Download as pdf or txt
Download as pdf or txt
You are on page 1of 10

Copyright © IFAC Dynamics and Control of Process Systems,

Corfu, Greece, 1998

MODELING, OPTIMIZATION AND CONTROL OF BATCH


CHEMICAL REACTORS IN FINE CHEMICAL PRODUCTION
M .V. LE LANN, M. CABASSUD, G. CASAMAITA
Ecole Nationale Superieure d'Ingenieurs de Genie Chimique
Laboratoire de Genie Chimique -UMR CNRS 5503
18, Chemin de la Loge, 31078 TOULOUSE Cedex 4 - France
Tel : 33 5 6225 23 68 - Fax : 33 5 62 25 23 18,
email: MarieVeronique.LeLann@ensigctJr

Abstract : Operation of a batch reactor covers a wide domain from


optimization of operating conditions to on-line control and monitoring. This
paper focuses on the overall methodology developed in our laboratory for
more than 10 years, in the field of fine chemical industry. This methodology
lies on tendency modeling for stoichiometry and kinetic determination,
optimization of operating conditions, and control of batch reactor by linear or
non-linear Model Predictive Control. Copyright © 1998 IFAC

Keyword : Batch reactor, Modeling, Optimization, Model Predictive


Control, Fine Chemistry

1. INTRODUCTION reactions are usually complex, their kinetics <J'e


poorly known. Classical kinetic studies are not
With the emergence of generic drugs, pharmaceutical possible because they are time-consuming in
industries which were not used to suffer from a severe comparison with the duration of the campaigns and
concurrence are now pushed to compete in order to be the economical objectives. An alternative approach is
the first on the market. So, they are now interested proposed which aims to conserve the numerical
in any technique or methodology which permits them approach of the optimization, based upon a structured
to reduce the time to market. The heart of a drug non-linear model. The model will not be a
manufacturing process is the batch reactor which is fundamental model but a "tendency model" i.e. an
still widely used in fine and pharmaceutical approximate stoichiometric and a pseudo-
industries. It is often characterized as a flexible art homogeneous kinetic model of the transformation
multipurpose equipment. That means that a same (Filippi, et al., 1986). The tendency model allows to
apparatus is used to carry out different reactions art propose a mathematical support to integrate available
operations under various operating conditions knowledge on the transformation thus providing a
involving chaining of sequences. better insight into the process than that offered by the
approach based on a black box model (Rastogi et al.,
In this paper, we want to focus on the overall 1992). The aim of this tool is first of all, to represent
methodology we have developed for modeling, satisfactorily the influence of the most important
optimization and control of batch reactor for more operating parameters in order to elaborate an
than 10 years . We have successfully applied these operating protocol leading to the improvement of a
developments to industrial processes particularly in given criterion. This approach will be detailed in the
the SANOFI company with which a long time first part.
partnership has been established ..
Once the stoichiometric and kinetic models have been
Operation of a batch reactor covers a wide domain determined, it is therefore possible to focus on the
from optimization of operating conditions to on-line optimization of operating conditions. Let us notice
control and monitoring. In each of these domains, that this optimization is rather different from the
computational procedures are generally based on a design procedure commonly used for continuous
process model. The numerical problem of batch processes. In our case, the industrial reactor in which
reactor performance optimization has attracted a lot of the reaction will be carried out already exits. So, it is
attention (Bonvin, 1997). Since fine chemical necessary to match this requirement by either

715
introducing a perfect description of the apparatus NR

itself inside the model used for the optimization or by leads to : Yjl = L V ij Xii (4)
defining physical constraints relative to the use of i=l

this specific reactor. This optimization procedure is This system may be written in a matrix form as
described in the third part. The success of such follows :
methodology in fine chemical industry depends on the (5)
time needed to carry out the experiments. To The stoichiometric vectors of each reaction and the
minimize this time we now develop a planning corresponding reaction extents have to be identified.
strategy to minimize the number of experiments. In Two algorithms have been proposed in the literature:
the fourth part this strategy is discussed for the
problem of kinetic parameters determination. The Step by step method
final part deals with the control of the reactor with a
particular emphasis on a new strategy developed for This method was developed by Filippi et al. (1986)
temperature control and recent experimental results and used for several industrial applications. It
obtained with Non-Linear Model Predictive Control. identifies the stoichiometric coefficients of one
reaction at a time. Firstly, the algorithm assumes
2. STOICHIOMETRY AND KINETICS that one unique reaction can fit the data. The extents
MODELING are computed by the conjugate gradient technique aM
the stoichiometric coefficients by a least square
The minimum necessary and essential data for the method. The search is stopped if the representation is
development of the tendency model are : the total sufficiently good. Otherwise, a new reaction is aihl
batch time, the reactor temperature, the initial and its stoichiometric coefficients are identified,
compositions, the feed rate of the reactant (in the case maintaining those of the previous reaction
of semi-batch operation) and the final concentration unchanged. The number of reactions is increased by
of the different components . In the case where one at each step. The procedure is stopped when the
material losses are observed due to unmeasured model accuracy, specified by the user. is obtained.
products, the definition of pseudo-constituents allows
one to satisfy the molar component balances. However, this technique only leads to nonsensical
stoichiometric network. The first reaction is a global
2.1. Identification of a stoichiometric model representation of the transformation and the others are
added to complete the mass balances. As result, the
The first step is the identification of an approximate obtained stoichiometric matrix is generally a linear
stoichiometric model of the transformation which combination of the real one (Garcia, 1993) and the
correctly fits the batch data. The stoichiometry of a difficulty lies in the recombination of the identified
reaction system involving NC species Aj (j;I,NC) matrix. It is important to notice that at each step, the
and NR reactions Ri (i;I,NR), can be written : operator can propose a stoichiometric vector which is
Ne either in agreement with the computed values (by
'vA
L..J .=0I) )
0) rounding the real values proposed by the algorithm)
j=l or not. In this way, it is possible to incorporate
where vij is the stoichiometric coefficient of Aj in available information on the transformation to assist
the reaction Ri (if vij>O then Aj is a product, if the determination of the stoichiometric network.
vij<O then Aj is a reactant, if Vij;O then Aj is not)
Singular Value Decomposition method
For a batch reactor and a data base of NE experiments
The second algorithm used factorial analysis
(1; I,NE) , the number of moles of the component Aj
techniques to separate stoichiometric vectors aM
in the chemical transformation represented by several extents of reaction from the data matrix Y . Hamer
reactions Ri, is given by : (1989) has applied this technique to fed batch
NR fermentation examples. Moreover, as in the previous
njl; nO jl + n° L. VijXil (2) method, they must be combined to obtain a realistic
i;1
model. Since information about the transformation is
where n jl is the number of mole of Aj in the often available from the chemist's experience, an
experiment I, n O jl is the initial number of mole of Aj approach called Target Factor Analysis (TF A)
in the experiment I, Xii is the extent of the reaction (Bonvin and Rippin, 1990). This technique enables
Ri in the experiment I and n° is a normalizing factor. one to know whether a postulated stoichiometry from
Introducing the molar ratio : a priori information is compatible with the abstract
n).1 _no )1 factors. If the postulated stoichiometry is consistent
Yj1 == Yjl - yO jl == ""':'-n-o""";- (3) with the data, the postulated and the transformed
matrix are therefore identical. Otherwise, the target

716
stoichiometry is not consistent with the step by step and SVD decomposition . The
observations. identification of kinetic parameters is performed using
SQP algorithm from IMSL. Designed for chemists,
In conclusion, with both methods, the main problem it has been implemented on 4 sites of SANOFI
is the combination of the identified vectors so as to company: Sisteron, Aramon, M ontpellier am
obtain physical meaningful vectors. In both cases, Budapest ( Hungary).
the problem may be solved by testing the proposed
stoichiometries : 2.3 Exnmple of complex synthesis mode ling
- in the iterative method, by forcing at each step
one reaction of the postulated stoichiometry To examplify the interest of this methodology, its
(beginning with the principal reaction followed by application to a complex chemical transformation is
the main side-reactions) presented. The alkaline degradation of fructose may be
- in the factor analysis method by testing postulated achhieved in a simple way by introducing fructose in
stoichiometry using the TFA . an alkaline solution. Nevertheless, the alkaline
degradation of fructose is a non-selective
2.2. Identification of a kinetic model transformation resulting, by means of a very complex
reaction scheme, in numerous side-products, such as :
If the reactor operates in batch or semi-batch mode a - the isomers of the fructose : glucose, mannose,
molar balance gives: sorbose, psicose, galactose, tagatose, gulose ...
dn j - the formic , acetic, glycolic, glyceric, lactic,
- = Fe + QV (6) threonic, 2,4 and 3,4 dihydroxybutyric, 2-
dt J' ~
deoxypentonic, saccharinic acids ...
where nj is the number of moles of Aj at instant t, products of degradation such as the
Fej is the feed rate of the compound Aj (j=I, NC) CIld cyclopentadion, furanic compounds, phenolic
V is the reactor volume. compounds, volatile compounds, oligomers,
NR polymers, coloured products ...
and we have: Rj = .L Vi/i where
For many years the degradation of fructose in alkaline
i=]
ri represents the rate of the reactions considered. medium has attracted the attention of chemists. For
simplification, we can say that the transformation can
In this work, the transformation is supposed to be a
be roughly decomposed in two kinds of reactions :
pseudo-homogeneous one and the kinetic law is
- isomerizations leading to the different
written as a classical Arrhenius' law. It is important
isomers of the fructose (reversible reactions),
to emphasize that the form of the kinetic law and its
- degradations leading to the different acids
degree of complexity depend on the user and the
(irreversible reactions).
desired accuracy of the tendency model. So, we have
-E Ne The first step of these two reactions is the formation
r I = kOI eRfIlC aj (7)
of an intermediate called enediol (enediol 2,1 or 2,3
J
j=!
.. . according to which carbon it is attached to). The
enolization may be followed by numerous reactions :
Preexponential factors (kOi), activation energies (Ei),
elimination, retroaldolization, rearrangements
order (aij) with respect to each of the reactants for Through the enediols, the different acids may
each reaction of the stoichiometric model and the originate from either fructose , glucose, mannose or
form of the kinetic laws are unknown. The orders are sorbose ... An experimental planning has been carried
assumed to be part of the data of the problem and are out by Dubois (1992) to improve the yield of lactic
chosen a priori to be equal to the absolute value of acid by the degradation of fructose. The experiments
the stoichiometric coefficients of every reactant. The have been run for a total batch time of 9 hours. The
problem is thus the determination of the amount of 6 sugars (fructose (F), glucose (G),
preexponential factors and the activation energies for man nose (M), sorbose (S), psicose (Ps), galactose
each reaction in order to minimize the difference (Gal)) and 5 adds (lactic add (AL), acetic add (AA) ,
between the experimental concentrations and those formic acid (AF) , glycolic acid (AGlyc) and glyceric
computed with the identified parameters for the acid (AGlyc) was determined by an HPLIC analysis.
different constituents. The identification is carried out The sugars essentially produced from fructose are :
over the entire NE experiments. Classical parameter glucose, mannose, sorbose and psicose. The analysis
identification techniques can be used. of the results issued from the experimental planning
indicates the following favourable conditions for
A software named BATC HM OD performing both lactic acid production : a high initial hydroxide
stoichiometry and kinetic parameters identification
sodium concentration ([NaOH]O = 2m01.l- 1), a low
has been developed in VISUAL Basic and Fortran. It
proposes both methods for stoichiometry research: initial fructose concentration ([F]0 = 25 g.t 1) and a
low temperature (T = 40°C). In optimal conditions,

717
the lactic acid selectivity is equal to 65.8% and its value by SVD of the data matrix composed by the
yield to 64.6%. The experimental data show dimensionless mole fraction of each constituent gives
sometimes a very important loss in raw materials, : (3 .6478 0.7820 0.2154 0 0.0123). The last
due to the non analyzed products. In order to take this two singular values are much smaller than the
parameter into account in the model, this quantity previous ones. This indicates that the data may be
was expressed by the mean of a pseudo-constituent described adequately by three reactions. According to
called P. Its molar concentration was roughly the available knowledge about the transformation we
estimated assuming that the production of lactic, have postulated a stoichiometry with 3 reactions :
acetic and formic acid is essentially achieved by - the first reaction represents the lactic acid
means of a triose and the production of glyceric arl formation from fructose and hydroxide sodium (one
glycolic acids by means of pentose or hexose ; thus mole of fructose gives two moles of lactic acid)
one obtains : - the second reaction represents the unknown or
Cp = COF-[CF + CG + CM + CS o + Cps + CGal + non-measured by-products formation only from
CAGlyce + CAGlyco + (CAL + CAA + CAF)/2) fructose , without hydroxide sodium consumption.
Some assumptions were made in order to decrease the - in the third reaction the secondary acids are
number of constituents present in the mixture. In this produced according to a ratio of 2 fructose moles for
work we have assumed that all the sugars, which leaj three sodium hydroxide moles to form three moles of
to an enediol intermediate, present similar acids.
sensitivities with respect to the operating variables.
All the hexoses were gathered to form a pseudo- NaOH FS AL A P
constituent FS : FS = F + G + M + Ps + Gal. RI -2 -1 2 0 0
The secondary acids were also grouped to obtain: R2 0 -1 0 0 1
A = AGlyco + AGlyce + AA + AF. R3 -3 -2 0 3 0
The sodium hydroxide and the lactic acid (product of Table 1 : Postulated Stoichiometric matrix
interest) were kept unchanged.
The identification of a kinetic model based upon the
This complex system has been reduced to 5 stoichiometry described above, requires the
components only : NaOH, FS, AL, A and P. The determination of six parameters: three preexponential
tendency model had to establish as well as possible factors and three activation energies. The kinetic law
the influence of the operating variables (temperature, is expressed as a classical Arrhenius' law and the
initial fructose and sodium hydroxide concentration) orders are assumed to be equal to the stoichiometric
on the result of the transformation through the coefficients of reactants in absolute value. The
behavior of these 5 constituents or pseudo- computed kinetic parameters are given in Table 2.
constituents. For the identification of an approximate
stoichiometric model of the transformation based on The stoichio-kinetic model has been identified using
the five components described above, the two data corresponding only to initial and final
techniques previously presented were used (step by concentrations. Nevertheless, the structure of the
step and SVD methods) (Garcia, 1993). Since the model allows to simulate time evolution of the
combination of the nonsensical reactions is a difficult concentrations as it has been shown in (Garcia et al.,
problem we have chosen to test an a priori 1997). Especially a good representation of the time-
stoichiometry stemmed from deductive reasoning evolution of the lactic acid concentration was
based on the available knowledge about the obtained.
transformation. The determination of the singular

El =65690 J .mol- 1 E2 =64698 J.morl E3 =78156 J.mol- 1


kOl =2.42 106 12.mor2 .s-1 k02 =2.00 106 s-1 k03 =2.70 108 14.mo l-4. s-1
Table 2 : kinetic parameters identified

3. OPTIMIZATION OF OPERATlNG reeded are dynamic ones. These two characters have
CONDmONS led to more complex problems in comparison to
optimization of continuous processes which generally
Optimization of an operation in the field of fine involves only steady-state models of the single
chemistry is characterized by the fact that generally chemical operation. This is the reason why the
the apparatus devoted to this operation already exits. optimization of batch process operations in this field
Therefore the optimization step has to include not has known development relatively recent compared to
only the model of the chemical operation involved works done on continuous processes (Terwiesch, er
but also a model of the apparatus. Moreover as the al., 1994 ;Bonvin, 1997). In the literature, works
operation are performed in batch mode, the models have been performed on batch processes optimization

718
by using the well-known Maximum Principle and (Zj_l-Zj) + bmin(tj-tj-J) < 0 (1)
technique. Nevertheless, as this approach is unable to - a constraint on the rate of the heat generated by
handle complex models and constraints, the results the reactions in order to avoid runaway.
are of poor interest since they are generally not Chemical constraints can be used as well :
experimentally applicable. Recent developments on - constraint to control the amount of introduced
non-linear optimization allowed to defme a new reactant
methodology in better agreement with the apparatus - constraints of purity restricting the formation of a
limitations . by-product W under a threshold HI:
Optimization of batch reactor consists in the - constraints on the formation of the desired product
determination of temperature or additional reactant P in order to obtain a concentration above a threshold
rate profiles, or the determination of these two
profties simultaneously. The different variables of the The objective function, given by a simulation model,
problem are lower and upper bounded. These bounds is an implicit function of the variables. It could be a
represent some physical limits (for example, simple objective function to be maximized, for
concerning the temperature, the lower bound can example the concentration of the desired product.
represent the minimum temperature of the reaction, Other types of functions could be chosen, depending
the upper bound the boiling point of the solvent) . upon the nature of the problem : it could be the
A classical optimization problem with bounded maximization of the yield versus to the total quantity
and constrained variables takes the form : of reactant initially present in the reactor or versus to
Minf(X) the total quantity of added reactant. Some other
objective functions including economic factors may

j C(x) = 0
I ~ x~ u
C: R m ----7 RP (m ~ p) (8)
be interesting, for example, when the cost of the
reactants is the controlling factor or when it is
economically advantageous to minimize the operation
where C is a vectoriallinear/non-linear continuous time or the operating costs .... The problem is now to
application, I and u are the lower and upper bounds. compute the variables Zj (temperature and feed rate) at
To treat the optimal control problem by non linear each bound of each subinterval, that minimize the
programming, a finite number of discrete parameters criterion subject to the bounds and imposed
representing the control function u(t), t E It(), tf] constraints. The optimal control problem is solved
(temperature or feed rate profties), has to be defined. by nonlinear programming . A sequential quadratic
For this purpose, the interval [t(),tf] is divided into a programming (SQP) procedure (Gill, 1985) has been
finite number nint of subintervals. In each used as it is considered as the most effective method
subinterval, the control function is represented by a for NLP problems. These studies led to the
function f as : development of a software called OPTIBATCH.
u( t) = f(t,Zj) t E [tj-l , tj] j= 1,nint (9)
The profiles are defined by the parameters Zj and the Example of chemical synthesis optimization
switching times tj . In order to avoid too much
This example corresponds to the optimization of the
sophisticated profiles, f is assumed to be a linear
lactic acid synthesis which the stoichiometry and the
function versus time in each subinterval. The
kinetics identification has been reported in paragraph
temperature and feed rate variations in the range [tj-l,
2.3 . The main objective was to simultaneously

giVu7.)b~ Zj-l + (t _ t j-l l( ~


improve the lactic acid yield and to minimize the
tjl :ne Z - Z j-1] (10) unknown by-products and acid by-products formation.
t - t
The objective function to maximize used is the
J J-1
C LA (2)
Note that this formulation allows for a discontinuity following : c=
CA +C p
and change of functional shape in each subinterval.
The optimization procedure needs a model of the In the conditions determined by Dubois (992) (batch
reactor. In the first studies the model used was and isotherm at 40°C) the lactic acid yield and
restricted to the mass balance assuming no change of selectivity are :
state, only modeling the kinetic aspect (Eq. 1 to 5 R~P = 64% and S ~P = 65 .8%
with cv=cl=O). Nevertheless to get feasible operating which leads to a value of the criterion described above
conditions it is necessary to consider the limitations equal to 2.577. One or more variables can be
of the heating and cooling capaCities of the system optimized using the optimization algorithm . The
fitted out the real industrial reactor. These limitations temperature and the feed rate of fructose will be
can be expressed by two ways : computed in order to improve the previous criterion.
- constraints on the rate of change in the The hydroxide sodium is kept at it maximum value
temperature : in the experimental range, (i.e. 2 moU I ) . The total
(Zj-Zj-J) - bmax(tj-tj-I> < 0 batch time is equal to 9 hours. The reaction volume

719
MF(~' e) = I (a ll(~i' 9)]T I:-l(~d (a T](~i ' e)]
is equal to 1 liter. The amount of fed fructose is fixed
to 25g. The following constraints and bounds on the
variables were employed : maximal temperature i=! a9 ae (14)
variations: ±2°C.min- 1, temperature bounds : There are several possibilities to express this
400 C ~ T ~ 100°C optimization problem . One solution generally
and feed rate bounds : 0 ~ F ~ 10-4 kgs- 1 . adopted is the maximization of the M F determinant.
The optimal conditions obtained are a fed batch Therefore, the determination of the optimal operating
operation with feed rate of fructose associated to an conditions ~D for parameters estimation can be
isothermal operation at the maximum temperature expressed as :
100°C. For these conditions one can notice a very ~D = Arg max ret(MF(~ ' 8) (5)
.;
important improvement in the criterion, since it has
This objective criterion is a non-linear function of the
improved from 2.577 to 5.785. The positive effect of
unknown parameters. So, a classical approach
a low initial fructose concentration, predicted by the
consists in alternating the estimation step procedure
analysis of the experimental planning, suggests that
and the procedure of experimental operating
the fructose may be introduced in semi-batch mode to
conditions determination. This approach is called the
favor the lactic acid production. The results of the
sequential experimental design strategy in which it is
optimization confirm this tendency.
necessary to include all the constraints limiting the
feasible experimental domain . For example, the
The reaction was camed out under these optimal
minimum time needed between two consecutive
operating conditions. A syrup with 75 % of fructose
measurements, the maximum time of the experiment
in water was fed during 9 hours (which corresponds to
and also all the technical constraints relative to the
25g of fructose . The theoretical and experimental
reactor in which the experiments are carried out (such
yields and selectivities (expressed in %) are given in
as maximum heating or cooling rate and/or
Table 3. Rather better experimental results that those
maximum reactant feeding, maximum released heat
predicted by the tendency model were observed.
capacity).
RE;.P Rt."'}. st;,P s[;
5. CONTROL OF BATCH REACTOR
Previous conditions 64.6 59.8 65.8 60.8
(batch - 40 0c) If we consider the optimization step described above
Feeding of fructose 76.3 74.1 83.0 74.4 the control of batch reactor should be a temperature
(maximum isotherm) control coupled with a concentration control.
Table 3 : Yields and selectivities in lactic acid Nevertheless excepted a few applications it appears
generally impossible to implement an on-line control
4. EXPERIMENTAL PLANNING STRATEGY of concentration. This is due essentially to the lack of
on-line concentration measurements. Even if new
In the previous example, the optimization strategy methods have revealed promising (Infra Red
was developed on the basis of existing data. Then spectrometry measurements (Amrhein et al ., 1996 ;
arised two novel questions: Wise and GalIagher, 1996» , the problem of batch
-Is-it possible to minimize the number of needed reactor control remains essentially a problem of
experiments? temperature control (Friedric and Perne, 1995). Thus,
- Is-it possible to design experiments affording the control performances are mainly dependent on the
maximum information for model identification ? heating-cooling system associated with the reactor
(Berber, 1995). A lot of studies have been performed
Both can be achieved by developing experimental on control problems and strategies in such a type of
planning strategy . This approach consists in the reactors. Chylla and Randall Hasse. (1993) and Juba
determination of the experimental operating and Hamer. (1986) give a good overview on the
conditions in such of way to maximize the confidence challenges in batch reactor control and suggest several
in the estimated parameters (Walter and Pronzato, control strategies.
1990). In practice this objective corresponds to the
minimization of the confidence region volume of the Let us only recall that the two widely used heating-
parameters which can be expressed as : cooling systems in industry are the multifluid and the
&- O*)T MF (~, 0)(0 - 0* ~ const (13)
monofluid. The former one consists in delivering to
the jacket alternatively utility fluids available at
where * is the parameter vector estimated for the
e given temperature. The latter is based on heating or
cooling of a single thermal fluid delivered to the
operating conditions ~ . and M F the Fisher' s
jacket.
information matrix given by:

720
In the field of fine chemistry the ideal controller temperature of the external fluid . The value of the
should be as flexible as the apparatus. In such a case flow rate doesn't influence this final state but affects
the controller tuning will be attached to the reactor the "speed" at which the final state is obtained. The
itself and should be able to handle different operating manipulative variable is in fact the time constant of
conditions or chemical reactions . the process. This is true for all processes where the
manipulated variable is the flow rate (i.e. heat
5.1 Linear Model Predictive Control exchangers for example . The parameters of this model
was identified on-line by using all the available
The temptation in the control of batch reactors has measurements such as : temperature inside the jacket.
been to use the simplest model i.e. the classical In practice, this to identify the heat transfer
input-output model (also called the black box coefficient. An estimator of the heat generation rate
model). In the first works performed in our was also developed from past values of reactor arx1
laboratory related to the thermal control of batch inlet/outlet jacket temperatures. This value, supposed
reactor, we applied the adaptive Generalized to be constant on the prediction horizon, was then
Predictive Controller with Reference Model used in the model (term ~H') . Such a technique
(GPCMR) on pilot and industrial reactors (1000 I allows also to elaborate a supervision structure which
glass-lined reactor (Le Lann et al., 1995» . The pilot was able to predict when the valve of hot fluid has to
plants' results were very satisfactory as we were able be cut-off to prevent from temperature overshoot at
to operate the reactor for different conditions the end of the preheating phase (Le Lann et aI. ,
(temperature profile, chemical reactions , .. .) with the 1995).
same controller design parameters (Le Lann et al.,
1995).We thought that we had reached our objective Controller based on the thermal flux analysis
to develop a controller as flexible as a batch reactor.
But, the results were not as good as expected for the By considering Eq.16 one can notice that « the
industrial reactor fitted out with the multifluid driving force » for a temperature change in the reactor
system. It revealed that the black box model didn't
model the transfers between the reaction mixture arx1
is the term U rj a rj ~j - Tr ) which represents the
the jacket. A particular problem was the "bain-marie" thermal flux exchanged between the reactor mixture
effect (effect corresponding to a zero utility flow rate and the jacket. A new strategy for temperature reactor
but with a jacket full of fluid). Let us recall that in control was developed by considering this thermal
the case of the multifluid system the manipulative flux as the manipulated variable. At a first level, a
variable is the fluid flow rate. The heat transfer predictive controller determines the thermal flux to be
between the jacket and the reaction mixture exchanged for temperature set-point tracking. This
represents the main limiting dynamics in an thermal flux is then used in a cascade-like structure to
industrial-scale reactor. determine the real manipulative variable : valve
opening, heating power ... This structure is based on
Realistic Model Based Predictive Controller a thermal model of the exchanges between the reactor
and its jacket and depends on the heating-cooling
A novel predictive controller called "realistic model- system used. A considerable improvement in
based controller" was developed (Le Lann et al., comparison to classical strategy is brought
1995). This controller is always based on the GPC concerning the supervision of the heating-cooling
structure but used a nonlinear model which is system . In fact, in addition to the temperature control
obtained from the heat balance on the reaction problem itself there is always a problem of
mixture (Eq.16) and the jacket (Eq.17). supervision whatever the heating-cooling system is .

=Urjarj~j - }~H'+
Indeed it is necessary to choose on-line either the
mrCPr : r Tr right utility fluid in the case of multifluid system or
FcCpc (Tc - Tr ) (16) the thermal device to be used (heat-exchanger, heating
device, ... ) in the case of monofluid system.
j Classically this supervision was performed using
m.Cp . df =F. p.Cp.fr. -T.)- temperature alarms or a priori defined range of
J J ci J J J ~m J
operation for each device. With the thermal flux
Urj arj ~j - Tr ) (17)
analysis strategy it is therefore possible to compute at
each sampling period the minimum and maximum
The manipulative variable was the flow rate delivered thermal exchange capacities of each fluid or each
to the jacket. In such a case, as shown by the model device depending on the system used. According to
the manipulated variable is not a classical one. The these thermal flux limits, it is therefore possible to
final temperature of the external fluid is independent determine which fluid or device is able to
of its flow rate, and thus, without a chemical deliver/remove the thermal flux computed by the
reaction, the reaction mixture will tend to the master controller. The choice is thus done properly as

721
it is based on the thermal capacity and not on a reactors) by monofluid one. Therefore, the solution
predefined temperature. This strategy can be applied a we adopted was to use a monofluid configuration
priori to any heating-cooling system. called intermediate fluid which is obtained by mixing
steam and cold water while this is possible (in
In parallel we developed a new heating-cooling practice from 25°C to 70°C) and to switch to steam
system called hybrid system which mixes both only when more heating is needed or to glycol-water
mono fluid and multifluid systems . The objective was in the cas of high cooling (Louleh et al.; 1997).
to define a thermal system which takes the advantages This system presents the advantages of a multifluid
of both systems. The multifluid is many used in thermal system (possibility to use utility fluids
industry as it demands a very low cost to be settled directly available on a plant) and those of monofluid
and offers a large heating (cooling) capabilities of thermal system (characterized by high dynamic
steam (glycol) when rapid heating (cooling) is needed. performance and continuity of thermal control). The
Nevertheless, it makes temperature control difficult switching condition is given by the on-line
due to changeover of fluids , to a non classical computation of the thermal exchange capacities of
manipulative variable as it has been explained each configuration (intermediate fluid, steam, glycol
previously. The monofluid system is therefore more water) determined as previously described. More
convenient in the controllability point of view since details on this procedure can be found in (Louleh et
it is better to use a single fluid circulating at a al. , 1997). The strategy based on the thermal flux
sufficiently rapid flow rate to get a good heat transfer analysis has been applied to a 16 liter-pilot plant
coefficient. Nevertheless, it remains expensive since fitted out with this system. On figure I a scheme of
as it has to be used in a large range of temperature, this pilot plant is given. An example of results of a
the thermal loop is often fitted-out with expensive temperature control experiment performed on this
devices as heat-exchangers, .... In an other hand it is pilot plant using the thermal flux analysis strategy
unrealistic to think that industry will replace every are plotted on figure 2.
multifluid system (which fits 90% of the batch

Air
--1:><:],-----
I

GlycolfWater
Tu gw

Cold Tuc
w water
Figure 1 : Scheme of the 16 liters pilot plant fitted out with the hybrid heating-cooling system

Temperature (0C) Valve opening B Thermal flux (KcaI.5-1)

70
FTr"I j O.5 5 5
4
60 L=2J·i 0.4
l~pl 103
3 r---------------------------~ 3
---(lc('llt --. -· Qrn;lx if - . ~ . Q['j'I:~" :
50 -Qmaxvap ·· · · -Qmmi f 2
2 -Qminvap ...,' - ~):I:.L\!;'h
40
.i
1
°. ,
.-
30

20 o..J : I
° °
0 , •.••••••••••••• 0• ••••••••••• • ••••
-2 -2
:WOO 4000 6000 8000 2000 4000 6000 8000
Time (5) Time (5)
Fig . 2 : Temperature and manipulated variable Fig . 3 : Limits and control value of the thermal flux

722
This figure shows the evolution of the temperature in including the filling-up ratio of the tank and the
the reactor. the set point and the valve opening degree nature of the fluid (air or utility fluid) . This technique
(in the case of no recycle of the intermediate fluid). A leads to a ODE system which is solved over the
good tracking performance of the cascaded model prediction horizon at every optimization iteration
based controller can be noticed. Figure 3 gives the step. A procedure for model error compensation and
evolution of the thermal flux computed by the master an estimator of the heat generation rate have been
predictive controller (GPCMR) and the upper m used for eliminating process-model mismatches
lower flux limit capacities determined on-line for each (Ettedgui et al. , 1997) .
configuration. One can see that all the different
configurations have to be used. In particular. at the The main advantage of such technique lies on
end of the heating phase. steam is required instead of considering the phases of jacket filling-up and jacket
intermediate fluid. Similarly. during the cooling purge as any normal operation phase. Before using
phase (around 8000 s). the higher cooling rate this model it was not possible to control the reactor
implies a changeover from intermediate fluid to during theses particular phases and consequently it
glycoUwater. This experiment shows the capacity of was very difficult for example to control temperature
such a methodology to handle changeovers. at a constant value when the reaction was weakly
exothermic and the reactor was well insulated. This
5.2 Non-Linear Model Predictive Control technique, firstly tested in simulation (Ettedgui et al.,
1997) has been implemented on a pilot plant (the 16
As a logical continuation to our studies on Linear liter pilot plant described previously but modified to
Predictive Control and to our works on off-line get a pure multifluid system). Figure 4 shows an
operating conditions optimization. we developed a example of experimental results obtained with such a
generic methodology of Non-Linear Model Predictive strategy. One can see that the temperature tracking
Control. This NLMPC is based on the minimization has been successfully performed especially during the
of a quadratic criterion (Eaton and Rawlings. 1990 ; constant level temperature step. One can notice that
Biegler. 1997): the predictive controller is able to perfectly anticipate
p 2 the changes in the temperature profile and
Min J = I(Y c(k + e + i) - ypred(k + e + i») (18)
u(I) .... u(N) i=l consequently it can insure a smooth control without
temperature overshoot. This success is mainly due to
subject to constraints: the accuracy of the model used for the prediction. An
- Upper and lower bounds on manipulated variable automatic supervision technique is being developed.
umin(i) ~ u(i) ~ umax(i) i = 1..... N (19) It is based on a similar procedure to the one described
- Upper and lower velocity bounds on manipulated previously : on a prevision horizon, the maximum
variable and minimum temperature reachable with all the
u(i-I) - ~umax(i) ~ u(i)~ u(i-I) + ~umax(j) i = 1, .. .,N fluids are computed .It is therefore possible to
(20) determine if the present fluid is accurate or if a
where Yc(.) represents the values of the reference changeover is needed and what fluid can be used. In
trajectory (the desired value for the output), Ypred (.) is such a case, this information is used in the predictive
the value of the output predicted by the model, u(.) is controller for the computation of the control law.
the future value of the manipulative variable (they are
Temperature ( cC) Manipulative
the decision variables in this optimization problem). 65r-~-.~~~~~.-~-.~~~v~a~r~ia~ble

The values predicted over the time horizon P are 55


obtained by the integration of an ODE or DAE 0.8
system. This strategy has been applied to the control 45 0,6
of temperature in a batch reactor fitted out with the
multifluid system. In fact, which such a system 35 0,4
there were cases where the control of the reactor 0,2
temperature can only be achieved by partly filling the 25
jacket. Therefore in such cases. the filling-up ratio of o
the jacket becomes a manipulated variable . The two 600 1200 1800 2400 3000 3600
equations (Eq. 16 and Eq . 17) are no more sufficient
to model the dynamic behavior of the reactor when Figure 4 : Experimental results of NLMPC applied
the jacket is not totally full . To model the dynamics to the 16 liter pilot plant (with multifluid system)
in this case. we have adopted the methodology
formerly developed for batch reactor simulation 6. CONCLUSION AND RJTURE WORKS
(Cabassud et al., 1994). The jacket is lumped in
This paper has surveyed more than 10 years of
several tanks . Heat transfer between each jacket-tank
and the reactor is modeled by an energy balance development we had in modeling, optimization and

723
control of batch reactors in the field of fine chemicals optimization techniques, Comput. Chem.
industry. These works have been performed in a close Engng., 14, 469-479
collaboration with industrial partners which permits Ettedgui B., M. Cabassud, M.Y. LeLann, NL
to the use of the developed methodology in industry. Ricker (1997)Regulation of a Fed-Batch Chemical
Our present studies deal with the development of the Reactor Incorporating Parameter Estimator,
experimental design strategy in order to reduce the Proceedings o/the lFAC ADCHEM '97, Banff
experimental effort. This is a key point to make this (Canada), 365-370
methodology widely applied in fine chemical Filippi c., JL Graffe , J. Bordet, J. Villermaux, J.L.
industry. Barney, P. Bonte and C. Gerogakis (1986).
Tendency modeling of semi batch reactors for
In an other hand. in our previous studies on batch optimization and control, Chem. Engng. Sci .,
reactor control, it was assumed that no chemical 41 , 913-920
reaction model was available. The chemical reaction Friedric M. and R. Perne, 1995, Design and control
was considered as a disturbance. The future works of batch reactors : an industrial viewpoint,
will focus to include tendency models in the model Comp. Chem. Engng 19, S357-S368 .
based controller. Then, a state estimator based on Garcia V. (1993), , PhD Thesis, I.N.P. Toulouse
temperature measurements coupled to an estimator of Garcia V., M.Cabassud, M.V. Le Lann., L. Rigal,
the reaction heats will be developed. G. Casamatta (1997), A global strategy for the
optimization of batch reactors : Application to the
We aim to adapt this methodology to other industrial lactic acid synthesis by the alkaline degradation of
fields : biochemical, pharmaceutical. fructose, lMACS (M2SABI) Mathematical
modelling and simlation in agricultural and bio-
REFERENCES industries, Budapest (Hongrie), , in Proc. 63-68
Gill P.E.,W. Murray, M.A. Saunders and M.H.
Amrhein M., B. Srinivasan, D. Bonvin and M. Wright (1985) Software and its relationship to
Schumacher (1996), Inferring concentrations on- methods in numerical optimization, PT.
line from near-infrared spectra : Nonlinear Boggs, R.H . Byrd and R.B . Schnabel, Eds.
calibration via mid-infrared measurements, SIAM, Philadelphia.
Comput. Chem. Engng, 20 , S975-980 Hamer J.W .(989), Stoichiometric interpretation of
Berber R.( 1995), Control of batch reactors : a multireaction data : application to fed batch
review, Methods 0/ Model Based Process Control, fermentation data, Chem. Engng Sci., 44 (10) ,
R. Berber (ed.), Kluwer Academic Publishers, 2363-2374
Oordrecht, 459-493 . Juba M.R. and J.W. Hamer, 1986, Progress and
Biegler L. (1997), Advances in Nonlinear challenges in batch process control. Proc. CPC
Programming Concepts for Process Control, lll. CACHE, Elsevier, Amsterdam.
Proceedings o/the lFAC ADCHEM '97. Banff Le Lann M.V ., M . Cabassud and G. Casamatta,
(Canada), 587-598 Adaptive Model Predictive Control (1995), in R.
Bonvin D. and Rippin D .W.T .. 1990, Target factor Berber (ed.) Methods 0/ Model Based Process
analysis for the identification of stoichiometric Control, Kluwer Academic Publishers,
models. Chem. Engng Sci., 45 (12), 3417-3426 Dordrecht, pp. 426-458, 1995
Bonvin D. (1997) Optimal operation of Louleh., Z . Cabassud, M.V. LeLann and G .
Discontinuous Reactors-A personal view Casamatta (1997) and G. Casamatta Experimental
Proceedings o/the lFAC ADCHEM'97. Banff Study of a New Thermal Control Strategy for
(Canada),155-169 Batch Reactors, Proceedings o/the lFAC
Cabassud M. , M.V. Le Lann, B. Euedgui and G. ADCHEM'97, Banff(Canada), 119-124
Casamatta (1994) A general simulation model of Rastogi A., J. Fotopoulos, C. Geaorgakis and H.G.
batch chemical , Chem. Engng Technol.,17 (4) , Stenger (1992), The identification of kinetic
255-268 expressions and the evolutionary optimization of
Chylla R. W. and D. Randall Hasse( 1993), specialty chemical batch reactors using tendency
Temperature control of semi-batch models, Chem. Engng., Sci ., 47 , 2487-2492
polymerization reactors, Comput. Chem. Engng Waiter E. and L. Pronzato,(1990) Qualitative and
17 W 3, 257-264. Quantitative Experiment Design for
Dubois M.F(l992), Valorisation chimique du 0- Phenomenological Models-A Survey
Fructose: synthese des acides levulinique, Automatica, 26, (2), pp. 195-213.
arabonique et lactique, PhD Thesis, I.N.P. Wise B.M. and N.B. Gallagher (1996) The process
Toulouse chemometrics approach to process monitoring aM
Eaton J.W . and J.B . Rawlings (1990), Feedback fault detection, J. Proc. Control, 6, 329-348
control of chemical processes using on-line

724

You might also like