7/27/2019 1-s2.0-S0009250904004282-main
1/21
7/27/2019 1-s2.0-S0009250904004282-main
2/21
4548 I.L. Chien et al. / Chemical Engineering Science 59 (2004) 45474567
x (H O)
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
y(HO
)
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
x (H O)
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
RelativeVolatility((yHO
/xHO
)/(y
/x
))
1.0
1.2
1.4
1.6
1.8
2.0
2.2
Fig. 1. VLE and relative volatility for HAc and water system.
alcohol + cyclohexane + water system. Experimental veri-
fication of the above control strategy can be seen in Chien
et al. (2000a). Chien et al. (2000b) proposed a simple oper-
ating procedure under inverse double loop control strategy
to automatically adjust the column heat duty and organic re-
flux to be at optimum operating point and also proposed an
improved decoupling control strategy for the double loops.
Ulrich and Morari (2002) examine the influence of fourth
component impurities on the operation and control of a het-
erogeneous azeotropic distillation column for dewatering a
heavy-boiling organic using methyl tert-butyl ether as a lightentrainer. None of the above papers studied the acetic acid
dehydration system.
Design of acetic acid dehydration system using an en-
trainer has been studied in several publications. In a review
paper, Othmer (1963) described an azeotropic distillation
system containing a dehydrating column, a decanter, and a
water column for the separation of acetic acid and water.
The entrainer used before 1932 was ethylene dichloride, and
later normal propyl acetate and normal butyl acetate were
used to reduce the organic reflux and heat duty used in the
dehydrating column. In the paper by Pham and Doherty
(1990), examples of using ethyl acetate (cf. Tanaka and
Yamada, 1972), n-propyl acetate (cf. Othmer, 1941), or n-
butyl acetate (cf. Othmer, 1941; Tanaka and Yamada, 1972)
as entrainer were listed in a table of examples of heteroge-
neous azeotropic separations. Siirola (1995) uses acetic acid
dehydration as an example to demonstrate a systematic pro-
cess synthesis technique to the conceptual design of process
flowsheet. Ethyl acetate as entrainer was used in the paperby Siirola (1995) to design a complete acetic acid dehydra-
tion process with multiple effect azeotropic distillation and
heat integration. More recently, Wasylkiewicz et al. (2000)
proposed using geometric method for optimum process de-
sign of an acetic acid dehydration column with n-butyl ac-
etate as entrainer.
All of the above papers on acetic acid dehydration system
are on the subject of process synthesis and design, very lit-
tle discussion about control strategy of this system has been
found in the literature. Luyben and Tyreus (1998) offered
a realistic vinyl acetate monomer example for academic re-
searchers pursuing simulation, design, and control studies.
In this example, an azeotropic distillation column with de-
canter is presented. Although the flowsheet of this column
system is similar to this study with components of acetic acid
and water, but since vinyl acetate is a product of the overall
process, an extra organic phase product is drawn-off from
the decanter which is different from the system which will be
studied in this paper. Kurooka et al. (2000) proposed a non-
linear control system for the acetic acid dehydration column
with n-butyl acetate as entrainer. The thermodynamic model
used in this work is questionable because a minimum-boiling
azeotrope is predicted between n-butyl acetate and acetic
acid though the mixture is zeotropic (cf. Horsley, 1973).
In their study, complicated exact inputoutput linearizationcontroller was used with values of some unmeasured state
variables needed for the calculations. The resulting control
performances under feed rate and composition changes are
not desirable because of large fluctuations in the manipu-
lated variables. Gaubert et al. (2001) studied operation of an
unnamed organic acid dehydration in the industry using an
immiscible entrainer. Multiple steady states are confirmed
for the heterogeneous column by simulation and experimen-
tal data for the industrial unit. However, dynamics and con-
trol of this system is not studied in their paper.
In this paper, a suitable entrainer for this acetic acid de-
hydration system will be selected from several candidate ac-etates. Steady-state tray by tray column simulation will be
used to determine the best entrainer with minimum total an-
nual cost. Optimum process design and operating condition
will be determined to keep high-purity bottom acetic acid
composition and also keep a small acetic acid loss through
top aqueous draw. The overall control strategy of this col-
umn system will be proposed to hold both bottom and top
product specifications in spite of feed rate and feed compo-
sition disturbances. In the control study, conventional con-
trol strategy using only tray temperature measurements will
be considered so that the result of this study can easily be
used directly in industry.
http://-/?-http://-/?-http://-/?-7/27/2019 1-s2.0-S0009250904004282-main
3/21
I.L. Chien et al. / Chemical Engineering Science 59 (2004) 4547 4567 4549
Table 1
Experimental physical properties of three candidate acetates
Acetate Normal Azeotropic Azeotropic Azeotropic Aqueous Organic
boiling component temp. composition phase phase
point ( C) ( C) (water, mol%) (acetate, mol%) (acetate, mol%)
Ethyl 77.15 Water 70.38 31.2% 1.40% 83.5%
acetate (@40 C) (@40 C)iso-butyl 117.2 Water 87.4 56.1% 0.127% 90.47%
acetate (@40 C) (@40 C)
n-butyl 126.2 Water 90.2 72.2% 0.3638% 86.31%
acetate (@25 C) (@25 C)
Table 2
Parameter values for the NRTL model
i, j Bij Bj i ij
1, 2 576.234 322.424 0.3
1, 3 416.124 1024.50 0.3067
2, 3 211.310 652.995 0.3
(1) Ethyl acetate, (2) acetic acid, (3) water.
2. Process simulation and entrainer selection
Three candidate acetates will be studied in detailed pro-
cess simulation to demonstrate the factors needed to be
considered in determining the suitable entrainer for this
system. The three candidate acetates to be considered are:
ethyl acetate, iso-butyl acetate, and n-butyl acetate. The
important experimental physical properties of these three
acetates at atmospheric pressure are listed in Table 1. The
azeotropic data is from Horsley (1973), the vaporliquidequilibrium data is from Gmehling and Onken (1977) with
the VLE date for acetic acidiso-butyl acetate system from
Christensen and Olson (1992). For the binary and ternary
liquidliquid equilibrium data, they are from SZrensen and
Arlt (1979, 1980). The nonrandom two-liquid (NRTL) ac-
tivity coefficient model (Renon and Prausnitz, 1968) was
used for the vaporliquidliquid equilibrium (VLLE) for
the ternary system. The HaydenOConnell (Hayden and
OConnell, 1975) second virial coefficient model with asso-
ciation parameters was used to account for the dimerization
of acetic acid in the vapor phase. The Aspen Plus (Aspen
Technology, Inc., 2001) built-in association parameters wereemployed to compute fugacity coefficients. The extended
Antoine equation is used to calculate the vapor pressure of
each component in the system. The Aspen Plus built-in
parameters were again used in the simulation. The set of
NRTL parameters are obtained to be capable of describing
well the binary and ternary, vaporliquid equilibrium (VLE)
and liquidliquid equilibrium (LLE) data. The set of NRTL
parameters for the ternary systems of acetic acidethyl
acetatewater, acetic acidiso-butyl acetatewater, and
acetic acidn-butyl acetatewater are listed in Tables 24.
All three candidate entrainers form a minimum-boiling
azeotrope with water. A heterogeneous azeotropic distilla-
Table 3
Parameter values for the NRTL model
i, j Bij Bj i ij
1, 2 194.416 90.268 0.3
1, 3 489.609 1809.079 0.2505
2, 3 211.310 652.995 0.3
(1) Iso-butyl acetate, (2) acetic acid, (3) water.
Table 4
Parameter values for the NRTL model
i, j Bij Bj i ij
1, 2 397.85 68.61 0.3
1, 3 354.31 2578.35 0.219
2, 3 211.31 652.995 0.3(1) n-Butyl acetate, (2) acetic acid, (3) water.
tion column can be designed to obtain high-purity acetic acidproduct (b.p. of 118 C) at the column bottom while obtain-
ing minimum boiling entrainerwater azeotrope at the top
of the column. With this column design by adding entrainer
into the system, the difficult tangent pinch of the pure wa-
ter side can be avoided at the top of the column. Since this
entrainerwater azeotrope is heterogeneous, the top column
vapor stream forms two liquid phases after condensation in
the decanter. The organic phase will be refluxed back to
the heterogeneous azeotropic column to provide enough en-
trainer inside of the column. The aqueous phase containing
mostly water will be assumed to be drawn out from the sys-
tem for further treatment or discharge. Some of the aqueousphase can be refluxed back to the heterogeneous azeotropic
column if the organic reflux is too small to fulfill the column
specifications. The conceptual design of this heterogeneous
azeotropic distillation column system is illustrated in Fig. 2.
The residue curve maps with the binodal curve of the
LLE of the three entrainer systems studied in this paper
are shown in Figs. 35. By observing these three figures,
the prediction for entrainer solubility in water and also the
azeotropic temperature match well with the experimental
data in Table 1. The azeotropic composition for the iso-butyl
acetate system gives the most discrepancy in comparison
with the experimental data in Table 1. The experimental
http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-7/27/2019 1-s2.0-S0009250904004282-main
4/21
4550 I.L. Chien et al. / Chemical Engineering Science 59 (2004) 45474567
Fig. 2. Conceptual design for the separation of acetic acid and water.
Fig. 3. Residual curve map for the system HAcEtAcH2O.
azeotropic composition is at 43.9mol% iso-butyl acetate but
the simulation predicted at 36.8mol%. This is mainly due to
the compromise of obtaining the NRTL model parameters by
fitting all binary and ternary VLE and LLE data while trying
to predict well the azeotropic temperature and composition.
The residue curve maps for the ethyl acetate and the iso-
butyl acetate systems are similar in nature with the two-
componentazeotropeas the lowest temperature in thesystem
and acetic acid as the highest temperature in the system. The
Fig. 4. Residual curve map for the system HAciBuAcH2O.
residue curve map for the n-butyl acetate system is different
than the other two systems. In the n-butyl acetate system,
the highest temperature in the system is n-butyl acetate (b.p.
126.2 C), not acetic acid (b.p. 118 C). Slippage of entrainer
into the bottom product stream is the situation needed to
be avoided for this system. The other two systems do not
need to worry about this situation because acetic acid is the
highest temperature in the system which should come out
of the column through bottom stream in ideal situation.
7/27/2019 1-s2.0-S0009250904004282-main
5/21
I.L. Chien et al. / Chemical Engineering Science 59 (2004) 4547 4567 4551
Fig. 5. Residual curve map for the system HAcnBuAcH2O.
Rigorous process simulation is performed to find the op-
timum design and operating conditions of these three en-
trainer systems. The feed composition of 50 mol% acetic
acid and 50 mol% water is considered for the Aspen Plus
simulation. The feed rate is assumed to be 500kg/h and it
is saturated liquid phase. The column pressure is assumed
to be at atmospheric pressure. The decanter temperature is
at 40 C. In the Aspen Plus simulation, the column bot-
tom product is kept at 99.9mol% acetic acid high purity by
varying the reboiler heat duty and the column top aqueousproduct is kept at 0.1 mol% acetic acid loss by varying the
entrainer makeup flow rate. If the high purity specifications
cannot be met, portions of the aqueous phase can be refluxed
back to the column to fulfill the column specifications. This
extra third degree of freedom (aqueous reflux flow rate) is
fixed at a value which will meet both top and bottom prod-
uct specifications while also minimize reboiler heat duty of
the column system.
Design variable of total number of trays is a compromise
between the total equipment cost and the total utility cost.
The optimum total number of trays and the feed tray location
are determined to minimize Total Annual Cost (TAC). Thecalculation procedure of Douglas (1988) is followed with
the annual capital charge factor of 1/3 was used. The utility
cost is calculated the same way as in Chiang et al. (2002).
The Aspen Plussimulation results for the three entrainers
are summarized in Tables 57.
Several observations can be made by comparing these
three tables. Firstly, for the system of acetic acidethyl
acetatewater, no aqueous reflux is necessary to meet the
product specifications while the other two systems need
aqueous reflux stream for the separation with higher aque-
ous reflux flow rate for the n-butyl acetate system. Secondly,
the organic reflux flow rate and also the reboiler heat duty
for the ethyl acetate system are much larger in compari-
son with the other two systems. This high organic reflux
flow rate in the ethyl acetate system can actually be pre-
dicted by the inner molar balance envelope in Fig. 2 with
the residue curve map plot of the ethyl acetate system in
Fig. 3. Assuming at ideal condition, the column top vapor
composition should be at the ethyl acetatewater azeotropeand the column bottom composition should be very close
to the pure acetic acid corner in Fig. 3. Because of the feed
composition is at 50 mol% acetic acid and 50 mol% water
and the other inlet stream to the column for the inner molar
balance envelope in Fig. 2 is the organic reflux (recall that
no aqueous reflux is necessary for this system), the inter-
ception of the two inlet and outlet molar balance lines can
be used to estimate the organic reflux flow rate. Since the
interception point is closer to the organic reflux composi-
tion point, the organic reflux flow rate is quite high. If the
feed is much richer in acetic acid, the organic reflux flow
rate will be lower than the current case.
Another observation by comparing these three tables is
that the makeup flow rate for the ethyl acetate system is the
highest while for the iso-butyl acetate system is the lowest.
This can be explained by the outer molar balance envelope
in Fig. 2 with the knowledge of the aqueous phase composi-
tion in Fig. 3. Assuming ideal situation for the ethyl acetate
system, the two outlet streams for the outer molar balance
envelope are at the points of aqueous phase composition and
pure acetic acid in Fig. 3. The two inlet streams are at the
points of feed composition and pure ethyl acetate (entrainer
makeup) point. How close the interception point of the two
molar balance lines to the feed composition point can be
used to determine the makeup flow rate since the feed flowrate is known. If this interception point is very close to the
feed composition point, the makeup flow rate will be small.
From this explanation, it is not difficult to conclude that the
ethyl acetate system will have the highest makeup flow rate
and the iso-butyl acetate system will have the lowest makeup
flow rate.
The comparison of the minimum attainable TAC for these
three systems as well as the acetic acid dehydration system
without any entrainer is shown in Table 8. From the table,
one can observe that the no entrainer system required the
most TAC and the iso-butyl acetate system is most favor-
able for this feed composition and product specification re-quirements. The TAC for the iso-butyl acetate system is only
about 55% of the no entrainer system which represents large
saving can be made by using the iso-butyl acetate system.
Notice that this finding is in general agreement with the in-
dustrial applications. (See patents by Costantini et al., 1981
and by Parten and Ure, 1999). The above two patents also
found iso-butyl acetate as a favorable entrainer for the sep-
aration of acetic acid and water. Another earlier patent by
Othmer (1936) found n-propyl acetate to be useful as an en-
trainer for this system. The patent by Mitsui Petro-Chemical
Industries, Ltd. (1980) found n-butyl acetate to be favor-
able for this system. Notice that in all the patents above, the
http://-/?-http://-/?-http://-/?-7/27/2019 1-s2.0-S0009250904004282-main
6/21
4552 I.L. Chien et al. / Chemical Engineering Science 59 (2004) 45474567
Table 5
Stream information for the system acetic acid(HAc)ethyl acetate(EtAc)water(H2O)
Feed Bottom Top Aqueous Organic Makeup Reboiler
product product reflux reflux stream energy
Flow rate 213.48 106.74 108.98 0 573.95 2.24
(mol/min)
HAc mole 0.5 0.999 1.00103 2.87103 0 fraction
H2O mole 0.5 9.90104 0.9785 0.17586 0
fraction
EtAc mole 0.0 1.0105 2.05102 0.82127 1
fraction
Heat duty (KW) 401.17
Table 6
Stream information for the system acetic acid(HAc)iso-butyl acetate(iBuAc)water(H2O)
Feed Bottom Top Aqueous Organic Makeup Reboiler
product product reflux reflux stream energy
Flow rate 213.48 106.74 106.90 33.36 92.71 0.16
(mol/min)
HAc mole 0.5 0.999 1.00103 1.00103 1.90103 0
fraction
H2O mole 0.5 5.90104 0.9979 0.9979 7.98102 0
fraction
iBuAc mole 0.0 4.10104 1.10103 1.10103 0.9183 1
fraction
Heat duty (KW) 167.01
Table 7
Stream information for the system acetic acid(HAc)n-butyl acetate(nBuAc)water(H2O)
Feed Bottom Top Aqueous Organic Makeup Reboilerproduct product reflux reflux stream energy
Flow rate 213.48 106.74 107.44 98.78 102.32 0.70
(mol/min)
HAc mole 0.5 0.999 1.00103 1.00103 1.50103 0
fraction
H2O mole 0.5 7.48104 0.9928 0.9928 0.1448 0
fraction
nBuAc mole 0.0 2.52104 6.20103 6.20103 0.8537 1
fraction
Heat duty (KW) 259.68
Table 8Comparison of total annual cost for the acetic acid dehydration systems
Entrainer Optimal Optimal Annualized Utility cost Entrainer TAC($)
total feed capital cost cost
stages stage
Ethyl acetate 16 2 6.84104 4.20104 5.40104 1.64105
iso-butyl 30 9 6.81104 1.80104 1.70104 1.03105
acetate
n-butyl 31 11 8.44104 2.78104 6.08104 1.73105
acetate
No entrainer 50 37 1.42105 4.37104 0 1.86105
7/27/2019 1-s2.0-S0009250904004282-main
7/21
I.L. Chien et al. / Chemical Engineering Science 59 (2004) 4547 4567 4553
Fig. 6. Vapor and liquid profiles for the optimum system HAc
iBuAcH2O.
designed feed compositions and the specified column bot-
tom and top purities are all different from this paper, thus
no direct comparison of results can be made.
The vapor and liquid profiles inside the column for the
optimized iso-butyl acetate system can be seen in Fig. 6.
Notice that the column behaves as what was designed. The
first five stages counting from the top of the column have two
liquid phases. The combined liquid compositions for thesefive stages are plotted in this figure. Another thing worth
mention in the figure is that the column composition profiles
bypassing the corner of pure water which is the region for
the tangent pinch to be avoided.
From this study, some important factors in determining the
suitable entrainer for the acetic acid dehydration system are
summarized below. The information needed for this qualita-
tive comparison can be illustrated by the residue curve maps
with the binodal curve of the LLE as shown in Figs. 35.
The suitability of the entrainer is actually a combination of
the following factors.
2.1. Azeotropic composition and organic phase composition
It is better to have the azeotropic composition containing
more water in this mixture. This means that this entrainer is
more capable of carrying water to the top of the column, thus
less entrainer is needed inside of the column. The distance
for the points between azeotropic composition and organic
phase composition is better to be further apart. This means
that besides that the azeotropic composition containing more
water, the organic phase composition should contain more
entrainer. The location of these two points in Figs. 35 have
to do with the organic reflux flow rate into the heterogeneous
column as explained above during estimating the organic
reflux flow rate for the ethyl acetate system. From Figs. 35,
the ethyl acetate is the worst entrainer if only considering
this factor.
2.2. Azeotropic temperature
The azeotropic temperature determines the temperature
difference between the top and the bottom of the column. A
large delta T of the azeotropic temperature to the pure acetic
acid temperature implies a good separability. Less column
stages will be needed for specific product purity specifica-
tions. In this regard, ethyl acetate is the best entrainer. This
interpretation is confirmed by Table 8 because ethyl acetate
system requires the least total number of stages for the same
separation.
2.3. Aqueous phase composition and entrainer pricing
The aqueous phase should contain as little entrainer as
feasible. The reason is because the aqueous phase stream
will be drawn out of the system, thus any entrainer loss
should be compensated by the makeup stream in Fig. 2.
This will correspond to a stream cost of the system as seen
in Table 8. The makeup flow rate can actually be estimated
using the outer molar balance envelope in Fig. 2 during ideal
situation as explained previously. In this regard, iso-butyl
acetate system results in the least makeup flow rate while
ethyl acetate systemrequires themost makeupflow rate. This
is confirmed by Tables 57. The annual cost of this stream
is not only related to its flow rate but also related to theentrainer pricing. In this regard, ethyl acetate is the cheapest
and iso-butyl acetate is the most expensive entrainer. With
the knowledge of the entrainer pricing and the calculation
of the makeup flow rates for the three systems, the entrainer
cost can be estimated as seen in Table 8 even without any
rigorous simulation.
Since the system with iso-butyl acetate as entrainer re-
sults in most economical process design, we will study the
dynamic and control strategy of this system in detail in the
next section. Before doing that, let us first explore the neces-
sity of the aqueous reflux stream under various feed com-
positions for the system using iso-butyl acetate as entrainer.Fig. 7 shows the collection of many simulation results under
various feed composition conditions. In all the simulation
runs, the total numbers of stages for the column are all fixed
the same as the one in Table 8 (30 stages including reboiler
but not the condenser). The column bottom product is kept
at 99.9 mol% acetic acid high purity by varying the reboiler
heat duty and the column top aqueous product is kept at
0.1 mol% acetic acid loss by varying the entrainer makeup
flow rate. The aqueous reflux flow rate is fixed at value
which will meet both top and bottom product specifications
while also minimize reboiler heat duty of this column sys-
tem. From the figure, one can observed that for feed water
7/27/2019 1-s2.0-S0009250904004282-main
8/21
4554 I.L. Chien et al. / Chemical Engineering Science 59 (2004) 45474567
Feed Water Composition
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
AqueousRe
fluxFraction
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
Fig. 7. Minimum aqueous reflux ratio under various feed compositions.
composition above 70 mol%, the aqueous reflux stream is
not needed. For the feed composition studied in this paper,
the aqueous reflux stream is necessary in order to properly
hold the bottom and top product specifications.
Table 9 shows the value of main operating variables in
keeping the top and bottom product purity at their specifi-
cations under various feed composition conditions. In this
table, the operating condition is not unique for feed water
contents from 10% to 60%. For these feed composition
cases, there are three degrees of freedom (extra one is the
aqueous reflux) for the system with only two product purity
specifications. The ones included in the table are the operat-
ing conditions that minimize reboiler heat duty by varying
aqueous reflux flow rate. One important thing which needs tobe pointed out from the table is that in order to hold product
specifications, aqueous reflux flow rate needs to be adjusted
in a very wide range. This means that this manipulated
variable (aqueous reflux) should not be fixed in the control
strategy when trying to reject unmeasured feed composition
disturbance. This manipulated variable is preferable to be
used in the inferred composition loop to hold product spec-
ifications. Another observation is that the reboiler duty goes
Table 9
Desired operating conditions under various feed compositions
Feed H2O Aqueous Organic Reboiler Entrainer Aqueous Bottom
composition reflux reflux duty makeup draw product
(mol%) (mol/min) (mol/min) (KW) (mol/min) (mol/min) (mol/min)
10 133 102 189 0.07 21 192
20 109 100 186 0.08 43 171
30 85 99 182 0.11 64 150
40 60 96 175 0.13 86 128
50 33 93 167 0.16 107 107
60 8 90 159 0.16 128 85
70 0 99 175 0.16 150 64
80 0 113 200 0.19 171 43
90 0 127 226 0.21 193 21Product specifications: Bottom at 99.9mol% HAc and aqueous draw at 0.1mol% HAc.
up dramatically for the cases with no aqueous reflux (70%,
80%, and 90% feed H2O compositions). This implies that
it may be better to add a preconcentrator column before the
heterogeneous azeotropic column to increase the acetic acid
content in the feed to the heterogeneous azeotropic column
if the fresh feed water composition is too high. Another
possible advantage may be to have extra degree of freedom(aqueous reflux) to be manipulated in the control strategy.
The importance of the aqueous reflux stream for manipula-
tion purpose will be shown in the next section.
3. Control strategy design
The heterogeneous azeotropic column system using iso-
butyl acetate as entrainer will be studied in detail in this
section. The overall control strategy of this system will be
developed in order to hold bottom and top product spec-
ifications in spite of feed flow rate and feed composition
changes. In the control strategy development, we will as-sume no online composition measurement is available. The
composition control loops will be inferred by some tray tem-
perature control strategy. This type of control strategy can
easily be implemented in industry for wider applications.
The Aspen Plus steady state simulation in the last
section is exported to the dynamic simulation of Aspen
DynamicsTM. The tray sizing option in Aspen Plus is uti-
lized to calculate the column diameter to be 0.3259m with
the tray spacing of 0.6096 m is assumed. Other equipment
sizing recommended by Luyben (2002) is used here. The
volume of the reboiler is sized to give 10 min holdup with
50% liquid level. The decanter is sized to be bigger to al-low for two liquid phases to separate. The holdup time of
20 min is used in the dynamic simulation. Pressure-driven
simulation in Aspen DynamicsTM is used with the top
pressure of the azeotropic column controlled at 1.1 atm to
allow for some pressure drop in the condenser and decanter
to give the decanter at atmospheric pressure. The pressure
drop inside the column is automatically calculated in Aspen
DynamicsTM. Since the tray pressures in the columns are
7/27/2019 1-s2.0-S0009250904004282-main
9/21
I.L. Chien et al. / Chemical Engineering Science 59 (2004) 4547 4567 4555
Table 10
Base case condition of the optimum flow sheet for dynamic tests
Total number of stage for the 30 (including reboiler but not
azeotropic column including condenser)
Feed stage 9 (counting from the top tray)
Fresh feed flow rate 213.48 mol/min
Fresh feed composition 50mol% HAc and 50mol% H2O
Column reboiler duty 162.27 KW
Organic reflux flow rate 92.21mol/min
Aqueous reflux flow rate 33.40mol/min
Entrainer makeup flow rate 0.165mol/min
Bottom flow rate 106.76 mol/min
Bottom composition 99.89 mol% HAc
0.0665mol% H2O
0.0458mol% iBuAc
Top aqueous outlet flow rate 106.89mol/min
Top aqueous outlet composition 0.0997mol% HAc
99.79 mol% H2O
0.11mol% iBuAc
different than the constant atmospheric pressure assump-
tion used in steady-state simulation, the established base
case condition in Aspen DynamicsTM will be slightly dif-
ferent than Table 6 in previous section. The final base case
steady-state condition used for control study can be seen in
Table 10.
There are two inventory control strategies which can be
used for this system. The first inventory control strategy (In-
ventory Strategy #1) uses entrainer makeup flow to control
theorganic phase level in thedecanter. This inventory control
strategy was successfully used in Chien et al. (1999b, 2000a)when controlling an isopropyl alcohol dehydration column.
The second inventory control strategy (Inventory Strategy
#2) uses organic reflux flow to control the organic phase
level in the decanter. This second inventory control strategy
is more intuitively sound because organic reflux flow rate is
much larger than the entrainer makeup flow rate, thus the
organic phase level control should be more effective. Other
inventory control loops which use the same pairings for ei-
ther of the above strategies are: using top aqueous product
flow to control the aqueous phase level in the decanter; us-
ing bottom product flow to control the column bottom level.
The column top pressure is controlled at 1.1 atm by manip-ulating the top vapor flow and the decanter temperature is
controlled at 40 C by manipulating the condenser duty.
After deciding the inventory control strategy, there are
three variables left and can be used in some composition
control strategy. The three candidate variables for Inventory
Strategy #1 are: organic reflux flow, aqueous reflux flow,
and the reboiler duty; while the three candidate variables for
Inventory Strategy #2 are: entrainer makeup, aqueous reflux
flow, and the reboiler duty. The control objective is to hold
the bottom and the top aqueous product specifications at
base case condition under 10% feed flow and 10% feed
H2O composition changes.
Stages
1 2 3 4 5 6 7 8 9 101112131415161718192021222324252627282930
80
90
100
110
120
130
140
Stages
1 2 3 4 5 6 7 8 9 101112131415161718192021222324252627282930
80
90
100
110
120
130
140Stages
1 2 3 4 5 6 7 8 9 101112131415161718192021222324252627282930
Te
mp(oC)
Temp(oC)
Temp(
oC)
80
90
100
110
120
130
140
Aqueous Reflux 0.1% Change
Heat Duty 0.1% Change
Organic Reflux 0.1% Change
Base case
+0.1%
-0.1%
Base case
+0.1%
-0.1%
Base case
+0.1%
-0.1%
Fig. 8. Sensitivity analysis of the three manipulated variables under Strat-
egy #1.
3.1. Dual temperature loop control strategy
Since product specifications at both bottom and top ends
are specified, we will consider dual-point temperature con-
trol structure first. The sensitivity analysis with small pertur-
bation of the manipulated variables will be performed next
in order to determine the two temperature control points.
Fig. 8 shows the sensitivity analysis of the three manipulated
variable changes under Inventory Strategy #1 and Fig. 9
shows the sensitivity analysis of the three manipulated vari-
able changes under Inventory Strategy #2. The numbering
of the stage in this column is counting from top to bottom
http://-/?-http://-/?-7/27/2019 1-s2.0-S0009250904004282-main
10/21
4556 I.L. Chien et al. / Chemical Engineering Science 59 (2004) 45474567
Stages
1 2 3 4 5 6 7 8 9 1 0 1 1 12 1 3 1 4 1 5 16 1 7 1 8 1 9 20 2 1 2 2 2 3 24 2 5 2 6 2 7 28 2 9 3 0
T
emp(C)
80
90
10 0
11 0
12 0
13 0
14 0
Base Case
+0.1%
-0.1%
Aqueous Reflux 0.1% Change
Stages
1 2 3 4 5 6 7 8 9 10 1 1 1 2 13 1 4 1 5 16 1 7 1 8 19 2 0 2 1 22 2 3 2 4 25 2 6 2 7 28 2 9 3 0
Temp(C)
80
90
10 0
11 0
12 0
13 0
14 0
Heat Duty 0.1% Change
Stages
1 2 3 4 5 6 7 8 9 10 1 1 1 2 13 1 4 1 5 16 1 7 1 8 19 2 0 2 1 22 2 3 2 4 25 2 6 2 7 28 2 9 3 0
Temp
(C)
80
90
10 0
11 0
12 0
13 0
14 0
Base Case
+10%
-10%
Makeup Flow 10% Change
Base Case
+0.1%
-0.1%
Fig. 9. Sensitivity analysis of the three manipulated variables under Strat-
egy #2.
with stage #1 as the top stage and stage #30 as the reboiler.When perturbing one manipulated variable, the other two
manipulated variables are fixed at base case condition. The
final steady-state conditions of Figs. 8 and 9 are obtained
by running dynamic simulation with the above mentioned
perturbations and then wait until the dynamic simulation to
reach final steady-state.
For organic reflux changes in Fig. 8, a process gain sign
reversing is observed between stages #12 and #13. Dynam-
ically, a large inverse response was observed for column
stages between stages #13 to column bottom. Similarly for
entrainer makeup changes in Fig. 9, a process gain sign re-
versing is also observed between stages #12 and #13. This
Stages
5 10 15 20 25 30
Zi=|U1i|-|U2i|
-0.2
-0.1
0.0
0.1
0.2
0.3
Fig. 10. Plot ofZi for control structure CS1.
also indicates that dynamically a large inverse response was
also observed for column stages between stages #13 to col-
umn bottom.
Since the numbers of the candidate manipulated variables
for each inventory control strategy are three, there will be
three alternative temperature control structures for each in-
ventory control strategy. From the open-loop data in Figs. 8
and 9, steady-state gain matrix for each alternative overall
control strategy can be obtained by averaging the positive
and negative manipulated variable changes. Each elements
of the steady-state gain matrix is made to be dimensionless
by the spans of the temperature sensors and the manipulated
variables. Singular-value decomposition (SVD) as described
by Moore (1992) can be made on the steady-state gain ma-trices as follows:
K= UVT, (1)
where K is a 302 steady-state gain matrix for each control
strategy. U= [U1|U2] is an 302 orthonormal matrix, the
columns of which are called the left singular vectors.VT is a
22 orthonormal matrix, the columns of which are called the
right singular vectors. is a 22 diagonal matrix of scalars
called the singular values and organized in descending order.
To trade off between sensorsensitivity and loop interaction,a
function was defined by the difference between the absolute
values of the elements of the U vectors as:
Zi = |U1i | |U2i |. (2)
The maximum and the minimum of this function as sug-
gested by Moore (1992) are selected as the two tray locations
for the temperature control points. For an example, Fig. 10
shows the Zi for the Inventory Strategy #1 with two manip-
ulated variables of aqueous reflux and reboiler duty. From
this figure, temperatures at stages #6 and #16 are selected
for the two controlled variables for the above two manip-
ulated variables. To compare among the alternative control
structures, condition number (CN) and relative gain array
7/27/2019 1-s2.0-S0009250904004282-main
11/21
I.L. Chien et al. / Chemical Engineering Science 59 (2004) 4547 4567 4557
Table 11
SVD and RGA analysis for each control structure
Control Controller pairing Singular values CN RGA(11)
structure
CS1 (Inventory T16aqueous reflux 1 = 102.9 10.05 2.64
Strategy #1) T6reboiler duty 2 = 10.24
CS2 (Inventory T7organic reflux 1 = 112.2 3.14 0.673Strategy #1) T17reboiler duty 2 = 35.76
CS3 (Inventory T16aqueous reflux 1 = 60.72 3.27 0.874
Strategy #1) T7organic reflux 2 = 18.57
CS4 (Inventory T6aqueous reflux 1 = 108.0 7.89 2.24
Strategy #2) T15reboiler duty 2 = 13.68
CS5 (Inventory T7entrainer makeup 1 = 101.6 529.2 0.594
Strategy #2) T16reboiler duty 2 = 0.192
CS6 (Inventory T16aqueous reflux 1 = 38.93 282.1 0.595
Strategy #2) T6entrainer makeup 2 = 0.138
(RGA) are also calculated for each control structure. The
candidate control structures are listed below:
CS1: using Inventory Strategy #1 with reboiler duty and
aqueous reflux as two manipulated variables for dual-
point temperature control.
CS2: using Inventory Strategy #1 with reboiler duty and
organic reflux as two manipulated variables for dual-
point temperature control.
CS3: using Inventory Strategy #1 with aqueous reflux and
organic reflux as two manipulated variables for dual-
point temperature control.
CS4: using Inventory Strategy #2 with reboiler duty and
aqueous reflux as two manipulated variables for dual-
point temperature control.CS5: using Inventory Strategy #2 with reboiler duty and
entrainer makeup as two manipulated variables for
dual-point temperature control.
CS6: using Inventory Strategy #2 with aqueous reflux and
entrainer makeup as two manipulated variables for
dual-point temperature control.
Table 11 summarizes the results of SVD and RGA anal-
ysis for the above six control structures. From the results of
this table, several guidelines as suggested in Moore (1992)
are followed to screen out the undesirable control structures
from this steady-state analysis. The guidelines are: to select
the smallest singular value as large as possible; to select
the CN as small as possible; and to select the RGA(11) as
close to unity as possible. From these guidelines, CS5 and
CS6 are dropped for further comparison. For the remain-
ing four control structures (CS14), further closed-loop dy-
namic evaluation will be made to determine which control
structure is the best.
For the manipulated variables not used for temperature
control purpose, it is preferable to design some kind of ratio
scheme in order to move these manipulated variables ac-
cording to the disturbance changes. For example, with con-
trol structure CS1, constant ratio of organic reflux flow rate
to feed flow rate is maintained throughout the closed-loop
simulation run. In order to compensate the feed disturbance
effect dynamically, a first-order lag with adjustable time con-
stant is also included in the ratio scheme.
Table 11 only shows the steady-state characteristics of
each control structure. However, good steady-state charac-
teristics are not a sufficient condition for good dynamic con-
trol system performance. Thus, Aspen DynamicsTM will be
used to evaluate the control system performance for the alter-
native control structures. Since the 10% unmeasured feed
composition changes are the more severe closed-loop test in
comparison with the feed rate changes, these load changes
will be made in the closed-loop dynamic simulations for
comparison. All level loops are assumed to be controlled by
P-only controller in order to smooth out their manipulatedvariables in the system. Controller gain of 2.0 as suggested
in Luyben (2002) is used in all the level loops. The PID tun-
ing constants for all the stage temperature control loops are
tuned using the same multiloop tuning guideline (cf. Chien
et al., 1999a), thus fair comparison can be made on the
closed-loop dynamic responses among the four candidate
control structures. Besides the dynamic response, the eval-
uation of which control structure is adequate will emphasis
more on the observation at final steady-state if the candidate
control structure will actually meet the final product speci-
fications in spite of the load disturbances.
The closed-loop dynamic responses of control structures
CS1CS4 with 10% changes in the feed H2O composition
are shown in Figs. 1114, respectively. The disturbances are
introduced at time = 0.5 h. With Inventory Strategy #1 (or-
ganic phase level to manipulate the entrainer makeup flow),
the maximum makeup flow is assumed to be larger than
twice of the steady-state flow rate in order to have better con-
trol of the organic phase level when this level is dropping.
Since the control loop pairing of CS1 is unconventional (re-
boiler duty to control temperature at a stage closer to the
top of the column), it is very difficult to find proper tuning
constants for the two temperature control loops. The tuning
guideline in Chien et al. (1999a) has to be further detuned
7/27/2019 1-s2.0-S0009250904004282-main
12/21
4558 I.L. Chien et al. / Chemical Engineering Science 59 (2004) 45474567
Time (hr)
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
T
(C)
92
94
96
98
100
102
104
106
108
Feed H O +10% Change
Feed H O -10% Change
Time (hr)
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
T(
C)
110
112
114
116
118
120
122
124
126
Feed H O +10% Change
Feed H O -10% Change
Time (hr)
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
ReboilerDuty(KW)
152
154
156
158
160
162
164
166
Time (hr)
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
AqueousReflux(mol/min)
15
20
25
30
35
40
45
50
Feed H O +10% Change
Feed H O -10% Change
Feed H O +10% Change
Feed H O -10% Change
Time (hr)
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
OrganicLevel(m)
0.40
0.45
0.50
0.55
0.60
0.65
0.70
0.75
0.80
Feed H O +10% Change
Feed H O -10% Change
Time (hr)
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
MakeupFlow(mo
l/min)
0.0
0.2
0.4
0.6
0.8
1.0
1.2
1.4
1.6
1.8
Feed H O +10% Change
Feed H O -10% Change
Fig. 11. Closed-loop dynamic simulation using control structure CS1 with 10% changes in the feed H2O composition.
in order to make the system stabilized. Notice that for feed
H2O 10% change in Fig. 11, entrainer makeup flow valve
has to be fully open at time = 2.4 h and stayed fully open
until time = 10.2 h. The controlled temperature points are
not steady yet at final simulation time of 20 h. For CS2 in
Fig. 12, the dynamic response of 10% feed H2O change
is quite satisfactory. Two controlled temperature points are
returned back to setpoints well before time = 20 h. How-
ever, the dynamic response is unacceptable for +10% feed
H2O change. Although two controlled temperature points
reach their setpoints at final simulation time, the dynamic
response of organic phase level is very bad. Its manipu-
lated variable switches from fully close to fully open for
the entire simulation run. Similar unacceptable dynamic re-
sponses are observed in Fig. 13 for CS3. In comparison with
the other three control structures, it is quite obvious from
Fig. 14 that CS4 is the best control structure in terms of the
dynamic response. All controlled and manipulated variables
reach new steady-state within 8h. The organic phase level
is also maintained much better than the other three control
structures.
Although not shown in this paper, dynamic closed-loop
tests for CS5 and CS6 are also performed for 10% changes
in the feed H2O composition. The closed-loop performance
7/27/2019 1-s2.0-S0009250904004282-main
13/21
I.L. Chien et al. / Chemical Engineering Science 59 (2004) 4547 4567 4559
Time (hr)
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
T
(C)
90
92
94
96
98
100
102
104
106
108
110
112
Feed H O +10% Change
Feed H O -10% Change
Time (hr)
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
T
(C)
119.0
119.2
119.4
119.6
119.8
120.0
120.2
120.4
Feed H O +10% Change
Feed H O -10% Change
Time (hr)
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
OrganicR
eflux(mol/min)
80
90
100
110
120
130
Time (hr)
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
ReboilerDuty(KW)
140
150
160
170
180
190
200
Feed H O +10% Change
Feed H O -10% Change
Feed H O +10% Change
Feed H O -10% Change
Time (hr)
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
OrganicLevel(m)
0.0
0.2
0.4
0.6
0.8
1.0
Feed H O +10% Change
Feed H O -10% Change
Time (hr)
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
MakeupFlow
(mol/min)
-0.2
0.0
0.2
0.4
0.6
0.8
1.0
1.2
1.4
1.6
1.8
Feed H O +10% Change
Feed H O -10% Change
Fig. 12. Closed-loop dynamic simulation using control structure CS2 with 10% changes in the feed H2O composition.
is also not satisfactory particularly for the temperature loop
using entrainer makeup flow as the manipulated variable.
This ineffectiveness of the entrainer makeup flow on the con-
trolled tray temperature can actually be seen in previous sen-
sitivity plot (Fig. 9). Large 10% changes in the entrainer
makeup only give comparable tray temperature perturbations
to 0.1% changes in the aqueous reflux flow. The large vari-
ations on the entrainer makeup flow rate as well as the con-
trolled tray temperature have detrimental effect on the prod-
uct composition specifications. One thing worth mention is
that although the closed-loop dynamic response of CS6 is
much worse than CS4 but it is performed somewhat better
than CS1, CS2, and CS3. This is another proof that good
steady-state characteristics are not a sufficient condition for
good dynamic control system performance of nonlinear sys-
tems. The condition number of CS6 (282.1 in Table 11) is
much larger than CS1CS3 but gives better dynamic control
performance.
The final objective of the alternative control structures
is to maintain the bottom and top products specifications
in spite of the load disturbances. Figs. 15 and 16 compare
the dynamic responses of these four control structures with
+10% and 10% feed H2O composition changes, respec-
tively. It is obvious from these two figures that CS4 is the
best control structure to reject feed H2O composition dis-
turbances. Both bottom and top product compositions are
7/27/2019 1-s2.0-S0009250904004282-main
14/21
4560 I.L. Chien et al. / Chemical Engineering Science 59 (2004) 45474567
Time (hr)
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
T
(C)
90
92
94
96
98
100
102
104
106
108
110
112
Feed H O +10% Change
Feed H O -10% Change
Time (hr)
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
T
(C)
95
100
105
110
115
120
125
Feed H O +10% Change
Feed H O -10% Change
Time (hr)
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
OrganicR
eflux(mol/min)
80
90
100
110
120
130
Time (hr)
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
AqueousReflux(mol/min)
-10
0
10
20
30
40
50
60
Feed H O +10% Change
Feed H O -10% Change
Feed H O +10% Change
Feed H O -10% Change
Time (hr)
0 1 2 3 4 5 6 7 8 9 1 0 11 12 13 14 15 16 17 18 19 20
OrganicLevel(m)
0.0
0.2
0.4
0.6
0.8
1.0
1.2
Feed H O +10% Change
Feed H O -10% Change
Time (hr)
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
MakeupFlow(mol/min)
-0.2
0.0
0.2
0.4
0.6
0.8
1.0
1.2
1.4
Feed H O +10% Change
Feed H O -10% Change
Fig. 13. Closed-loop dynamic simulation using control structure CS3 with 10% changes in the feed H2O composition.
quickly returned back to specifications much faster than the
other three alternative control structures. Control structure
CS3 dynamically departs the furthest for both bottom and
top product compositions to their specifications for +10%
feed H2O composition change and control structure CS1
dynamically departs the furthest for 10% feed H2O com-
position change. In terms of the final steady-state value,
control structure CS2 departs the furthest to bottom prod-
uct specification (see Fig. 16). This control structure of CS2
makes the aqueous reflux flow rate fixed during the dy-
namic runs thus lose the ability to adjust the aqueous flow
rate upward or downward to cope with the feed H2O com-
position changes. Attempts have been made to introduce a
ratio scheme to maintain constant aqueous reflux ratio in-
stead of ratio to feed flow. The dynamic results are even
worse than fixing the aqueous reflux flow rate during load
disturbances.
Another important closed-loop test for this control system
is the feed flow rate changes. These changes are necessary
in order to adjust the production rate upward or downward.
Fig. 17 shows the dynamic responses for the proposed con-
trol structure CS4 under 10% feed rate changes. Notice
again that the closed-loop dynamic response is very sat-
isfactory. Although not shown in the paper, both product
specifications are maintained in spite of the production rate
changes.
7/27/2019 1-s2.0-S0009250904004282-main
15/21
I.L. Chien et al. / Chemical Engineering Science 59 (2004) 4547 4567 4561
Time (hr)
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
T
(C)
95
96
97
98
99
Feed H O +10% Change
Feed H O -10% Change
Time (hr)
0 1 2 3 4 5 6 7 8 9 1 0 11 12 13 14 15 16 17 18 19 20
T
(C
)
114
115
116
117
118
Feed H O +10% Change
Feed H O -10% Change
Time (hr)
0 1 2 3 4 5 6 7 8 9 1 0 11 12 13 14 15 16 17 18 19 20
AqueousR
eflux(mol/min)
20
25
30
35
40
45
50
Time (hr)
0 1 2 3 4 5 6 7 8 9 1 0 11 12 13 14 15 16 17 18 19 20
ReboilerDuty(KW)
156
158
160
162
164
166
168
170
Feed H O +10% Change
Feed H O -10% Change
Feed H O +10% Change
Feed H O -10% Change
Time (hr)
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
OrganicLevel(m
)
0.718
0.720
0.722
0.724
0.726
0.728
Feed H O +10% Change
Feed H O -10% Change
Time (hr)
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
OrganicReflux(mol/min)
88
90
92
94
96
98
Feed H O +10% Change
Feed H O -10% Change
Fig. 14. Closed-loop dynamic simulation using control structure CS4 with 10% changes in the feed H2O composition.
3.1.1. Summary of dynamic simulation runs
Some concluding remarks can be made after the above
dynamic runs. Firstly, good steady-state characteristics on
CS1CS3 (see Table 11) are not a sufficient condition for
good dynamic control system performance of nonlinear
systems. These three control structures gave much un-
acceptable closed-loop performance than CS4. Secondly,
dynamic simulations show that it is better to avoid using
entrainer makeup flow as the manipulated variable for either
of the organic phase level loop (CS1CS3) or the controlled
tray temperature loop (CS5 and CS6). In this system, there
are three free manipulated variables that can be cho-
sen for the dual temperature loop. If including the organic
phase level loop, there are total of four manipulated
variables that can be chosen from. Thus, it is possible to
select a control strategy not using entrainer makeup flow
as manipulated variable. This entrainer makeup flow rate
is fixed at the base case value and ratio to fresh feed rate
changes.
Fixing this entrainer makeup flow at base case value un-
der various feed composition changes can still meet prod-
uct purity specifications. This can be demonstrated by some
steady-state simulation runs as in previous Table 9. In that ta-
ble, desirable entrainer makeup flow rate is at 0.16mol/min
7/27/2019 1-s2.0-S0009250904004282-main
16/21
4562 I.L. Chien et al. / Chemical Engineering Science 59 (2004) 45474567
Time(hr)
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
TopAceticAc
idComposition
0.00
0.02
0.04
0.06
0.08
0.10
0.12
0.14
0.16
CS1
CS2
CS3
CS4
Time(hr)
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
BottomAceticAcidCom
position
0.84
0.86
0.88
0.90
0.92
0.94
0.96
0.98
1.00
1.02
CS1
CS2
CS3
CS4
Fig. 15. Comparison of bottom product acetic acid compositions for the
four control structures with +10% feed H2O composition change.
for 50% feed H2O composition. If feed H2O composition
is changed to 40%, from the table, the desirable entrainer
makeup flow rate is at 0.13 mol/min. This does not mean that
entrainer makeup flow rate has to be at this value to meet
two product purity specifications. As mentioned previously,
there are multiple steady-state conditions which can meet
both product purity specifications because there are three
degrees of freedom. In fact, a steady-state run at 40% feed
H2O composition can be made with fixing of the two prod-
uct purity specifications by varying reboiler duty and aque-
ous reflux and also fixing the entrainer makeup flow rateat 0.16 mol/min. The resulting steady-state condition gave a
little bit more on the value of the reboiler duty (176 KW vs.
175KW in Table 9) but still holding two product purity spec-
ifications. Also, noticeably the impurity of the bottom prod-
uct shifted to more in iBuAc and less in H2O but the total
impurity (iBuAc+H2O) is still at 0.1 mol% as desired. This
demonstrates that control structure to fix entrainer makeup
flow rate is workable.
Since the existence of this extra degree of freedom (aque-
ous reflux) to let the entrainer makeup flow released from
controlling the organic phase level or tray temperature, it is
desirable to have this extra degree of freedom in the sys-
Time(hr)
0 1 2 3 4 5 6 7 8 9 1 0 11 12 13 14 15 16 17 18 19 20
TopAceticAcidComposition
0.000
0.005
0.010
0.015
0.020CS1
CS2
CS3
CS4
Time(hr)
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
BottomAceticAcidCom
position
0.970
0.975
0.980
0.985
0.990
0.995
1.000
1.005
CS1
CS2
CS3
CS4
Fig. 16. Comparison of bottom product acetic acid compositions for the
four control structures with 10% feed H2O composition change.
tem. This extra degree of freedom can provide desirable
closed-loop dynamic response, thus if the feed composi-
tion is rich in water (70mol% H2O), it is preferable to
add a preconcentrator column before the feed stream to im-
prove the dynamic behavior. This suggestion is supported
from previous Fig. 7 and the dynamic simulation runs in this
section.
From the dynamic responses of CS4 (Fig. 14), another
important observation can be made. With disturbances like
10% changes in the feed H2O composition, the aqueous
reflux flow rate will be adjusted upward or downward ac-
cordingly in order to maintain about the same overall H2O
composition into the column. However, the reboiler duty is
actually returned back close to their original steady-state af-
ter some dynamic transients. This inspires the thinking if
simpler single temperature control strategy will work or not?
The attempts of using single temperature control strategy
will be explored next.
3.2. Simpler single temperature loop control strategy
The idea of the simpler single temperature loop control
strategy is to use aqueous reflux flow rate to hold some tray
7/27/2019 1-s2.0-S0009250904004282-main
17/21
I.L. Chien et al. / Chemical Engineering Science 59 (2004) 4547 4567 4563
Time(hr)
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
T(C)
97
Feed Rate +10% Change
Feed Rate -10% Change
Time(hr)
0 1 2 3 4 5 6 7 8 9 1 0 11 12 13 14 15 16 17 18 19 20
T
(C
)
113
114
115
116
117
118
119
120
Time(hr)
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
AqueousReflux(mol/min)
30
31
32
33
34
35
36
37
38
Time (hr)
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
ReboilerDuty
(KW)
140
150
160
170
180
190
Time(hr)
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
OrganicLevel(m)
0.710
0.715
0.720
0.725
0.730
0.735
Time(hr)
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
OrganicReflux(mol/min)
80
82
84
86
88
90
92
94
96
98
100
102
104
Feed Rate +10% Change
Feed Rate -10% Change
Feed Rate +10% ChangeFeed Rate -10% Change
Feed Rate +10% Change
Feed Rate -10% Change
Feed Rate +10% Change
Feed Rate -10% Change
Feed Rate +10% Change
Feed Rate -10% Change
Fig. 17. Closed-loop dynamic simulation using control structure CS4 with 10% changes in the feed flow rate.
temperature inside the column and to keep the other two
free manipulated variables (reboiler duty and entrainer
makeup) to maintain ratioed to the feed flow rate. From the
earlier sensitivity analysis in Fig. 9, the most sensitive con-
trol point inside the column is stage #6 which is closer to
the top of the column. Since the main acetic acid product is
drawn from the bottom of the column, an alternative con-
trol point is to select the second most sensitive control point
which is closer to the bottom of the column. This alterna-
tive control point will be stage #16. Thus two single loop
control structures will be deduced for closed-loop dynamic
test. They are:
CS7: using Inventory Strategy #2 in the previous subsection
with temperature at stage #6 controlled by manipulat-
ing aqueous reflux flow, while the other two manip-
ulated variables, reboiler duty and entrainer makeup,
maintain constant ratios to the feed flow rate.
CS8: using Inventory Strategy #2 in the previous subsec-
tion with temperature at stage #16 controlled by ma-
nipulating aqueous reflux flow, while the other two
manipulated variables, reboiler duty and entrainer
7/27/2019 1-s2.0-S0009250904004282-main
18/21
4564 I.L. Chien et al. / Chemical Engineering Science 59 (2004) 45474567
Time (hr)
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
T
(C)
94
95
96
97
98
99
100
Time (hr)
0 1 2 3 4 5 6 7 8 9 1 0 11 12 13 14 15 16 17 18 19 20
AqueousReflux(mol/min)
20
25
30
35
40
45
50
Feed H O +10% Change
Feed H O -10% Change
Time (hr)
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
OrganicLevel(m)
0.7215
0.7220
0.7225
0.7230
0.7235
0.7240
0.7245
0.7250
0.7255
Feed H O +10% Change
Feed H O -10% Change
Time (hr)
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
OrganicReflux(mol/min)
90.4
90.6
90.8
91.0
91.2
91.4
91.6
91.8
92.0
92.2
92.4
92.6
92.8
93.0
93.2
93.4
93.6
Feed H O +10% Change
Feed H O -10% Change
Feed H O +10% Change
Feed H O -10% Change
Fig. 18. Closed-loop dynamic simulation using control structure CS7 with 10% changes in the feed H2O composition.
Time (hr)
0 1 2 3 4 5 6 7 8 9 1 0 11 12 13 14 15 16 17 18 19 20
T
(C)
112
114
116
118
120
122
124
Time (hr)
0 1 2 3 4 5 6 7 8 9 1 0 11 12 13 14 15 16 17 18 19 20
AqueousReflux(mol/min)
15
20
25
30
35
40
45
50
Feed H O +10% Change
Feed H O -10% Change
Time (hr)
0 1 2 3 4 5 6 7 8 9 1 0 11 12 13 14 15 16 17 18 19 20
OrganicLevel(m)
0.7220
0.7225
0.7230
0.7235
0.7240
0.7245
0.7250
Feed H O +10% Change
Feed H O -10% Change
Time (hr)
0 1 2 3 4 5 6 7 8 9 1 0 11 12 13 14 15 16 17 18 19 20
OrganicReflux(mol/min)
90.6
90.8
91.0
91.2
91.4
91.6
91.8
92.0
92.2
92.4
92.6
92.8
93.0
93.2 Feed H O +10% Change
Feed H O -10% Change
Feed H O +10% Change
Feed H O -10% Change
Fig. 19. Closed-loop dynamic simulation using control structure CS8 with 10% changes in the feed H2O composition.
7/27/2019 1-s2.0-S0009250904004282-main
19/21
I.L. Chien et al. / Chemical Engineering Science 59 (2004) 4547 4567 4565
Time (hr)
0 1 2 3 4 5 6 7 8 9 1 0 11 12 13 14 15 16 17 18 19 20
TopAceticAc
idComposition
0.0007
0.0008
0.0009
0.0010
0.0011
0.0012
0.0013
CS7
CS8
CS4
Time (hr)0 1 2 3 4 5 6 7 8 9 1 0 11 12 13 14 15 16 17 18 19 20
BottomAceticAcidCo
mposition
0.9978
0.9980
0.9982
0.9984
0.9986
0.9988
0.9990
0.9992
0.9994
CS7CS8
CS4
Fig. 20. Comparison of bottom product acetic acid compositions for the
double loop control structure of CS4 with single loop control structures
of CS7 and CS8 with +10% feed H2O composition change.
makeup, maintain constant ratios to the feed flow
rate.
Figs. 18 and 19 show the closed-loop dynamic responses
for CS7 and CS8 under 10% feed H2O composition
changes, respectively. Notice that the dynamic responses
are all quite satisfactory with all variables settled out at
new steady-state values even faster than CS4 (comparing to
Fig. 14). The dynamic responses of the most important bot-
tom and top product compositions are shown in Figs. 20 and
21 for +10% and
10% changes in the feed H2O compo-sition, respectively. Notice first that the scaling ofFigs. 20
and 21 are much smaller than previous Figs. 15 and 16
indicating these two single loop control structures perform
much better than previous CS1, CS2, and CS3. Comparing
to more complex double loop control structure CS4, CS7
performs very satisfactory. Both product compositions are
maintained at tight specifications even the control point is
far away from the column bottom. On the contrary, the
acetic acid loss through the column top cannot be main-
tained at tight specification for control structure CS8. The
proposed control structure CS7 also performs very well for
10% feed rate disturbances. The dynamic responses for
Time(hr)
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
TopAceticAc
idComposition
0.000
0.001
0.002
0.003
0.004
0.005
0.006
CS7CS8
CS4
Time(hr)
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
BottomAceticAcidComposition
0.9972
0.9974
0.9976
0.9978
0.9980
0.9982
0.9984
0.9986
0.9988
0.9990
CS7
CS8
CS4
Fig. 21. Comparison of bottom product acetic acid compositions for the
double loop control structure of CS4 with single loop control structures
of CS7 and CS8 with 10% feed H2O composition change.
the temperature loop and the organic level loop are shown in
Fig. 22. All controlled and manipulated variables reach new
steady-state values after a short dynamic transient.Although
not shown in the paper, both product compositions are also
maintained at tight specifications. The final proposed simple
single temperature loop control structure of CS7 is shown in
Fig. 23. Although the two product purities are assumed not
to be measured on-line, but if they can be measured in-
frequently in quality lab, small trimming of the controlled
temperature setpoint or small changes of the reboiler heat
duty can be made to even more precisely to hold the prod-uct purities at their specifications during sustained feed
disturbances.
4. Conclusions
Three candidate entrainers (ethyl acetate, iso-butyl ac-
etate, and n-butyl acetate) are considered for acetic acid
dehydration via heterogeneous azeotriopic distillation. The
factors needed to be considered in selecting the proper en-
trainer are illustrated for this example system. Optimum col-
umn designs and operating conditions are obtained for these
7/27/2019 1-s2.0-S0009250904004282-main
20/21
4566 I.L. Chien et al. / Chemical Engineering Science 59 (2004) 45474567
Time (hr)
0 1 2 3 4 5 6 7 8 9 1 0 11 12 13 14 15 16 17 18 19 20
T
(C)
94
96
98
100
102
Time (hr)
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
AqueousR
eflux(mol/min)
25
30
35
40
45
Time (hr)
0 1 2 3 4 5 6 7 8 9 1 0 11 12 13 14 15 16 17 18 19 20
OrganicLevel(m)
0.71
0.72
0.73
0.74
Time (hr)
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
OrganicReflux
(mol/min)
75
80
85
90
95
100
105
110
Feed Rate +10% Change
Feed Rate -10% Change
Feed Rate +10% Change
Feed Rate -10% Change
Feed Rate +10% Change
Feed Rate -10% Change
Feed Rate +10% Change
Feed Rate -10% Change
Fig. 22. Closed-loop dynamic simulation using control structure CS7 with 10% changes in the feed flow rate.
Fig. 23. Schematic diagram of the proposed control structure CS7.
7/27/2019 1-s2.0-S0009250904004282-main
21/21
I.L. Chien et al. / Chemical Engineering Science 59 (2004) 4547 4567 4567
three candidate systems using rigorous process simulation.
Total Annual Cost (TAC) is used as the objective function in
determining the optimum column designs andoperating con-
ditions for these three candidate systems. Iso-butyl acetate
was found to be the best entrainer with resulting TAC only
about 55% of the system with no entrainer. The optimum
overall control strategy is also proposed for this column sys-tem to hold both bottom and top product specifications in
spite of10% feed rate and 10% feed H2O composition
load disturbances. Several alternative control structures are
compared using dynamic simulation. The proposed overall
control strategy is very simple requiring only one tray tem-
perature control loop inside the column. This simple overall
control strategy can easily be implemented in industry for
wider applications.
Acknowledgements
This work is supported by National Science Council ofR. O. C. under grant nos. NSC 89-2214-E-011-025 and NSC
90-2214-E-011-013. Helpful suggestions from anonymous
reviewers are gratefully acknowledged.
References
Aspen Technology, Inc., 2001. Aspen Plus Users Manual 11.1. Aspen
Technology, Inc., Cambridge.
Bozenhardt, H.F., 1988. Modern control tricks solve distillation problems.
Hydrocarbon Processing 67 (6), 4750.
Chiang, S.F., Kuo, C.L., Yu, C.C., Wong, D.S.H., 2002. Design alternatives
for the amyl acetate process: coupled reactor/column and reactive
distillation. Industrial and Engineering Chemistry Research 41 (13),
32333246.
Chien, I.L., Huang, H.P., Yang, J.C., 1999a. A simple multiloop tuning
method for PID controllers with no proportional kick. Industrial and
Engineering Chemistry Research 38 (4), 14561468.
Chien, I.L., Wang, C.J., Wong, D.S.H., 1999b. Dynamics and control of
a heterogeneous azeotropic distillation column: conventional control
approach. Industrial and Engineering Chemistry Research 38 (2),
468478.
Chien, I.L., Wang, C.J., Wong, D.S.H., Lee, C.H., Cheng, S.H., Shih, R.F.,
Liu, W.T., Tsai, C.S., 2000a. Experimental investigation of conventional
control strategies for a heterogeneous azeotropic distillation column.
Journal of Process Control 10 (4), 333340.
Chien, I.L., Chen, W.H., Chang, T.S., 2000b. Operation and decoupling
control of a heterogeneous azeotropic distillation column. Computers
and Chemical Engineering 24 (27), 893899.
Christensen, S.P., Olson, J.D., 1992. Phase equilibria and multiple
azeotrope of the acetic acidisobutyl acetate system. Fluid Phase
Equilibria 79, 187199.
Costantini, G., Serafini, M., Paoli, P., 1981. Process for the recovery of
the solvent and of the by-produced methylacetate in the synthesis of
terephthalic acid. U.S. Patent 4, 250, 330.
Douglas, J.M., 1988. Conceptual Process Design. McGraw-Hill, New
York.
Gaubert, M.A., Gerbaud, V., Joulia, X., Peyrigain, P.S., Pons, M., 2001.
Analysis and multiple steady states of an industrial heterogeneous
azeotropic distillation. Industrial and Engineering Chemistry Research
40 (13), 29142924.
Gmehling, J., Onken, U., 1977. Vaporliquid equilibrium data collection.
In: Behrens, D., Eckermann, R. (Eds.), DECHEMA Chemistry Data
Series. DECHEMA Publishers, Frankfurt, Germany.
Hayden, J.G., OConnell, J.P., 1975. A generalized method for predicting
second virial coefficients. Industrial and Engineering Chemistry Process
Design and Development 14, 209.
Horsley, L.H., 1973. Azeotropic Data III. Advances in Chemistry Series
No. 116. American Chemical Society, Washington, DC.Kurooka, T., Yamashita, Y., Nishitani, H., Hashimoto, Y., Yoshida, M.,
Numata, M., 2000. Dynamic simulation and nonlinear control system
design of a heterogeneous azeotropic distillation column. Computers
and Chemical Engineering 24 (27), 887892.
Luyben, M.L., Tyreus, B.D., 1998. An industrial design/control study
for the vinyl acetate monomer process. Computers and Chemical
Engineering 22 (78), 867877.
Luyben, W.L., 2002. Plantwide Dynamic Simulators in Chemical
Processing and Control. Marcel Dekker, New York.
Moore, C.F., 1992. Selection of controlled and manipulated variables.
In: Luyben, W.L. (Ed.), Practical Distillation Control. Van Nostrand
Reinhold, New York.
Mitsui Petro-Chemical Industries, Ltd., 1980. Process for azeotropic
distillation. United Kingdom Patent 1, 576, 787.
Othmer, D.F., 1936. Process for dehydration of acetic acid and otherlower fatty acids. U.S. Patent 2, 050, 234.
Othmer, D.F., 1941. Azeotropic distillation for dehydrating acetic acid.
Chemical and Metallurgical Engineering 40, 9195.
Othmer, D.F., 1963. Azeotropic separation. Chemical Engineering Progress
59 (6), 6778.
Parten, W.D., Ure, A.M., 1999. Dehydration of acetic acid by azeotropic
distillation in the production of an aromatic acid. U.S. Patent 5, 980,
696.
Pham, H.N., Doherty, M.F., 1990. Design and synthesis of heterogeneous
azeotropic distillations-III. column sequences. Chemical Engineering
Science 45 (7), 18451854.
Renon, H., Prausnitz, J.M., 1968. Local compositions in thermodynamics
excess functions for liquid mixtures. A.I.Ch.E. Journal 14, 135.
Rovaglio, M., Faravelli, T., Biardi, G., Gaffuri, P., Soccol, S., 1993.
The key role of entrainer inventory for operation and control ofheterogeneous azeotropic distillation column towers. Computers and
Chemical Engineering 17 (5), 535.
Siirola, J.J., 1995. An industrial perspective on process synthesis. In:
Biegler, L.T., Doherty, M.F. (Eds.), A.I.Ch.E. Symposium Series No.
304, vol. 91, 222233.
SZrensen, J.M., Arlt, W., 1979. Liquidliquid equilibrium data collection
binary systems. In: Behrens, D., Eckermann, R. (Eds.), DECHEMA
Chemistry Data Series. DECHEMA Publishers, Frankfurt, Germany.
SZrensen, J.M., Arlt, W., 1980. Liquidliquid equilibrium data collection
ternary systems. In: Behrens, D., Eckermann, R. (Eds.), DECHEMA
Chemistry Data Series. DECHEMA Publishers, Frankfurt, Germany.
Tanaka, S., Yamada, J., 1972. Graphical calculation method for minimum
reflux ratio in azeotropic distillation. Journal of Chemical Engineering
of Japan 5, 2026.
Ulrich, J., Morari, M., 2002. Influence of impurities on the controlof heterogeneous azeotropic distillation columns. Industrial and
Engineering Chemistry Research 41 (2), 230250.
Wasylkiewicz, S.K., Kobylka, L.C., Castillo, F.J.L., 2000. Optimal design
of complex azeotropic distillation columns. Chemical Engineering
Journal 79, 219227.
Widagdo, S., Seider, W.D., 1996. Azeotropic distillation. A.I.Ch.E. Journal
42 (1), 96130.