Ind. Eng. Chem. Res.

1990, 29, 1349-1363 1349

Multicomponent Three-phase Azeotropic Distillation. 1. Extensive

Experimental Data and Simulation Results
Brett P. Cairns and Ian A. Furzer*
Department of Chemical Engineering, University of Sydney, Sydney, New South Wales, Australia 2006

The first comprehensive data set of compositions of a 13-component mixture on each tray of a
distillation column operated in the 3-phase region is presented in this paper. The design of the
experiments has been aimed a t covering all the significant regions of the composition space. Dis-
tillation columns producing a near heteroazeotrope as overheads can produce widely different bottom
products depending on the feed conditions. The composition profiles have been simulated with a
modified Naphtali-Sandholm method that includes a phase splitting algorithm. A wide range of
group contribution methods has been critically examined for the evaluation of VLE and VLLE data.
A comparison of the experimental and predicted composition profiles along the column indicates
an overall column efficiency of 70%. Attempts to calculate Murphree vapor efficiencies require a
thermodynamic model and except for the major components lead to widely varying numerical values.
Overall column efficiencies are preferred for column simulation studies. The collected data set should
be valuable to test all three phase distillation models.

Three-phase azeotropic distillation is an important N

separation technique which finds wide use in industry. FxF,~= mnxn,i i = 1, ..., n, (2)
n=l
The process is so-named because the key azeotrope taken
overhead condenses to form two liquid phases. This where m, is the molar liquid holdup at each stage n. This
azeotrope is often called a heteroazeotrope, and the process assumes that the vapor holdup is negligible. Finally, the
is sometimes referred to as heteroazeotropic distillation. mole fraction summation for each stage must be held:
The overhead liquid-phase split raises the possibility of nc

a similar occurrence on the column trays, and therefore, Cy,,i = 1 n = 1, ...,N (3)
i=l
effective simulation of the process can only be accom-
plished by a three-phase distillation model. Equations 1-3 represent N ( n , + 1) equations in N(n, +
Despite the importance of azeotropic distillation and the 1) unknowns, Nn, liquid-phase mole fractions (x,,$, and
numerous other examples of three-phase distillation found N stage temperatures ( T J .
in industry, there appears to be virtually no experimental By use of the Naphthali-Sandholm (1971) approach of
tray-to-tray composition data available for columns op- Furzer (1986),the following deviation functions are written:
erating with two liquid phases. Kovach and Seider Fn,i = Y n + l , i - xn,i (4)
(1987a,b) have reported experimental results of the feed,
n,
distillate, and bottoms compositions but not the tray-to-
tray compositions in a three-phase distillation column. (5)
This, combined with scarcity of vapor-liquid-liquid
equilibrium (VLLE) data, makes the results obtained from
simulation models difficult to accept.
This paper presents an extensive set of tray-to-tray
multicomponent three-phase azeotropic experimental data (7)
at total reflux. The first comprehensive data set for the
system ethanol-water-2,2,4-trimethylpentanewas given These deviation functions equal zero at the solution.
by Furzer (1985). The new data in this paper complement Grouping the equations as given in eqs 4-7 gives the fol-
the previous work of Cairns and Furzer (1987a,b). The lowing block structure, which has also been presented by
system investigated was the dehydration of ethanol using Hirose et al. (1980):
an impure isooctane fraction, containing 11 components,
as an entrainer. The experimental results are compared
with a newly developed three-phase distillation model.
Further, the predictions of the currently available group
contribution activity coefficient models-UNIFAC-VLE,
UNIFAC-LLE, modified UNIFAC, ASOG, modified
ASOG-are critically assessed. Furzer (1988) has made
a critical examination of the extremely limited VLLE data
published and the modified UNIFAC model.
or
Simulation JAx = -F
A column operating at total reflux is depicted in Figure
1. At total reflux, it can be shown that “passing streams The Newton-Raphson scheme then updates the variables
are equal”; that is, Xk+l = xk + Ax k = iteration counter (10)
yn+i,l= x,,~ i = 1, ..., n, n = 1, ..., N - 1 (1)
and the process is terminated once the deviation function
Further, because no streams are withdrawn, the following residuals have reached a suitable tolerance. Maximum
restriction on the initial charge to the still must be satis- projection limits are applied to the corrections at each
fied: application of eq 10. The limits are user supplied as AX,
osss-~~a~~~o~~~0
~ 1990
~ - ~American
~ 4 9 $Chemical
0 ~ . ~ Society
0/0
1350 Ind. Eng. Chem. Res., Vol. 29, No. 7, 1990

This in effect reduces eq 8 to

1 Condenser

Stage 1
Bn Cn

BN-l cN-l
-

--

D2* D3 Dn DN-~ DN-

1 X
1,i 4 This process is carried out N - 1times to give the following

i
Stage 2 C1

1l!
B2 c
2

x2, i L*
B" C n
BN-l

v[:yN-2,i N-3,i LN-3

which allows the correction vectors Axn to be calculated
Stage N-2 by back-substitution using Gaussian elimination.
In general, eqs 18 and 19 can be written as
vN-l yN-1,i N-2,i
L
N-2 Dn+l*= Dn+l*- DnBn-lCn n = 1, ...,N - 1 (22)
FN*(n) = FN*(n-l)- DnB;lFn
(23)
Furzer (1986) points out that the inverse of the B matrix
N
' IyN,i 4 X
N-1,i LN-l does not have to be calculated since eqs 22 and 23 can be
written as
I Reboiler I Dn+l*= Dn+l- DnZn+l n = 1, ...,N - 1 (24)
1 FN*(n) = FN*(n-l) - DnWn+1 (25)
Figure 1. Total reflux column. where
and ATmm with suggested values of 0.2 mole fraction and
10 K, respectively, and if or
I&l> h m a x (11)
I A q > ATma (12) and
for any iteration the projections are reset to the maximum Wn+l = B,-lF,,
values. Furthermore, if the projections of the Newton-
Raphson scheme produce negative mole fractions, these or
are reset as small positive numbers, typically lo*. Cairns
and Furzer (198813) have shown that this technique pro-
vides a robust and flexible way of constraining the New- This allows standard Gaussian elimination to be used to
ton-Raphson iterations. solve eqs 26 and 27 and retrieve Zn+land Wn+l.
Equation 9 can be solved for Ax by using simple The reason that the sum of the y values is used in eq
Gaussian elimination. The row of D matrices are elimi- 3 and not the sum of the x values is that, if the latter is
nated by expanding eq 9 to give used, the B matrices have a column of zeroes due to the
temperature partial derivative. This arises because at total
Bl&l+ Clk2 = -F1 (13) reflux the component deviation functions do not involve
D1&1+ D2&2 + ... + DN&N = -FN (14) the vapor composition leaving the stage but rather the
vapor entering the stage, that is, leaving the stage below,
Multiplying eq 13 by the product DIB1-l,we have which contributes to the C matrices. Using eq 3 introduces
the temperature derivative via the equilibrium condition
Dl&i+ DIB1-lC1&P = -DlBl-lFl (15)
Y n i = Kn,ixnj (28)
Subtracting eq 15 from eq 14 gives
since Kni is a function of temperature. For an ideal gas,
+ D3h3 + ... + DN&N =
(D2 - D1B1-1C1)~2
pa
-(FN -.D1B1-'FJ (16) .A,
n,i 1 n,i
.

Kn,i = -
D2*&2 + D3&3 + ... + DN&N = -FN*(') (17)
P

where where P",i is evaluated by the Antoine equation and Y~~

is calculated by using an appropriate model. The models
D2* = D2 - DIBrlC1 (18) investigated in this work are UNIFAC (Fredenslund et al.,
1975) with parameters correlated against VLE, UNIFAC-
FN*(') = F N - DlBl-'Fl (19) VLE (Gmehling et al., 1982), and LLE; UNIFAC-LLE
 Ind. Eng. Chem. Res., Vol. 29, NO. 7, 1990 1351

Table I. Total Reflux Feed Compositions for

Methanol-AcetoneChloroform (Mole Fractions)
run methanol acetone chloroform
1 0.90 0.05 0.05

F. fi 2
3
4
0.83
0.05
0.05
0.12
0.40
0.38
0.05
0.55
0.57

D = DISTILLATE
.-8 0.8 B = BOlTOMS
0

02 -
01 -
" " "
00 . ' * " ' '
00 01 02 03 04 05 06 07 08 09 IO
CHLOROFORM (mole fraction)
Figure 3. Total reflux simulation, UNIFAC thermodynamics:
methanol (1)-acetone (2)-chloroform (3).

in Table I. A 25 ideal stage column, including reboiler,

operating at atmospheric pressure was used in the simu-
lation. The reboiler holdup was assumed to be 85% of the
overall mixture, and the remaining 15% was distributed
evenly over the other stages. The results are shown in
Figure 2. Bubble point temperature surface, methanol (1)-acetone Figure 3 for UNIFAC-VLE thermodynamics with Antoine
(2)-chloroform (3). Temperature-composition axes. A = pure
chloroform (61.2 "C), B = chloroform-acetone azeotrope (64.5 "C), coefficients taken from Sinnot et al. (1983).
C = pure acetone (56.4 "C), D = acetone-methanol azeotrope (54.6 The total reflux distillation paths shown in Figure 3
"C),E = pure methanol (64.7 "C), F = methanol-chloroform azeo- clearly illustrate the effect of the saddle azeotrope. Run
trope (53.5 "C). 1, for example, had a mixture that was in a zone that
produced a distillation path approaching pure methanol
(Magnussen et al., 1981);modified UNIFAC (Larsen et al., at the bottom and the methanol-chloroform azeotrope at
1987); ASOG (Kojima and Tochigi, 1979); and modified the top. On the other hand, run 2 was such that the top
ASOG (Tochigi et al., 1981). of the column approached the methanol-acetone azeotrope
while again producing methanol a t the bottoms. Runs 3
Phase Splitting and 4 illustrate the same effect for a bottom that ap-
The inclusion of liquid-phase splitting has been outlined proaches the maximum boiling binary acetone-chloroform
in part 2 (Cairns and Furzer, 1990). The liquid phase is azeotrope. Cairns and Furzer (198810) have recently used
checked for stability every time the activity coefficients this example to show that different thermodynamic models
are evaluated. Derivatives with respect to the average could place the same overall feed mixtures in different
liquid-phase mole fractions (f,,Jfor unstable phases are distillation regions and, as illustrated in Figure 3, predict
calculated analytically as described in part 2. very different distillation paths.
Examples. Two different azeotropic mixtures have Several workers, such as Hoffman (1964) and more re-
been simulated at total reflux to illustrate the scope of the cently Van Dongen and Doherty (1985) have noted that
method. The first, methanol-acetone-chloroform, repre- the simple distillation residue curve which passes through
sents a homogeneous mixture which exhibits a saddle the bottom composition of a column coincides, in part, with
azeotrope. The second, ethanol-water-2,2,4-trimethyl- some portion of the total reflux composition profile. In-
pentane, represents a three-phase azeotropic application deed, Maga et al. (1986) have reported experimental res-
which has relevance for the proposed experimental work. idue curves for the methanol-acetone-chloroform system
Methanol-Acetonexhloroform. Possibly the most that largely match those shown in Figure 3.
investigated multicomponent homogeneous azeotropic Ethanol-Water-2,2,4-Trimethylpentane. The bubble
mixture is methanol-acetone-chloroform, which was dis- point surface for the heteroazeotropic mixture ethanol-
covered by Ewe11 and Welch (1945). This system is in- water-2,2,4-trimethylpentane,based on UNIFAC-VLE, is
teresting because it exhibits a ternary saddle point azeo- represented in Figure 4. The dew and bubble point
trope as well as binary pair maximum and minimum surfaces would touch at the pure component corners of the
azeotropes. To illustrate how these azeotropes combine ternary diagram, as well as a t the azeotropes. The het-
to create distillation zones, Figure 2 shows a representation erogeneous azeotropes exhibit minimum lines of temper-
of the bubble point surface based on the predictions of ature rather than the minimum and maximum points, as
UNIFAC-VLE. The figure clearly illustrates the saddle shown by the homogeneous example in Figure 3. This is
point and the relationships of the binary azeotropes. illustrated by the binary heteroazeotrope between 2,2,4-
Four total reflux runs have been simulated in the dif- trimethylpentane and water. The small solubility of the
ferent distillation zones, with the feed compositions given two components is reflected in the constant temperature
1352 Ind. Eng. Chem. Res., Vol. 29, No. 7 , 1990

Figure 5. Expanded view of bubble point surface along heteroaz-

eotropic tie line, Temperature-composition axes.

+ - - - + LIQUD-LlQUlDTE.LwE
t-.OVERALL LIQUD-VAPOUR TIE-LNf
*- VAPOURLINE

Figure 4. Bubble point temperature, surface, ethanol (1)-water

(2)-2,2,4-trimethylpentane(3). Temperaturecomposition axes. A
= pure 2,2,4-trimethylpente (99.22 "C), B = 2,2,4-trimethyl- 00 0I 0.2 0.3 0.4 0.5 0.6 07
2.2.4.TRIMETHYLPEhTAhT (mole fraction)
OS 0.9 10

pentane-water azeotrope (78.8 "C), C = pure water (100.0 "C),D =

water-ethanol azeotrope (78.2 "C), E = pure ethanol (78.3 "0,F= Figure 6. Vapor line: ethanol (1)-water (2)-2,2,4-trimethylpnhne
ethanol-2,2,4-trimethylpentaneazeotrope (71.8 "C), G = ternary (3).
heteroazeotrope (68.95 "C).
pendent of the overall liquid composition as long as the
of the miscribility gap that sharply rises to the pure com- equilibrium liquid-phase compositions remain unchanged.
ponent temperatures at the diagram corners. This implies that a mixture that lies on any given liq-
The diagram also shows the "imprint" of the two-liq- uid-liquid tie line has the same vapor composition and
uid-phase region and that only very small changes in the bubble point temperature irrespective of its position on
bubble point temperature occur within this region. Large the line. If this is extended for all liquid-liquid tie lines,
increases in temperature occur at the binodal curve as the then the series of vapor compositions can be linked to form
mixture becomes homogeneous. what is called the "vapor line". This is illustrated in Figure
The ethanol corner of the diagram corresponds to the 6 for ethanol-water-2,2,4-trimethylpentaneusing UNI-
region in which an ethanol dehydration distillation column FAC-VLE.
would operate. The surface shows that, if the distillation Figure 6 shows different vapor compositions in equi-
path of the column approached the pure ethanol corner librium with various overall liquid compositions which lie
along the zero water axis, a steep temperature front would within the three-phase region. The line connecting the
be experienced in the bottom section of the column. On vapor points is the vapor line. Any mixture that lies within
the other hand, if the path approached along a constant the three-phase region would boil to produce a vapor
water composition line, the steep front would be at the top composition that lies on the vapor line. Therefore, large
of the column, leaving the bottom to slowly increase in differences in composition between the overall liquid and
temperature. vapor can be experienced for mixtures that exist just inside
The ternary heteroazeotrope is less obvious in Figure the binodal curve.
4 than the saddle azeotrope in Figure 3. This is because The vapor line intersects the water-2,2,4-trimethyl-
of the very small temperature gradients in the three-phase p e n m e axis at the binary heteroazeotropic point. Moving
region. Figure 5 shows an expanded view of the heter- away from this axis, the bubble point temperature of each
oazeotropic tie line in which the temperature difference ternary liquid-liquid tie line decreases, as shown in Figure
between the minimum heteroazeotropic plane and the 4, until the minimum heteroazeotropic tie line is reached.
maximum points shown is only 0.2 "C. The figure illus- The bubble point temperature then increases as illustrated
trates that the bubble point temperature is a minimum in the magnified view shown in Figure 5. The intersection
along this line. Only at the point where the dew point of the vapor line and the heteroazeotropic tie line defines
surface touches this line is the ternary mixture at the the heteroazeotropic point. For mixtures that lie on the
heteroazeotrope and the overall liquid composition equal heteroazeotropic tie line, the vapor composition would lie
to the vapor composition. on the vapor line as well as the liquid-liquid tie line. Only
A feature of three-phase mixtures is that the vapor at the heteroazeotropic point, however, will the overall
composition in equilibrium with two liquid phases remains liquid and vapor compositions be equal.
the same irrespective of the amount of each liquid phase. Although perhaps less apparent from the bubble point
Furthermore, the bubble point of the mixture is inde- surface shown in Figure 4, the heteroazeotropic system
 Ind. Eng. Chem. Res., Vol. 29, No. 7 , 1990 1353

Table 11. Total Reflux Feed Compositions for

Et hanol-Water-2,2,4-Trimethylpentane(Mole Fractions) t HETEROAZEOTROPE

run ethanol water 2,2,4-trimethylpente

______ BINODAL CURVE AT BUBBLE POINT
DISTILLATION PATH
1 0.940 0.005 0.055
2 0.950 0.025 0.025
3 0.850 0.125 0.025
4 0.500 0.475 0.025
5 0.050 0.925 0.025
6 0.010 0.965 0.025
7 0.010 0.025 0.965
8 0.040 0.010 0.950
02 -
ethanol-water-2,2,4-trimethylpentanealso has distillation
zones that control the end products from a column.
The total reflux method was again used to simulate the
distillation paths for a variety of different feed mixtures, 2.2.4TMC5 (mole fraction)

which are listed in Table 11. The, distillation paths are Figure 7. Total reflux simulations, UNIFAC-VLE thermodynamics:
shown in Figure 7 and were calculated for a five-ideal-stage ethanol (1)-water (2)-2,2,4-trimethylpentane (3).
atmospheric column, including reboiler, assuming 85%
reboiler holdup. UNIFAC-VLE thermodynamics and from stage to stage can result because of this requirement,
Antoine coefficients from Sinnott et al. (1983) were again which is shown by the paths for runs 7 and 8. These runs
used. have an overall feed composition rich in 2,2,4-trimethyl-
Figure 7 shows that, as expected, all distillate compo- pentane, and the total reflux paths show that the heter-
sitions tend toward the heteroazeotrope; however, de- oazeotrope can be approached along the vapor line from
pending on the feed composition, the bottoms approach either above or below depending on the amount of ethanol
different corners of the ternary diagram. Run 1, for ex- in the system (Table 11).
ample, had a high ethanol feeed content with only traces
of water, and the simulation predicts that the total reflux Experimental Section
column would approach pure ethanol at the bottom, The objective of this section of the work was to collect
leaving the top to tend toward the heteroazotrope. Sim- at total reflux an extensive set of data on three-phase
ilarly run 2 also approached the ethanol of the ternary azeotropic distillation. This was to include measurements
diagram; however, because more water and less 2,2,4-tri- of stage temperatures, liquid-phase compositions, and split
methylpentane was present in the system, the distillation ratios, which would allow distillation paths to be con-
path shows an approach along a constant water line which structed. The mixture investigated was ethanol, water, and
intersects the binary ethanol-water axis. an industrial isooctane (2,2,4-trimethylpentane) fraction
As the ethanol content in the feed mixture is lowered, referred to as “alkylate”. This mixture was chosen to
the total reflux paths show the influence of the binary examine the potential use of isooctane as an entrainer in
ethanol-water azeotrope, and the bottoms approach the the production of dehydrated ethanol. Two batches of the
pure water corner of the ternary diagram (runs 3 and 4). alkylate mixture were obtained and used in two distillation
For mixtures very low in ethanol and rich in water (runs columns.
5 and 6), the total reflux path approaches the heteroaz- Alkylate Analysis. An analysis utilizing high-per-
eotrope from below, as shown in Figure 7 . formance gas chromatography and mass spectrometry was
The distinguishing feature of heteroazeotropic systems performed to identify and quantify each component in the
is the liquid-phase split. Figure 7 illustrates the effect of alkylate batches. Details of the method can be found in
the vapor line and demonstrates how it essentially divides Cairns et al. (1987).
the composition space in half. The requirement that The first batch (batch 1)was found to contain 10 com-
passing streams be equal for total reflux means that the ponents ranging from C, to C9 with a boiling point range
distillation paths in the three-phase region must approach of 89.7-124.1 “C. The second batch (batch 2) was found
the heteroazeotrope along the vapor line. This is because to have one extra C, component with a boiling point of 80.5
any ideal stage that has two liquid phases must be in OC. The results of the analyses are summarized in Table
equilibrium with a vapor that lies on the vapor line, and 111.
since the liquid leaving the stage above is equal to the Distillation Columns. Two distillation columns were
vapor entering that stage, the liquid must also lie on the used in these experiments. The first was a 104-mm in-
vapor line. This is well illustrated in Figure 7 , which shows ternal diameter glass column containing nine stainless steel
that all runs tend toward the heteroazeotrope along the dual-flow trays at 320-mm tray spacing of 20.19% free area
vapor line. Furthermore, large changes in composition with 8.0-mm-diameter holes. The second was a 610-mm
Table 111. Alkylate Composition
batch 1, batch 2,
comDonent name symbol bp mole fraction mole fraction
1 2,4-dimethylpentane 2,4-DMCS 80.5 0.0087
2 2,3-dimethylpentane 2,3-DMC6 89.7 0.0105 0.0172
3 2,2,4-trimethylpentane 2,2,4-TMCS 99.2 0.6203 0.6800
4 2,5-dimethylhexane 2,5-DMC6 109.1 0.0453 0.0254
5 2,4-dimethylhexane 2,4-DMC6 109.4 0.0589 0.0422
6 2,2,3-trimethylpentane 2,2,3-TMCS 109.8 0.0127 0.0078
7 2,3,4-trimethylpentane 2,3,4-TMC, 113.4 0.1215 0.1237
8 2,3,3-trimethylpentane 2,3,3-TMCS 114.7 0.0346 0.0318
9 2,3-dimethylhexane 2,3-DMC6 115.6 0.0830 0.0579
10 3,4-dimethylhexane 3,4-DMCB 117.7 0.0062 0.0032
11 2,2,5-trimethylhexane 2,2,5-TMCe 124.1 0.0070 0.0021
1354 Ind. Eng. Chem. Res., Vol. 29, No. 7 , 1990

09

0.R
L
2n7
L
Y nh
2 0.5
0
04

n3
n2

0I
' I , t 8
0.0
0 1 2 3 4 5 6 7 8 9 10
ALKYLATE (mole fraction) P L A T E NUMBER

Figure 8. Summary of typical experimental distillation paths.

1.0 ,
internal diameter mild steel column containing four
stainless steel sieve trays with downcomers on a 610-mm
tray spacing. The trays contained 10% free with 13.0-
mm-diameter holes with 50-mm weirs.
0.8
".y I
Liquid samples were withdrawn from each tray, sepa-
rated into two liquid phases where necessary, and analyzed
for water by an automated Karl Fischer titration method
and for ethanol and hydrocarbons with a GC equipped
with a 50-m fused silica open tubular column. Full details - c - 2.2.4TMC.5
of the distillation columns and analysis are given by Cairns
(1988).
0.I 2.3.4TMC5
Experimental Results 0.0
0 1 2 3 4 5 6 1 8 9 1
Glass Column, 104-mm Diameter. A total of 29 total
reflux runs, as described by Cairns (1988), were performed
in the various distillation regions of the compositional

1
diagram. Figure 8 shows 12 typical runs on a quasi-ternary
diagram with all the hydrocarbon components summed to 2,3DMC6
give an overall alkylate composition. The figure shows that - L2ADM.6
all runs tended toward a common heterogeneous azeotrope no40
0.045

t // I
z
overhead and that the bottoms approached different c 0.035
compositions along varying paths which were dependent e2 on30
on the makeup of the overall feed mixture. Li

An example of the full results is given for run 1 in Table 3 0025

0
1V. The complete set of results for all 29 runs are available 0020
as supplementary material. Run 1 was wholly within the
0.015
three-phase region with the bottoms composition ap-
proaching pure alkylate. Table IV shows the composition 0.010
and fraction of each phase a t each stage. Also given are 0.005
the stage temperatures and estimated froth heights, along 0.W
with the calculated F numbers ( F = up,"2). The raw
measurements of the volumetric liquid flow rate leaving
the bottom plate, plate 9, and the reboiler steam pressure
are given, and an average mass balance error is also shown.
This figure represents the average of the absolute devia- 0.07 -
tions from 100% obtained when the independent water f2,3DMC6
and gas chromatograph results are summed. The distil-
lation paths have been shown graphically on quasi-ternary 2,4DMC6
diagrams, and where a liquid-phase split was recorded, the
tielines linking the liquid phases are also shown. 2.5DMC6
Bottoms Approaching Pure Alkylate. Runs 5 and
6 shown in Figure 8 resulted in the entire column operating
in the three-phase region, with an approach to pure al- 0.02
kylate at the bottoms and an approach to the heteroaz-
eotrope at the distillate. Run 28, shown in part on Figure o ni
8, avoided the three-phase region, with the compositions
0.00
near the zero water axis and the bottoms approaching pure 0 I 2 3 4 5 6 7 8 9 IO I1 12
PLATE NUMBER
alkylate. Composition profiles along the column for runs
6 and 28 are shown for the major components in Figure Figure 10. Comparison of minor component profiles for runs 6 (top)
9 and the minor components in Figure 10. The regular and 28 (bottom).
profiles obtained for all components highlight the accuracy Bottoms Approaching Pure Water. Runs 7 and 8
of the analytical techniques. shown on Figure 8 resulted in the column also operating
 Ind. Eng. Chem. Res., Vol. 29, No. 7, 1990 1355

5
hl
1.0 0.050 I 1
A - - - - - l
0.8
PHASE
REGION
0.045

0.040
THREE
PHASE
REGION
2,3DMC6
I
z
8 0.035
0030
E
3 0.025
0
0.020

0.015

0.2 0.010

0.1 0.005

0.0 - - - .
O.OO0 c - - .
0 1 2 3 4 5 6 7 8 9 IO 0 1 2 3 4 5 6 7 8 9 1 0
PLATE NUMBER PLATE NUMBER

1.o 0.025
THREE
0.9 THREE I PHASE I
PHAPF ! REGION
0.8 0.020

0.7

2 OA 3 0.015
U.
3 0.5
w
d
E 0.4 0.010

0.3

0.2 0.005

0.1

0.0 O.Oo0
3

Figure 11. Comparison of major component profiles for runs 7 (top) Figure 12. Comparison of minor component profiles for NIU7 (top)
and 16 (bottom). and 16 (bottom).

in the three-phase region, but this time the bottoms ap-

proached pure water. Runs 14,16, and 18 shown in part
in Figure 8 contained compositions near the zero alkylate 0.8
axis, exhibited maxima in ethanol composition, and ap- 0.7
proached the heteroazeotrope from above. E
Composition profiles showing the change from a two-
phase region to a three-phase region are shown for runs
7 and 16 for the major and minor components in Figures
11 and 12.
Bottoms Approaching Pure Ethanol. Runs 26 and
27 shown in Figure 8 resulted in a bottom product ap-
proaching pure ethanol. In run 26, a three-phase region
o,2&

0.1 WATER
2,3,4T C
~ ~ j
existed on the top trays of the column. In both runs 26 0.0
and 27, the compositions were close to the zero water axis 0 1 2 3 4 5
PLATE NUMBER
6 7 8 9 IO
on many trays. Composition profiles for the major com-
ponents are shown in Figure 13, and the pronounced
maxima for the minor components are shown in Figure 14.
Temperature Profiles. The experimental temperature
profiles are shown in Figure 15 for some of the typical runs
in the different distillation regions. Figure 15a shows that
the region around the pseudoternary heteroazeotrope(runs
3-6) had only a small temperature gradient. This is con-
sistent with the ternary bubble point surfaces of Figure 3 0.5
4 for ethanol-water and 2,2,4-TMCS. Runs 7-9 in Figure
15a illustrate a sharp drop in temperature at the top of 2,2,4TMCS
the column due to the presence of the alkylate components.
This is also consistent with Figure 4 where a steep drop
in temperature is shown for only small additions of 2.3.4TMC5
2,2,4-TMC5to dilute ethanol-water mixtures. 0.0
The deep penetration of runs 5 and 6 toward the alkylate 0 1 2 3 4 5
PLATE NUMBER
6 7 8 9 0
corner of the quasi-ternary diagram is reflected in Figure
Figure 13. Comparison of major component profiles for runs 26
15a by the sharp rise in temperature on the lower stages (top) and 27 (bottom).
of the column. This compares with runs 3 and 4, which
show only small temperature variations due to the com- the column. Furthermore, the differences between runs
paratively small composition changes over the length of 5 and 6 and runs 3 and 4 highlight the steep temperature
1356 Ind. Eng. Chem. Res., Vol. 29, No. 7 , 1990

0027.5
0.02s0
1 ,
&PHASE
,
THREE

REGION-
I

TRAY SPACIFiC

50
~ I , ~, .
0.0075 0
1 2 3 4 5 6 7 8 9
PLATE NUMBER

0.0025 Figure 16. Froth height profile, run 8.

0 . m
n ! 2 3 4 5 6 7 R 9 10
PL4TE NL-MBER Figure 15b shows that comparatively gentle temperature
gradients were recorded for the typical runs that ap-
0.0250 I I
proached the pseudoternary heteroazeotrope from above.

0.0200
2 I Runs 14 and 16 tended toward pure water in the reboiler
and, therefore, had higher temperatures than runs 26 and
z 27, which both approached a pure ethanol bottom com-
8 0.0175
2 0.0150
position. The differences between the distillation paths
of runs 26 and 27 are also highlighted in Figure 15b. Run
U
9 0.0125 27 had only a small temperature gradient because it ap-
3 proached the ethanol corner of the diagram along the zero
2 n.0100
water axis, whereas run 26 approached the same compo-
0.0075 sition along a constant water composition, leading to the
0.0050 ethanol-water axis. These differences in temperature
profiles are well illustrated in the bubble point surfaces
ao(125
of ethanol-water and 2,2,4-TMC6 shown in Figure 4.
O . m
I. 1 2 3 4 5 6 7 R 9 10
Operating Observations. The observations of Davies
PLATE NUMBER et al. (1987) and the work of Ross and Nishioka (1975,
Figure 14. Comparison of minor component profiles for runs 26 1981) indicate that the capacity of a three-phase distillation
(top) and 27 (bottom). column is limited by foaming on the trays that have a
composition close to the binodal curve. The froth height
100.0 1
observations during the experimental runs tend to agree
RUN 9
with this, although in some cases large foams were recorded
95.0 t / // I on trays away from the binodal curve. This is illustrated
w.O in Figure 16 for run 8 where the upper most tray had a
froth height twice that of a normal tray spacing. It was
noted that the froth was very foamy, while the lower stages
had a spray/foam mixture with clearly visible liquid dro-
2 80.0 plets. Figure 8 shows that the top section of the column
for run 8 was well away from the binodal curve.
In general, the froths on the trays with three-phase
mixtures tended to appear foamy with no visible small
spray droplets. The samples withdrawn for analysis always
entered the syringe as a cloudy emulsion, which then
n i 2 3 4 5 6 7 s 9 in 11 separated into two phases in the syringe barrel. It was not
PLATE NUMBER possible to distinguish if one phase was dispersed in the
other, and at no time was a tray observed to have a alkylate
85.0
layer resting above an aqueous layer.
1 Runs 1-6 where the entire column operated within the
I- p RUN14 I three-phase region were, in general, difficult runs to per-
80.0 form. This was because the liquid-phase split in the
RLW 16
Y bottom section of the column created problems with the
$ - RUN 26
reboiler during start-up. Although a large thermosyphon
circulation of liquid was usually established, this did not
k! 75.0

5 . R L 3 27
appear to be sufficient to mix the liquid phases, and often
phase separation was observed in the pipework leading to
70.0 - the reboiler.
The effect of a layered flow into the reboiler at the
beginning of the run was to set up a cyclic vapor flow
pattern which could only be stabilized by manipulating the
65.0
0 I 2 3 4 5 6 7 8 9 10 11
steam load to the reboiler.
PLATE NUMBER The problem was that the hydrocarbon phase tended
Figure 15. Experimental temperature profiles (a, top; b, bottom). to vaporize first, sending a large pulse of vapor to the top
of the column and flooding that section. As the water
front as the pure alkylate corner is approached (see Figure phase entered the reboiler, less vapor was generated, and
4). the result was that the hydrocarbon at the top of the
 Ind. Eng. Chem. Res., Vol. 29, No. 7, 1990 1357

Table IV. Ethanol-Water-Alkylate Distillation Experimental Results"

DISTILLATION PATH
RUN I

0.8 t----a TlEUHB

0.2-

ai -
0.0 0.1 0.2 0,4 0.5 0.6 0.7 0.8 0.9 1.0
ALKYLATE (MOLE FfUCnOh')

stage
component cond 1 2 3 4 5 6 7 8 9 rboil
Aqueous Phase, Mole Fraction
water 0.342 79 0.328 44 0.330 99 0.33856 0.34846 0.37837 0.398 12 0.443 75 0.546 09 0.728 76 0.960 88
ethanol 0.639 36 0.644 59 0.643 37 0.636 65 0.630 89 0.605 84 0.589 20 0.547 55 0.451 27 0.271 24 0.039 12
2,3-DMC, 0.001 15 0.001 56 0.001 25 0.000 95 0.000 62 0.000 36 0.000 26 0.000 16 0 0 0
2,2,4-TMCS 0.016 31 0.024 36 0.023 02 0.021 94 0.017 77 0.013 56 0.01062 0.006 94 0.002 10 0 0
2.5-DMCa 0.000 10 0.000 22 0.000 26 0.000 31 0.000 34 0.000 30 0.000 26 0.000 18 0.00008 0 0
0.000 13 0.000 30 0.000 37 0.000 43 0.000 55 0.000 40 0.000 37 0.000 30 0.000 11 0 0
0 0 0 0.00015 0.00020 0 0 0 0 0 0
0.000 16 0.000 33 0.00041 0.000 60 0.000 64 0.000 60 0.000 63 0.000 54 0.000 22 0 0
0 0 0 0 0.000 10 0.000 11 0.000 11 0.000 14 0 0 0
0 0.000 21 0.000 23 0.000 40 0.000 44 0.000 45 0.000 43 0.000 44 0.000 12 0 0
0 0 0 0 0 0 0 0 0 0 0
0 0 0.000 10 0 0 0 0 0 0 0 0
Alkylate Phase, Mole Fraction
water 0.007 30 0.008 62 0.008 89 0.008 37 0.007 94 0.007 94 0.006 01 0.004 85 0.002 96 0.001 50 O.OO0 57
ethanol 0.159 83 0.145 85 0.109 70 0.100 33 0.094 49 0.082 96 0.076 01 0.062 43 0.041 40 0.021 32 0.004 17
2,3-DMC5 0.05308 0.043 49 0.037 85 0.030 39 0.024 89 0.019 67 0.017 40 0.014 57 0.011 57 0.008 71 0.005 69
2,2,4-TMC5 0.749 77 0.760 43 0.790 76 0.785 41 0.775 75 0.761 81 0.751 84 0.739 91 0.711 23 0.668 02 0.565 37
2,5-DMC6 0.006 56 0.008 62 0.01044 0.013 84 0.017 18 0.021 15 0.023 44 0.027 29 0.033 74 0.040 15 0.04890
2,4-DMCe 0.008 06 0.011 12 0.013 42 0.017 58 0.02208 0.027 24 0.03144 0.035 78 0.043 83 0.05343 0.066 56
2,2,3-TMCS 0.001 72 0.002 87 0.002 97 0.003 89 0.004 77 0.005 84 0.006 86 0.007 72 0.009 53 0.011 38 0.014 27
2,3,4-TMC5 0.007 99 0.011 11 0.014 86 0.022 19 0.028 87 0.039 28 0.046 39 0.055 84 0.074 41 0.097 58 0.13832
2,3,3-TMCS 0.001 17 0.00181 0.002 5 0.004 53 0.005 88 0.008 56 0.010 82 0.012 93 0.018 46 0.026 70 0.041 46
2,3-DMCe 0.004 51 0.006 08 0.008 56 0.012 91 0.017 39 0.023 99 0.02846 0.035 37 0.047 66 0.063 18 0.098 81
3,4-DMC6 0 0 0 0.000 55 0.000 77 0.001 55 0.001 33 0.001 95 0.002 99 0.004 44 0.007 51
2.2.5-TMCa 0 0 0 0 0 0 0 0.001 35 0.002 25 0.003 58 0.007 71
phase splits, mole fraction
staee aaueous alkvlate tema. "C froth heieht. cm F. SI units
cond 0.631 23 0.368 77
1 0.639 24 0.360 76 68.0 15.5 1.704
2 0.622 57 0.377 43 68.2 13.5 1.703
3 0.644 23 0.355 77 12.5 1.721
4 0.613 37 0.386 63 68.5 11.5 1.695
5 0.642 26 0.357 74 69.1 10.5 1.715
6 0.60000 0.400 00 10.5 1.675
7 0.624 60 0.375 40 69.8 10.0 1.706
8 0.536 30 0.463 70 71.4 10.0 1.671
9 0.373 00 0.627 00 77.6 10.5 1.752
rboil 0.211 57 0.78843
"Run 1;date performed, 20/5/1986; comments, none. Column operating conditions: reboiler steam pressure = 16.0 kPa g; liquid flow rate
leaving plate 9 = 2.34 L/min; column average F = 1.705 SI units; average mass balance error = 1.317%.

column could not be sustained and would fall back toward of the energy needed to vaporize the same quantity of
the reboiler. Eventually the hydrocarbon would reenter water. By manipulating the steam flow to the reboiler
the reboiler and the pulsing cycle would repeat. It is during the upward surge of hydrocarbon vapor, it was
significant to note that to vaporize 1 mol of 2,2,4-tri- possible, after some time, to arrive at a condition in which
methylpentane only requires approximately three-quarters the two-liquid-phase flow into the reboiler produced a
1358 Ind. Eng. Chem. Res., Vol. 29, No. 7 , 1990

Table V. Composition of Simulation Feed Mixtures .

feed composition, mole u
0.9 - - SMWLATION
fraction
comDonent run 27 run 1
-
0.001 92 0.207 50
0.887 15 0.048 66
0.000 54 0.005 07
0.057 45 0.432 33
0.006 44
0.007 99
0.001 73
0.035 60
0.04841
0.01038
o.6
".,
OM
t 0.05 0.10 0.1s 0.20 0.2s 0.30
* \
0.35 0.40 0.4.5 050
0.01771 0.099 98 ALKYLATE (mole fraction)
0.00504 0.029 86
0.012 03 0.071 31 1.0 1 1
0.000 91 0.005 39
0.001 08 0.005 51

steady exit vapor flow.

Simulation of Experimental Results
Runs 27 and 1were simulated by using the total reflux
method described above. All 12 components were included
in the simulations, and the predictions of the group con-
tribution thermodynamic models (UNIFAC, modified
UNIFAC, ASOG, and modified ASOG) were examined.
The binary interaction models NRTL (Renon and
Prausnitz, 1968) and UNIQUAC (Abrams and Prausnitz,
1975) could not be used, as no data exist to cover all
possible interactions. Antoine coefficients for all compo-
nents were taken from Sinnott et al. (1983), and a 1 kPa
pressure drop per stage was assumed. The reboiler was
assumed to retain 90% of the overall liquid mixture, with
the remaining 10% distributed evenly amongst the other
stages.
Seven ideal stages were found to best represent the ex-
perimental data for both runs, and hence the simulations
indicated that the column was operating at 70% overall . -
column efficiency. PLATE NUMBER
The results of the simulations for both runs 27 and 1 Figure 17. Simulation Vs,run 27. (a, top) Seven ideal stages, 90%
are discussed in more detail below. The compositions of reboiler holdup, UNIFAC-VLE thermodynamics. (b, middle) Major
components, 70% overall efficiency, UNIFAC-VLE thermodynam-
the feed mixtures are given in Table V. ics. (c, bottom) Minor components, 70% overall efficiency, UNI-
Run 27 Simulation Comparisons. Figure 17a com- FAC-VLE thermodynamics.
pares the experimental distillation path with the simula-
tion path obtained by using UNIFAC-VLE. This figure 1.0 I

represents an expanded view of the distillation path, with

all the hydrocarbon components being summed to give an
0.9
0 0.8 F ETHANOL

alkylate composition a t each stage. The composition of 2 0.7

the heteroazeotrope predicted by UNIFAC-VLE appears 0.6

to be richer in hydrocarbon than suggested by the ex- 3 OS - .
W A C
---- MODIJNIFAC
perimental path. The result is that the simulation path ' 0.4
P---
2.2.4TMC5
------ ASOO
MODASOO
0.3 ==--=
does not match the observed values a t the top of the Y-

0.2
column, as shown in Figure 17a.
0.1 WATER
The results of scaling up the simulation UNIFAC-VLE
nn
predictions for a seven-ideal-stage column using an overall 0 1 2 3 4 5 6 7 8 9 IO
PLATE NUMBER
column efficiency of 70% are shown in the component
profiles of Figure 17b,c. The predicted major component
profiles (Figure 17b) show good agreement with the ex- 800 I ,
perimental values for the bottom section of the column,
while some deviations are apparent toward the top. This
'2-7 7 . 5 1
is due to the errors of UNIFAC-VLE in predicting the
composition of the heteroazeotrope. The predictions of
the minor component profiles (Figure 17c) show excellent
agreement considering the small quantities involved, and - UNlFAC
again the largest errors occur in the top section of the ___-
-_ _ -_ _ MOD
ASOG
UNIFAC
MOD A S M
column.
The predictions of the other thermodynamic models are 650 ' I
0 1 2 3 4 5 6 7 8 9 IO
shown in Figure 18 for 70% overall column efficiency. All PLATE NUMBER
models predict an accurate bottom composition; however, Figure 18. Simulation V's, run 27. Comparison of thermodynamic
the overhead vapor compositions varied. The predictions models. (a, top) Composition profiles. (b, bottom) Temperature
obtained by using modified UNIFAC gave the best esti- profiles.
 Ind. Eng. Chem. Res., Vol. 29, NO. 7, 1990 1359
1.0 I

p 0.8
- EXPERIMENr
__-.
SIMULATION

0.1
0.5 0.0
0.0 0.1 0.2 0.3 0.4 0.5 0 1 2 3 4 5 6 7 8 9 10
ALKYLATE (mole fraction) PLATE NUMBER

-
.-S
1.0
I
0.0800

z o.Ln00
1 12.3DMC6 I
-
-e8
EYPERIMENT
2 0.9
EXP"
WAC-VLE
WAC-LLE
g 0.06oO
v E 0.0500
0.8 Y
0 0.0000
z r
<
0.7 0.0300
0.0200
0.6 * \ 0.0100

O.oo00
0.5 ~
0 1 2 3 4 5 6 7 8 9 1 0 1 1 1 2
00 01 02 03 04 05 PLATE NUMBER
ALKYLATE (mole fraction)
Figure 20. Simulation V's, run 1. (a, top) Major components, 70%
overall efficiency, UNIFAC-VLE thermodynamics. (b, bottom)
-a
:
2--.
80'o
77.5 I
t
Minor components, 70% overall efficiency, UNIFAC-VLEthermo-
dynamics.

phase region, whereas ASOG only predicted that the ov-

erhead vapor would be heterogeneous when condensed.
70.0 - Modified ASOG, on the other hand, predicted that all
EXPERIMENT
- tXYEWMENI
_-__MOD UMFAC (PHASE SPLI'ITING)
MOD UMFAC (NO PHASE SPLITTING)
stages, including the condenser liquid, would have only a
- - UNIFAC-LLE single liquid phase.
650 '
0 1 2 3 4 5 6 7 8 9
I
IO
The effect of ignoring this predicted liquid-phase split
is shown in Figure 19a for a seven-stage column with
PLAIT. NUMBER
modified UNIFAC. This figure shows that a greater sep-
Figure 19. Simulation V's, run 27. (a, top) Effect of ignoring liq- aration is predicted when the phase splitting is ignored,
uid-phase splitting, modified UNIFAC thermodynamics. (b, middle) which corresponds to a lower overall column efficiency. It
Comparison of UNIFAC-VLE and UNIFAC-LLE. (c, bottom)
Temperature profile comparisons. is interesting to note that by including the phase split the
predicted distillation path more closely approximates the
mate of the observed overhead vapor composition; how- homogeneous experimental run. This is also illustrated
ever, they provided less accurate results for the profiles by the differences in the temperature profiles shown in
in the lower section of the column. ASOG, on the other Figure 19c where the homogeneous predictions of modified
hand, produced excellent agreement in the lower section UNIFAC show large errors at the top of the column.
but predicted an overhead vapor that was weaker in A unique feature of the original UNIFAC thermody-
ethanol than observed. Modified ASOG also provided namic model is that there exists two parameter sets; one
good agreement in the lower section of the column but for predicting VLE (UNIFAC-VLE), and the other de-
failed to produce an accurate overhead composition by only signed for LLE calculations (UNIFAC-LLE). The LLE
predicting small quantities of water at the top of the parameters were formulated by correlating the group in-
column (Figure 18a). teractions against experimental liquid-liquid equilibria.
The predictions of temperatures at 70% overall column Figure 19b shows the results obtained using these param-
efficiency for all models are compared in Figure 18b. eters for a seven-stage column. The predicted distillation
UNIFAC-VLE provided the best estimates at the top of path fails to follow the experimental path, and also the
the column, while modified UNIFAC was more accurate predicted column temperatures are much higher than
in the lower section. It is significant to note that modified observed (Figure 19c). This was probably to be expected
UNIFAC predicted temperatures that were consistently as the LLE parameters are only recommended for the
lower than UNIFAC-VLE. Both versions of ASOG pre- range 10-40 "C (Magnussen et al. 1981). UNIFAC-LLE
dicted higher temperature values throughout the column, also predicted that the entire column was in the two-phase
with modified ASOG failing to correct the profile at the region.
condenser. Run 1 Simulation Comparisons. Figure 20 compares
An interesting feature of these simulations similar to the experimental and simulation profiles based on UNI-
those of Kovach and Seider (1987b) was that, even though FAC-VLE for an overall column efficiency of 70%. This
the experimental results were totally within the two-phase figure shows the overall composition for each component
region, some of the thermodynamic models predicted some in run 1that was wholly within the three-phase region; see
stages would lie in the three-phase region. Modified Table IV. The major component predictions, shown in
UNIFAC predicted that the top five stages of a seven-stage Figure 20a, are in good agreement with the experimental
column would have two liquid phases. UNIFAC-VLE data in the top portion of the column; however, errors are
predicted that the top two stages would be in the three- apparent in the bottom section. The minor components,
1360 Ind. Eng. Chem. Res., Vol. 29, No. 7, 1990

-
250.0

UNIFAC-VLE
$2jw . o e----* MODUNlFAC
+ C---. ASOO
+----+ MODASOG
150.0

6 100.0

0.0 '
1 2 3 4 5 6 7 8 9
I
10
PLATE NUMBER

85.0 I 1 250.0 I 1
FXPERIMENT
-MODUNIFAC
----_.
200.0
--- UNIFAC
ASOO
MOD ASOG 4W 150.0
Y
5 100.0
50.0

65.0 '
0 1 2 3 4 5 6 7 8 9
I
IO
0.0 '
I 2 3 4 5 6 7 8 9
J
IO
P L A m NUMBER PLATE NUMBER
Figure 21. Simulation V's, run 1. (a, top) Composition profiles,
thermodynamic models. (b, bottom) Temperature profiles, ther-
modynamic models. -
C---.
UNIFAC-VLE
e----* MODUNIPAC
ASOO
Figure 20b, show remarkable agreement considering the *----+ MODASOG
150.0
small quantities involved and the fact that two liquid
phases were found on all stages.
The distillation path predictions of the other thermo-
dynamic models for a seven-ideal-stage column are shown
in Figure 21a. All models predict a similar vapor line
position, with some variations in the location of the het-
eroazeotrope. The simulations show that, because the 1 2 3 4 5
PUTE"ER
6 7 8 9 IO
reboiler is within the three-phase region, the stage above
is constrained to lie on the vapor line, and hence, there is Figure 22. Murphree vapor efficiencies. (a, top) Ethanol profiles,
run 27. (b, middle) Water profiles, run 27. (c, bottom) 2,2,4-TMC5
a large change in composition between these two stages. profiles, run 27.
The experimental points of Figure 21a do not appear
to show the same rapid compositional changes and prob- would leave stage n in equilibrium with the liquid on that
ably reflect a different position of the vapor line and stage.
possibly the inefficiencies of the trays, with the vapor By use of the conditions of total reflux, the above
leaving the two-liquid-phase mixtures not being in equi- equation can be rewritten as
librium, as assumed in the simulations. This is particularly
relevant in the bottom section of the column and explains
the errors illustrated in Figure 20a. Further, because
Figure 20b shows that the minor components were well
predicted over the length of the column, this tends to which allows the experimentally measured stage compo-
indicate that these components played only a small role sitions given in the supplementary material to be used to
in establishing the position of the vapor line. calculate efficiencies for each component on all trays. The
The temperature profile predictions for all models are prediction of Y*,,,~ requires the use of a thermodynamic
shown in Figure 21b. As with the simulation of run 27, model and, therefore, makes the values of the efficiencies
ASOG produced higher temperatures, whereas modified model dependent.
UNIFAC again predicted consistently lower values than Murphree vapor efficiencies based on eq 31 have been
UNIFAC-VLE in the top portion of the column. All calculated for runs 27 and 16 by using UNIFAC-VLE,
models forecast the observed small temperature gradient modified UNIFAC, ASOG, and modified ASOG. In all
leading toward the heteroazeotrope. cases, the equilibrium vapor composition has been calcu-
lated assuming an ideal gas at atmospheric pressure. The
Murphree Vapor Efficiencies Murphree tray efficiencies for run 27 show that many of
The experimental data collected allow for an investi- the predicted values were greater than 100%. This cor-
gation of the Murphree tray efficiencies for each compo- responds to the experimental data producing a larger en-
nent. The Murphree vapor efficiency of component i on richment (or stripping) of the components than predicted
stage n is given by by the equilibrium estimates. Parts a-c of Figure 22 show
the efficiency column profiles for the major components
ethanol, water, and 2,2,4-trimethylpentane for each of the
thermodynamic models. The ethanol efficiencies, Figure
22a, varied by more than 30% over the length of the
where represents the composition of a vapor that column. Modified ASOG produced a consistently higher
 Ind. Eng. Chem. Res., Vol. 29, No. 7, 1990 1361
1 efficiencies of Figure 23a suggest that the measured de-
crease in concentration from plate 4 to plate 5 was much
8 m.0 greater than the equilibrium estimates. A significant
e feature of Figure 23a, however, is that modified ASOG only
@ 150.0 indicated an efficiency for ethanol of the fifth tray of
i 100.0
110%. This would appear to correspond with the generally
good simulation results, where modified ASOG provided
50.0
a comparatively accurate estimate of the experimental
distillation path at the turning point.
0.0 '
1 2 3 4 5 6 7 8 9
J
10
The efficiencies of water, Figure 23b, also show a major
deviation. ASOG estimated that the efficiency of water
PLATE NUMBW
on the second plate was less than -3700%, while all other
, models indicated efficiencies of greater than 100%. Again,
-
250.0 ;

'VNIFAC-VLE
upon closer examination of the water composition profile,
e----* MODUNIFAC
L - --t ASOO
it would appear that this has occurred in the vicinity of
e----+ MODASOO a turning point. It is also significant that the simulation
results show that ASOG failed to adequately predict the
distillation path in the top section of the column and that
the model produced poor estimates of the water concen-
tration around the second plate.
The efficiencies of all models for 2,2,4-TMC, are very
0.0 '
1 2
L
3 4 5 6 7 8 9
I
10
similar (Figure 23c) and show an approximate 40% vari-
ation on the top six trays where the hydrocarbon appeared
PLATE NUMBER
in detectable quantities. A negative efficiency for 2,3,4-
TMC, was indicated by all models on the top tray, and this
-
250.0
corresponds to a maximum in the observed concentration
UNIFAC-VLE
8 200.0 D----*
+--a
MODWFAC
ASOG
profile.
+----+ MODASOO As a final remark on the efficiences for this run, it is
perhaps interesting to note that the essentially binary
ethanol-water values for the lower trays of the column
fluctuate around 70% for all models, which matches the
observed overall column efficiencies deduced in the sim-
ulation of the experimental results. The phase split was
0.0 '
1 2 3 4 5 6 7 8 9
J
10
taken into consideration in these calculations for the
Murphree vapor efficiencies. If this phase split is ignored,
PLATE NUMBER
then different efficiencies will be produced.
Figure 23. Murphree vapor efficiencies. (a, top) Ethanol profiles, Steel Column, 610-mm Diameter. A number of runs
run 16. (b, middle) Water profiles, run 16. (c, bottom) 2,2,4-TMC6
profiles, run 16. at total reflux were made in a 610-mm-diameter column
with provision to withdraw samples from the midpoint of
efficiency, and the UNIFAC models predicted generally the plates for analysis. Provision was also made to measure
lower values. the mid-point temperatures of the plates. The composition
Figure 22b shows the water efficiency profiles, and all and temperature profiles for the system rich in ethanol and
models, other than modified ASOG, indicate a similar water are shown in parts a and b of Figure 24, respectively,
trend. The larger estimated efficiencies by modified ASOG together with the predicted values from UNIFAC, modi-
are consistent with the simulation results. All models, fied UNIFAC, ASOG, and modified ASOG. The most
however, predict that a greater than equilibrium concen- suitable overall column efficiency was found to be 75%.
tration of water was observed in the top section of the All thermodynamic models predict only a single liquid
column, resulting in efficiencies of more than 100% on the phase throughout the column.
top tray. The efficiency profiles of the major hydrocarbon The Murphree vapor efficiencies were calculated for all
component, 2,2,4-TMC6,are shown in Figure 22c. Only components on all plates using the above thermodynamic
small differences are predicted by the modifications to each models. For the major components, ethanol and water,
of the thermodynamic models, and ASOG generally esti- the Murphree efficiencies were approximately 75% for all
mated higher efficiencies than UNIFAC. Interestingly, the models. Wider variations were observed for the minor
top stages that correspond to an increasing water compo- components. The overall column efficiency was found to
sition show a rapid fall in the efficiency of 2,2,4-TMC,, with be similar to the efficiency of the major components.
all models, indicating the top tray was only operating at The column was also equipped for liquid sampling along
approximately 8% of its equilibrium potential. several lines between weirs on a tray described by Stevens
Run 16 had the top stages of the column in the three- and Furzer (1989). These weir-weir composition profiles
phase region, with the lower trays consisting of mainly provided valuable information on liquid mixing and in-
hydrocarbon-free ethanol-water mixtures. The Murphree formation as to the appearance of two liquid phases at a
tray efficiencies for this run show an enormous range of point on the tray. These profiles are shown in parts a, b,
-2724% to 3200%. and c of Figure 25 for water, ethanol, and 2,2,4-TMC6,
The major component profiles are illustrated in Figure respectively. The midpoint sampling position is shown at
23. The ethanol profile shows that all models indicate a 0.50 fractional distance across the tray, and the measured
large jump in efficiency on the fifth plate. Close exami- composition is in reasonable agreement with the value
nation of the experimental distillation path and the ethanol shown on the centerline profile. The composition differ-
concentration profile reveals that the ethanol concentration ences between the centerline and an off-center profile
in the column passed through a maximum at plate 4. The should be noted on these figures. The tray F number was
1362 Ind. Eng. Chem. Res., Vol. 29, No. 7 , 1990

OX
I

6 07
E2
0
.
I
.

2 05
06

04
I -UNIFAC
..
-__-
-- --- MOD UNIFAC 112 2

0 20
0 0
IrjLET
0 25 0.50 0.75
FRACTIONAL DISTANCE ACROSS PLATE
I M
01 ITET

i5,
03
0x0,

_I
02 I
0.1 t 2.2PTMCS
00
! 2 3
PLATE NUMBER
4 5
= 074 : ,I
0 72
,
0 70
OW 0 25 0.50 0.75 1.sa
INLET FRACTIONAL DISTANCE ACROSS PLATE 0CTLE-T

65.0
, , , , ,-y 1
- - - --- - -
_ . _ MOD
__ ASOG
___
MOD W A C

0 25
i

0 50
\

0 75 1 0
I

I
0 1 2 3 4 5 IVLET FRACTIOAAL DISTANCE ACROSS PLATE OUnET
PLATE NUMBER
Figure 24. Simulation V's experiment. Comparison of thermody- Figure 25. Plate composition profile. (a, top) Water, plate 1. (b,
namic models. (a, top) Composition profiles. (b, bottom) Temper- middle) Ethanol, plate 1. (c, bottom) 2,2,4-TMCS,plate 1.
ature profiles. The agreement between the simulation and the exper-
iment was dependent on the quality of the group contri-
varied, and the weep point was well predicted by sieve tray bution liquid-phase activity coefficient model to predict
correlations, but the flooding velocity was only 50% of the the number of liquid phases present and the composition
normal predicted value. of the heteroazeotrope. Most group contribution models
were especially good in predicting the composition profile
Conclusions along the column for the small, minor hydrocarbon com-
The simulation of a three-phase distillation column at ponents. The three models mentioned above provided a
total reflux has been completed by using a modified good prediction of most of the experimental data. Models
Naphtali-Sandholm (1971) technique with a phase-split- requiring binary interaction coefficients, such as NRTL
ting algorithm. The modifications include the summation and UNIQUAC, could not be used with confidence due to
of the vapor-phase mole fractions rather than the corre- the large number of interaction coefficients required.
sponding sum of the liquid-phase mole fractions. This is Modified ASOG and UNIFAC-LLE models had limita-
significant in the subsequent Gaussian block elimination tions in predicting water concentrations and temperature
procedures. An extensive critical examination has been profiels, respectively. The overall column efficiency rec-
made of the use of the group contribution methods for ommended for use in a simulation model using UNIFAC-
predicting phase equilibria in VLE and VLLE systems. VLE, modified UNIFAC, or ASOG thermodynamics is
The most suitable liquid-phase activity coefficient models 70% and includes both two- and three-phase distillation
are UNIFAC-VLE, modified UNIFAC, and to a lesser regions.
extent, ASOG. There is a real need for an accurate VLLE Murphree vapor efficiencies are dependent on the
predictive model for three-phase distillation modeling. thermodynamic model used to calculate the phase equi-
The experimental studies have consisted of 29 runs on libria. The experimental tray compositions plus a suitable
a 104-mm-diameter column containing nine dual flow trays group contribution model permitted the Murphree effi-
for the three-phase system, water-ethanol-alkylate. The ciencies to be obtained for each component on each tray.
column has been deliberately operated to cover the full For the dominant or key components, the Murphree ef-
composition space of the system. The overheads tended ficiencies were near 70%,the overall column efficiency.
toward the heteroazeotrope. The bottoms tended toward However, for the other components, the Murphree effi-
either water or ethanol or alkylate depending on the initial ciencies could be large, positive, or even negative numbers.
feed composition charge to the still. The multicomponent These results are similar to those obtained by Ognisty and
system consisted of 12 components. A very extensive data Sakata (1987), for the near ideal system isobutane-n-bu-
set of all components on all trays for all 29 runs had been tane-propane. We would not recommend the introduction
included as supplementary material. Additional runs were of a Murphree efficiency into the simulation equations, as
also made on a 610-nm-diameter column. its numerical value is not only highly variable but is de-
 Ind. Eng. Chem. Res., Vol. 29, No. 7, 1990 1363

pendent on the thermodynamic model. Instead we would Cairns, B. P.; Furzer, I. A. Rapid-Robust Calculational Procedures
recommend the simpler procedure of use for an overall for Multicomponent Distillation Problems. Chem. Eng. Aust.
column efficiency and for a normal tray layout is likely to 1988b,ChE. 13 (2),13.
Cairns, B.P.; Furzer, I. A. Multicomponent Thru-Phase Azeotropic
be near 70%. Distillation. 2. Phase Stability and Phase-Splitting Algorithms.
Further experimental data were collected on a 610" Ind. Eng. Chem. Res. 1990,following paper in this issue.
diameter column indicating overall column efficiencies of Cairns, B. P.; Nobbs, D. M., Furzer, I. A., Vine, J. H. The Identifi-
75% 9 cation and Analysis of the C7 to Cg Fraction from Petroleum
Finally we hope that the extensive data on three-phase Distillation using a Capillary Gas Chromatography Mass-Spec-
distillation in the supplementary material will be valuable trometry Data System. Proc. 9th Aust. Anal. Chem. Conf. (Syd-
ney) 1987,257.
to those persons who are actively engaged in modeling Davies, B.;Ali, Z.;Porter, K. E. Distillation of Systems Containing
three-phase columns. Modeling real experimental data Two Liquid Phases. AIChE J. 1987,33, 161.
requires an excellent modeling technique. Ewell, R. H.; Welch, L. M. Rectification in Ternary Systems Con-
taining Binary Azeotropes. Ind. Eng. Chem. 1945,1224.
Nomenclature Fredenslund, Aa.; Jones, R. L.; Prausnitz, J. M. Group-Contribution
Estimation of Activity Coefficients in Non-Ideal Liquid Mixtures.
B,C, D = matrices in the Naphtali-Sandholm method AZChE J. 1975,21 (6),1087.
E = Murphree vapor efficiency Furzer, I. A. Ethanol Dehydration Column Efficiencies Using UNI-
F = feed mass (kmol) or deviation function FAC. AZChE J. 1985,31 (81,1389.
J = Jacobian matrix Furzer, I. A. Distillation for University Students; Furzer, I. A.,
K = equilibrium ratio Publisher; Department of Chemical Engineering, University of
L = liquid flow rate, kmol/s Sydney: Sydney, NSW, Australia, 1986.
m = stage liquid holdup, kmol Furzer, I. A. CHEMECA Aust. 88 Bicent. Znt. Conf. Proc. 1988,
n, = number of components 1068-1072.
Gmehling, J.; Rasmussen, P.; Fredenslund, Aa. Vapour-Liquid
N = total number of stages Equilibria by UNIFAC Group Contribution. Revision and Ex-
P = pressure, Pa tension 2. Ind. Eng. Chem. Process Des. Dev. 1982,21,118.
T = temperature, K Hoffman, E. J. Azeotropic and Extractive Distillation; Wiley-In-
V = vapor flow rate, kmol/s terscience: New York, 1964.
W = defined by eq 27 Hirose, Y.; Kanase, Y.; Sampei, K.; Kawai, T. A New Approach to
x = liquid mole fraction Distillation Column at Total Reflux. Comput. Chem. Eng. 1980,
y = vapor mole fraction 4,133.
Z = defined by eq 26 Kojima, K.; Tochigi, K. Prediction of Vapour-Liquid Equilibria by
the ASOG Method; Elsevier: Amsterdam, The Netherlands, 1979.
Subscripts Kovach, J. W., 111; Seider, W. D. Heterogeneous Azeotropic Distil-
i = component lation: Experimental and Simulation Results. AZChE J. 1987a,
33 (8),1300.
n = stage number Kovach, J. W., 111; Seider, W. D. Heterogeneous Azeotropic Distil-
N = reboiler lation: Homotopy-Continuation Methods. Comp. Chem. Eng.
v = vapor 1987b. 11 (6).593.
Larsen, B. L.; Rhmussen, P.; Fredenslund, Aa. A Modified UNIFAC
Superscripts Group-Contributions Model for Prediction of Phase Equilibria
k = iteration counter and Heats of Mixing. Znd Eng. Chem. Res. 1987,26,2274.
s = saturated condition Magnussen, T.;Rasmussen, P.; Fredenslund, Aa. UNIFAC Param-
- = average eter Table for Prediction of Liquid-Liquid Equilibria. Znd. Eng.
Chem. Process Des. Dev. 1981,20,331.
Greek Letters Maga, L.;Arrighi, G.; Costa, P.; Canepa, B.;Reverberi A. Theoretical
y = activity coefficient (liquid) and Experimental Analyius of Distillation Paths for Three Com-
p= density ponent Mixtures. Proc. World Congress Chem. Eng. (Tokyo),3rd
1986,paper 8c-152,608.
Supplementary Material Available: Tables A9.1-A9.29, Naphtali, L. M.; Sandholm, D. P. Multicomponent Separation Cal-
culations by Linearization. AZChE J. 1971,17 (I), 1148.
listing the aqueous phase, alkylate phase, phase split, tem- Ognisty, T. P.; Sakata, M. Multicomponent Diffusion: Theory vs
perature, froth height, and F distillation experimental results Industrial Data. Chem. Eng. Prog. 1987,83 (3),60.
for the ethanol-water-alkylate system (58 pages). Ordering Renon, H.; Prausnitz, J. M. Local Compositions in Thermodynamic
information is given on any current masthead page. Excess Functions for Liquid Mixtures. AIChE J. 1968,14(l),135.
Ross, S.; Nishioka, G. J. Foaminess of Binary and Ternary Solutions.
Literature Cited J Phys. Chem. 1975, 79 (15),1561.
ROB, S.;Nishioka, G. J. Dynamic Foam: A Problem in Fractionation
Abrams, D. S.; Prausnitz, J. M. Statistical Thermodynamics of Liq- and Distillation Towers. Chem. Ind. 1981,47.
uid Mixtures: A New Expression for the Excess Gibbs Energy of Sinnott, R. K.; Coulson, J. M.; Richardson, J. F. Design. Chemical
Partly or Completely Miscible Systems. AIChE J. 1975,21 (l), Engineering; Pergamon Press: New York, 1983;Vol. 6.
116. Stevens, R.; Furzer, I. A. Incipient Three Phase Distillation: Ex-
Cairns, B. P. Three Phase Azeotropic Distillation. Ph.D. Thesis, perimental Tray Composition Profiles. AIChE J. 1989,35 (7),
University of Sydney, Australia, 1988. 1199.
Cairns, B. P.; Furzer, I. A. Three Phase Azeotropic Distillation. Tochigi, K.; Lu, B.C.-Y.; Ochi, K.; Kogima, K. On the Temperature
Experimental Results. Distillation and Absorption. Inst. Chem. Dependence of ASOG Parameters for VLE Calculations. AIChE
Eng. Symp Ser. 1987a,104,B505. J. 1981,27 (6),1022.
Cairns, B. P.; Furzer, I. A. Azeotropic Distillation-Simulation and Van Dongen, D. B.; Doherty, M. F. Design and Synthesis of Ho-
Experiment. Proc. 15th Aust. Chem. Eng. Conf. (Melbourne) mogeneous Azeotropic Distillation. I: Problem Formulation for
1987b,57.1. a Single Column. Ind. Eng. Chem. Fundam. 1985,24,454.
Cairns, B.P.; Furzer, I. A. Sensitivity Testing with the Predictive
Thermodynamic Models NRTL, UNIQUAC, ASOG and UNIFAC Received for review February 21, 1989
in Multicomponent Separations of Methanol-Acetone-Chloro- Revised manuscript received August 30, 1989
form. Chem. Eng. Sci. 1988a,43 (3),495. Accepted October 27, 1989