You are on page 1of 9

378

Ind. Eng. Chem. Res. 2000, 39, 378-386

PROCESS DESIGN AND CONTROL


Output-Feedback Control of Reactive Batch Distillation Columns
Rosendo Monroy-Loperena and Jose Alvarez-Ramirez*,
Estrategia Sinergica S.A. de C.V., Paseo de los Pirules 124, Col. Paseos de Taxquen a,
Mexico D.F., 04250 Mexico, and Departamento de Ingenieria de Procesos, Universidad Autonoma
MetropolitanasIztapalapa, Apartado Postal 55-534, Mexico D.F., 09340 Mexico

In this work, an output-feedback control for the regulation of distillate purity via manipulations
of the reflux ratio in reactive batch distillation is designed. The approach is based on an
approximate model of the composition dynamics and makes use of a reduced-order observer to
estimate the modeling error. An input/output linearizing feedback is proposed where the
estimated modeling error is included to achieve robust tracking of a composition reference. It is
shown that the resulting controller has the structure of a proportional-integral derivative (PID)
controller with antireset windup. The controller performance is tested using a simulation example
including strong uncertainties in the reaction model. An interesting finding is that the required
reflux ratio policy to reach asymptotically a constant reference resembles the reflux ratio policy
obtained from posing an optimization technique (Mujtaba, I. M.; Macchietto, S. Ind. Eng. Chem.
Res. 1997, 36, 2287-2295).
1. Introduction
Batch distillation with or without chemical reaction
is used in industry for the production of small amounts
of products with high added value and for processes
where flexibility is required. Distillation with chemical
reaction is well suited for processes where one of the
products has a lower boiling point than other products
and reactants. The higher volatility of this product
induces a decrease of its concentration in the liquid
phase, thus leading to higher reactant conversions than
with reaction alone.
Batch processes, as reactive batch distillation, are
inherently dynamic. In a rough sense, they can be seen
as composed of a stationary process driven by an
integrator. While both the dynamic modeling of reactive
batch distillation (RBD) and its optimization have been
studied to some extent,1-3 issues related to the feedback
regulation of this process have rarely been addressed.
A method for computing operational policies for RBD
systems using simulation techniques was reported by
Albet et al.1 Sorensen and Skogestad4 developed control
strategies by repeated simulations with the SPEEDUP
package. Some aspects related to the controllability of
RBD columns were also addressed. It is clear that a
drawback of this approach is that extensive simulations
using a simple model are required, thus leading to a
high computational burden. In a recent work, Mujtaba
and Macchietto3 reported a method for obtaining optimal operational policies for the RBD process. Their
approach is presented as a proper dynamic optimization
problem incorporating a detailed dynamic model, which
* Corresponding author. E-mail: cdp@xanum.uam.mx.
Fax: +52-5-7244900.
Estrategia Sinergica S.A. de C.V.
Universidad Autonoma MetropolitanasIztapalapa.

resulted in a nonlinear programming problem with the


dynamics of the RBD process as constraints. In general,
the solution of the resulting optimization problem can
be computationally very expensive3 and therefore may
not be suitable for on-line implementation. To alleviate
this problem, Mujtaba and Macchietto3 used polynomial
curve-fitting techniques. As a result, it was shown that
by judicious use of repeated solutions of the dynamic
optimization problem a priori, an algebraic representation of the optimal solutions can be obtained and very
efficient calculation of the optimal reflux ratio can be
performed.
A criticism to optimal policies approaches, such as
that reported by Mujtaba and Macchietto,3 is that the
optimal solution (optimal reflux ratio) depends strongly
on the model. Because models may present strong
uncertainties in the parameters and reaction rates,
these may lead to serious robustness problems and
performance degradation with respect to the computed
optimal profit. In fact, Mujtaba and Macchietto3 have
shown that changes in system parameters can significantly change the operating condition (the reflux rate
policies) of the plant.
During the past decade, the feedback linearization
control technique has been successfully used to address
some of the practical control problems, such as the
control of nonlinear fermentation processes,5 polymerization processes,6,7 and pH neutralization processes.8
Successful applications to nonminimum-phase nonlinear
systems have been reported.9-11 The important practical
case of output-feedback regulation has also been
addressed,11-14 where the idea is to use some kind of
state estimation within an input/output feedback linearization framework. In the spirit of these papers, our
work addresses several aspects of the output control of
RBD processes. Because the profitability of the process
is closely related to the distillate composition,3 the

10.1021/ie990382l CCC: $19.00 2000 American Chemical Society


Published on Web 12/16/1999

Ind. Eng. Chem. Res., Vol. 39, No. 2, 2000 379


Table 1. Input Data for Ethanol Esterification Using
Conventional Batch Distillation
no. of ideal separation stages (including
reboiler and total condenser)
total fresh feed (kmol)
feed composition (acetic acid, ethanol,
ethyl acetate, water) (mole fraction)
internal plates holdup (kmol)
condenser holdup (kmol)
condenser vapor load (kmol/h)
column pressure (bar)

10
5
0.45, 0.45,
0.0, 0.1
0.1
0.0125
2.5
1.013

Table 2. Vapor-Liquid Equilibrium and Kinetic Data for


Ethanol Esterification
Vapor-Liquid Equilibrium
acetic acid (1) + ethanol (2) a ethyl acetate (3) + water (4)
K1 ) (2.25 10-2)T - 7.812, T > 347.6 K
K1 ) 0.001, T e 347.6 K
log K2 ) -2.3 103/T + 6.588
log K3 ) -2.3 103/T + 6.742
log K4 ) -2.3 103/T + 6.484
Kinetic Data
rate of reaction, gmol/(L min); r ) k1C1C2 - k2C3C4, where
rate constants are k1 ) 4.76 10-4 and k2 ) 1.63 10-4 and Ci
stands for concentration in gmol/L for the ith component

Figure 1. Schematic diagram of the reactive batch reactifier.

control objective is to track a prescribed distillate


composition via manipulations of the reflux ratio. It is
assumed that the condenser duty is used for pressure
control and the distillate flow for condenser level control.
The main interests are (i) establishing an outputfeedback strategy with guaranteed tracking properties
despite strong uncertainties in the dynamics of the RBD
process and (ii) showing via a specific example from
Mujtaba and Macchietto3 that the resulting reflux ratio
policy approaches that obtained via optimization techniques. The control design is based on an approximate
model of the composition dynamics and makes use of a
reduced-order observer to estimate the modeling error.
An input/output linearizing feedback is proposed where
the estimated modeling error is included to achieve
robust tracking of a composition reference. It is shown
that the resulting controller has the structure of a PID
controller with an antireset windup scheme. The controller performance is tested using a simulation model
including strong uncertainties in the reaction model. An
interesting finding is that the required reflux ratio
policy to reach asymptotically a constant reference
approximates very closely that obtained from using
optimization techniques.3
2. Problem Statement
A schematic diagram of a RBD column is presented
in Figure 1. A dynamic model for an n-stage RBD
process consists of the mass- and energy-balance equations.2,15,16 The model includes column holdup, rigorous
phase equilibria, and chemical reaction on the plates.
The model is fairly detailed and assumes negligible
vapor holdup on plates, perfect mixing on trays, fast
energy dynamics, constant operating pressure, and total
condensation with no subcooling. The stages are counted
from top to bottom. Subindex D is assigned to the

condenser drum, and subindex R is assigned to the


reboiler drum. A complete description of the model is
made by Cuille and Reklaitis15 and Mujtaba and Macchietto.3 For the sake of completeness, the model and
its main variables are described in the appendix. It
should be remarked that although formation of azeotropes are quite common in reactive distillation, this
situation is not considered in the model, for convenience
and simplicity in presentation.
The worked example presented by Mujtaba and
Macchietto3 considers the esterification of ethanol and
acetic acid. The reaction products are ethyl acetate and
water. The reversible reaction scheme is the following:

acetic acid (1) + ethanol (2) a


ethyl acetate (3) + water (4)
The boiling temperatures are respectively 391.1, 351.5,
350.3, and 373.2 K. Ethyl acetate, the main product,
has the lowest boiling temperature in the mixture and
consequently has the highest volatility. The continuous
removal of this product by distillation will shift the
chemical equilibrium further to the right and will
improve conversion of reactants. The data defining the
column configuration, feed, feed composition, etc., for
the example are given in Table 1 of Mujtaba and
Macchietto.3 Table 2 of the same work presents the
vapor-liquid equilibria and kinetic data.
Although our results are presented for this example,
they are intended for the general case of RBD processes.
Let xD,3 be the concentration of ethyl acetate in the
distillate flow. In addition, let wref(t) be a reference
trajectory. The control problem is to track the wref(t) by
manipulations of the reflux ratio rf ) L0/V1.
3. A State Feedback Control Design
In this section, we will build a feedback control under
the assumption of complete knowledge and state measurement. To this end, we will follow the methodology
described by Barolo and Berto17 for nonreactive distillation columns.
For simplicity in algebraical manipulations, let us
take the reflux flow rate L0 as the manipulated variable.

380

Ind. Eng. Chem. Res., Vol. 39, No. 2, 2000

Moreover, let us assume that the vapor flow rate in the


first plate V1 and the condenser drum holdup hD vary
slowly; i.e., dV1/dt 0 and dhD/dt 0. The reflux ratio
is then given by rf ) L0/V1. The dynamics of the distillate
composition xD,3 are given by (see the appendix)

hD

dxD,3
) V1(y1,3 - xD,3) + hDRD(XD,TD)
dt

dx1,3
) V2y2,3 + L0xD,3 - V1y1,3 - L1x1,3 +
h1
dt
dh1
h1R1(X1,T1) - x1,3
(2)
dt
For simplicity in notation, introduce the following
functions:

fD,3 ) [V1(y1,3 - xD,3) + hDRD(XD,TD)]/hD

(3a)

f1,3 )
V2y2,3 - V1y1,3 - L1x1,3 + H1R1(X1,T1) - x1,3

dh1
/h1
dt
(3b)

Because y1,3 ) E(x1,3) (liquid-vapor equilibria relationship), we have that

hD

d2xD,3

)
dt2
dE(x1,3)
dRD
V1
[f1,3 + (x3,D/h1)L0] - fD,3 + hD
(4)
x1,3
dt

[(

where

dRD RD dXD RD dTD


)
+
dt
XD dt
TD dt
and dXD/dt and dTD/dt are given by the mass and energy
balances in the condenser. In eq 4, the control input L0
affects directly the dynamics of the controlled concentration xD,3 via the second-order derivative d2xD,3/dt2,
and therefore the relative degree of the system is 2.17
Let e ) xD,3 - wref be the tracking error. Suppose the
following stable error trajectory description:18-20

de
d2e
+ 2cc-1
+ c-2e ) 0
dt
dt2

d2xD,3
dt

d2wref
dt

+ 2cc-1

(1)

where XD ) (xD,1, ..., xD,4) is the distillate composition,


y1,3 is the ethyl acetate mole fraction in the vapor
leaving the first stage (the column stages are counted
from top to bottom), RD(XD,TD) is the reaction rate, and
TD denotes the temperature in the condenser drum.
Following the methodology described by Barolo and
Berto,17 the relative degree of the system is not 1
because the dynamics of xD,3 are not directly affected
by the control input L0. On the other hand, the dynamics
of the ethyl acetate concentration in the first stage are
given by

for xD,3 are given by

(5)

where c and c are respectively the closed-loop damping


factor and time constant. Then, the desired dynamics

dwref
+ c-2wref dt
dxD,3
2cc-1
- c-2xD,3 (6)
dt

From eqs 4 and 6, we obtain the theoretical control


input LT0 , which provides the trajectory tracking error
behavior (5) and is given by

LT0 ) - +

d2wref
dt

+ 2cc-1

dwref
+ c-2wref dt

2cc-1

dxD,3
- c-2xD,3 / (7)
dt

where

[(

) (h1hD)-1 V1

dE(x1,3)
dRD
f1,3 - fD,3 + hD
dx1,3
dt

(8a)

and

) (V1/hDh1)xD,3

dE(x1,3)
dx1,3

(8b)

The control input LT0 is well-defined provided that (t)


* 0, for all t > 0. The main feature of the feedback
function (7) is that it leads to asymptotic convergence
of the tracking error to zero within a mean operating
time c. In this way, tuning of the controller (7) can be
easily made just by choosing the closed-loop time
constant c and damping coefficient c. To avoid unrealistic situations due to hard input bounds, a saturated version of eq 7 is proposed:
T
) Sat(LT0 ; L0,max, L0,min)
L0,sat

(9)

where Sat is a standard saturating function with upper


and lower limits L0,max and L0,min, respectively.
A. Numerical Simulations with the Theoretical
Control Law (9). We have carried out several numerical simulations with the example presented by Mujtaba
and Macchietto.3 Figure 2 shows the dynamics of the
uncontrolled distillate composition for several values of
the reflux ratio rf ) L0/V1. It is noted that the concentration of the reactant acetic acid in the distillate flow
goes to zero immediately. However, the concentration
of this reactant is very high in the bottom plates (not
shown). This behavior is due to the fact that the acetic
acid has the higher boiling temperature, so that it is
maintained at the bottom plates where the chemical
reaction mechanism becomes more important than the
distillate separation mechanism. For total reflux operation, the maximum achievable ethyl acetate concentration is 0.964 mole fraction. This value imposes a limit
in the achievable product purity under batch operation.
In fact, for batch operation (0 < rf < 1), the ethyl acetate
mole fraction increases in the first part of the batch
time, achieves a maximum value, and then decreases
until the end of the operation. The first part of the batch
operation where the ethyl acetate mole fraction increases can be called the reaction phase because the
chemical reaction is the main drive of ethyl acetate in
the distillate product. Compared with nonreactive batch

Ind. Eng. Chem. Res., Vol. 39, No. 2, 2000 381

Figure 2. Dynamics of the uncontrolled distillate composition for several values of the reflux ratio rf ) L0/V1.

distillation where the mole fraction of the more volatile


component decreases along the operation of the uncontrolled batch distillation, the last part of the batch
operation can be called the separation phase. In fact,
the process is controlled by the separation during the
second phase. It is also noted that the smaller the reflux
ratio, the smaller the operating time where the maximum ethyl acetate mole fraction is achieved. Hence,
inefficient batch operation is obtained with smaller
values of the reflux ratio.
As in Mujtaba and Macchietto,3 assume that the
control objective is to regulate the distillate composition
at a given constant product purity wref. We have carried
out numerical simulations for two given purities, wref
) 0.7 and 0.8. The reflux ratio level was computed with

the feedback control law (8) for batch time tf ) 40 h,


although there is not an inherent restriction to compute
the reflux flow rate for any batch time. The reboiler heat
duty was kept constant. The control parameters were
chosen as c ) 1 and c ) 0.01 h. Besides, L0,max ) V1
and L0,min ) 0.8 V1 (i.e., rf,max ) 1 and rf,min ) 0.8). Figure
3 shows the dynamics of the computed reflux ratio and
product purity. After about 2 h, the product purity is
maintained at its reference value until the end of the
batch time. The reflux ratio attains its upper limit
(L0,satT/V1 ) 1) during the first part of the batch
operation. This behavior is due to the fact that the
column is started up with zero mole fraction of products
(ethyl acetate and water) in all trays. During the
reaction phase, the reflux ratio decreases to compensate

382

Ind. Eng. Chem. Res., Vol. 39, No. 2, 2000

Figure 3. Dynamics of the computed reflux ratio and product


purity under the theoretical feedback control law (9).

for the high capacity of the process to give product at


the desired purity. This is expected because, as the
product species are withdrawn by distillation, the
reaction goes further to the right. As Mujtaba and
Macchietto3 have pointed out, this increase is very sharp
at the beginning because it is easier to shift equilibrium
by eliminating the plentiful product at the given purity.
The curve is flattened near the end of the batch time
because it is progressively more difficult to remove the
product at the given purity. At a certain batch time
(about 6 h), the separation mechanism starts to control
the process dynamics and the column behaves like a
nonreactive batch column. During the separation phase,
the reflux ratio increases to compensate the decrease
in the reaction drive to generate ethyl acetate. This
behavior is maintained until the end of the batch time.
Compare the behavior of the reflux ratio vs batch time
in Figure 2 with that in Figure 5 of Mujtaba and
Macchiettos paper.3 Both reflux ratio policies display
the same shape with almost the same quantitative
evolution. An advantage of the feedback control approach (9) is that this is easy to compute. On the
contrary, the computation of the dynamic optimization
problem seems to be computer time-consuming.
4. Robust Output Feedback Control Design
As in Mujtaba and Macchiettos approach, given an
accurate model of the RBD process, the computed reflux
ratio policy can be computed off-line and used in the
open loop. A drawback of this approach is that reaction
rate parameters may be highly uncertain, which may
lead to serious degradation of the optimized performance (e.g., maximum profit or maximum conversion).

Figure 4. Time evolution of the reflux ratio and product purity


for wref ) 0.7 and two different values of the estimation time
constant e.

On the other hand, an on-line computation of eq 7 would


require measurements of vapor and liquid composition
which is expensive. In this section, we will design a
feedback control for on-line implementation. To this end,
we will assume that the only concentration measurement is the product purity. This measurement can be
accomplished through a chromatograph or state estimators via temperature measurements (see the interesting
work by Quintero-Marmol and Luyben21 on this topic).
Equation 4 can be written as

d2xD,3
dt2

) + L0

(10)

It is noted that the function involves a set of quite


complex and uncertain functions, such as the time
derivative of the reaction rate dR1/dt. As a worst case
control design, assume that the function is unknown.
On the other hand, the function involves the gradient
of the vapor-liquid equilibrium relationship dE(x1,3)/
dx1,3. In general, it is expected that dE(x3,1)/dx3,1 be
positive. Let
j be an estimate of , which can be taken
as

j ) (V1/hDh1)wref dE(wref)/dx1,3

(11)

where dE(wref)/dx1,3 is the gradient of E(x) evaluated at


the reference value wref and V1/hDh1 can be computed
from the nominal design values V1, hD, and h1. Introduce the modeling error function

) + ( -
j )L0

(12)

Ind. Eng. Chem. Res., Vol. 39, No. 2, 2000 383

Then, eq 10 can be written as

d2xD,3
dt2

)+
j L0

(13)

Following the ideas described in section 3, the theoretical feedback control leading to desired closed-loop
behavior (5) is

LT0 ) - +

d2wref
dt

+ 2cc-1

dwref
+ c-2wref dt

2cc-1

dxD,3
- c-2xD,3 /
j (14)
dt

This feedback control cannot be implemented just as it


is because the modeling error and the time derivative
dxD,3/dt are not available for feedback. An alternative
is to use estimates of and dxD,3/dt. To this end,
introduce the variable z ) dxD,3/dt. Then, eq 12 can be
written as

dxD,3
)z
dt
dz
)+
j L0
dt
Let zj and
j be estimates of z and
j , respectively. The
following estimator is proposed:

dxD,3
dzj
)
j+
j L0 + 2ee-1
- zj
dt
dt

dxD,3
d
j
) e-2
- zj
dt
dt

Figure 5. Time evolution of the reflux ratio and product purity


for wref ) 0.8 and two different values of the estimation time
constant e.

(15)

where c and c are respectively the estimation damping


factor and time constant. In fact, in the case that )
constant, the following stable estimation error trajectory
description is obtained:
2

d ee
dt

+ 2ee-1

LP0 ) -
j+

+ 2cc-1

dwref
+ c-2wref dt

such that a saturated version becomes


P
) Sat(LP0 ; L0,max,L0,min)
L0,sat

where ee ) dxD,3/dt - zj is the estimation error. The


estimator (15) plays the role of a reduced-order observer
for the unmeasured states z and . To implement (15),
j define the variables q1 ) zj - 2ee-1xD,3 and q2 )
j are computed from
e-2xD,3. Then, the estimates zj and
the following equations:

dq1
)
j+
j L0 - 2ee-1zj
dt
(16)

(19)

In this way, the reflux ratio policy is computed online with the feedback function (19) and the estimators
(16) and (17). This control law has an interesting
structure. It can be interpreted as a PID-like control
law with an antireset windup (ARW) scheme.22,23 In fact,
after straightforward algebraic manipulations, it can be
concluded that the feedback control (16)-(19) can be
written as

LP0

where

-1

)
j

d2wref
dt2

- CPID(s) F(s) er P
GARW(s) (LP0 - L0,sat
) (20)

where CPID(s) is a classical PID controller with control


gain, integral, and derivative time constants given by

zj ) q1 + 2ce-1 dxD,3/dt

j ) q2 + c-2 dxD,3/dt

dt

j (18)
2cc-1zj - c-2xD,3 /

dee
+ e-2ee ) 0
dt

dq2
) -e-2zj
dt

d2wref

(17)

The initial conditions for eq 16 can be chosen as


follows. Because the signals z(t) and (t) are unknown,
take q1(0) ) -2ce-1xD,3(0) and q2(0) ) -e - 2xD,3(0).
In this way, the practical feedback control is given by

j -1
Kc )
D )

cc + ee
eec2 + cce2

c2 + 4ecec + e2
2(cc + ee)

384

Ind. Eng. Chem. Res., Vol. 39, No. 2, 2000

I ) 2(cc + ee)

(21)

F(s) is a first-order filter (i.e., F(s) ) 1/(fs+1)) with filter


time constant given by

f )

ce
2(ec + ce)

(22)

and GARW(s) is the following ARW operator acting on


P
:
the saturation error LP0 - L0,sat

GARW(s) )

c - 2ce2s
s(e2cs + 2e[ec + ce])

(23)

Regarding the structure of the PID control configuration


(20), the following comments are in order:
(a) If
j ) constant, the feedback function (16)-(19)
is a linear controller that can be easily implemented in
actual inexpensive technologies (e.g., programmable
logic controllers).
(b) When the actuator saturates, the feedback signal
P
P
) tries to drive the error LP0 - L0,sat
GARW(s) (LP0 - L0,sat
to zero by recomputing the integral action, such that
the controller output becomes exactly at the saturation
limit. This prevents the controller from windup.22,23
(c) It is noted that the PID control parameters are
symmetric functions of the nominal closed-loop parameters {c,c} and the observer parameters {e,e}. In
other words, the PID control parameters are invariant
under the shifts {c,c} f {e,e} and {e,e} f {c,c}.
This means that the reference model (5) and the
estimators (16) and (17) have the same effects on the
PID performance.
(d) Although the PID representation (20) and the
control law (16)-(19) are input/output equivalent to
each other, probably the key advantage of the proposed
PID control configuration (16)-(19) lies in the fact that
the controller states are meaningful variables as estimates of the physical plant states and the model/plant
mistmaches. It follows that the estimates zj and
j can
be used to monitor the performance of the process or
detect failures of actuators and sensors.
(e) Dead time in the input channel imposes serious
limitations in the achievable closed-loop performance.23
Following some internal model control ideas (see Morari
and Zafiriou23) and given an upper bound for the dead
time, the estimation and closed-loop time constants
should be set at values not smaller than the dead-time
upper bound. This tuning guideline will be used in
numerical simulations below.
A. Numerical Simulations with the Practical
Control Law (16)-(19). We have carried out several
numerical simulations to illustrate the performance of
the proposed robust control design. From nominal
design parameters and the vapor-liquid equilibrium
relationship, we have chosen
j ) 18.28 (see eq 11). The
closed-loop damping factor and time constant, c and
c, have been chosen as in the simulation above. The
estimation damping factor has been chosen as e ) 1.
Figure 4 shows the time evolution of the reflux ratio
and product purity for wref ) 0.7 and two different
values of the estimation time constant e. For comparison, the ideal behavior under the theoretical feedback
control (9) is also shown. It is noted that the smaller
the estimation time constant, the closer the behavior
to the ideal one. This behavior is also observed for the

Figure 6. Dynamics of the controlled RBD column for e ) 0.025


h and three different values of the input dead time.

case wref ) 0.8 (see Figure 5). This implies that, in


principle, the ideal behavior under the theoretical
feedback control (9) could be achieved as e f 0. Of
course, this is not possible in the presence of dead times.
Figure 6 shows the dynamics of the controlled RBD
column for c ) e ) 0.025 h and three different values
of the dead time. It is noted in this case that perfect
regulation of the product purity is not achieved. After
control input saturation, product purity is maintained
below the required purity. This behavior is induced by
the delayed information used by the control input, which
induces smaller reflux ratios. Such a regulation offset
cannot be completely removed despite the presence of
the integral action in the output feedback control law
(see eqs 21 and 22) because the nonregulated RBD
dynamics (i.e., the internal dynamics) are nonvanishing.
That is, because these uncontrolled dynamics correspond to batch processes, they induce a time-varying
behavior in the modeling error function . Anyway, the
maximum deviation from the reference is about 3%.
The numerical simulations above have been carried
out for the feed composition3 0.45/0.45/0.0/0.1. However,
within a batch distillation process, the feed composition
can change from one batch to another. If the control of
product purity is based on optimization techniques3 and
implemented in the open loop, the reflux ratio policy
should be computed for every batch to be processed. This
is not the case when feedback control is used. Assume
that the feed composition is 0.2/0.3/0.0/0.5. Figure 7
shows the dynamics of the controlled RBD column using
the same parameters as in Figure 6. Because the feed
of acetic acid and ethanol is not stoichiometric, the
production of ethyl acetate is less than that in the

Ind. Eng. Chem. Res., Vol. 39, No. 2, 2000 385

1. The vapor-phase holdup is assumed to be negligible


compared to the liquid-phase holdup on each phase.
2. Chemical reactions in the vapor phase are neglected.
3. The initial state of the column is the steady-state
total reflux condition with no reactions.
4. The liquid volumetric holdups on the plates will
be assumed to be constant. Thus, the model is directed
at simulating the dynamics of the main production
period during which the hydrodynamic conditions are
not widely varying.
5. The pressure drops and the plate efficiencies are
constant during the operation.
6. The control of levels in condenser and reboiler
drums is perfect.
Consider the following notation:

Figure 7. Dynamics of the controlled RBD column for e ) 0.025


h and three different values of the input deadtime. The feed
composition was 0.2/0.3/0.0/0.5.

former case. This is represented by the fact that the


reaction phase is shorter than that in the stoichiometric
feed case.
5. Conclusions
The regulation of product purity in reactive batch
distillation columns has been studied in this work. It
has been found that a reflux ratio policy computed from
a feedback control function resembles the reflux ratio
policy obtained from a nonlinear optimization problem.
The proposed control policy should be implemented in
a closed-loop fashion to avoid loss of performance due
to strong uncertainties in the process model. To this end,
a robust control design has been proposed, which for
implementation only requires measurement or estimation of the product purity. It has been shown that the
resulting control law is equivalent to a classical PID
controller with an ARW scheme. Several numerical
simulations have been presented to illustrate the performance of the controlled column under strong uncertainties in the process dynamics and important dead
times due to measurement and estimation.
Appendix. Theoretical Model
In this appendix we consider the mass- and energybalance equations for a batch rectification column shown
in Figure 1. The set of mass-balance differential equations consists of the total mass balance and mass
balance on each component.15 The energy-balance differential equation merely consists of the application of
the first law of thermodynamics. The model is valid
under the following assumptions:

D ) overhead product flow rate


Hl ) molar enthalpy of the liquid phase
Hv ) molar enthalpy of the vapor phase
L ) liquid flow rate
M ) molar mass
n ) number of plates
Q ) rate of heat generation by chemical reaction activity
R ) vector of reaction rates
T ) thermodynamic temperature
t ) time
V ) vapor flow rate
X ) vector of mole fractions (liquid)
Y ) vector of mole fraction (vapor)
F ) density
Also, consider the following subscripts:
D ) condenser
1 ) plate 1 (top)
j ) plate j
n ) plate n (bottom)
B ) reboiler
i ) component i

The following differential equations result.

Condenser
dhD
) V1 - L0 - D + hD
dt

RD,i(XD,TD)

(A.1)

d(hDXD)
) V1Y1 - L0XD - DXD + hDRD(XD,TD) (A.2)
dt
d(hDHl,D)
) V1Hv,1 - (L0 + D)Hv,D + QD(XD,TD)
dt
(A.3)
Plate j, 1 e j e n
dhj
) Vj+1 + Lj-1 - Vj - Lj + hj
dt

Rj,i(Xj,Tj)

(A.4)

d(hjXj)
) Vj+1Yj+1 + Lj-1Xj-1 - VjXj - LjYj +
dt
hjRj(Xj,Tj) (A.5)
d(hjHl,j)
) Vj+1Hv,j+1 + Lj-1Hl,j-1 - VjHv,j - LjHl,j +
dt
Qj(Xj,Tj) (A.6)

386

Ind. Eng. Chem. Res., Vol. 39, No. 2, 2000

Reboiler
dhB
) Ln - VB + hB
dt

RB,i(XB,TB)

d(hBXB)
) LnXn - VBYB + hBRB
dt

(A.7)

(2) Mujtaba, I. M.; Macchietto, S. Simultaneous optimization


of design and operation of multicomponent batch distillation
column-single and multiple separation duties. J. Process Control
1996, 6, 27.

(A.8)

(3) Mujtaba, I. M.; Macchietto, S. Efficient optimization of batch


distillation with chemical reaction using polynomial curve fitting
techniques. Ind. Eng. Chem. Res. 1997, 36, 2287.

d(hBHl,B)
) LnHl,n - VBHv,B + QB(XB,TB) (A.9)
dt
To obtain a fully determined system, the variables
appearing in these balance equations must also satisfy
the following equations.
Constraint in the Volume. The volume of the liquid
phase in the condenser and on each plate is assumed
to be constant. Consequently, the corresponding molar
holdups are functions only of the temperature, the
pressure, and the compositions:

hj ) jFj/Mj

(A.10)

(4) Sorensen, E.; Skogestad, S. Control strategies for reactive


batch distillation. J. Process Control 1994, 4, 205.
(5) Henson, M. A.; Seborg, D. E. Nonlinear control strategies
for continuous fermenters. Chem. Eng. Sci. 1992, 47, 821.
(6) Schork, F. J.; Deshpande, P. B.; Leffew, K. W. Control of
Polymerization Reactors; Marcel Dekker: New York, 1993.
(7) Soroush, M.; Kravaris, C. Nonlinear control of a polymerization CSTR with singular characteristic matrix. AIChE J. 1994,
40, 6.
(8) Henson, M. A.; Seborg, D. E. Adaptive nonlinear control of
a pH neutralization process. IEEE Trans. Control Syst. Tech. 1994,
2, 169.

Constraint due to Vapor-Liquid Equilibrium

(9) Kravaris, C.; Daoutidis, P. Nonlinear state feedback control


of second-order nonminimum-phase nonlinear systems. Comput.
Chem. Eng. 1990, 49, 439.

Yj ) KjXj

(10) Wright, R. A.; Kravaris, C. Nonminimum-phase compensation for nonlinear processes. AIChE J. 1992, 38, 26.

(A.11)

with
nc

yj,i ) 1

i)1

(A12)

YB ) KBXB

(A.13)

for the reboiler,

with
nc

(11) Kravaris, C.; Daoutidis, P.; Wright, M. A. Output feedback


control of nonminimum-phase nonlinear processes. Chem. Eng.
Sci. 1994, 49, 2107.
(12) Limquenco, L. C.; Kantor, J. C. Nonlinear output feedback
control of an exhotermic reactor. Comput. Chem. Eng. 1990, 14,
427.
(13) Wu, W.; Chou, Y. S. Robust output regulation for nonlinear
chemical processes with unmesurable disturbances. AIChE J.
1995, 41, 2565.
(14) Soroush, M.; Kravaris, C. Nonlinear control of a batch
polymerization reactor: an experimental study. AIChE J. 1992,
38, 1429.

yB,i ) 1

i)1

(A.14)

YD ) KBXB

(A.15)

(16) Bosely, J. R., Jr.; Edgar, T. F. Appropriate modeling


assumptions for batch distillation optimization and control. In
Proceedings of 5th International Seminar on Process Systems
Engineering, Kyongju, Korea, May 30-June 3, 1994; Vol. 1, p 477.

(A.16)

(17) Barolo, M.; Berto, F. Composition control in batch distillation: binary and multicomponent mixtures. Ind. Eng. Chem. Res.
1998, 37, 4689-4698.

Constraint in the Enthalpy. If the temperature and


the composition of the liquid phase are known, then the
molar enthalpy of this liquid phase is determined. At
each time, the molar enthalpy calculated according to
the value of the composition (given by the integration
of the mass balance equations) and the temperature
(given by the liquid-vapor equilibrium constraint equations) must be equal to the molar enthalpy calculated
by the integration of the energy-mass balance equations:

(18) Bartusiak, R. D.; Georgakis, C.; Reilly, M. J. Nonlinear


feedforward/feedback control structures designed by reference
systems synthesis. Chem. Eng. Sci. 1989, 44, 1837-1851.

(15) Cuille, P. E.; Reklaitis, G. V. Dynamic simulation of


multicomponent batch rectification with chemical reaction. Comput. Chem. Eng. 1986, 10, 389.

for the condenser,

with
nc

yD,i ) 1

i)1

Hl,j ) Hl(Xj,Tj,Pj)

(A.17)

Total enthalpies are used in this formulation; thus,


no heat of reaction terms are required in eqs A.3, A.6,
and A.9.
Literature Cited
(1) Albet, J.; Le Lann, J. M.; Julia, X.; Koehret, B. Rigorous
simulation of multicomponent multisequence batch reactive distillation. Proceedings of Computed-Oriented Process Engineering;
Elsevier Science Publishers B. V.: Amsterdam, The Netherlands,
1991; p 75.

(19) McLellan, P. J.; Harris, T. J.; Bacon, D. W. Error trajectory


descriptions of nonlinear controller designs. Chem. Eng. Sci. 1990,
45, 3017-3034.
(20) Lee, P. L.; Sullivan, G. R. Generic Model Control (GMC).
Comput. Chem. Eng. 1988, 12, 573-580.
(21) Quintero-Marmol, E.; Luyben, W. L. Inferential modelbased control of multicomponent batch distillation. Chem. Eng.
Sci. 1992, 47, 887.
(22) Kothare, M. V.; Campo, P. J.; Morari, M.; Nett, N. N. A
unified framework for the study of anti-windup designs. Automatica 1994, 30, 1869.
(23) Morari, E.; Zafiriou, E. Robust Process Control; PrenticeHall: New York, 1989.

Received for review June 1, 1999


Revised manuscript received October 11, 1999
Accepted October 21, 1999
IE990382L

You might also like