Continuous Stirred Tank Reactors
Continuous Stirred Tank Reactors
Continuous Stirred Tank Reactors
R. Antonelli
Centre for Process Systems Engineering
Imperial College of Science, Technology and Medicine
London, SW7 2BY, UK
A. Astol*
Dept. of Electrical and Electronic Engineering
Imperial College of Science, Technology and Medicine
London, SW7 2BT, UK
and
Dipartimento di Elettronica e Informazione
Politecnico di Milano
Milano, ITALY
April 4, 2001
*
Author to whom correspondence should be addressed. Fax:+44-20-7594-6282, E-
mail:a.astol@ic.ac.uk.
1
Abstract
The objective of this work is to make use of methodologies and tools
from nonlinear Lyapunov theory in the design of bounded control laws
for the temperature stabilisation of Continuous Stirred Tank Reactors
(CSTRs) with exothermic or endothermic reactions. Simple output feed-
back control laws, i.e. control laws that feed back only the available mea-
surements, are designed. It is shown that these controllers yield Global
Asymptotic Stability (GAS) and do not require precise knowledge of the
system parameters. The capabilities of the proposed method to handle
control input constraints and model uncertainty are also discussed. The
work is then completed with the analysis of two case studies: the classical
rst order irreversible reaction A ! B and the more complex van der
Vusse reaction mechanism.
2
1 Introduction
Many processes of industrial interest are dicult to control for their inherent
nonlinear behaviour, presence of input constraints and lack of measurements.
Among these systems, chemical processes have been widely studied and tradi-
tionally controlled using linear system analysis and design tools together with
linearised models 11, 32, 34], or nonlinear methods based on linearisation tech-
niques 3, 5, 16, 17, 22, 29, 31]. However, the use of linear techniques is very
limiting if the analysed chemical process is highly nonlinear, whereas nonlinear
control design needs, in general, too much information about the process.
Chemical reactors exhibit a challenging control problem due to their non-
linearity, chaotic behaviour and the presence of several stable and unstable
equilibrium points. The uncertainty aecting the kinetic parameters is also not
negligible: modelling of kinetic reactions is dicult and, as a result, the model
mismatch may be signicant. If the controller design utilises too much infor-
mation, the control performance may be severely degraded. Furthermore, most
nonlinear control techniques assume that all state variables are measured or ac-
curately estimated. In industrial practice, the estimation of unmeasured state
variables and unknown parameters is not a trivial problem. Usually, only the
temperatures and owrates are easily measurable and, although advanced meth-
ods for accurate measurement of the concentrations in a chemical reactor have
been developed, these have not been used frequently in industrial plants be-
cause of very high operating costs. Finally, the hard constraints imposed on the
control input may adversely aect the eciency of the controller and degrade
the overall performance of the control system.
During the past decade there has been a signicant increase in interest
by many industries in the study of appropriate control strategies for batch,
semi-batch or continuous reactors. Several researchers have studied the reactor
control problem 1, 2, 3, 4, 5, 6, 7, 11, 13, 15, 16, 17, 19, 23, 24, 28, 29, 30, 31,
35, 36].
Conventional Proportional-Integral-Derivative (PID) controllers have been
rst applied for the temperature regulation of chemical reactors. They are usu-
ally robust to model uncertainty since their design is partly independent from
the model of the system, but dependent on tuning. The temperature stabili-
sation of an open stirred tank reactor with a rst order exothermic reaction
was extensively studied using a proportional controller 11]. This process shows
3
three open loop steady states, two of which are stable and one unstable. Aris
and Amundson 11], studied the dynamic behaviour of a reactor and used the
linearisation method to show that the open loop unstable equilibrium point
becomes locally asymptotically stable in closed loop.
Feedback linearisation techniques have also been used. In particular, a full
state feedback linearisation solution is presented for an exothermic reactor with
one reaction 3, 31]. Global stabilisation by state feedback has been numerically
demonstrated using the coolant temperature as a manipulated variable and the
addition of a state observer 31]. Stabilisation by state feedback has been proved
with respect to a well dened domain 3]. Input-output linearising control, an
analytical design approach which aims to reduce the original nonlinear control
problem to a simpler linear control problem, has been extensively investigated
15, 30]. An input-output linearisation strategy for constrained nonlinear pro-
cesses is proposed by combining the benets of feedback linearisation and model
predictive control 30], but robustness to model uncertainty is not guaranteed
since feedback linearisation techniques require exact knowledge of the process
model.
Various solutions have been proposed to overcome this problem, in particular
the addition of an integral action to the controller obtained from input-output
linearisation has been proposed 29]. Lyapunov-based adaptive linearising tech-
niques have been successively developed 1, 2, 28]. First, global stabilisation by
state feedback of a reactor with an exothermic reaction was proved 1]. Then,
the same case was considered in presence of a state observer 2]. Finally, global
stabilisation of a reactor with complex kinetics was dealt with 28], still con-
sidering a state observer. Good control performance was obtained with a state
observer in the case of a reactor with two sequential exothermic reactions 16].
Important studies have been made by Alvarez et al. 3], who used nonlin-
ear geometric control tools to stabilise unstable steady states of a CSTR and
considered the performance of Proportional (P) and Proportional-Integral (PI)
controllers. Lately, they studied the behavior of nonlinear controllers for the
specic case of an exothermic, rst order, irreversible reactor 4]. Moreover,
they showed the global boundedness of trajectories and the practical stabili-
sation of a chemical reactor when driven by an input-output linearising type
feedback, under parametric uncertainty 5]. Finally, they proved that a CSTR
with unknown reaction rates can be stabilised by applying a high-gain feedback
law and introduced a state observer to solve the problem of input saturation
4
and poor closed loop performance 7].
Robust control techniques have been applied to a reactor with a rst or-
der exothermic reaction 5, 13, 17]. Semi-global practical stabilisation has been
proved in the case of parametric uncertainty on the kinetics by means of a
robust input-state linearising state feedback 13]. In recent years, interesting
attempts to include the process nonlinearities into an adaptive controller have
been made with application to a two sequential reactions: the zero dynamics
are analysed and a model reference adaptive controller is designed to control a
highly nonlinear CSTR 2, 16].
Still recently, some signicant results were also obtained for the robust global
stabilisation under input constraints. Viel et al. 36] designed a state feedback
controller to stabilise the temperature at an arbitrary set-point in spite of un-
certainties on the reaction kinetics and proposed a robust state observer to
overcome the problem of partial state measurement. Then they extended these
results to an exothermic chemical reactor considering the feed concentration
as additional input 35]. Another approach has been proposed by Ortega et al.
19]. Therein a semi-adaptive state feedback controller was applied in the region
of the state space where convexity can be used to reduce parameter uncertainty
and it is then frozen and switched to a robust scheme outside this region.
It must be noted that to be implemented, all the above control laws need
the on-line measurement of the full state of the system, represented by the set
of concentrations and the temperatures. In industrial practice it is very easy to
measure temperatures, but usually not concentrations.
The problem of state variable measurement or estimation can be overcome
if we select an output feedback controller 8, 9, 10] able to regulate and glob-
ally stabilise the reactor temperature at its set-point without requiring any
knowledge of the concentrations. Interesting results in this direction have been
obtained 6, 23, 24]. A rst order compensator with modelling error compensa-
tion is designed and its ability to stabilise the temperature of a chemical reactor
is proved 6]. Then, the robust stabilisation of a class of CSTRs with unknown
kinetics and control input saturations is achieved by means of modelling error
compensation techniques which consist of interpreting the modelling error signal
as a new state, whose dynamics are observable from output measurements 24].
Finally the stabilisation of chemical reactors by output feedback with PI-type
controllers has been reviewed and treated in detail 23].
The objective of this work is to use Lyapunov-based methods to design non-
5
linear output feedback control laws for the temperature stabilisation of a wide
class of chemical reactors. The proposed controllers yield robust stability and
guaranteed performance also in the presence of input constraints and uncer-
tainty on the kinetic parameters. The paper is organised as follows. Section 2
briey discusses the main features of chemical reactors and, in particular, fo-
cuses on the dynamic model of non-isothermal continuous stirred tank reactors.
Section 3 and Section 4 are respectively concerned with the problem statement,
some preliminary assumptions and an introductory lemma used to derive the
main results presented in Section 5, i.e. dynamic output feedback stabilising
controllers for single and multi-input systems. Section 6 presents some simula-
tions results based on two case studies: a simple irreversible rst order reaction
and the reaction mechanism known as van der Vusse reaction. Finally Section 7
discusses the main goal achieved in the paper giving some conclusions.
6
From a dynamic viewpoint, in chemical reactions, the actual reacted concentra-
tion of the generic component Aj can be determined by evaluating the reaction
rate, which may have a dierent numerical value according to the way it is de-
ned and measured. In general, the reaction rate is a nonlinear function of the
temperature T , usually expressed in Kelvin, and of the vector xr of the concen-
trations of the reacting species. It expresses the rate of reactant consumption
and product formation in the reactor. Usually the reaction rate is non-negative,
possibly a time-varying function. Just for simplicity, this work will consider the
case where the reaction rate vector r(xr T ) depends only upon xr and T and
is separable with respect to its arguments, i.e.
r(xr T ) = '(xr )k(T ) (2)
where the vector k(T ) 2 IRm is called rate constant of reaction and it is com-
posed of positive bounded functions of the temperature, while '(xr ) 2 IRm m
is a diagonal matrix whose elements are functions of the vector xr of the concen-
trations of the reactants. This is related to the vector x of the n concentrations
of chemical species involved in the reaction network (reactants and products),
by a projection operator, i.e.
xri = Pir x (3)
where Pir is the operator that projects the concentrations of the reactants of
the i-th reaction into the vector xri , such that xr = xr1 : : : xrm] 2 IRm . In
matrix form, one has Pir = diag(0 : : : 1 0 1 : : : 1 : : : 0). Moreover, note that
the vector k(T ) is a continuous and strictly positive function for all T > 0 and
that
lim+ k(T ) = 0
T !0
in accordance with the thermodynamic principle that a chemical reaction cannot
occur at absolute zero. In the common case when the process kinetics follow
the Arrhenius law, the reaction rate vector can be expressed as 16]
2 r 3 2 ;E =RT 3
66 '1(x1) 0 r : : : 0
77 66 kk01ee;E =RT 77
1
..
r(xr T ) = 66 0 '2(x2 ) : : : . 77 66 02 . 2
77 (4)
64 ..
. ..
. ::: 0 75 64 .. 75
0 0 : : : 'm (xrm) k0me;Em =RT
where k0i , for i = 1 : : :m, are the constant pre-exponential factors, Ei, for i =
1 : : :m, are the activation energies and R is the Joule constant. The functions
7
'i (xri), for i = 1 : : :m, are generally nonlinear functions of the concentration of
the reactants involved in the ith reaction, i.e.
Y
'i (xri ) = ij (xj ) i = 1 : : : m (5)
j 2r
where, in the presence of elementary reactions 21], ij (xj ) represents the mass
action law
ij (xj ) = xj ji ji 2 IN: (6)
The above considerations are important to better illustrate the main concepts
related to the dynamic behaviour of continuous non-isothermal stirred tank
reactors, as presented in what follows.
8
REACTANTS
MIXER
HEATING/COOLING
IN
REACTOR
Tj T
JACKET
OUT
CONTROL
PRODUCTS
Figure 1: Block scheme of a CSTR. The external jacket is used as a heat source
or sink. The reactor temperature T and the jacket temperature Tj are controlled
by means of the heating or cooling owrate.
the recent years 14, 18]. Unfortunately in many cases the structure of the reac-
tion kinetics still continues to be uncertain. The presence of many uncertainties
renders ineective a control scheme that uses too much information about the
process. Similarly to other research works in this area (see e.g. 23, 27]), in this
paper some assumptions have been introduced in the attempt to reduce the
model complexity. In particular we will assume the following.
Hp1) There is a complete uniformity of concentration and temperature within
the reactor and a complete uniformity of temperature inside the jacket.
Hp2) The reactor and jacket volumes are constant.
Hp4) The thermal exchange between the reactor and the jacket is expressed
by a constant global heat exchange coecient U and the amount of heat
retained in the reactor walls is considered negligible.
Hp5) The reaction enthalpies HRi are independent of temperature, positive
for endothermic reactions and negative for exothermic reactions.
Hp6) The heat capacities, cp1 and cp2, of the process uids, i.e. reactor content
and heat transfer medium, are constant.
Under the above assumptions, the material and energy balances applied to the
9
jacket reactor give the following mathematical model 20]:
8
>> x_ = Cr(xr T ) + VQ (xin ; x)
>> 1
< T_ = HrT r(xr T ) + UA t (T ; T ) + Q (T in ; T )
>> 1V1cp1
j V1 (7)
>> T_j = ; UAt (Tj ; T ) + D (Tjin ; Tj )
: 2 V2cp2 V2
that can be rewritten in compact form as
8
>> x_ = Cr(xr T ) + d(xin ; x)
<
>> T_ = B (xr T ) + q1 (Tj ; T ) + d(T in ; T ) (8)
: _
Tj = ;q2 (Tj ; T ) + u
where x 2 IRn is the vector of the n concentrations of reactants r and products p
xr 2 IRr is the vector of the r concentrations of reactants C 2 IRn m is the char-
acteristic matrix of the stoichiometric coecients r(xr T ) 2 IRm is the vector of
reaction kinetics, expressed as in equation (2) HrT = (; cp HR1) : : : (;HRm) =
m cpm
1 1
b1 b2 : : : bm] 2 IRm is the vector of reaction enthalpies T 2 IR+ is the reactor
temperature d 2 IR+ is the dilution rate, that is the ratio between the feed
stream owrate and the volume of the reactor xin 2 Rn is the vector of non-
negative reactor feed concentrations B (xr T ) = HrT r(xr T ) 2 IR is the global
reaction heat q1 2 IR+ is the side reactor heat transfer parameter q2 2 IR+
is the side jacket heat transfer parameter Tj 2 IR+ is the jacket temperature
T in 2 IR+ is the reactor feed temperature Tjin 2 IR+ is the jacket uid stream
feed temperature u 2 IR is the manipulated heat. By assumption Hp5), the
components bi of the vector of reaction enthalpies, HrT , are negative for en-
dothermic reactions and positive for exothermic reactions. Finally the physical
parameters of the model are listed in Table 1.
Let xss Tss Tjss uss be steady state values of the state and input variables
respectively, and dene, as in 6], the deviation variables
= x ; xss 2 IRn
= T ; Tss 2 IR
(9)
= Tj ; Tjss 2 IR
u = u ; uss 2 IR:
10
Parameter Description
At heat exchange area, m2
V1 reactor volume, m3
V2 jacket volume, m3
cp1 reactor content heat capacity, kJ=kg K
cp2 heat transfer uid heat capacity, kJ=kg K
1 reactor content density, kg=m3
2 heat transfer uid density, kg=m3
Q volumetric owrate of the feed stream, m3=min
D volumetric owrate of the heat transfer uid, m3 =min
U global heat exchange coecient, kJ=Km2 min
HRi heat of the i-th reaction at temperature T , kJ=mol
Table 1: Denition of the physical parameters for the CSTR model (7).
Dene
ir = Pir
i (10)
where Pir is as in (3), and (
r )T =
1r : : :
mr ] 2 IRm , then the system (8) can
be rewritten as
8
>>
_ = C r~(
r ) + d(
in ;
)
<
> _ = ;(q1 + d) + q1 + (Tss Tjss) + B~ (
r ) (11)
>: _ = q2 ; q2 ; q2 (Tjss ; Tss) + uss +
u
where (Tss Tjss) = q1 (Tjss ; Tss )+ d(T in ; Tss ), r~(
r ) = r(
r + xrss + Tss ),
66 " # 7
66 ' (xr )k e;E =RT ; e;E =RTss 777
2 2
13
As already observed, in industrial practice, concentration measurement devices
are expensive, while temperatures and owrates are simple to measure with
devices that are accurate and reliable.
In what follows the system (8) and the Assumptions 1 to 4 are considered.
The objective consists of nding, if possible, a static control law, described by
equations of the form
w = w(y ) (22)
where wT = u d T in] and y T = T Tj T in d], or a dynamics control law with
state , described by equations of the form
_ = ( y ) (23)
w = w( y )
such that the closed loop system (8)-(22) or (8)-(23) is globally asymptotically
stable or locally asymptotically stable with a computable bound on the region
of attraction. Moreover, the manipulated variables wT = u d T in] take value
inside a compact set D IR3 which depends on the specic application.
4 A Preliminary Result
This section provides a simple result, which is instrumental in proving the main
results of this work.
Lemma 1 Consider the system
z_ = f (z ) + g (z )u (24)
with state z 2 IRn and control u 2 IR. Suppose that there exist an unknown
constant control u = uss and a positive denite and proper function V (z ) such
that
V_ (z ) = @V f (z) + @V g (z)uss < 0
@z @z (25)
for all nonzero z . Then, for any u and u such that u uss u, there exists a
dynamic control law, which does not require knowledge of uss , namely
_ = (z )
(26)
u =
with (0) 2 (u u), such that the closed loop system (24)-(26) is stable in the
sense of Lyapunov, the variable z converges to zero and, along the trajectories
of the closed loop system, (t) = u(t) 2 u u].
14
Proof of Lemma 1 Consider the function
V (z ) + ! ( ) (27)
where
! () = ln(u ; )=k ; ln( ; u)( +1)=k (28)
with k > 0, = uuss;;uu < 0 and + 1 > 0. Note that V (z ) + ! () is positive
denite and proper in z = f(z ) 2 IR(n+1) j u < < ug. Then, dierentiation
with respect to time gives
V_ (z ) + !_ () = @V @! _
@z f (z) + g (z)] + @
(29)
= @V @V @! _
@z f (z) + g (z)uss] + @z g (z )( ; uss ) + @ :
Setting
_ = @V
@z g (z )()
with
() = k( ; u)( ; u) (30)
and noting that
; uss + @!
@ () = 0
yields
V_ (z) + !_ () = @V
@z f ( z ) + g ( z ) uss ] + @V g(z) ; u + @! ()
@z ss @
(31)
= @V f (z ) + g (z )uss ] < 0
@z
for all nonzero z . As a result the closed loop system
z_ = f (z) + g (z)u
_ = (z ) (32)
u =
is locally stable and any trajectory starting in the set z is such that the
variable z converges to zero and (t) = u(t) 2 u u] for all t. /
From equation (30) it is easy to show that, if
y = @V
@z g (z)
15
is the measurable variable, then the dynamic control law (30) may be interpreted
as a nonlinear PI controller given by
u = u uu(0) ; u] ; u u(0) ; u] nl (y)
(0) ; u ; u(0) ; u] nl (y )
h R i
where nl (y ) = exp k(u ; u) 0tf y (t)dt and u(0) is the value of the control
input at the initial time.
5 Main Results
In this section it is shown how the problem of asymptotic stabilisation for system
(24) can be solved using dynamic control laws which require only partial state
information and provide a bounded control action.
The result established in Lemma 1 will be instrumental in deriving the
control laws. However, as already discussed, the application of Lemma 1 requires
the property that the system can be stabilised with constant control action. This
16
property is closely related to the features of the temperature sub-system and,
in particular, to the degree of exothermicity of the reaction taking place in the
reactor. These are in turn related to the properties of the mappings B1 and B2
given in equations (14) and (16).
It must be noted that a CSTR can be controlled using dierent variables. To
begin with, the heat supplied or removed from the jacket, u, is considered to be
the manipulated variable and the jacket temperature Tj is assumed to be the
measured variable. Then, multivariable extensions, in which the manipulated
variables are the pair (u T in) or (u d) and the measured variables are (T Tj d),
are considered. In all cases the control laws provide bounded control action and,
in particular, the positivity constraints associated with some variables, e.g. the
dilution rate d, will be satised for all time.
Remark 4 The results described in this section are mainly global, i.e. under
certain assumptions, the closed loop system is globally asymptotically stable in
its domain of denition. The only exception is the case of the manipulated vari-
ables (u T in) and the measured variables (T Tj ) where only local stability, with
an estimation of the region of attraction, is proved. This is because, as detailed
in the proof of Proposition 2, the positivity constraints on the manipulated
variable T in cannot be satised in the whole state space.
Remark 5 It will be apparent from the forthcoming discussion that the condi-
tion needed to guarantee global or local asymptotic stability of the closed loop
system depend upon the control pattern. In particular, the assumption required
in the case of the single manipulated variable u can be substantially weakened
if other control variables, such as T in or d, are employed.
17
where B 2 = max B2 , M is as in (21), d is the dilution rate, 2 and 3 are as
8T 2IR+
in (19) and (20) respectively, and = max
8xr T
k(P r )T B1T k with P r as in (10).
Then there exist a constant feedback law
u = uss (35)
two real-valued functions, W (
) and V ( ), which are positive denite and
proper, and a real positive constant such that
W_ + V_ < 0
for all (
) 6= (0 0 0). Therefore, the constant control (35) globally asymp-
totically stabilises the system (11) at the equilibrium point (0 0 0).
Remark 6 Before presenting the proof of Proposition 1, it is convenient to
briey discuss condition (34). This condition implies that B 2 < d which is
always veried if the reaction network is globally endothermic (in fact, for en-
dothermic reactions B2 < 0 for all T 2 IR+ ). By a simple property of any
projection operator, condition (34) can be simplied as
k T 2 (d ; B ):
max B 1k<
r
8x T M 3
2
Moreover, kB1T k and B2 are bounded functions of their arguments and this
assumption can be satised if the dilution rate d is suciently large, which is
in agreement with common process industry experience. It is also worth noting
that the functions B1 and B2 and the variables M , 2 and 3 can be, in practical
applications, experimentally estimated.
Proof of Proposition 1 Let W (
) be as in Assumption 2 and let V ( ) be
dened as
V ( ) = q22 (T ; Tss )2 + q21 (Tj ; Tjss )2 = q22 2 + q21 2: (36)
Then
@W T
W_ = @x Cr(xr T ) + d(xin ; x)
@W T @W T
= @x Cr(xr T ss) + d(xin ; x) + @x Cr(xr T ) ; Cr(xr T ss )]
@W
;2 + @x kCr(xr T ) ; Cr(xr T ss)k
2
2 2 2 2
2 2
p p (40)
; q1q2 ; q1q2 2
; q2(d ; B2) ; 23M ; q2 k(P2) B1 k 2:
2 2 2 2 rT T 2
2 2
The rst three terms of equation (40) are non-positive for any non-zero value
of the variables
, and , while the sign of the fourth term is unknown and
depends on the choice of . However, this last term can be made negative if
there exists a > 0 such that
q2(d ; B2 ) ; 23M ; q2 k(P2) B1 k > 0
2 2 2 2 r T T 2
(41)
2 2
or equivalently
q22k(P r )T B1T k22 ; 2q2 2 (d ; B2) + 23 M 2 < 0 (42)
Such a indeed exists and it is given by
2(d ; B 2 )
= 2 > 0: (43)
q2 max k(P r )T B1T k
8xr T
In fact, by condition (34), one has
q22 k(P r )T B1T k2 ()2 ; 2q22 (d ; B2 ) + 23M 2 < 0: (44)
19
It is straightforward to conclude that the function
W (
) + V ( )
is a Lyapunov function for the system (11) with u = uss , hence the equilibrium
point (0 0 0) of the closed loop system (11)-(35) is GAS. /
Proposition 1 can be combined with the result established in Lemma 1 as ex-
pressed in the following corollary, in which, instead of the constant control
u = uss , a dynamic control law, which requires only knowledge of the jacket
temperature and of its set-point value, is used.
Corollary 1 Consider the CSTR described in equation (11) and suppose that
Assumptions 1 to 4 hold and that condition (34) is veried. Assume moreover
that u is the only manipulated variable and that u 2 (u u) and uss 2 u u].
Then the initialized dynamic output feedback control law
8
<u =
: _ (45)
= k ( ; u)( ; u) = k ( ; u)( ; u)(Tj ; Tjss )
with k > 0, is such that the equilibrium (0 0 0 uss) of the closed loop system
(11)-(45) is locally stable and, for any initial conditions (
(0) (0) (0) (0))
with (0) 2 (u u), (
(t) (t) (t)) converges asymptotically to (0 0 0) and the
control variable (t) 2 (u u).
Proof of Corollary 1 The proof is a simple consequence of the results proved
in Lemma 1 and Proposition 1. Simple manipulations show that the positive
denite and proper function
W (
) + V ( ) + !() (46)
where W (
) and V ( ) are as in the proof of Proposition 1,
! () = q1 ln(u ; )=k ; q1 ln( ; u)( +1)=k (47)
with
= uuss;;uu < 0
is such that
W_ + V_ + !_ ;2 k
k2 + 3 M k
kj j + q2B1 P r
; q2 (q1 + d ; B2) 2
+2q1 q2 ; q1q2 2
(48)
which is non-positive as shown in the proof of Proposition 1, hence the claim. /
20
The above corollary shows that, for the reactor (8), temperature regulation
can be achieved by a bounded control action without any knowledge of the
steady state control input uss , but only with the knowledge of the set-point
jacket temperature Tjss . This set-point value is in general simpler to compute
via experiments than uss .
The stabilising control has a very simple structure, however it requires that
condition (34), concerning the level of exothermicity of the reaction network, is
satised. This condition, which might be too restrictive, can be relaxed increas-
ing the complexity of the control structure, i.e. introducing extra manipulated
variables to compensate the physical limitations of the heat exchange when a
highly exothermic reaction takes place, as discussed hereafter.
21
Then there exists two positive constants uss and Tssin such that the equilibrium
(0 0 0) of the system (49) in closed loop with the feedback law
u = uss
(51)
T in = Tssin ; kin(T ; Tss )
is globally asymptotically stable. Moreover, if
in r
0 kin Tss 3 M q q+2 q (52)
2 1 2
with
= max k
k, then all trajectories of the closed loop system (49)-(51) such
that T (0) and Tj (0) are in the set
( in 2 )
H = (T Tj) j q22 (T ; Tss)2 + q21 (Tj ; Tjss)2 q22 Tkss
in
remain in the set H, and are such that
T in = Tssin ; kin (T ; Tss )
is non-negative for all t 0.
Proof of Proposition 2 We break up the proof in two steps.
Step 1: Global asymptotic stability. Consider the positive denite and
proper function
W (
) + V ( ) (53)
where W (
) and V ( ) are as in the proof of Proposition 1 and is a positive
constant to be determined. Then, along the trajectories of the closed loop system
(49)-(51) one has
W_ + V_ ;2 2 + 3 M k
kj j + q2B1
r ; q2 (q1 + d ; B2 ) 2
+q2 d
T in ; q1 q2 2 + 2q1q2
(54)
= ; 2 k
k2 + 3M k
kj j ; 2 k
k2 + q2B1 P r
2 2
;q2(q1 + d + kind ; B2) 2 ; q1q22 + 2q1q2
22
and, after simple manipulation,
2 32 2
6 r 7
r r T
3M q2(B1 P )
W + V ; 4 2 k
k ; r j j5 ; 2
; r
_ _ 6 2 7 2
2 2 2 2
2 2
(55)
; pq q ; pq q 2
1 2 1 2
2M 2 2q 2 k(P r )T B T k2
; q2(d + kind ; B2 ) ; 2 ; 2 2 1 2:
3
2 2
23
T in are the manipulated variables, that u 2 (u u) and T in 2 (T in T in ), and
that uss 2 (u u) and Tssin 2 (T in T in ).
Then the initialized dynamic output feedback control law
8
>> u =
>>
>< _ = k ( ; u)( ; u) = k ( ; u)( ; u)(Tj ; Tjss )
>> T in = ; kin (T ; Tss )
>>
>:
_ = k ( ; T in )( ; T in ) = k ( ; T in )( ; T in )(T ; Tss )
(58)
with k > 0, kin 0 and k > 0, is such that the equilibrium point (0 0 0 uss Tssin)
of the closed loop (49)-(58) is locally stable and, for any initial conditions
(
(0) (0) (0) (0) (0)) with (0) 2 (u u) and (0) 2 (T in T in ), (
(t) (t) (t))
converges asymptotically to (0 0 0) and the control variables are such that
(t) 2 (u u) and (t) 2 (T in T in ).
Moreover, if r
in
kin Tss 3 M q q+1 q
2 1 2
then every trajectory of the closed loop system starting in H remains in H and
T in 0 for all t 0.
24
heat u, and introduce the new deviation variable
d = d ; dss 2 IR: (59)
As a result, the system (8) becomes
8 _
>>
= C r~(
r ) + d(
in ;
)
< _
= ;(q1 + dss +
d ; B2 ) + q1 +
d(T in ; Tss ) + B1 P r
>> (60)
: _ = ;q2 + q2 +
u
In the following proposition it is shown how this system can be stabilised at the
desired equilibrium and how condition (34) is modied by the introduction of
a further control input, namely the dilution rate d.
Proposition 3 Consider the CSTR described in (60) and suppose that As-
sumptions 1 to 4 hold. Consider the reaction heat B~ (
r ) as in (12), i.e.
B~ (
r ) = B0 + B1
r + B2: (61)
Suppose that
B2(T in) + 3 M in dss (62)
2
where B2 (T in ) is as in (16) with T = T in , d is the dilution rate, 2 is as in (19),
3 is as in (20), M is as in (21), in = max 8xr
k(P r)T B1T (T in)k and B1 (T in) is
as in (14) with T = T in . Then there exist a feedback control law
u = uss
d = dss + kd sat(T ; Tss )sat(T ; T in)
(63)
with kd = min(dss ; d d ; dss ) and sat the saturation function, two real-valued
functions, W (
) and V ( ), which are positive denite and proper, and a real
constant such that
W_ + V_ < 0
for all (
) 6= (0 0 0). Therefore, the control law (63) globally asymptotically
stabilises the system (60) at the equilibrium point (0 0 0).
Proof of Proposition 3 The proof is similar to the Proof of Proposition 1.
Consider the positive denite and proper function
W (
) + V ( ) (64)
25
where W (
) and V ( ) are as in Assumption 2 and equation (36) respectively.
Dierentiating the function (64) with respect to time yields
W_ + V_ ; 22 k
k2 + 3M k
kj j ; 22 k
k2 + q2B1P r
;q2(q1 + dss ; B2) 2 ; q1q22 + 2q1q2 + q2(T in ; T )
d
(65)
and, after some manipulation, one has
2 32 2
r r
_W + V_ ; 664 2 k
k ; r3 M j j775 ; 2
; q2(r
B1 P r )T
2
2 2 2 2 2
2 2
; pq1q2 ; pq1q2 2
; q2 dss + ((TT ;; TT ))
d ; B2 ; 23M ; q2 k(P2) B1 k 2:
in 2 2 2 2 r T T 2
ss 2 2
(66)
It is easy to check that equation (66) is non-positive if
q2 dss + ((TT ;; TT ))
d ; B2 ; 23M ; q2 k(P2) B1 k > 0
in 2 2 2 2 r T T 2
(67)
2 2
6 Case Studies
The control methodology developed so far is applied to stabilise the temperature
in two chemical reactors. It is assumed that the classic rst order irreversible
reaction A ! B takes place in the rst case study, while in the second the
reaction network is modelled in accordance with the complex van der Vusse
reaction mechanism. The performance of the proposed output feedback con-
trol laws is evaluated in both cases through numerical simulations and some
comparisons with the result in 24] and a classical PI. As main advantage, the
control strategy shows the capability to handle input constraints, still maintain-
ing the computational simplicity typical of conventional regulators. The results
illustrate that the controllers lead to good performance in terms of overshoot,
settling time and control eort.
6.1 A ! B Reaction
Consider the dynamic behavior of a classic example of continuous stirred tank
reactor in which a rst order, irreversible endothermic or exothermic reaction
A ! B takes place. The reactor is continuously fed by a liquid ow with the
temperature T in containing the reactant A with the concentration xin . In the
liquid phase inside the reactor, which is assumed to be of constant volume, the
component A reacts giving the desired product B . The control problem is to
27
370 16
360 14 u
350 12
Tj
340 10
u (K / min)
T (K)
330 T 8
320 6
310 4
u
300 2
290 0
0 5 10 15 20 25 30 35 40 45 50 0 5 10 15 20 25 30 35 40 45 50
Time (min) Time (min)
Figure 2: Simulation results for the chemical reactor (69) with endothermic
reaction A ! B when the output feedback control law (70) is active. Reactor
and jacket temperature (left) and control action (right) for k = 0:07 min=K
and dierent initial values of the controller state (K=min): (0) = 1:1u
(dashed line), (0) = 0:5(u + u) (solid line) and (0) = 0:9u (dotted line).
perform reactor start-up and operate the reactor at the steady state (Tss Tjss)
by controlling the inlet uid owrate to the jacket. Let E be the activation
energy and k0 the constant pre-exponential factor. The chemical reaction rate
is assumed to follow a rst order Arrhenius type kinetic
E
k(T ) = k0 exp ; RT
and the reactor is modelled in accordance with the Hypothesis Hp1 to Hp6
considered in Section 2.1. In this special case the system (8) becomes
8
>> x_ A = ;k(T )xA + d(xin
A ; xA )
>>
< x_ B B ; xB )
= k(T )xA + d(xin
>> T_ (69)
= bk(T )xA + q1 (Tj ; T ) + d(T in ; T )
>>
: T_j = ;q2 (Tj ; T ) + u
where xA and xB are the concentrations of the component A and B , and simple
considerations show that Assumptions 1 to 4 of Section 3 hold.
28
6.1.1 Endothermic Case (b < 0)
The output feedback control law (45), namely
8<
u =
: _ (70)
= k ( ; u)( ; u)(Tj ; Tjss )
is implemented on the reactor described by the equations (69) and the closed
loop process is simulated, assuming for the model parameters the following
values (see 20]): xA (0) = 100 mol=m3, xB (0) = 0 mol=m3, T (0) = 303 K ,
Tj (0) = 298 K , k0 = 1:3677 108 min;1 , E=R = 7290:11 K , d = 0:01 min;1,
xinA = 100 mol=m3, xinB = 0 mol=m3, T in = 300 K , b = ;0:01196 Km3=mol,
q1 = 0:191 min;1, q2 = 0:574 min;1, Tss = 350 K , Tjss = 352:6 K , xssA = 7:5
mol=m3, u = 15 K=min, u = 0:5 K=min.
Figures 2 and 3 show the jacket and reactor temperatures of the controlled
system and the control input behaviour for dierent initial values of the con-
troller state and for dierent values of the controller gain k , respectively. It
should be noted that the manipulated variable u is always bounded between
the upper and lower bounds, represented by the dotted lines. Observe also the
ability of the controller to stabilise the jacket and reactor temperatures in a rel-
atively short time and the good control performance achieved. It is worthwhile
to note that the only information on the model required to implement the con-
troller is the jacket temperature set-point Tjss . Finally, the nonlinear controller
(70) yields superior regulatory performance (see Figure 4) when compared to a
standard saturated PI controller
1 Zt
u(t) = u + k e(t) +
c
0
e(t )dt0 (71)
0
I 0
with e(t) = Tj (t) ; Tjss and tuning parameters kc = 0:24 min;1 K ;1 and
I = 3 min.
29
370 16
360 14
u
350 Tj 12
340 10
u (K / min)
T (K)
330 8
T
320 6
310 4
u
300 2
290 0
0 5 10 15 20 25 30 35 40 45 50 0 5 10 15 20 25 30 35 40 45 50
Time (min) Time (min)
Figure 3: Simulation results for the chemical reactor (69) with endothermic
reaction A ! B when the output feedback control law (70) is active. Reactor
and jacket temperature (left) and control action (right) for (0) = 0:5(u + u)
K=min and dierent values of the controller gain k (min=K ): k = 0:03
(dotted line), k = 0:07 (solid line) and k = 0:09 (dashed line).
380 16
370 T (PI)
j 14 u
Tj
360
12
350
10
u (K / min)
T
T , T (K)
340
T (PI)
8
j
330
6
320
PI
4
310
u
300 2
290 0
0 50 100 150 0 50 100 150
Time (min) Time (min)
30
measurements, the controller (58), i.e.
8
>> u =
>>
>< _ = k ( ; u)( ; u)(Tj ; Tjss )
>> T in (72)
>> = ; kin (T ; Tss )
>:
_= k ( ; T in )( ; T in )(T ; Tss )
may be implemented and interpreted as an extension of the dynamic scheme
(70) with the addition of a second update law that keeps the estimate of the
unknown parameter Tssin inside the interval limited by the upper and lower
bounds T in T in ]. Alternatively, the dilution rate d may be considered as an
additional control variable and the dynamic output feedback controller (68),
i.e. 8
>> u =
>>
>< _ = k ( ; u)( ; u)(Tj ; Tjss )
>> d (73)
>> = + kd sat(T ; Tss)sat(T ; T in )
>:
_= k (T in ; Tss )( ; d)( ; d)(T ; Tss )
implemented. The closed loop system with the controllers (70), (72) and (73) is
simulated in order to examine the performance of the controllers in the presence
of the input constraints and to show the considerable improvement in the closed
loop response for the case of an exothermic reaction when the multivariable
controllers (72) and (73) are used. Figures 5, 6 and 7 show the start-up proles
of the measured outputs and manipulated input for the exothermic system (69)
when the control input u, the control inputs (u T in ) and, nally, the control
inputs (u d) are considered. In the presence of tight input constraints, the
single input controller (70) shows very poor performance while the controllers
(72) and (73) exhibit signicantly better performance and faster settling times.
The solid, dashed and dotted lines refer to the results of the implementation of
the controllers (70), (72) and (73), respectively. For the exothermic reactor, the
process parameters are (see 20]): xA (0) = 100 mol=m3, xB (0) = 0 mol=m3,
T (0) = 380 K , Tj (0) = 370 K , (0) = ;5 K=min, (0) = 320 K , (0) = 1:8
min;1, k0 = 15:155 108 min;1 , k = 0:01 min= K 2, k = 0:01 K ;2min;1 ,
k = 0:01 K ;2, 0 = 7216:74 K , xinA = 100 mol=m3, xinB = 0 mol=m3,
b = 0:00717 K m3=mol, q1 = 0:191 min;1, q2 = 0:574 min;1 , Tss = 350 K ,
31
385
380
375
370
365
T (K)
360
u
355
(u,d) in
(u,T )
350
345
340
0 20 40 60 80 100 120 140 160 180 200
Time (min)
Tjss = 349:19 K , Tssin = 360 K , dss = 0:0144 min;1 , uss = ;0:4636 K=min,
xssA = 0:85 mol=m3, kin = 20, kd = 0:0134, T in = 400 K , T in = 280 K ,
u = ;0:1 K=min, u = ;10 K=min, d = 3 min;1 , d = 0:001 min;1 .
6.1.3 Comparison between di
erent controllers
In this section, the controller (70), tested through computer simulations in the
previous paragraph, is compared with a similar controller proposed by F. Jadot,
G. Bastin and F.Viel 24]. In their work, robust global stabilisation of stirred
tank reactors is achieved by means of saturated output feedback control laws.
The process considered is again the rst order, irreversible exothermic reaction
X1 ! X2, but the model studied is slightly dierent from the system (69), since
it does not take into account the jacket dynamics. The exothermic reactor is
described in this case by a dynamic model of the form
8
>> x_ 1 = ;r(x1 y ) + d(xin
1 ; x1 )
<
>> x_ 2 2 ; x2 )
= r(x1 y ) + d(xin (74)
: y_ = r(x1 y ) ; (d + e)y + u
32
375
370
365
360
355 in u
Tj (K)
(u,d) (u,T )
350
345
340
335
330
0 20 40 60 80 100 120 140 160 180 200
Time (min)
−1
−2 in
(u,T )
−3 (u,d)
−4
u (K / min)
−5
−6
−7
u
−8
−9
−10
Figure 7: Comparison of the control input u proles for the exothermic reaction
A ! B described in (69) with the control laws (70) (solid line), (72) (dashed
line) and (73) (dotted line), respectively.
33
380 100
90
375
80
370
70
365
Conc. (mol / m )
3
60
T (K)
360 50
40
355
30
350
20
345
10
340 0
0 5 10 15 20 25 30 0 5 10 15 20 25 30
Time (min) Time (min)
180 m
σ
175
(K / min)
170
jbv
u
165
u
0
160
0 5 10 15 20 25 30
Time (min)
34
where x1 and x2 are the concentrations of the component X1 and X2 , r(x1 y )
is the reaction rate, expressed by the Arrhenius law
r(x1 y ) = x1 1 exp(;2 =y )
y is the reactor temperature, d is the dilution rate, xin1 is the reactor feed
concentration, is the reaction enthalpy, e is the heat transfer coecient and
u is the manipulated heat.
It is easy to understand the relationship between the symbolism used respec-
tively in system (69) and in system (74). In particular, the manipulated variable
selected by Jadot et al., and renamed here for simplicity u = ujbv , is equivalent
to
36
Parameter Value Description
k10 1:287 1012 h;1 collision factor for reaction k1
k20 1:287 1012 h;1 collision factor for reaction k2
k30 9:043 109 1=molA h collision factor for reaction k3
E1 ;9:758:3 K activation energy for reaction k1
E2 ;9:758:3 K activation energy for reaction k2
E3 ;8560 K activation energy for reaction k3
HAB 4:2 kJ=molA h enthalpies of reaction k1
HBC ;11:0 kJ=molB h enthalpies of reaction k2
HAD ;41:85 kJ=molA h enthalpies of reaction k3
0:9342 kg=l density
Cp 3:01 kJ=kg K heat capacity
kw 4032 kJ=h m2 K heat transfer parameter for cooling jacket
AR 0:215 m2 surface of cooling jacket
VR 0:01 m3 reactor volume
mK 5:0 kg coolant mass
CPK 2:0 kJ=kg K heat capacity of coolant
Table 3: Physical parameters for the van der Vusse reactor (77).
37
12
10
6
xB
0
(xss,xss)
A B
0 2 4 6 8 10 12
xA
7 Conclusions
This work has considered the output feedback regulation of endothermic and
exothermic chemical reactors. In particular, the problem of global temperature
38
180
170
160
150
T (C)
140
130
(u,Tin)
120
(u,d)
110
u
100
0 10 20 30 40 50 60
Time (min)
Figure 10: Comparison of the reactor temperature T proles for the van der
Vusse reaction described in (77) with the control laws (70)(solid line), (72)
(dashed line) and (73) (dotted line), respectively.
150
140
130
Tj (C)
120 (u,Tin)
(u,d)
110
100 u
90
0 10 20 30 40 50 60
Time (min)
Figure 11: Comparison of the jacket temperature Tj proles for the van der
Vusse reaction described in (77) with the control laws (70)(solid line), (72)
(dashed line) and (73) (dotted line), respectively.
39
0
(u,d) u
u (KJ / min) −50
in
(u,T )
−100
−150
0 10 20 30 40 50 60
Time (min)
Figure 12: Comparison of the manipulated variable u proles for the van der
Vusse reaction described in (77) with the control laws (70)(solid line), (72)
(dashed line) and (73) (dotted line), respectively.
1.4
(u,d)
1.2
(u,Tin)
0.8
xB (mol / l)
0.6
0.4
0.2
0
0 10 20 30 40 50 60
Time (min)
Figure 13: Comparison of the product B concentration xB proles for the van
der Vusse reaction described in (77) with the control laws (70)(solid line), (72)
(dashed line) and (73) (dotted line), respectively.
40
control for a general class of chemical reactors has been addressed and solved
using Lyapunov techniques. A nonlinear reactor model has been used to synthe-
size output feedback control laws, i.e. feedback control laws which generate the
control signals on the basis of the available measurements only. Under reason-
able and easy to test assumptions, global temperature control has been achieved
in presence of general and tight control bounds. The proposed approach oers
some important advantages including constraint handling capability and com-
putational simplicity: the only condition to be fullled is a simple and natural
feasibility condition, namely that the steady state control is admissible. The
result is that the diculty in the measurement or the estimation of some state
variables has been overcome given that all the proposed controllers are based
only on the measured variable. Finally some numerical simulations, illustrating
the theoretical results, are presented for a simple and a more complex reaction
mechanism together with some comparisons with existing methods.
41
m m
Tint1 Tint2
Tj
V=c
V<0
V>0
Tss
B
j A
V<0
H
T ss Tss
j
Tss α 2 ψ
in
Tss Tss Tss T
α M k in
3
42
in r
and
m = T + Tss
Tint q2 : (84)
2 ss k q + q1 m2
in 2
To avoid the intersection of the two ellipses (81) and (80) it is necessary that
m Tm
Tint 1 int2
43
References
1] A. K. Adebekun and F. J. Schork. On the robust global stabilization of
nth order reactions. Chem. Eng. Commun., 101:1{15, 1991.
2] K. Adebekun. The robust global stabilisation of a stirred tank reactor.
AIChE Journal, 38(5):651{659, 1992.
3] J. Alvarez, J. Alvarez, and E. Gonzales. Global nonlinear control of a con-
tinuous stirred reactor. Chemical Engineering Science, 44(5):1147{1160,
1989.
4] J. Alvarez, J. Alvarez, and R. Suarez. Nonlinear bounded control for a
class of continuous agitated tank reactors. Chemical Engineering Science,
46(12):3235{3249, 1991.
5] J. Alvarez-Ramirez. Stability of a class of uncertain continuous stirred
chemical reactors with a nonlinear feedback. Chemical Engineering Sci-
ence, 49(11):1743{1748, 1994.
6] J. Alvarez-Ramirez and R. Femat. Robust PI stabilization of a class of
chemical reactors. Systems and Control Letters, 38(4):219{225, 1999.
7] J. Alvarez-Ramirez, R. Suarez, and R. Femat. Control of continuous-stirred
tank reactors: stabilization with unknown reaction rates. Chemical Engi-
neering Science, 51(17):4183{4188, 1996.
8] R. Antonelli and A. Astol. Global temperature stabilization of chemical
reactors with bounded control. Proc. AIChE Annual Meeting, November
15-20, Miami Beach, Florida, USA, 1998.
9] R. Antonelli and A. Astol. Global temperature stabilization of continuous
stirred tank reactors with bounded output feedback control. Proc. 6th
Mediterranean Control Conference, June 9-11, Alghero, Italy, 1998.
10] R. Antonelli and A. Astol. Adaptive output feedback stabilization of a
class of uncertain continuous stirred tank reactors. Proc. 14th IFAC World
Congress, Beijing, China, 1999.
11] R. Aris and N. Amundson. An analysis of chemical reactor stability and
control. Chemical Engineering Science, 7(8):121{131, 1958.
44
12] H. Chen, A. Kremling, and F. Allgower. Nonlinear predictive control of a
benchmark CSTR. Proceedings of 3rd European Control Conference, pages
3247{3252, Rome, Italy, 1995.
13] Y.-S Chou and W. Wu. Robust controller design for uncertain nonlinear
systems via feedback linearization. Chem. Eng. Sci., 50:1429{1439, 1995.
14] J.M. Coulson, J.F. Richardson, J.R. Backhurst, and J.H. Harker. Coulson
and Richardson's chemical engineering. Volume 1: uid ow, heat transfer
and mass transfer. Butterworth-Heinemann, Oxford, Boston, 1999.
15] P. Daoutidis and C. Kravaris. Dynamic output feedback control of
minimum-phase nonlinear processes. Chem. Eng. Sci., 47:837{843, 1992.
16] D. Dochain. Design of adaptive linearizing controllers for non-isothermal
reactors. Int. J. Control, 59:689{710, 1994.
17] F.J. Doyle III, A.K. Packard, and M. Morari. Robust controller design for
a nonlinear CSTR. Chemical Engineering Science, 44(9):1929{1947, 1989.
18] J.H Espenson. Chemical kinetics and reaction mechanisms. 2nd ed.,
McGraw-Hill, New York London, 1995.
19] A. Fradkov, R. Ortega, and G. Bastin. Semi-adaptive control of convexly
parameterized systems with application to temperature regulation of chem-
ical reactors. European Control Conference, 1997.
20] A.M. Gibon-Fargeot, H. Hammouri, and F. Celle. Nonlinear observers for
chemical reactors. Chemical Engineering Science, 49(14):2287{2300, 1994.
21] C.G. Hill. An introduction to chemical engineering kinetics and reactor
design. John Wiley and Sons, New York, 1977.
22] K. A. Hoo and J. C. Kantor. An exothermic continuous stirred tank reactor
is feedback equivalent to a linear system. Chem. Eng. Commun., 37:1{10,
1985.
23] F. Jadot. Dynamics and robust nonlinear PI control of stirred tank reactors.
Ph.D. Thesis, CESAME, Universite Catholique de Louvain, Belgium, 1996.
24] F. Jadot, G. Bastin, and F. Viel. Robust global stabilisation of stirred
tank reactors by saturated output feedback. European Journal of Control,
5:361{371, 1999.
45
25] H. K. Khalil. Nonlinear systems. 2nd Edition, Prentice Hall, Upper Saddle
River, New Jersey, 1996.
26] H.K. Khalil and F. Esfandiari. Semiglobal stabilization of a class of non-
linear systems using output feedback. IEEE Transactions on Automatic
Control, 38(9):1412{1415, 1993.
27] P. Kittisupakorn. The use of nonlinear model predictive control techniques
for the control of a reactor with exothermic reactions. Ph.D. Thesis, Chem-
ical Engineering Department, Imperial College of Science, Technology and
Medicine, London, U.K., 1996.
28] K. A. Kosanovich, M. J. Piovoso, V. Rokhlenko, and A. Guez. Nonlinear
adaptive control with parameter estimation of a CSTR. J. Proc. Contr.,
5:133{148, 1995.
29] C. Kravaris and C.-B. Chung. Nonlinear state feedback synthesis by global
input/output linearization. AIChE Journal, 33:592{599, 1987.
30] M.J. Kurtz and M.A. Henson. Input-output linearizing control of con-
strained nonlinear processes. Journal of Process Control, 7(1):3{17, 1997.
31] L. C. Limqueco and J. C. Kantor. Nonlinear output feedback control of an
exothermic reactor. Computers Chem. Eng., 14:427{437, 1990.
32] M. Morari and E. Zariou. Robust Process Control. Prentice-Hall, Engle-
wood Clis, NJ, 1989.
33] R. H. Perry and D. Green. Perry's Chemical Engineers' Handbook. Sixth
Edition, McGraw-Hill, Tokio London, 1984.
34] A. Uppal, W.H. Ray, and A.B. Poore. On the dynamic behaviour of con-
tinuous stirred tank reactors. Chem. Eng. Sci., 29:967{985, 1976.
35] F. Viel, F. Jadot, and G. Bastin. Global stabilization of exothermic chem-
ical reactor under input constraints. Automatica, 33(8):1437{1448, 1997.
36] F. Viel, F. Jadot, and G. Bastin. Robust feedback stabilization of chemical
reactors. IEEE Trans. Autom. Control, 42(4):473{481, 1997.
46