Ind. Eng. Chem. Res.

2010, 49, 91959208 9195

Time-Optimal Control of Dividing-Wall Distillation Columns

Alexandru Woinaroschy* and Raluca Isopescu
Department of Chemical Engineering, Politehnica UniVersity of Bucharest, 1-5 Polizu Street, Bucharest, Romania

Time-optimal control of startup traditional distillation columns by iterative programming proposed by

Woinaroschy for ideal [Ind. Eng. Chem. Res. 2008, 47, 4158] and nonideal mixtures [Ind. Eng. Chem. Res.
2009, 48, 3873] is extended to the case of dividing-wall distillation columns. The minimization of distillation
startup time is performed by iterative dynamic programming employing randomly chosen candidates for
admissible control. The control variables are the reflux ratio, the reboiler heat duty, and the side-draw flow
rate. The dynamic distillation model proposed by the author in the previous papers is applied. Two illustrative
case studies for the separation in a dividing-wall column with sieve trays and lateral downcomers are presented
as follows: the separation of an ideal benzene-toluene-ethylbenzene ternary mixture and the separation of
a nonideal methanol-ethanol-1-propanol mixture. In another case study, a conventional two-column system
is presented in comparison to the dividing-wall column. As in the cases of traditional distillation columns,
the startup time decrease and the corresponding reboiler energy savings are significant for each of the control
variables.

Introduction
Distillation is the most commonly used separation process
in chemical and petrochemical plants. Its main disadvantage is
high energy consumption, and this is the reason why thermally
coupled distillation systems are studied to provide challenging
solutions for industrial implementations. The dividing-wall
distillation column (DWC), which represents the Petlyuk column
built in a single shell, is a very promising alternative for both
energy and cost savings. Several theoretical studies have
reported important energy savings in thermally coupled columns,
including the DWC, as compared to classical separation
schemes.1-5 Some practical applications of this technology are
also reported, underlying the thermal efficiency and cost
reduction.6-8 The most important parameters that are mentioned
to influence the energy consumption in a DWC are the feed
composition and pressure.8 The DWC led to about 40% energy
savings, and the efficiency of DWC increases when the middle
component is in a large amount in the feed.9 A case study for
a common hydrocarbon mixture separation in oil refiners
demonstrates that the heat transfer units reduction and the use
of a single column shell provide a decrease up to 23% of the
capital cost as compared to a classical two-column separation
sequence.10
Despite the advantages of the DWC, the industry is still
reluctant to introduce it on a large scale due to not yet well-
established design procedures and fear of control issues.5 The
design of a full thermally coupled column can not follow the
conventional multicomponent design procedures when informa-
tion about interlinking streams is unknown.11 A design proce-
dure based on a three-column model was developed by
Triantafyllou and Smith.12 In their model, the final structure
has to be iteratively adjusted, as the number of trays on both
sides of the dividing wall must be the same. This problem also
arises when a commercial simulator is used to establish the final
column topology. A structural design methodology for full
thermally coupled distillation columns is developed by Kim,11
based on tray to tray calculations, assuming very large reflux
ratios in the main column and ideal trays. From an operational
* To whom correspondence should be addressed. Tel: +40-21-
4023902. Fax: +40-21-3185900. E-mail: A_WOINAROSCHY@
chim.upb.ro.
10.1021/ie100090p 2010 American Chemical Society
Published on Web 08/25/2010
point of view, the thermally coupled columns seem to raise more
problems than simple distillation columns. The operational
complexity of a DWC derives from the increased number of
freedom degrees as compared to a simple column with a side
draw. Halvorsen and Skogestad13 found that the liquid and vapor
split ratios, considered as two additional freedom degrees, are
important for the thermal efficiency of the DWC. An optimal
solution can be established only for certain correlations of the
liquid and vapor split. Moreover, Wang and Wong14 demon-
strated that for an infinite number of stages column, the energy
efficiency depends on the liquid and vapor split values, with a
trade-off between energy efficiency and controllability of end
products composition.
Dividing-wall columns still have rare applications, due to
some uncertainties when dealing with the dynamics of the
process. In general, the heat integration brings supplementary
difficulties when column dynamics and start up policies are
implied. High values of startup time have been found to
characterize a heat-integrated two-pressure column system,
implying higher costs.15
The dynamic simulation of a distillation column has proved
to be an efficient theoretical tool in providing startup policies.16,17
The comparison with experimental studies concerning startup
and transient regimes is also of crucial importance when
operating conditions change and a new steady state must be
reached.17,18
Commercial simulators, such as HYSYS or ASPEN, provide
reliable environments for steady state and dynamic simulation
of distillation columns, including conventional PID controllers.
A special structure as the DWC can also be implemented using
thermodynamic equivalent schemes. For such a scheme, con-
centration control loops19 and temperature control loops20 were
defined to handle feed composition or feed flow variations.
When special control or optimization procedures are applied
for the analysis of heat-integrated structures, the simulators are
used to provide steady state solutions in the frame of an
optimization loop21 or stand-alone simulation programs are built,
using a convenient mathematical model for the column dynam-
ics, efficient in terms of reliability and computation time for
the optimization method or controller type adopted.22,23
9196 Ind. Eng. Chem. Res., Vol. 49, No. 19, 2010

In the present work, the DWC topology was designed in the
frame of HYSYS simulator, a dynamic model was elaborated
based on this structure, and the optimal startup time was defined,
using iterative dynamic programming as the optimization
procedure.
Dynamic Simulation Model
The dynamic distillation models (DDMs) presented in
literature are algebraic-differential models. The dynamic of
DWC was simulated with a rigorous DDM proposed by
Woinaroschy.24,25 The advantage and originality of this model
Therefore, knowing the liquid composition and temperature on
each tray, the vapor composition is obtained by straightforward
computation, without iterative calculations.
The model core consists of n equations (eq 1), n(m - 1)
equations (eq 2), and n equations (eq 3), therefore, a total
number of n(m + 1) differential equations. The rest of the
variables involved in these equations are explicitly calculated
as follows:
The vapor flow rate is obtained from the total energy balance:
Vj )
1
[
L (h - hj) + Vj+1(Hj+1 - hj) ( FL,j(hF,j - hj) (
Hj j-1 j-1

]
consist in the fact that the iterative algebraic equations are dhj dTj
avoided, the model being a good compromise between the FV,j(HF,j - hj) - qj - Nj (4)
dTj dt
complexity degree and the correctness.
The following simplification assumptions are present in this The liquid flow rate is obtained on the base of Francis
model: correlation for a plate weir, leading to

( )
(i) The molar vapor holdup is negligible as compared to
m 1.5
the molar liquid holdup.
(ii) Interphase heat transfer is considered much more intense
1.84lwFL,j
Nj Mx i i,j
i)1
than interphase mass transport; consequently, the liquid Lj ) - zW (5)
and vapor leaving each plate are in thermal equilibrium
m L,jApFL,j
at the boiling temperature, corresponding to the liquid Mx i i,j
i)1
composition.
(iii) The liquid and vapor are perfectly mixed on each plate, To avoid iterative calculations, the vapor flow rates and
and Murphree plate efficiency can be used for real trays compositions are computed in the j ) n, n - 1, ..., 1 order,
applications. while the liquid flow rates are calculated in the j ) 1, 2, ..., n
(iv) Entrainment and weeping rates, plate flooding, down- order.
comer holdup, and delay time between plates are The total pressure variation during one time integration step
neglected. is much smaller than the composition and temperature variations.
Thus for a given tray, j, when neglecting the molar vapor To simplify the procedure, the pressure is considered constant
holdup, mass-balance equations are defined as: along the time integration step; it is recomputed at the beginning
- total mass balance around plate j of each new time step
dNj pj ) pj+1 - pj (6)
) Lj-1 + Vj+1 - Lj - Vj ( FL,j ( FV,j (1)
dt
where the pressure drop is calculated on the base of hydraulic
- mass balance around plate j for component i correlations, specific for the plate type. For the sieve tray, this
is

( )
dxi,j 1
) [Lj-1(xi,j-1 - xi,j) + Vj+1(yi,j+1 - xi,j) ( m 2
dt Nj
FL,j(xF,i,j - xi,j) ( FV,j(yF,i,j - xi,j)] (2)
QG,j yi,j Mi
j
0.00683 i)1
pj ) - - 0.02175 -
FG,j Ao,j do
The model original feature (due to the fact that iterative 0.07352FL,jzst,j (7)
solving of nonlinear algebraic equations is avoided; for details,
see Woinaroschy24,25) consists in the temperature calculation where

( )
on each plate with the equation
m 0.667

x
( )
m
i,jPi,j xi,j di,j dxi,j Lj i,j Mi

dTj
pj
1+
i,j dxi,j dt zst,j ) 0.65zW + 0.8
i)1
(8)
)-
i)1
(3.1) lW,jFL,j
dt m
xi,ji,j dPi,j
pj dTj
The state variables are Ni, xi,j, and Tj for i ) 1, ..., m and j )
1, ..., n. The control variables considered in the case studies
i)1
are the reflux ratio, the reboiler heat duty, and the side-draw
This equation is applied for ideal and nonideal mixtures in flow rate. These control variables are implied in DDM as
the liquid phase. For nonideal mixtures, in liquid and/or vapor follows:
phase, the temperature variation can be computed with the (i) The reflux ratio is involved in eq 1, which describes the
equation.26 total mass balance around plate 1 in eq 2, expressing the mass
balance around plate 1 for each component i, and in eq 4, when
m calculating the vapor flow rate over plate 1.
dxi,j
dTj
i,j
dt If the total condenser type is considered, the material balance
i)1 over tray 1, the condenser, and condensate splitter gives
)- (3.2)
dt m
di,j
dTj L0 )
R
V (9)
i)1 R+1 1

Therefore for j ) 1, eqs 1, 2, and 4 become
dN1 V1
) V 2 - L1 - (1.1)
dt R+1
dxi,1
1 R
dt
) [ V (x - xi,1) + V2(yi,2 - xi,1)
N1 R + 1 1 i,D ] operating regime procedure begins at the end of step 7. This is
(2.1)
dh1 dT
V2(H2 - h1) - N1
dT dt
V1 ) (4.1)
R
(h - hD) + H1 - h1
R+1 1
The value of the liquid flow rate L1 (eq 5 for j ) 1) is established
according to the dependence of the liquid holdup per plate 1
versus the reflux ratio.
(ii) Reboiler heat duty is involved in eq 4, giving the vapor
flow rate for the bottom (j ) n), where qn ) -qB.
(iii) The side-draw flow rate is -FL,j in eqs 1, 2, and 4 for
the side-draw tray j.
Lower and upper bounds are imposed for each control
variable. The equilibrium, thermodynamic data and physical
properties correlations are selected according to the mixture
nature. Details and model validation for traditional distillation
columns are presented by Woinaroschy.25
Startup Simulation
The startup of distillation columns is a very challenging
control and simulation problem, due to theoretical and practical
aspects. In fact, the startup is the most dynamic operation
stage, involving important economic and safety considerations.
Several researchers have studied this subject: Fieg and Wozny,17
Woinaroschy,24-26 Yasuoka et al.,27 Ruiz et al.,28 Barolo et
al.,29,30 and Han and Park.31 None of these papers consider the
case of thermally coupled distillation columns. Recently, the
minimization of startup time in a heat and mass integrated two
columns system was analyzed by Verbanov et al.,32 as a
generalization of the two heat integrated-two pressure column
system.15 An investigation on optimal startup control of DWC
was briefly reported by Woinaroschy and Isopescu.16
A general sequence of actions, forming the basis for different
startup procedures for sieve plate columns, was formulated by
Ruiz et al.28
Step 0: The column is empty, and the liquid feed is intro-
duced (only liquid feed is considered).
Step 1: The liquid starts to weep to the plate below through
the plate holes rather than through the downcomer. The
liquid reaches the bottom of the column (or reboiler),
thus increasing the liquid level there.
Step 2: Heat is introduced in the reboiler, and vapors start
to go up.
Step 3: The condenser starts to operate as the vapor phase
reaches the top plate, and the reflux drum starts to
fill up.
Step 4: The reflux is introduced into the column, and
operation at total reflux begins.
Step 5: The vapor flow through the plate holes seals the
plates in terms of liquid weeping through the plate
holes. This starts to increase the liquid holdup on
the plates.
Step 6: All plates have enough liquid holdup so that the
liquid can start to fall down the downcomers. The
Ind. Eng. Chem. Res., Vol. 49, No. 19, 2010 9197
downcomers are sealed, and vapors cannot pass
upward through them.
Step 7: The column operation is changed from the total
reflux. Distillate, bottom product, and the possible
side streams are taken out.
In the frame of the present paper, the effective start-up transient
the objective of the simulation. The liquid composition is
considered the same on all plates at the beginning of simulation,
being equal to the feed composition. Traditionally, the column
is operated at constant values of control parameters (feed rate,
feed thermal state, reflux ratio, reboiler heat duty, bottom
pressure, etc.). These values are those corresponding to the final
desired stationary regime. All state variables values (concentra-
tions, temperatures, pressures on plates, and in output material
fluxes) vary in time. The fluctuations amplitude decreases in
time and will vanish at steady state regime. The products
compositions are, of course, different from the desired values
up to the stationary regime, the products being collected, mixed,
and recycled to the feed. Apart from the chemical reactors case,
the momentum, energy, and mass transfer processes, which are
not accomplished by chemical reactions, do not have multiple
or nonstable stationary states.
Time-Optimal Control (TOC)
The TOC problem is to minimize the final time, tf, and to
determine the corresponding control variables, ui(k), i ) 1,
2, ..., nu; k ) 1, 2, ..., S, and length of time stages V(k), k ) 1,
2, ..., S restricted to
si(tf) ) ssi i ) 1, 2, ..., ns (10)
where the desired final stationary state values, ssi, are computed
from the differential equations of the continuous dynamic system
dsi
) V(k)Sfi(s1, s2, ..., sns, u1, u2, ..., unu) i ) 1, 2, ..., ns;
d
k ) 1, 2, ..., S (11)
Practically, eq 11 corresponds to eqs 1-3, after introducing the
normalized time variable, , so that ) t/tf, takes discrete values
k ) k/S, k ) 0, 1, ..., S. Each stage is of equal length 1/S in the
transformed time domain.
To minimize the final time, tf, subject to the constraint given
in eq 10, the penalty performance index was formulated as
| |
ns
si(tf)
1-
i)1
ssi
I ) tf + (12)
ns
where > 0 is a penalty coefficient.
The application of TOC is the same as in the previous
papers,25,26 using the algorithm proposed by Bojkov and Luus,33
based on the IDP procedure given by Bojkov and Luus,34,35
which employs randomly chosen candidates for the admissible
control.36 The system of differential equations was numerically
integrated by a fourth order Runge-Kutta-Gill method, which
automatically chooses the time step to satisfy local error
tolerance. It can be appreciated that the corresponding computer
time increase due to the small values of the integration step is
justified, thus avoiding the iterative solution of algebraic
equations, which are even more computer time consumers.25,26
The computer programs were coded in FORTRAN.
Case Study 1: A Ternary Ideal Mixture
Configuration of DWC. The ternary ideal mixture considered
is a common one in oil refineries: benzene, toluene, and
9198 Ind. Eng. Chem. Res., Vol. 49, No. 19, 2010

Figure 1. DWC topology: Simulation flowsheet in HYSYSsCase Figure 2. DWC topology: Designed structure of the DWCsCase Study 1.
Study 1.

ethylbenzene. The feed compositions, in mole fractions, were Table 1. General Data for Case Study 1
0.6/0.3/0.1. An estimation of the number of trays in each section parameters and operating conditions value
of the DWC, feed tray, and side-draw location as well as the feed flow rate (kmol/h) 126
position of interlinking vapor and liquid streams were obtained feed composition (mole fractions) benzene, 0.6
using the short-cut design option available in HYSYS, which toluene, 0.3
applies Fenske-Underwood-Gilliland-Kirkbride equations. ethylbenzene, 0.1
side-draw flow ratea (kmol/h) 36
This structure is largely used in the design step, as generally
thermal state of the feed stream liquid at boiling
mentioned in the literature.12 temperature
A four-column equivalent structure was derived from this thermal state of the side-draw stream liquid at boiling
initial design (Figure 1). This structure is more appropriate for temperature
condenser type total
dynamic simulations due to its capabilities of reflecting the pressure in the top of column and 760
internal streams and geometrical characteristics of the tray in condenser (mm Hg)
sections, to be used in the dynamic simulation and in the TOC reflux ratioa 2
procedure. reboiler type total
reboiler heat dutya (kW) 2095
The final structure, corresponding to the DWC, consists of a bottom liquid volume (m3) 0.911
top tray section placed above the dividing wall, the trays in the tray area for top and bottom 2.8143
prefractionator, the trays in the side-draw region, and the trays sections (m2)
below the dividing wall. The number of trays in each section, tray area for prefractionator and 1.4071
the feed tray, side-draw location, the position of thermal side-draw (m2)
hole area per plate for top and 0.2815
coupling streams, and the reflux ratio as identified by short-cut bottom sections (m2)
design were slightly modified by repeated rigorous simulations hole area per plate for prefractionator 0.1407
aiming to fulfill the purity requirements (more than 0.95 mol and side draw (m2)
fraction) at a low reboiler duty. This final design was considered weir height (m) 0.025
hole diameter (m) 0.005
a good DWC design, but this is not an optimal one in terms of tray number of top section 11
minimum energy consumption. The Antoine model was used tray number of prefractionator 14
to calculate the thermodynamical properties of the system. tray number of side draw 25
Figure 1 presents the final topology, as resulting after rigorous tray number of bottom section 13
simulation in the frame of HYSYS. feed tray position 12
side-draw tray position 30
As shown in Figure 2, the number of trays on either part of liquid split fraction in prefractionator 0.6
the DWC is not the same, unlike the generally used approach, a
with an equal number of trays in both sides of the wall.12,20 Prescribed values for the desired stationary regime.
From the constructive point of view, an equal trays number between the reflux flow rate in the prefractionator and the total
would be preferable. In our analyses, the dividing wall is meant reflux flow rate in the column, was adjusted during the dynamic
to ensure the same cross-section in the prefractionator, as in simulation step, to provide the products purities. The vapor split
the side-draw region. The liquid split, defined as the ratio is free to adjust by imposing the same pressure drop on both
 Ind. Eng. Chem. Res., Vol. 49, No. 19, 2010 9199
Table 2. Initial State and Desired State for the Case Study 1
initial state desired state
tray j x1,j x2,j x3,j Tj pj x1,j x2,j x3,j Tj pj
1 0.6 0.2 0.1 94.87 867.3 0.9190 0.0810 0.0000 86.05 757.3
2 0.6 0.2 0.1 94.94 869.1 0.8562 0.1438 0.0000 87.52 761.1
3 0.6 0.2 0.1 95.01 871.0 0.7708 0.2291 0.0000 89.67 765.0
4 0.6 0.2 0.1 95.09 872.8 0.6729 0.3270 0.0001 92.35 768.8
5 0.6 0.2 0.1 95.16 874.7 0.5803 0.4195 0.0002 95.09 772.7
6 0.6 0.2 0.1 95.24 876.5 0.5045 0.4949 0.0006 97.51 776.5
7 0.6 0.2 0.1 95.31 878.4 0.4493 0.5494 0.0013 99.39 780.3
8 0.6 0.2 0.1 95.38 880.2 0.4122 0.5850 0.0028 100.72 784.1
9 0.6 0.2 0.1 95.46 882.1 0.3881 0.6058 0.0061 101.64 787.9
10 0.6 0.2 0.1 95.53 883.9 0.3724 0.6147 0.0129 102.34 791.7
11 0.6 0.2 0.1 95.60 885.7 0.3609 0.6124 0.0267 102.99 795.5
12 0.6 0.2 0.1 95.81 890.8 0.5071 0.4159 0.0770 98.59 798.3
13 0.6 0.2 0.1 95.89 893.0 0.4705 0.4508 0.0787 99.94 804.2
14 0.6 0.2 0.1 95.97 895.2 0.4173 0.5017 0.0810 101.95 810.1
15 0.6 0.2 0.1 96.06 897.4 0.3481 0.5680 0.0839 104.68 816.0
16 0.6 0.2 0.1 96.15 899.6 0.2697 0.6433 0.0871 107.97 822.0
17 0.6 0.2 0.1 96.24 901.8 0.1935 0.7165 0.0900 111.40 828.0
18 0.6 0.2 0.1 96.32 904.0 0.1296 0.7781 0.0924 114.46 834.0
19 0.6 0.2 0.1 96.41 906.2 0.0821 0.8238 0.0941 116.85 840.0
20 0.6 0.2 0.1 96.49 908.4 0.0499 0.8546 0.0955 118.51 846.1
21 0.6 0.2 0.1 96.57 910.6 0.0294 0.8736 0.0970 119.57 852.3
22 0.6 0.2 0.1 96.66 912.8 0.0165 0.8857 0.0978 120.20 858.4
23 0.6 0.2 0.1 96.74 915.0 0.0090 0.8931 0.0980 120.51 864.6
24 0.6 0.2 0.1 96.83 917.1 0.0046 0.8975 0.0979 120.64 870.8
25 0.6 0.2 0.1 96.91 919.3 0.0022 0.9000 0.0978 120.62 877.0
26 0.6 0.2 0.1 95.49 882.8 0.1794 0.7907 0.0298 110.44 800.7
27 0.6 0.2 0.1 95.55 884.4 0.0803 0.8868 0.0329 115.34 804.5
28 0.6 0.2 0.1 95.61 885.9 0.0319 0.9334 0.0347 117.94 808.4
29 0.6 0.2 0.1 95.67 887.5 0.0118 0.9528 0.0354 119.02 812.2
30 0.6 0.2 0.1 95.73 889.1 0.0042 0.9600 0.0358 119.40 816.1
31 0.6 0.2 0.1 95.79 890.6 0.0015 0.9621 0.0364 119.55 819.3
32 0.6 0.2 0.1 95.86 892.2 0.0009 0.9620 0.0371 119.59 822.5
33 0.6 0.2 0.1 95.92 893.7 0.0008 0.9613 0.0379 119.60 825.8
34 0.6 0.2 0.1 95.98 895.3 0.0007 0.9604 0.0389 119.59 829.0
35 0.6 0.2 0.1 96.04 896.8 0.0007 0.9592 0.0400 119.58 832.2
36 0.6 0.2 0.1 96.10 898.4 0.0007 0.9578 0.0414 119.58 835.4
37 0.6 0.2 0.1 96.16 899.9 0.0008 0.9563 0.0430 119.57 838.6
38 0.6 0.2 0.1 96.22 901.5 0.0008 0.9545 0.0448 119.58 841.8
39 0.6 0.2 0.1 96.28 903.0 0.0008 0.9524 0.0468 119.57 845.0
40 0.6 0.2 0.1 96.34 904.6 0.0008 0.9501 0.0491 119.59 848.2
41 0.6 0.2 0.1 96.40 906.1 0.0008 0.9475 0.0517 119.60 851.4
42 0.6 0.2 0.1 96.46 907.7 0.0008 0.9446 0.0546 119.64 854.6
43 0.6 0.2 0.1 96.52 909.2 0.0008 0.9414 0.0578 119.70 857.8
44 0.6 0.2 0.1 96.58 910.7 0.0008 0.9379 0.0613 119.75 860.9
45 0.6 0.2 0.1 96.64 912.3 0.0008 0.9340 0.0652 119.83 864.1
46 0.6 0.2 0.1 96.70 913.8 0.0008 0.9297 0.0695 119.91 867.3
47 0.6 0.2 0.1 96.76 915.4 0.0008 0.9250 0.0742 120.01 870.5
48 0.6 0.2 0.1 96.82 916.9 0.0008 0.9198 0.0794 120.11 873.7
49 0.6 0.2 0.1 96.88 918.4 0.0008 0.9141 0.0851 120.24 876.8
50 0.6 0.2 0.1 96.94 920.0 0.0008 0.9079 0.0913 120.37 880.0
51 0.6 0.2 0.1 97.00 921.5 0.0008 0.9011 0.0981 120.52 883.2
52 0.6 0.2 0.1 97.07 923.3 0.0004 0.8975 0.1022 120.50 887.9
53 0.6 0.2 0.1 97.14 925.1 0.0002 0.8894 0.1105 120.62 892.7
54 0.6 0.2 0.1 97.21 927.0 0.0001 0.8740 0.1260 120.88 897.4
55 0.6 0.2 0.1 97.27 928.8 0.0000 0.8458 0.1542 121.45 902.1
56 0.6 0.2 0.1 97.34 930.6 0.0000 0.7968 0.2032 122.47 906.8
57 0.6 0.2 0.1 97.41 932.4 0.0000 0.7177 0.2823 124.19 911.6
58 0.6 0.2 0.1 97.48 934.2 0.0000 0.6046 0.3954 126.82 916.3
59 0.6 0.2 0.1 97.55 936.0 0.0000 0.4669 0.5331 130.27 921.0
60 0.6 0.2 0.1 97.62 937.8 0.0000 0.3281 0.6719 134.06 925.7
61 0.6 0.2 0.1 97.69 939.6 0.0000 0.2115 0.7885 137.54 930.4
62 0.6 0.2 0.1 97.76 941.4 0.0000 0.1274 0.8726 140.24 935.2
63 0.6 0.2 0.1 97.82 943.2 0.0000 0.0730 0.9270 141.95 940.1
B 0.6 0.2 0.1 97.89 945.0 0.0000 0.0402 0.9598 143.00 945.0

sides of the dividing wall, which implied a different number of
trays. In practice, this issue of nonequal number of trays can
be overcome by using packing elements of different heights.37
Apart from the main design and operating parameters obtained
from the HYSYS simulation, several geometrical and mechan-
ical characteristics are also considered. Table 1 presents the
detailed data used in the dynamic analysis.
The vapor and liquid flow rate values calculated with DDM
at the stationary regime are practically the same as the HYSYS
values. This comparison was necessary to validate the solution
obtained with the proposed mathematical model. As the HYSYS
simulation was performed using an equivalent scheme without
including heat transfer through the dividing wall, the same
hypothesis was adopted in DDM, where no heat transfer between
the prefractionator and the side draw was considered. The heat
rate losses per plate, qj, were also neglected.
Startup Simulation of DWC. The dynamic model was
applied for DWC startup simulation. The stationary state was
considered, as in all case studies presented in this paper, when
the average absolute value of the derivatives is less than 10-6.
During the startup, all control variables were maintained constant
at the values indicated in Table 1. The initial state and the
desired state are indicated in Table 2. The final steady state
values for product concentrations are very close to those that
resulted from HYSYS simulation (Table 3). The evolutions
during the startup transient regime of benzene concentration in
the top product, toluene in the side product, and ethylbenzene
in the bottom product are presented in Figure 3. As shown, the
9200 Ind. Eng. Chem. Res., Vol. 49, No. 19, 2010

Figure 3. Case Study 1: Evolutions during the traditional startup transient

regime of benzene concentration in top product (- - -), toluene in side Figure 4. Case Study 1: The best reflux control for 5 stages ( ) and 10
product (s), and ethylbenzene in bottom product (- -). stages (s).

Table 3. Mole Fractions of Product Components for Case Study 1 The TOC algorithm was applied for 5 and 10 time stages.
benzene toluene ethylbenzene The grid point numbers were ns ) 5, nu ) 9, and nV ) 9. The
region contraction factor was set to 0.8, and the total number
top product
of IDP iterations was 10. The penalty function values and the
HYSYS 0.9613 0.0387 10-6 tolerance in having reached the desired state are given in Table
DDM at stationary state 0.9632 0.0368 10-6
4. The tolerance was calculated as the average relative differ-
side product
ences between computed and desired values of components
HYSYS 0.0106 0.9612 0.0282 concentrations on all DWC trays. The tolerance values are the
DDM at stationary state 0.0042 0.9600 0.0358
consequence of the stopping criterion (the average absolute value
bottom product
of the derivatives less than 10-6).
HYSYS 10-8 0.0463 0.9537 The best control policies are presented in Figures 4-6. The
DDM at stationary state 10-8 0.0402 0.9598
5 or 10 steps cannot always be distinguished, because two or
steady state is reached in more than 630 min, and reaching more successive policy values are very close or identical. The
the stationary regime is decided by the bottom response. The corresponding evolutions of benzene concentration in the top
explanation consists of the bottom position and its volume, product, toluene concentration in the side product, and ethyl-
which is much greater than the trays volume. benzene concentration in the bottom product for the 10 time
TOC of DWC. A set of DDM simulations indicated that the stages policies are presented in Figures 7-9. These figures show
best control parameters for the optimal startup of the DWCs that the control policy is very much dependent on the number
are the reflux ratio, reboiler heat duty, and the side-draw flow of time stages. If another number of time stages is used, it would
rate. The bounds on these control variables are as follows: be possible to obtain different results. Therefore, the math-
ematical optimal control has not been reached. For practical
1.8 e R e 2.5 (13) reasons of DWC control, the best (not necessary optimal)
solutions obtained are also useful. It can be observed that the
responses in the bottom concentrations are determinant for the
1.676 106 e qB e 2.514 106 (14)
values of startup time, as expected.
The performances of best control policies are given in Table
25.77 e Fs e 37.14 (15) 5. The decreases of the startup time and the reboiler energy
savings are related to the traditional startup at time constant
These bounds are in agreement with technological and control values of control variables (indicated in Table 1). The perfor-
reasons; at the same time, the control domains limited by these mances of the 5 stages side-draw flow rate control and 10 stages
restrictions are large enough. For a total number of 64 trays side-draw flow rate control are very close. The first policy is
(including the bottom), the number of state variables is 256. easier and therefore preferred. Among the policies with one

Table 4. Values of Penalty Function and Tolerance in Having Reached the Desired State for Case Study 1
penalty function at the
number penalty function at the end of the 10th IDP tolerance at the end of
control policy of stages end of first IDP iteration iteration the 10th IDP iteration
reflux ratio 5 28396.73 9928.07 9.88 10-4
reflux ratio 10 22910.74 7899.85 9.39 10-4
reboiler heat duty 5 22442.24 13889.87 3.51 10-3
reboiler heat duty 10 25706.15 11798.57 3.28 10-3
side-draw flow rate 5 25667.71 15347.52 1.25 10-3
side-draw flow rate 10 22623.48 15126.25 1.21 10-3
reflux ratio and reboiler heat duty 10 17288.19 7366.22 1.19 10-3

Figure 5. Case Study 1: The best reboiler heat duty control for 5 stages
( ) and 10 stages (s).
Figure 6. Case Study 1: The best side-draw flow rate control for 5 stages
( ) and 10 stages (s).
Figure 7. Case Study 1: Evolutions during transient regime for the best
reflux control (10 stages) of benzene concentration in top product
(- - -), toluene in side product (s), and ethylbenzene in bottom
product (- -).
Ind. Eng. Chem. Res., Vol. 49, No. 19, 2010 9201
Figure 8. Case Study 1: Evolutions during transient regime for the best
reboiler heat duty control (10 stages) of benzene concentration in top product
(- - -), toluene in side product (s), and ethylbenzene in bottom product
(- -).
Figure 9. Case Study 1: Evolutions during transient regime for the best
side-draw flow rate control (10 stages) of benzene concentration in top
product (- - -), toluene in side product (s), and ethylbenzene in bottom
product (- -).
control variable, the 10 stages reflux control is the best. For
this policy, the decrease in startup time (and the corresponding
reboiler energy saving) is 81.65% from the time of the traditional
startup procedure.
An improved result is given by the simultaneous reflux and
reboiler heat duty control. The TOC algorithm was applied for
10 time stages, and the grid point numbers were ns ) 5, nu1 )
9, nu2 ) 9, and nV ) 9. As in the case of policies with one
control variable, the region contraction factor was set to 0.8,
and the total number of IDP iterations was 10. The decrease in
startup time for this policy is 83.70% from the time of the
traditional startup procedure, and the corresponding reboiler
energy saving is 83.23%. This control policy is presented in
Figure 10. The corresponding evolutions of benzene concentra-
tion in the top product, toluene concentration in the side product,
and ethylbenzene concentration in the bottom product are
presented in Figure 11.
A slight improvement of these results is expected when all
three controls will be used simultaneously. In this case, the
estimated computer execution time is overwhelming (approxi-
9202 Ind. Eng. Chem. Res., Vol. 49, No. 19, 2010

Table 5. Performances of the Best Control Policies for Case Study 1

reduction of reboiler energy reboiler energy computer execution
control policy number of stages startup time (min) startup time (%) consumption (kW h) savings (%) time (h)
traditional procedure 632 22067
reflux ratio 5 149 76.42 5202 76.42 8.35
reflux ratio 10 116 81.65 4050 81.65 15.79
reboiler heat duty 5 173 72.63 6005 72.79 7.48
reboiler heat duty 10 142 77.53 5078 76.98 13.24
side-draw flow rate 5 235 62.82 8078 62.82 8.78
side-draw flow rate 10 232 63.29 8100 63.29 14.01
reflux ratio and reboiler heat duty 10 103 83.70 3700 83.23 46.90

matively 140 h for 10 IDP iterations). The attempt to reduce
the computer execution time by decreasing the grid point
numbers ns, nu1, nu2, nu3, and nV did not give good results. In
fact, the corresponding decrease of startup time and reboiler
energy consumption is expected to be minor, because the side-
draw flow rate is the control variable with the smallest effect
(Table 5).
Although there are systems for which more than one state
grid point is necessary, Luus36 showed that for most systems
that he investigated, a single state grid point is a good choice.
Therefore, the state grid in the present work was reduced to
one point, which gave the possibility to increase the number of
points in the control grid and time grid to 25. The number of
time stages was 10, the region contraction factor was set to 0.8,
and the total number of IDP iterations was increased to 20. The
best solution obtained for the reflux control corresponds to a
startup time of 116 min, a value identical to that previously
obtained (Table 5). The computer execution time was 29.47 h
for 20 IDP iterations, about twice the time for a double number
of IDP iterations. The time evolution of control variable is
different from the best reflux control policy presented in Figure
4 for 10 time stages. The best solution obtained for the reboiler
heat duty control corresponds to a startup time of 162.5 min,
which is worse than the solution previously obtained (Table
5). The computer execution time was 37.99 h for 20 IDP
iterations, about 2.87 times larger for a double number of IDP
iterations. These results show that despite the use of a higher
number of points in the control grid and in the time grid and a
double number of IDP iterations, one state grid point is not a
good choice for the TOC of DWC. A similar result was obtained
for the piecewise linear control of distillation columns,26 when
the use of one state grid point was also considered.

Case Study 2: Comparison with a Conventional

Figure 10. Case Study 1: The best simultaneous reflux (thick line) and
reboiler heat duty (thin line) control.
Two-Column System
The aim of this case study is to compare the optimal startup
results of DWC to those for a conventional two-column system
(CCS). The same mixture, feed composition, and flow rate as
in Case Study 1 were considered. In the first column, benzene
is separated as a top product, while toluene and ethylbenzene
are separated in the second column. The general data for this
case study are given in Table 6. Products compositions and flow
rates are presented in Table 7. The structures of the two columns

Table 6. General Data for Case Study 2

parameters and
operating conditions column 1 column 2

feed flow rate (kmol/h) 126 49.82

feed composition (mole fractions) benzene, 0.6 benzene, 0.0231
toluene, 0.3 toluene, 0.7240
ethylbenzene, 0.1 ethybenzene, 0.2529
thermal state of the feed liquid at boiling liquid at boiling
temperature temperature
condenser type total total
reboiler type total total
reboiler heat dutya (kW) 2100 1250
pressure in the top of column 760 760
and in condenser (mm Hg)
reflux ratioa 2.6 1.66
column diameter (m) 2 0.9
bottom liquid volume (m3) 0.911 0.364
tray area (m2) 2.8143 1.257
hole area per plate (m2) 0.2815 0.1126
weir height (m) 0.025 0.025
hole diameter (m) 0.005 0.005
Figure 11. Case Study 1: Evolutions during the transient regime for the number of trays 20 35
best simultaneous reflux and reboiler heat duty control of benzene feed tray position 15 24
concentration in top product (- - -), toluene in side product (s), and
ethylbenzene in bottom product (- -). a
Prescribed values for the desired stationary regime.
 Ind. Eng. Chem. Res., Vol. 49, No. 19, 2010 9203

Figure 12. Case Study 2: Evolutions during the traditional startup transient
regime of concentrations of benzene (- - -), toluene (s), and ethylbenzene Figure 13. Case Study 2: The best reflux control for column 1 (s) and
(- -). column 2 ( ).
The startup and TOC of the second column are applied when
Table 7. Products Compositions and Flow Rates for Case Study 2
the first column reaches the stationary regime. This startup
top product bottom product procedure is also applied in industrial operations.
column 1 For the total number of 20 trays in the first column (including
composition (mole fractions): the bottom), the number of state variables is 80, and for 35
benzene 0.9777 0.0231 trays in the second column, there are 140 state variables. This
toluene 0.0223 0.7240 difference of the number of state variables is accountable for
ethylbenzene 0 0.2529 the higher value of computer execution time for the second
flow rate (kmol/h) 76.18 49.82
column (Table 8).
column 2 The TOC algorithm was applied for each column considering
composition (mole fractions): 10 time stages. The grid point numbers were ns ) 5, nu ) 9,
benzene 0.0311 0 and nV ) 9. The region contraction factor was set to 0.8, and
toluene 0.9617 0.0333
the total number of IDP iterations was 20. The best reflux control
ethylbenzene 0.0072 0.9667
flow rate (kmol/h) 37.06 12.76 policies of the two columns are presented in Figure 13, and the
corresponding evolutions of products composition are given in
were established iteratively with the DDM to obtain at final Figure 14. The performances of the best control policies are
stationary state products compositions similar to those consid- given in Table 8. The comparison of these results to the best
ered in the DWC (Case Study 1). The total number of trays in reflux control of DWC for 10 time stages (Table 5) indicates
the CCS was close to the number of trays in the DWC. These that the decreases by TOC of startup time (and corresponding
considerations allow a proper comparison of the two cases. The energy savings) have close values: 86.20% for the first column,
corresponding values of the reflux ratios of the two columns 75.91% for the second column, and 81.65% for the DWC. The
resulted from these conditions and not from an optimum design. total startup time for the best reflux control of the CCS is 93
As for Case Study 1, the Antoine model was used to calculate min, which is 20% shorter than the corresponding startup of
the thermodynamical properties of the system. the DWC (116 min). This result does not affect the use of DWC,
The evolutions during the traditional startup transient regime which has the advantages of energy savings in stationary regime
of products concentrations are presented in Figure 12. The reflux and lower capital cost. The reboiler duty for steady state in DWC
ratio was selected as the control variable for TOC in the CCS. (2095 kW in Table 1) is about 63% from the total reboilers
The bounds on this control variable are: duties in CCS (3350 kW, the sum of the reboiler duty in column
- For the first column: 1 and reboiler duty in column 2 in Table 6) as expected due to
the thermal coupling that reduces energy consumption. During
1.5 e R e 4 (16.1) the startup, the energy consumption is less in CCS because the
startup time is also shorter (632 min for DWC in traditional
- For the second column: operating mode and 116 min for DWC in reflux ratio control
policy in Table 5, as compared to 510 min for CCS in traditional
1.3 e R e 2.2 (16.2) operating mode and 93 min reflux ratio policy in Table 8). The

Table 8. Performances of the Best Control Policies for Case Study 2

reboiler energy
number of startup reduction of consumption reboiler energy computer execution
control policy stages time (min) startup time (%) (kW h) savings (%) time (h)
column 1 traditional procedure 290 10150
column 1 reflux ratio 10 40 86.20 1400 86.20 8.54
column 2 traditional procedure 220 4583
column 2 reflux ratio 10 53 75.91 1104 75.91 20.48
9204 Ind. Eng. Chem. Res., Vol. 49, No. 19, 2010

Table 9. General Data for Case Study 3

parameters and
operating conditions value
feed flow rate (kmol/h) 126
feed composition (mole fractions) methanol, 0.25
ethanol, 0.45
1-propanol, 0.3
side-draw flow ratea (kmol/h) 56.7
thermal state of the feed stream liquid at boiling
temperature
thermal state of the side-draw stream liquid at boiling
temperature
condenser type total
pressure in the top of column and in 760
condenser (mm Hg)
reflux ratioa 9
reboiler type total
reboiler heat dutya (kW) 2864
bottom liquid volume (m3) 0.410
tray area for top and bottom sections (m2) 1.2636
tray area for prefractionator 0.6318
Figure 14. Case Study 2: Evolutions during the transient regime for the and side-draw (m2)
best reflux control of concentrations of benzene (- - -), toluene (s), and hole area per plate for top and 0.1266
ethylbenzene (- -). bottom sections (m2)
hole area per plate for prefractionator 0.0633
and side-draw (m2)
higher energy consumption in DWC during the startup is weir height (m) 0.025
overcome by the advantage of DWC at steady state operation hole diameter (m) 0.005
when important energy savings are possible. The comparison tray number of top section 16
between DWC startup (Case 1) and CCS startup (Case 2) aimed tray number of prefractionator 36
tray number of side draw 36
to prove that the startup policies bring comparable energy tray number of bottom section 9
savings in both cases (81.65% for DWC, Table 5, and 83% for feed tray position 25
CCS, Table 8). side-draw tray position 79
liquid split fraction in prefractionator 0.59
Case Study 3: A Ternary Nonideal Mixture a
Prescribed values for the desired stationary regime.

A ternary nonideal mixture of alcohols, respectively, metha- Table 10. Mole Fractions of Products Components for Case Study 3
nol, ethanol, and 1-propanol, is separated in a DWC. The feed methanol ethanol 1-propanol
flow rate and feed composition are given in Table 9. The design
of DWC was done using HYSYS in a similar manner to Case top product
Study 1. The general data resulted from the design procedure 0.9967 0.0033 10-8
are presented in Table 9. Because of higher values of the number side product
of trays and reflux ratio, the separation degrees of the compo- 0.0043 0.9759 0.0198
nents (Table 10) are better than in Case Study 1.
bottom product
The equilibrium constants for the mixture methanol-
10-6 0.0036 0.9964
ethanol-1-propanol for eq 3.2 were obtained by regression of
equilibrium data, calculated in the frame of HYSYS, using the
where the coefficient of determination ) 0.997 and standard
UNIQUAC model. This procedure is similar to that used by
error ) 0.0092.
Woinaroschy,26 for the equilibrium data of propene-propane
The evolutions during the traditional startup transient regime
mixture using the Soave, Redlich, and Kwong equation of state.
of products concentrations are presented in Figure 15. As this
The regressed expressions of the equilibrium constants and
figure shows, the steady state is reached in 760 min.
corresponding regression statistics parameters are as follows:
For TOC the control parameters were reflux ratio and reboiler
1,j ) 5173.230 - 46.87533Tj + 8.8297247 10-2Tj2 + heat duty. The bounds imposed on these control variables are
as follows:
3.3168898 10-4Tj3 - 1.4083716 10-6Tj4 +
1.3958698 10-9Tj5 (17) 8.5 e R e 9.5 (20)

where the coefficient of determination ) 0.998 and standard

2.30 106 e qB e 3.44 106 (21)
error ) 0.0174.

2,j ) 4933.838 - 0.3914401Tj2 + 2.2053993 10-3Tj3 -
4.6604342 10-6Tj4 + 3.5028422 10-9Tj5 (18)
where the coefficient of determination ) 0.997 and standard
error ) 0.0177.
3,j ) 6504.007 - 0.5150276Tj2 + 2.8969788 10-3Tj3 -
6.1087665 10-6Tj4 + 4.5788536 10-9Tj5 (19)
As in Case Study 1, these bounds are in agreement to the
technological and control reasons, but at the same time, the
control domains, limited by these restrictions, are large enough.
For the total number of 98 trays (including the bottom), the
number of state variables is 392.
The TOC algorithm was applied for 10 time stages. The grid
point numbers were ns ) 5, nu ) 9, and nV ) 9. The region
contraction factor was set to 0.8, and the total number of IDP
iterations was 20.

Figure 15. Case Study 3: Evolutions during the traditional startup transient
regime of methanol concentration in top product (- - -), ethanol in side
product (s), and 1-propanol in bottom product (- -).
Figure 16. Case Study 3: The best reflux control.
Figure 17. Case Study 3: The best reboiler heat duty control.
The best control policies are presented in Figures 16 and 17,
and the corresponding evolutions of methanol concentration in
Ind. Eng. Chem. Res., Vol. 49, No. 19, 2010 9205
Figure 18. Case Study 3: Evolutions during transient regime for the best
reflux control of methanol concentration in top product (- - -), ethanol
in side product (s), and 1-propanol in bottom product (- -).
Figure 19. Case Study 3: Evolutions during transient regime for the best
reboiler duty control of methanol concentration in top product (- - -),
ethanol in side product (s), and 1-propanol in bottom product (- -).
top product, ethanol in side product, and 1-propanol in bottom
product are given in Figures 18 and 19. The ethanol concentra-
tion in the side product reaches a minimum value after 15 min
in the case of the best reflux control and at 30 min in the case
of the best reboiler duty control. The performances of the best
control policies are similar (Table 11) but less significant than
for Case Study 1. Despite the higher number of state variables
and a double number of IDP iterations, the computer execution
time is shorter than for Case Study 1. The explanation consists
of the simpler expressions of equilibrium constants and their
derivatives.
Discussion
The computer execution times are given for a computer with
a I7-965 processor (3.2 GHz), 3 GB DDR2 (677 MHz) memory,
and Microsoft Windows XP Professional SP3 operating system.
This computer is approximately 3.4 times faster than the IBM
Intel Pentium IV computer (2.66 GHz processor, 1 GB memory,
and Microsoft Windows XP Professional SP2 operating system)
that we have used in previous TOC work.26 With traditional
9206 Ind. Eng. Chem. Res., Vol. 49, No. 19, 2010
Table 11. Performances of the Best Control Policies for Case Study 3
control number of startup reduction of
policy stages time (min) startup time (%)
traditional procedure 760
reflux ratio 10 297 60.92
reboiler heat duty 10 309 59.34
nonparallel computer programming, the CPU usage of this
4-core processor is only 13%. By parallel, computer program-
ming as shown by Keil and Mandel,38 the computation time
can be reduced, since many calculations in IDP can be done in
parallel fashion.
The main goal of this paper is to present the possibility to
use IDP for TOC of DWC. Detailed investigations concerning
the best choice of IDP parameters (number of time stages, the
grids points numbers, region contraction factor, penalty coef-
ficients, total number of IDP iterations, etc.) were not considered.
The total number of state variables and differential equations
of the DDM in the TOC of DWC is too high for these
investigations. Long computing times would have been required
despite our access to a high-performance computer. There are
excellent works36,39 in the literature that performed this research
with simple differential models characterized by a low number
of state variables. In many cases, a good selection of IDP
parameters improves the performance index mathematically but
with very low or no practical consequences. For example, in
an application presented by Dadebo and McAuley40 by increas-
ing the number of time stages from 5 to 17 and the number of
state grid points from 10 to 20, the CPU time increased 7.29
times, and the performance index improvement was 1.77%.
Dadebo and McAuley,41 in a parallel reactions problem with
two state variables and one control parameter, managed to
improve the performance index only with 0.22%, by increasing
the number of time stages from 10 to 80. In a more recent work,
for a fed-batch reactor control with four state variables and one
control parameter, Luus42 obtained an improvement of the
performance index from 20842.278 to 20842.282, by increasing
the number of time stages from 44 to 324, and the best result
for an increased number of state grid points from 1 to 19 led to
an improvement of 0.33% of the performance index.
For each case study, the widely accepted thermodynamical
model type was selected. The benzene-toluene-ethylbenzene
mixture, at normal pressure, can be considered an ideal one,
and the use of Antoine model is reasonable. UNIQUAC (as in
Case Study 3) or NRTL models are usually used for vapor-liquid
equilibrium concerning nonideal mixture of alcohols, such as
methanol-ethanol-1-propanol. It was not the aim of the present
work to use several thermodynamic models for vapor-liquid
equilibrium in the frame of each case study to compare the
influence of the thermodynamic model on the results. The DDM
and the application of IDP for TOC of DWC accept any
vapor-liquid equilibrium model, and their proper selection is
a thermodynamic problem.
At the industrial level, it is important to consider the time
delay. According to the simplification assumption (iv) of DDM,
the time delay between plates was neglected in our study. To
include the time delay, a DDM with more detailed tray
hydraulics would be required. It is impractical to include
complex hydraulic calculations in the TOC procedure due to
excessive CPU time.
The selected control variables (reflux, reboiler heat duty, and
side-draw flow rate) can be easily manipulated in practical
applications. Unlike the cases of reflux or side-draw flow rate
control, when the switch off the control variables values can
reboiler energy
consumption reboiler energy computer execution
(kW h) savings (%) time (h)
36277
14177 60.92 5.97
14144 61.01 8.25
be done instantaneously, the modification of the reboiler heat
duty has a time delay, and the startup time and state variables
will be affected. These effects are not included in the above
case studies.
In a DWC, the reflux must be high enough to build-up the
reflux on both sides of the dividing wall. The value of the reflux
ratio has a big impact on the separation on both sides of the
dividing wall; hence, it affects the concentration profiles along
the trays. Not only the distillate and bottom product will be
influenced by the variation of reflux ratio but also the side-
draw stream, which is desired to contain the intermediate
component at high concentration. This consideration can lead
to the conclusion that a correct reflux policy will bring the DWC
in stable operating condition in a reasonable time. In Case Study
1, the values chosen for product purities are not very high.
Higher values for product purities would impose a larger number
of trays, leading to longer computing time requirements. Similar
values for products composition are used in other case studies
referring to the separation of hydrocarbon mixtures in a DWC.
For instance, Premkumar and Rangaiah43 mention values from
industrial sources used as specifications of the main component
concentration in the top, side and bottom products of 0.995,
0.91, and 0.92 for the separation of benzene-toluene-styrene
mixture, and 0.995, 0.96, and 0.96 for benzene-toluene-ethyl-
benzene mixture; they are very close with the products purities
in Case Study 1. With regard to the separation of methanol-
ethanol-1-propanol (Case Study 3), product concentrations in
the main component of 0.9968, 0.9766, and 0.9962 are reason-
able. In fact, this is a separation by distillation, not by high-
purity advanced separation techniques. The selected case studies
refer to mixtures not easily separable at high purities. For
benzene-toluene-ethylbenzene mixture, ethylbenzene increases
the difficulties to obtain high purities. For methanol-ethanol-1-
propanol mixture, methanol increases the difficulty to obtain
higher degrees of separation. Of course, by increasing the plates
number and/or reflux ratio, the products purities can be
improved. As the aim of the present paper was to formulate
and test a method for startup time minimization in a DWC, the
computation effort was not increased by imposing very high
product purities.
The liquid split, a practical tool for controlling the side-draw
stream purity (Mutalib and Smith44), was set at the values
obtained in steady state, as the feed composition of DWCs did
not vary in the present study. The influence of the liquid split
variation along the startup period did not affect significantly
the startup time. With regard to the vapor split, it was left to
adjust according to the temperature distribution along the trays
and pressure drop.
Luus45 investigated obtaining a bang-bang optimal control
policy for a binary distillation column. The application contains
a control model consisting of only 11 linear differential
equations. The IDP programming algorithm was different from
the classical IDP procedure (used in the cited works33-36 and
also in many other papers): He used a nested iterative
supplementary loop inside the IDP passes to reduce the control
variables domain. The use of a bang-bang control for distil-
 Ind. Eng. Chem. Res., Vol. 49, No. 19, 2010 9207
lation columns with its deep technological implications (e.g., T ) temperature (C)
hydrodynamic regime) is a challenging subject. t ) time (s)
As in the case of startup optimization of classical distillation u ) control vector
columns, it is possible to avoid undesirable secondary bang-bang V ) vapor flow rate (kmol s-1)
effects of the piecewise constant control by replacing it with V ) time stage (s)
piecewise linear control.25 The practical implementation of the x ) liquid mole fraction on plate
optimal policies obtained here can lead to some hydrodynamic y ) vapor mole fraction on plate
problems (flooding, weeping, liquid entrainment, etc). Therefore, zst ) static height of the liquid (m)
it is useful to test these control policies by simulations, using a zW ) weir height (m)
DDM with more detailed plate hydraulics, taking into account
the possibility of including the time delay. This way, some Greek Letters
suboptimal control policies, avoiding wrong hydrodynamic ) activity coefficient
regimes, can be identified by suitable corrections. Of course, a p ) pressure drop (mm Hg)
more useful but more difficult way to validate these control ) volumetric fraction
policies, based exclusively on simulation, is to test the approach ) equilibrium constant
on a pilot plant. F ) density (kg m-3)
) superficial tension (kg s-2)
Conclusions ) normalized time variable
In this work, the ability of IDP to solve high dimensional ) penalty coefficient
TOC problems, involving complex models, was demonstrated
Subscripts
once more. TOC of startup traditional distillation columns by
iterative programming proposed by Woinaroschy for ideal25 and B ) bottom, reboiler
nonideal mixtures26 was extended to the case of DWCs. The D ) distillate
proposed DDM proved to represent correctly the separation of F ) feed
an ideal ternary hydrocarbon mixture and of a nonideal ternary f ) final
alcohol mixture in a DWC. The values of internal flows and i ) component
temperature distributions along the trays at steady state were in j ) plate
good agreement with the simulations obtained in the frame of k ) time stage
commercial simulators. Usage of reflux ratio, reboiler heat duty, L ) liquid
or side-draw flow rate as control variables enabled a decrease S ) side-draw
of the startup time and a corresponding reboiler energy saving s ) stationary
up to about 80%, as compared to classical startup procedures. V ) vapor
For the case of DWCs, these reductions are of a similar order
of magnitude to the traditional distillation columns.
Literature Cited
Note (1) Rev, E.; Emtir, M.; Szitkai, Z.; Mizsey, P.; Fonyo, Z. Energy savings
of integrated and coupled distillation systems. Comput. Chem. Eng. 2001,
In the equations, the units of time are seconds, and the units 25, 119.
of power are Watts. In the tables and figures, for practical (2) Amminudin, K. A.; Smith, R. Design and optimization of fully
reasons, the units of time are minutes or hours, and the units of thermally coupled distillation columns. Part 2: Application of the dividing
wall in retrofit. Trans. IChemE. Part A 2001, 79, 716.
power are kiloWatts. (3) Doherty, M. F.; Malone, M. F. Conceptual Design of Distillation
Systems; McGraw Hill: Boston, 2001.
Nomenclature (4) Hernandez, J. G.; Segovia-Hernandez, V.; Ramrez, R. Thermody-
namicaly equivalent distillation schemes to the Petlyuk column for ternary
A0 ) hole area per plate (m2) mixtures. Energy 2006, 31, 2176.
Ap ) total area per plate (m2) (5) Emtir, M.; Rev, E.; Fonyo, Z. Rigorous simulation of energy
d0 ) hole diameter (m) integrated and thermal coupled distillation schemes for ternary mixtures.
Appl. Therm. Eng. 2001, 21, 1299.
F ) feed stream or side-draw streamflow rate (kmol s-1)
(6) Schultz, M. A.; Stewart, D. G.; Harris, J. M.; Rosenblum, S. P.;
h ) liquid enthalpy (J kmol-1) Shakur, M. S.; OBrien, D. E. Reduce costs with dividing wall columns.
H ) vapor enthalpy (J kmol-1) Chem. Eng. Prog. 2002, 98, 64.
I ) performance index (7) Schoenmakers, A. T. H.; Boll, M. Model predictive control of
L ) liquid flow rate (kmol s-1) integrated unit operation: Control of a dividing-wall column. Chem. Eng.
Process. 2004, 43, 347.
lW ) weir length (m)
(8) Errico, M.; Tola, G.; Rong, B. G.; Demurtas, D.; Turunen, I. Energy
m ) number of components saving and capital cost evaluation in distillation column sequences with
N ) liquid holdup per plate (kmol) dividing-wall column. Chem. Eng. Res. Des. 2009, 87, 1649.
n ) number of plates (9) Isopescu, R.; Woinaroschy, A.; Draghiciu, L. Energy Reduction in
ns ) dimension of state vector a Dividing-Wall Distillation Column. ReV. Chim. 2008, 59, 812.
(10) Draghiciu, L.; Isopescu, R.; Woinaroschy, A. Capital Cost Reduc-
nu ) dimension of control vector tion by the Use of Dividing-Wall Distillation Column. ReV. Chim. 2009,
nV ) dimension of time vector 60, 1056.
P ) vapor pressure (mm Hg) (11) Kim, Y. H. Structural design and operation of fully thermally
p ) total pressure on plate (mm Hg) coupled distillation column. Chem. Eng. J. 2002, 85, 289.
q ) heat rate losses per plate (W) (12) Triantafyllou, C.; Smith, R. The design and optimization of fully
thermally coupled distillation columns. Trans. IChemE. Part A 1992, 70,
R ) reflux ratio 118.
S ) number of time stages (13) Halvorsen, I. J.; Skogestad, S. Shortcut analysis of optimal operation
s ) state vector of Petlyuk distillation. Ind. Eng. Chem. Res. 2004, 43, 3994.
9208 Ind. Eng. Chem. Res., Vol. 49, No. 19, 2010
(14) Wang, S. H.; Wong, D. S. H. Controllability and energy efficiency
of a high purity dividing-wall column. Chem. Eng. Sci. 2007, 62, 1010.
(15) Wendt, M.; Konigseder, R.; Li, P.; Wozny, G. Theoretical and
experimental studies on start up strategies for heat-integrated distillation
column system. Chem. Eng. Res. Des. 2003, 81, 153.
(16) Woinaroschy, A.; Isopescu, R. Dividing-Wall Distillation Column:
Dynamic Modelling and Control. 18th European Symposium on Computer
Aided Process Engineering, Lyon, France, June 1-4; Computer-Aided
Chemical Engineering; Elsevier: Oxford, 2008; Vol. 25, p 313.
(17) Fieg, G.; Wozny, G. Experimental and theoretical studies of the
dynamics of startup and product switchover operation of distillation columns.
Chem. Eng. Process. 1993, 32, 283.
(18) Wozny, G.; Li, P. Optimization and experimental verification of
startup policies for distillation columns. Comput. Chem. Eng. 2004, 28,
253.
(19) Ling, H.; Luyben, W. L. New control structure for divided-wall
columns. Ind. Eng. Chem. Res. 2009, 48, 6034.
(20) Ling, H.; Luyben, W. L. Temperature Control of the BTX Divided-
Wall Column. Ind. Eng. Chem. Res. 2010, 49, 198.
(21) Vasquez-Castillo, J. A.; Venegaz-Sanchez, J. A.; Seqovia-Hernan-
dez, J. G.; Hernandez-Escota, H.; Hernandez, S.; Gutierrez-Antonio, C.;
Brionez-Ramirez, A. Design and optimization, using genetic algorithms of
intensified distillation systems for a class of quaternary mixtures. Comput.
Chem. Eng. 2009, 33, 1841.
(22) Ho, T. J.; Huang, C. T.; Lin, J. M.; Leea, L. S. Dynamic simulation
for internally heat-integrated distillation columns (HIDiC) for propylene-
propane system. Comput. Chem. Eng. 2009, 33, 1187.
(23) van Dieggelen, R. C.; Kiss, A. A.; Heemink, A. W. Comparison
of Control Strategies for Dividing-Wall Columns. Ind. Eng. Chem. Res.
2010, 49, 288.
(24) Woinaroschy, A. A New Model for the Dynamic Simulation of
the Rectification Processes. I. Development of the Calculation Model and
Algorithm. ReV. Chim. 1986, 37, 697.
(25) Woinaroschy, A. Time-Optimal Control of Startup Distillation
Column by Iterative Dynamic Programming. Ind. Eng. Chem. Res. 2008,
47, 4158.
(26) Woinaroschy, A. Time Optimal Control of Startup Distillation of
Nonideal Mixtures. Ind. Eng. Chem. Res. 2009, 48, 3837.
(27) Yasuoka, H.; Nakanishi, E.; Kunikita, E. Design of an On-line
Startup System for Distillation Column Based on a Simple Algorithm. Int.
Chem. Eng. 1987, 27, 466.
(28) Ruiz, C. A.; Cameron, I. T.; Gani, R. A Generalized Model for
Distillation Columns. III. Study of Startup Operations. Comput. Chem. Eng.
1988, 12, 1.
(29) Barolo, M.; Guaries, G. B.; Rienzi, S.; Trotta, A. On-line Startup
of a Distillation Column Using Generic Model Control. Comput. Chem.
Eng. 1993, 17S, 349.
(30) Barolo, M.; Guaries, G. B.; Rienzi, S.; Trotta, A. Nonlinear Model-
Based Startup and Operational Control of a Distillation Column: An
Experimental Study. Ind. Eng. Chem. Res. 1994, 33, 3160.
(31) Han, M.; Park, S. Startup of Distillation Columns Using Profile
Position Control Based on a Nonlinear Wave Model. Ind. Eng. Chem. Res.
1999, 38, 1565.
(32) Verbanov, P.; Klein, A.; Repke, J. U.; Wozny, G. Minimising the
startup duration for mass and heat integrated two-column distillation
systems: A conceptual approach. Chem. Eng. Process. 2008, 47, 1456.
(33) Bojkov, B.; Luus, R. Time-Optimal Control by Iterative Dynamic
Programming. Ind. Eng. Chem. Res. 1994, 33, 1486.
(34) Bojkov, B.; Luus, R. Use of Random Admissible Values for Control
in Iterative Dynamic Programming. Ind. Eng. Chem. Res. 1992, 31, 1308.
(35) Bojkov, B.; Luus, R. Evaluation of the Parameters Used in Iterative
Dynamic Programming. Can. J. Chem. Eng. 1993, 71, 451.
(36) Luus, R. IteratiVe Dynamic Programming; Chapman & Hall/CRC:
London, 2000.
(37) Olujie, Z.; Keibel, B.; Jansen, H.; Rietfort, T.; Zich, E.; Fery, G.
Distillation Columns Internals. Configuration for process Intensification.
Chem. Biochem. Eng. Q. 2003, 17, 301.
(38) Keil, F. J.; Mandel, K. Parallelization of iterative dynamic program-
ming. In Recent DeVelopments in Optimization and Optimal Control in
Chemical Engineering; Luus, R., Ed.; Research Signpost: India, 2002.
(39) Luus, R. Parametrization in nonlinear optimal control problems.
Optimization 2006, 55, 65.
(40) Dadebo, S. A.; McAuley, K. B. A Simultaneous Iterative Solution
Technique for Time-Optimal Control Using Dynamic Programming. Ind.
Eng. Chem. Res. 1995, 34, 2077.
(41) Dadebo, S. A.; McAuley, K. B. Dynamic Optimization of Con-
strained Chemical Engineering Problems Using Dynamic Programming.
Comput. Chem. Eng. 1995, 19, 513.
(42) Luus, R. Choosing Grid Points in Solving Singular Optimal Control
Problems by Iterative Dynamic Programming. Proceedings of the 10th
IASTED International Conference Intelligent Systems and Control, No-
vember 19-21, 2007, Cambridge, MA.
(43) Premkumar, R.; Rangaiah, G. P. Retrofitting conventional columns
systems to dividing-wall columns. Chem. Eng. Res. Des. 2009, 87, 47.
(44) Mutalib, M. I. A.; Smith, R. Operation and Control of Dividing
Wall Distillation Columns. Trans. IChemE. Part A 1998, 76, 318.
(45) Luus, R. Time Optimal Control of a Binary Distillation Column.
DeV. Chem. Eng. Mineral Process. 2002, 10, 19.
ReceiVed for reView January 14, 2010
ReVised manuscript receiVed July 25, 2010
Accepted July 26, 2010
IE100090P