I n d . Eng. C h e m . Res.

1990,29, 1383-1395 1383

Multicomponent Three-phase Azeotropic Distillation. 3. Modern

Thermodynamic Models and Multiple Solutions
Brett P. Cairns and Ian A. Furzer*
Department of Chemical Engineering, University of Sydney, Sydney, New South Wales, Australia 2006

A new simulation method using an extended phase-stability and phase-splitting algorithm has been
programmed, and new three-phase regions are identified in published azeotropic distillation examples.
If phase splitting is ignored, then an erroneous composition profile can be produced. The modern
thermodynamic models UNIFAC-VLE, UNIFAC-LLE, modified UNIFAC, UNIQUAC, NRTL,
ASOG, and modified ASOG have been investigated in the simulation model. The sensitivity of the
selection of thermodynamic model on temperature and composition profiles has been studied for
a wide range of systems. The changes in the number of theoretical stages required for the separation
is dependent on the choice of thermodynamic model. A worked example of ethanol dehydration
with 2,2,4-trimethylpentane shows the importance of reflux ratio, feed plate location, and number
of stages on the separation. Multiple solutions for azeotropic distillation columns reported in the
literature can be rationalized by the inclusion of a phase-splitting algorithm in the simulation model.
A new set of multiple solutions is reported for the system ethanol-water-2,2,4-trimethylpentane.

Part 2 of this publication by Cairns and Furzer (1990b) Table I. Initial Temperature Estimates, Maximum
was concerned with reviewing the previous simulation Variable Corrections, and Number of Iterations
models and worked examples for three-phase distillation top plate reboiler no. of
and the development using a modified phase-stability fig no. temp, “C temp, O C AT” AXmaX iterations
analysis of a new simulation method. This new method 1 88.0 93.0 10 0.10 4
has been programmed and named NSHET and used to 2 88.0 100.0 10 0.10 7
predict the separation that had been previously estimated 3 -55.0 0.0 10 0.10 7
by those worked examples. The examples simulated are 4 46.0 50.0 10 0.10 4
given in part 2, Table I, numbers 3 and 5-10 covering a 4 46.0 160.0 10 0.10 11
4 46.0 160.0 20 0.10 7
wide range of system components. Figure 1 shows the 4 160.0 46.0 20 0.10 8
composition profile along the column for the system 1- 5 83.0 85.0 10 0.10 6
propanol (1)-water (2)-l-butanol (3), Figure 2 for the 6 70.0 77.0 10 0.10 5
system ethanol (1)-water (2)-l-butanol (3),and Figure 3 7 70.0 82.0 10 0.10 7
for the system propylene (1)-benzene (2)-n-hexane (3)- 8 70.0 82.0 10 0.10 9
water (4) using UNIFAC-VLE thermodynamics. The full 9 63.0 78.0 10 0.10 10
graphical results are given in Figures 1-9, and where phase
splitting was detected, the overall liquid compositions have sixth plate. This compares with Pratt’s original hand
been shown. All solutions were obtained by starting the calculation which shows two liquid phases on the top two
Newton-Raphson iterations from the overall feed com- stages. The second case, Figure 8 however, matches Pratt’s
position on all stages. The initial temperature map was hand calculations with two liquid phases on the top two
set by linear interpolation between the top plate and re- plates.
boiler estimates shown in Table I. Antoine coefficients Figure 1 shows the solution of the l-propanol-water-
were taken from Sinnott et al. (1983). The number of butanol example proposed by Block and Hegner (1976)
iterations required to reach the solution for a tolerance of using UNIFAC-VLE thermodynamics. The butanol con-
1.0 X lo4 is also given in Table I, along with the maximum centrates sufficiently on the lower plates to initiate a liq-
allowable variable corrections. uid-phase split. Figure 2 shows the solution for a similar
The number of iterations required for all problems was problem, given by Ross (1979), where the propanol of the
generally less than 10. Boston and Shah (1979) solved the Block and Hegner (1976) example is replaced by ethanol.
acetone-chloroform-water example of Figure 4 for initial Again the three-phase region is found in the bottom section
temperature estimates of up to 150 K outside the solution. of the column.
Table I shows a similar analysis where the bottoms tem- The results shown in Figure 3 are for the four-compo-
perature estimate is some 100 K away from the final so- nent hydrocarbon and water systems treated by Boston
lution. For a ATMAxequal to 10 K and AXMM,equal to and Shah (1979). Using UNIFAC-VLE, NSHET predicts
0.10 mole fraction, NSHET converged in 11 iterations. that the bottom two stages are within the three-phase
When the maximum temperature correction was doubled region as water concentrates toward the reboiler, leaving
to give ATMAX equal to 20 K, seven iterations were re- propylene to be collected overhead. This compares with
quired. The algorithm also converged in only eight iter- Boston and Shah (1979) who used the Margules equation
ations for the poor starting point where the reboiler and to calculate the liquid-phase activity coefficients and found
top plate estimates were reversed, Table I. that the bottom four stages had two liquid phases. Figure
The examples of Figures 7 and 8 are taken from the 5 shows the results for the final problem presented by
work of Pratt (1947), who gives results for hand calcula- Boston and Shah (19791, which involves separating l-bu-
tions for both problems. The UNIQUAC parameters have tanol, water, and butyl acetate in a five-stage column.
been taken from Gmehling and Onken (1977) and are NSHET, using UNIFAC-VLE thermodynamics, predicts
directly fitted to the experimental data measured by Pratt that the top 80% of the column has two liquid phases due
(1947). The example of Figure 7 shows that NSHET to the high concentration of water. Boston and Shah
predicts that the three-phase region extends down to the (1979) used NRTL and found that only the top two stages
Osss-5a85/90/2629-1383~02.50/0 0 1990 American Chemical Society
1384 Ind. Eng. Chem. Res., Vol. 29, No. 7, 1990

z o 9: THREE
- PHASE
.,OR
(07 - REGION

$06

605
'04

O3
:L
RUTYLACETATE
02
BUTnACETATE j
01 - "-BUTANOL
0.0' . ' ' ' ' I nn I
O 1 2 3 4 5 6 7 8 9 10 11 \,."

STAGE NUMBER 0 I 2 3 4 S
STAGE NUMBER
Figure 1. 1-Propanol (1)-water (2)-l-butanol (3) UNIFAC-VLE Figure 5. 1-Butanol (1)-water (2)-butyl acetate (3) UNIFAC-VLE
thermodynamics. thermodynamics
IO, IO L I

L I

4 07
$ 06
6 05
' 04

01
02 t
01

0 1 2 3 4 5 6 7 8 9 10 11 0 I 2 3 4 5 6 7 8 9 10 11 12 13 14 I5 16 17 18
STAGE NUMBER
STAGE NUMBER

Figure 2. Ethanol (1)-water (2j-1-butanol (3) UNIFAC-VLE Figure 6. Acrylonitrile (1)-acetonitrile (2)-water (3) UNIFAC-VLE
thermodynamics. thermodynamics.

, " I .
07
REGION

0 05

0.4
0.3 - 03 - \ ! TRICHLOROETHYLENE
02 - 02 - WATER /
0.1 - 01 -
0.0 ~

1 2 4 4
~ STAGENUMBER STAGE NUMBER

Figure 3. Propylene (1)-benzene (2)-n-hexane (3)-water (4) UNI- Figure 7. Acetonitrile (1)-trichloroethylene (2)-water (3) UNI-
FAC-VLE thermodynamics. QUAC thermodynamics.
10
1.0 .
WHOLE
COLUMN
m
THREE
PHASE
RUjlON
6 0.5
0 0.5 ACETONE

WATER
1
' 0.4

0.3 - 0.3

0.2 CHLOROFORM

O0.0
1 0 I 1 2 3 4 5

STAGE NUMBER
6 7 8 9 10
0 1 2 3 4 5 6 7
STAGE NUMBER
8 9 10 11 12 13

Figure 4. Acetone (1)-chloroform (2)-water (3) UNIFAC-VLE Figure 8. Acetonitrile (1)-trichloroethylene (2)-water (3) UNI-
thermodynamics. QUAC thermodynamics.

were within the three-phase region. profiles, generated with UNIFAC-VLE, match those cal-
Figure 6 shows the results of the system considered by culated by Baden (1984).
Buzzi Ferraris and Morbidelli (1982) to illustrate their If the possibility of liquid-phase splitting is ignored, a
approximate model. Using NRTL, Buzzi Ferraris and two-phase solution which always exists for a distillation
Morbidelli (1982) calculated that only the top 4 stages problem, may in fact be satisfied by a three-phase solution.
would experience a liquid-phase split, whereas, UNIFAC- In many cases, depending on the thermodynamic model,
VLE and NSHET predict that the three-phase region the pseudo-two-phase solution is not very different from
extends to the 15th stage. the actual three-phase solution. Indeed, many authors
Finally, Figure 9 shows the results for an ethanol- suggest using the two-phase solution as the initial condi-
water-benzene azeotropic column which has approximately tions for the full three-phase calculations. However, in
70% of the column in the three-phase region. These other cases, the three-phase solution may differ markedly,
 Ind. Eng. Chem. Res., Vol. 29, No. 7, 1990 1385
1.0 L I
0.9
W E
PHASE
WION
- THREEPHASE
0.8
7 ------- TWOPHASE

03
02
0.1 WAT€R
0.0
0 2 4 6 8 1 0 1 2 1 4 1 6 1 8 ~ ~ 2 4
STAGE NUMBER

Figure 9. Benzene (1)-water (2)-ethanol (3) UNIFAC-VLE ther-

modynamics.

"."
0 2 4 6 8 IO 12 14 16 18 M 22 24

I METHANOL STAGE NUMBER

Figure 13. Benzene (1)-water (2)-ethanol(3) UNIFAC-VLE ther-
modynamics.

- THREEPHASE

0 2 4 6 8 10 12 14 16 I8 20
STAGE NUMBER
Figure 10. Propane (1)-butane (2)-pentane (3)-methanol (4)-hy-
drogen sulfide (5) UNIFAC-VLE thermodynamics.

tII5-J
62

60
0 2 4 6 8 10 12 14
STAGE NUMBER
16
UNIFAC-VLE PREDICTED
AZEOTROPIC'TEMPERATURE

18 20 22 24

0.3
0.2
0.1
1 ACETONE
Figure 14. Benzene (1)-water (2)-ethanol(3)UNIFAC-VLE ther-
modynamics.
0.0
0
'
1
'
2
'
3
'

4 5
' '

6
'-'
7 8 9
' '
1 0 1 1 I 2 1 3 1 4 I 5
' '
found on a single plate around the feed. The effect of
STAGE N U M B E R ignoring this phase split is minimal as shown in Figure 10.
Figure 11. Water (1)-acetone (2)-furfural(3) UNIFAC-VLE ther- The example shown in Figure 11 was first proposed by
modynamics. Kinoshita et al. (1983) and later by Vickery and Taylor
(1986) for a single liquid phase only. The problem involves
removing water from acetone in the presence of furfural.
- THREEPHASE Vickery and Taylor (1986) have noted that the furfural
----- TW'OPWSE tends to concentrate almost exclusively within the bottom
section of the column, and this has been seen to adversely
affect convergence. In fact, when phase splitting is con-
sidered, the reboiler is within the three-phase region. The
effect of ignoring this phase split results in some compo-
nent profile deviations around the lower four stages (Figure
11).
0.00
&I I 1 2 3 4 5 6 7 8 9 10 The differences in the predictions become more im-
portant as the number of stages within the three-phase
STAGE NUMBER

Figure 12. Acetone (1)-chloroform (2)-water (3) UNIFAC-VLE region increase. Figure 12 shows the two- and three-phase
thermodynamics. solution for the acetone-chloroform-water problem in part
2 (Table 11,number 7) in which the entire column is within
and care should be taken before accepting a two-phase the three-phase region. Although the two-phase solution
solution. generally follows the three-phase profiles, large differences
To illustrate these differences, Figures 10-14 show the are apparent toward the top of the column.
results obtained by NSHET for various separations when The last example, Figure 13, which has also been in-
phase splitting was ignored (two phase) and when it was vestigated by Baden (1984), shows marked differences that
considered (three phase). Figure 10 presents the compo- can result when liquid-phase splitting is ignored in azeo-
nent profiles for the column proposed by Kistenmacher tropic distillation. If the two-phase solution is analyzed,
(1982) to desulfurize methanol in the presence of trace it is found that only the top stage is unstable and should
hydrocarbons. The solution shown is for a column without be split to form two liquid phases. This compares with
side streams and with a total condenser, whereas Baden the three-phase solution where the top 17 stages have two
(1984) solves the problem with two side streams and a liquid phases. Furthermore, as shown in Figure 14, the
partial condenser. In both cases, two liquid phases are only temperature in the top section of the column for the
1386 Ind. Eng. Chem. Res., Vol. 29, No. 7 , 1990

r YO\ I 1 c I
- -MOD UKIFAC
UNlFAC.VLE

NRTL

1
0 I 2 3 4 5 6 7 x q 10 I
S T A G E NUMBER

Figure 15. 1-Propanol (1)-water (2)-l-butanol(3) 1-propanol pro- Figure 16. 1-Propanol (1)-water (2)-l-butanol (3) water profile.
file.

two-phase solution is actually infeasible, as it is below the

predicted azeotropic minimum temperature. This high- UNIFAC-VLE
lights the importance of performing the full three-phase
calculations, rather than relying on the profiles that can MOD A S K
NRTZ
be generated by ignoring the liquid-phase split.
Effect of Thermodynamic Model -
02
The variety of different activity coefficient models
available raises concern over which model will provide I
0 I 2 3 4 5 6 7 X 9 10
accurate equilibrium predictions for use in distillation STAGE NUMBER
I

calculations. To quantify expected errors, the usual pro- Figure 17. 1-Propanol (1)-water (2)-l-butanol(3) 1-butanol profile.
cedure is to compare the predictions of a particular model
and parameter set to experimentally measured equilibrium
data and arrive at a mean deviation. This, however, can 95 0
be misleading, especially for highly nonideal systems, be-
cause it is possible that the overall mean error may appear
acceptable and yet some specific regions of the concen-
tration space may be relatively poorly predicted. Furzer -
(1988) has made a critical examination of the predicted 3
s
910

VLLE data in the system ethanol (1)-water (2)-benzene 8910

(3) using modified UNIFAC and the published experi- r:
mental data for this system. e 900

Cairns and Furzer (1988) showed that for the metha- 89 0

nol-acetone-chloroform system different thermodynamic
8 %0
models could place the same overall feed mixture in dif-
ferent distillation regions. The result was that very dif- 87 0

ferent product compositions were predicted for the same

feed composition. This sensitivity was due to relatively 0 I 2 3 4 s 6
STAGE " 4 B E . R
7 8 9 10

poor correlation of experimental data by some of the

Figure 18. 1-Propanol (1)-water (2)-l-butanol (3) temperature
models for dilute methanol concentrations, even though profile.
the overall mean deviations were of the order of 0.02 mole
fraction.
The problem for three-phase distillation is further
complicated because of the liquid-phase split. The scarcity
of VLLE and experimental three-phase distillation data
make comparisons difficult. Table I in part 2 shows that
the most often attempted three-phase distillation problem
in the literature has been the 1-butanol-water-1-propanol
example given by Block and Hegner (1976), which has
commonly been solved by using the NRTL model. To
examine the predictions of other models, NSHET was used
to solve the problem using UNIFAC-VLE (Gmehling et
al., 1982),UNIFAC-LLE (Magnussen et al., 1981), ASOG
(Kojima and Tochigi, 1979), modified UNIFAC (Larsen
et al., 19871, modified ASOG (Tochigi et al., 19811, UNI-
QUAC (Abrams and Prausnitz, 1975),and NRTL (Renon -
and Prausnitz, 1968). UNIQUAC and NRTL parameters C

were taken from Gmehling and Onken (1977) and Antoine
coefficients were taken from Sinnott et al. (1983).
The results for the system propanol (1)-water (21-1-
butanol (3) for all models are shown in Figures 15-19. The
composition predictions of UNIFAC-VLE and modified
UNIFAC were virtually the same, although the tempera-
ture profiles show that modified UNIFAC predicted con-
Figure 19. Comparison of predicted number of stages with three
phases.
sistently higher temperatures than UNIFAC-VLE.
UNIFAC-LLE, however, provided very poor composition
profiles and also predicted much higher temperatures than
the other models. This is probably to be expected, as the
 Ind. Eng. Chem. Res., Vol. 29, No. 7, 1990 1387

Table 11. Predicted Absolute Mean Vapor Composition and

Temperature Errors for I-Propanol-l-Butanol-Water
mean comp - .
error, mole mean temp g00.25 -
model fraction error, K d3 0.20 -.
UNIFAC-VLE 0.0054 0.342
UNIFAC-LLE 0.0390 2.753 a .
modified UNIFAC 0.0033 0.463 0.10 -
ASOG 0.0156 0.287
modified ASOG 0.0077 0.680 0.05 -
NRTL 0.0118 0.628 -
"M
UNIQUAC 0.0126 0.499 Y.W
0.6 0.62s 0.65 0.67s 0.7 0.725 0.75 0,775 0.8
WATER (mole fraction)

Figure 21. 1-Propanol (1)-1-butanol (2)-water (3) VLLE predic-

tions and vapor compositions.

$0.35
Y
-

h
0.10
-
0 EWWUMENTAL
IJNIPAC-VLE
---- UNIFAC-LLB
MODIFIED W A C '\
f
3
0.30

.25 -
E0 0.25
g 0.30
.
1
0 05

000
06 0625 065 067s 07 0725
\', \ I
,
07s 077s
,<,J

08
b
0.10
'
-
---- "IQUAC
WATER (mole fraction)
0.05 -
Figure 20. 1-Propanol (1)-1-butanol (2)-water (3) VLLE predic-
0.00
tions and vapor compositions.

UNIFAC-LLE parameters are only recommended for Figure 22. 1-Propanol (1)-1-butanol (2)-water (3) VLLE predic-
10-40 O C . tions and vapor compositions.
Modified ASOG appeared to produce composition pre-
dictions that differed at the ends of the column compared
to ASOG and the other models. The temperature pre-
dictions of the ASOG models (Figure 18) also showed some - 6 .
variations with approximately 1.5 "C difference at the 5 0.20 -

column top. UNIQUAC and NRTL predicted very similar g2 0.15 L-

results which essentially matched those of UNIFAC, al- 0 .
though some differences in the temperature predictions 0.10 -
are evident.
-
The prediction of the number of stages within the 0.05

three-phase region varied as shown in Figure 19. UNI- 003

FAC-LLE indicated the whole column would have only a 0.4 0.5 06 07
WATER (mole fraction)
08 09 I0

single liquid phase, whereas ASOG, modified ASOG, UN- Figure 23. 1-Propanol (1)-1-butanol (2)-water (3) bubble point
IQUAC, and NRTL all predicted that 9 of the 11 stages binodal curve and UNIFAC thermodynamics.
would have two liquid phases. UNIFAC-VLE indicated
that six stages would be in the three-phase region, and
modified UNIFAC only predicted that the bottom four
stages would have two liquid phases. 2025 -
5 .
These differences and the prediction of the temperature g0.20 -
and composition profiles are difficult to accept without 4 .-
experimental data. A search of the literature reveals that $015
0 .
some VLLE measurements have been made for this system
by Newsham and Vahdat (1977). The authors present Eo10 -
EXPERIMENTAL
three-phase measurements as well as some two-phase data 005 -
just outside the binodal curve. These data have been used
to examine the predictions of each of the models, and the 0.03
0.4 0.5 06 0.7 0.8 0.9 1.0
absolute mean vapor composition and temperature errors WATER (mole fraction)
over the experimental range are given in Table 11. A Figure 24. 1-Propanol (1)-1-butanol (2)-water (3) bubble point
comparison of the experimental vapor compositions and binodal curve and ASOG thermodynamics.
the predicted values for each model is shown in Figures
20-22. also provided good estimates and produced a mean tem-
The results of Table I1 show that UNIFAC-LLE pro- perature error less than modified UNIFAC (Table 11). The
vided a very poor correlation of the experimental data with predictions of the ASOG models, as well as UNIQUAC and
a mean composition error of approximately 4.0 mol % and NRTL, were not as good as UNIFAC-VLE and modified
a mean temperature error of 2.7 "C. This highlights the UNIFAC, although modified ASOG provided a mean
restriction on these parameters and corresponds to the composition error of only 0.77 mol 70and ASOG provided
poor column profiles shown in Figures 15-18. The most the most accurate temperature estimates at only 0.287 K
accurate of the models appears to be modified UNIFAC, mean error.
which produced excellent agreement with the experimental These results, however, do not explain the differences
data over the range shown in Figure 20. UNIFAC-VLE in the number of predicted three-phase stages as shown
1388 Ind. Eng. Chem. Res., Vol. 29, No. 7, 1990

V = 708.4 v = 122.10
Y1 = 0.4643
Y2 = 0.1731 R = 10 R = 10 Y1 = 0.5245
Y3 = 0.3626 Y2 = 0.1498
1498
,I I 4 T-69.06C I
I
I
rd = 0.3257
Y3 u.3257

P o l 01.325 kPa 101.325

49.05

0.05 - @..@@ PRASE1 I X3

-- 0.2581
0.07371 I

0.00
0.4 05 0.6
a
a @
.

0.7
. . .

0.8
.

0.9 .o
1
x2
x3
= 0.11
= 0.00
-Li\T_E2
I
3.103 x
2 PRASE
STRIPPII
WATER (mole fraction) x1 = 0.0 COLUMN
x2 = 0.0
Figure 25. 1-Propanol (1)-1-butanol (2)-water (3) bubble point
PHASE x3 =
binodal curve and UNIQUAC and NRTL thermodynamics. AZEOTROPIC
COLUMN 122.10

in Figure 19. To illustrate the discrepancy in the model

predictions, Figures 23-25 show the binodal curves a t the
bubble point predicted by each of the models. All models
overestimate the three-phase region. For the UNIQUAC,
37*95
NRTL, and ASOG models, the errors are extreme, with X1 = 0.7102

NRTL predicting that the two-liquid-phase region extends

up to nearly 0.3 mole fraction of 1-propanol compared to
the experimental observation of only approximately 0.06
mole fraction of propanol. This corresponds to the sim-
ulation results which indicate that nine stages are in the
three-phase region. The predictions of modified UNIFAC
again appear to be the most accurate (see Figure 23), while
UNIFAC-LLE provides a better estimate of the three-
phase region than UNIFAC-VLE.
It is important to realize that these differences in the
predicted size of the three-phase region have occurred even
though all models, except UNIFAC-LLE, provided rea-
sonably accurate predictions of the experimental vapor
compositions and bubble point temperatures. This illus-
trates the importance of not merely relying on the com-
parison of experimental and predicted vapor compositions
for three-phase systems. This has particular relevance
considering that the butanol-water-propanol example of
Block and Hegner (1976) has been used extensively in the
literature to illustrate what effect an aqueous phase dra-
woff would have on the column separation. These results
would be meaningless if the thermodynamic model did not
accurately predict the extent of the three-phase region.
Moreover, it must be stressed that the NRTL and UNI-
QUAC parameters given by Gmehling and Onken (1977)
are correlated against the VLE data of Dawe et al. (1973)
and not that of Newshaw and Vahdat (1977). However,
Gmehling and Onken (1977) show that the mean vapor
error for these data for both models is approximately 0.014
mole fraction, while the temperature error is approximately
1.4 K. These numbers are similar to those given in Table
11.
This underscores the danger of using these measures as
a selection criterion for choosing a model to be used in
potential three-phase distillation calculations, as there is
no information on the accuracy of the predicted three-
phase region.
Ethanol Dehydration with Isooctane
Part 1 by Cairns and Furzer (1990a) contains experi-
mental data on the separation of ethanol and water by the
addition of a mixture consisting of 11octane isomers. The
simulation considered in this section is with a single ad-
ditional component, isooctane or 2,2,4-trimethylpentane.
The flow sheet for isooctane recovery and recycle and the
compositions of the various streams is shown on Figure 26
and follows the detail given by Furzer (1986).
The results were obtained by using the program NSHET
and UNIFAC-VLE thermodynamics. The initial flow rates
Table 111. HeteroazeotroDesUsing UNIFAC-VLEO
composition, mole fraction
component overall phase I phase I1
1. Ethanol-Water-Benzene: Temperature = 64.03 "C;Phase I
Fraction = 0.3842
ethanol 0.2706 0.4504 0.1585
water 0.1943 0.4773 0.0177
benzene 0.5351 0.0723 0.8238
2. Ethanol-Water-2,2,4-Trimethylpentane: Temperature =
69.05 OC;Phase I Fraction = 0.6387
ethanol 0.4584 0.6567 0.1077
water 0.1786 0.2764 0.0056
2,2,4-TMC, 0.3630 0.0669 0.8867
a Antoine coefficients from Sinnott et al. (1983).
were calculated by an approximate mass balance assuming
that D was a t the heteroazeotrope and F2 had the com-
position of the corresponding entrainer phase (Table 111).
Thirty ideal stages, including the reboiler, were assumed
for the azeotropic column, and 20 stages were used for the
stripping tower. The ethanol-water feet stream (PI)was
introduced to the azeotropic tower on the fourth stage from
the top, and the decanter entrainer phase was refluxed to
the top stage. The aqueous phase from the decanter was
fed to the top stage of the stripping column. A relatively
high reflux ratio of 10 was assumed for both columns, and
a 0.5-kPa pressure drop per stage was included.
The mass balance for the complex was closed by exam-
ining the decanter products after each simulation. Initially,
the flow rate and composition of the overhead stream
entering the decanter from the stripping column were
assumed, and the azeotropic tower and decanter were
simulated. The distillate flow was adjusted to ensure that
nearly pure ethanol was obtained for a constant F2. The
resultant entrainer-phase reflux flow rate and the com-
position from the decanter were compared with the as-
sumed values for F2, and F2was adjusted accordingly. The
aqueous phase from the decanter was then used as feed
to the stripping column, which was simulated for a bottoms
flow rate that preserved the overall system mass balance.
The results from this simulation provided a new estimate
of the overhead stream entering the decanter. The azeo-
tropic column and decanter were then simulated with these
 Ind. Eng. Chem. Res., Vol. 29, No. 7,1990 1389
85 I
I I
0.9

0.8

0.7

h
g 0.6
-3
l! I
A 0.5 I

z
I
I
II

$
W

0.4

ELrl 0.3

0.2 0.8
2

0.1 8< 0.7

0.6
E
3 0.5
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
2.2.4-TRIMETHYLPENTANE (mole fraction)
1.0 '
0
0.4

Figure 27. Mass balance construction.

1.0 I I

0.0' O L ' . ' . ' . " ' . ' . ' .

0 2 4 6 8 10 12 14 16 18
PLATE NUMBER
z
Figure 30. Stripping column component profile.
8 O.'

23 OA
0.5
0
0.4

0.3

0.2 L i \
0 2 4 6 8 10 12 14 16 18 20 22 24 26 28
PLATE NUMBER
Figure 28. Azeotropic column component profiles.

new updates, and the process was repeated several times

until the mass balance was satisfactorily closed. t ! I
" ~ ' ~ ' " ~ ' ~ ' ~ ' ~ ' .
The results show that approximately70% of the ethanol 65
0 2 4 6 8 10 12 14 16 18
is recovered by the azeotropic tower in high purity. The PLATE NUMBER
bottoms of the stripping tower, therefore, contain signif- Figure 3 1. Stripping column temperature profile.
icant quantities of alcohol (0.71 mol %) that would have
to be recovered by another unit. stripping column (V); see Figure 27. The decanter bubble
Figure 27 shows that the distillate composition from the point temperature, therefore, is close to the heteroazeo-
azeotropic column (D)is very close to the heteroazeotrope. tropic temperature (see Figure 26 and Table 111).
Furthermore, Figure 27 shows that the entrainer reflux Virtually no water appears in the ethanol product
phase from the decanter (F2)is also very close to the stream, while no entrainer appears in the bottoms from
corresponding phase of the heteroazeotrope. This the the stripping column. A small amount of isooctane is lost
direct result of the steep tie-line slope of the ethanol- in the bottoms from the azeotropic column, and this is
water-2,2,4-trimethylpentanesystem. This implies that replaced by the make-up stream shown in Figure 26.
the recovery from the column is close to the maximum The top 13 stages of the azeotropic tower were detected
possible for an 89 mol % ethanol feed using the scheme to have two liquid phases, while no phase splitting was
shown in Figure 26. predicted for the stripping column. Figures 28-31 show
The overall decanter composition (C) is near the dis- the component and temperature profiles for both units.
tillate composition (D)and, hence, the heteroazeotrope, The profiles for the azeotropic column (Figures 28 and 29)
because of the relatively small overhead flow rate from the show that the characteristic steep front is located toward
1390 Ind. Eng. Chem. Res., Vol. 29, No. 7 , 1990
1.o

z 0.9 c HFlTROAZEOTXOPE

2E
- --- REFLUX RATIO = 0 5
-.
-
- - REFLUX u n o = 5 o
0.8
_- REFLUXRATIO = ino

05
0 9 4 " "
no IO 20
"

30 40
'
50
'
60
'
70
'
80
REFLUX RATIO
"
90 loo
"

IIO no
"

120 140
'
10

Figure 34. Effect of reflux ratio on bottoms ethanol.

03' ' ' ' ' . " " " '

00 01 02 03 04 05 06 (1 7
2,2,4TMC5 MOLE FRACTION

Figure 32. Effect of reflux ratio on distillation path. Thirty stages,

feed stage 4.

00 IO 20 30 40 50 60 70 80 90 ion 110 120 i ? n 140 isn

REFLUX RATIO
Figure 35. Effect of reflux ratio on bottoms water.

<I 3

PLATE NUMBER
Figure 33. Effect of reflux ratio component profiles. Thirty stages,
feed stage 4.
the bottom of the column with a smaller front below the
feed stage. The entrainer composition is virtually constant
over the middle section of the tower at approximately 0.33
mole fraction. This profile is consistent with the obser-
vations of King (19711, where the water is stripped from
the ethanol in a pinch zone of relatively high entrainer
composition. The three-phase region, therefore, extends
down the column until the water and ethanol compositions
have attained values that allow the isooctane to exist in
a single phase. The temperature profile for the azeotropic
column (Figure 29) reflects the composition fronts and
shows that the top section is near the heteroazeotropic
temperature.
The profiles for the stripping column (Figures 30 and
31) show that all the entrainer is located in the top section
of the tower as expected. Furthermore, little separation
is achieved over much of the column, which indicates that
the tower is probably overdesigned and that a similar result
could be obtained with fewer stages and a lower reflux
ratio.
Effect of Reflux Ratio. The effect of the reflux ratio
on the distillation path of the azeotropic column for a
constant number of stages and feed plate location is shown
in Figure 32. As the reflux ratio is reduced, the distillation
path leaves the zero water axis of the ternary diagram and
the bottoms become contaminated with water. This effect
has also been observed for 2-propanol dehydration with
benzene by Bril' et al. (1977).
Figure 32 shows that at a very low reflux ratio of 0.5 the
stages above the feed are dominated by the entrainer reflux
"oo io 20 30 40 50 so 7.0 8.0 9.0 io0 1i.o 12.0 130 140 150
REFLUX RATIO
Figure 36. Effect of reflux ratio on bottoms 2,2,4-TMC5.
phase from the decanter. However, the figure shows that,
since the top plate is in the three-phase region, the vappor
composition lies on the vapor line above the heteroazeo-
trope. As the reflux ratio is increased, the composition of
the stages above the feed approaches the distillate com-
position (which lies on the vapor line), and the distillation
path moves toward the zero water axis.
The reason for this is that at low reflux ratios the level
of entrainer in the column is not enough to exaggerate the
differences in volatility between ethanol and water. This
is illustrated by the component profiles shown in Figure
33. At high reflux ratios, the pinch zone has a relatively
high level of isooctane, and this enables water to be the
first component stripped from the ethanol. However, as
the reflux ratio is decreased, the level of isooctane in the
pinch zone is reduced, and eventually the difference be-
tween the volatilities of ethanol and water is not enough
to strip the water from the system in the given number
of stages. The entrainer becomes the first component to
be removed from the ethanol, and the separation is re-
duced to a binary ethanol-water split on the ethanol side
of the azeotrope. This can also be seen in Figure 32 by
the way the inflection point in the distillation path, which
corresponds to the pinch zone, increases in isooctane
composition for increasing reflux ratios.
Perhaps, as expected, there is an upper limit to this
effect. This is shown in Figures 34-36, which illustrate the
change in the bottoms composition for increasing reflux
ratio. Above a reflux ratio of 10, no change in the com-
 Ind. Eng. Chem. Res., Vol. 29, No. 7, 1990 1391

d L I

650 2 4 6 8 IO 12 14 16 18 20 22 24 26 28 30
PLATE NUMBER

Figure 37. Effect of reflux ratio on temperature profile.

"0 1 2 3 4 5 6 7 8 ,9 1 0 1 1 1 2
FEED PLATE

Figure 41. Effect of feed plate on the number of plates with two
liquid phases.

100

:F
&I00

2Eo99 .-
d
00 10 20 30 40 50 60 70 80

REFLUX RATIO
90 100 110 120 130 1 4 0 I50
3099 --
t ; .-
Figure 38. Effect of reflux ratio on the number of plates with two 098
liquid phases.
098 -

0 91

045
I
036 03604 03608 03612 03616 0362 03624 03628 03632 03036 0361
2,2,4TMCS MOLE FRACTION IN DISTILLATE
Figure 39. Approach of distillate toward heteroazeotrope with in-
creasing reflux ratio.

1.0

-
FEEDPLATE = I
FEEDPLATE I IO
f ..- OOOO 0
O'Oo5 i\ 1 2 3 4

Figure 43. Effect of feed plate on bottoms water.

5 FEED6 PLATE
7 8 9 IO 11 12

0.8

0.7

$ 0.6

2 O,5
E 0.4

0.3

0.2

0.1

0.0 0 . W
0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 0 1 2 3 4 5 6 1 8 9 IO II
P L A T E NUMBER FEED PLATE

Figure 40. Effect of feed plate on component profiles. Thirty Figure 44. Effect of feed plate on bottoms 2,2,4-TMCb
stages, reflux ratio = 10.0.
when the reflux ratio is reduced due to the increasing
position is obtained and the bottoms remain virtually pure presence of water. Furthermore, the increase of water in
in ethanol. the column a t low reflux ratios results in more stages in
The effect of reflux ratio on the temperature profile of the three-phase region as shown in Figure 38. For reflux
the azeotropic column is shown in Figure 37. This illus- ratios of less than 4, the top 75% of the column has two
trates that the temperature front moves down the column liquid phases.
1392 Ind. Eng. Chem. Res., Vol. 29, No. 7 , 1990
Finally, Figure 39 shows the approach of the distillate
composition along the vapor line toward the heteroazeo-
trope for increasing reflux ratios. A feature of this figure
is that very little change in the distillate composition occurs
because of the constraint of the vapor line.
Effect of Feed Plate Location. The effect of altering
the stage at which the ethanol-water feed is introduced
to the azeotropic column is shown in Figures 40-44 for a
reflux ratio of 10 and 30 ideal stages.
Figure 40 shows the component profiles for feed stages
at 1 and 10. When the feed is introduced at the top stage,
the bottoms become contaminated with water. The col-
umn appears to be pinched with very little separation for
the top 20 stages. Eventually the water composition begins
to decrease; however, the entrainer is the first component
to be removed, leaving the ethanol with approximately 2
mol 70water. As the feed stage is lowered, the water front
moves up the column and eventually becomes the first
component to be stripped from the ethanol. This is il-
lustrated in Figures 42-44, which show the effect of the
feed plate location on the bottoms composition, Little
change is experienced for feed stages lower than plate four.
There is, however, a limit to what the feed plate can be
lowered. This is illustrated in Figure 41, which shows how
the number of three-phase stages varies with feed plate
location. The figure shows that a minimum occurs around
stage seven, and that as the feed stage is further lowered,
the number of three-phase plates increases. This is due
to the water front moving down the column and thus in-
itiating the phase split. This is probably to be expected
considering the feed stream contains 11mol % water and
indicates that water would contaminate the bottoms
product if the feed stage were too low.
Multiple Solutions
Et hanol-Water-Benzene, Magnussen et al. (1979)
were the first to report the presence of multiple steady-
state solutions for azeotropic distillation of the ethanol-
water-benzene system. Magnussen and co-workers found
that, depending on the initialization procedure, a constant
molal overflow Naphtali-Sandholm algorithm would con-
verge to two distinct solutions for the same column spec-
ifications.
The first of these solutions (solution 1)resembled the
classic profiles where water was the first component to be
stripped from the ethanol. A large pinch zone of benzene
and ethanol existed before the benzene was removed from
the bottoms product in a steep front in the stripping
section. The bottoms were found to be nearly pure in
ethanol, with only traces of water and benzene. The sec-
ond solution (solution 3) was calculated from a different
starting point to the first and had steep concentration
fronts in the rectifying section. Benzene was stripped first,
leaving an almost constant water composition throughout
the column. The bottoms were found to contain significant
quantities of water.
A third solution (solution 2) was also located and found
to lie between solutions 1 and 3. This solution had a
constant water composition over much of the column at
a level less than solution 2 and steep concentration fronts
in the rectifying section. Importantly, solution 2 could not
be located by using the Naphtali-Sandholm method and
was found only by using a stage-to-stage algorithm.
Magnussen et al. (1979) showed that the three solutions
existed over B narrow range of column specifications.
Outside this range, only solution 1 or solution 3 could be
found. Similar multiplicity was further reported by the
authors for ethanol-water-pentane, although none could
be found for toluene-heptane-2-butanone.
Prokopakis et al. (1981) also encountered multiple so-
lutions for the system 2-propanol-water-cyclohexane using
the algorithm of Ross and Seider (1981). Initial attempts
to confirm the solutions for ethanol-water-benzene, how-
ever, were frustrated by the variations in thermodynamic
model predictions. Magnussen et al. (1979) used the UN-
IQUAC model with parameters from Gmehling and Onken
(1977),which overestimated the slope of the liquid-liquid
tie lines. This forced Prokopakis and Seider (1983) to
abandon their search for multiple solutions because the
tie-line errors resulted in a violation of some of the con-
straints used in their minimization approach. Prokopakis
and Seider (1983) attempted to use the parameters from
Prausnitz et al. (1980) and failed to find any multiplicity.
Comparison of the results was further complicated by the
fact that Magnussen et al. (1979) had assumed constant
molol overflow, whereas Prokopakis and Seider (1983)
included the energy balance equations.
Very recently, however, both Venkataraman and Lucia
(1988) and Kovach and Seider (1987) have confirmed the
existence of multiple solutions for the tower proposed by
Prokopakis and Seider (1983). Venkataraman and Lucia
(1988) showed that three steady-state solutions can be
found by using the parameters of Prausnitz et al. (1980)
with an algorithm that includes the energy balance equa-
tions. The authors also obtained a qualitatively similar
region of multiplicity to that of Magnussen et al. (1979)
and further noted that solution 2 had a very small domain
of convergence. Further complexities of the solution are
given by Kingsley and Lucia (1988).
Kovach and Seider (1987)used a homotopy-continuation
method which detected a very narrow range of specifica-
tions in which five steady-state solutions could be located.
Over a slightly larger range, three solutions were possible.
Also noted by Kovach and Seider (1987) was that solution
2 existed near a limit point in the homotopy path, which
accounted for the small domain of convergence experienced
by Magnussen et al. (1979)and later by Venkataraman and
Lucia (1988).
Some explanations have been given to account for this
solution multiplicity. In an eloquent examination of the
problem, Venkataraman and Lucia (1988) showed that the
initial estimates of the column composition and temper-
ature can place the iterations in thermodynamic regions
that produce different results. This is due to the very flat
nature of the bubble point surface for heteroazeotropic
systems and the fact that for these systems there exists
regions that have significantly different liquid compositions
in equilibrium with similar vapor compositions. If the
initial estimate of the liquid in the column is spread over
these regions, then a different solution may result com-
pared to an initial estimate wholly within one region.
Venkataraman and Lucia use these observations to ra-
tionalize how solutions 1 and 3 are obtained and explain
why solution 3 has a larger domain of convergence than
solution 1.
Earlier, Van Dongen et al. (1983) postulated that ig-
noring possible liquid-phase splitting may be linked to the
solution multiplicity. Van Dongen and co-workers showed
that a pseudohomogeneous liquid phase exhibits a spurious
homogeneous azeotrope, which has a temperature lower
than that of the true heterogeneous azeotrope. Further-
more, the authors showed that neglecting phase splitting
admits regions where multiple metastable equilibria can
be calculated.
To explore the relationship of phase splitting and
multiple solutions, the program NSHET was used to in-
vestigate the original column proposed by Magnussen et
 Ind. Eng. Chem. Res., Vol. 29, No. 7, 1990 1393
Table IV. Multiple Solutions for Ethanol (1)-Water 80
(2)-Benzene (3)
ethanol in
78 -
/
_-- ,'
estimate
(toD/ bottom)
bottoms, -76 -
mole 2 .
estimate x1 x2 x3 method fraction solution
1 0.3 0.2 0.5 2 phase 0.99764 1 $74:
-
1.0 0.0 0.0 3 phase 0.92879 3 2 72
2 0.3 0.2 0.5 2 phase 0.99764 1
0.9 0.1 0.0 3 phase failed
3 0.5 0.5 0.0 2 phase 0.92879 3
0.9 0.1 0.0 3 phase 0.92879 3
4 0.3 0.4 0.3 2 phase 0.96827 2
0.9 0.1 0.0 3 phase failed
5 overall mixed feed 2 phase 0.99764 1
3 phase failed
6 2 " " " " " " " " " " " " " '
0 2 4 6 8 10 12 14 16 18 20 22 24 26
STAGE NUMBER
Figure 46. Multiple steady-state solution temperature profiles.
Ethanol-water-benzene UNIFAC-VLE thermodynamics.
stripping section, as shown in Figure 45. The bottoms
ethanol product is predicted to be nearly pure in ethanol,
and there is a large pinch zone of ethanol and benzene in
the middle section of the column. When phase splitting
was considered, however, NSHET converged to solution
3, which indicated steep fronts in the rectifying section and
a nearly constant water composition throughout the col-
umn. Consequently, the bottoms product is heavily con-
0.2 L '\\ \ taminated with water, as shown in Figure 45. Importantly,
WATER '*: I even though phase splitting was considered, this solution
has only a single liquid phase on all stages.
_-- \ 1
\ - -
.
0.0
0
--.
2 4 6 8 10 12 14 16 18 20 22 24
m
26
Observing the intermediate output for the three-phase
STAGE NUVBER approach to solution 3 revealed that during the course of
Figure 45. Multiple steady-state solution component profiles. the iterations unstable phases were detected on some of
Ethanol-water-benzene UNIFAC-VLE thermodynamics. the top stages. This forced the iterations away from so-
lution l and toward solution 3. Examination of solution
al. (1979) using UNIFAC-VLE thermodynamics and An- 1 shows that the top stage is unstable and would not exist
toine coefficients taken from Sinnott et al. (1983). as a single liquid phase. This implies that solution 1 is
Table IV summarizes the results obtained when phase fictitious and only exists because of the errors associated
splitting was ignored (two phase) and when it was con- with incorrectly assuming a single phase.
sidered (three phase). Initial estimates were linearly in- This is further reinforced in Figure 46, which shows the
terpolated between the top and bottom compositions temperature profiles for the three solutions, as well as the
shown, except for estimate five, which used the overall feed heteroazeotropic temperature of the system predicted by
composition on all stages. The temperature estimates for using UNIFAC-VLE and Antoine coefficients from Sinnot
all cases were linearly interpolated between 63 "C at the et al. (1983). The profile for solution 1 shows that the
top and 78 "C a t the bottom. The maximum change in temperature at the top stage is below the heteroazeotropic
temperature between iterations (AT-) was 10.0 "C, and temperature and, therefore, is impossible. If the bubble
the maximum composition change (AX-) was 0.10 mole point of the overhead vapor is calculated, the condensate
fraction. is found to have two liquid phases and a temperature
The results show that three distinct solutions were lo- slightly above the heteroazeotrope. This gives rise to the
cated, and the composition profiles for each are illustrated impossible situation of the top plate being cooler than the
in Figure 45. These profiles are consistent with those overhead condensate bubble point temperature, as shown
published by Magnussen et al. (1979); however, it should in Figure 46.
be pointed out that the solutions were obtained for a The fact that solution 1 had a temperature lower than
distillate rate of 64.0 and not 63.1 as used by the original the predicted heteroazeotrope tends to support the ob-
authors. Multiple solutions could not be found for this servation of Van Dongen et al. (1983). Indeed, if the
latter value, and therefore, it would appear that the region bubble point of the overhead for solution 1 is calculated
of multiplicity is slightly different for UNIFAC-VLE without phase splitting, the temperature is found to be
thermodynamics. This is consistent with the observations even lower at approximately 61 "C. These results are also
of Magnussen and co-workers who found that the region consistent with those of Figures 13 and 14, which show the
of multiplicity differed for the predictions of UNIQUAC two- and three-phase solutions for a similar column run
and NRTL. Apart from this alteration, all other column at slightly different specifications. In this case, a three-
specifications used to obtain the solutions shown in Figure phase solution existed, and when phase splitting was ig-
45 were the same as in the original paper by Magnussen nored, the predicted composition and temperature profiles
et al. were again very different. Furthermore, Figure 14 shows
Table IV shows that, for the initial conditions of esti- that the two-phase solution is infeasible because the top
mate one, the two- and three-phase methods converged to stage temperatures are below the predicted heteroazeo-
different solutions. When phase splitting was ignored, tropic temperature.
NSHET produced solution 1, which has the classic azeo- When the initial condition was changed to give slightly
tropic profiles with steep concentration fronts in the less ethanol and more water in the system (estimate 2,
1394 Ind. Eng. Chem. Res., Vol. 29, No. 7 , 1990

Table IV), the two-phase method again converged to so- Table V. Multiple Solutions for Ethanol (1)-Water
lution 1; however, the three-phase method failed to find (2b2.2.4-Trimethsloentane (3)
a solution. Observation of the iteration pattern for this ethanol in
estimate bottoms,
method showed that unstable phases were again detected (toD/ bottom)
but the calculations moved into a region where water was mole
estimate x1 x2 x3 method fraction solution
simultaneously concentrated at the top and bottom of the
column. As the iterations continued, the maximum com- 1 overall mixed feed 2 phase failed
position change was applied to both water and benzene 3 phase 0.99495 1
2 0.9 0.0 0.1 2 phase 0.91897 2
over much of the column, with the result that large sections 1.0 0.0 0.0 3 phase failed
of the column were either stripped or enriched in both of 3 solution 2 3 phase failed
the components in an erratic manner. These observations
appear to match those of Venkataraman and Lucia (19881,
who have rationalized such behavior by noting that, if
1.0 , I

during the course of an approach to a solution the com-

positions in the column move into regions that straddle
an azeotropic point, failure can occur because the com-
ponents exhibit very different volatilities in each section.
Initial estimate three had a higher level of ethanol and
water in the column than estimates one and two and
started the calculations with no benzene present in the
system. This resulted in both the two- and three-phase
methods converging to solution 3 as shown in Table IV.
As more benzene was added to the starting composition
(estimate four), the three-phase method again failed in a
simlar fashion to that described above. Interestingly,
however, the two-phase method converged to the so-called
middle solution 2. This intermediate solution is shown in
Figure 45, which illustrates that the steep fronts have
moved slightly from solution 3 toward solution 1 but re-
main in the rectifying section. This results in a constant
water composition throughout the column of slightly less
than in solution 3 and an ethanol product that is ap-
proximately 97 mol % pure. It is interesting to note that
solution 2 was obtained with the Naphtali-Sandholm
method of NSHET. Magnussen et al. (1979) could only
find this solution by using a stage-to-stage algorithm, and
Venkataraman and Lucia (1988) were forced to use simple
continuation in order to approach the solution from pre-
viously calculated solutions. It was noted, however, that
the domain of attraction to solution 2 was small, and slight
variations in the initial estimates produced either solution
1 or 3.
Close examination of the final results of solution 2 re-
veals that the top stage is unstable and would split to form
a benzene phase that would occupy approximately 22 mol
% of the original liquid. This has happened even though
the temperature profiles of Figure 45 suggest that the
solution is feasible. Observations showed that this ac-
counted for the failure of the three-phase method which
detected the instability and was forced into regions that
did not permit convergence.
The final starting point examined (estimate 5) was the
standard feed condition that places the overall feed com-
position for the column on all stages. Table IV indicates
that the two-phase method converged to solution 1,
whereas the three-phase method failed in a similar manner
as previously noted.
In summary, it would appear that only solution 3 is
feasible using UNIFAC-VLE thermodynamics. Solution
1and 2 are fictitious and only result when phase splitting
is ignored. Solution 1 not only has unstable phases on the
top stages but also produces thermodynamically impossible
temperatures.
It is interesting to note that there have been no reports
of multiple solutions for a column in which two liquid
phases exist. Indeed, it is not clear what stability checks
were applied to the previous findings of multiple solution.
Magnussen et al. (1979) state that "for all three solutions,
z
E O.'
06
E
y 0.5
0
E 0.4
\
2.2.4TMCS
0.2 - <,
0 2 4 6 8 IO 12 14 16 18 20 22 24 26 28
STAGE NUMBER
Figure 47. Multiple steady-state solution component profiles.
Ethanol-Water-2,2,4-TMCS UNIFAC-VLE thermodynamics.
the liquid phase compositions on all stages were checked
for miscibility, and it was verified that in all cases there
is only one liquid phase." However, it is not immediately
obvious if metastability was included in the stability test.
Furthermore, it is not clear whether the stability test was
only applied to the final solutions or during the actual
iterations. If the stability test used by Magnussen et al.
(1979) was only based on the convexity condition, for ex-
ample, then this could admit the multiple metastable
equilibrium conditions discussed by Van Dongen et al.
(1983). Comparison, however, with the above results is
difficult because Magnussen et al. (1979) used UNIQUAC,
whereas the solutions shown in Figures 45 and 46 are based
on UNIFAC-VLE.
Although not explicitly stated, Kovach and Seider (1987)
appear to have encountered a similar situation to that
described above. Kovach and Seider used the UNIQUAC
parameters of Prausnitz et al. (1980) to find 5 solutions
for the 41-stage column of Prokopakis and Seider (1983).
One of the solutions closely resembles solution 1, while
another was similar to solution 3. Kovach and Seider state
that when the three-phase algorithm of Ross and Seider
(1981) was initialized with solution 1, for a tight conver-
gence condition, the iterations diverged and eventually
converged to solution 3. The authors make no reference
to phase splitting during the calculations.
Ethanol-Water-2,2,4-Trimethylpentane.Motivated
by the results for benzene, a similar examination was un-
dertaken for the column designed using isooctane or
2,2,4-trimethylpentane. Several starting points were used,
and the results are summarized in Table V. Two solutions
were found, and the profiles are shown in Figure 47.
These results were obtained for initial temperatures line-
arized between 70 "C at the top and 80 O C a t the bottom,
ATMm equal to 10 "C, and AX- equal to 0.1 mole
fraction. All other specifications are shown in Figure 26.
Solution 1corresponds to the three-phase result that was
achieved from the overall feed initial condition (estimate

one, Table V); however, when phase splitting was ignored
from this starting point, the iterations failed to find a
solution. Observation of the iteration pattern showed that,
as before, the calculations appeared to become split across
the azeotrope.
When the initial condition was changed to give a starting
point rich in ethanol with no water and very little 2,2,4-
trimethylpentane (estimate two), the two-phase method
found solution 2. This solution resembles solution 3 for
the benzene problem with a relatively high constant water
composition throughout the column (see Figure 47). Ex-
amination of this solution showed that the predicted tem-
peratures were above the heteroazeotropic temperature;
however, the top stage was detected to just lie within the
three-phase region (0.22 mol % isooctane phase).
Conclusions
The simulation method developed in part 2 of this series
has been used to predict the separation of many of the
three-phase distillation examples found in the literature.
Various thermodynamic models were investigated, and if
phase splitting was ignored, then erroneous composition
and temperature profiles can be produced. The 1-
propanol-water-1-butanol example of Block and Hegner
(1976) was used to show that, even though thermodynamic
models can provide good estimates of the vapor compo-
sitions and bubble points, the extent of the three-phase
region is very poorly predicted. This means that attempts
to locate and withdraw an aqueous side stream are
meaningless if accurate three-phase data are not used.
A marked example of ethanol dehydration with 2,2,4-
trimethylpentane has been used to illustrate the impor-
tance of reflux ratio, feed plate location, and number of
stages on the separation. The multiple solutions for
azeotropic distillation reported in the literature were in-
vestigated and rationalized by the inclusion of a phase-
splitting algorithm in the simulation model. This analysis
showed that two solution sets could be determined for the
ethanol-water-2,2,4-trimethylpentanesystem.
Registry No. Ethanol, 64-17-5;2,2,4-trimethylpentane, 540-
84-1.
Literature Cited
Abrams, D. S.; Prausnitz, J. M. Statistical Thermodynamics of Liq-
uid Mixtures: A New Expression for the Excess Gibbs Energy of
Partly or Completely Miscible Systems. AIChE J . 1975,21 (l),
116.
Baden, N. Ph.D. Thesis, Danmarks Tekniske Hojskole, 1984.
Block, U.; Hegner, B. Development and Application of a Simulation
Model for Three-phase Distillation. AIChE J . 1976,22(3), 532.
Boston, J. F.; Shah, V. B. An Algorithm for Rigorous Distillation
Calculations with Two Liquid Phases. Paper presented at the
AIChE Meeting, Houston, TX, April 1979.
Bril', Zh. A.; Mozzhukhin, A. S.; Petlyuk, F. B.; Serafimov, L. A.
Investigation of Optimal Conditions of Heteroazeotropic Rectifi-
cation. Theo. Found. Chem. Eng. 1977,11 (6),675.
Buzzi Ferraris, G.; Morbidelli, M. An Approximate Mathematical
Model for Three-phase Multistaged Separators. AIChE J . 1982,
28 (l),49.
Cairns, B. P.; Furzer, I. A. Sensitivity Testing with the Predictive
Thermodynamic Models NRTL, UNIQUAC, ASOG and UNIFAC
in Multicomponent Separations of Methanol-Acetone-Chloro-
form. Chem. Eng. Sci. 1988,43 (3), 495.
Cairns, B. P.; Furzer, I. A. Multicomponent Three Phase Azeotropic
Distillation. 1. Extensive Experimental Data and Simulation
Results. Ind. Eng. Chem. Res. 1990a,first in a series of three in
this issue.
Cairns, B. P.; Furzer, I. A. Multicomponent Three Phase Azeotropic
Distillation. Phase-Stability and Phase-Splitting Algorithms.
Ind. Eng. Chem. Res. 1990b,second in a series of three in this
issue.
Ind. Eng. Chem. Res., Vol. 29, No. 7, 1990 1395
Dawe, R. A.; Newsham, D. M. T.; Ng, S. B. Vapour-Liquid Equi-
libria in Mixtures of Water, n-Propanol, and n-Butanol. J . Chem.
Eng. Data 1973,18 (l),44.
Furzer, I. A. In Distillation for University Students; Furzer, I. A,,
Publisher; Department of Chemical Engineering, University of
Sydney: Sydney, Australia, 1986.
Furzer, I. A. Predicted VLLE Data Using Modified UNIFAC,
CHEMECA 88,16th Australian Chemical Engineering Confer-
ence, 1988.
Gmehling, J.; Onken, U. Vapour-Liquid Equilibrium Data Collec-
tion. DECHEMA Chem. Ser. 1977,1.
Gmehling, J.; Rasmussen, P.; Fredenslund, Aa. Vapour-Liquid
Equilibrium by UNIFAC Group Contribution. Revision and Ex-
tension 2. Ind. Eng. Chem. Process Des. Deu. 1982,21,118.
King, J. C. Separation Processes; McGraw-Hill: New York, 1971.
Kinglsey, J. P.; Lucia, A. Simulation and Optimization of Three-
Phase Distillation Processes. Ind. Eng. Chem. Res. 1988,27,1900.
Kinoshita, M.; Hashimoto, I.; Takamutsu, T. A. Simulation Proce-
dure for Multicomponent Distillation Column Processing Noni-
deal Solutions or Reactive Solutions. J . Chem. Eng. Jpn. 1983,
16 (6),515.
Kistenmacher, H. Private Communication as cited by Baden (1984),
1982.
Kojima, K.; Tochigi, K. Prediction of Vapour-Liquid Equilibria by
the ASOG Method; Elsevier: Amsterdam, 1979.
Kovach, J. W., 111; Seider, W. D. Heterogeneous Azeotropic
Distillation-Homotopy-Continuation Methods. Comput.
Chem. Eng. 1987,11 (6),593.
Larsen, B. L.; Rasmussen, P.; Fredenslund, Aa. A Modified UNIFAC
Group-Contributions Model for Prediction of Phase Equilibria
and Heats of Mixing. Ind. Eng. Chem. Res. 1987,26,227.
Magnussen, T.; Michelsen, M. L.; Fredenslund, Aa. Azeotropic Dis-
tillation using UNIFAC. Inst. Chem. Eng. Symp. Ser. 1979,56,
4.211.
Magnussen, T.; Rasmussen, P.; Fredenslund, Aa. UNIFAC Param-
eter Table for Prediction of Liquid-Liquid Equilibria. Ind. Eng.
Chem Process Des Dev. 1981,20,331.
Newsham, D. M. T.; Vahdat, N. Prediction of Vapour-Liquid-Liquid
Equilibria from Liquid-Liquid Equilibria. Part I: Experimental
Results for the Systems Methanol-Water-n-Butanol, Ethanol-
Water-n-Butanol and n-Propanol-Water-n-Butanol. Chem. Eng.
J . 1977,13, 27.
Pratt, H. R. C. Continuous Purification and Azeotropic Dehydration
of Acetonitrile Produced by the Catalytic Acetic Acid-Ammonia
Reaction. Trans. Inst. Chem. Eng. 1947,25, 43.
Prausnitz, J. M.; Anderson, T.; Grens, E.; Eckert, C.; Hsieh, R.;
O'Connell, T. Computer Calculations for Multicomponent Vap-
our-Liquid and Liquid-Liquid Equilibria; Prentice-Hall: En-
glewood Cliffs, NJ, 1980.
Prokopakis, G. J.; Seider, W. D. Feasible Specifications in Azeotropic
Distillation. AIChE J. 1983,29 (l),49.
Prokopakis, G. J.; Seider, W. D.; Ross, B. A. Azeotropic Distillation
Towers with Two Liquid Phases. Foundations of Computer-
Aided Chemical Process Design; Mah, R. S. H., Seider, W. D.,
Eds.; AIChE: Washington, DC, 1981;Vol. 2,p 239.
Renon, H.; Prausnitz, J. M. Local Compositions in Thermodynamic
Excess Functions for Liquid Mixtures. AIChE J . 1968,14(11,135.
Ross, B. A. Simulation of Three-phase Distillation Towers. Ph.D.
Dissertation, University of Pennsylvania, University Park, 1979.
Ross, B. A.; Seider, W. D. Simulation of Three-phase Distillation
Towers. Comput. Chem. Eng. 1981,5 (l),7.
Sinnott, R. K.;Coulson, J. M.; Richardson, J. F. Chemical Engi-
neering; Pergamon, Press: New York 1983;Vol. 6.
Tochigi, K.; Lu, B. C.-Y.; Ochi, K.; Kojima, K. On the Temperature
Dependence of ASOG Parameters for VLE Calculations. AIChE
J . 1981,27 (61,1022.
Van Dongen, D. B.; Doherty, M. F.; Halght, J. R. Material Stability
of Multicomponent Mixtures and the Multiplicity of Solutions to
Phase-Equilibrium Equations. 1. Non-Reaction Systems. Ind.
Eng. Chem. Fundam. 1983,22,412.
Venkataraman, S.; Lucia, A. Solving Distillation Problems by New-
ton-Like Methods. Comput. Chem. Eng. 1988,lZ(l),55.
Vickery, D. J.; Taylor, R. Path-Following Approaches to the Solution
of Multicomponent, Multistage Separation Problems. AIChE J .
1986,32 (41,547.
Received f o r review May 17,1989
Revised manuscript received October 4, 1989
Accepted October 27, 1989