Professional Documents
Culture Documents
C) [1].
First of all, an PEMFC system must be accurately modeled in
order to apply a suitable nonlinear control scheme. Models have
been reported so far in the literature for the PEMFC, ranging
from stationary and dynamic models [3][9], [13][17], [20],
[24] for the control design applied to a fuel cell vehicle and a
distributed generation system [11], [12].
Unfortunately, those models are mainly for experimental ver-
ications other than control design [1][3] or for prediction of
the fuel cell phenomenon by analyzing an electrochemical re-
action, the thermodynamics, and the uid mechanics. Recently,
Purkrushpan et al. [3] developed a control-oriented PEMFC
model that includes ow characteristics and dynamics of the
compressor and the manifold (anode and cathode), reactant par-
tial pressures, and membrane humidity. However, because of
the nonlinear relationship between stack voltage and load cur-
rent shown in Fig. 1 [1] and the state equations [3], [20], it is
a challenge to develop a nonlinear controller for the PEMFC.
Because of the operational parametric uncertainties such as the
parametric coefcients for each cell on kinetic, thermodynamic,
and electrochemical foundations, and the resistivity of the mem-
brane for the electron ow, the linear PEMFC models proposed
by Purkrushpan and coworkers [3][5] and Chiu et al. [20]
using Jacobian linearization via a Taylor series expansion at
the nominal operating point cannot easily achieve satisfactory
dynamic performance under large disturbances. An accurate
nonlinear dynamic model needs to be developed for the fuel
cell system as well as an advanced controller design technique,
considering the nonlinearity and uncertainty that need to be
proposed.
A fuzzy control system for a boost dc/dc converter of a fuel
cell system was developed in [18]. Neural optimal control was
presented for the PEMFC by using an articial neural network
(ANN) in [19]. However, instead of controlling the PEM fuel
cell system, the neural optimal control is mainly used to derive a
new architecture to synthesize an approximated optimal control
0885-8969/$25.00 2008 IEEE
by means of the ANN, where the PEM fuel cell was chosen as
a test bed.
In this paper, feedback linearization, a well-known nonlinear
approach, is applied to design a controller, based directly on
the nonlinear dynamic fuel cell model, to achieve more robust
transient behavior. Furthermore, the fuel cell stack life can be
prolonged and stack systems can be protected by minimizing the
deviations between the hydrogen and oxygen partial pressures
[2], [4].
In the last few years, feedback linearization for nonlinear dy-
namic models has been widely used [26][29], [31]. Feedback
linearization uses a nonlinear transformation to transform an
original nonlinear dynamic model into a linear model by diffeo-
morphism mapping [26][29], [31]. An optimal control theory
is also applied to obtain a linear control that is transformed back
to the original space by using the nonlinear mapping.
In this paper, the nonlinear dynamic model developed in
[3][5] and the small signal model of an PEMFC in [20] are
considered together to obtain a new dynamic nonlinear model
that is appropriate for developing a nonlinear controller. The
proposed controller, which is expected to perform rapid tran-
sient responses under load variations, is tested in MATLAB
simulink environment.
The paper is organized as follows. Section II gives a brief
introduction to feedback linearization. The PEMFC dynamic
model based on [3][5] and [20] is proposed in Section III, and
Section IV addresses the design of a nonlinear controller for
an PEMFC. Section V provides the simulation results for the
proposed controller, and Section VI concludes the paper.
II. NONLINEAR CONTROL BY FEEDBACK LINEARIZATION
For decades, signicant progress has been made in control
designs based on nonlinear concepts. In particular, nonlinear
control theory developed from differential geometry, known as
exact linearization or feedback linearization, has become more
and more attractive for chemical process control because many
chemical processes are basically of high nonlinearity [29], [34].
Hence, one of the main motivations of utilizing feedback lin-
earization for a fuel cell system is that the operation of PEMFC
is inherently a nonlinear chemical process. In this section, feed-
back linearization of nonlinear systems is briey introduced.
More details of nonlinear control based on differential geome-
try are available in [26][28], [34].
A. Feedback or Exact Linearization
Consider a single-input single-output (SISO) nonlinear sys-
tem described by the state equation
x = f(x) + g(x)u
y = h(x) (1)
where x is an n-dimensional state vector that is assumed to be
measurable, u is a scalar input, and y is a scalar output.
The objective of feedback linearization is to create a linear
differential relation between the output y and a newly dened
input v. The notation and concepts of differential geometry are
essential to understand this approach.
The Lie derivative of a scalar function h(x) with respect to a
vector function f(x) is dened as
L
f
h(x) = hf =
h(x)
x
f(x). (2)
Repeated Lie derivatives can be dened recursively as
L
k
f
h(x) = L
f
_
L
k1
f
h
_
=
_
L
k1
f
h
_
f
L
0
f
h(x) = h(x) (3)
for k = 1, 2, . . .
Similarly, in case of another vector eld g
L
g
L
f
h = (L
f
h)g. (4)
The output needs to be differentiated for r times until it is
directly related to the input u. The number r is called the relative
degree of the system.
The system is said to have a relative r at a point x
0
if
1) L
g
L
k
f
h(x) = 0 for all x in the neighborhood of x
0
and for
k < r 1;
2) L
g
L
r1
f
h(x
0
) = 0.
Thus, according to the aforementioned condition, with a de-
ned relative degree, the r time derivatives of y are described
as
y
(k)
= L
k
f
(x)h(x), for k = 0, 1, . . . , r 1,
y
(r)
= L
r
f
(x)h(x) + L
g
L
r1
f
h(x)u. (5)
The control law is
u =
1
L
g
L
r1
f
h(x)
_
L
r
f
h(x) + v
_
(6)
where y
(r)
= v. This control law can transform the nonlinear
system into a linear one. In addition, a nonlinear transformation
of a coordinate in the state space
z = (x) (7)
is called a local diffeomorphism, in which the map between
the new input v and the output is exactly linear for all x in the
neighborhood of x
0
.
This feedback linearization theory can be used to design
multiple-input multiple-output (MIMO) nonlinear system:
x = f(x) +
m
i=1
g
i
(x)u
i
y
i
= h(x), i = 1, 2, . . . , m (8)
where x is an n-dimensional state vector and u and y are m-
dimensional input and output vectors. The system is said to
have a vector relative degree {r
1
, r
2
, . . . , r
m
} at a point x
0
if
1) L
g
j
L
k
f
h
i
(x) = 0 for all 1 j m, all 1 i m, k <
r
i
1, and for all x in the neighborhood of x
0
;
2) mm matrix
A(x) =
_
_
L
g
1
L
r
1
1
f
h
1
(x) L
g
m
L
r
1
1
f
h
1
(x)
.
.
.
.
.
.
L
g
1
L
r
m
1
f
h
m
(x) L
g
m
L
r
m
1
f
h
m
(x)
_
_
(9)
is nonsingular at x = x
0
, which is called as a decoupling matrix.
Based on the dened relative degree, the control law of an
MIMO nonlinear system is dened as
u = A
1
(x)b(x) + A
1
(x)v (10)
where
b(x) =
_
_
L
r
1
f
h
i
(x)
.
.
.
L
r
m
f
h
m
(x)
_
_ v =
_
_
v
1
.
.
.
v
m
_
_ =
_
_
y
(r
1
)
1
.
.
.
y
(r
m
)
1
_
_
.
Note that the control lawin (10) transforms the nonlinear sys-
tem into a linear one in which the aforementioned inputoutput
relation is linearized and decoupled. In this paper, feedback lin-
earization for an MIMO nonlinear system is utilized due to the
MIMO dynamic nonlinear model of the PEMFC.
III. PEM FUEL CELLS AND DYNAMIC MODEL
A. PEM Fuel Cells
An PEM fuel cell consists of a polymer electrolyte mem-
brane sandwiched between two electrodes (anode and cathode).
In the electrolyte, only ions can exit and electrons are not al-
lowed to pass through. So, the ow of electrons needs a path
like an external circuit from the anode to the cathode to pro-
duce electricity because of a potential difference between the
anode and cathode. The overall electrochemical reactions for an
PEM fuel cell fed with a hydrogen-containing anode gas and an
oxygen-containing cathode gas are as follows:
Anode: 2H
2
4H
+
+ 4e
Cathode: O
2
+ 4H
+
+ 4e
2H
2
O
Overall: 2H
2
+ O
2
2H
2
O+ electricity + heat.
In practice, a 5 kW fuel cell stack, such as a Ballard MK5-E
PEMFC stack, uses a pressurized hydrogen tank at 10 atm and
oxygen taken from atmospheric air [10], [32].
On the anode side, a fuel processor, a so-called reformer,
that generates hydrogen through reforming methane or other
fuels like natural gas, can be used instead of the pressurized
hydrogen tank. Apressure regulator and purging of the hydrogen
component are also needed. On the cathode side, an air supply
system containing a compressor, an air lter, and an air ow
controller are required to maintain the oxygen partial pressure
[1], [4], [12], [25]. On both sides, a humidier is needed to
prevent dehydration of fuel cell membrane [1], [4], [25]. In
addition, a heat exchanger, a water tank, a water separator, and
a pump may be needed for water and heat management in the
FC systems [1], [4], [25].
To produce a higher voltage, multiple cells have to be con-
nected in series. Typically, a single cell produces voltage be-
TABLE I
CELL VOLTAGE PARAMETERS [1]
tween 0 and 1 V based on the polarization IV curve, which
expresses the relationship between stack voltage and load cur-
rent [1], [25]. Fig. 1 shows that their relationship is nonlinear
and mainly depends on current density, cell temperature, reac-
tant partial pressure, and membrane humidity [1], [25].
The output stack voltage V
st
[1] is dened as a function of the
stack current, reactant partial pressures, fuel cell temperature,
and membrane humidity:
V
st
= E V
activation
V
ohmic
V
concentration
. (11)
In the aforementioned equation, E = N
o
[V
o
+ (RT/2F)
ln(P
H
2
_
P
O
2
/P
H
2
O
c
)] is the thermodynamic potential of the
cell or reversible voltage based on the Nernst equation [1],
V
activation
is the voltage loss due to the rate of reactions on
the surface of the electrodes, V
ohmic
is the ohmic voltage drop
from the resistances of proton ow in the electrolyte, and
V
concentration
is the voltage loss from the reduction in concen-
tration gases or the transport of mass of oxygen and hydrogen.
Their equations are given as follows:
V
activation
= N
RT
2F
ln
_
I
fc
+ I
n
I
o
_
(12)
V
ohm
= NI
fc
r (13)
V
concentration
= Nm exp(nI
fc
). (14)
In (11), P
H
2
, P
O
2
, and P
H
2
O
c
are the partial pressures of hy-
drogen, oxygen, and water, respectively. Subscript c means the
water partial pressure, which is vented from the cathode side.
See Table I for cell voltage parameters.
Adetailedexplanationof eachvoltage loss canbe foundin[1],
and other voltage descriptions are also reported in [3], [4], [6],
[8], [10], where the fuel cell voltage is mainly expressed by the
combination of physical and empirical relationships in which
many parametric coefcients of the membrane water content,
humidity, and temperate, as well as the reactant concentrations
are involved. In our paper, the general voltage form as given
in (11) will be used because water and temperature factors are
assumed to be constant due to their slow response time.
Fig. 2. Illustration gas ows of PEMFC.
B. State-Space PEMFC Dynamic Model
The following assumptions are made when deriving the sim-
plied dynamic nonlinear model for the PEMFC.
1) Due to a slow response time regarding the stack tempera-
ture (about 102 s [1], [23], [25], [35]), the operating stack
temperature is assumed to be constant.
2) The fuel cell is well humidied on both the anode and
cathode sides.
3) The stoichiometry for hydrogen is set as 2 to provide
exibility in operation.
4) The stoichiometriy for oxygen ranges from 2 to 2.5 for
normal operation.
5) For the faster response (like a battery), this fuel cell model
has to have a continuous supply of reactants adjusted by
the current drawn from the stack.
6) For water management, it is assumed that the liquid water
does not leave the stack and that it evaporates into the
cathode or anode gas if humidity on either side drops
below 100% [3].
7) Perfectly controlled humidier and temperature con-
troller are assumed because of their slow response time
[3].
8) The mole fractions of the inlet reactants are assumed to be
constant to build the simplied dynamic PEMFCmodel. In
other words, pure hydrogen (99.99%) is fed to the anode,
and air that is uniformly mixed with nitrogen and oxygen
by a ratio of say 21:79 is supplied to the cathode.
The ideal gas law and mole conservation rule are applied
by supposing that all gases are ideal. The partial pressures of
hydrogen and water on the anode, and the partial pressures of
oxygen, nitrogen, and water on the cathode are dened as state
variables of the PEMFC. The relationship between inlet gases
and outgases is illustrated in Fig. 2 [20].
According to the ideal gas law and the mode conservation
rule, the partial pressure of each gas is balanced by the gas inlet
ow rate minus the gas consumption and the gas outlet ow
rate. The partial pressure derivatives are given as follows.
Anode mole conservation:
dP
H
2
dt
=
RT
V
A
(H
2in
H
2used
H
2out
)
dP
H
2
O
A
dt
=
RT
V
A
(H
2
O
Ain
H
2
O
Aout
H
2
O
mbr
). (15)
Cathode mole conservation:
dP
O
2
dt
=
RT
V
C
(O
2in
O
2used
O
2out
)
dP
N
2
dt
=
RT
V
C
(N
2in
N
2out
)
dP
H
2
O
C
dt
=
RT
V
C
(H
2
O
Cin
+ H
2
O
Cproduced
H
2
O
Cout
+ H
2
O
mbr
) (16)
where H
2in
, O
2in
, N
2in
, H
2
O
Ain
, and H
2
O
Cin
are the inlet ow
rates of hydrogen, oxygen, nitrogen, the anode side water, and
the cathode side water, respectively.
In addition, H
2out
, O
2out
, N
2out
, H
2
O
Aout
, and H
2
O
Ciout
are the outlet ow rates of each gas. H
2used
, O
2used
, and
H
2
O
Ciproduced
are the usage and the production of the gases.
Normally, the membrane water inlet ow rate H
2
O
mbr
across
the membrane is a function of the stack current and the
relative humidity. However, because we assumed that the
humidity is constant with the membrane average water content
m
= 14 [3], [21] under 100%of the relative humidity, H
2
O
mbr
is dened as a function of the stack current only: H
2
O
mbr
=
1.2684(NA
fc
I
fc
/F) [3], [21], where A
fc
(inverse square
centimeter) is the fuel cell active area, N is the number of the
fuel cells, and I
fc
the cell current density. All units of ow
rates, usages, and production of gases are dened as mole per
second. V
a
and V
c
are the anode and the cathode volumes,
respectively, and their units are cubic meter. Based on the basic
electrochemical relationships, the usage and production of the
gases are a function of the stack current:
H
2used
= 2O
2used
= H
2
O
Cproduced
=
NA
fc
I
fc
2F
. (17)
For simplicity, let us dene
NA
fc
2F
= C
1
, and 1.2684
NA
fc
F
= C
2
.
With the measured inlet ow rates and the stack current, the
outlet ow rates are given by the summations of anode and
cathode inlet ow ratesthat is, Anode
in
and Cath
in
, minus
the usage and production of gases. Anode
in
is dened by H
2in
+
H
2
O
Ain
, and Cath
in
is dened by O
2in
+ N
2in
+ H
2
O
Cin
.
The outlet ow rates on the anode side are
H
2out
= (Anode
in
C
1
I
fc
)F
H
2
H
2
O
Aout
= (Anode
in
C
2
I
fc
)F
H
2
O
A
. (18)
The outlet ow rates on the cathode side are
O
2out
=
_
Cath
in
C
1
2
I
fc
_
F
O
2
N
2out
= Cath
in
F
N
2
H
2
O
Cout
= (Cath
in
+ C
1
I
fc
+ C
2
I
fc
)F
H
2
O
C
(19)
where F
H
2
, F
H
2
O
A
, F
O
2
, F
N
2
, and F
H
2
O
C
are the pressure frac-
tions of gases inside the fuel cell, given as follows [20]:
F
H
2
=
P
H
2
P
H
2
+ P
H
2
O
A
F
H
2
O
A
=
P
H
2
O
A
P
H
2
+ P
H
2
O
A
,
F
O
2
=
P
O
2
P
O
2
+ P
N
2
+ P
H
2
O
C
F
N
2
=
P
N
2
P
O
2
+ P
N
2
+ P
H
2
O
C
F
H
2
O
C
=
P
H
2
O
C
P
O
2
+ P
N
2
+ P
H
2
O
C
.
(20)
To analyze the transient behavior of fuel cells, we take into
account the pressure fraction of each gas in the Department of
Energy (DoE) model [33]. The summation of anode and cathode
pressures is constant. The fuel cell model with the pressure frac-
tions of the gases was originally proposed by Chiu et al. [20].
In [20], only three pressure fractionsF
H
2
, F
O
2
, and F
H
2
O
C
are considered. In [3], [9], and [10], the outlet ow rates were
described by the difference between upstream pressures (par-
tial pressures) and downstream pressures (regulated pressures)
with a constant ow. In those models, the ow constants and
downstreams were set at 0.065 mol s
1
(in atmosphere) and 3
(in atmosphere), respectively. However, these xed values are
not appropriate for a dynamic model of a fuel cell. In our paper,
since all pressure fractions of gases, including the partial pres-
sures of water on the anode side and nitrogen on the cathode
side, are considered, a more accurate model and transient behav-
ior of fuel cells can be obtained, as compared with [3], [9], [10],
and [20].
The state equations as shown in (21) and (22) are obtained by
substituting (17)(20) into (15)(16).
State equations on the anode side:
dP
H
2
dt
=
RT
V
A
[H
2in
C
1
I
fc
(Anode
in
C
1
I
fc
)F
H
2
]
dP
H
2
O
a
dt
=
RT
V
A
[H
2
O
Ain
(Anode
in
C
2
I
fc
)F
H
2
O
A
C
2
I
fc
].
(21)
State equations on the cathode side:
dP
O
2
dt
=
RT
V
C
_
O
2in
C
1
2
I
fc
_
Cath
in
C
1
2
I
fc
_
F
O
2
_
dP
N
2
dt
=
RT
V
C
[N
2in
Cath
in
F
N
2
]
dP
H
2
O
C
dt
=
RT
V
C
_
H
2
O
Cin
+ C
1
I
fc
(Cath
in
+ C
1
I
fc
+ C
2
I
fc
)F
H
2
O
C
+ C
2
I
fc
_
.
(22)
For the input values H
2in
, O
2in
, and N
2in
, since the initial
mole fractions Y
H
2
, Y
O
2
, and Y
N
2
are set to be 0.99, 0.21, and
0.79, respectively [1], [3], and [9], the inputs are dened by the
mole fractions, given as:
H
2in
= Y
H
2
Anode
in
O
2in
= Y
O
2
Anode
in
N
2in
= Y
N
2
Cath
in
. (23)
The water inlet ow rates on the anode and the cathode are
expressed in terms of the relative humidity, saturation pressure,
and total pressure on each side, as follows [25]
H
2
O
Ain
=
a
P
vs
P
A
a
P
vs
Anode
in
H
2
O
Cin
=
c
P
vs
P
C
c
P
vs
Cath
in
, (24)
where
a
and
c
are the relative humidities on the anode and
the cathode sides, respectively; P
A
= P
H
2
+ P
H
2
O
A
is the sum-
mation of partial pressures of the anode; and P
C
= P
O
2
+
P
N
2
+ P
H
2
O
C
is the summation of partial pressures of the
cathode. P
vs
is the saturation pressure, which can be found
in the thermodynamics tables [30]. In addition, Anode
in
and
Cath
in
are dened as the product of input control variables u
a
and u
c
and the conversion factors k
a
and k
c
[9], [10] on each
side, which are from the standard litre per minutes (SLPM) to
mole per second:
Anode
in
= u
a
k
a
Cath
in
= u
c
k
c
. (25a)
These control variables are also affected by the hydrogen
and the air stoichiometry. However, the hydrogen stoichiome-
try does not have a big impact on the cell efciency in case of
operating with pure hydrogen [1]. Then, it can be assumed to
be constant such that a simple relationship between the fuel cell
efciency and its voltage is obtained. A typical air stoichiom-
etry ranges from 2 to 2.5, and its value is set as 2, which is
kept owing through the stack [23]. Hence, both reactants are
continuously fed to the fuel cell at sufciently high ow rate.
This fuel cell control systemis mainly dependent upon the input
control variables u
a
and u
c
. The inlet ow rates, Anode
in
and
Cath
in
, can be transformed into (25b) by multiplying each con-
stant stoichiometry to them. In designing the fuel cell control,
our main focus is on the control of hydrogen and oxygen par-
tial pressures, which can avoid unwanted pressure uctuation
and prevent the membrane electrode assemblies (MEAs) from
collapsing by minimizing the pressure difference between the
anode and the cathode, as follows:
Anode
in
= u
a
k
a
H
2
Cath
in
= u
c
k
c
air
. (25b)
By substituting (23)(25) into (21)(22), we can reconstruct
the state equations as follows.
New state equations on the anode side:
dP
H
2
dt
=
RT
V
A
_
u
a
k
a
Y
H
2
H
2
C
1
I
fc
(u
a
k
a
H
2
C
1
I
fc
)F
H
2
_
dP
H
2
O
a
dt
=
RT
V
A
_
_
u
a
k
a
H
2
a
P
vs
P
H
2
+P
H
2
O
A
a
P
vs
(u
a
k
a
H
2
C
2
I
fc
)F
H
2
O
A
C
2
I
fc
_
_
. (26)
New state equations on the cathode side:
dP
O
2
dt
=
RT
V
C
_
u
c
k
c
Y
O
2
air
C
1
2
I
fc
_
u
c
k
c
air
C
1
2
I
fc
_
F
O
2
_
dP
O
2
dt
=
RT
V
C
[u
c
k
c
Y
N
2
air
u
c
k
c
air
F
N
2
]
dP
O
2
dt
=
RT
V
C
_
_
u
c
k
c
air
c
P
vs
P
O
2
+P
N
2
+P
H
2
O
C
c
P
vs
+C
1
I
fc
(u
c
k
c
air
+C
1
I
fc
+C
2
I
fc
)F
H
2
O
C
+C
2
I
fc
_
_
.
(27)
IV. NONLINEAR CONTROL OF MIMO PEMFC
In this section, an MIMO dynamic nonlinear model of
PEMFC is derived from (26) and (27), and it is then used to
design a nonlinear controller by using feedback linearization in
order to minimize the difference P between the hydrogen and
oxygen partial pressures. The main purpose of keeping P in
a certain small range is to protect the membrane from damage,
and therefore, prolong the fuel cell stack life [2], [3]. In addi-
tion, the pressures have bigger impact on the performance of
fuel cells than other parameters [2], [3]. Because the fuel cell
voltage is a function of the pressures, each pressure needs to
be appropriately controlled to avoid a detrimental degradation
of the fuel cell voltage. To achieve this goal, it is necessary
to minimize the pressure deviation between the anode and the
cathode by using precise actuators like an accurate valve con-
troller. Normally, the optimal pressure controller consists of a
pressure sensor and a solenoid ow-control valve. In this pa-
per, we focus more on developing a pressure-control algorithm
for the whole fuel cell system instead of designing the pressure
sensors and ow controllers inside the fuel cell. In addition, the
fuel cell voltage is not considered as a control output because of
the characteristics of the VI polarization curve, which makes it
difcult to perform the voltage control in a short period of time.
Therefore, only the partial pressures of hydrogen and oxygen
are chosen as the outputs. The stack current is considered as a
disturbance to the system instead of an external input [20].
Consider the following MIMO nonlinear system with a
disturbance:
X = f(X) +
m
i=1
g
i
(X)u
i
+ p(X)d, i = 1, 2, . . . , m
y
1
= h
1
(X)
.
.
.
y
m
= h
m
(X) (28)
where X R
n
is the state vector, U R
m
is the input or
control vector, y R
P
is the output vector, and f(x) and g(x),
i = 1, 2, . . . , m, are n-dimensional smooth vector elds. The d
represents the disturbance variables, and p(x) the dimensional
vector eld directly related to the disturbance.
Considering (26) and (27), the dened outputs and the dis-
turbance, the nonlinear dynamic system model of PEMFC is
rewritten as follows:
X = f(x) + g
1
(x)u
1
+ g
2
(x)u
2
+ p(x)d
_
y
1
y
2
_
=
_
x
1
x
3
_
=
_
h
1
(x)
h
2
(x)
_
(29)
where
X =
_
_
P
H
2
P
H
2
O
A
P
O
2
P
N
2
P
H
2
O
C
_
_
; U =
_
u
a
u
c
_
; Y =
_
P
H
2
P
O
2
_
;
d = I
f ct
; f(x) = 0;
g
1
(x) = RT
H
2
_
_
k
a
Y
H
2
V
A
k
a
V
A
x
1
x
1
+ x
2
k
a
a
P
vs
V
A
(x
1
+ x
2
a
P
vs
)
k
a
V
A
x
1
x
1
+ x
2
0
0
0
_
_
g
2
(x) = RT
air
_
0
0
k
c
Y
O
2
V
C
k
c
V
C
x
3
x
3
+ x
4
+ x
5
k
c
Y
N
2
V
C
k
c
V
C
x
4
x
3
+ x
4
+ x
5
k
c
c
P
vs
V
C
(x
3
+ x
4
+ x
5
c
P
vs
)
k
c
V
C
x
5
x
3
+ x
4
+ x
5
_
_
p(x) = RT
C
1
V
A
+
C
1
x
1
V
A
(x
1
+ x
2
)
C
1
x
2
V
A
(x
1
+ x
2
)
C
1
V
A
C
1
2V
C
+
C
1
x
2
2V
C
(x
3
+ x
4
+ x
5
)
0
C
1
V
C
C
1
x
5
V
C
(x
3
+ x
4
+ x
5
)
C
2
x
5
V
C
(x
3
+ x
4
+ x
5
)
+
C
2
V
C
_
_
Equation (29) implies that the inputoutput behavior of the
system is nonlinear and coupled. In order to achieve the control
objective, two steps need to follow.
1) Obtaining a nonlinear control law that not only can com-
pensate nonlinearities but also can decouple and linearize
the input and output behaviors.
2) Imposing the poles of the closed loop so that the outputs
P
H
2
and P
O
2
track asymptotically the desired trajectory 3
atm by adding a proportional integral controller.
Fig. 3. Overall control block diagram of PEMFC.
From (29), an MIMO nonlinear system is ready to develop a
nonlinear control law. Normally, the disturbance d in (29) cannot
be directly used in the control design because an additional
necessary conditionthat the disturbance can be measured and
the feedfoward action is allowedhas to be satised [26], [34].
Otherwise, the linearized map between the new input v and the
output y does not exist. The condition renders the following
control law by using the measurement of the disturbance:
U = A
1
(x)b(x) + A
1
(x)v A
1
(x)p(x)d. (30)
As shown in (29), f(x) = 0 leads to b(x) = L
r
f
h(x) = 0, and
so the control law is written as
U = A
1
(x)v A
1
(x)p(x)d. (31)
Because each control variable u shows up after the rst deriva-
tive of each y
1
= x
1
and y
2
= x
3
, the relative degree vector
[r
1
r
2
] is [1 1], and the decoupling matrix A(x) is dened as
A(x) =
_
L
g
1
h
1
(x) L
g
2
h
1
(x)
L
g
1
h
2
(x) L
g
2
h
2
(x)
_
(32)
A(x)
=
_
k
a
Y
H
2
H
2
V
A
k
a
H
2
V
A
x
1
x
1
+x
2
0
0
k
c
Y
O
2
a i r
V
C
k
c
a i r
V
C
x
3
x
3
+x
4
+x
5
_
(33)
which is nonsingular at x = x
0
. Additionally, the matrix v and
p(x) in (31) are given as follows:
v =
_
y
1
y
2
_
p(x) = RT
_
C
1
V
A
+
C
1
x
1
V
A
(x
1
+ x
2
)
C
1
2V
C
+
C
1
x
2
2V
C
(x
3
+ x
4
+ x
5
)
_
_
.
(34)
The control law given in (30) yields decoupled and linearized
inputoutput behavior (see Fig. 3):
P
H
2
= v
1
P
O
2
= v
2
(35)
The outputs P
H
2
and P
O
2
are decoupled in terms of the
new inputs v
1
and v
2
. Thus, two linear subsystems, which are
between the input v
1
and the hydrogen partial pressure y
1
=
P
H
2
, and between the input v
2
and the oxygen partial pressure
y
2
= P
O
2
, are obtained. Furthermore, note that y
1
= x
1
and
y
2
= x
3
, and so in order to ensure that y
1
and y
2
are adjusted
to the desired values 3 (in atmospher) of y
1 S
and y
2 S
, the
stabilizing controller is designed by linear control theory using
the pole-placement strategy [27]. The new control inputs are
given by
_
v
1
v
2
_
=
_
y
1 S
k
11
e
1
y
2 S
k
21
e
2
_
where e
1
= y
1
y
1 S
and e
2
= y
2
y
2 S
.
Even though the nonlinear system PEMFC is exactly lin-
earized by feedback linearization, there may exist a tracking
error in the variation of the parameters, especially when the
load changes. To eliminate this tracking error, the integral terms
are added in the closed-loop error equation as in [27] and [34]:
_
v
1
v
2
_
=
_
_
y
1 S
k
11
e
1
k
12
_
e
1
dt
y
2 S
k
21
e
2
k
22
_
e
2
dt
_
_. (36)
From (36), the error dynamics can be obtained as follows:
e
1
+ k
11
e
1
+ k
12
e
1
= 0
e
2
+ k
21
e
2
+ k
22
e
2
= 0. (37)
By appropriately choosing the roots of the characteristics
of s
2
+ k
11
s + k
12
and s
2
+ k
21
s + k
22
, asymptotic tracking
is achieved, so that P
H
2
y
1 s
and P
O
y
2 s
as t .
The overshoots also become small by choosing k
2
11
4k
12
and
k
2
21
4k
22
[27], [34].
As shown in Fig. 3, the main objective of this control scheme
is to design a nonlinear controller by appropriately dening
a transformation mapping scheme that transforms the original
nonlinear system into a linear and controllable (closed) system,
at which point a linear controller can be designed using the
pole-placement technique for tracking purposes. However, the
control law in Eq. (30) will be unobservable because the entire
dynamics has a ve order (P
H
2
, . . . , P
H
2
O
C
), whereas only a
two order (P
H
2
and P
O
2
) are observed in the outputs. So, we
may have a problem of internal dynamics. In other words, the
internal dynamics of P
H
2
O
A
, P
N
2
, and P
H
2
O
C
must be stable
so that the states of the tracking controllers in (37) are held in
a bounded region during tracking. Otherwise, with external as
well as internal dynamics, this control law cannot enhance the
overall system performance. However, it is difcult to directly
determine the internal dynamics of the system because it is non-
linear, nonautonomous, and coupled to the external closed-loop
dynamics, as seen in (29)(37). In this paper, simulation results
are used to verify whether each state remains within the rea-
sonable bounded area. The comparison between the simulation
results and experimental data in [23] was used to verify the
performance of the control law.
V. MODEL VALIDATION AND SIMULATION RESULTS
In this section, the proposed dynamic PEMFCmodel and con-
trol method are tested through simulation in a MatlabSimulink
environment. For simplicity, the fuel processor, water and heat
management, and air compressor models are not considered in
Fig. 4. PEMFC stack based on PGS-105B system [23].
the simulation, as mentioned previously in Section III. Experi-
mental data in [23] are used to justify the validity of the proposed
dynamic PEMFC model.
The fuel cell system Ballard MK5-E-based PGS105B system
in Fig. 4 is used to test the proposed nonlinear controller. This
systemhas a total of 35 cells, connected in series, with a cell sur-
face area of 232 cm
2
. The MEA consists of a graphite electrode
and a Dow membrane. The reactant gases (hydrogen and air)
are humidied inside the stack, and the hydrogen is recirculated
at the anode while the air is owing through the cathode. The
hydrogen pressure is regulated to 3 atm at the anode inlet by a
pressure regulator, and a back-pressure regulator at the air out-
let also maintains 3 atm through the Ballard fuel cell stack. The
oxidant owrate is automatically adjusted to a constant value of
4.5 l s
1
, based on a programmable load, via a mass owmeter
to ensure sufcient water removal at the cathode. Hydrogen is
replenished at the same rate as it is consumed. The stack temper-
ature measured at the air outlet is maintained between 72
Cand
75