PEM Model

i n t e r n a t i o n a l j o u r n a l o f h y d r o g e n e n e r g y x x x ( 2 0 1 7 ) 1 e3 3
Available online at www.sciencedirect.com
ScienceDirect
journal homepage: www.elsevier.com/locate/he
Dynamic and multiphysic PEM electrolysis system

modelling: A bond graph approach
Pierre Olivier a,*, Cyril Bourasseau a, Belkacem Bouamama b

a
Univ Grenoble Alpes CEA/LITEN, 17 rue des Martyrs 38054 Grenoble, France
b
Lille University, CRIStAL, UMR 9189, 59650 Villeneuve d'Ascq, France
article info abstract
Article history: Proton exchange membrane (PEM) electrolysis technology appears as a key technology in
Received 9 June 2016 the development of hydrogen-energy market applications such as energy storage or fuel for
Received in revised form mobility. Its coupling with renewable electrical sources involves some issues related to the
16 February 2017 intermittent operation of PEM electrolysis systems. Within this framework, modelling is an
Accepted 1 March 2017 essential tool to understand these issues, provide a thorough analysis and suggest some
Available online xxx design optimization. A bibliographic analysis was carried out to identify existing models.
State of the art highlighted that, although it is critical for the conception of such systems,
Keywords: only a few models take into account the dynamic of the whole system including balance of
Modelling plant. Therefore, in this paper a new dynamic and multiphysic model of a proton exchange
Bond graph membrane electrolysis system is presented. It was first developed under a graphical
Water electrolysis system modelling formalism: the bond graphs (BG). Regarding dynamic and multiphysic modelling
Hydrogen of complex systems, the BG have many advantages: it involves four levels of modelling
PEM using only one tool; it is a unified multiphysic approach; the parameters used have a
Simulation physical meaning; the BG model can be refined very easily by adding new elements without
having to start again the modelling process. Finally, because of its causal and structural
properties, BG is suitable for modelling but also for control, sizing and diagnosis analysis.
The model was then transcribed systematically into block diagrams in modular fashion for
reutilisability of model libraries components. After a validation process, the model was
proved to describe accurately the dynamic behaviour of a semi-industrial PEM electrolysis
system (25 kW). The dynamic model can now be used to achieve some analyses through BG
structural properties and simulations. Thus, it is a powerful tool to improve the design of
PEM electrolysis systems powered by intermittent electrical sources.
© 2017 Hydrogen Energy Publications LLC. Published by Elsevier Ltd. All rights reserved.
policies promote the massive development of renewable en-

Introduction ergy such as photovoltaic or wind energy. The intermittent
production profiles of such technologies involve the need of
In order to face climate change and mitigate anthropic energy storage in order to face electrical network stability is-
greenhouse gas emissions, fossil energy resources have to be sues [2]. Proton exchange membrane (PEM) electrolysis tech-
substituted [1]. In the field of electricity production, European nology enables to cope with these issues by converting
* Corresponding author.
E-mail address: pierre.olivier44@gmail.com (P. Olivier).
http://dx.doi.org/10.1016/j.ijhydene.2017.03.002
0360-3199/© 2017 Hydrogen Energy Publications LLC. Published by Elsevier Ltd. All rights reserved.
Please cite this article in press as: Olivier P, et al., Dynamic and multiphysic PEM electrolysis system modelling: A bond graph approach,
International Journal of Hydrogen Energy (2017), http://dx.doi.org/10.1016/j.ijhydene.2017.03.002
2 i n t e r n a t i o n a l j o u r n a l o f h y d r o g e n e n e r g y x x x ( 2 0 1 7 ) 1 e3 3
electricity excess productions into hydrogen. Regarding other parameters used have a physical meaning and the BG model
electrolysis technologies (alkaline and solid oxide electrolysis can be refined very easily by adding new elements without
technologies), PEM technology appears as the most suitable having to start again the modelling process. Finally, due to its
for an intermittent operating mode [3]. structural and causal properties, the bond graph is suitable for
Besides, its zero-pollutant emission during utilization and control analysis, dimensioning, and diagnosis analysis [37].
its wide range of uses [4e6], especially as a fuel for mobility, Because of the advantages exposed below, some authors
make hydrogen a good alternative to fossil energy [5]. In the developed their PEM electrolyser models using BG or other
climate change context, hydrogen-energy is then a good way graphical modelling formalism such as Energetic Macroscopic
to facilitate the transition towards a low-carbon society Representation (EMR, more suitable for electric-mechanical
[7e10]. Nevertheless, this hypothetical substitution involves systems). In particular, Laplace laboratory in Toulouse
large-scale production of carbon free hydrogen. Therefore, (France) developed BG models [18,19] with a dynamic and
PEM electrolysis systems powered with renewable electricity coupled description of electrochemical, electrical, thermal and
[11], in order to produce and store hydrogen, is a good op- mass transfer phenomena in a PEM electrolysis cell. The model
portunity to face two challenges: (i) Fulfil the need of energy developed by Agbli using EMR [20,21] involves the description
storage caused by the integration of renewable energy sources of electrochemical, electrical and thermal phenomena. It en-
on the European electricity network; (ii) Cope with the ables to show on a laboratory scale system the existing thermal
growing need of carbon free hydrogen caused by the future coupling between the stack and the water tank (here anode
market applications of hydrogen energy (especially in side recirculation system including pipes and water-oxygen
mobility usage). separator) and the necessity to have a systemic approach in
There are some issues related to the operation of an elec- order to obtain a good description of stack thermal behaviour.
trolysis system at nominal point (temperature regulation Finally, Zhou et al. [14,15] developed a PEM electrolysis system
[12,13], pressure [14,15] and gas purity regulation [16,17], etc.). model using EMR. This model provides a description of elec-
Some of these ones are aggravated by coupling the system trical and pneumatic phenomena. It enables to study and
with intermittent electrical sources. As industrial electro- validate regulation loops of stack voltage and hydrogen pres-
lysers were first designed to operate at nominal conditions, it sure through Hardware in the Loop simulations (HiL).
is crucial to understand these issues, provide a thorough All PEM electrolysis models found in the literature are not
analysis and suggest some design improvements (both hard- based on graphical modelling formalism. Beyond these
ware and software). Within this framework, numerous particular publications, most of the reviewed models rely on
experimental works are achieved to investigate impacts of the mathematical description of stack polarization curve
materials, catalyst loadings, membrane electrode assemblies (steady-state electrochemical models, either empirical or
(MEAs) preparation process, operating conditions, stack as- analytical) in order to compute stack efficiency. Some of them
sembly and design, etc. on PEM electrolysis cells and stack include a dynamic thermal model of the stack [22e26], and/or
efficiency and reliability. In addition, modelling enables to mass transfer descriptions [26e29] in order to take into account
investigate design and operating situations by spending less the dynamic impact of temperature and pressures/concentra-
time and effort than experiments. It is a critical and powerful tions on the steady-state stack electrochemical response. Be-
tool for phenomena understanding, control analysis, sizing, sides, only a few of them include systemic considerations such
energy management and optimisation. Computational as: balance of plant consumption and in particular converters
modelling field applied to PEM electrolysis technology was [30], thermal integration of the stack into an electrolysis sys-
really launched in the late 2000s. tem, or control of converters or pneumatic valves [14,15]. These
Dynamic modelling of PEM electrolysis systems is com- systemic considerations remain incomplete regarding PEM
plex. The complexity is due to the coupling of different kind of electrolysis system modelling. As a result, only a few models
energies (electrochemical, electrical, thermos fluidic…) and its are dedicated to control and monitoring analysis [14,15,24].
non-stationarity. The innovative interest of proposed Finally, state of the art highlights that although it is critical
research is the use of bond graphs (BG) as a multidisciplinary for the conception of such systems, none of the reviewed
tool for dynamic modelling of a PEM electrolysis system taking models takes into account the dynamic of the whole system
into account coupling of all types of energies occurring in the including balance of plant (BoP: circulating pumps, tempera-
process. ture regulation system, gas purification system, etc.).
Regarding dynamic and multiphysic modelling of complex Furthermore, most of the models do not describe all the en-
systems, the BG have many advantages. First, it is a unified ergies involved in such systems (electrical, thermal, chemical,
multiphysic approach which involves four levels of modelling and fluidic).
using only one tool (technological level, physical level, math- Given the limits of literature, BG modelling formalism was
ematical level and algorithmic level). Indeed to deal with the used to build a complete multiphysic and dynamic model of a
enormous amount of equations describing the dynamic PEM electrolysis system. Its aim is to analyse and simulate the
behaviour of the different phenomena occurring in the system behaviour of the whole system, including BoP, when operated
(mathematical level), the bond graph, by its graphical nature, under intermittent solicitation. The innovative interest of this
enables to display precisely the exchange of power in a sys- paper is the development of a new model able to represent a
tem, including storage and transformation (physical level). whole PEM electrolysis system including stacks and BoP,
The advantage here is to deduce systematically equations model which was proved to describe accurately the dynamic
(mathematical level) from a graphical description of the power behaviour of an experimental system. Therefore, it is orga-
exchange in the system (physical level). Moreover, the nized as follows. After this brief introduction, the second
i n t e r n a t i o n a l j o u r n a l o f h y d r o g e n e n e r g y x x x ( 2 0 1 7 ) 1 e3 3 3
section is dedicated to the presentation of the electrolysis consumed water flow by reaction. As water is a great heat
technology and the general behaviour of a PEM electrolysis carrier, heat exchanger and cooling circuit are included to
system. The BG modelling approach applied to our system is control stack temperature. These components are located on
detailed in section “PEM electrolysis system modelling”. First, the anode side because of the irreversibilities more important
basic knowledge on BG is exposed. Then, models of each on the anode side than on cathode side [3].
component of the system are presented with respect to some At stack outputs, unreacted water and produced gases are
mentioned hypotheses. Finally, section “Results and mixed in a biphasic fluid. Therefore, separators are placed at
discussion” deals with simulation results, experimental vali- stack outputs in order to separate these two phases. The water
dation, and perspectives are given regarding the numerous levels in these separators are not constant. Indeed, water is
possibilities offered by the model. consumed on the anode side and it migrates through mem-
brane from anode to cathode in the cells (by electro-osmosis
[26e28]). Therefore, a controlled valve enables to feed oxy-
Description of the electrolysis technology gen separator with purified water from an input water con-
ditioning unit. A second controlled valve is placed between
The water electrolysis reaction is an electrolytic process separators in order to control hydrogen separator level.
which decomposes water into dioxygen and dihydrogen with After the separators, the gases have to be conditioned
the help of a DC electrical source (1). depending on their uses. Their purity and pressure are
controlled through output conditioning units. Finally, each
1
H2 O ¼ H2 þ O2 (1) PEM electrolysis system is composed of different sensors,
2
safety devices, fans, air conditioning to control enclosure
It takes place in an electrolysis cell composed of: (i) an temperature and a control and monitoring system.
electrolyte (ionic conductor), (ii) active layers with catalysts at An electrolysis system is then a complex system as it in-
which the redox reactions occur and (iii) current and materials volves multiphysic phenomena (Fig. 2) and numerous com-
collectors (electronic conductors) enabling the electricity ponents (Fig. 3).
supply and the supply and collection of reagents and prod-
ucts. In a PEM electrolyser, the electrolyte is a proton ex-
change membrane. PEM electrolysis system modelling
Water is supplied at the anode reaction interface where
molecules are split into oxygen, protons and electrons. Elec- Before the modelling, a short presentation of the modelling
trons move to the cathode via the outer electrical circuit while tool and the description of the experimental device are
protons migrate from anode to cathode through the PEM introduced.
membrane. At the cathode reaction interface, hydrogen is
formed by electrons and proton recombination (Fig. 1). Further
details about PEM electrolysis technology (cell conception,
used materials, etc.) can be found in Refs. [3,31e34].
In order to produce hydrogen at large scale, cells are
assembled into a stack by the mean of bipolar plates. These
ones contain channels in order to feed the cell with reagents
and evacuate the gases produced. An electrolysis stack is then
an assembly of numerous open electrochemical reactors in
which numerous multiphysic phenomena occur simulta-
neously (Fig. 2).
In order to ensure the good operation of these reactors,
several functions are required and cannot be accomplished by
the stack on its own. That is why an electrolysis system must
include auxiliary components and circuits in order to carry
out: (i) temperature regulation of the stack(s); (ii) anode and
cathode input conditioning (control the purity and the flow
rate of input water); (iii) anode and cathode output condi-
tioning (gas evacuation, pressure and purity regulation); (iv)
power supply regulation: adapt the input power to the stack
currentevoltage characteristics.
The power supply regulation is achieved by a converter. It
enables to control stack voltage (stack operating point). Then,
closed water circulation circuits composed of pipes and
circulating pumps allow feeding the stack in reagent. Even if
water is only consumed on the anode side, cathode circulation
circuit is usually included in order to facilitate produced
hydrogen evacuation from the stack. For this reason, among
other ones, circulated water flow is much higher than the Fig. 1 e A PEM electrolysis cell.
Fig. 2 e Coupled multiphysic phenomena involved in an electrolysis stack (inspired from Ref. [20]).
Fig. 3 e Technological scheme of a PEM electrolysis system.
Short presentation of bond graph modelling formalism between subsystems. BG is a multidisciplinary tool using a
unified formalism for modelling multi-physic systems.
Bond graph is an energetic representation, initially invented
by Paynter [35] and then developed by Karnopp and Rosenberg Basic elements of bond graph
[36] that allows showing graphically the exchanges of power BG is a unified graphical language for multi-physic domains. A
within a modelled system and the dynamical interaction BG is a graph G(S, A), where the nodes S represent physical
(1-junction), or flow (0-junction). The conservative energy laws

are obtained from the latter. TF, and GY are used to represent
transformers and gyrators, respectively to model trans-
formation of energy from one domain into another (Table 2).
Advantages regarding PEM electrolysis system modelling

The Bond graph modelling formalism is an inherent dynamic
approach which enables to represent any kind of power ex-
change in the same way. Furthermore the BG model is subject
to evolution: the model can be refined by adding BG elements
without having to start again the modelling process. This is
why this tool is well suited for modelling PEM electrolysis
Fig. 4 e Bond graph representations (a1 and a2) and system characterized by interaction of different energies (e.g.
corresponding causalities (b1 and b2): in (a1) and (b1) thermal, electro-chemical, electrical, thermo-fluidic…). More-
system A imposes flow on B; in (a2) and (b2) system A over, by its graphical nature, BG enables to display precisely the
imposes effort on B. exchange of power in a system. The equations describing the
dynamic behaviour of the different phenomena occurring in
components, subsystems, and other basic elements called the system can be then deduced systematically from this
junctions while the edges A called power bonds represent the graphical representation. This is an important aspect regarding
exchanged powers labelled by two conjugated variables called bond graph advantages given the amount of equations
effort (e) and flow (f) (Fig. 4). The product of those variables is involved. Finally, BG can be used not only for modelling but also
the instantaneous transferred power which is graphically for control and diagnosis purposes because of its structural and
represented by a half arrow. The effort and flow variables have causal properties. Therefore, many research works have been
a different meaning depending on the physical field that is published on BG theory for modelling and further for supervi-
considered (Table 1). Many systems such as thermo-fluid, sion, control analysis and automated modelling [37].
biological and chemical engineering can be also represented
usefully as pseudo-bond graphs where the product of effort Description of the modelled PEM electrolysis system
and flow variable is not a power but the theory of BG is un-
changed (Table 1). One important structural property of the BG The considered system in this paper is a 27 kW powered PEM
is its causality concept. Indeed, the determination of causes electrolysis system with 2 stacks of 48 cells. Its architecture is
and effects in the system is directly deduced from the the same than the one presented in Fig. 3 with water recir-
graphical representation. In the BG, it is denoted by the causal culation at both electrodes.
stroke indicating that the effort acts to the right. The side of Many components are specific to each stack. Only the
the causal stroke for the flow is in the reverse direction. following subsystems are common for the two stacks: input
Assigned causality means that system A imposes efforts on B water conditioning, hydrogen output conditioning and oxygen
as indicated by corresponding block diagram given by Fig. 4. output conditioning. The input water conditioning is
Independently of the causality, the direction of the positive composed of a water purification unit by reverse osmosis and
power is indicated by the half-arrow on the bond. a purified water tank. The oxygen output conditioning con-
sists only of pipes and exhaust valves given produced oxygen
Bond graph elements is not recovered and its pressure not controlled. The hydrogen
Only a limited number of elements are necessary to describe output conditioning is composed of dryer and controlled
the majority of systems. The R-element represents a passive valves in order to regulate hydrogen pressure. The cooling
energy dissipation phenomenon, while C, and I model the circuit consists of a coolant circulation circuit and a chiller.
passive energy storage (under potential and kinetic format) Regarding control of the system, the following variables are
elements. (Se), and (Sf) are the sources of effort (intensive regulated: (i) stack temperature through the control of a
variable) and flow (derivative of extensive variable), respec- chiller; (ii) hydrogen pressure through valves; (iii) separator
tively. Sensors are represented by flow (Df), and effort (De) water levels through the control of valves opening; (iv) stack
detectors. Finally, J (which can be a zero or a one junction), is voltage through the control of stack converter; (v) circulating
used to connect the elements having the same effort mass flows through the control of circulating pumps.
Table 1 e Bond graph variables.

Physical domain Effort (e) Flux (f)
Electrical Voltage U (V) Current intensity I (A)
Fluidic Pressure P (Pa) Volume flow rate (m3 s1)
Fluidic (pseudo bond graph) Pressure P (Pa) Mass flow (kg s1)
Thermal Temperature T (K) Entropy flow (J K1 S1)
Thermal (pseudo bond graph) Temperature T (K) Thermal flow (J s1)
Chemical (transformation) Chemical potential (J mol1) Molar flow (mol s1)
Chemical (kinetic) Chemical affinity A (J mol1) Reaction speed flow rate (mol s1)
Technical specifications and operating conditions of the sys- always taken in most of the model reviewed. It remains valid
tem are detailed in Table 3. as long as the number of cell is quite enough and the behav-
iour deviation of extreme cells is then negligible considering
Bond graph modelling of the PEM electrolysis system the behaviour of the whole stack.
Bond graph modelling enables to integrate four levels of Electrochemical model. Regarding their response time in a PEM
modelling using only one tool: (i) the technological level given electrolysis system, electrical and electrochemical phenom-
by the word bond graph which represents the architecture of ena are considered instantaneous (steady-state). Therefore,
the system in a modular fashion; (ii) the physical level using these phenomena are described through algebraic equations.
an energy description of the physical phenomena based on BG In BG elements, the stack is described through non-linear
elements; (iii) the mathematical level which describes the resistive elements which generate heat and a transformer
system behaviour by writing the constitutive equations of BG element which represent the conversion of electrical energy
elements; (iv) the algorithmic level indicating how the math- into chemical energy (Fig. 6).
ematical models are calculated. From 1-junctions, equations (2)e(4) are deduced.
As a result of these structural and causal properties, a dy- X
namic model under state equation or block diagram format is hk ¼ Ecell Erev (2)
k
deduced systematically from the graphical model using
dedicated software for simulation and analysis. For this last X
Eohm ¼ hk Eact;ano Eact;cat (3)
reason, once the BG model was built, it was then transcribed
k
into block diagrams and implemented into simulation soft-
ware Simulink®. The different steps of the modelling process
Eohm ¼ Ecell Erev Eact;ano Eact;cat (4)
are described in the following subsections.
Indeed, at non-zero current, the cell voltage is composed
Word bond graph by two terms, the reversible potential Erev which is given by the
P
The first step in the BG modelling approach is to build the thermochemical model and the overvoltages hk caused by
k
word bond graph which is a representation of the architecture the irreversible dissipation generated by the phenomena
of the system with all its components and the nature of the linked to the current flow in the cell. The major dissipation
exchange power between each of them. The word bond graph mechanisms in a PEM electrolysis cell are the following: (i) the
of the PEM electrolysis modelled is represented in Fig. 5. In this activation overvoltages Eact;k due to the chemical reaction at
figure, the blue lines represent fluidic energy and the red lines, both electrodes k; (ii) the ohmic overvoltage Eohm due to the ion
thermal energy. The green ones represent both fluidic and transport in the electrolyte and the electron transport in the
thermal energy (thermos fluidic) related to hydrogen flow, and electrode; (iii) the diffusion overvoltage due to the transport of
orange lines represent thermos fluidic coupling energy related electro-active species to the electrochemical reaction loca-
to oxygen flow. tion. This last overvoltage appears only at high current density
The meaning of each variable indexes can be found in and it is not considered in the model. All these terms are
Table 4. developed in the following subsections.
In the next step, models are built for each component using
an energy description of the physical phenomena based on BG Electrochemical transformer and reversible potential. From
elements. Based on these BG models, constitutive equations the Tf-element, the reaction speed flow rate x_ is deduced from
are deduced. Finally, the causality, which indicates how the the cell electric current Icell and the reversible potential is
mathematical models are calculated, enables to transcribe BG computed from the free enthalpy of water dissociation DrG
models into block diagrams. This modelling process is (equation (5) and (6)). This transformer element represents the
exposed for each component of the system. conversion of electrical energy into chemical energy and en-
ables to describe the Faraday's law1 (i.e. the conversion of elec-
General hypotheses trical energy into chemical energy, with F the Faraday constant).
Each modelling work involves some hypotheses. Regarding
Icell
the PEM electrolysis system model exposed in this paper, the x_ ¼ (5)
2$F
general hypotheses taken are detailed below:
DrG
The model uses ordinary differential equations: lumped Erev ¼ (6)
2$F
parameters are involved.
The reversible potential is the minimum voltage to apply
Pumps, chillers and fans are considered perfect mass flow
between the two electrodes of an electrolysis cell in order to
sources.
make the water dissociation reaction occur. Its analytical
The gases are assumed to be ideal
expression can be obtained from energy balance on the ther-
modynamic system “stack” [38]. It is given here by the
Equivalent mean cell/stack
The stack is represented by a unique cell, assuming all cells 1
This equation can be easily established from mass balances
have the same average behaviour (equivalent mean cell hy-
on an electrolysis cell computed at steady-state. It is assumed
pothesis). Therefore, a perfect distribution of current and there is no leakage current and faraday efficiency is considered
liquid flows between cells is considered. These hypothesis is equal to one.
Table 2 e Bond graph elements.

Elements Symbols Constitutive equations Designations
Sources Source of effort
Source of flow
Passive elements
Dissipative elements FR ðe; f Þ ¼ 0 Resistance
R
Storage elements FC ðe; fdtÞ ¼ 0 Capacitance
R
FI ðe; edtÞ ¼ 0 Inductance
Junctions
e1 ¼ me2
Transformers Transformer
f2 ¼ mf1

e1 ¼ rf2
Gyrator
f2 ¼ rf1

e1 ¼ e2 ¼ e3
Junctions «Zero» junction same effort
f1 f2 þ f3 ¼ 0

f1 ¼ f2 ¼ f3
«One» junction same flow
e1 e2 þ e3 ¼ 0
(continued on next page)
Table 2 e (continued )
Elements Symbols Constitutive equations Designations

e ¼ eðtÞ
Sensors Flow sensors (Df) and
f ¼0
effort sensors (De)

f ¼ f ðtÞ
e¼0
thermochemical model. Considering the expressions of free With PH2 the hydrogen pressure, PO2 the oxygen pressure
enthalpy of reaction (7) and chemical potentials mi (8), equa- and aH2 O the chemical activity of water. The first term is the
tion (9) is obtained. reversible potential in standard conditions E0rev at a considered
X temperature, and the second one describes the deviation from
DrG ¼ ni $mi (7) standard operating conditions. The value of the standard
i
reversible potential at 298 K is 1.229 V. This potential is
ref
temperature-dependent. Its temperature dependency can be
mi ¼ mi þ R$T$lnðai Þ (8)
computed through Kirchhoff laws (equation (13)).
!
ref ref ref
mH2 þ 0; 5$mO2 mH2 O R$T Y n DrG0 ðTÞ
Erev ¼ þ $ln ai i (9) E0rev ðTÞ¼
2$F 2$F 2$F
i Z T Z T DrC0 ðTÞ
p
DrH0 ðT0 ÞT0 $DrS0 ðT0 Þþ DrC0p ðTÞ$dTT$ $dT
With mref
the standard chemical potential of species i, mi the T0 T0 T
i ¼
chemical potential of species i, ai the chemical activity of 2$F
species i, ni the stoichiometric coefficient of species i, T the (13)
temperature and R the ideal gas constant. With DrH0 ðT0 Þ the standard enthalpy of reaction at tem-
The standard free enthalpy of reaction can be isolated (10) perature T0 , DrS0 ðT0 Þ the standard entropy of reaction at
and the general expression of reversible potential is obtained temperature T0 and DrC0p ðTÞ the standard specific heat of re-
(11). action at temperature T.
ref ref ref Besides, numerous empiric correlations based or not on
mH2 þ 0; 5$mO2 mH2 O DrG0
E0rev ¼ ¼ (10) thermodynamic tables can be found in the literature
2$F 2$F
[21,23,27,28,39e44]. The one used in the model is the following
! (14) [21,23,28].
R$T Y n
Erev ¼ E0rev þ $ln ai i (11)
2$F i E0 ðTÞ ¼ 1; 5184 1; 5421$103 $T þ 9; 523$105 $T$lnðTÞ
With DrG0 and E0rev respectively the standard free enthalpy þ 9; 84$108 $T2 (14)
of reaction and the standard reversible potential. The following subsections enable to establish the intrinsic
Considering the gases are ideal, the following expression is equations of RS-elements (both resistive and entropy source
obtained (12). element) of the BG model.
pffiffiffiffiffiffiffi!
R$T PH2 $ PO2
Erev ¼ E0rev þ ln (12) Activation overvoltages. The activation overvoltages come
2$F aH2 O
from the kinetics of the electronic charge transfer reactions
that occur at electrodes reaction sites [38]. Applying the acti-
vated complex theory to the half-reactions, the Butlere
Table 3 e Technical specifications of experimental device. Volmer equation can be established in equation (15).
Electrolysis technology PEM
Power 27 kW CPE ak $z$F$Eact;k CRE ð1 ak Þ$z$F$Eact;k
Jcell ¼ Jk $exp $exp
Number of stacks 2 CPS R$T CRS R$T
Number of cells by stacks 48 (15)
Cell Area 130 cm2
Nominal hydrogen flow 5 Nm3 h1 With Jcell the current density of electric cell, Eact;k the acti-
Nominal current density 0.92 A cm2 vation overvoltages due to the chemical reaction at both
Nominal temperature 55 C electrodes k, CPE , CRE , CPS and CRS the different concentrations
Nominal hydrogen pressure 13 bars of products (P) and reagents (R) respectively near the electrode
Operating range 10e100% (E) and in the solution (S), ak and Jk respectively the charge
transfer coefficient and the current exchange density of the

Table 4 e Index of variables used in the bond graph
half-reaction occurring at electrode k, z the number of elec-
models.
trons involved in the half-reaction, R the ideal gas constant, T
Variables index Variables nature
the temperature and F the Faraday constant.
A first hypothesis is to consider diffusion is not a limiting IN Input electrical variables
process during the reaction. As a result, the concentrations in CELL/STACK Cell/Stack electrical variables
TANK Water tank variables
the bulk and near the triple-bound contact are assumed to be
ANO Anode side variables
equal and diffusion overvoltages are not considered in this CAT Cathode side variables
model. This hypothesis remains valid as long as the current SEPO Oxygen separator water variables
density is not too high (equation (16)). Indeed, this phenom- SEPH Hydrogen separator water
enon appears only at high current density (above 2 A cm2) variables
and the maximal current density of the experimental device is OXYG Oxygen separator gas variables
HYDR Hydrogen separator gas variables
0.92 A cm2.
SEPV Separator vessels valve variables
RECO Anode side recirculation circuit
ak $z$F$Eact;k ð1 ak Þ$z$F$Eact;k
Jcell ¼ Jk exp exp (16) variables
R$T R$T
RECH Cathode side recirculation circuit
A second hypothesis is to consider the charge transfer variables
coefficient equal to 0.5 (symmetry in the processes of the SPVA Stack pressure valve variables
electronic transfer, charge equally distributed, classical PURI Purification system variables
PCVA Pressure control valve variables
assumption in most of the models). Therefore, equation (17) is
EXHA Exhaust valve variables
obtained to compute activation overvoltages [40,41,45,46]. SAFE Safe valve variables
Other Butler-Volmer equation derivatives can be found but OUTH Hydrogen output variables
whether the equations obtained cannot be used for small OUTO Oxygen output variables
current densities (Tafel relations [23,24,27,29,42,47e49]), or ENCL Enclosure variables
they are obtained with faulty hypotheses or mathematical COOL Cooling circuit variables
COLD Cold tank variables
development [21,26,28,39,43,44,50].
EXCH Heat exchanger variables
Fig. 5 e Word bond graph of the PEM electrolysis system considered.
Fig. 6 e Bond graph electrochemical model of the PEM electrolysis cell.
Equation (17) is the intrinsic equation of the non-linear RS- With LPEM the length of proton exchange membrane, APEM
elements Ract;ano and Ract;cat .The current exchange density pa- its section, and Rothers the ohmic resistance of the cell expect
rameters for each electrode are fitting parameters of the membrane.
model, and the cell temperature is given by the stack thermal An empiric correlation (equation (20)) is used to determine
model. the conductivity of proton exchange membrane sPEM
depending on cell temperature Tcell and water content of the
R$T 1 Jcell
Eact;k ¼ sinh (17) membrane l. It was first developed in Ref. [51] and used in
F 2$Jk
many PEM electrolysis models [21,24,26e28,39,42,45,48,50].
Others ohmic contributions are aggregated into a unique
Ohmic overvoltage. The ohmic overvoltage is caused by
resistance Rothers which value is considered a fitting parameter
non-infinite conductivity of electrolysis cells. The different
of the model.
sources of ohmic overvoltage are electrolyte, electrodes, bi-
polar plates, contact resistances, etc. The equation is given by
1 1
sPEM ¼ 0:005139$l 0:0326 1268 (20)
R-element Rohm . Following the causality of the BG model, it 303 Tcell
enables to compute the cell electric current Icell depending of
Finally, the global equation of the stack polarization curve
the ohmic resistance Rohm of the cell and the ohmic over-
of the stack electrochemical model can be written (21).
voltage Eohm (equation (18)), which is given by equation (4).
pffiffiffiffiffiffiffi!
Eohm R$Tcell PH2 $ PO2 R$Tcell 1 Jcell
Icell ¼ (18) Ecell ¼ þ
E0rev ln þ sinh
Rohm 2$F aH2 O F 2$Jano

The ohmic resistance of the cell is composed of the resis- R$Tcell 1 Jcell LPEM
þ sinh þ þ Rothers $Icell (21)
tance of the proton exchange membrane and the other ohmic F 2$Jcat sPEM $APEM
contributions (equation (19)).
Thermochemical model. The global bond graph thermo-
LPEM electrochemical model of the electrolysis stack in Fig. 7.
Rohm ¼ þ Rothers (19)
sPEM $APEM
Fig. 7 e Global bond graph thermo-electrochemical model of the PEM electrolysis cell.
Tf-elements of the thermochemical model enable to is considered for the stack which is assumed to be constant.
compute mass flows m_ i produced and consumed by the As a result, the stack temperature is assumed to be homoge-
reaction. neous. It is assumed the water evaporation has a negligible
impact on stack temperature evolution (indeed, the nominal
Icell
m_ i ¼ ni $Mi $x_ ¼ ni $Mi $ (22) temperature of the experimental device is 55 C, and the water
2$F
evaporation was proved to have a negligible impact on the
With Mi the molar mass of species i, ni the stoichiometry stack thermal behaviour at this temperature). Considering the
coefficient of species i, x_ the reaction speed flow rate, Icell the different inputeoutput enthalpy flow on electrolysis stack
electric current of the cell and F the Faraday constant. (Fig. 8) the resulting BG model is exposed in Fig. 9.
From 0-junction equation (23) is deduced.
Thermal model. Temperature has a beneficial impact on the Z
1
value of reversible potential. Moreover, as activation, ionic TSTACK ¼ HRECH þ H_ RECO H_ CAT H_ ANO þ Q_ STA þ Q_ ENTH
diffusion, electronic diffusion and mass transfer are thermal Cp
activated phenomena, the temperature has a beneficial þ Q_ ENTR Q_ ATM $dt (23)
impact on the global cell voltage [52,53]. To capture its influ-
ence, a thermal model was developed and coupled to the The enthalpy flows related to input and output flow of
electrochemical model (Figs. 7, 9). matter both at anode and cathode side (H_ RECH , H_ RECO , H_ CAT and
According to general hypotheses, the model involves or- H_ ANO ) depend on the molar flow n_i and the molar enthalpy Hi of
dinary differential equations. A lumped thermal capacitance each species involved.
Fig. 8 e Input and output enthalpy flows of a PEM electrolysis stack.
Fig. 9 e Bond graph thermal model of the PEM electrolysis cell.
H_ RECH þ H_ RECO H_ CAT H_ ANO ¼ n_RECH RECH

_RECO RECO
H2 O $HH2 O þ nH2 O $HH2 O Q_ ENTH ¼
Icell
HH2 þ 0:5HO2 HH2 O (32)
2$F
n_H2 O $HCAT n_H2 $HCAT
CAT CAT
H2 O H2
n_O2 $HCAT _
CAT ANO
O2 nH2 O $HH2 O
ANO
Icell
Q_ ENTH ¼ $DrH (33)
2$F
n_ANO _ANO ANO
H2 $HH2 nO2 $HO2
ANO
(24)
The thermal energy exchanged in the stack during the
For each speciesi, the molar enthalpy depends on the
electrochemical reaction i.e. the reversible entropy heat is
temperature and the heat capacity.
integrated into the heat flow source Q_ ENTR . It is modulated by
the electric current of the cell. This term enables to take into
Hi ¼ H0i þ Cpi $Ti (25)
account the endothermic nature of the water electrolysis
Assuming the output temperatures are both equal to stack reaction.
temperature, species balances at steady-state enable to make
appear standard enthalpy of reaction. Q_ ENTR ¼ n_H2 O SH2 O n_H2 $SH2 n_O2 SO2 þ n_H2 O SH2 O (34)
H_ RECH þ H_ RECO H_ CAT H_ ANO ¼ n_H2 O $CpH2 O $ðTRECH TSTACK Þ

RECH
Icell
Q_ ENTR ¼ SH2 O SH2 0:5$SO2 (35)
þ n_RECO
H2 O $CpH2 O $ðTRECO TSTACK Þ
2$F
I
$ HH2 þ 0:5HO2 HH2 O Icell
2:F Q_ ENTR ¼ DrS (36)
2$F
(26)
The constitutive equation of R-element in the thermal BG
H_ RECH þ H_ RECO H_ CAT H_ ANO ¼ n_RECH model enables to compute the losses term Q_ ATM from the stack
H2 O $CpH2 O $ðTRECH TSTACK Þ
to the system enclosure.
þ n_H2 O $CpH2 O $ðTRECO TSTACK Þ
RECO
I TSTACK TENCL
$DrH Q_ ATM ¼ (37)
2$F Rth cell
(27)
With TENCL the system enclosure temperature and Rth cell the
The heat production term from irreversibilities Q_ STA is heat transfer coefficient of the cells.
deduced from RS-elements (both resistive element entropy By recombining equations (27), (30), (33), (36) and (37), the
source element) of the bond graph electrochemical model. following thermal balance is obtained.
Q_ STA ¼ Q_ ohm þ Q_ act;ano þ Q_ act;cat (28)
0 1
I
n_H2 O $CpH2 O $ðTRECH TSTACK Þ þ n_H2 O $CpH2 O $ðTRECO TSTACK Þ
RECH RECO
Z $DrH
1 B 2$F C
TSTACK ¼ B C$dt (38)
Cp @ I I TSTACK TENCL A
þðEcell Erev Þ$I þ $DrH TSTACK $DrS
2$F 2$F Rth cell
Z
1
Q_ STA ¼ Eohm $Icell þ Eact;ano $Icell þ Eact;cat $Icell (29) TSTACK ¼
Cp
0 RECH 1
nH2 O $CpH2 O $ðTRECH TSTACK Þ þ n_RECO
Q_ STA ¼ ðEcell Erev Þ$Icell (30) B C
B C$dt
@ DrG þ T$DrS TSTACK TENCL A
Otherwise, thermal coupling with reaction is difficult to þI$ Ecell
2$F Rth cell
represent in bond graph. Therefore, two heat flow sources are
(39)
integrated in order to represent (i) the thermal energy trans-
ported by the different species consumed and produced dur- Z
1
ing the reaction, and (ii) the endothermic nature of the TSTACK ¼
Cp
reaction. The reaction term Q_ ENTH which take into account the 0 RECH 1
consumption of water and the production of oxygen and n_H2 O $CpH2 O $ðTRECH TSTACK Þ þ n_RECO
B C
B C$dt
hydrogen is obtained by writing the enthalpy flows of the @ DrH TSTACK TENCL A
different species involved in the reaction. It appears in the BG þI$ Ecell
2$F Rth cell
model as a heat flow source modulated by the electric current (40)
of the cell.
In equation (40), the thermoneutral voltage Etn appears.
Q_ ENTH ¼ n_H2 $HH2 þ n_O2 HO2 n_H2 O HH2 O (31)
DrH
Etn ¼ (41)
2$F
Finally, equation (42) is obtained. It has the same structure

as thermal balances found in the PEM electrolysis models m_ ANO ¼ m_ RECO þ m_ O2 prod þ m_ H2 diff þ m_ H2 O diff m_ H2 O cons m_ O2 diff
reviewed [22e26]. The heat capacity of the stack and its heat
m_ H2 O eos
transfer coefficient are fitting parameters of the model.
(44)
Z
1
TSTACK ¼ With m_ i diff the diffusion flow though the proton ex-
Cp
0 RECH 1 change membrane of species i, m_ i prod the production flow of
n_ $CpH2 O $ðTRECH TSTACK Þ þ n_H2 O $CpH2 O $ðTRECO TSTACK Þ
RECO
B H2 O C hydrogen or oxygen, m_ H2 O cons the water mass flow consumed
B C$dt by the reaction and m_ H2 O eos the water mass flow by electro-
@ TSTACK TENCL A
þI$ðEcell Etn Þ osmosis.
Rth cell
(42) As shown before, the production and consumption terms
are deduced directly from BG electrochemical model of the
Mass transfer and fluidic model. Fluidic and mass transfer stack (i.e. Faraday's law).
phenomena in a PEM electrolysis cell cannot be ignored as
Icell
they have impacts on cell efficiency and global behaviour of m_ i ¼ ni Mi (45)
2$F
the whole system. As a consequence, a mass transfer and
fluidic model was built and coupled to electrochemical and With Icell the stack current and Mi the molar mass of
thermal models presented before. Three ancillaries are species i.
considered: the anode side, the cathode side and the mem- The diffusion terms m_ i diff for each species i are given by BG
brane. No fluidic accumulation is considered in the stack. The resistive elements: they are proportional to the partial pres-
gases produced are assumed to be continuously, instanta- sure difference between the two sides of the PEM membrane
neously and homogenously evacuated from reactive sites to DPi [16].
channels and from channels to stack outputs. Thus, the mass DPi
transfers considered in this model are the ones trough the m_ i diff ¼ (46)
Ri diff
PEM membrane: (i) water transfer from anode to cathode by
electro-osmosis and from cathode to anode by diffusion; (ii) The coefficient Ri diff used in equation (46) depends on the
hydrogen and oxygen molecules transfer respectively from Henry's constant Hi of species i, the diffusion coefficients Di
cathode to anode and vice versa due to their non-zero solu- and the length of the PEM membrane LPEM [16]. Henry's con-
bility and mobility in water and proton exchange membrane. stants and diffusion coefficients values are deduced from
The resulting fluidic BG model of the stack is exposed in literature [54].
Fig. 10.
Hi
The constitute equation of 0-junctions enables to write: Ri diff ¼ $LPEM (47)
Di
m_ CAT ¼ m_ RECH þ m_ H2 prod þ m_ O2 diff þ m_ H2 O eos m_ H2 diff m_ H2 O diff Finally, electro-osmotic flow is due to water molecules
(43) dragged by protons which migrate from anode to cathode
through the PEM membrane. Electro osmotic drag does not
Fig. 10 e Fluidic bond graph model of the PEM electrolysis cell.
depend on the membrane thickness and seems to be the most Converter

important phenomenon involved in water transfer across the The converter enables to adapt the input power to the stack
membrane. It depends on the electric current being passed currentevoltage characteristics and to control its voltage (i.e.
through the electrolysis and the number of water molecules its operating point). Regarding the dynamic of electrical phe-
dragged by one proton. The number is called the electro- nomena in a converter (time response less than 0.1 s), these
osmosis coefficient neos . The electro-osmosis term is given by phenomena are considered instantaneous. Therefore, the
a BG resistive element modulated by electric current of the converter is represented with a transformer element and a
cell. This term is expressed in equation (48). Values from 0.27 resistive element to take into account its electrical efficiency
[28] to 5 [50] were found in the literature for the electro- (Fig. 12).
osmosis coefficient. In the present model, its value was fixed Constitutive equations of this model are equations
to 3. (49)e(51).
I Estack ¼ a$Ein (49)

m_ H2 O eos ¼ neos MH O (48)
2$F 2
Pin ¼ Pstack þ Pconv (50)
Global bond graph model. The global bond graph model of the
stack is presented in Fig. 11. Even if they are actually inte-
Pconv ¼ Pin $ð1 hconv Þ (51)
grated, mass transfers trough the PEM membrane are not
represented in order to improve readability of the BG model. With Ein the input voltage, a the transformer coefficient of
Once the BG stack model is established, the same modelling the converter, Pin the input power, Pstack the stack power, Pconv the
process is used to build models for all other components of the power losses by the converter and hconv the converter efficiency.
system identified in the word bond graph (Fig. 5): converter, Even if the electric phenomena in the converter are
separator vessels, recirculation circuits (with exchanger and considered instantaneous, it is necessary to take into ac-
cooling circuit on the oxygen side), purification unit and count the fact its efficiency is not equal to one. In order to
valves. describe its efficiency on the whole operating range of the
Fig. 11 e Simplified global bond graph of the PEM electrolysis cell.
Fig. 12 e Bond graph model of the converter.
converter, a particular model was used [30,55]. The electrical With m_ SEPO the output water mass flow, m_ SEPV the water
efficiency of the converter is then deduced from equations mass flow from separator valve and m_ RES the water mass flow
(52)e(54). from water tank.
hy
The hydraulic capacitance CSEPO is expressed in equation (59).
Ps =Pnom
h¼ (52)
Ps =Pnom þ n0 þ m$ðPs =Pnom Þ2 hy ASEPO
CSEPO ¼ (60)
g
10=h10 þ 1=h100 9 With ASEPO the section of the vessel andg the gravitational
n0 ¼ (53)
99 constant.
Furthermore, the water pressure is proportional to water
1
m¼ n0 1 (54) level LSEPO in the separator vessel, the section of the vessel and
h10
the density of water rH2 O .
With Ps the output power of the modelled converter, Pnom its
nominal power, h100 its efficiency at nominal power and h10 its PSEPO ¼ rH2 O $g$LSEPO (61)
efficiency at 10% of its nominal power. Therefore, equations (62) and (63) are obtained.
Z
Separators vessels g
rH2 O $g$LSEPO ¼ xliq $m_ ano þ m_ SEPV þ m_ RES m_ SEPO $dt (62)
The purpose of separator vessels is to separate gas and liquid ASEPO
phases from stack output biphasic flows. For each stack, there Z
are two separator vessels, one for each electrode side. Some 1
LSEPO ¼ xliq $m_ ano þ m_ SEPV þ m_ RES m_ SEPO $dt (63)
hypotheses are taken in order to model these components. ASEPO $rH2 O
The vessels are assumed to be perfectly cylindrical to compute The water mass flows m_ ano , m_ SEPO and m_ SEPV are given
water levels. In accordance with general hypotheses, a lum- respectively by stack mode, anode side recirculation circuit
ped thermal capacitance is considered. Its value is not con- model and separator valve model. The water mass flow from
stant and it is assumed to depend on the water level but not on water tank is computed through the resistive element RRES .
the gas mass balance (because the contribution of gases in This element represents the water tank valve which is
thermal capacitance is negligible). Finally, the heat transfer controlled by the water level in the vessel [56].
coefficient is assumed to be constant. The resulting BG models
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
are exposed in Figs. 13 and 14. PRES PSEPO
m_ RES ¼ (64)
RRES
Fluidic part. The transformer elements xliq and xgaz used in this By recombining equations (63) and (64), equation (65) is
model are not power conservative. Indeed, we used a math- obtained.
ematical artifice to obtain equations (55)e(58) for these
Z pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
particular transformer elements. 1 PRES PSEPO
LSEPO ¼ xliq $m_ ano þ m_ SEPV þ
ASEPO $rH2 O RRES
m_ liq ¼ m_ ano $xliq (55)
m_ SEPO $dt (65)
m_ gaz ¼ m_ ano $xgaz (56)
Chemical part. In the same way, from 0-junction in chemical
m_ ano ¼ m_ liq þ m_ gaz (57) part, equation (66) is obtained.
Z pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Pano ¼ Pliq ¼ Pgaz 1 PSEPO POXYG
(58) PiSEPO ¼ m_ ANO $xgas $xi xi $dt (66)
Cchem
SEPO Rsep ano
From 0-Junction in fluidic part, equation (59) is deduced.
With xi the mass fraction of gas i, xgas the mass fraction of
Z
1 gas in the anode output biphasic flow,Rsep ano the pneumatic
PSEPO ¼ hy
m_ liq þ m_ SEPV þ m_ RES m_ SEPO $dt (59)
CSEPO
Fig. 13 e Bond graph model of the oxygen separator vessel.
resistance of the oxygen separator vessel and POXYG the pres- Thermal part. From 0-junction in thermal part, the thermal
sure of the oxygen circuit. balance of the oxygen separator vessel is exposed in equation
Following the assumption that all gases considered are (70).
ideal, the capacitance is expressed in equation (67). Z
1
TSEPO ¼ H_ SEPV þ H_ ANO þ H_ RES H_ SEPO H_ OXYG Q_ ENC $dt
Mi $Vgas Cth
SEPO ¼
Cchem SEPO
(67)
R$TSEPO (70)
The volume available for the gas Vgas is deduced from the With the enthalpy flow from separator valve H_ SEPV , the
water level in the separator vessel and equation (69) is enthalpy flow from anode H_ ANO , the water output enthalpy
obtained. flow H_ SEPO respectively given by the separator valve model, the
stack model, and the anode side recirculation circuit model.
Vgas ¼ VSEPO LSEPO $ASEPO (68)
The expression of enthalpy flow from water tank H_ RES ,
Z gases output enthalpy flow H_ OXYG and are respectively given
R$TSEPO
PiSEPO ¼ m_ ANO $xgas $xi by equations (71) and (72) [56].
Mi $ðVSEPO LSEPO $ASEPO Þ
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
PSEPO POXYG H_ RES ¼ m_ RES $CpH2 O $TRES (71)
xi $dt (69)
Rsep ano
Fig. 14 e Bond graph model of the hydrogen separator vessel.
X 0 1
H_ OXYG ¼ m_ OXYG $TOXYG $ xi $Cpi (72) Z H_ SEPV þ H_ ANO H_ SEPO þ m_ RES $CpH2 O $TRES
1 B pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi C
i
TSEPO ¼ B X TSEPO TENCL C
Cth @ PSEPO POXYG A$dt
The thermal losses Q_ ENC are linked to the resistive element SEPO $TSEPO $ xi $Cpi
Rsep ano i
Rcond sep ano
Rcond sep ano , and are computed through equation (73).
(74)
TSEPO TENCL
Q_ ENC ¼ (73) The exactly same modelling process for hydrogen sepa-
Rcond sep ano
rator vessel enables to write equations (75)e(77) obtained from
With Rcond sep ano the heat resistance of the oxygen separator the BG model exposed in Fig. 14.
vessel and TENCL the temperature of the system enclosure Z pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1 PSEPH PSEPO
given by the thermal model of the enclosure. LSEPH ¼ m_ CAT $xliq m_ SEPH $dt
rH2 O $ASEPH Rvalve sep
By recombining equations (70)e(73), equation (74) is
obtained. (75)
Recirculation circuits and cooling circuit

Z Recirculation circuits enable to feed stacks with water. On the
R$TSEPH
PiSEPH ¼ m_ CAT $xgas $xi anode side, a heat exchanger with its own cooling circuit is
Mi $ðVSEPH LSEPH $ASEPH Þ
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi integrated in order to control anode side input water tem-
PSEPH PHYDR
xi $dt (76) perature and consequently stack temperature. Regarding the
Rsep cat
modelling of these components, some hypotheses are taken.
0 pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 1 The pumps are considered perfect mass flow sources and
H_ CAT H_ SEPH PSEPH PSEPO $CpH O $TSEPH
Z B Rvalve sep 2 C neither fluidic inertia nor accumulation is considered. Cooling
1 B C
TSEPH ¼ th B pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi C$dt circuit is assumed to be thermally insulated. The cooling unit
CSEPH B@ SEPH
P P HYDR
X TSEPH T C
ENCL A
$TSEPH $ xi $Cpi is considered a perfect heat flow source and its coefficient of
Rsep cat Rcond sep cat
i performance is assumed to be constant. In accordance with
(77) general hypotheses, lumped thermal capacitances are
With ASEPH the section of the hydrogen separator vessel, considered. The resulting bond graph models of anode side
VSEPH its volume, Rsep cat its pneumatic resistance, Rcond sep cat its recirculation circuit and cathode recirculation circuit are
the heat resistance, Rvalve sep the hydraulic resistance of the respectively exposed in Figs. 15 and 16.
separator valve which is controlled by the water level in the From 0-junction of the anode side recirculation circuit,
hydrogen separator vessel, m_ CAT and H_ CAT respectively the equation (78) is obtained.
mass flow and the enthalpy flow from the cathode given by Z
1
the stack model, m_ SEPH and H_ SEPH respectively the output water TRECO ¼ H_ SEPO H_ RECO Q_ COOL Q_ ENC $dt (78)
Cth
RECO
mass flow and the output enthalpy flow given by the cathode
recirculation side model PHYDR the pressure of the hydrogen H_ SEPO and H_ RECO are computed respectively with equations
circuit given by the purification unit model and TENCL the (79) and (80) [56].
temperature of the system enclosure given by the thermal
H_ SEPO ¼ m_ RECO $CpH2 O $TSEPO (79)
model of the enclosure.
Fig. 15 e Bond graph model of the anode side recirculation circuit and the cooling circuit.
Fig. 16 e Bond graph model of the cathode side recirculation circuit.
Following the same process for other 0-junctions, thermal

H_ RECO ¼ m_ RECO $CpH2 O $TRECO (80) balances for the cooling circuit are expressed in equations (83)
and (84).
The thermal losses Q_ ENC are described through a BG resis-
Z
tive element, and equation (81) is deduced. 1
TCOOL ¼ m_ COOL $CpH2 O $ðTCOLD TCOOL Þ þ Q_ COOL $dt (83)
Cth
COOL
TRECO TENCL
Q_ ENC ¼ (81) Z
Rcond recir ano 1
TCOLD ¼ m_ COOL $CpH2 O $ðTCOOL TCOLD Þ þ Q_ COLD $dt (84)
Finally, the Number of Transfer Unit (NTU) method was Cth
COLD
used to compute the exchanged heat flow Q_ COOL in function of

With Cth th
COOL and CCOLD respectively the heat capacities of the
mass flows on both sides of the heat exchanger. The resulting
cooling circuit and the cooling tank, and m_ COOL the mass flow
thermal balance is exposed in equation (82).
of the coolant liquid.
Z In the same way, for the cathode side recirculation circuit,
1 TRECO TENCL
TRECO ¼ m_ RECO $CpH2 O $ðTSEPO TRECO Þ the BG model is exposed in Fig. 16 and equation (85) is
Cth
RECO Rcond recir ano
obtained.
Q_ COOL $dt
Z
1 TRECH TENCL
(82) TRECH ¼ m_ RECH $CpH2 O $ðTSEPH TRECH Þ $dt
Cth
RECH Rcond recir cat
With Cth
RECO the heat capacity of the anode side recirculation (85)
circuit, Rcond recir ano its thermal resistance, TSEPO the oxygen
With CthRECH the heat capacity of the cathode side recircula-
separator vessel temperature which is given by the oxygen
tion circuit, Rcond recir cat its heat resistance, TSEPH the hydrogen
separator vessel model and TENCL the temperature of the sys-
separator vessel temperature which is given by the hydrogen
tem enclosure given by the thermal model of the enclosure.
separator vessel model and TENCL the temperature of the sys- unit and m_ SPVA the mass flow from the stack pressure valve and
tem enclosure given by the thermal model of the enclosure. given by the stack pressure valve model.
The purification unit temperature TDRY is given by the
Purification unit thermal part of the BG model. From the 0-junction in the
The purification unit installed in the experimental device thermal part, the thermal balance is obtained and it is
consists only in a dryer. This component enables to adsorb expressed in equation (90).
water fraction contained in hydrogen. In accordance with Z pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1 PDRY PPURI
general hypotheses, a lumped thermal capacitance is H_ SPVA þ n_ads
H2 O
TDRY ¼ $DHads $TDRY $xi $Cpi
considered. The resulting BG model is expressed in Fig. 17. Cth
DRY RDRY

From the chemical part of the BG model, equations TDRY TENCL
$dt
(86)e(89) are obtained. Rcond dry
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi (90)
Z
R$TDRY PDRY PPURI
PHDRY
2
¼ m_ SPVA $xH2 xH2 $dt (86) With Cth
the heat capacity of the purification unit, Rcond dry
MH2 $VDRY RDRY DRY
its thermal resistance, DHads the enthalpy of the water
Z pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi adsorption reaction, H_ SPVA the enthalpy flow from the stack
R$TDRY PDRY PPURI
PODRY
2
¼ m_ SPVA $xO2 xO2 $dt (87) pressure valve and given by the stack pressure valve model
MO2 $VDRY RDRY
and TENCL the temperature of the system enclosure given by
Z pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi the thermal model of the enclosure.
R$TDRY PDRY PPURI
m_ SPVA $xH2 O xH2 O n_ads
H2 O H2 O
PDRY ¼ $dt
MH2 O $VDRY RDRY
Valves
(88)
The numerous valves included in the electrolysis system en-
ables to control water levels in the separator vessels or control
PDRY ¼ PODRY
2
þ PHDRY
2 H2 O
þ PDRY (89) gas pressure. The corresponding BG model is expressed in
With VDRY the volume of the purification unit, RDRY its Fig. 18.
pneumatic resistance, n_ads The constitutive equations of this model are equations
H2 O
the molar flow of water adsorbed in
the column which depends on the kinetic law selected for the (91)e(93) [56]. The resistive coefficients are not constant and
adsorption reaction, PPURI the output pressure of the purification depend on: (i) the opening of controlled valves; (ii) the
Fig. 17 e Bond graph model of the purification unit.
fans are considered perfect mass flow sources. The resulting

BG model is represented in Fig. 19.
From 0-junction, the thermal balance of the enclosure is
deduced and it is expressed in equation (94).
Z X
1
TENC ¼ Q_ m_ FANS $Cpair $ðTENC TATM Þ
Cth
ENC

TENC TATM
$dt (94)
Rcond enc
P _
With Q the sum of the thermal losses of the different
component modelled before, Cth ENC the heat capacity of the
enclosure, Cpair the air thermal capacity, m_ FANS the mass flow
given by the fans Rcond enc , the heat resistance of the enclosure
and TATM the atmospheric temperature.
Block diagram conversion

Fig. 18 e Bond graph model of the valves.
As a result of bond graph's structural and causal properties, a
dynamic model in a block diagrams format is deduced sys-
experimental laws m_ VALVE ¼ f ðDPÞ obtained for the different tematically from the graphical model using dedicated soft-
valves (when available). ware for simulation. Therefore, once the BG models were built
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi for each component of the system, they were transcribed into
Pin Pout block diagrams and implemented on Simulink. Thus, model
m_ VALVE ¼ (91)
RVALVE libraries were built for each component both under BG
formalism and block diagram format. The hardware part of
H_ VALVE ¼ m_ VALVE $CpH2 O $Tin (92) the model under block diagram format is represented in
Fig. 20.
H_ VALVE ¼ m_ VALVE $xi Cpi $Tin (93)
Control system
Enclosure
The system is conditioned into an enclosure. In order to obtain In order to obtain a complete representation of the dynamic
a good representation of the thermal behaviour of the whole behaviour of the PEM electrolysis system, the control system
system it is necessary to include a dynamic model of this of the experimental device was modelled in a block diagram
enclosure. Only the thermal part is considered, the fluidic part format in addition to the modelling of the hardware part
is assumed to be at steady-state. The enclosure is vented and (Fig. 20).
Fig. 19 e Bond graph model of the enclosure.
Fig. 20 e Global scheme of the Simulink interface.
Fig. 21 e Functional scheme of the control system implemented in the PEM electrolysis model.
The control system implemented in the model can be and thermal parameters). All these parameters were then
decomposed in 3 levels presented in the following subsections used with a dynamic operating profile. The model data ob-
[15,20] (Fig. 21). The first level is the mode control unit. It de- tained on this particular profile were compared to experi-
termines the operating mode of the system in function of the mental ones in order to check the accuracy of the model. As
system input (hydrogen flow production set point or available the experimental device is connected to the grid and
power) and the state of the system: ON, OFF, powering on, controlled by a hydrogen flow production set point, the two
powering off. The experimental device is controlled by a profiles used in this validation process (a nominal operating
hydrogen production flow set point and is connected to the profile and a dynamic operating profile) are hydrogen flow
grid. Therefore the power consumed by the system results production set point profiles.
from the hydrogen production flow set point of the system. On
the experimental device, the hydrogen flow production set Identification of electrochemical parameters
point is converted into a corresponding global current in-
tensity which has to be distributed between the two stacks Regarding the wide range of values of electrochemical pa-
installed. This repartition is ensured by the second level of rameters available in the literature, a fitting parameters pro-
control which is the power control unit: it ensures the repar- cess was achieved based on experimental results obtained
tition of the power in the system by defining the references with the experimental device presented before. The fitting
value for the regulated variables depending on the operating process relies on a Matlab function based on Lev-
mode and the power references given by the mode control enbergeMarquardt algorithm (nonlinear method) in order to
unit. minimize quadratic error between experimental and model
The third one is the automatic control unit which provides data.
command for the actuators based on the comparison of ref- pffiffiffiffiffiffiffi!
erences value and measured value of the regulated variables R$TSTACK PH2 $ PO2 R$TSTACK 1 Jcell
Ecell ¼ þ
E0rev ln þ sinh
of the system. The regulated variables, the control loops used 2$F aH2 O F 2$Jano

and the associated equipment controlled are summarized in R$TSTACK 1 Jcell LPEM
þ sinh þ þ Rothers $Icell
Table 5. F 2$Jcat sPEM $APEM
Finally, the switching control unit enables to define the (95)
transistor's ON-OFF state of the power converters based on the
The equation of the polarization curve integrated in the
module indexes given by the automatic control unit [15,20].
model is expressed in equation (95). The electrochemical pa-
Modulation frequencies of such control units (from one to
rameters fitted are the equivalent electric resistance Rothers ,
dozen kHz) are not consistent with the representation of the
corresponding to all ohmic contributions in the cell excepting
dynamic behaviour of the system over long-term periods. As a
the proton exchange membrane, and the current exchange
result, it is not integrated in the model.
density at each electrode (Jcat and Jano ). As temperature has an
impact on electrochemical phenomena and therefore on the
polarization curve of the electrolysis cells, the fitting process
Results and discussion
was performed for different temperatures. The results for
these three parameters are exposed in Table 6.
The model described in previous part was developed based on
the architecture of a semi-industrial PEM electrolysis system.
This experimental device was presented before and its tech-
nical specifications are summarized in Table 3. It is a semi- Table 6 e Results of fitting procedure for the
industrial PEM electrolysis system which has design specific- electrochemical parameters at different temperature.
ities of high-power systems. Validation of the model was then Temperature (K) Req (ohm) Jano (A cm2) Jcat (A cm2)
possible by confrontation of simulation results with experi- 3 7
298.15 1,80$10 2,27$10 4,90$102
mental results. The validation process is exposed in the 303.15 1,70$103 3,64$107 6,35$102
following subsections. At first, electrochemical parameters of 308.15 1,60$103 5,54$107 7,63$102
the stack are identified. Then, experimental data from a 313.15 1,50$103 8,08$107 8,69$102
nominal operating profile were used in order to obtain and 318.15 1,40$103 1,13$106 9,60$102
identify values of others model parameters (electric, fluidic 323.15 1,30$103 1,53$106 1,05$101
Table 5 e Regulated variables, control loop and associated equipment controlled in the PEM electrolysis system.
Regulated variable Loop control system Control loop Equipment controlled
Stack voltage Closed PID Stack converter
Stack Temperature Closed ONeOFF Control Cooling unit and Cooling pump
Water level in oxygen separator vessel Closed ONeOFF Control Water tank valve
Water level in hydrogen separator vessel Closed ONeOFF Control Separator valve
Mass flow in oxygen side recirculation circuit Open x Oxygen side recirculation pump
Mass flow in hydrogen side recirculation circuit Open x Hydrogen side recirculation pump
Mass flow in cooling circuit Open x Cooling pump
It was reported in the literature the correlation exposed in profile, the start-up phase of the system is achieved: the
equation (96) for the current exchange densities depending on stacks produce hydrogen which enable to pressurize the sys-
the temperature [40]. Then, the current exchange densities tem. During this phase, no hydrogen is available at system
values obtained for each electrode at different temperatures output. After a pressure test during which the stacks stop
(Table 6) are injected in equation (96) in order to evaluate the producing hydrogen in order to detect any hydrogen leak, a
standard current exchange density J0k at electrode k at tem- hydrogen flow is available at system output. The system is
perature T0 and the activation energy Eak of the half-reaction then operating at its nominal hydrogen flow production set
occurring at electrode k. point: the current densities of the stacks are kept at their
nominal values. The identification process of model parame-
1 1
Jk ¼ J0k $exp Eak $ (96) ters is addressed by physical domain, as each physical domain
T T0
has been validated separately.
The values obtained for these two parameters are exposed
in Table 7 as well as determination coefficient for each elec- Electrical parameters
trode. The values of current exchange densities exposed in Once the electrical behaviour of the stack is well described, it
Table 7 are found to be in agreement with those found in the is necessary to characterize the consumption of auxiliaries’
literature (Table 8). electrical components of the system (power consumed by the
For the equivalent resistance, a linear function of temper- programmable logical controller PPLC , the power consumed by
ature was obtained with a great determination coefficient stack fans Pstack fans and system fans Psystem fans , the power
(R2 ¼ 0.9999). The empiric correlation obtained is exposed in consumed by anode recirculation pump Ppump ano , cathode
equation (97). recirculation pump Ppump cat and cooling pump Ppump cool ). This
characterization was achieved with component datasheets
Req ¼ 2:105 $T þ 0:0078 (97) when available and experimental measurements.
Finally, the comparison between experimental and In order to validate the electrical part of the model,
modelled polarization curves at different temperature is rep- experimental and model data regarding system power were
resented in Fig. 22. The experimental polarization curves were compared on about 6000 points, for each second of the nom-
obtained with a data processing of raw current-voltage- inal profile. The results are exposed in Fig. 23. As explained in
temperature measurements (indeed the experimental device section “Control system”, the experimental device is
does not enable to control stack temperature except above controlled by a hydrogen production flow set point and is
55 C). connected to the grid. Therefore, the system power is not the
A great correlation between experimental and model data power available in a hypothetical stand-alone system, but
is obtained which is approved by the values of statistical co- results from the hydrogen production flow set point and the
efficients exposed in Table 9. operating conditions of the electrolysis system.
A good agreement between experimental and model data
Identification of model parameters is obtained (Table 11). The gap observed from 3000 to 5500 s
comes from the power consumed by the air conditioning
The identification of model parameters process is performed related to the regulation of the electrical enclosure tempera-
with a nominal operating profile. During the first 1000 s of this ture which is not integrated in the model.
Fluidic parameters
The fluidic parameters of the model are summarised in Table 10.
Table 7 e Values for standard exchange current density Some of these parameters are deduced from components
and activation energy obtained for each electrode. datasheets such as mass flows delivered by the different
J0 (A cm2) Ea (J mol1) R2 pumps and fans, and the different volumes and areas in the
Anode 2,00$107 6,11$104 0,9966 system. The other ones, all fluidic resistance values, are
Cathode 0,0447 2,38$104 0,9639 determined based on experimental results.
Table 8 e Values of current exchange densities found in the literature.

Authors Temperature (K) Anode (A cm2) Cathode (A cm2)
Biaku et al. [57] [293; 333] [4$103;5$104] [0.287; 0.391]
Choi et al. [41] 353 1012 (Pt), 107 (PteIr) 103
Han et al. [58] 353 2$106 101
Chandesris et al. [54] [313; 353] [1$107; 2,36$106] x
Abdin et al. [39] [283; 363] 107 101
Agbli et al. [21] [300,11; 302,19] 0.1548$102 0.3539.101
Dale et al. [43] [283; 333] [0,76$105; 4,93$105] [0,18; 0,39]
Sawada et al. [46] [323; 353] [5$109; 1,1$103] x
Marangio et al. [28] [313; 328] 107 103
Harrisson et al. [59] Not mentioned 1.65$108 9.102
Ito et al. [60] 353 [0,85$106; 1,70$106] x
Fig. 22 e Comparison between experimental and modelled polarization curves for different temperatures.
Table 9 e Statistical data about comparison of experimental and electrochemical model data.
Temperature ( C) 25 30 35 40 45 50
Correlation coefficient 0,9998 0,9997 0,9997 0,9997 0,9997 0,9998
Mean square error (MSE) 1,20$104 5,93$105 4,09$105 3,86$105 4,66$105 6,73$105
35.00
30.00
25.00
System Power (kW)
20.00
15.00 Experimental data

Model data
10.00
5.00
0.00
0.00 2000.00 4000.00 6000.00
Time (s)
Fig. 23 e Comparison between experimental and modelled system power on a nominal profile.
The fluidic variables are the following: stack pressure, Nevertheless, the experimental data available are limited
system pressure, output hydrogen flow and water levels in the by the sensors installed on the experimental device: no data is
separator vessels. The pressure difference between stack part available regarding the water level in the separator vessels.
and system part enables to determine stack pressure valve. In Therefore, a specific methodology was used to identify some
the same way, pressure control valve fluidic resistance is fluidic parameters. It relies on some characteristic value of the
determined by plotting the difference between system pres- experimental device linked to hydrogen separator vessel
sure and output pressure in function of the hydrogen output drain. Indeed, due to electro-osmosis drag, the water level in
mass flow. hydrogen separator vessel increases. This level is regulated by
the opening of the separator valve which enables to drain

Table 10 e Fluidic parameters of the model.
periodically the excess of water in the vessel. The opening of
Fluidic resistances Tank valve RRES the separator valve is linked to water level thresholds. This
(Pa s m3) Separator valve Rvalve sep
drain leads to system pressure drop and output hydrogen flow
Stack pressure valve Rvalve stack
drops. These characteristic values are the following:
System pressure valve Rvalve system
Pressure control valve Rvalve pressure
Area (m2) Separator vessels ASEPO and ASEPH - The frequency of the drains which enables to deduce the
Volumes (m3) Separator vessels VSEPO and VSEPH difference between the two water level thresholds;
Purification unit VDRY - The duration of the drains which allows obtaining the
Mass flows (kg s1) Pumps m_ RECO and m_ RECH value of the fluidic resistance of the separator valve. The
and m_ COOL
same value was used for the fluidic resistance of the water
Fans m_ FANS
tank valve.
16.00
14.00
12.00
Pressure (bar)
10.00
8.00 System Pressure
experimental data
6.00
System pressure
4.00 model data
2.00
0.00
0.00 2000.00 4000.00 6000.00
Time (s)
Fig. 24 e Comparison between experimental and modelled system pressure on a nominal profile.
7.00
6.00
Hydrogen flow (Nm3.h-1)
5.00
4.00
Hydrogen flow
3.00 experimental data
2.00 Hydrogen flow model

data
1.00
0.00
0.00 2000.00 4000.00 6000.00
Time (s)
Fig. 25 e Comparison between experimental and modelled output hydrogen flow on a nominal profile.
Table 11 e Statistical data about comparison of experimental and model data regarding fluidic behaviour on a nominal
profile.
System power System pressure H2 flow Stack 1 pressure Stack 2 pressure
Correlation coefficient 0,9896 0,9962 0,9548 0,9929 0,9909
MSE 6,98$101 1,22$101 3,03$101 1,28$101 1,76$101
% error 2,9% 3,4% 5,2% 2,1% 2,4%
Thermal parameters
Table 12 e Thermal parameters of the model.
Thermal parameters of the model are summarized in Table 12.
Heat capacities Stack Ccell All these parameters have to be determined experimen-
Separator vessels CSEPO and CSEPH
tally. They were estimated through: (i) a previous identifica-
Recirculation circuits CRECO and CRECH
Cooling circuit CCOOL
tion process on the stack only, which had enabled to obtain
Cold tank CCOLD stack thermal parameters; (ii) the manufacturer data
Purification unit CDRY regarding the materials of each components in order to eval-
Enclosure CENC uate thermal capacitances; (iii) the individual thermal
Heat transfer Stack Rth cell behaviour of each component with the experimental data. As
coefficients Separator vessels Rcond sep ano and Rcond sep cat
a result, a good agreement between experimental and model
Recirculation circuits Rcond recir ano and Rcond recir cat
data regarding stack temperatures and enclosure temperature
Heat exchanger Rech
Purification unit Rcond dry is obtained (Fig. 26 and Table 13).
Enclosure Rcond enc
Balance on the efficiency of the system
Once the accuracy of the model regarding the description of
the dynamic behaviour of the system for each physical
With the same nominal profile than the one used for the domain is proved, the global model validation has to be
validation of electrical behaviour of the system, compari- checked through energy balances. The good agreement be-
son between experimental and model data is exposed in tween experimental and model data regarding hydrogen
Fig. 24 for system pressure and in Fig. 25 for output produced, electrical energy consumed, and specific con-
hydrogen flow. sumption of the system are then exposed in Table 14.
The agreement between experimental and model data is
quite good (Table 11). The typical fluidic behaviour linked to Validation on a dynamic operating profile
periodic hydrogen separator vessel drains is well represented.
The differences come from the desynchronized drain of the A set of parameters was obtained through the fitting process
two vessels observed on the experimental data. achieved on a nominal operating profile. This set of
335.00
Atmospheric
330.00 temperature
325.00 Enclosure temperature
Temperature (K)
315.00 Enclosure temperature

model data
310.00
Stack 1 temperature
300.00 Stack 2 temperature
experimental data
295.00
0.00 2000.00 4000.00 6000.00 Stack temperature
model data
Time (s)
Fig. 26 e Comparison between experimental and modelled temperatures of the system on a nominal profile.
Table 13 e Statistical data about comparison of experimental and model data regarding thermal behaviour on a nominal
profile.
Enclosure temperature Stack 1 temperature Stack 2 temperature
Correlation coefficient 0,9921 0,9801 0,9719
MSE 1,87$101 6,94$101 4,51$101
% error 0,1% 0,2% 0,1%
Table 14 e Comparison between experimental and model data regarding efficiency of the system on a nominal profile.
Hydrogen produced (Nm3) Energy consumed (kWh) Specific consumption (kWh Nm3)
Experimental data 6,11 43,34 7,09
Model data 5,99 42,58 7,11
Fig. 27 e Comparison between experimental and modelled system power on a dynamic operating profile.
Fig. 28 e Comparison between experimental and modelled system pressure on a dynamic operating profile.
Fig. 29 e Comparison between experimental and modelled output hydrogen flow on a dynamic operating profile.
Fig. 30 e Comparison between experimental and modelled temperatures of the system on a dynamic operating profile.
Table 15 e Statistical data about comparison of experimental and model data on a dynamic operating profile.
System System H2 output Enclosure Stack 1 Stack 2
power pressure flow temperature temperature temperature
Correlation 0,9977 0,9853 0,9813 0,9692 0,9912 0,9702
coefficient
MSE 3,18$101 2,19$101 1,06$101 2,04 1,55 5,82
% error 2,9% 3,6% 14,0% 0,4% 0,3% 0,6%
regarding system pressure, (iii) Fig. 29 regarding hydrogen

Table 16 e Comparison between experimental and model
output flow, (iv) Fig. 30 regarding stacks and enclosure tem-
data regarding efficiency of the system on a dynamic
operating profile. peratures. For each of these variables, a good accuracy is ob-
tained which is approved by the statistical data exposed
Hydrogen Energy Specific
in Table 15. Finally, regarding energy balances, the compari-
produced consumed consumption
(Nm3) (kWh) (kWh Nm3) son between experimental and model data are exposed
in Table 16. The model is then proved to represent accurately
Experimental 19,15 143,66 7,50
data
the dynamic behaviour of a PEM electrolysis system.
Model data 19,13 143,75 7,51
Discussion and improvements
parameters is then used with a dynamic operating profile in To obtain a better accuracy of the model regarding the
order to check the accuracy of the parameters identification description of the dynamic behaviour of a PEM electrolysis
process and validate the model. As well as the nominal profile, system, some improvements have been identified:
the start-up phase of the system is achieved during the first
1000 s of this profile. The system is then operating at its - Not consider pumps, cooling units and fans as perfect flow
nominal hydrogen production flow set point during the next sources anymore and achieve the modelling of the elec-
4000 s. After that, the dynamic operation of the system comes trical circuit for each of these electrical components in
from the variations of the hydrogen production flow set point order to study more precisely the dynamic of the power
which is set so as to simulate a photovoltaic power supply of consumed by the system.
the electrolysis system (with this profile, as well as with the - Improve the modelling of the fluidic part of the system
nominal profile, the experimental device is connected to the especially the stack and the pumps to take into account the
grid and the power consumed by the electrolysis system re- issues related to bubbles and biphasic flow inside the
sults from its hydrogen production flow set point). electrolysis cells: covering of active surface, bubbles evac-
The comparison between experimental and model data is uation and impacts on species diffusion, etc.
carried out for each point of the profile (one point per second, - Develop the model with distributed parameters, especially
profile of about 25,000 s). As well as for the nominal profile, stack and enclosure models to obtain a better accuracy of
each physical domain is addressed separately. The compari- the dynamic thermal behaviour of the system. All these
son between experimental and model data are exposed in: (i) improvements will be studied and developed in further
Fig. 27 regarding the system electrical power, (ii) Fig. 28 work.
[5] Singh S, Jain S, Venkateswaran PS, Tiwari AK, Nouni MR,

Conclusions and perspectives Pandey JK, et al. Hydrogen: a sustainable fuel for future of the
transport sector. Renew Sustain Energy Rev 2015;51:623e33.
Following the identification of lacks and limits of literature http://dx.doi.org/10.1016/j.rser.2015.06.040.
regarding the system scale dynamic modelling of electrolysis [6] Gahleitner G. Hydrogen from renewable electricity: an
international review of power-to-gas pilot plants for
systems, a new PEM electrolysis system model was developed
stationary applications. Int J Hydrogen Energy
under bond graph modelling formalism. This model was then
2013;38:2039e61. http://dx.doi.org/10.1016/
transcribed into block diagrams and, after a fitting process, j.ijhydene.2012.12.010.
was proved to represent accurately the dynamic behaviour of [7] Veziro]lu TN, Barbir F. Hydrogen: the wonder fuel. Int J
an experimental PEM water electrolysis system of 25 kW. It Hydrogen Energy 1992;17:391e404. http://dx.doi.org/10.1016/
was shown in the introduction of this paper that most of 0360-3199(92)90183-W.
existing models are focused only on the description of the [8] Barbir F. Transition to renewable energy systems with
hydrogen as an energy carrier. Energy 2009;34:308e12. http://
electrochemical behaviour of the stack. Therefore, the great
dx.doi.org/10.1016/j.energy.2008.07.007.
innovative interest of this model is to represent an entire [9] Lund H, Østergaard PA, Stadler I. Towards 100% renewable
system including stacks and balance of plant components. energy systems. Appl Energy 2011;88:419e21. http://
Regarding the modelling of such complex system, ener- dx.doi.org/10.1016/j.apenergy.2010.10.013.
getic modelling approach are really suitable. Within this [10] Mathiesen BV, Lund H, Karlsson K. 100% Renewable energy
framework, the use of BG modelling formalism enabled to systems, climate mitigation and economic growth. Appl
build model libraries for each component. As a consequence, Energy 2011;88:488e501. http://dx.doi.org/10.1016/
j.apenergy.2010.03.001.
the model presented in this paper is flexible and versatile, able
[11] Barbir F. PEM electrolysis for production of hydrogen from
to represent any PEM electrolysis system from laboratory to renewable energy sources. Sol Energy 2005;78:661e9. http://
industrial scale with a minimum adaptation. Due to the dx.doi.org/10.1016/j.solener.2004.09.003.
scalable nature of the built model, numerous improvements [12] Briguglio N, Brunaccini G, Siracusano S, Randazzo N,
can then be made regarding modelling of some components Dispenza G, Ferraro M, et al. Design and testing of a compact
such as stack, pump and cooling unit. PEM electrolyzer system. Int J Hydrogen Energy
2013;38:11519e29. http://dx.doi.org/10.1016/
Furthermore, the BG is a graphical tool which provides the
j.ijhydene.2013.04.091.
easy reading of the model. Unlike other graphical modelling
[13] Bergen A, Pitt L, Rowe A, Wild P, Djilali N. Transient
approach such as Macroscopic Energetic Representation, its electrolyser response in a renewable-regenerative energy
structural properties enable to obtain data about observ- system. Int J Hydrogen Energy 2009;34:64e70. http://
ability, controllability, monitorability of the PEM electrolysis dx.doi.org/10.1016/j.ijhydene.2008.10.007.
system. As a result of the development of this PEM electrolysis [14] Zhou Tao, Francois Bruno. Modeling and control design of
system model under BG modelling formalism, the analysis of hydrogen production process for an active hydrogen/
wind hybrid power system. Int J Hydrogen Energy
these properties are being performed and the results obtained
2009;34:21e30. http://dx.doi.org/10.1016/j.ijhydene.2008.10.030.
will be the subject of further publications. [15] Zhou Tao. Commande et supervision e nerge tique d’un
Finally, the accuracy of the model regarding the repre- ge ne
rateur hybride actif e olien incluant du stockage sous
sentation of PEM electrolysis system enables to perform sim- forme d’hydroge ne et des super-condensateurs pour
ulations in order to analyse its dynamic behaviour and gration dans le syste
l’inte me electrique d’un micro re seau.
achieve some design sensibility analyses. The representation Ecole Centrale de Lille; 2009.
of the whole system enables to consider numerous design [16] Grigoriev SA, Porembskiy VI, Korobtsev SV, Fateev VN,
Aupre ^tre F, Millet P. High-pressure PEM water electrolysis
changes and to assess precisely all their impact on the dy-
and corresponding safety issues. Int J Hydrogen Energy
namic behaviour of the system. Thus, it is a modular and 2011;36:2721e8. http://dx.doi.org/10.1016/
powerful tool to improve the conception of PEM electrolysis j.ijhydene.2010.03.058.
systems, which will be used into further works. [17] Grigoriev SA, Millet P, Korobtsev SV, Porembskiy VI, Pepic M,
Etievant C, et al. Hydrogen safety aspects related to high-
pressure polymer electrolyte membrane water electrolysis.
references Int J Hydrogen Energy 2009;34:5986e91. http://dx.doi.org/
10.1016/j.ijhydene.2009.01.047.
[18] Rabih Samer. Contribution a la mode lisation de systemes
reversibles de types e lectrolyseur et pile a hydroge ne en vue
[1] IPCC. Summary for policymakers. In: Climate change 2013:
de leur couplage aux ge ne
rateurs photovoltaı̈ques. phd. 2008.
the physical science basis. Cambridge, United Kingdom and
[19] Rallieres Olivier. Mode lisation et caracte risation de Piles A
New York, NY, USA: Cambridge University Press; 2013.
Combustible et Electrolyseurs PEM. phd. 2011.
[2] Albadi MH, El-Saadany EF. Overview of wind power
[20] Agbli KS. Mode lisation multiphysique des flux e nerge
tiques
intermittency impacts on power systems. Electr Power Syst Res
d’un couplage photovoltaı̈que-e lectrolyseur PEM-pile a
2010;80:627e32. http://dx.doi.org/10.1016/j.epsr.2009.10.035.
combustible PEM en vue d’une application stationnaire. Co ^ te
[3] Carmo M, Fritz DL, Mergel J, Stolten D. A comprehensive
d’Ivoire: Universite de Franche-Comte , Universite de
review on PEM water electrolysis. Int J Hydrogen Energy
Cocody-Abidjan; 2012.
2013;38:4901e34. http://dx.doi.org/10.1016/
[21] Agbli KS, Pe ra MC, Hissel D, Rallie res O, Turpin C, Doumbia I.
j.ijhydene.2013.01.151.
Multiphysics simulation of a PEM electrolyser: energetic
[4] Yilanci A, Dincer I, Ozturk HK. A review on solar-hydrogen/
macroscopic representation approach. Int J Hydrogen Energy
fuel cell hybrid energy systems for stationary applications.
2011;36:1382e98. http://dx.doi.org/10.1016/
Prog Energy Combust Sci 2009;35:231e44. http://dx.doi.org/
j.ijhydene.2010.10.069.
10.1016/j.pecs.2008.07.004.
[22] Die guez PM, Ursúa A, Sanchis P, Sopena C, Guelbenzu E, technol. energy manag. John Wiley & Sons, Inc.; 2013.
Gandı́a LM. Thermal performance of a commercial alkaline p. 179e262.
water electrolyzer: experimental study and mathematical [39] Abdin Z, Webb CJ, Gray EM. Modelling and simulation of a
modeling. Int J Hydrogen Energy 2008;33:7338e54. http:// proton exchange membrane (PEM) electrolyser cell. Int J
dx.doi.org/10.1016/j.ijhydene.2008.09.051. Hydrogen Energy 2015;40:13243e57. http://dx.doi.org/
[23] Garcı́a-Valverde R, Espinosa N, Urbina A. Simple PEM water 10.1016/j.ijhydene.2015.07.129.
electrolyser model and experimental validation. Int J [40] Shen Muzhong, Bennett Nick, Ding Yulong, Scott Keith. A
Hydrogen Energy 2012;37:1927e38. http://dx.doi.org/10.1016/ concise model for evaluating water electrolysis. Int J
j.ijhydene.2011.09.027. Hydrogen Energy 2011;36:14335e41. http://dx.doi.org/
[24] Lebbal ME, Lecœuche S. Identification and monitoring of a 10.1016/j.ijhydene.2010.12.029.
PEM electrolyser based on dynamical modelling. Int J [41] Choi Pyoungho, Bessarabov DG, Datta R. A simple model for
Hydrogen Energy 2009;34:5992e9. http://dx.doi.org/10.1016/ solid polymer electrolyte (SPE) water electrolysis. Solid State
j.ijhydene.2009.02.003. Ion 2004;175:535e9. http://dx.doi.org/10.1016/j.ssi.2004.01.076.
[25] Onda Kazuo, Murakami Toshio, Hikosaka Takeshi, [42] Zhang Houchen, Su Shanhe, Lin Guoxing, Chen Jincan.
Kobayashi Misaki, Notu Ryouhei, Ito Kohei. Performance Efficiency calculation and configuration design of a PEM
analysis of polymer-electrolyte water electrolysis cell at a electrolyzer system for hydrogen production. Int J
small-unit test cell and performance prediction of large Electrochem 2013;7:4143e57.
stacked cell. J Electrochem Soc 2002;149:A1069e78. http:// [43] Dale NV, Mann MD, Salehfar H. Semiempirical model based
dx.doi.org/10.1149/1.1492287. on thermodynamic principles for determining 6 kW proton
[26] Kim Huiyong, Park Mikyoung, Lee Kwang Soon. One- exchange membrane electrolyzer stack characteristics. J
dimensional dynamic modeling of a high-pressure water Power Sources 2008;185:1348e53. http://dx.doi.org/10.1016/
electrolysis system for hydrogen production. Int J Hydrogen j.jpowsour.2008.08.054.
Energy 2013;38:2596e609. http://dx.doi.org/10.1016/ [44] da Costa Lopes F, Watanabe EH. Experimental and
j.ijhydene.2012.12.006. theoretical development of a PEM electrolyzer model applied
[27] Go € rgün H. Dynamic modelling of a proton exchange membrane to energy storage systems. In: Power electron. conf. 2009
(PEM) electrolyzer. Int J Hydrogen Energy 2006;31:29e38. http:// COBEP 09 braz.; 2009. p. 775e82. http://dx.doi.org/10.1109/
dx.doi.org/10.1016/j.ijhydene.2005.04.001. COBEP.2009.5347619.
[28] Marangio F, Santarelli M, Calı̀ M. Theoretical model and [45] Ni Meng, Leung Michael KH, Leung Dennis YC. Energy and
experimental analysis of a high pressure PEM water exergy analysis of hydrogen production by a proton
electrolyser for hydrogen production. Int J Hydrogen Energy exchange membrane (PEM) electrolyzer plant. Energy
2009;34:1143e58. http://dx.doi.org/10.1016/ Convers Manag 2008;49:2748e56. http://dx.doi.org/10.1016/
j.ijhydene.2008.11.083. j.enconman.2008.03.018.
[29] Grigoriev SA, Kalinnikov AA, Millet P, Porembsky VI, [46] Sawada S, Yamaki T, Maeno T, Asano M, Suzuki A, Terai T,
Fateev VN. Mathematical modeling of high-pressure PEM et al. Solid polymer electrolyte water electrolysis systems for
water electrolysis. J Appl Electrochem 2010;40:921e32. http:// hydrogen production based on our newly developed
dx.doi.org/10.1007/s10800-009-0031-z. membranes, Part I: analysis of voltageecurrent
[30] Darras C. Mode lisation de syste
mes hybrides characteristics. Prog Nucl Energy 2008;50:443e8. http://
photovoltaı̈que/hydroge ne: applications site isole, dx.doi.org/10.1016/j.pnucene.2007.11.029.
micro-re seau et connexion au re seau e
lectrique dans le [47] Henao Christian, Agbossou Kodjo, Hammoudi Mhamed,
cadre du projet PEPITE (ANR PAN-H). Universite Pascal Dube Yves, Cardenas Alben. Simulation tool based on a
Paoli; 2010. physics model and an electrical analogy for an alkaline
[31] Grigoriev SA, Porembsky VI, Fateev VN. Pure hydrogen electrolyser. J Power Sources 2014;250:58e67. http://
production by PEM electrolysis for hydrogen energy. Int J dx.doi.org/10.1016/j.jpowsour.2013.10.086.
Hydrogen Energy 2006;31:171e5. http://dx.doi.org/10.1016/ [48] Brown Tim M, Jacob Brouwer G, Samuelsen Scott,
j.ijhydene.2005.04.038. Holcomb Franklin H, King Joel. Dynamic first principles
[32] Grigoriev SA, Millet P, Volobuev SA, Fateev VN. Optimization model of a complete reversible fuel cell system. J Power
of porous current collectors for PEM water electrolysers. Int J Sources 2008;182:240e53. http://dx.doi.org/10.1016/
Hydrogen Energy 2009;34:4968e73. http://dx.doi.org/10.1016/ j.jpowsour.2008.03.077.
j.ijhydene.2008.11.056. [49] Lee Bonghwan, Park Kiwon, Kim Hyung-Man. Dynamic
[33] Marangio F, Pagani M, Santarelli M, Calı̀ M. Concept of a high simulation of PEM water electrolysis and comparison with
pressure PEM electrolyser prototype. Int J Hydrogen Energy experiments. Int J Electrochem 2012;8:235e48.
2011;36:7807e15. http://dx.doi.org/10.1016/ [50] Awasthi A, Scott K, Basu S. Dynamic modeling and simulation
j.ijhydene.2011.01.091. of a proton exchange membrane electrolyzer for hydrogen
[34] Millet P, Ngameni R, Grigoriev SA, Mbemba N, Brisset F, production. Int J Hydrogen Energy 2011;36:14779e86. http://
Ranjbari A, et al. PEM water electrolyzers: from dx.doi.org/10.1016/j.ijhydene.2011.03.045.
electrocatalysis to stack development. Int J Hydrogen Energy [51] Springer TE, Zawodzinski TA, Gottesfeld S. Polymer
2010;35:5043e52. http://dx.doi.org/10.1016/ electrolyte fuel cell model. J Electrochem Soc
j.ijhydene.2009.09.015. 1991;138:2334e42. http://dx.doi.org/10.1149/1.2085971.
[35] Paynter HA. Analysis and design of engineering systems. S.l. [52] Selamet OF,€ Acar MC, Mat MD, Kaplan Y. Effects of operating
Mit Press; 1960. parameters on the performance of a high-pressure proton
[36] Karnopp D, Rosenberg RC. System dynamics: a unified exchange membrane electrolyzer. Int J Energy Res
approach. New York: Wiley; 1975. 2013;37:457e67. http://dx.doi.org/10.1002/er.2942.
[37] Ould-Bouamama B, El Harabi R, Abdelkrim MN, Ben [53] Kothari R, Buddhi D, Sawhney RL. Studies on the effect of
Gayed MK. Bond graphs for the diagnosis of chemical temperature of the electrolytes on the rate of production of
processes. Comput Chem Eng 2012;36:301e24. http:// hydrogen. Int J Hydrogen Energy 2005;30:261e3. http://
dx.doi.org/10.1016/j.compchemeng.2011.07.008. dx.doi.org/10.1016/j.ijhydene.2004.03.030.
[38] Baurens P, Serre-Combe P, Poirot-Crouvezier J-P. Fuel cells. [54] Chandesris M, Me deau V, Guillet N, Chelghoum S, Thoby D,
In: Sabonnadie re J-C, editor. Low emiss. power gener. Fouda-Onana F. Membrane degradation in PEM water
electrolyzer: numerical modeling and experimental evidence J0k : Standard exchange current density at electrode k, A cm2
of the influence of temperature and current density. Int J LPEM : Length of the PEM membrane, m
Hydrogen Energy 2015;40:1353e66. http://dx.doi.org/10.1016/ Ncell : Number of cells in a stack
j.ijhydene.2014.11.111. Pnom : Converter nominal power, W
[55] Labbe Julien. L'Hydrogene e
lectrolytique comme moyen de Rohm : Ohmic resistance of the electrolysis cell, U
stockage d’e lectricite
pour syste
mes photovoltaı̈ques isole
s. Rothers : Ohmic resistance of other contributions except PEM

Ecole Nationale Supe rieure des Mines de Paris; 2006. membrane, U
[56] Bouamama BO. Bond graph approach as analysis tool in Rcond dry : Thermal resistance of purification unit, K$W1
thermofluid model library conception. J Frankl Inst Rcond enc : Thermal resistance of enclosure, K$W1
2003;340:1e23. http://dx.doi.org/10.1016/S0016-0032(02)00051-0. Rcond recir ano : Thermal resistance of anodic recirculation circuit,
[57] Biaku CY, Dale NV, Mann MD, Salehfar H, Peters AJ, Han T. A K$W1
semiempirical study of the temperature dependence of the Rcond recir cat : Thermal resistance of cathodic recirculation circuit,
anode charge transfer coefficient of a 6 kW PEM electrolyzer. K$W1
Int J Hydrogen Energy 2008;33:4247e54. http://dx.doi.org/ Rcond sep ano : Thermal resistance of oxygen separator vessel, K$W1
10.1016/j.ijhydene.2008.06.006. Rcond sep cat : Thermal resistance of hydrogen separator vessel, K$W1
[58] Han B, Steen III SM, Mo J, Zhang F-Y. Electrochemical Rth cell : Thermal resistance of electrolysis cells, K$W1
performance modeling of a proton exchange membrane RDRY : Fluidic resistance of purification unit, Pa s m3
electrolyzer cell for hydrogen energy. Int J Hydrogen Energy RRES : Fluidic resistance of water tank, Pa s m3
2015;40:7006e16. http://dx.doi.org/10.1016/ Rsep ano : Fluidic resistance of oxygen separator vessel, Pa s m3
j.ijhydene.2015.03.164. Rsep cat : Fluidic resistance of hydrogen separator vessel, Pa s m3
[59] Harrison KW, Herna ndez-Pacheco E, Mann M, Salehfar H. Rvalve sep : Fluidic resistance of separator valve, Pa s m3
Semiempirical model for determining PEM electrolyzer stack VDRY : Purification unit volume, m3
characteristics. J Fuel Cell Sci Technol 2005;3:220e3. http:// VSEPH : Hydrogen separator vessel volume, m3
dx.doi.org/10.1115/1.2174072. VSEPO : Oxygen separator vessel volume, m3
[60] Ito H, Maeda T, Nakano A, Kato A, Yoshida T. Influence of ak : Charge transfer coefficient at electrode k
pore structural properties of current collectors on the l: Water content of the membrane
performance of proton exchange membrane electrolyzer. heos : Electro-osmosis coefficient
Electrochimica Acta 2013;100:242e8. http://dx.doi.org/ h10 : Converter efficiency at 10% of its nominal power, %
10.1016/j.electacta.2012.05.068. h100 : Converter efficiency at nominal power, %
Cpi : Molar thermal capacitance of species i, J K1 mol1
Mi : Molar mass of species i, kg mol1
Nomenclature
ri : Density of species i, kg m3
ni : Stoichiometry coefficient of species i in water dissociation
reaction
Physico-chemical parameters
Model variables
DrG: Free enthalpy of water dissociation reaction, J mol1
DrG0 : Standard free enthalpy of water dissociation reaction,
E: Tension, V
J mol1
I: Current intensity, A
DrH: Enthalpy of water dissociation reaction, J mol1
J: Current density, A m2
DrH0 : Standard enthalpy of water dissociation reaction, J mol1
P: Pressure, Pa
DHads : Enthalpy of the water adsorption reaction, J mol1 _ Mass flow, kg s1
m:
DrS: Entropy of water dissociation reaction, J K1 mol1
T: Temperature, K
DrS0 : Standard entropy of water dissociation reaction, J K1 mol1 _ Heat flow, W
Q:
DrC0p : Standard specific heat of reaction, J K1 mol1 _ Enthalpy flow, W
H:
F: Faraday constant, C mol1
A: Chemical affinity, J mol1
g: Gravitational acceleration, m s2
m: Chemical potential, J mol1
R: Ideal gas constant, J mol1 K1 _ Reaction speed flow rate, mol s1
x:
z: Number of electrons involved in the reaction _ Molar flow, mol s1
n:
CPE : Products concentration near electrode, mol m3
Model parameters CPS : Products concentration in the solution, mol m3
CRE : Reagents concentration near electrode, mol m3
APEM : Section of the PEM membrane, m2 CRS : Reagents concentration in the solution, mol m3
ASEPH : Section of hydrogen separator vessel, m2 Eact;ano : Anode activation overvoltage, V
ASEPO : Section of oxygen separator vessel, m2 Eact;cat : Cathode activation overvoltage, V
hy
CSEPH : Hydraulic capacitive element of hydrogen separator, m s2 Ecell : Cell voltage, V
hy
CSEPO : Hydraulic capacitive element of oxygen separator, m s2 Eohm : Ohmic overvoltage, V
Cp : Heat capacity of stacks, J K1 Erev : Reversible potential, V
1
Cth
COLD : Heat capacity of cold tank, J K E0rev : Standard reversible potential, V
1
Cth
COOL : Heat capacity of cooling circuit, J K E : Thermoneutral voltage, V
1 Ptn
Cth
DRY : Heat capacity of purification unit, J K hk : Sum of overvoltages, V
1
Cth
ENC : Heat capacity of enclosure, J K LkSEPH : Water level in hydrogen separator, m
1
Cth
RECH : Heat capacity of cathodic recirculation circuit, J K LSEPO : Water level in oxygen separator, m
1
Cth
RECO : Heat capacity of anodic recirculation circuit, J K a: Converter transformer coefficient
Eak : Activation energy of half-reaction occurring at electrode k, hconv : Converter efficiency, %
J mol1 h_ eos : Electro-osmosis flow, mol s1
Jk : Current exchange density of the half-reaction occurring at sPEM : Proton exchange membrane conductivity, S m1
electrode k, A m2 ai : Chemical activity of species i
m_ i : Mass flow of species i, mol s1 xi : Mass fraction of species i

n_ i : Molar flow of species i, mol s1 Di : Diffusion coefficient of species i, m2 s1
mi : Chemical potential of species i, J mol1 Hi : Henry constant of species i, m3 Pa mol1
mi : Standard chemical potential of species i, J mol1
ref
Ri diff : Resistive element relative to diffusion of species i
Pi : Partial pressure of species i, Pa Hi : Molar enthalpy of species i, J mol1
ni : Molar quantity of species i, mol Si : Molar entropy of species i, J K1 mol1
mi : Mass quantity of species i, mol

PEM Model

Uploaded by

Copyright:

Available Formats

PEM Model

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

PEM Model

Uploaded by

Copyright:

Available Formats

i n t e r n a t i o n a l j o u r n a l o f h y d r o g e n e n e r g y x x x ( 2 0 1 7 ) 1 e3 3

Available online at www.sciencedirect.com

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

Dynamic and multiphysic PEM electrolysis system

Pierre Olivier a,*, Cyril Bourasseau a, Belkacem Bouamama b

article info abstract

policies promote the massive development of renewable en-

Fig. 3 e Technological scheme of a PEM electrolysis system.

(1-junction), or flow (0-junction). The conservative energy laws

Advantages regarding PEM electrolysis system modelling

Table 1 e Bond graph variables.

Table 2 e Bond graph elements.

(continued on next page)

transfer coefficient and the current exchange density of the

Fig. 5 e Word bond graph of the PEM electrolysis system considered.

Fig. 6 e Bond graph electrochemical model of the PEM electrolysis cell.

Fig. 8 e Input and output enthalpy flows of a PEM electrolysis stack.

Fig. 9 e Bond graph thermal model of the PEM electrolysis cell.

H_ RECH þ H_ RECO H_ CAT H_ ANO ¼ n_RECH RECH

H_ RECH þ H_ RECO H_ CAT H_ ANO ¼ n_H2 O $CpH2 O $ðTRECH TSTACK Þ

Q_ STA ¼ Q_ ohm þ Q_ act;ano þ Q_ act;cat (28)

Finally, equation (42) is obtained. It has the same structure

Fig. 10 e Fluidic bond graph model of the PEM electrolysis cell.

depend on the membrane thickness and seems to be the most Converter

I Estack ¼ a$Ein (49)

Fig. 11 e Simplified global bond graph of the PEM electrolysis cell.

Fig. 12 e Bond graph model of the converter.

Fig. 13 e Bond graph model of the oxygen separator vessel.

Fig. 14 e Bond graph model of the hydrogen separator vessel.

Recirculation circuits and cooling circuit

Fig. 16 e Bond graph model of the cathode side recirculation circuit.

Following the same process for other 0-junctions, thermal

used to compute the exchanged heat flow Q_ COOL in function of

Fig. 17 e Bond graph model of the purification unit.

fans are considered perfect mass flow sources. The resulting

Block diagram conversion

Fig. 19 e Bond graph model of the enclosure.

Fig. 20 e Global scheme of the Simulink interface.

Table 8 e Values of current exchange densities found in the literature.

15.00 Experimental data

the opening of the separator valve which enables to drain

2.00 Hydrogen flow model

315.00 Enclosure temperature

regarding system pressure, (iii) Fig. 29 regarding hydrogen

[5] Singh S, Jain S, Venkateswaran PS, Tiwari AK, Nouni MR,

m_ i : Mass flow of species i, mol s1 xi : Mass fraction of species i

You might also like