110 kWhe/kgH2 (a 21 % reduction with respect to the con- 2022).
ventional constant flow rate case). A further improvement is obtained [13] Clean Hydrogen JU, “Strategic Research and Innovation Agenda (SRIA) 2021-

F
with a smarter regeneration strategy of the PSA columns. The average 2027,” 2022.
[14] Grigoriev S.A, Porembskiy V.I, Korobtsev S.V, Fateev V.N, Aupre F, Millet P.
net specific consumption at 0.3 A/cm2 is equal to 70 kWhe/kgH2 (a High-pressure PEM water electrolysis and corresponding safety issues. Int J
50 % reduction) when, instead of regenerating the PSA columns after a Hydrogen Energy 2010;6:3–10. https://doi.org/10.1016/j.ijhydene.2010.03.058.

OO
fixed amount of time (18 min), the regeneration is performed after that [15] “PRETZEL project website.” http://pretzel-electrolyzer.eu/.
[16] Carmo M, Fritz D.L, Mergel J, Stolten D. A comprehensive review on PEM water
a given amount of H2 has been dried. Such results highlight the impor- electrolysis. Int J Hydrogen Energy 2013;8(1):pp. https://doi.org/10.1016/
tance of the modeling tool proposed in this work to explore higher effi- j.ijhydene.2013.01.151.
ciency management strategies and configurations. Indeed, the adoption [17] Bessarabov D, Wang H, Li H, Zhao N. Electrolysis for Hydrogen Production -.
Principles and Application 2016.
of these control strategies in the real system allows decreasing the spe-
[18] Falcao D.S, Pinto A.M.F.R. A review on PEM electrolyzer modelling : Guidelines
cific system consumption, thus decreasing the operational cost and in- for beginners. J Clean Prod 2020;261. https://doi.org/10.1016/
creasing its economic competitiveness. Additionally, the proposed j.jclepro.2020.121184.

PR
[19] Olivier P, Bourasseau C, Bouamama P.B. Low-temperature electrolysis system
model can be used to simulate variations in the system integrated de-
modelling : A review. Renew Sustain Energy Rev 2017;78(March):280–300.
sign to help further improvements, with a focus on intermittent and https://doi.org/10.1016/j.rser.2017.03.099.
part load operation that are not the main design criteria of current [20] Grigoriev S.A, Kalinnikov A.A, Millet P, Porembsky V.I, Fateev V.N.
units, whose optimization is mainly based on stacks performance. Mathematical modeling of high-pressure PEM water electrolysis. J Appl
Electrochem 2010 https://doi.org/10.1007/s10800-009-0031-z.
[Online].Available:
CRediT authorship contribution statement [21] Görgün H. Dynamic modelling of a proton exchange membrane (PEM)
electrolyzer. Int J Hydrogen Energy 2006 https://doi.org/10.1016/j.ijhydene.%
0A2005.04.001. [Online].Available:
Elena Crespi : Conceptualization, Formal analysis, Investiga-
D
[22] Garcıa-Valverde R, Espinosa N, Urbina A. Simple PEM water electrolyser model
tion, Methodology, Software, Visualization, Writing – original and experimental validation. Int J Hydrogen Energy 2011;7. https://doi.org/
draft. Giulio Guandalini : Conceptualization, Visualization, Writing 10.1016/j.ijhydene.2011.09.027.
[23] Marangio F, Santarelli M, Calı M. “Theoretical model and experimental analysis
– review & editing. Luca Mastropasqua : Conceptualization, Visual-
of a high pressure PEM water electrolyser for hydrogen production” 2009;34:
ization, Writing – review & editing. Stefano Campanari : Supervi-
TE

1143–58. https://doi.org/10.1016/j.ijhydene.2008.11.083.
sion, Writing – review & editing. Jacob Brouwer : Supervision, Re- [24] Huiyong K, Mikyoung P, Soon L.K. One-dimensional dynamic modeling of a
high-pressure water electrolysis system for hydrogen production. Int J Hydrogen
sources, Writing – review & editing.
Energy 2013 https://doi.org/10.1016/j.ijhydene.%0A2012.12.006.
[Online].Available:
Declaration of Competing Interest [25] Kazou O, Toshio M, Takeshi H, Misaki K, Pyouhei N, Kohei I. Performance
analysis of polymer-electrolyte water electrolysis cell at a small-unit test cell and
performance prediction of large stacked cell. -journal Electrochem Soc 2002
EC

The authors declare that they have no known competing financial https://doi.org/10.1149/1.1492287. [Online].Available:
interests or personal relationships that could have appeared to influ- [26] Santarelli M, Medina P, Calì M. Fitting regression model and experimental
ence the work reported in this paper. validation for a high-pressure PEM electrolyzer. Int J Hydrogen Energy 2009
https://doi.org/10.1016/j.ijhydene.2008.11.036. [Online].Available:
[27] O. Rallières, “Modélisation et caractérisation de piles a combustible et
Data availability electrolyseurs PEM,” 2008.
[28] S. Rabih, “Contribution à la modélisation de systèmes réversibles de types
électrolyseur et pile à hydrogène en vue de leur couplage aux générateurs
Data will be made available on request.
RR

photovoltaïques,” 2008.
[29] Yigit T, Selamet O.F. Mathematical modeling and dynamic Simulink simulation
Acknowledgment of high-pressure PEM electrolyzer system. Int J Hydrogen Energy 2016;41(32):
13901–14. https://doi.org/10.1016/j.ijhydene.2016.06.022.
[30] Olivier P, Bourasseau C, Bouamama B. Dynamic and multiphysic PEM
The authors wish to thank John Stansberry and Robert Hayden Lav- electrolysis system modelling : A bond graph approach. Int J Hydrogen Energy
iguer for their support in the experimental activity and Simone Molho 2017;42(22):14872–904. https://doi.org/10.1016/j.ijhydene.2017.03.002.
for his contribution to the modelling activity. [31] Sood S, et al. Generic Dynamical Model of PEM Electrolyser under Intermittent
Sources. Energies 2020;1–34. https://doi.org/10.3390/en13246556.
CO

[32] Stansberry J.M, Brouwer J. Experimental dynamic dispatch of a 60 kW proton


References exchange membrane electrolyzer in power-to-gas application. Int J Hydrogen
Energy 2020;45(16):9305–16. https://doi.org/10.1016/j.ijhydene.2020.01.228.
[1] European Commission, “A Clean Planet for all. A European long-term strategic [33] B. J. McBride, S. Gordon, and M. A. Reno, “Coefficients for Calculating
vision for a prosperous, modern, competitive and climate neutral economy,” Com Thermodynamic and Transport Properties of Individual Species,” Nasa Tech.
(2018) 773, p. 114, 2018. Memo., vol. 4513, no. NASA-TM-4513, p. 98, 1993, [Online]. Available: http://
[2] IEA - International Energy Agency, “Harnessing Variable Renewables - A Guide ntrs.nasa.gov/archive/nasa/casi.ntrs.nasa.gov/19940013151_1994013151.pdf.
to the Balancing Challenge,” 2011. doi: 612011171P1. [34] Wagner W, Pruss A. International Equations for the Saturation Properties of
[3] Impram S, Varbak S, Oral B. “Challenges of renewable energy penetration on Ordinary Water Substance. Revised According to the International Temperature
power system flexibility : A survey”, Energy. Strateg Rev 2020;31, August:100539. Scale of 1990. J Phys Chem Ref Data 1993;22(3):pp. https://doi.org/10.1063/
https://doi.org/10.1016/j.esr.2020.100539. 1.555926.
[4] Hydrogen Council, “How hydrogen empowers the energy transition,” 2017. [35] Abdin Z, Webb C.J, Gray E.M. Modelling and simulation of a proton exchange
[5] Lehner M, Tichler R, Steinmueller H, Koppe M. Power-to-Gas: Technology and membrane (PEM) electrolyser cell. Int J Hydrogen Energy 2015;40(39):13243–57.
Business Models 2014. https://doi.org/10.1016/j.ijhydene.2015.07.129.
[6] Crespi E. Energy storage with Power-to-Power systems relying on photovoltaic [36] Santarelli M.G, Torchio M.F, Cochis P. Parameters estimation of a PEM fuel cell
and hydrogen: modelling the operation with secondary reserve provision. J Energy polarization curve and analysis of their behavior with temperature. J Power
Storage 2022. https://doi.org/10.1016/j.est.2022.105613. Sources 2006;159:824–35. https://doi.org/10.1016/j.jpowsour.2005.11.099.
[7] S. J. Davis et al., “Net-zero emissions energy systems,” vol. 9793, no. June, 2018, [37] S. Dutta, S. Shimpalee, and J. W. Van Zee, “Numerical Prediction of Mass-
doi: 10.1126/science.aas9793. Exchange between Cathode and Anode Channels in a PEM Fuel Cell Numerical
[8] IRENA, “Hydrogen from renewable power: technology outlook for the energy prediction of mass-exchange between cathode and anode channels in a PEM fuel

15
E. Crespi et al. Energy Conversion and Management xxx (xxxx) 116622

cell,” vol. 9310, no. June, 2001, doi: 10.1016/S0017-9310(00)00257-X. [43] Espinosa-Lopez M, et al. Modelling and experimental validation of a 46 kW PEM
[38] Incropera, DeWitt, and B. Lavine, Fundamentals of Heat and Mass Transfer. . high pressure water electrolyzer. Renew Energy 2018;119:160–73. https://
[39] Lebbal M.E, Lecœuche S. Identification and monitoring of a PEM electrolyser doi.org/10.1016/j.renene.2017.11.081.
based on dynamical modelling. Int J Hydrogen Energy 2009;34(14):5992–9. [44] Shiva Kumar S, Himabindu V. Materials Science for Energy Technologies
https://doi.org/10.1016/j.ijhydene.2009.02.003. Hydrogen production by PEM water electrolysis – A review. Mater Sci Energy
[40] Harrison K.W, Hernández-Pacheco E, Mann M, Salehfar H. Semiempirical model Technol 2019;2(3):442–54. https://doi.org/10.1016/j.mset.2019.03.002.
for determining PEM electrolyzer stack characteristics. J fuel cell Sci Technol 2006; [45] P. Colbertaldo, S. L. Gómez Aláez, and S. Campanari, “Zero-dimensional dynamic
3(May):3–6. https://doi.org/10.1115/1.2174072. modeling of PEM electrolyzers,” in Energy Procedia, 2017, vol. 142, pp. 1468–1473,
[41] A. S. Tijani and A. H. A. Rahim, “Numerical Modeling the Effect of Operating doi: 10.1016/j.egypro.2017.12.594.
Variables on Faraday Efficiency in PEM Electrolyzer,” in 3rd International [46] D. Bessarabov and P. Millet, PEM Water Electrolysis, 1st ed. Academic Press, 2018.
Conference on System-integrated Intelligence: New Challenges for Product and [47] B. Yodwong, D. Guilbert, M. Phattanasak, W. Kaewmanee, M. Hinaje, and G.
Production Engineering, SysInt 2016, 2016, vol. 26, pp. 419–427, doi: 10.1016/ Vitale, “Faraday’s Efficiency Modeling of a Proton Exchange Membrane
j.protcy.2016.08.054. Electrolyzer Based on Experimental Data,” pp. 1–14, 2020.

F
[42] Ni M, Leung M.K.H, Leung D.Y.C. Energy and exergy analysis of hydrogen [48] R. G. Cunningham, “Orifice Meters with Supercritical Compressible Flow.” pp.
production by a proton exchange membrane (PEM) electrolyzer plant. Energy 625–638, 1951.
Convers Manag 2008;49:2748–56. https://doi.org/10.1016/
j.enconman.2008.03.018.

OO
PR
D
TE
EC
RR
CO

16

You might also like