IDENTIFICATION AND MONITORING OF A PEM ELECTROLYSER

BASED ON DYNAMICAL MODELLING

Lebbal, M.E.1 [email protected] and Lecœuche, S.1,2 [email protected]
1
Department A.I, Ecole des Mines de Douai, 941 rue Charles Bourseul, 59508 Douai Cedex France
2
Laboratoire LAGIS, USTL, Cité Scientifique-Bâtiment P2, 59655 Villeneuve d'Ascq Cedex France

ABSTRACT
Hydrogen from water electrolysis associated with renewable energies is one of the most attractive
solutions for the green energy storage. To improve the efficiency and the safety of such stations, some
technological studies are still under investigation both on methods and materials. As methods, control,
monitoring and diagnosis algorithms are relevant tools. These methods are efficient when they use an
accurate mathematical model representing the real behaviour of hydrogen production system. This
work focuses on the dynamical modelling and the monitoring of Proton Exchange Membrane (PEM)
electrolyser. Our contribution consists in three parts: to develop an analytical dynamical PEM
electrolyser model dedicated to the control and the monitoring; to identify the model parameters and to
propose adequate monitoring tools. The proposed model is deduced from physical laws and
electrochemical equations and consists in a steady-state electric model coupled with a dynamical
thermal model. The estimation of the model parameters is achieved using identification and data
fitting techniques based on experimental measurements. Taking into account the information given by
the proposed analytical model and the experimentation data (temperature T, voltage U and current I)
given by a PEM electrolyser composed of seven cells, the model parameters are identified. After
estimating the dynamical model, model based diagnosis approach is used in order to monitoring the
PEM electrolyser and to ensure its safety. We illustrate how our algorithm can detect and isolate faults
on actuators, on sensors or on electrolyser system.

1.0 INTRODUCTION

In recent decades, the global warming increases the average temperature of the air and oceans near the
earth surface. This problem caused by CO2 gas and several polluting wastes continuing to affect the
lives in the world. In order to overcome this problem, the use of renewable energy and its optimization
become a humanity challenge [3]. An attractive solution is to integrate efficient energy storages. The
hydrogen is one of most promising vectors to store green energy. In the last years, numerous stations
including renewable energy and electrolyser have been developed in order to optimise the electric
energy production by increasing the storage capacity. The key idea is to convert the hydrogen into
electricity using Fuel Cell (FC) when the renewable energy is off (no wind, no sun). In order to have
this fuel, during the high potential periods, the extra renewable energy is converted using a Proton
Exchange Membrane (PEM) electrolyser into H2. The global efficiency and safety of such installations
(renewable energy source, fuel cell and electrolyser) lead to important research works in modelling,
control [1] and monitoring. More precisely, it is necessary to propose an efficient supervision system.
It permits to the user to decide if the hydrogen production station is faulty and if risk for itself and its
environment could occur. For example, when sensor or actuator fault is detected, control laws could be
adapted in order to put the system in fallback options. On the same way, maintenance could be planed
in order to avoid major system failure (e.g. membrane breaking). In the aim to optimise control and
monitoring of the hydrogen production and exploitation under safe conditions, the PEM cell modelling
became very important and necessary.

This work is dedicated to the electrolyser safety for the development of monitoring and diagnosis
tools. From the point of view our aim is to detect drifts that could lead to the critical failures by
employing Fault Detection and Isolation (FDI) methods. Our contribution consists in three parts:
modelling, identification and monitoring of the PEM electrolyser.
According to the recent literature, the modelling task still needs improvements in order to develop
models able to represent accurately the real dynamical equipment behaviour. Many electrolysis cell
models have been proposed, these models are deduced from the thermal, the electrical and the fluid
dynamics behaviours. Electrical model represents the relation between voltage U, current I and
temperature T [7]. Thermal model represents the temperature T variation and heat exchange [9]. The
fluid dynamics model characterizes the chemical components movement, the pressure and the fluid
volume [11]. Despite models variety, simplified electrical and thermal are mostly used [9]. Few
advanced monitoring and control algorithms using a efficient modelling are really used in real
conditions [10] . Our proposed model consists in a steady-state electrical model and a dynamic thermal
model. These models are deduced from physical laws and electrochemical equations. The steady-state
electric model represents algebra relations between voltage U, current I and temperature T, it reflects
steady behaviour. In order to reduce the model imprecision, these models need a parameters estimation
using experimental data. This imprecision is a real limitation for the monitoring and diagnosis tools
because the faults indicators and their sensitivity are strongly influenced by the parameters values. The
identification of the electrical model parameters employs a non linear least square algorithm [5] and
experimentation data (temperature T, voltage U and current I) given by a PEM electrolyser composed
of seven cells. The dynamic thermal model structure is inspired from the analytical heat transfer. Due
to the fact that this model is linear and of order one. The identification of the thermal model
parameters is developed using voltage U, current I, and temperature step responses.

Using these accurate models, it could be possible to develop efficient control system and on-line
monitoring stage. In order to detect and isolate actuators, sensors or system faults, specific tools have
to be designed [4]. For fault detection and isolation (FDI), we use model-based diagnosis approach.
Based on the identified PEM electrolyser models, residuals are generated in order to check non-
consistency between the theoretical information and the measurements. These residuals contain only
known or measured variables and are statistically nil in the absence of faults and different of zero
when a fault occurs. In this paper we show modelling and identification results and illustrate that our
monitoring algorithm (FDI) is able to detect and isolate faults.

The paper is organized in five sections. The first section concerns the introduction. The second section
presents the PEM electrolyser principle. The third section exposes the electric and thermal behaviours
and details our proposed models. The parameters estimation and the description of the used
identification techniques are developed in fourth section. The monitoring approach is detailed and
illustrates by results in the fifth part.

2.0 THE PEM ELECTROLYZER PRINCIPLE

The water electrolysis operation is an electrolytic process which decomposes water H2O molecule into
oxygen O2 and hydrogen H2 gasses with the help of an electric current (figure 1). The electrodes
submerged in an electrolyte (conductive medium) form the electrolysis cell. In PEM electrolysis cell,
the electrolyte is a solid called membrane. In this latter, the H+ ions are used for electricity
conductivity. We resume the water decomposition in followings Redox (Reduction and Oxidation)
chemical reactions:

2H2O + electric energy → 2H2 + O2 (1)

• Oxidation Describes the electrons (e-) loss (or the protons gain) by a molecule, atom or ion. In
water electrolysis, at the anode electrode (+) (oxygen production), the oxidation equation is:

2H2O → O2 + 4H++ 4 e- (2)

• Reduction Describes the electrons (e-) gain (or the protons loss) by a molecule, atom or ion. In
water electrolysis, at the cathode electrode (-) (hydrogen production), the reduction equation is:
4 H+ + 4 e- → 2 H2 (3)

Electric energy
Cathode (-) Anode (+)
Reduction → e- Oxidation ← e-

Hydrogen out Water in

e- H + e -

H2 H2O

Hydrogen out O2 Oxygen out

Membrane
Electrodes
Figure 1: PEM electrolyser device

For splitting the water molecule, the electric energy is consumed by the Redox chemical reaction. As
consequent, the change in enthalpy ∆H, entropy ∆S (heat exchange=irreversible energy), and Gibbs
energy (reversible energy) ∆G are closely related to the electric and the thermal energy generated by
cell [9]. Assuming that the PEM electrolyser cell temperature T [K] is given in Kelvin, for the water
splitting, the change in enthalpy can be given according to the following equation:

∆H =∆G + T ∆S (4)

According to this equation, the electrical and thermal behaviours are combined. From the view of
point of the electrolyser destruction, the current and the temperature are very significant. In one hand,
high current could cause: electrode destruction, membrane melting, membrane drying or electrode
pressure augmentation [8]. In other hand, a high temperature could induce: membrane hot point and
membrane tear. In addition, these faults could cause a H2 (respectively O2) migration to O2
(respectively H2) side.

All these faults could be dangerous for the electrolyser and its environment. In order to detect, isolate
and correct faults and then to improve the quality and the safety of the hydrogen production, this paper
proposes a model-based diagnosis of a PEM electrolyser. This approach needs three processing steps :
electrolyser modelling, parameters estimation and electrolyser monitoring which are detailed in the
next sections.

3.0 MODELLING APPROACH

Using the energy change equation (4) and electrochemical process aspect, we can divide our
electrolyser model into an electrical model completed with a thermal model. The electrical model is
based on the energy exchange in link with the Gibbs energy and the electric source. The heat transfer
equation is composed by the energy generated by entropy and the energy losses (or gained) by the
external temperature and chemical components movement. In our proposed modelling approach,
steady-state electrical model and dynamic thermal model are given based on energy conservation. The
electrical model is deduced from an algebra relation between the current cell I and voltage cell U for a
given temperature value T. The thermal model represents the dynamic temperature behaviours for both
current I and voltage U. The following figure illustrates our PEM electrolyser model.

Steady-state T Dynamic
electrical thermal
I and U
model model
 Figure 2:PEM electrolyser model

3.1 Electrical model

When the input voltage is applied to the PEM cell several voltage drops appear. These drops are
characterized by an reversible drop Vrev, an activation drop Vact, an diffusion drop Vdiff and an ohmic
losses Vohm. The current circulating into the cell is influenced by these drops which are most of them
non linear functions of the current. The electric schema is given in figure 3.

Vrev Vact Vdiff Vohm

U

Figure 3 : Electric schema

From the figure 3, using circuit laws, we can write the following equation :

U = Vrev + Vact + Vdiff + Vohm (5)

In order to build our steady-state electric model, we will study separately each voltage drop.

3.1.1 Reversible voltage

This voltage drop called sometimes open voltage is caused by the chemical Redox reaction. It
represents the cell potential when the current is nil (I=0). This reversible voltage Vrev is calculated
from the Gibbs energy, it can be defined by the Nernst equation [3] as follow:

RT  PH 2 PO12/ 2 
Vrev = V0 + ln  (6)
2 F  aH 2 O 

Where R = 8,314 4 [J / mol K], F =96 487 [C /mol], V0 =1.23 [V], aH2O =1 (for liquid water), PH2 and
PO2 [atm] are universal gas constant, Faraday constant, standard reversible voltage, water activity the
partially pressures of hydrogen and oxygen respectively.

3.1.2 Activation voltage drop

The activation voltage drop results from the proton transfer and chemical reaction velocity, it
represents the electrochemical kinetic behaviours [2]. The activation voltage drop Vact relation can be
deduced from Butler-Volmer and Tafel laws [2]. Based on the work of Fonts [2], Vact can be rewritten
for an electrolyser, in function of the current I as the following equation :

R ⋅T I
Vact = ln( ) (7)
α ⋅n⋅F I0

Where α, I0, are the transfer coefficient and the exchange current respectively.

3.1.3 Diffusion voltage drop

The fluids (gas and water) concentration and diffusion close to the electrodes influence the current
values. Indeed diffusion behaviour causes variations of partial pressures and chemical reaction
velocity. These changes induce voltage diffusion drop Vdiff. Based on an adaptation of diffusion
equation proposed by Fonts [2] for a PEM fuel cell, the diffusion voltage Vdiff can be given by the
following equation.
 R ⋅T I
Vdiff = ln( 1 + ) (8)
β ⋅n⋅F I lim

Where β and Ilim are the constant coefficient and the diffusion limit current respectively

3.1.4 Ohmic voltage drop

The resistor polymer membrane Rmem is the mainly cause of Ohmic voltage drop. The resistance value
is often given by empirical relation (9) proposed by Springer and al [6]. As given in relation (9) Rmem is
function of membrane section area Am [cm2], membrane thickness lm [cm], hydration ratio λm (=7 dry
enough =14 good hydration , =22 bathed) and temperature T.

lm
Rmem = (9)
1 1
Am ⋅ ( 0.005139 ⋅ λm + 0.00326 ) ⋅ exp( 1267( − ))
303 T

then the ohmic loss is given by :

lm
Vohm = Rmem ⋅ I = ⋅I (10)
1 1
Am ⋅ ( 0.005139 ⋅ λm + 0.00326 ) ⋅ exp( 1267( − ))
303 T

3.1.5 Electric model

Substitute the voltage drops given by equations (6), (7), (8) and (10) in (5), we can deduced the steady
state electrical model given by equation (11).

RT  PH 2 PO12/ 2  R ⋅T I R ⋅T I
U = V0 + ln  + ln( ) + ln( 1 + ) + Rmem ⋅ I (11)
2 F  aH 2 O  α ⋅ 2 ⋅ F I0 β ⋅n⋅F I lim

The hydrogen flow is given in [g/s] by :

I
F H 2= (12)
2F

3.2 Thermal model

The electrolysis cell temperature affects the relation between U and I. In order to represent
temperature behaviour and build a thermal model, we define all induced heat sources. We can consider
four principal heat powers : chemical reaction (entropy), chemical components thermodynamic (gasses
and water), external ambient temperature Ta and joule effect caused by current circulation. In the aim
to build the thermal model we use the heat energy conservation principle. For calculate chemical
reaction heat energy, we use the thermoneutral cell voltage Vth=1.48 [V] given by expression (13).
This voltage remains almost constant or slightly changes with temperature. In addition we assume that
the joule effect is neglected and the external temperature Ta is constant. Then the thermal model can be
written as the continuous dynamic equation (14):

∆H
Vth = (13)
2F

d ( T − Ta )
Cp = ( U − Vth ) ⋅ I − h ⋅ ( T − Ta ) (14)
dt
Where Cp [J/K], h [W/K], u = ( U − V ) ⋅ I and h ⋅ ( T − T ) are the electrolyser overall thermal
th a
capacity, the overall thermal admittance of electrolyser, the heat transfer rate generated by chemical
reaction (caused by entropy energy) and the heat transfer rate caused by external temperature and
fluids movement respectively.

Assuming that u, x=(T-Ta) and y=x are input, state and output of dynamic thermal model. The state
and output equations in continuous case are given by the first order linear dynamic model (15):

 dx h 1
 =− x+ u
 dt Cp Cp (15)
 y = x

Our modelling step allows having a PEM electrolyser model structure presented both by the steady-
state electrical model (11) and the dynamic thermal model (15). We can note that several parameters
(α, β, I0 Ilim ,Rmem, Cp and h, ) in the model are difficult to fix, because these values are specific to
each electrolyser device. In order to use these models, their parameters have to be estimated using
experimental data and identification techniques. In next part we propose and apply our identification
approach for PEM electrolyzer.

4.0 IDENTIFICATION APPROACH

In this section, electrical and thermal models parameters are estimated using two distinct identification
techniques. Electrical model parameters are estimated through a non-linear least square method [5]
and thermal model parameters are identified using the properties of a first order linear model. Next
sections present briefly these techniques and the estimated models.

4.1 Electrical model identification

4.1.1 Non-linear least square identification principle

We assume that we have N measurements (yk, gk(θ, r) ), k = 0,..,N, as samples, r represents the input
and output regressor vector and θ∈IRp is the parameter vector to be identified. Letting that a non linear
relation gk between yk and θ at any time k is given by :

yk = gk(θ, r) (16)

The parameter vector estimated by a non linear least square (NLS) is θ̂ ∈IRp which satisfy (17).

(
G( θˆ )T y − g( θˆ ,r ) = 0) (17)

Where, the observations vector y∈IRN×1 is composed by yk, the regressor vector g ( θˆ ,r ) ∈IRN×1 contain

[ ]
observations function g ( θˆ ,r ) and the G( θˆ , r ) = G ( θˆ , r ) with G = ∂g k ( θ , r ) . At the last of
ˆ
∂θˆi
k ki ki

identification, we have θˆ → θ ( θˆ = θ ). For find θ̂ the optimal solution of (17), recursive algorithms
can be used. We use in our case Gauss-Newton regression defined as:

(
θˆ ( j + 1 ) = θˆ + GT ( θˆ , r ) ⋅ G( θˆ , r ) )−1 ⋅ GT ( θˆ ,r ) ⋅ (y − g( θˆ ,r )) ˆ (18)
θ = θˆ ( j )

where θ̂ ( j ) and θ ( j + 1 ) are the value of θ̂ at j and j+1 iteration respectively.

ˆ
4.1.2 Electrical model parameters identification

We consider the static electrical model described by the equation (11). The basic idea is to fix the
operating temperature T=338 [k] and collect measurements (U, I). From these measurements, the
unknown parameters (α, β, I0 Ilim and Rmem) are identified applying the NLS method. The estimated
T
parameters are defined by θ =  1 1 1 1 
R mem  . Then the elements used for

α I0 β I lim 
identification are defined as following


[ ]
T
1 1 1 1  T
θ = 
ˆ R̂mem  = θˆ1 θˆ2 θˆ3 θˆ4 θˆ5
αˆ Î 0 β Î lim
 ˆ


 RT  PH 2 PO12/ 2 
 yk = U ( k ) − V0 − ln  (19)
 2 F a
 H 2O 
 R ⋅T ˆ R ⋅T
 g k = θ1 ⋅ 2 ⋅ F ln( θ 2 ⋅ I ( k )) − θ 3 n ⋅ F ln( 1 − θ 4 ⋅ I ( k )) + θ 5 ⋅ I ( k )
ˆ ˆ ˆ ˆ


Figure 4: Real and identified electrical model

Applying the non linear least square method at real data given by the PEM electrolyser cell, we obtain
the following parameters values: α=0.452; I0=0.13⋅10-3; β=0.04; Ilim =120; and Rmem=3.2⋅10-3. The
figure 4 illustrates a comparison between the identified electrical model and the real device data. We
note that our model only represents electric behaviours of PEM electrolyser. Indeed the error between
our model and real data is very small, the average relative error equals to 0.32%.

4.3 Thermal model identification

The parameters to be identified in the linear thermal model given by (15) are Cp and h. From this
model we can deduce the following Laplace transfer function TP:

1/ h
TP = (20)
 Cp 
 P + 1 
 h 
 
The basic idea is to collect measurements (U, I, T), then unknown Cp and h are identified using static
gain Sg and response time tr properties. Indeed in the first order linear dynamic system, Sg and tr are
related to Cp and h by following equation:

 1 final value of y final value of T − Ta 1 ( )

S g = h = final value of u = (
U − Vth ⋅ I
⇒h=
) Sg
 (21)
t = 3 ⋅ C p ⇒ C = t r ⋅ h
 r h
p
3

Using this identification approach we have identified values Cp=68544 and h=10.71. The following
figures illustrate the identified thermal model and real temperature data error for (U=1.74, I=24) at
external temperature Ta=298 [K]. The identified thermal model is validated for a whole of admissible
temperature variation (283-333 [K]).

Figure 5: Real and identified thermal model for U=1.74 and I=24

As example from figure 5, we note that our thermal model represent accurately the temperature
dynamic. Indeed the error between our model and real data is very small, the average relative error
equals to 0.032%.

Both the modelling and identification approach are developed in order to monitor and supervise a
PEM electrolyser. In the next part our monitoring approach is proposed by introducing model based
diagnosis algorithms.

5.0 MONITORING APPROACH

In order to improve the electrolyser safety, the monitoring is the supervision operation which permits
to the user to analyse if the real system is (or would be) faulty or healthy (no-faults) [4]. In the aim to
detect and isolate actuator, sensor or system faults, we use a model-based diagnosis approach. Based
on our PEM electrolyser model, residuals are generated in order to check non-consistency between the
theoretical information and the measurements. In an ideal situation, these residuals should remain zero
in the healthy case and non-zero when a fault occurs. Among the methods used to generate residuals,
the Analytical Redundancy Relations (ARR) permit to have equation where all the variables are known
[4]. In other words, these relations are functions of only the measured inputs and outputs without
requiring the use of the unmeasured states. In order to monitor the PEM electrolyser, the residuals are
generated from electrical and thermal models using a parity space diagnosis approach [4]. Our
monitoring structure (figure 6) and approach is described as follow.
 U
PEM electrolyser +- R1
Û
I
T
PEM electrolyser model + T̂ -+ R2
Û and T̂ generation en

Figure 6: Monitoring approach

Using as measurement the voltage U, the current I and the temperature T, two residuals R1 and R2 are
generated according to the PEM electrolyser model (11) and (15). The electrical model (11) can be
induced the residual R1 at each sample time k as :

 RT ( k )  PH 2 PO12/ 2  R ⋅ T ( k ) I( k ) R ⋅T( k ) I( k )
Û = V0 + ln  + ln( )− ln( 1 − ) + Rmem ⋅ I ( k )
 2F  aH 2O  α ⋅ 2 ⋅ F I0 β ⋅n⋅ F I lim (22)
 R1 = U − Û


where (α, β, I0 Ilim ,Rmem) are identified parameters

Using the thermal model (15) and the parity space diagnosis algorithm, the residual R2 could be
deduced. Indeed from (15) and (20) we can have linear discrete affined representation written as :

a 
y( k ) = a ⋅ y( k − 1 ) + b ⋅ u( k − 1 ) = [ y( k − 1 ) u( k − 1 )]  (23)
b 

where in our case for a sample time te=300s b=0.0043 and a=0.9542.

then the residual R2 can be given by the following equations:

( ) ( )
T̂ ( k ) = a ⋅ T ( k − 1 ) − Ta + b ⋅ U ( k − 1 ) − Vth ⋅ I ( k − 1 ) + Ta
 (24)
 R 2 = T ( k ) − T̂ ( k )

We use model parameters (α, β, I0 Ilim ,Rmem, a, b) estimated in the section 4 and measurement
(U(k),I(k),T(k),T(k-1)) to calculate instantaneous values of residuals R1(U,I,T,α, β, I0 Ilim ,Rmem) and
R2(U,I,T, a, b). When residuals are nil, it could be estimated that the actuators, the sensors and the
PEM electrolyser system are healthy. In the opposite, non zero residuals imply fault presences. The
fault detection and isolate could be diagnosed using a simple thresholding logic. A test is achieved on
the residual by comparison with a threshold S fixed according to the measurement noise and the
modelling uncertainties. A basic fault isolation decision is given in the table 1.

Table 1: Faults isolation

Non fault Thermal model electrical model Sensors or actuators are
System is healthy is faulty is faulty faulty
R1 R1<S R1<S R1>S R1>S
R2 R2<S R2>S R2<S R2>S
In the following figures we illustrate our simulation for healthy and faulty cases. The results are given
for current input I=2 [A], with electrodes pressure equal to 1 [atm]. The faults occur at time 12000s
and vanish at time 60000s.

Figure 7: Residuals R1 and R2 in healthy case

From figure 7, corresponding to the healthy behaviours we note that the residuals R1 and R2 are both
close to zero.

To illustrate our monitoring algorithm efficiency, we represent in the following figures, the residuals
(R1 and R2) evolutions when three faults occur. Figure 8 shows the behaviours when the desired
current actuator value is deviated by a fault equal to 0.3 A. Figure 9 illustrates the residuals behaviours
when the overall thermal admittance h parameter value is deviated. It takes a value equal to (h +90)
corresponding to a heat exchange fault. In figure 10, residuals evolutions due to a membrane fault are
represented. In this case, the membrane resistor deviation equals to 3Ω (Rmem+3).

Figure 8: Residuals R1 and R2 in case of an actuator fault

 Figure 9: Residuals R1 and R2 when one thermal model parameter is faulty

Figure 10: Residuals R1 and R2 when one electrical model parameter is faulty

In figure 8, 9 and 10, we remark that for the two residuals R1 and R2, the choice of two threshold
permit the fault detection. The optimal threshold values are defined in order to increase the fault
detection and decrease the false alarm. In figure 8 and 9, we note that the sensor and thermal model
fault can be isolated according to the table 1. But in figure 10 the electrical model fault can be isolated
and confirmed after a delay corresponding to the residual R2 evolution. This delay is due to the
thermal model response time seeing that the electrical model and thermal are closely coupled. Indeed
in this case electrical model is influenced by temperature and thermal model is influenced by the faulty
voltage sensor. Our monitoring results illustrate that we can avoid several critical and dangerous cases
as presented in §2. Indeed the current actuator monitoring permits to avoid electrode destruction,
membrane melting, membrane drying or electrode pressure augmentation [8]. The temperature sensor
monitoring allows avoiding membrane hot point, membrane tear. In this sense, the membrane
monitoring is an important tool to ensure the electrolyser safety.
6.0 CONCLUSIONS

In this work we have developed for PEM electrolyser a model, an identification approach and a
diagnosis algorithm. Our model consists in steady-state electrical model and linear dynamic thermal
model. The PEM electrolyser model is validated with an identification approach through real
experiments. The identification algorithm uses a non linear least square method for electrical model
parameters estimation and the thermal model parameters are estimated using the first order response
properties. In order to have a safe electrolyser operating mode, our identified model is used for
developing monitoring algorithms. Our monitoring approach is based on a model based diagnosis
method, it permits the detection and the isolation of sensors, actuators and (or) system faults. Our
diagnosis approach can be used to avoid electrode destruction, membrane melting, membrane drying,
electrode pressure augmentation, membrane hot point and membrane tear. Several results are
presented in order to illustrate the advantage of our modelling, identifying and monitoring methods.
The perspective of this work is to improve monitoring approach by the use of adaptive threshold
detection and the study fault detectors (residuals) sensitivity to a several PEM electrolyser parameters.

REFERENCES

1. Chiu, L.Y, Diong, B and Gemmen, R.S. An improved small-signal model of the dynamic behavior
of pem fuel cells. IEEE transactions on industry applications, vol. 40, no. 4, pp970-977,
july/august 2004.
2. Fontes, G. Modélisation et caractérisation de la pile PEM pour l’étude des interactions avec les
convertisseurs statiques. Thèse doctorat, laboratoire d’électrotechnique et d’électronique
industrielle de l’ENSEEIHT UMR CNRS, l’institut national polytechnique de Toulouse, 16
septembre 2005.
3. Francois, B, Hissel, D and Iqbal, M.T. Dynamic modelling of a fuel cell and wind turbine DC-
linked power system Electrimacs conference, Hammamet, Tunisia, April 17-20 2005
4. Lebbal. M. Contribution à la modélisation et au diagnosis des systèmes à commutations. PHD
thèse de l’université du havre, Le Havre, France 2006.
5. Ljung,L. System identification: Theory for the user. Prentice-Hall information and system
sciences series, Prentice-Hall, Englewood Cliffs, New Jersey, 1999
6. Springer, T. E, Zawodinski, T. A and Gottesfeld, S. Polymer electrolyte fuel cell model, Journal of
the Electrochemical Society, vol. 138, no 8, pp. 2334–2342,1991.
7. Wang, C, Nehrir, M.H and Shaw, S.R. Dynamic models and model validation for PEM fuel cells
using electrical circuits. IEEE transactions on energy conversion, vol. 20, no. 2, pp 442 452, June
2005.
8. Tekin, M, Hissel, D, Pera, M and Kauffmann, J. Energy consumption reduction of a PEM fuel cell
motor-compressor group thanks to efficient control laws. Journal of Power Sources 156, pp 57–63
(2006).
9. Ulleberg, O. Modeling of advanced alkaline electrolyzers: a system simulation approach.
International Journal of Hydrogen Energy 28, pp21-33 , 2003.
10. Vosen, S.R. Keller, J.O. Hybrid energy storage systems for stand-alone electric power systems:
optimization of system performance and cost through control strategies International Journal of
Hydrogen Energy 24, pp1139-1156, 1999.
11. Yu, Q, Srivastava, A.K, Choe, S.Y and Gao, W. Improved modeling and control of a PEM fuel
cell power system for vehicles”. Proceedings of the IEEE Southeast Conference, pp 331- 336 ,
March 31 - April 2, 2006.