Computers

&Chemical
Engineering
Computers and Chemical Engineering 24 (2000) 1569-1575
www.elsevier.com/locate/compchemeng

Optimal operation of a semi-batch reactive distillation column

Gregor Fernholz *, Sebastian Engell, Lars-Ulrich Kreul, Andrzej Gorak
Department of Chemical Engineering, University of Dortmund, Dortmund, Germany

Abstract

By using reactive distillation processes, limitations due to reaction equilibria and due to azeotropes can be overcome, resulting
in an intensified overall process. We here present work on the optimization of a semi-batch process, the heterogeneously catalyzed
esterification of methanol and acetic acid. Two different cost functions, minimal batch time and maximal productivity of the
process are compared. Constraints are added to ensure the purity of the product and sufficient conversion of the raw materials.
The analysis of the influence of the prescribed conversion rates on the productivity helps to detect the crucial parameters of the
optimization problem, giving deeper insight into the process. 0 2000 Elsevier Science Ltd. All rights reserved.

Keywords: Optimization; Reactive distillation; Batch distillation; Optimal operation

1. Introduction al. (1979) and Li et al. (1998) investigated the semi-

batch mode and used the reflux ratio as well as the feed
The integration of reaction and separation into one rate of one reactant as optimization variables. Sprrensen
apparatus reduces the investment costs and facilitates et al. (1996) considered the batch case and optimized
to use the heat of reaction. By reactive distillation the reflux ratio and the heat duty of the reactor.
processes, limitations due to reaction equilibria and This study deals with the heterogeneously catalyzed
azeotropes can be overcome by removing the products esterification of methanol and acetic acid. The reaction
from the reaction zone and by consuming azeotrope takes place inside the distillation column. In contrast to
forming components in the reaction (DeGarmo, previous studies (Egly et al., 1979; Sorensen et al., 1996;
Parulekar & Pinjala, 1992). Discontinuous processes Li et al., 1998) where the two units reactor and distilla-
offer a higher degree of freedom to adapt e.g. to tion column were integrated into one apparatus, but the
changing customer specifications. process steps reaction and separation were kept segre-
The optimal operation policy for batch processes can gated, a fully integrated process is considered here, and
be calculated off-line if a reliable process model is the reflux ratio, the heat duty and the feed rate are
optimized simultaneously.
available. For batch distillation processes several ap-
For the optimization of batch distillation processes
proaches were proposed for the calculation of the opti-
three different formulations of the problem are dis-
mal policy (e.g. Diwekar, 1992; Srar&sen, 1994;
cussed in the literature:
Logsdon, Diwekar & Biegler, 1990). The investigation
- Maximum distillate problem, maximize the
of batch distillation processes was extended to the
amount of distillate with a specified final concentra-
reactive case by Egly, Ruby and Seid (1979). In their
tion for a specified batch time (e.g. Converse &
study a chemical reactor with a distillation column on
Gross, 1963; Diwekar, Malik & Madhavan, 1987).
top was considered and the reaction took place in the _ Minimum time problem, minimize the batch time
reboiler only. Similar plant configurations were consid- required to produce a prescribed amount of distillate
ered by Sorensen, Macchietto, Stuart and Skogestad of a specified concentration (e.g. Coward, 1967; Egly
(1996), Li, Garcia, Wozny and Reuter (1998). Egly et et al., 1979).
- Maximum profit problem, maximize a profit func-
tion (e.g. Kerkhof & Visser, 1978; Logsdon et al.,
* Corresponding author. 1990).

0098-1354/00/$ - see front matter 0 2000 Elsevier Science Ltd. All rights reserved.
PII: SOO98-1354(00)00553-6
1570 G. Fernholz et al. /Computers and Chemical Engineering 24 (2000) 1569-1575

In order to determine the influence of the cost func- - Start-up, the reboiler is filled with a specified
tion on the solution of the optimization problem, two amount of methanol and heated until the complete
different cost functions are investigated here. The first column is filled with boiling methanol.
formulation uses the batch time as a performance index - Production, the manipulated variables in each time
because the batch time is often critical for multiple interval are set to their optimal values.
purpose plants. In the second setting we use the pro- - Shutdown, the heat supply and the feed flow are
ductivity which we define as the amount of product stopped and the system is cooled down.
divided by the total batch time. In both cases, con-
straints on the product purity and on the conversion of The start-up and shutdown procedures are not in-
the raw materials are included. cluded in the optimization because validated models for
these phases are not yet available.

2. Description of the process 3. Process model

Methyl acetate can be produced either by homoge- Mass and energy balances for all elements of the
neous catalysis via sulfonic acid or by heterogeneous plant give rise to a differential-algebraic equation sys-
catalysis using a solid catalyst. A detailed description of tem of approximately 2400 equations. The main fea-
the homogeneously catalyzed reactive distillation for tures of the model are:
the production of methyl acetate can be found in
- The structured packings are treated as a number of
Agreda, Partin and Heise (1990). Using a solid catalyst theoretical plates using the HETP-value (height
avoids material problems caused by the sulfonic acid as equivalent to a theoretical plate).
well as the removal of the catalyst at the end of the - The vapor and the liquid phase are assumed to be
batch (Krafczyk & Gmehling, 1994). By placing the in thermodynamic equilibrium.
catalyst inside the column limitations of raw material - The vapor holdup is negligible and the vapor
conversion and product purity caused by azeotropes behavior is ideal.
and the reaction equilibrium can be overcome because - The molar holdup in the condenser is constant.
the product methyl acetate is removed from the liquid _ The dynamics of the tray hydraulics and of the
phase and methanol is consumed by the reaction. liquid enthalpy are included.
The scheme of the process, which we consider here, is - The pressure drop of the packing is calculated by
shown in Fig. 1. The lOO-mm column consists of three the equation of Mackowiak (1991).
sections. Two identical catalytic structured packings of _ The holdup of the packing is determined by and
l-m length are located in the lower part of the column, experimentally verified correlation.
while the upper part contains a non-catalytic packing. - The phase equilibrium is calculated using the
Acetic acid is fed above the catalytic packing to ensure Wilson equations. The dimerisation of acetic acid in
that both raw materials are present in the catalytic zone the vapor phase is included.
in sufficient concentrations. The semi-batch mode is _ The reaction kinetics is formulated by a quasi-ho-
preferable for this reactive system because methanol is mogeneous correlation.
much more volatile than acetic acid and this would lead
to low concentrations of acetic acid in most parts of the The kinetics of the heterogeneously catalyzed reac-
catalytic section in a batch operation. tion (Eq. (1)) of acetic acid and methanol to methyl
The plant is operated in the following manner: acetate and water
CH,COOH + CH,OH ++CH,COOCH, + HZ0 (1)
is described by
*PrOdUOt
yi = ui C&l XHAcXMeoH - &~eAc+o
(2)
1 + khleAcXMvieAc ’
All constants k, in Eq. (2) depend on the temperature
as

ki = ko,i exp ; . (3)

0
The rate expression Eq. (2) is identical to first order
kinetics as suggested by Agreda et al. (1990) if the
Fig. 1. Scheme of the semi-batch column. denominator is ignored. Eq. (2) implies that high con-
 G. Fernholz et al. /Computers and Chemical Engineering 24 (2000) 1569-1575 1571

Table 1 Qi,minI Qi< Qi,max, i= 1, a.-> N (6)

Boundsfor the time optimalproblem
Ri,min~Ri<Ri,max, i=l, . . .. N (7)
pin pin c-$ye
MeOH “AC
Fi,minI F< Fi,;;,,,,, i= 1, . . . . N (8)
80% 88% 0.8%
x$gggytJ 2 xg& (9)
Ch.fA&f)2 CFEc (10)
version rates can be achieved by removing methyl ac-
etate from the liquid phase. A more detailed description CMeOH 2 G&H (11)

of the reaction kinetics and the determination of the

parameters of Eqs. (2) and (3) is given in Kreul, G6rak, 2 ti=tf
i=l
(12)
Dittrich and Barton (1998) where the simulation model
is compared with experimental data from a 100 mm where, N is the number of intervals. Additionally the
pilot plant, showing good agreement. initial conditions of the DAE-system and the transition
conditions from one interval to next must be met. For
the initial values of the DAE-system (Eq. (5)) we as-
4. Optimization problem sume that the column is filled with boiling methanol.
For the time optimal problem, constraints on the
Two principal approaches are currently used to solve conversions CHAc and CMeoH were added to obtain a
dynamic optimization by transformation into nonlinear practically relevant solution. The values of the bounds
programs, control vector parametrization (Vassiliadis, Eqs. (9)-(11) are given in Table 1. The desired product
Sargent & Pantelides, 1994a,b) and full parametrization purity XMeAc= 0.8 is larger than the concentration of
of the control variables and of the state variables the binary azeotrope of methanol and methyl acetate at
(Cuthrell & Biegler, 1987). x azeotrope= 0.66.
The first approach gives a small number of optimiza- For the maximal productivity problem the cost func-
tion variables at the expense of simulating the full batch tion f is defined by
run for each step of the optimizer while the second
nMeAc
leads to a uniform formulation with algebraic equations f= (13)
tf + bwtup + tshutdown
only but a large number of variables. As a dynamic
simulation model implemented in gPROMS was avail- where, tstartup and tshutdownare the times for filling,
able, we used the first approach as implemented in the heating up, cooling down and cleaning the plant and
gOPT tool of gPROMS (gPROMS User’s Guide, nMeAc is the total amount of methyl acetate in the
1997). The time horizon is divided into intervals in disillate tank at the end of the batch. For the bound on
which the free variables are parametrized functions of the product concentration the same value as in Table 1
time, in our case piecewise constant. The lengths of the was assumed, but the constraints on the conversion of
intervals are optimization variables whereas their num- the raw materials were varied. The initial amount of
ber is fixed. The optimization problem is solved using methanol nMeoH in the reboiler was added to the list of
an SQP-based algorithm (DAEOPT, Vassiliadis, 1992). optimization variables with an upper bound of 2.37
The decision variables of the optimization problem kmol in this case.
are the heat duty Q supplied to the reboiler, the feed
flow rate F of acetic acid, the reflux ratio R and the
final time ti. They are constrained by upper and lower 5. Results
bounds to ensure that the column is used within its
limits of operation. A lower bound for the mole frac- In the solution of the optimization problem (Eqs.
tion of methyl acetate in the fraction of methyl acetate (4)-(12)) the number of control intervals has to be fixed
in the distillate tank at the end of the batch is pre- a priori. Increasing the number of control intervals
scribed. Additionally lower bounds for the conversions allows a better approximation of the continuous opti-
C,,, and CMeoH of the raw materials are added. Thus, mal control profiles. On the other hand, the dimension
the optimization problem is given by of the optimization problem increases with the number
of intervals leading to longer computational times. In
min f (4) order to estimate the suitable number of control inter-
tcQi&.Fi
vals, the time optimal optimization problem was solved
subject to the plant dynamics
for different number of intervals. The results are shown
W(t), x(t), v(t), u(t)> = 0 in Fig. 2. The use of more than five intervals increases
(5)
the computational time but does not yield significantly
and the constraints shorter batch times. In order to compare the results for
1572 G. Fernholz et al. /Computers and Chemical Engineering 24 (2000) 1569-1575

Table 3
Optimal interval lengths (hmin)

1 2 3 4 5 6

0O:ll 00:16 00:24 00:28 00:18 00:33

07:57 07:54 07:57 07157 07:56 07:58
05:38 05:36 05:35 05:37 05:38 05:40
02:48 02:46 02:43 02:45 02:47 02:48
00:45 00:46 00:37 00:32 00:42 00:27

Fig. 2. Influence of the number of intervals.

Table 4
different cost functions, the number of intervals for the Optimal feed rates (l/h)
maximal productivity problem was fixed at five. The
1 2 3 4 5 6
results which were obtained for the different optimiza-
tion problems are summarized in Table 2, the corre- 12.44 12.44 9.37 7.34 9.84 6.82
sponding values of the free variables are shown in 7.32 12.44 7.34 4.95 5.97 51.2
Tables 3-6. 7.21 10.20 8.10 5.37 6.76 5.78
For all problems the resulting batch time is almost 7.05 5.02 9.14 6.67 6.84 5.68
3.73 6.88 3.73 3.73 3.73 3.73
identical while the initial amounts of methanol, the final
amounts of methyl acetate, and the productivity differ
significantly. The influence of the bound on the conver- Table 5
sion of acetic acid on the productivity can be seen by Optimal reflux ratios (-)
comparing the results obtained for problems 2-4. Not
1 2 3 4 5 6
imposing a bound yields 74% conversion and a produc-
tivity of 30.6 mmol/s. An increase of the demanded acid 0.300 1.ooo 1.000 1.000 1.000 1.000
conversion results in a lower yield. Thus the productiv- 0.596 0.651 0.605 0.586 0.595 0.587
ity must be traded against the effort required to sepa- 0.608 0.632 0.599 0.586 0.599 0.588
rate the mixture in the reboiler after the batch run. 0.637 0.679 0.587 0.548 0.618 0.618
0.853 0.748 0.764 0.739 0.705 0.519
The effect of the bound on the conversion of
methanol is less pronounced. The comparison of the
solutions to problems 1, 3, 4, and 6 shows that the Table 6
solution algorithm gets stuck in local optima. Optimal heat duties (kW)
As more product is obtained for the same batch time,
1 2 3 4 5 6
the optimization of the productivity of the process is
preferable over the solution of the time optimal prob- 1.00 1.00 1.09 1.12 1.oo 1.15
lem. This is not due to the problem of local minima in 3.10 4.21 2.88 2.00 2.38 2.04
the time-optimal problem but can be attributed directly 3.29 4.56 3.55 2.38 3.03 2.57
to the formulation of the cost function. In the time 3.14 4.49 4.74 3.35 3.63 3.01
8.00 8.00 1.93 7.81 7.95 7.64
optimal problem high amounts of product are not
rewarded, only long batch times are penalized. Minimal
raw material conversion rates must be ensured by con-
straints. In contrast, the productivity optimal formula-
tion makes a trade-off between the amount of product

Table 2
Optimization results

Problem 1 2 3 4 5 6

C;i& (%) 88 _ 88 95 92 95
G,, WI 88 74 88 95 92 95
c$:o, (%) 80 _ _ _ 92 95
G&o” (“4 80 96.5 86.7 87.9 92 95
9 (h) 17.3 17.3 17.3 17.3 17.4 17.4
f bmol/s) 25.2 30.6 27.5 20.6 23.4 20.8
nMeOH(kmol) 2.37 2.37 2.37 1.75 1.90 1.65
VMe.& (1) 173 205 187 140 160 141
 G. Fernholz et al. /Computers and Chemical Engineering 24 (2000) 1569-1575 1573

and the batch time. This is reflected by a comparison of

the compositions of the final product in Table 7. If the
productivity is maximized, the final product contains
more water and less methanol.
An analysis of the trajectories of the optimization
variables for all problems indicates that the perfor-

Table 7
Product composition for Eqs. (1) and (3)

Mole fraction Eq. (1) Eq. (3)

Fig. 6. Temperature profiles.
Acetic acid 0.000 0.000
Methanol 0.110 0.086 mance of the process is limited by the reaction kinetics
Methyl acetate 0.800 0.800 rather than by the separation of the mixture: the feed
Water 0.090 0.114
rates and the heat duties decrease with increasing con-
version rates of acetic acid. Low feed rates and heat
duties increase the residence time of the reactants in the
column yielding higher conversion rates. The slow reac-
tion kinetics demand high residence times to ensure
high conversion rates.

6. Optimal trajectories

In order to gain insight into the dynamic behavior of

the process for the different solutions, simulation re-
sults are presented in this section. First the simulation
Fig. 3. Methanol concentrations (liquid phase).
results for the time optimal problem are shown. Ini-
tially the column is filled with methanol. During the
acmlcadd batch run methanol is evaporated in the reboiler and
0.3 ; consumed by the reaction in the lower part of the
o,7: .,.. ., ;.__ j... ;..i .;. .: .
: : :; :... column yielding a low concentration of methanol on
I..
0.1) ,. ,,,..... _i. .,;... .:+
the feed stage and in the product steam (cf. Fig. 3). On
:.
the other hand, the concentrations of acetic acid on the
.; feed stage and on the bottom stage are nearly constant
:. over the time (cf. Fig. 4). Most of the acid is used up
. while flowing down the catalytic packing. The uncon-
0.1’ j&&
.$t!m... ‘. sumed acid accumulates in the reboiler. Acetic acid is
-
‘c .-
completely removed from the mixture in the upper part
Oo 3 5 12 15 15
t&h1 of the column (Fig. 5).
The changes of the concentration profiles during the
Fig. 4. Acetic acid concentrations (liquid phase). batch are reflected in the temperature trajectories in
Fig. 6. High temperatures below the feed stage indicate
high concentrations of acetic acid and water. Since
water and acetic acid are removed in the upper part of
the column the temperature strongly drops from the
feed stage (stage 8) to the top (stage 12). The rise of the
reboiler temperature corresponds to the increasing
amount of water formed by the reaction.
The production rate of the ester in the catalytic
packing is another important aspect besides the concen-
tration and the temperature profiles. The production
rate in the column changes only slightly during most of
the batch time. A typical profile in the catalytic zone of
Fig. 5. Methyl acetate concentrations (liquid phase). the packing is shown in Fig. 7.
1574 G. Fernholz et al. /Computers and Chemical Engineering 24 (2000) 1569-1575

While the production rate peaks in the middle part likely that a higher catalytic packing would yield
of the packing, the formation of the ester is still sig- higher conversion rates of methanol and acetic acid.
nificant at the top and at the bottom. Thus, it seems The simultaneous determination of the optimal design
and the optimal dynamic operation of such columns
are a challenging task for research in this area.
In the sequel a comparison to the evolutions of the
concentration of the components in the reflux stream
for optimization Eqs. (l), (3), (4) and (6) is shown in
Figs. 8- 10.
It can be seen that the time-optimal and the pro-
ductivity-optimal concentration trajectories differ con-
siderably. In the second interval, the product
concentration is much higher for the optimal yield
and the methanol concentration is lower, whereas to-
Fig. 7. Formation rate of methyl acetate. wards the end of the batch the water concentration ia
much higher. In the second interval, the inputs to the
process are rather similar, indicating a high sensitivity
to the operation parameters. This necessitates an effi-
cient, multivariable control scheme to steer the pro-
cess along the optimal trajectory in the presence of
model uncertainties and disturbances. This is dis-
cussed by Fernholz and Engell (2000) and Fernholz,
Wang, Engell, Fougner and Bredehiift (1999).

7. Summary

Fig. 8. Methyl acetate concentration trajectories. Based on an equilibrium stage model of a reactive
semi-batch distillation column a dynamic optimization
memmol
problem was formulated and solved. Time optimal
and maximal productivity problems were set up, in-
cluding constraints on the product purity as well as
on the conversions of the raw materials and on the
manipulated variables.
The optimization problem was solved by
parametrization of the manipulated variables and
transformation of the dynamic problem into a NLP
using the gOPT tool of gPROMS (gPROMS User’s
Guide, 1997). In addition to the reflux ratio, the heat
duty, the feed rate, and the initial amount of
Fig. 9. Methanol concentration trajectories. methanol were optimized.
A comparison of the solutions of the time-optimal
and of the productivity-optimal problem formulation
0.4.
showed that the latter leads to higher conversion
-
--
l.praml
apmblm
rates for identical batch times, making the productiv-
,, -
0.3. ._...&- ity optimal formulation more adequate. The variation
025
-:r_1
of the constraints on the raw material conversion
f
rates showed that bounds posed on the conversion of
dI 0.2
0.15 acetic acid affect the productivity more than bounds
0.1. on the conversion of methanol. There is a trade-off
between high conversion of acetic acid and the pro-
15 ductivity of the process. The results indicate that the
process is limited by the reaction kinetics rather than
Fig. 10. Water concentration trajectories. by the separation.
 G. Fernholz et al. /Computers and Chemical Engineering 24 (2000) 1569-1575 1.515

Acknowledgements tion column. Proceedings of IEEE Znterantional Conference on

Control Applications (pp. 397-402).
gPROMS User’s Guide (1997). Process System Enterprise Ltd.,
The support of the research reported here by the London, UK.
Volkswagen Stiftung is gratefully acknowledged. Kerkhof, L. H. J., & Visser, H. J. M. (1978). On the profit of
optimum control in batch distillation. Chemical Engineering Sci-
ence, 33, 961-970.
References Krafczyk, J., & Gmehling, J. (1994). Einsatz von Katalysatorpack-
ungen fur die Herstellung von Methylacetat durch reaktive Rek-
tifikation. Chemical Engineering Technology, 66(10), 1372-1375.
Agreda, V. H., Partin, L. R., & Heise, W. H. (1990). High purity
methyl acetate via reactive distillation. Chemical Engineering Kreul, L. U., Gorak, A., Dittrich, C., & Barton, P. I. (1998).
Progress, 86(2), 40-46.
Dynamic catalytic distillation: advanced simulation and experi-
Converse, A. O., & Gross, G. D. (1963). Optimal distillate-rate mental validation. Computers & Chemical Engineering, 22,
policy in batch distillation. Industrial Engineering & Chemical S371-S378.
Fundamentals, 4, 415-477. Li, P., Garcia, A., Wozny, G., & Reuter, E. (1998). Optimization
Coward, I. (1967). The time-optimal problem in binary batch distil- of a semibatch distillation process with model validation on the
lation. Chemical Engineering Science, 22, 503-516. industrial site. Industrial Engineering & Chemical Research, 37,
Cuthrell, J. E., & Biegler, L. T. (1987). On the optimization of 1341-1350.
differential-algebraic process systems. American Institute of Logsdon, J. S., Diwekar, U. M., & Biegler, L. T. (1990). On the
Chemical Engineering Journal, 33(8), 1257-1270. simultaneous optimal design and operation of batch distillation
DeGarmo, J. L., Parulekar, V. N., & Pinjala, V. (1992). Consider columns. Chemical Engineering Research Design, 68, 434-444.
reactive distillation. Chemical Engineering Progress, 88(3), 43- Mackowiak, J. (1991). Fluiddynamik von Kolonnen mit modernen
50. Fiillkijrpern und Packungen fur Gas/Fliissigsysteme, 1. Auflage,
Diwekar, U. M. (1992). Unified approach to solving optimal de- Otto-Salle-Verlag, Frankfurt.
sign-control problems in batch distillation. American Institute of Sorensen, E. (1994). Studies on optimal operation and control of
Chemical Engineering Journal, 38, 1551- 1563. batch distillation columns. Ph.D. thesis, Norwegian Institute of
Diwekar, U. M., Malik, R. K., & Madhavan, K. P. (1987). Opti- Technology, Trondheim.
mal reflux rate policy determination for multicomponent batch Sorensen, E., Macchietto, S., Stuart, G., & Skogestad, S. (1996).
distillation columns. Chemical Engineering Progresss, 88(3), 43- Optimal control and on-line operation of reactive batch distilla-
50. tion. Computers & Chemical Engineering, 20(12), 1491- 1498.
Egly, H., Ruby, V., & Seid, B. (1979). Optimum design and opera- Vassiliadis, VS. (1992). DAEOPT-A differential-algebraic optimal
tion of batch rectification accompanied by chemical reaction. control problem solver - version 2.0. Internal report, Imperial
Computers di Chemical Engineering, 3, 169-174. College, London.
Fernholz, G., & Engell, S. (2000). Nonlinear model predictive con- Vassiliadis, V. S., Sargent, R. W. H., & Pantelides, C. C. (1994ab).
trol of a semibatch reactive distillation column. Process Control Solution of a class of multistage dynamic optimization prob-
& Instrumentation, 7 (pp. 26-28), Glasgow, Scotland (in press). lems. 1. Problems without path constraints, 2. Problems with
Fernholz, G., Wang, W., Engell, S., Fougner, K, & Bredehoft, J. path constraints. Industrial Engineering & Chemical Research,
(1999). Operation and control of a semibatch reactive distilla- 33. 2111-2133.