1-s2.0-S0009250904004282-main

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


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


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


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


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


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


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


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


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


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


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


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


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


Strategy #1) T7organic reflux 2 = 18.57



CS5 (Inventory T7entrainer makeup 1 = 101.6 529.2 0.594



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.


organic reflux as two manipulated variables for dual-


CS3: using Inventory Strategy #1 with aqueous reflux and

organic reflux as two manipulated variables for dual-



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


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



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





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



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



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


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



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



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





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



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




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


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



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



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


-10

0

10

20

30

40

50

60





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



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




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


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



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



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





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



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




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


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


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


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


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


80

82

84

86

88

90

92

94

96

98

100

102

104



Feed Rate +10% ChangeFeed 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


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


20

25

30

35

40

45

50



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



Time (hr)

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20


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






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


15

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

OrganicLevel(m)

0.7220

0.7225

0.7230

0.7235

0.7240

0.7245

0.7250



Time (hr)

0 1 2 3 4 5 6 7 8 9 1 0 11 12 13 14 15 16 17 18 19 20


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





7/27/2019 1-s2.0-S0009250904004282-main

19/21


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


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


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


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









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


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.

1-s2.0-S0009250904004282-main

Documents