Energy Conversion and Management 65 (2013) 408–419

Contents lists available at SciVerse ScienceDirect

Energy Conversion and Management

journal homepage: [Link]/locate/enconman

Robust voltage tracking control for proton exchange membrane fuel cells
Pang-Chia Chen ⇑
Department of Electro-Optical Science and Engineering and Department of Electrical Engineering, Kao Yuan University, No. 1821, Jhong-Shan Rd., Lu-Jhu District,
Kao-Hsiung City, Taiwan

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

Article history: This paper presents a robust control approach for proton exchange membrane (PEM) fuel cell systems. In
Received 3 August 2012 a linear parameter varying system representation of the nonlinear PEM fuel cell dynamics, the system
Received in revised form 6 September 2012 matrices are dependent on the system varying parameters, the output current and the stack temperature.
Accepted 6 September 2012
To obtain guaranteed design performance, system uncertainties caused by the variational system param-
Available online 17 October 2012
eters are addressed during controller design. The voltage tracking performance is expressed in terms of
H1 optimization of the ratio of the tracking error to the issued command. The controller is constructed
Keywords:
numerically in terms of the convex tractable linear matrix inequalities. Due to the parameter-dependent
Proton exchange membrane fuel cell
Robust voltage control
system matrices of the PEM fuel cells, the formulated matrix inequalities in denoting various design spec-
Parameter-dependent matrix inequality ifications are also dependent on the system varying parameters. Using the affinely dependent property of
Constrained control these matrix inequalities, design performance can be established by evaluating only the matrix inequal-
ities in the extremes of the varying parameters. Both nominal and robust controller designs are verified
through time response simulation for both nominal PEM fuel cell and nonlinear PEM fuel cell dynamics.
Ó 2012 Elsevier Ltd. All rights reserved.

1. Introduction only time-dependent behavior, it is known as a zero-dimensional or

lumped model. On the other hand, if the effects of spatial changes
Fuel cell and hydrogen related technologies have been develop- are also considered, the resulting models can be either one dimen-
ing toward technological maturity and practical market commer- sional, two dimensional or three dimensional [4].
cialization. As long as sufficient hydrogen can be supplied, a fuel The work in [10] proposed a performance modeling of the Bal-
cell can produce electricity consistently, with high efficiency and lard Mark IV 5 kW PEM fuel cell. The model was developed based
low pollution factors. Therefore, fuel cell technology is attracting on a combination of mechanistic and empirical modeling tech-
considerable international interest. The proton exchange mem- niques. The dependence of fuel cell parameters such as the activa-
brane (PEM) fuel cell possesses the advantages of maneuverable tion overvoltage and internal resistance was investigated. The
output power, low operation temperature and quick start, which modeling approach is considered to be effective and efficient. In
makes PEM fuel cells suitable for transportation applications and [11], a new dynamic model of a 1.2 kW polymer electrolyte mem-
residential usages [1,2]. brane fuel cell was developed and validated through a series of
PEM fuel cell systems present nonlinear dynamics [3,4] and pos- experiments. The characteristics of temperature, no load voltage
sess characteristics that are affected by the operating conditions [5], and equivalent internal resistance were developed. The proposed
which include the effects of the output current [6], the temperature model was simple, yet could obtain reasonable accuracy in evalu-
[7] and the humidity [8]. From the viewpoint of control, the fuel ating the fuel cell system performance. However in [10,11], the ef-
cell’s operation can be changed by the hydrogen and air (oxygen) fects of the control inputs, i.e. hydrogen and air (oxygen), on the
supply, which are considered as the systems’ maneuvering or en- fuel cell output were not addressed; thus the model was not suit-
ergy inputs. Moreover, as demonstrated in [9], fuel cell performance able for applications regarding fuel cell control system design.
has a positive correlation with the anode and cathode pressures, The works of [12] investigated a generic steady-state model for
which can be considered as the system dynamics’ state variables. the PEM fuel cell. In addition to the operating variables of gas feed,
Mathematical models have been developed for purposes of de- pressure, cell temperature and current density, other parameters
sign, optimization, monitoring and control of PEM fuel cell systems. including fuel cell active area, membrane thickness and membrane
Modeling of PEM fuel cell gas has been performed for both the stea- aging were considered and evaluated by means of the fuel cell’s
dy-state and dynamic behaviors. For fuel cell modeling considering experimental results. In recent work [13], a nonlinear state-space
model composed of 11 states for a 500 W PEM fuel cell was
⇑ Tel.: +886 7 6077285; fax: +886 7 6077009. developed. The thermodynamic energy was considered along with
E-mail address: patsuchen@[Link] four state variables directly related to the thermodynamics,

0196-8904/$ - see front matter Ó 2012 Elsevier Ltd. All rights reserved.
[Link]
 P.-C. Chen / Energy Conversion and Management 65 (2013) 408–419 409

Nomenclature

CH2 hydrogen concentration at the anode/membrane inter- e ce11 cell voltage, its equilibrium, and deviation
V cell ; V cell ; V
face V ohmic ; V ohmic ; V e ohmic ohmic overvoltage, its equilibrium, and
CO2 oxygen concentration at cathode/membrane interface deviation
ENernst thermodynamic potential V stack ; V stack ; Ve stack stack voltage, its equilibrium, and deviation
I, dI output current and its deviation gact cathodic activation drop
m_ H2;in ; m ~_
_ H2;in ; m H2;in hydrogen molar input flowrate, its equilib- fi, i = 1, 2, 3, 4 parametric coefficients of cathodic activation drop
rium, and deviation xf ; xf ; x
~ f state vector, its equilibrium, and deviation of PEMFC
m_ H2;used ; m _ H2;out hydrogen molar consumed, and output flowrate u; u ; u ~ control vector, its equilibrium, and deviation
m_ O2;in ; m ~_
_ O2;in ; m ~ i;max ; i ¼ 1; 2 magnitude bounds on deviation of control efforts
O2;in oxygen input flowrate, its equilibrium, and u
deviation h system varying parameters of PEMFC
m_ O2;used ; m _ O2;out oxygen molar consumed, and output flowrate Af ðxf ; hÞ system matrix of PEMFC
PH2 ; PH2 ; P e H2 partial pressure of hydrogen at the anode, its equi- Bf(h) input matrix of PEMFC
librium, and deviation Cf ðxf ; hÞ output matrix of PEMFC
PO2 ; PO2 ; P e O2 partial pressure of oxygen at the cathode, its equi- xp, ze, ei state vector, tracking error, and integral tracking error of
librium, and deviation the augmented PEMFC
Pstack stack power output vdc, vd external, and low-passed actual voltage tracking com-
Ract activation resistance mand
Rint internal resistance attributed to the electrolyte mem- v d;max ; v d;max nominal and actual bound of voltage tracking com-
brane mand
Rload equivalent load resistance A, B, Bd, Ce, De system matrices of the augmented PEMFC
rm membrane resistivity K integral tracking control law
T, dT temperature and its deviation in absolute degree P, Q, L, g auxiliary variables for controller synthesis
V act ; V act ; Ve act activation overvoltage, its equilibrium, and c H1 performance level
deviation

including the stack temperature. However, as pointed out in [4], in PEM fuel cells with the objective of controlling and maintaining
multi-time-scale PEM fuel cell systems, electronic components the output voltage during load variation. The coefficients of the
have the fastest responses, while thermal processes usually have PID controller were optimized based on genetic algorithms. The
the slowest responses. From the viewpoint of control, the elec- anode and cathode inlet pressure were chosen as control inputs.
tronic and the thermal state variables should be controlled sepa- In the work of [18], a fuzzy-PI controller was used to improve both
rately or addressed by an inner and outer loop control structure. PEM fuel cells and solid oxide fuel cells as distribution generations
For the control of PEM fuel cell systems, controllers are de- under voltage flicker. The control objective was to regulate the
signed with the objective of regulating the various system vari- deviation of the transient output power.
ables into a normal range and possibly achieving optimization of Also, control of PEM fuel cells directly based on the systems’
certain performance variables while accommodating varying oper- nonlinear dynamic models has been proposed. In [19], based on
ating conditions. Controller designs can be produced by use of the control-oriented dynamic thermal affine model identified by
intelligent control approaches such as artificial neural networks an optimization algorithm, a novel variable structure control was
[14] or fuzzy reasoning [15], in which no explicit governing models presented for the temperature control system of PEM fuel cells.
for the PEM fuel cell system are needed. In [14], a recurrent net- In [20], feedback linearization was applied to regulate the devia-
work controller was used to manipulate the anode and cathode tions of hydrogen and oxygen pressure during disturbances or load
water mole fractions in order to fix cell voltage and preserve cell variations. The hydrogen and oxygen inlet flow rates were defined
water content within a satisfactory interval regardless of the vary- as manipulated control inputs. In the work of [21], the state feed-
ing cell current. In the work of [15], a fuzzy controller was applied back exact linearization approach was used to obtain a robust con-
for PEM fuel cells described by a nonlinear model. The purpose was trol strategy based on the PEM fuel cell’s dynamic nonlinear model.
to maintain the output voltage level under varying load through The control variables were the anode and cathode inputs.
gas pressure control. However, from the viewpoint of energy bal- Model predictive control featuring on-line computation for real-
ance, the output voltage related to the fuel cell’s electrical output time objective optimization and constraints accommodation have
energy is much more appropriately manipulated via the input also been investigated for the control of PEM fuel cell systems. In
hydrogen and oxygen, which represent the chemical input energy, [22], under the multi-loop nonlinear predictive control framework,
as opposed to use of the water mole fractions as in [14]. Also, the air (oxygen) and water flowrates as control inputs are manipulated
gas pressures as the PEM fuel cells’ internal variables in [15] are to regulate both the stack temperature and oxygen excess ratio. In
better considered as controlled state variables, instead of as the [23], an internal model control law was optimized on-line to oper-
control inputs to be manipulated directly. ate a PEM fuel cell at its peak power density and continually track
Alternatively, PEM fuel cell systems can be controlled via mod- the shifting optimum. The cell voltage and inlet molar flow rate
el-based control approaches, which are directly related to evalua- were selected as manipulated inputs. The average power density
tion of the mathematical model derivation and also to the realm and temperature were the controlled outputs.
investigated in this present work. The well known PID-related con- Robust control approaches have been proposed to handle the is-
trol approaches were employed for the PEM fuel cell control sys- sues of system uncertainties for the fuel cell dynamic models. In
tem design. In [16], a dynamic model for PEM fuel cell systems [24], the controller design for voltage regulation under small load
was derived, after which a fuzzy PID controller and a PID controller variation was performed based on the numerically convex tracta-
were used to control the output voltage. The control input was the ble linear matrix inequality (LMI) [25] algorithm. Different linear
hydrogen pressure. In [17], a PID controller was investigated for models of the PEM fuel cell were identified under different loads
410 P.-C. Chen / Energy Conversion and Management 65 (2013) 408–419

and used for robust controller design. In [26], the third-order non- Although the concentration overvoltage can be included in the volt-
linear dynamics of a PEM fuel cell were derived and equivalently age Eq. (1) for completeness such as in [3,13], nevertheless, as
represented as a linear parameter varying (LPV) [27] system ame- pointed out in [28], the most important losses are due to the activa-
nable to controller design via linear control methodologies. Then, tion and ohmic losses. The expression of the output voltage (1)
controller design was performed based on a fixed linear time agrees with the model used in the works of [10,22,29], in which
invariant PEM fuel cell model obtained from evaluation of the for- the concentration overvoltage is neglected for simplicity. The total
mulated LPV systems around the half load operation condition and stack voltage including the specification of cell number N = 35 in
the regulated stack temperature. series is
In this present paper, a robust control is investigated for PEM
fuel cell systems as in the previously proposed LPV formulation V stack ¼ NV cell : ð2Þ
in [26]. Instead of considering only the nominal operating condi-
The thermodynamic potential, ENernst, is expressed as:
tion defined by the half load output current and the regulated
nominal stack temperature, the whole range of the system varying  
RT
parameters, the output current and the stack temperature, all of ENernst ¼ 1:229  8:5  104 ðT  298:15Þ þ ln PH2 ðPO2 Þ0:5 ; ð3Þ
2F
which affect the elements of the state-space PEM fuel cell model,
are addressed explicitly during the process of controller design.
where 298.15 is the assumed room temperature in absolute de-
The objective of controller design for the PEM fuel cell system is
grees. The activation overvoltage, Vact, is described by a first-order
to achieve a robust voltage tracking control which takes into ac-
dynamic with the effects of double layer capacitance, Cdl, charging
count the variations of system matrices corresponding to the spec-
at the electrode–electrolyte interfaces
ified range of output current and stack temperature. During
controller design, the magnitudes of the desired tracking command
and the allowed gas flowrates are specified correlatively. The volt-
dV act V act I
¼ þ ; ð4Þ
age tracking performance is expressed in terms of H1 performance dt Ract C dl C dl
level regarding the issued command to the tracking error. More-
over, suitable pole placement constraints are imposed to address where the activation resistance is
closed-loop system behavior. The controller is constructed numer- Ract ¼ gact =I; ð5Þ
ically by using LMI algorithms to meet the desired various con-
straints on system variables and optimization of tracking and the cathodic activation drop, gact, is expressed by the empirical
capability, as well as the system uncertainties caused by variation model as:
of operation current. gact ¼ f1 þ f2 T þ f3 T lnðC O2 Þ þ f4 T ln I; ð6Þ
Both a nominal controller considering only the nominal varying
parameters and a robust controller for whole the specified range of with the parametric coefficients
varying parameters are designed and verified through time re- f1 ¼ 0:948;
sponse simulation. The nominal controller design can achieve bet-
ter design performance and faster time response under the same f2 ¼ 0:00286 þ 0:0002 lnðAcell Þ þ 4:3  105 lnðC H2 Þ;
ð7Þ
magnitude of gas flowrates than the robust controller. The robust f3 ¼ 7:6  105 ;
controller can obtain guaranteed design performance under non-
f4 ¼ 1:93  104 :
linear PEM fuel cell dynamics with varying system matrices. In
the time response simulations, both the nominal PEM fuel cell sys- The hydrogen concentration at the anode/membrane interface, CH2,
tem with constant system matrices and the nonlinear PEM fuel cell and the oxygen concentration at cathode/membrane interface, CO2,
dynamics are evaluated. As shown, although more gas flowrates are determined by Henry’s law as:
are exerted when the nonlinear PEM fuel cell dynamics are used,  
77
the responses exhibit larger integral tracking errors than the nom- C H2 ¼ 9:174  107 PH2 exp ;
inal PEM fuel cell system for both the nominal and robust designed T
  ð8Þ
controllers. 498
C O2 ¼ 1:97  107 P O2 exp :
The rest of the work is organized as follows: Section 2 presents T
the dynamics of the PEM fuel cell, including the nonlinear dynam-
By substituting the activation resistance in (5) into (4), the dynam-
ics, the equilibrium operation conditions, the reformulated LPV
ics of activation overvoltage are obtained as:
system representation and uncertainty of system matrices. In Sec-
tion 3, the control problem is formulated, and then the design of dV act V act I I
robust state-feedback control law to achieve tracking performance, ¼ þ : ð9Þ
dt gact C dl C dl
under the specified constraints of tracking command magnitude
and allowable gas flowrates, is represented in terms of parame- The ohmic overvoltage is
ter-dependent matrix inequalities. Section 4 is the time response
V ohmic ¼ IRint ; ð10Þ
simulation of the PEM fuel cell system equipped with the con-
structed controllers. Finally, Section 5 gives the conclusions. where the internal resistance attributed to the electrolyte mem-
brane is expressed as:
2. PEM fuel cell system dynamics rm ‘m
Rint ¼ : ð11Þ
Acell
2.1. The nonlinear PEM fuel cell dynamics
An empirical expression for the membrane resistivity is
The output voltage of a single fuel cell can be defined by the    
 T 2  I 2:5
sum of three voltage terms: the thermodynamic potential, ENernst, 181:6 1 þ 0:03 A I þ 0:062 303 A
cell cell
the activation overvoltage, Vact, and the ohmic overvoltage, Vohmic: rm ¼ h  i  T303 : ð12Þ
I
11:866  3 A exp 4:18 T
V cell ¼ ENernst  V act  V ohmic : ð1Þ cell
 P.-C. Chen / Energy Conversion and Management 65 (2013) 408–419 411

Fig. 1. Diagram of the PEM fuel cell system.

The dynamics of the partial pressure of hydrogen at the anode NI

_ H2;used ¼
m ; ð17Þ
and the partial pressure of oxygen at the cathode are given, respec- 2F
tively, by: NI
_ O2;used
m ¼ : ð18Þ
4F
V an dP H2
¼m _ H2;out  m
_ H2;in  m _ H2;used ; ð13Þ
RT dt By taking into account (15)–(18), the dynamics of the partial pres-
V ca dP O2 sure of hydrogen in (13) and oxygen in (14) are obtained as:
¼m _ O2;out  m
_ O2;in  m _ O2;used : ð14Þ
RT dt V an dPH2 NI
_ H2;in  kan ðP H2  P H2;tank Þ 
¼m ; ð19Þ
As shown in Fig. 1 for the structure of the studied PEM fuel cell con- RT dt 2F
trol system, the excess hydrogen is re-circulated to the anode and V ca dPO2 NI
_ O2;in  kca ðPO2  PO2;BP Þ 
¼m : ð20Þ
the throttle is assisted to regulate the inlet flowrate of hydrogen RT dt 4F
at the proper tank pressure, PH2,tank. Therefore, the hydrogen molar By denoting the state variables including the activation over-
outlet flowrate is voltage, the hydrogen partial pressure and the oxygen partial pres-
T
_ H2;out ¼ kan ðPH2  PH2;tank Þ:
m ð15Þ sure as xf ¼ ð V act P H2 PO2 Þ , the control inputs including the
hydrogen and oxygen input flowrate as u ¼ ð m _ O2;in ÞT and
_ H2;in m
Because the oxygen pressure at the outlet is regulated at the desired the varying system parameters including the current and temper-
back pressure, PO2,BP, the oxygen molar outlet flowrate is expressed ature as h ¼ ð I T ÞT , the system nonlinear dynamics consist of (9),
as (19) and (20) can be expressed as:
_ O2;out ¼ kca ðPO2  PO2;BP Þ:
m ð16Þ
x_ f ¼ fðxf ; u; hÞ: ð21Þ
In this study, the tank pressures of the hydrogen and oxygen back
pressures are regulated as PH2,tank = 3 atm and PO2,BP = 3 atm, which
It is noted that among the system varying parameters, the current is
are listed in Table 1 for the system parameter values of the Ballard
mainly determined by the fuel cell control inputs and the external
5 kW PEM fuel cell.
varying electrical load demand. On the other hand, the temperature
The consumptions of hydrogen and oxygen depend on electrical
is to be regulated in a reasonable range to assure operating safety
load and are respectively expressed as:
and optimal performance. As depicted in Fig. 1, the temperature
can be regulated by a suitable heat exchanger and the fresh air pro-
vided by a blower. Also, as pointed out in [4] and explained previ-
ously in Section 1, in PEM fuel cell system dynamics, the electric
Table 1 variables controlled in this paper (i.e. the voltage and current) have
System parameters of the Ballard 5 kW PEMFC. the fastest responses, while the regulated stack temperature usually
Symbol Description Value Unit
has the slowest response. For the control of electric variables as
done in this work, the slow dynamics of the temperature related
Acell Cell active area 232 cm2
to the system’s thermodynamic energy balance are neglected and
Cdl Double layer capacitance 0.035  232 F
F Faraday constant 96,485 Cmol1 only the range of the temperature is properly addressed during con-
kan Anode flow constant 0.065 mol s1 atm1 troller design.
kca Cathode flow constant 0.065 mol s1 atm1
lm Membrane thickness 178  104 cm
2.2. The equilibrium operation conditions and LPV system
N Number of cells 35
PH2,tank Hydrogen tank pressure 3 atm representation
PO2,BP Oxygen back pressure 3 atm
R Universal gas constant 8.314 J mol1 K1 Consider the equilibrium operating conditions of the nonlinear
Van Anode volume 0.005 m3 dynamics of the PEM fuel cell in (21). By denoting dVact/dt = 0, the
Vca Cathode volume 0.01 m3
equilibrium value of the activation overvoltage is:
k Oxygen excess ratio kmin = 2, knominal = 3, kmax = 4
kJ mol1
 
DH Hydrogen enthalpy 285.5 V act ¼ gact PH2 ; P O2 ; I; T :¼ V act PH2 ; PO2 ; I; T : ð22Þ
412 P.-C. Chen / Energy Conversion and Management 65 (2013) 408–419

By considering dPH2/dt = 0, the steady hydrogen partial pressure is From the nonlinear dynamics (9), the equilibrium condition in
obtained as: (22), and the denotation in (27), the dynamics regarding the devi-
  ation of activation overvoltage can be obtained as:
1 NI
PH2 ¼ m_ H2;in  þ PH2;in :¼ P H2 ðm
_ H2;in ; IÞ: ð23Þ  
kan 2F e_ act ¼ a11 P ; P ; I; T V
V e act þ a12 P ; P  ; I; T P
e H2
H2 O2 H2 O2
 
Also from dPO2/dt = 0, we can have the oxygen partial pressure writ- þ a13 PH2 ; PO2 ; I; T Pe O2 ; ð29Þ
ten as:
  where the parameterized coefficients are computed as [26]
1 NI
PO2 ¼ m_ O2;in  þ PO2;out :¼ PO2 ðm
_ O2;in ; IÞ: ð24Þ  I
kca 4F a11 PH2 ; P O2 ; I; T ¼  ; ð30aÞ
gact PH2 ; PO2 ; I; T C dl
Let the equilibrium values of the state variables in (22)–(24) be 
T a12 PH2 ; P O2 ; I; T
denoted as, xf ¼ ð V act P H2 PO2 Þ , which are defined correspond-     
 
ing to the control inputs, u ¼ m _ O2;in T . The equilibrium
_ H2;in m IT  4:3  105  C1H2  9:174  107 exp 77 T
state variables are also functions of the varying parameters, ¼  ; ð30bÞ
gact PH2 ; PO2 ; I; T C dl
h ¼ ð I T ÞT , where the current is determined by the required elec-   
tric load and the temperature is usually regulated by an air fan   ITf3  C1O2  1:97  107 exp 498 T

a13 PH2 ; P O2 ; I; T ¼  : ð30cÞ
cooler or heat exchanger. gact P H2 ; PO2 ; I; T C dl
As indexed by the current parameter, from (17) and (18), the
equilibrium values of hydrogen and oxygen input flowrate are gi- Similarly, by (19), (23),(27) and (28), the deviation of hydrogen par-
ven as tial pressure can be described by

N  I N  I e_ H2 ¼ a22 ðTÞ P
P e H2 þ b21 ðTÞm
~_ H2;in ; ð31Þ
_ H2;in ¼
m _ O2;in ¼ knominal 
;m ; knominal ¼ 3: ð25Þ
2F 4F
where the coefficients are
Under the equilibrium values of current, I⁄ = [0, 50, 100, 150, 200,
250, 300]A, and temperature regulated at T = To = 70 °C, the corre-
a22 ðTÞ ¼ kan RT=V an ; b21 ðTÞ ¼ RT=V an : ð32Þ
sponding equilibrium system variables can be obtained as shown As well, by (20), (24), (27) and (28), the deviation of oxygen partial
in Table 2, including the state variables V act ; P H2 ; PO2 from pressure is described by:
(22)–(24), the hydrogen and oxygen input flow rates
_ H2;in ; m
m _ O2;in from (25), the stack voltage from (1) and (2) e_ O2 ¼ a33 ðTÞ P
P e O2 þ b32 ðTÞm
~_ O2;in ; ð33Þ

V stack ¼ N ENernst  V act  V ohmic ; ð26Þ where the coefficients are

the stack power output by V stack I , and the equivalent load resis- a33 ðTÞ ¼ kca RT=V ca ; b32 ðTÞ ¼ RT=V ca : ð34Þ
tance by Rload ¼ V stack =I . By collecting (29), (31), and (33), the state dynamics of the PEM fuel
Let the deviations of the state variables and control inputs from cell in an LPV system representation can be written as:
their equilibrium operation conditions be respectively denoted as 0 1 0 1
a11 ðxf ; hÞ a12 ðxf ; hÞ a13 ðxf ; hÞ 0 0

e act ¼ V act 
V V act PH2 ; P O2 ; I; T ; ~_ ¼ B
x @ 0 a22 ðhÞ 0
C~ B
Axf þ @ b21 ðhÞ 0
C
A
f
e
P H2 ¼ PH2  PH2 ðm
_ H2;in ; IÞ; ð27Þ 0 0 a33 ðhÞ 0 b32 ðhÞ
e O2 ¼ PO2  P ðm
_ O2;in ; IÞ; ~ :¼ Af ðxf ; hÞx
u ~:
~ þ Bf ðhÞu ð35Þ
P O2

In considering the objective of output voltage control or regula-

and
tion, the stack voltage by (1) and (2) is expressed as:
~_ H2;in ¼ m
m _ H2;in ;
_ H2;in  m
ð28Þ V stack ¼ NV cell ¼ NðENernst  V act  V ohmic Þ: ð36Þ
~_ O2;in ¼ m
m _ O2;in :
_ O2;in  m
Denote the deviation quantities,
The PEM fuel cell system dynamics (21) can be expressed in terms
 T e stack ¼ V stack  V  ;
V
~f ¼ V e act Pe H2 P
e O2 stack
of the deviations of state variables, x and
 T e ce11 ¼ V ce11  V  ;
V cell
~¼ m
deviations of control inputs, u ~_ H2;in m~_ O2;in , as defined in ð37Þ
e
E Nernst ¼ ENernst  ENernst ;
(27) and (28), respectively, by following the linearization of nonlin-
e ohmic ¼ V ohmic  V 
V
ear system dynamics around the equilibrium trajectories defined in ohmic :

(22)–(24).

Table 2
Operation conditions of the Ballard 5 kW PEMFC.

#1 #2 #3 #4 #5 #6 #7
I 0 50 100 150 200 250 300
Vstack 38.5152 28.3296 25.3307 22.7538 20.0849 17.0445 13.3636
Rload 1 0.5666 0.2533 0.1517 0.1004 0.0682 0.0445
Pstack 0 1.4165 2.5331 3.4131 4.0170 4.2611 4.0091
V act 0.0984 0.3637 0.4098 0.4363 0.4508 0.4690 0.4804
P H2 3 3 3 3 3 3 3
P O2 3 3.1395 3.2790 3.4186 3.5581 3.6976 3.8371
m_ H2;in 0 0.0091 0.0181 0.0272 0.0363 0.0453 0.0544
m_ O2;in 0 0.0136 0.0272 0.0408 0.0544 0.0680 0.0816
 P.-C. Chen / Energy Conversion and Management 65 (2013) 408–419 413

where the equilibrium and the deviation of stack voltage are writ- current and nominal temperature, whereas the lines denoted by
ten, respectively, as: ‘‘’’ are obtained from nominal current and varying temperature.
 The x-axis represents the deviation of operating points and the y-
V stack ¼ N ENernst  V act  V ohmic ; e stack
V axis denotes the numerical values of matrix elements. As shown,
¼ Nð e e act  V
E Nernst  V e ohmic Þ: ð38Þ the matrix elements a11(t), a12(t), a13(t), c3(t) vary along both the
values of current and temperature, while the elements of a22(t),
By introducing the deviations of the thermodynamic potential a33(t), b21(t), b32(t),c12(t) depend only on the values of temperature.
e e ohmic
E Nernst and ohmic overvoltage V Moreover, the values of these elements present linear dependence
  on the varying parameters I(t) and T(t). Therefore, the varying ele-
e RT 1 e 0:5 e e ohmic ¼ 0;
E Nernst ¼  P H2 þ  P O2 ; V ð39Þ ments and thus the parameter-dependent system matrices can be
2F P H2 PO2
well represented by a first-order Taylor expansion along the nom-
 T
the deviation of stack voltage in (38) can be expressed as: inal operation condition corresponding to ho ¼ Io T o .
 Let the system matrices in (35) and (40) be properly repre-
e stack ¼ c1
V ~f :¼ Cf ðxf ; hÞx
c2 ðxf ; hÞ c3 ðxf ; hÞ x ~f ; ð40Þ
sented as:
where the coefficients are 
@Af ðxf ; hÞ @Af xf ; h
NRT 1  0:5NRT 1 Af ðxf ; hÞ ffi Af ðxof ; ho Þ þ dI  jh¼ho þ dT  jh¼ho
c1 ¼ N; c2 ðxf ; hÞ ¼  ; c3 xf ; h ¼   : ð41Þ @I @T
2F PH2 2F P O2 o
:¼ Af ðxof ; h Þ þ dI  Af;I þ dT  Af;T ; ð42aÞ
It is noted that the derivations of the nonlinear PEM fuel cell
@Bf ðhÞ
dynamics and their LPV representation can be found coherently in Bf ðhÞ ffi Bf ðho Þ þ dT  j o :¼ Bf ðho Þ þ dT  Bf;T ; ð42bÞ
@T h¼h
the previous studies for the nominal state-feedback controller de-
sign in [26] and the nominal output-feedback controller design in @Cf ðxf ; hÞ @Cf ðxf ; hÞ
[30]. Cf ðxf ; hÞ ffi Cf ðxof ; ho Þ þ dI  jh¼ho þ dT  jh¼ho
@I @T
o o
:¼ Cf ðxf ; h Þ þ dI  Cf;I þ dT  Cf;T :
2.3. The parameters variation analysis of the PEM fuel cell system
The nominal system matrices in (42) can be obtained as:
For a nominal oxygen excess ratio knominal = 3 as given in Table 1, 0 1
4:2340  101 2:1423  102 4:1452  101
the equilibrium control inputs u⁄ are corresponding to the output B C
B C
current I = I⁄ as described in (25). Therefore, the parameter-depen- Af ðxof ; ho Þ ¼ B 0 3:8153  104 0 C;
@ A
dent variational elements in the system matrices of (35) and (40)
0 0 1:9076  104
can be determined solely by the output current I and stack temper-
ð43aÞ
ature T.
Let the equilibrium nominal operation condition correspond to 0 1
0 0
the output current in #4 operating point I = Io = 150A and stack B C
temperature regulated at T = To = 70 °C. During real-time operation,
o B
Bf ðh Þ ¼ @ 5:8697  105 0 C; ð43bÞ
A
the deviations of operation current and temperature away from the 5
0 2:9348  10
nominal values are denoted as dI(t) = I(t)  Io and dT(t) = T(t)  To.
Fig. 2 shows the values of non-zero elements in the system matri-  
Cf xof ; ho ¼ 3:5  101 1:7744  101 7:7856  102 :
ces Af(I(t), T(t)), Bf(T(t)) and Cf(I(t), T(t)) when the currents are
chosen as IðtÞ ¼ I;#1 I;#4 I;#7 ¼ ½ 0 150 300 A and the ð43cÞ
temperature are assumed as TðtÞ ¼ ½ 40 70 100  C. The lines According to the matrices elements defined in Section 2.2, the cur-
denoted by ‘‘h’’ represent the values corresponding to varying rent derivative matrices in (42) can be computed as:
0 1
2:3819  101 1:2052  103 1:8693  103
B C
Af;I ¼ B
@ 0 0 0 C;
A
0 0 0
ð44aÞ

Cf;I ¼ 0 0 6:3549  105 : ð44bÞ
The temperature derivative matrices in (42) can be obtained as:
0 1
2:3508  101 1:3984  103 6:3565  104
B C
Af;T ¼ B
@ 0 1:0808  102 0 C;
A
0 0 5:4041  101
ð45aÞ
0 1
0 0
B C
Bf;T B
¼ @ 1:6628  103 0 C; ð45bÞ
A
2
0 8:3140  10
Fig. 2. Variation of system matrix elements under three different values of current 
and temperature.
Cf;T ¼ 0 5:0265  104 2:2055  104 : ð45cÞ
414 P.-C. Chen / Energy Conversion and Management 65 (2013) 408–419

The PEM fuel cell described in (35) and (40) possesses system where the nominal matrices, current derivative matrices and tem-
matrices with dependence on the varying parameters, output cur- perature derivative matrices are:
rent I(t) and stack temperature T(t). For the purpose of PEM fuel 0 1 0 1 0 1
Af;o 0 Af;I 0 Af;T 0
cell control, the designated performance needs to be assured in
Ao ¼ @ A; A I ¼ @ A; A T ¼ @ A; ð50dÞ
the real-time operation for all values of varying parameters satisfy-
Cf;o 0 Cf;I 0 Cf;T 0
ing the deviation ranges dI(t) 2 [dImin, dImax] and dT(t) 2 [dTmin, -
dTmax], that need to be addressed during controller design. The    
Bf;o Bf;T
representation of parameter-dependent system matrices in (42) Bo ¼ ; BT ¼ ; ð50eÞ
has an affine or linear dependence on deviations of the varying
0 0
parameters.
Ce;o ¼ ð Cf;o 0 Þ; Ce;I ¼ ð Cf;I 0 Þ; Ce;T ¼ ð Cf;T 0 Þ: ð50fÞ
3. Robust integral voltage tracking control with varying system According to the operation conditions of the Ballard 5 kW PEM
parameters fuel cell as shown in Table 2, consider the quantities of system
variables corresponding to the operation conditions neighboring
3.1. The control problem formulation to the half load #4 operation condition. The nominal voltage track-
ing command in (48) is given as:
The objective of controller design for the PEM fuel cell systems n o
is to achieve a robust voltage tracking control, which takes into ac- jv d j 6 max V ;#3 ;#4 ;#4 ;#5
stack  V stack ; V stack  V stack ¼ 2:6689 :
count the variations of system matrices corresponding to the spec-
¼ v d;max : ð51Þ
ified range of output current and stack temperature. Specifically,
the variations of system matrices will be represented as affinely- Also, regarding the corresponding control efforts, the nominal mag-
dependent on the deviations of varying current and temperature nitude constraints on the hydrogen and oxygen input flowrates are
away from their nominal values. The magnitudes of the desired chosen as,
tracking command and the allowed gas flowrates are coordinately n o
e_ H2;in j 6 m
~ 1j ¼ jm
ju e_ H2;in;max ¼ max m_ ;#5 _ ;#4 _ ;#4 _ ;#3
considered during controller design. The voltage tracking perfor- H2;in  mH2;in ; mH2;in  mH2;in
mance is expressed in terms of H1 performance level regarding
the issued command to the tracking error. ~ 1;max ;
¼ 0:0091 :¼ u
The tracking error and integral tracking error for the voltage
control can be, respectively, written as n o
e_ O2;in j 6 m
~ 2j ¼ jm
ju e_ O2;in;max ¼ max m_ ;#5 _ ;#4 _ ;#4 _ ;#3
O2;in  mO2;in ; mO2;in  mO2;in
e stack ;
ze :¼ v d  V ð46Þ
and ~ 2;max :
¼ 0:0136 :¼ u
Z
e stack Þds; ð52Þ
ei :¼ ðv d  V ð47Þ

where vd is the voltage command. By introducing the expression of 3.2. The robust voltage tracking controller design
e stack from (40), the tracking error in (46) can also be written as:
V
The synthesized robust voltage tracking state-feedback control
ze :¼ v d  Cf ðxf ; hÞx
~f : law is denoted as:

By considering the integral tracking error as additional state, the ~ ¼ Kxp :

u ð53Þ
augmented state dynamics incorporating (35) and (47) can be writ-
The closed-loop controlled systems of the augmented state dynam-
ten as,
ics in (48) becomes:
~ þ Bd v d ;
x_ p ¼ Axp þ Bu ð48aÞ
x_ p ¼ ðA þ BKÞxp þ Bd v d :¼ Acl xp þ Bcl v d : ð54Þ
with
! During the controller optimization, the voltage tracking error in
      
~f
x Af xf ; h 0 Bf ðhÞ 0 (49) is the regulated output to be minimized, denoted as:
xp ¼ ;A ¼  ;B ¼ ; Bd ¼ :
ei Cf xf ; h 0 0 1 ze ¼ Ce xp þ De v d :¼ Ccl xp þ Dcl v d : ð55Þ
ð48bÞ
The tracking error to be regulated during controller design can be 3.2.1. The H1 voltage tracking performance
expressed as: The H1 performance is used to minimize the upper bound on
the energy ratio between tracking error and tracking command,
ze ¼ Ce xp þ De v d : ð49aÞ defined as kT v d ze k1 , supv d ðtÞ–0 kze ðtÞk2 =kv d ðtÞk2 . According to the
with Bounded-Real Lemma [25], the stabilization and H1 performance
  level kT v d ze k1 < c for the closed-loop system (54) and (55) can be
Ce ¼ Cf xf ; h 0 ; De ¼ 1: ð49bÞ achieved if the following matrix inequality holds under the posi-
tive Lyapunov matrix P,
Based on the representation of affine-dependence system
0 1
matrices in (42) for the PEM fuel cell system, the system matrices PAc1 þ ATc1 P PBc1 CTc1
in the augmented system (48) and (49) for controller design can be B C
@ BTc1 P cI DTcl A < 0: ð56Þ
accordingly expressed as:
Cc1 Dcl c
A :¼ Ao þ dI  AI þ dT  AT ; ð50aÞ
By denoting Q = P1, L = KQ, and multiplying the congruence matri-
B :¼ Bo þ dT  BT ; ð50bÞ
ces diag (Q, I, I) and diag (Q, I, I) from left and right hand side, the
Ce :¼ Ce; o þ dI  Ce; I þ dT  Ce; T ; ð50cÞ inequality (56) can be rewritten as:
 P.-C. Chen / Energy Conversion and Management 65 (2013) 408–419 415

0 1
AQ þ BL þ ðÞT ðÞT ðÞ T Linvar ðQ ; L; g; dI; dTÞ :¼ Linvar;o þ dI  Linvar;I þ dT  Linvar;T < 0; ð65aÞ
B C T
@ BTd cI ðÞ A < 0; ð57Þ where
Ce Q De cI !
Ao Q þ Bo L þ ðÞT þ gQ v 2d;max ðÞT
where () denotes for symmetry matrix block. By substituting the Linvar;o ¼ < 0; ð65bÞ
BTd gI
expressions of system matrices in (50), (57) can be expressed as:
!
L1 ðQ ; L; c; dI; dTÞ :¼ L1;o þ dI  L1;I þ dT  L1;T < 0; ð58aÞ AI Q þ ðÞT ðÞT
Linvar;I ¼ < 0; ð65cÞ
where 0 0
0 1
Ao Q þ Bo L þ ðÞT ðÞT ðÞT !
B CT AT Q þ BT L þ ðÞT ðÞT
L1;o ¼@ BTd cI ðÞ A; ð58bÞ Linvar;T ¼ < 0: ð65dÞ
0 0
Ce;o Q De cI
0 1
AI Q þ ðÞT ðÞT ðÞT 3.2.3. The gas flowrate constraints
B C The magnitude constraints on the control efforts of gas flowrate
L1;I ¼ @ 0 0 ðÞT A; ð58cÞ
are given as ju ~ij 6 u
~ i; max ; i ¼ 1; 2, where the values of u ~ i; max are
Ce;I Q 0 0
specified as in (52). For the voltage tracking control law in (53),
0 1 the magnitude constraint ju ~ij 6 u
~ i; max , is equivalent to
AT Q þ BT L þ ðÞT ðÞT ðÞT
B C xTp KTi Ki xp 6 u
~ 2i;max : ð66Þ
L1;T ¼ B
@ 0 0 ðÞT C
A: ð58dÞ
CQe;T 0 0 By (63) and P = Q , K = LQ 1 1
, the magnitude constraints in (66)
can be expressed as:

3.2.2. The state invariant set under magnitude bounded tracking 1

xT Q 1 LTi Li Q 1 xp < xTp Q 1 xp ;
command ~ 2i;max p
u
The time derivative of the Lyapunov function along the trajec-
tory of the closed-loop fuel cell dynamics (54) and (55) can be writ- which by Schur’s complement can be rewritten as the LMI condition
ten as:
!
Q LTi
h i Lu;i ðQ ; LÞ :¼ > 0: ð67Þ
V_ ¼ xTp PAcl þ ATcl P xp þ xTp PBcl v d þ v Td BTcl Pxp : ð59Þ Li ~ 2i;max
u

By the fact
4. Design results
1
xTp PBcl v d þ v Td BTcl Pxp 6 xTp PBcl BTcl Pxp þ gv Td v d ;
g 4.1. Controller construction
for all g > 0, the derivative of Lyapunov function in (59) is upper
bounded by: The integral voltage tracking control law (53) is constructed to
  achieve the optimal H1 performance level relating the command
1
V_ 6 xTp PAcl þ ATcl P þ PBcl BTcl P xp þ gv Td v d : ð60Þ input to the tracking error via the LMI (58), to address the magni-
g tude of input command via (65), and the constraints on gas flow-
Let the considered tracking command satisfy, rates via (67). The synthesis variables including the matrices
v d 6 kv d;max :¼ v d;max , where k is an adjustable positive parameter. Q > 0, L and scalars g > 0 are to be searched to minimize the H1
If the following inequality is true performance level c for all the deviation of varying parameters that
are located in the specified ranges dI(t) 2 [dImin, dImax] and
1 dT(t) 2 [dTmin, dTmax].
PAcl þ ATcl P þ PBcl BTcl P þ gPv 2d;max < 0: ð61Þ
g In the parameter-dependent matrix inequalities (58) and (65),
then the derivative of Lyapunov function as defined in (60) satisfies the feasibility must be established for all the values of varying
h i parameters dIðtÞ 2 ½dImin ; dImax  and dTðtÞ 2 ½dT min ; dT max . Due to
V_ < xTp gPv 2d;max xp þ gv 2d;max : ð62Þ the affine dependence of these matrix inequalities on the deviation
of varying parameters dI(t) and dT(t), only the vertices of the vary-
From (62), the condition of a decaying Lyapunov function V_ < 0, can ing parameters need to be evaluated, then via the property of con-
be obtained for xTp Pxp ¼ 1. Therefore, the state variables of the vex combination, the feasibility of these matrix inequalities can be
closed-loop controlled system will always be restricted to the fol- established for all the variation ranges. Therefore, the four vertices
lowing ellipsoidal invariant set of varying parameters

xTp Pxp 6 1: ð63Þ dhðtÞ ¼ fdIðtÞ; dTðtÞg

By the defined synthesis matrix variables Q = P ,L = KQ, the 1 2 ffdImin ; dT min g; fdImin ; dT max g; fdImax ; dT min g; fdImax ; dT max gg;
desired inequality (61) by using Schur’s complement [25] can be ð68Þ
written as:
! need to be evaluated for the these parameter-dependent matrix
AQ þ BL þ ðÞT þ gQ v 2d;max Bd inequalities (58) and (65). As in the parameters analysis in Sec-
< 0: ð64Þ tion 2.3, the nominal varying parameters are chosen as the out-
BTd gI
put current in #4 operating point I = Io = 150A and stack
By substituting the expression of system matrices in (50), (64) can temperature regulated at T = To = 70 °C. By considering the param-
be expressed as: eters ranges as IðtÞ ¼ I#1 I#4 I#7 ¼ ½ 0 150 300 A and
416 P.-C. Chen / Energy Conversion and Management 65 (2013) 408–419

TðtÞ ¼ ½ 40 70 100  C, the extreme values of the parameters tively. For the nominal design, the values of relative stability are
deviation can be obtained as: specified as a = [2.0, 1.8, 1.6, 1.4]. One the other hand, a milder rel-
ative stability a = [0.5, 0.45, 0.4, 0.35] is considered for the robust
dIðtÞ 2 ½dImin ; dImax  ¼ ½I#1  I#4 ; I#7  I#4  ¼ ½150; 150A; ð69aÞ design to obtain comparable H1 performance levels for the voltage
dTðtÞ 2 ½dT min ; dT max  ¼ ½40—70; 100—70 ¼ ½30; 30 C: ð69bÞ tracking capability.
Since the nominal oxygen excess ratio is utilized for the con-
The considered magnitude of tracking command is chosen as ducted gas flowrates control as mentioned in (25), the operated
one-tenth of the largest voltage interval regarding the neighboring oxygen should be knominal/2 times of the hydrogen. To achieve this,
operation conditions of the half load, #4, operation point. The mag- the control law relating the oxygen can be implemented as knominal/
nitude of tracking command in (65) is specified as v  d;max ¼ 2 times of the hydrogen. By the control law K = LQ1 with L and Q
0:1v d;max ¼ 0:26689 V, with vd,max described in (51). In order to ob- as synthesis matrix variables, the variable L can be structurally
tain a comparative H1 performance, the magnitude constraints on specified using the LMI Toolbox [31] as:
input gas flowrates are specified for u ~ i 6 ru
~ i; max with the scaling    
lh lh
factor chosen as r = [10,11,12,13,14,15] and the nominal magni- L¼ ¼ : ð70Þ
tude constraints u ~ 2;max ¼ 0:0136 (mols1) as men-
~ 1;max ¼ 0:0091; u lo knominal =2  lh
tioned in (52). The resulting control laws for both the nominal design Kit and ro-
Both the cases of nominal and robust control laws are con- bust design Krit, based on the synthesis matrix inequalities (58),
structed. In the nominal design, the terms related to parameter (65) and (67), are constructed, respectively, as the following
deviations dI and dT are simply zero. For the robust design, the !
parameter deviations as defined in (68) and (69) are evaluated in 1:9244  101 2:1635  106 5:1391  106 2:4021  101
Kit ¼ ;
the synthesis matrix inequalities. With the relative stability a as 2:8865  101 3:2453  106 7:6979  106 3:6032  101
design parameter, the achieved H1 performance levels for both ð71aÞ
the nominal and robust designs are shown in Fig. 3a and b, respec-
!
1:7098  101 8:9038  107 3:1613  107 1:0533  101
Krit ¼ :
2:5648  101 1:3356  106 4:7419  107 1:5799  101
ð71bÞ

4.2. Time response simulation

The controller constructed under the gas flowrate constraint

10u ~ i; max for both the nominal design with a = 2.0 and robust design
with a = 0.5 are also utilized for the time response simulation, with
the achieved H1 performance levels c = 2.1421 and c = 2.0129,
respectively, as shown in Fig. 3. In the simulation, the PEM fuel cell
system is initially assumed in its half load, # 4, operation condition
with equilibrium current I#4 = 150A. The external voltage tracking
command with magnitude as specified in controller design,
v dc ¼ v d;max ¼ 0:26689 V, is issued at 1 s as a step signal, which is
low-passed by a first-order filter with bandwidth 10 rads1 to ob-
tain the true tracking command vd.
The resulting time responses are shown in Figs. 4 and 5, where
Fig. 4 is based on the nominal PEM fuel cell system with constant
(a) nominal design system matrices and Fig. 5 is obtained by using the nonlinear PEM
fuel cell dynamics. In Fig. 5, the stack temperature T = To = 80 °C is
assumed and a sinusoidal-variation stack temperature
T = (80 + 10sin 0.5t)°C is used, which is also illustrated in Fig. 6.
For comparison, in the time responses of Figs. 4 and 5, the nominal
design using Kit is denoted with solid lines and the robust design
using Krit is denoted with pointed solid lines. Figs. 4a and 5a show
the external step command vdc, the true tracking command vd, the
deviation of stack voltage V e stack , the tracking error ze and the inte-
gral tracking error ei. Figs. 4b and 5b show the deviation of fuel cell
states, including the activation overvoltage V e act , partial pressure of
hydrogen P e H2 and partial pressure of oxygen P e O2 . Figs. 4c and 5c
show the deviation of control efforts, including hydrogen flowrate
me_ H2;in and oxygen flowrate m e_ O2;in . Figs. 4d and 5d show the stack
voltage Vstack, output current I and stack power Pstack.
In Figs. 4d and 5d, the value of stack voltage is obtained as
V stack ¼ V e stack þ V  e
stack ¼ V stack þ 22:7538, with the deviation of
stack voltage V e stack from (40). The output current I under the equiv-
alent load resistance of the half load #4 operating condition is
computed as I = Vstack/Rload = Vstack/0.1517. The corresponding stack
(b) robust design
power output is Pstack = VstackI.
Fig. 3. The integral tracking H1 performance level under varying bound on gas As shown in Figs. 4a and 5a, the controlled PEM fuel cells can
inputs. achieve a decaying tracking error ze under the suitable gas
 P.-C. Chen / Energy Conversion and Management 65 (2013) 408–419 417

(a) The Vfc , ze , and zi (b) The deviation of Vact , PH2 , and PO2

(c) The deviation of m H2,in and m O2,in (d) The stack voltage, output current, and stack power
Fig. 4. Time response simulations based on nominal PEM fuel cell model.

flowrates on gas flowrates specified during controller design. It is cell shown in Fig. 4a, the integral tracking errors are obtained as
true that the formulation of LMI algorithms admits a certain degree eiti ðtÞ ¼ 3:8357  101 and erit 1
i ðtÞ ¼ 8:5941  10 , respectively, for
of conservatism. The magnitude constraints 10u ~ i; max are chosen to the nominal and robust controllers. Additionally, when the time
obtain the constructed controllers as shown in (71). However, as varying temperature as shown in Fig. 6 is assumed, the integral
shown in Fig. 4c for time responses based on nominal PEM fuel tracking errors under the nonlinear PEM fuel cell dynamics for
cell, the ratio between the exerted gas flowrates and nominal the nominal controller are obtained as eiti ðtÞ ¼ 7:9420  101 and,
flowrate constraints u ~ i; max is computed as maxð m e_ H2;in =u
~ 1;max Þ ffi for the robust controller, are obtained as erit
i ðtÞ ¼ 1:7827, as seen
maxð m e_ O2;in =u
~ 2;max Þ ¼ 1:0088 obtained from using both the nominal in Fig. 5a.
and robust designed controllers. From Fig. 5c for time response As seen from Figs. 4 and 5, the nominal controller design can
with temperature variation under the nonlinear PEM fuel achieve faster time response under the same gas flowrates as al-
cell dynamics, the ratio is obtained as maxð m e_ H2;in =u
~ 1;max Þ ffi lowed to the robust design. In contrast, under the same gas flow
e_ ~
maxð m O2;in =u2;max Þ ¼ 2:1005. Therefore, it can be observed that rates, the robust controller can obtain guaranteed design perfor-
there are more gas flowrates exerted when the nonlinear PEM fuel mance under nonlinear PEM fuel cell dynamics with varying sys-
cell dynamics are used for evaluation. tem matrices. The responses for both the nominal and robust
Also, from the comparative time responses in Figs. 4 and 5, the controller designs exhibit integral tracking errors that are larger
nominal controller design Kit can achieve better response speed than when using nominal fuel cell dynamics. When the varying
than the robust controller Krit, which agrees with the designated stack temperatures of Fig. 6 are included in the simulation, the
relative stability and the control gain magnitudes of (71). In the time responses in Fig. 5 show that the system variables exhibit cor-
simulation based on the constant system matrices of the PEM fuel responding ripple effects.
418 P.-C. Chen / Energy Conversion and Management 65 (2013) 408–419

(a) The Vfc , ze , and zi (b) The deviation of Vact , PH2 , and PO2

(c) The deviation of mH2,in and mO2,in (d) The stack voltage, output current, and stack power
Fig. 5. Time response simulations with temperature variations based on nonlinear PEM fuel cell dynamics.

5. Conclusions

This paper has presented a robust control approach for proton

exchange membrane (PEM) fuel cell systems. To obtain a guaran-
teed design performance for a linear parameter varying system
representation of nonlinear PEM fuel cell dynamics, this study
has addressed the system uncertainties caused by system parame-
ter variations. The voltage tracking performance was expressed in
terms of H1 performance optimization with regard to the ratio of
the issued command to the tracking error. The controller was con-
structed numerically in terms of convex tractable linear matrix
inequalities. By the affinely dependent property of the formulated
parameter-dependent matrix inequalities related to various design
specifications, the desired performance was established by evalu-
ating only the matrix inequalities in the extremes of the varying
parameters. Both nominal and robust controller designs were
constructed. The designs were verified through time response
simulation for both the nominal PEM fuel cell and nonlinear PEM
Fig. 6. The variation of temperature during simulation. fuel cell dynamics.
 P.-C. Chen / Energy Conversion and Management 65 (2013) 408–419 419

Acknowledgment [15] Sedighizadeh M, Fathian K. Dynamic modeling and adaptive control of voltage
in proton exchange membrane fuel cell using water management. Int J Energy
Res 2011; doi:[Link]
This work was sponsored by the National Science Council of [16] Sedighizadeh M, Fathian K. Dynamic modeling and adaptive control of voltage
Republic of China under contract NSC-101-2221-E-244-003. in proton exchange membrane fuel cell using water management. Int J Energy
Res 2011; doi:[Link]
[17] Sedighizadeh M, Fathian K. Dynamic modeling and adaptive control of voltage
References in proton exchange membrane fuel cell using water management. Int J Energy
Res 2011; doi:[Link]
[1] Wang Y, Chen KS, Mishler J, Cho SC, Adroher XC. A review of polymer [18] Marzooghi H, Raoofat M. Improving the performance of proton exchange
electrolyte fuel cells: technology, applications, and needs on fundamental membrane and solid oxide fuel cells under voltage flicker using fuzzy-PI
research. Appl Energy 2011;88:981–1007. controller. Int J Hydro Energy 2012;37:7796–804.
[2] Corbo P, Migliardini F, Veneri O. Experimental analysis and management issues [19] Li X, Deng ZH, Wei D, Xu CS, Cao GY. Novel variable structure control for the
of a hydrogen fuel cell system for stationary and mobile application. Energy temperature of PEM fuel cell stack based on the dynamic thermal affine model.
Convers Manage 2007;48:2365–74. Energy Convers Manage 2011;52:3265–74.
[3] Nehrir MH, Wang C. Modeling and control of fuel cells distributed generation [20] Na WK, Gou B. Feedback-Linearization-based nonlinear control for PEM fuel
applications. New Jersey, USA: John Wiley & Sons; 2009. cells. IEEE Trans Energy Convers 2008;23:179–90.
[4] Bavarian M, Soroush M, Kevrekidis IG, Benziger JB. Mathematical modeling, [21] Li Q, Chen W, Wang Y, Jia J, Han M. Nonlinear robust control of proton
steady-state and dynamic behavior, and control of fuel cells: a review. Ind Eng exchange membrane fuel cell by state feedback exact linearization. J Power
Chem Res 2010;49:7922–50. Sour 2009;194:338–48.
[5] Cho J, Kim HS, Min K. Transient response of a unit proton-exchange membrane [22] Wu W, Xu JP, Hwang JJ. Multi-loop nonlinear predictive control scheme for a
fuel cell under various operating conditions. J Pow Sour 2008;185:118–28. simplistic hybrid energy system. Int J Hydro Energy 2009;34:3953–64.
[6] Tang Y, Yuan W, Pan M, Li Z, Chen G, Li Y. Experimental investigation of [23] Methekar RN, Patwardhan SC, Gudi RD, Prasad V. Adaptive peak seeking
dynamic performance and transient responses of a kW-class PEM fuel cell control of a proton exchange member fuel cell. J Process Control 2010;20:
stack under various load changes. Appl Energy 2010;87:1410–7. 73–82.
[7] Shan Y, Choe SY. Modeling and simulation of a PEMFC fuel cell stack [24] Kazmi H, Bhatti AI. Robust controller using LMI framework for PEM fuel cell
considering temperature effects. J Pow Sour 2006;158:274–86. system. In: 2009 International conference on emerging technologies,
[8] Zhang Z, Jia L, Wang X, Ba L. Effects of inlet humidification on PEM fuel cell December 9. p. 136–41.
dynamic behaviors. Int J Energy Res 2011;35:376–88. [25] Boyd S, Ghaoui LE, Feron E, Balakrishnan V. Linear matrix inequalities in
[9] Tafaoli-Masoule M, Shakeri M, Esmaili Q, Bahrami A. PEM fuel cell modeling system and control theory. Philadelphia, USA: SIAM; 1994.
and pressure investigation. Energy Sour, Part A 2011;33:2291–302. [26] Chen PC. The dynamics analysis and controller design for the PEM fuel cell
[10] Amphlett JC, Baumert RM, Mann RF, Peppley BA, Roberge PR. Performance under gas flowrate constraints. Int J Hydro Energy 2011;36:3110–22.
modeling of the Ballard mark IV solid polymer electrolyte fuel cell. J [27] Rugh WJ, Shamma JS. Research on gain scheduling. Automatica 2000;6:
Electrochem Soc 1995;142:9–15. 1401–25.
[11] Soltani M, Bathaee SMT. Development of an empirical dynamic model for a [28] Forrai A, Funato H, Yanagita Y, Kato Y. Fuel-cell parameter estimation and
Nexa PEM fuel cell power module. Energy Convers Manage 2010;51: diagnostics. IEEE Trans Energy Convers 2005;20:668–74.
2492–500. [29] Khan MJ, Iqbal MT. Modelling and analysis of electrochemical, thermal, and
[12] Mann RF, Amphlett JC, Hopper MAI, Jensen HM, Peppley BA, Roberge PR. reactant flow dynamics for a PEM fuel cell system. Fuel Cells 2005;5:463–75.
Development and application of a generalized steady-state electrochemical [30] Chen PC. Output-feedback voltage tracking control for input-constrained PEM
model for a PEM fuel cell. J Power Sour 2000;86:173–80. fuel cell systems. Int J Hydro Energy 2011;36:14608–21.
[13] Puranik SV, Keyhani A, Khorrami F. State-space modeling of proton exchange [31] Gahinet P, Nemirovski A, Laub A, Chilali M. The LMI control
membrane fuel cell. IEEE Trans Energy Convers 2010;25:804–13. toolbox. Massachusetts, USA: The MathWorks; 1995.
[14] Sedighizadeh M, Fathian K. Dynamic modeling and adaptive control of voltage
in proton exchange membrane fuel cell using water management. Int J Energy
Res 2011; doi:[Link]