PROCESS DESIGN AND CONTROL OF A REACTIVE DISTILLATION SYSTEM

Pinky Panjwani, Myrian Schenk1, Michael C. Georgiadis and Efstratios N. Pistikopoulos

Imperial College London,

Department of Chemical Engineering, Centre for Process Systems Engineering, London SW7 2AZ, UK

Abstract: A Mixed-Integer Dynamic Optimization (MIDO) model and solution strategy to

study the interactions between process design and process control/operability of an ethyl
acetate reactive distillation system is presented. It is shown that the resulting design and
control scheme can guarantee feasible operation under bounded uncertainty at a minimum
total average cost, representing 17 to 20% savings over the original design. Copyright
2004 IFAC
Keywords: Process design; Control; Reactive distillation; Mixed integer optimization.

1. INTRODUCTION The control of reactive distillation has received some

attent ion in recent years. Sneesby et al. (1999)
In recent years chemical process industries have discussed the advantages of a combined composition
shown an increasing interest in the development of and conversion scheme for an ETBE column.
reactive distillation processes combining reaction Monroy-Loperena et al. (2000) studied the control
and separation mechanisms into a single, integrated problem of an ethyl glycol reactive distillation
unit. Such processes offer several important system and proposed a robust PI control
advantages such as reduction in total costs and configuration. Sneesby et al. (2000) proposed an
energy consumption, overcoming of thermodynamic integrated control scheme for an ETBE column by
limitations, (e.g. azeotropes) and increased reaction permitting the control objectives to be changed on-
yield and selectivity. However, reactive distillation is line in order to reflect changing economic contracts.
not extensively used in the chemical process Vora and Daoutidis (2001) studied ht e dynamics and
industries since its operation is more difficult and control of an ethyl acetate reactive distillation system
poses higher requirements on the quality of the and proposed a new feed configuration for the two
design and control than conventional flowsheets in reactants that allows higher conversion and purity
which a reactor is typically followed by a train of than the conventional configuration, which involves
distillation columns (Engell and Fernholz, 2003). feeding in a single tray. Al-Arfaj and Luyben
(2002) compared the control of an ideal reactive
Modelling of reactive distillation has received distillation column with that of a similar real
considerable attention over the last 10 years and chemical system, the production of methyl acetate. A
several key contributions have appeared in the number of control structures were evaluated for both
literature (see the book of Doherty and Malone, 2001 systems. Gruner et al. (2003) proposed a non-linear
and the overview of Noeres et al., 2003). Pilavachi et control scheme for an industrial reactive distillation
al. (1997) presented an extensive discussion of column operated by Bayer AG. Bisowarno et al.
several important aspects that affect the accurate (2003) investigated model gain scheduling for an
modelling of reactive distillation processes. Schenk ETBE reactive distillation column. It was illustrated
et al. (1999) described in considerable detail a that the proposed control strategy outperforms the
hybrid-modelling environment in which a reactive standard PI control.
distillation process can be simulated using a
combination of equilibrium and mass transfer The work in the area of integration of design and
models, both in steady state and dynamic modes. control in reactive distillation systems is very limited.
Heath et al. (2000) studied the interactions of design

1
Present address, Air Products PLC, Molesey Road,
Walton-on-Thames, Surrey, U.K.
and control in an ethyl glycol reactive distillation ! slow-moving disturbance in the cooling
system assuming that the process structure is fixed water inlet temperature representing diurnal,
(e.g. number of trays, feed tray location, etc). Schenk ambient variations;
(1999) and Georgiadis et al. (2002) compared ! product quality specifications;
sequential and simultaneous approaches to design ! minimum column diameter to avoid
and control of a reactive distillation system, using flooding;
advanced optimization techniques. Again the ! temperature driving forces in the reboiler
process and control structure were assumed fixed. and condenser;
! fractional entrainment limits;
The aim of this work is to explore the synergistic ! control structure specifications.
benefits between process design, process control and
operability in a reactive distillation column for the The solution of this problem would determine:
production of ethyl acetate. The process structure (a) the optimal process design, i.e., number of trays
(number of trays, feed tray location), the process and feed location (discrete decisions), the
design (column diameter, reboiler and condenser diameter of the column, and the surfaces areas of
surface areas) and control structure/design (pairings reboiler and condenser (continuous decisions);
of measurements and manipulated variables, (b) the optimal PI control design, i.e., the pairings of
controller tuning parameters) are all simultaneously manipulated and controlled variables (discrete
optimized. The problem is posed as a Mixed-Integer decisions) and the tuning parameters (gain,
Dynamic Optimization (MIDO) problem and solved integral time) of the PI controllers (continuous
using recent algorithmic advances developed in our decisions).
group at Imperial College (Bansal, et al., 2003).
Note that an original design of diameter 5.2 m,
2. PROBLEM DEFINITION reboiler area 195 m2 and condenser area 495 m2 is
considered as a base case for comparison of the
The reactive system of ethyl acetate was considered results obtained in this paper.
in this work, since experimental data is openly
Slow disturbance
available in the literature. This enables to validate the Twin = 298.15*sin(2*PI/3600*DisFeed)
modelling results as well as the application of the
proposed optimization framework.

The main problems encountered in achieving high 1

purity products in the ethyl acetate reactive

distillation system are summarized below (Bock et
al. (1997); Chang and Seader (1988)): Optimal Feed Location ?
Distillate Spec
x d,ethylacetate > 0.50
! unfavourable reactant conversion;
! similar K-values of ethanol, water and ethyl Optimal Control Scheme ?
6
acetate;
! temperature profile in the column; Feed Flow rate = 4885 Kmol/hr
Temperature = 325 K

! the system is strongly non-ideal due to the Fast Disturbance in feed

compostion of Acetic Acid
presence of ethanol, acetic acid and water. Optimal Number of Trays ?
The separation of pure components is very
difficult due to the existence of five normal
azeotropes, namely, ethanol-water; water-
N
acetic acid; ethyl acetate-ethanol; ethyl
acetate-water; ethanol-ethyl acetate-water. Bottom Spec
x b,ethylacetate < 0.22

Full details of this system as well as a rigorous

dynamic model together with the thermodynamic
method employed are presented in Georgiadis et al.
(2002).
Fig. 1 Ethyl acetate column superstructure
Figure 1 depicts the ethyl acetate column
superstructure considered where a single feed
configuration is assumed. The objective is then to 3. SOLUTION STRATEGY
design the column and its required PI control scheme
at minimum total annualised cost (capital cost and The methodology employed in this work relies on a
operating cost), which can guarantee feasible systematic mixed-integer dynamic optimization
operation over a finite time horizon, subject to a set (MIDO) framework exploring simultaneously the
of constraints (Georgiadis et al., 2002) such as: interactions between process and control system
! model equations; design (Bansal et al., 2003). Two strategies were
! high-frequency sinusoidal disturbance in the employed. In the first strategy, the control scheme is
acetic acid inlet composition; fixed, while the design of the column is optimized.
The column design includes both discrete decisions
such as number of trays and feed location, as well as now also includes the control structure selection as
continuous decisions (column diameter, exchanger additional optimization decisions.
areas, controller tuning parameters). The second
strategy additionally incorporates the control Some process knowledge restrictions were imposed
structure selection (discrete decision). In this way, for the solution. In the master problems, based on the
the economic and operability benefits of fact that a very long time response is undesirable
simultaneously optimizing the process and control between the action of a manipulated variable and its
design can be revealed and further assessed. Full effect on a controlled variable, pairings such as
details and files can be obtained from the authors by Reflux rate Bottom ethyl acetate composition,
request. Distillate flowrate Bottom ethyl acetate
composition, Steam flowrate Top ethyl acetate
3.1 MIDO Strategy with Fixed Control Structure composition, Bottoms flowrate Top ethyl acetate
composition, are not considered from the search
The ethyl acetate column superstructure consisted of space; and, due to the available degrees of freedom in
13 trays with feed located on the 6th tray. Some the system, only two controllers are allowed.
restrictions were imposed for the solution of this
large-scale problem. Constraints like feed and reflux Table 1 Progress of Iterations for the MIDO strategy
entering in only one tray; and, feed entering below with fixed control structure
the reflux were included in the mathematical
framework. The fixed control scheme incorporated Iteration No 1 2 3
had two proportional integral controllers, as normally Primal
used in distillation systems, namely: Solutions
Discrete
# PI1: Reflux rate (R) Top ethyl acetate Decisions
composition (x4,d) No of trays 13 12 11
# PI2: Steam flow rate (Fs) Bottom ethyl Feed location 6 6 5
acetate composition (x4,r) Process Design
Diameter of
Since the period of the slow-moving disturbance in column 5.00 4.95
the cooling water temperature is 3600 hours, it is Reboiler --Inf--
deemed that this time is adequate as time horizon surface area 170.00 168.00 (Infeasible
while rendering the large -scale dynamic optimization Condenser point)
problem tractable. The objective function is defined surface area 480.00 478.00
as follows, Controller
gains
min ns PI1 25 25 --Inf--
( Expected )Cost = C cap + w j C op, j (1) PI2 5726.00 5725.66
d j =1
Controller
Reset times
PI1 60 60 --Inf-
For this particular case, the resulting MIDO problem PI2 4936.00 4935.54
consists of 1652 model equations (81 differential and Controller Bias
1571 algebraic) with 26 binary (0-1) variables and 36 PI1
optimization parameters (9 inequality constraints and PI2 3065.00 3074.54 -Inf--
27 equality constraints). The MIDO problem is 59000.0 58950.0
decomposed into primal problem and master Capital Cost 297000 286500
problem. The dynamic optimization primal problems Operating Cost 249280 249625 --Inf--
(NLPs) were solved using gPROMS/gOPT (Process Final Cost 546280 536125
Systems Enterprise Ltd., 1999) and the master UB 546280 536125
problems (MILPs) were solved using Master
GAMS/CPLEX. (Brooke et al., 1992). Solutions
No of trays 12 11
Final convergence, i.e. the solution of the problem Feed tray 6 5
was obtained in two iterations. Table 1 presents the LB 534330 535890
full details of the primal and master solutions for UB-
both iterations. The optimal column configuration LB<Tolerance NO NO STOP
has 12 trays with the feed located on the 6th tray and
the corresponding final total cost is 536125 $/yr.
The solution to this problem is summarised in Table
2. The optimal column configuration consists of 12
3.2 Overall MIDO Strategy
trays with optimal feed located on 7th tray. The final
The overall MIDO strategy considers the same total cost obtained is 517000 $/yr. The optimal
column superstructure, objective function and control scheme obtained is pairing reflux flowrate
optimization variables as the previous strategy but with top temperature of the column (Tc) and steam
flowrate with feed tray temperature (T7). The
noticeable fact, from this result, is that in order to as reactive. By introducing the binary variables to
obtain better purity and maximum conversion, in the model reactive and non-reactive trays in the
case of reactive distillation, it is important to control superstructure, it is expected that the performance of
the temperature of the column, both feed tray as well the design obtained will be better than that of the
as product (top column temperature, for this fully reactive distillation column.
particular case). This somehow contrasts with the
conventional control scheme in distillation columns,
Reaction occuring in all trays
where top and bottom compositions (or
temperatures) are controlled in order to attain better tray1
90

Mole Fraction (Ethyl Acetate)

performance. Moreover, for this particular case the 80
tray2
composition of ethyl acetate is reduced in the bottom 70 tray3
60 tray4
of the column by almost 10%, indicating less product
50 tray5
losses and better separation when using this control 40 tray6
scheme. 30 tray7
20 tray8
Figs. 2 and 3 show the top and bottom product 10
tray9
0
composition at the optimal solution. -10 0 1000 2000 3000 4000
tray10
Time Horizon (hrs)
tray11
Top composition (Ethyl Acetate) tray12
Fig. 4 Reaction profile in all the trays
0.535
Mole fraction (Ethyl acetate

0.53 3.3 Discussion of Results

0.525
0.52
The results obtained by using integrated MIDO
0.515
approach are summarised as follows:
0.51
0.505 The simultaneous consideration of process
0.5 design and control results in economic savings
0.495 of around 20% compared to the original design
0 500 1000 1500 2000 2500 3000 3500 4000 thus highlighting the importance of addressing
Timehorizon(hrs) the interactions and synergistic benefits that
exist between process design and process control
Fig. 2 Ethyl acetate top composition in the early design stage. For the case when only
process design is considered the savings are
around 17%.
Bottom composition (Ethyl acetate) It was also found that by determining the design
0.172 and control structure simultaneously, savings of
Mole fraction (Ethyl acetate

0.17 almost 3.7% are obtained compared to the case

0.168
0.166 where the control structure is fixed. The
0.164 reduction in the capital cost and the operating
0.162
0.16 cost are approximately 1.5% and 6%
0.158 respectively. There is also a considerable
0.156
0.154 reduction in the reboiler duty and the condenser
0.152
0 500 1000 1500 2000 2500 3000 3500 4000
duty, in the order of 10%. The reboiler and the
Time horizon (hrs)
condenser areas are significantly reduced.
Moreover, the reflux and steam flow rates are
Fig. 3 Ethyl acetate bottom composition reduced. This strongly portrays the impact that
the interactions between design and control have
Both figures illustrate that the ethyl acetate on the operability.
composition satisfies the product quality constraints Simultaneous design and control strategy results
over the entire time horizon (at least 50% of ethyl in the reduced composition of ethyl acetate in
acetate at the top; and, at the most 20% at the bottom the bottom stream, which implies less product
of the column). losses, than with fixed control structure strategy.
For this reactive distillation system, it is required
The temperature profile of a reactive distillation to control the temperature of the reactive tray in
column is closely related to the extent of reaction or which maximum extent of reaction takes place.
reaction rate. Figure 4 depicts the reaction rate Controlling feed tray temperature emphasizes the
throughout the column on the solution of the MIDO fact that the behaviour of reactive distillation
problem. It is observed that most of the reaction is column is different than the conventional
actually taking place at the bottom part of the distillation, where normally the composition of
column. Since reaction takes place mainly in the products (or top and bottom temperatures) are
bottom section of the column with tray 7 (feed tray) controlled to obtain better purity. The
having the maximum extent of reaction, a new temperature in the reactive section needs to be
column configuration could be considered where top controlled to obtain maximum conversion and
trays are considered as non-reactive and bottom trays better purity of the product.
Table 2 Progress of Iterations for the overall MIDO 4. CONCLUSIONS
strategy
This work considers an advanced MIDO framework
Iteration No 1 2 3 for studying the interaction between process design,
process control and operability in a reactive
Primal
Solutions distillation system. The results demonstrate the
Discrete benefits from pursuing a simultaneous approach in
Decisions process and control design rather than the traditional
sequential approach. Economic advantages of the
13 12 11 order of 3.7% are obtained with feasible dynamic
No of trays
performance, production specifications and
Feed location 6 7 3 operational restrictions despite the presence of
st
1 controller R-x4,d R-Tc R-x4,d rapidly varying disturbances.
nd Fs-x4,r Fs-T7 Fs-Tc
2 controller Although this work has focused on the design of
traditional PI controllers, more advanced control
Process Design
schemes (e.g. model-based parametric controllers)
could be incorporated based on recent advances in
Diameter of our group (Sakizlis et al., 2003; Sakizlis, 2003). It is
column (m) 5.00 5.10
thus expected that even more inexpensive designs
could be achieved with proven dynamic operation
Reboiler surface and economics over the simultaneous process and
area (m2) 170.00 140.00 --Inf--
control design featuring conventional controllers.
Furthermore, the importance of new binary variables
Condenser in the MIDO framework for the automatic
surface area 480.00 470.00
identification of the type of tray (reactive or non-
(m2) reactive) is expected to lead to more economic
Controller gains benefits since the cost of catalyst would be reduced.
1st 25 42.52 --Inf--
nd ACKNOWLEDGEMENT
2 5726.00 5174.00
Controller Reset The authors would like to thank financial support
times from the European Union, Shell Chemicals and Air
Products and Chemicals. P. Panjwani would also like
1st 60 28.00 --Inf-- to acknowledge financial support from the British
nd Chevening scholarship shared with Imperial College.
2 4936.00 4966.00
REFERENCES
Controller Bias
1st 3065.00 2790.00 Al-Arfaj and W.L. Luyben (2002). Comparative
nd control study of ideal and methyl acetate reactive
2 59000.0 55830.0 --Inf-- distillation. Chemical Engineering Science, 57,
Capital Cost 297000 282000 5039-5050.
Bansal, V., V. Sakizlis, R. Ross, J.D. Perkins and
Operating Cost 249280 235000 --Inf-- E.N. Pistikopoulos (2003). New algorithms for
Final Cost 546280 517000 Mixed-Integer Dynamic Optimization,
Computers and Chemical Engineering, 27, 647-
UB 546280 517000 668.
Master Bisowarno B.H., Y.C. Tian and M.O. Tade (2003).
Solutions Model gain scheduling control of an ethyl tert -
butyl ether reactive distillation column,
No of trays 12 11 Industrial and Engineering Chemistry Research,
42, 3584-3591.
Feed tray 7 3 Bock, H., Jimoh, M. and G. Wozni (1997). Analysis
st
1 controller R-Tc R-x4,d of reactive distillation using the esterification of
nd
acetic acid as an example. Chemical Engineering
2 controller Fs-T7 Fs-Tc Technology, 20, 182-191.
LB 445135 470000 Brooke, A., D. Kendrick and A. Meeraus (1992)
GAMS Release 2.25: A Userss Guide. The
UB- NO NO STOP Scientific Press. San Francisco.
LB<Tolerance Chang, Y. A. and J.D. Seader (1988). Simulation of
continuous reactive distillation by a homotopy-
 continuation method. Computers and Chemical Vora, N. and P. Daoutidis (2001). Dynamics and
Engineering, 12, 1243-1257. control of an ethyl acetate reactive distillation
Doherty, M.F., and M.F. Malone (2001). Conceptual column, Industrial and Engineering Chemistry
design of distillation systems. McGraw-Hill Research, 40, 833
Chemical Engineering Series. New York.
Engell, S., and G. Fernholz (2003). Control of a
reactive distillation process. Chemical
Engineering and Processing, 42, 201-210.
Georgiadis, M.C., M. Schenk, E.N. Pistikopoulos
and R. Gani (2002). The interactions of design,
control and operability in reactive distillation
systems, Computers and Chemical Engineering,
26, 735-746.
gPROMS/gOPT (1999) User Guide. Process
Systems Enterprise Ltd..
Gruner, S., K.D. Mohl, A. Kienle, E.D. Gilles, G.
Fernholz and M. Friedrich (2003). Non-linear
control of a reactive distillation column, Control
Engineering Practice, 11, 915-925.
Heath, J.A., I. Kookos and J.D. Perkins (2000).
Process control structure selection based on
economics, American Institute of Chemical
Engineers Journal, 46, 1998-2016.
Monroy -Lopera, R., E. Perez -Cisneros and J.
Alvarez-Ramirez (2000). A robust PI control
configurat ion for a high purity ethylene glycol
reactive distillation column, Chemical
Engineering Science, 55, 4925-4937.
Noeres, C., E.Y. Kenig and A. Gorak, (2003).
Modelling of reactive separation processes:
reactive absorption and reactive distillation,
Chemical Engineering and Processing, 42, 157-
178.
Pilavachi, P., M. Schenk, E. Perez-Cisneros and R.
Gani (1997). Modelling and Simulation of
Reactive Distillation Operations, Industrial and
Engineering Chemistry Research, 36, 3188-
3197.
Sakizlis, V., J.D. Perkins and E.N. Pistikopoulos
(2003). Parametric controllers in simultaneous
process and control design optimization,
Industrial and Engineering Chemistry Research,
42, 4545-4563.
Sakizlis, V., (2003) Design of Parametric Controllers
via Parametric Programming. PhD Thesis.
Imperial College London, University of London.
Schenk, M.A., (1999), Design of operable reactive
distillation columns. PhD Thesis. University of
London.
Schenk, M., R. Gani, D. Bogle and E.N.
Pistikopoulos (1999). A Hybrid Modelling
Approach for Separation Systems Involving
Distillation, Chemical Engineering Research
and Design, Part A, 77,519-534.
Sneesby, M.G., O.M. Tade and T.N. Smith (1999).
Two-point control of a reactive distillation
column for composition and conversion, Journal
of Process Control, 9, 19-31.
Sneesby, M.G., M.O. Tade, T.N. Smith (2000). A
multi-objective control scheme for an ETBE
reactive distillation column, Chemical
Engineering Research and Design, 78 (A2),
283-292.