You are on page 1of 13

Computerized Design of Multicomponent Distillation Columns Using

the UNIFAC Group Contribution Method


for Calculation of Activity Coefficients
Aage Fredenslund," Jurgen Gmehling,2Michael L. Mlchelsen,' Peter Rasmussen,' and
John M. Prau~nitz"~
lnstituttet for Kemiteknik, The Technical University of Denmark, DK-2800 Lyngby, Denmark, Abteilung Chemietechnik, Lehrstuhi
Technische Chemie B, University of Dortmund, 46 Dortmund 50. West Germany, and Department of Chemical Engineering, University of
California, Berkeley, California 94720

The UNIFAC group-contribution method for estimating activity coefficients provides the process engineer with a
rapid and reliable method for predicting equilibrium conditions required in distillation-column design.

Introduction nents as considered here, fin is the fugacity of pure liquid i a t


A large part of chemical engineering design is concerned system temperature T and pressure P ; as shown elsewhere
with separation operations. Many of these are diffusional (Prausnitz, 1969) it is calculated from
operations of the phase-contacting type, and distillation,
absorption, and extraction are the most common. For rational
(3)
design of such separation processes, we require quantitative
information on phase equilibria in multicomponent mixtures.
Satisfactory experimental equilibrium data are only seldom where, for pure liquid i, Pis is the saturation (vapor) pressure,
available for the particular conditions of temperature, pres- pisis the fugacity coefficient a t saturation, and u, is the molar
sure, and composition required in a particular design problem. liquid volume, all a t temperature T . Only pure-component
It is therefore necessary to interpolate or extrapolate existing data are required to evaluate f i n .
mixture data or, when suitable data are lacking, to estimate The fugacity coefficients cpi (in the mixture) and pIS(pure
the desired equilibria from some appropriate correlation. A i a t saturation) are found from vapor-phase volumetric
very useful correlation for this purpose, UNIFAC, was recently properties. Normally, a t the low pressures considered here,
proposed. It is the aim of this work (1)to review the essential these fugacity coefficients do not deviate much from unity.
ideas of IJNIFAC, to report extensive modifications and ex- While various methods are available to calculate fugacity
tensions of the UNIFAC parameters required for its use, and coefficients, we have here used the method of Hayden and
to present illustrative examples that indicate the extent and O'Connell(l975) as briefly summarized in Appendix C.
reliability of UNIFAC for making quantitative estimates of T o determine K factors, the most difficult-to-estimate
phase equilibria; and (2) to show how UNIFAC, coupled with quantity is the activity coefficient y. UNIFAC provides a
material- and energy balances, may be used to design multi- method for estimating activity coefficients in nonelectrolyte
component distillation columns. liquid mixtures remote from critical conditions. T o use this
method, no experimental data are required for the particular
The UNIFAC Method mixture of interest. In addition to the temperature and com-
position of the system, it is necessary only to know the mo-
UNIFAC is an abbreviation that indicates Universal
quasichemical Functional Group Activity Coefficients. It is lecular structure of every component in the mixture and the
based on the quasichemical theory of liquid solutions proposed necessary group parameters. A large number of group-inter-
by Guggenheim (1952), generalized by Abrams and Prausnitz action parameters is given here.
The large advantage of a group-contribution method is that
(1975), and applied to functional groups within the molecules
(rather than the molecules themselves) by Fredenslund et al. it enables systematic interpolation and extrapolation of
(1975). vapor-liquid equilibrium data simultaneously for many
In process design, the required phase equilibrium infor- chemically related mixtures. Most important, it provides a
mation is commonly expressed by K factors reasonable method for predicting properties of mixtures where
no mixture data at all are available. For such mixtures it is not
necessary to measure the intermolecular interactions because
these can be calculated whenever appropriate group-inter-
where y , is the mole fraction of component i in the vapor phase action parameters are known. These, however, are found from
and x , is the mole fraction of component i in the liquid phase. experimental data containing not the same molecules as those
Using standard thermodynamics in the mixture of interest, but containing the same groups. The
Tlfl O main advantage, then, is a form of "molecular scale-up." While
K, = - (2) there are many thousands of liquid nonelectrolyte mixtures
PlP
of interest in chemical technology, these mixtures can be
where y, is the liquid-phase activity coefficient, f L O is the constituted from a much smaller (perhaps 50 or 100) number
standard-state fugacity, pL is the vapor-phase fugacity coef- of functional groups.
ficient, and P is the total pressure. For condensable compo- Langmuir (1925) suggested that physical (van der Waals)
' Technical University of Denmark. interactions between polyfunctional molecules may be esti-
mated by summing interactions between the functional groups
University of Dortmund.
University of California. that constitute the molecules. Typical functional groups might

450 Ind. Eng. Chem., Process Des. Dev., Vol. 16, No. 4, 1977
Table I. Group Volume and Surface Area Parameters
Group name0 Rk Qk Sample group assignment
0.9011 0.848 Butane: 2 CH,, 2 CH,
“CH,” 0.6744 0.540
0.4469 0.228 i-Butane: 3 CH,, 1 CH
0.2195 0.000 2,2-Dimethylpropane 4 CH,, 1 C
CH,=CH 1.3454 1.176 1-Hexene: 1 CH,, 3 CH,
1 CH,=CH
CH=CH 1.1167 0.867 2-Hexene: 2 CH,, 2 CH,,
“C=C”
1 CH=CH
CH=C 0.8886 0.676 2-Methylbutene-2: 3 CH,, 1 CH=C
1.1173 0.988 2-Methylbutene-1: 2 CH,, 1 CH,, 1 CH,=C
“ ACH”
0.5313 0.400 Benzene: 6 ACH
0.3652 0.120 Styrene: 1 CH,=CH, 5 ACH, 1 AC
ACCH, 1.2663 0.968 Toluene: 5 ACH, 1 ACCH,
“ACCH,” ACCH, 1.0396 0.660 Ethylbenzene: 1 CH,, 5 ACH, 1 ACCH,
0.8121 0.348 Cumene: 2 CH,, 5 ACH, 1 ACCH
1.8788 1.664 1-Propanol: 1 CH,, 1 CH,CH,OH
1.8780 1.660 2-Butanol: 1 CH,, 1 CH,, 1 CHOHCH,
“CCOH” CHOHCH, 1.6513 1.352 3-Octanol: 2 CH,, 4 CH,, 1 CHOHCH,
CH,CH,OH 2.1055 1.972 Ethanol: 1 CH,CH,OH
CHCH,OH 1.6513 1.352 Isobutanol: 2 CH,, 1 CHCH,OH
CH,OH 1.4311 1.432 Methanol: 1 CH,OH
H,O 0.92 1.40 Water: 1 H,O
ACOH 0.8952 0.680 Phenol: 5 ACH, 1 ACOH
CH,CO 1.6724 1.488 Ketone group is 2nd carbon;
2-Butanone: 1 CH,, 1 CH,, 1 CH,CO
“CH ,C0” CH,CO 1.4457 1.180 Ketone group is any other carbon;
3-Pentanone: 2 CH,, 1 CH,, 1 CH,CO
CHO 0.9980 0.948 Acetaldehyde: 1 CH,, 1 CHO
CH,COO 1.9031 1.728 Butyl acetate: 1 CH,, 3 CH,, 1 CH,COO
“COOC” CH,COO 1.6764 1.420 Butyl propanoate: 2 CH,, 3 CH,,
1 CH,COO
1.1450 1.088 Dimethyl ether: 1 CH,, 1 CH,O
“CH,O” 0.9183 0.780 Diethyl ether: 2 CH,, 1 CH,, 1 CH,O
0.6908 0.468 Diisopropyl ether: 4 CH,, 1 CH, 1 CH-0
0.9183 (1.1) Tetrahydrofuran: 3 CH,, 1 FCH,O
CH,NH, 1.5959 1.544 Methylamine: 1 CH,NH,
“CNH,” 1.3692 1.236 n-Propylamine: 1 CH,, 1 CH,, 1 CH,NH,
1.1417 0.924 Isopropylamine: 2 CH,, 1 CHNH,

“CNH” { CH,NH
CH,NH
CHNH
1.4337
1.2070
0.9795
1.244
0.936
0.624
Dimethylamine: 1 CH,, 1 CH,NH
Diethylamine: 2 CH,, 1 CH,, 1 CH,NH
Diisopropylamine: 4 CH,, 1 CH, 1 CHNH
’ ACNH, 1.0600 0.816 Aniline: 5 ACH, 1 ACNH,
CH,CN 1.8701 1.724 Acetonitrile: 1 CH,CN
“CCN”
{CH,CN 1.6434 1.416 Propionitrile: 1 CH,, 1 CH,CN
“COOH”
3COOH
HCOOH
1.3013
1.5280
1.4654
1.224
1.532
Acetic acid: 1 CH,, 1 COOH
Formic acid: 1 HCOOH
1.264 Butylchloride: 1 CH,, 2 CH,, 1 CH,Cl
“CC1” 1.2380 0.952 Isopropyl chloride: 2 CH,, 1 CHCl
0.7910 0.7 24 tert-Butyl chloride: 3 CH,, 1 CC1
2.2564 1.988 Dichloromethane: 1 CH,Cl,
“ CCl, ” CHC1, 2.0606 1.684 1,l-Dichloroethane: 1 CH,, 1 CHC1,
1.8016 1.448 2,2-Dichloropropane: 2 CH,, 1 CCl,
1‘CCl ,” CHC1, 2.8700 2.410 Chloroform: 1 CHCl,
2.6401 2.184 l,l,l-Trichloroethane: 1 CH,, 1 CC1,
cc1, 3.3900 2.910 Carbon tetrachloride: 1 CC1,
ACCl 1.1562 0.844 Chlorobenzene: 5 ACH, 1 ACCl
CH,NO, 2.0086 1.868 Nitromethane: 1 CH,NO,
“CNO, ” CH,NO, 1.7818 1.560 1-Nitropropane: 1 CH,, 1 CH,, 1 CH,NO,
CHNO, 1.5544 1.248 2-Nitropropane: 2 CH,, 1 CHNO,
ACNO, 1.4199 1.104 Nitrobenzene: 5 ACH, 1 ACNO,
CS 2 2.057 (1.65) Carbon disulfide: 1 CS,
a Note: Subgroups (for example CH,=CH and CH=CH) within the same main group (here C=C) have different values
for R k and Q k but identical group interaction parameters, a w n . b Not t o be used with oxygen-containing hydrocarbons or
amines .

be CH3, OH, C1, N02, “ 2 , COOH, etc. Langmuir stated that able model, t o estimate thermodynamic properties, in par-
if it is possible t o characterize quantitatively the physical in- ticular, activity coefficients. UNIFAC provides such a
teractions between such groups, it should then be possible t o model.
estimate intermolecular interactions and finally, with a suit- Langmuir’s suggestion received little attention until Wilson

Ind. Eng. Chem., Process Des. Dev., Vol. 16, No. 4, 1977 451
and Deal (1962), Derr and Deal (1969), and Ronc and Ratcliff
(1975) proposed the ASOG model, which is similar to UNI-
FAC in principle but not in detail. From the user’s point of
1 ATM. A
l
view, UNIFAC provides three advantages: (1)flexibility, be-
cause UNIFAC, provides a priori estimates of sizes and surface
areas; (2) simplicity, because UNIFAC parameters, unlike
ASOG parameters, are essentially independent of temperature
for the temperature range considered here, typically 30-125
OC; and finally (3) due to results reported in this work, UNI-
FAC parameters are now available for a considerably larger
number of functional groups than ASOG parameters; there-
fore, the range of applicability for UNIFAC is larger.
All group-contribution methods are necessarily approxi-
mations because any group within a molecule is not completely
independent of the other groups within that molecule. But it
is precisely this independence which is the essential basis of
every group-contribution method. We can allow for interde-
pendence of groups within a molecule by our definition of what
atoms constitute a group. Increasing distinction of groups,
however, also increases the number of group interactions that
must be characterized. Ultimately, if we carry group distinc-
tion to the limit, we recover the individual molecules. In that 0 .2 .4 .6 .% 1.0
event, the advantage of the group-contribution method is lost. Mole Fraction W a t e r , x
Judgment and experience must tell us how to define func- H 20

tional groups so as to achieve a compromise between accuracy Figure 1. Experimental and calculated activity coefficients for organic
acid-water systems exhibit both positive and negative deviations from
of prediction and engineering utility. Raoult’s law: calculated, -; experimental: formic acid (1) 0-water
(4) 0 , Ito and Yoshida (1963); acetic acid (2) 0-water (4) m, Sebas-
tiani et al. (1957); propionic acid (3) A-Water (4)A ,Ito and Yoshida
Advances in UNIFAC Prediction of Activity (1963).
Coefficients
The UNIFAC group-contribution method is summarized
in Appendix A. Fredenslund et al. (1975) showed that the model to the experimental activity coefficients. The values
UNIFAC method may be used to predict fluid-phase equi- of U C O O H , C H and
~ UCH~,COOHmust be known prior to obtaining
libria in systems containing hydrocarbons (saturated, un- U H ~ O , C O O Hand U C O O H , H ~ Ofrom data on systems containing
saturated, and aromatic), alcohols, water, ketones, esters, organic acids and water. In some cases it is advantageous to
ethers, chlorinated hydrocarbons, amines, and nitriles. The evaluate several pairs of parameters, e.g., U C H ~ , C O O H ,
original work considered 25 different functional groups. Since U C O O H , C H ~U, C O O H , H ~ O and
, ~H~O,COOH simultaneously
, from
then, work on the UNIFAC method has continued in an effort the sum of the two sets of data.
to extend reliability and applicability for separation-process The new UNIFAC parameters are given in Tables I and 11.
design. Organic acids, secondary alcohols, methanol, most of the
We have now determined UNIFAC group-interaction pa- chlorinated hydrocarbons, carbon disulfide, and compounds
rameters for more than 50 different groups. The basis of the with the nitro group were not included previously. The re-
UNIFAC method now covers 70% of all published vapor-liq- maining parts of these tables indicate the revised parameters.
uid equilibrium data for nonelectrolyte mixtures a t low to Note that the ketone, ester, ether, secondary amines, and al-
moderate pressures. cohol groups are redefined by the addition of another meth-
The revised table of group-interaction parameters is based ylene group.
on vapor-liquid equilibrium data for roughly 2500 binary Prediction of Activity Coefficients. A detailed account
systems. The data were chosen from a data set of more than of the agreement between experimental and calculated ac-
4000 binary systems stored on magnetic tape and collected a t tivity coefficients for all of the groups included in UNIFAC
the University of Dortmund (Gmehling and Onken, 1977). will be published elsewhere (Fredenslund et al., 1977). In this
These data are, where appropriate, supplemented with data section we only give a few examples of activity coefficients
from Wichterle et al. (1973). predicted by the UNIFAC method.
The availability of this large data base has enabled us to Tables 111-VI show some of the results from the data cor-
obtain more reliable values for group-interaction parameters relation. We show typical experimental and calculated activity
than those published originally. I t has also made it possible coefficients in the dilute regions for several of the binary
better to ascertain which choice of functional groups results systems. Tables 111-VI clearly show UNIFAC’s ability to
in the best correlation. represent a wide range of mixtures with only few parameters.
The determination of group-interaction parameters from For example, activity coefficients at infinite dilution for binary
the data base proceeds as follows. systems of alkanes and aliphatic ethers with one, two, or three
(1) For thermodynamically consistent data (Christiansen ether groups or dioxane or tetrahydrofuran (Table 111) are all
and Fredenslund, 1975), activity coefficients are calculated correlated extremely well using only two parameters. For those
from isothermal P-x, or isobaric T-x,-y, data. Vapor-phase systems, the infinite-dilution activity coefficients range from
nonidealities are accounted for as shown by Hayden and roughly 1.1to 3.5. Similarly, Table IV shows that no more than
O’Connell (1975). two parameters are needed to represent the phase equilibria
(2) A pair of group-interaction parameters, e.g., U C O O H , C H ~ of alkanes with ketones ranging from acetone to nonanone.
and U C H ~ , C O O Hare
, obtained from a collection of all available, Tables V and VI show excellent agreement between predicted
consistent, relevant experimental data, in this case for systems and observed equilibria in, respectively, alkane-alcohol and
with alkanes (including isomers) and organic acids. The alkane-amine systems, Figure 1 shows results for systems
simplex method (Nelder and Mead, 1965) is used to fit the containing water and the lower organic acids. Again, only two

452 Ind. Eng. Chem., Process Des. Dev., Vol. 16, No. 4, 1977
h

- 1
1 -I
-I

Ind. Eng. Chem., Process Des. Dev., Vol. 16, No. 4, 1977 453
Table 111. Experimental and Calculated Activity Coefficients for Alkane-Ether Systems
Component 1 Component 2 T ,K x1 yl(expt1) yl(ca1cd)
n-Propyl ether n-Heptane 343.2 0 1.09 1.19
n-Heptane n-Propyl ether 343.2 0 1.08 1.19
Dimethoxyethane n-Heptane 343.2 0 2.32 2.16
Heptane Dimethoxyethane 343.2 0 2.11 2.45
Bis(2-methoxyethyl) ether n-Decane 393.2 0 2.62 2.94
n-Decane Bis(2-methoxyethyl) ether 393.2 0 3.06 3.46
Dioxane n-Hexane 353.2 0.024 2.67 2.13
n-Hexane Dioxane 353.2 0.030 2.94 3.11
Dioxane n-Octane 353.2 0.021 2.33 2.14
n-Octane Dioxane 353.2 0.046 3.32 2.93
Dioxane Cyclohexane 353.2 0 2.87 2.49
Cyclohexane Dioxane 353.2 0 2.55 2.58
Tetrahydrofuran Cyclohexane 298.2 0.1 1.77 1.75
Cyclohexane Tetrahydrofuran 298.2 0.1 1.70 1.71

Table IV. Experimental and Calculated Activity Coefficients for Alkane-Ketone Systems
Component 1 Component 2 T ,K x1 yl(expt1) 71(calcd)

Acetone n-Hexane 318.2 0.1 3.37 3.82


n-Hexane Acetone 318.2 0.1 3.40 3.76
2-Butanone n -Heptane 366.9 0.027 3.53 3.29
n-Heptane 2-Butanone 351.9 0.027 3.15 3.53
3-Pentanone n -Heptane 353.2 0.056 2.32 2.04
n-Heptane 3-Pentanone 353.2 0.051 2.24 1.98
5-Nonanone n-Hexane 333.2 0.1 1.68 1.69
n-Hexane 5-Nonanone 333.2 0.1 1.45 1.50

Table V. Experimental and Predicted Activity Coefficients for Alkane-Alcohol Systems"


Component 1 Component 2 T ,K x1 yl(expt1) yl(calcd)
n-Hexane 2-Butanol 333.15 0.094 3.34 3.27
2-Butanol n-Hexane 333.15 0.069 5.17 6.41
n-Heptane 1-Octanol 293.15 0.1 2.49 2.31
n-Heptane 2-Octanol 293.15 0.1 2.30 2.30
n-Heptane 3-Octanol 293.15 0.1 2.24 1.93
n-Heptane 4-Octanol 293.15 0.1 2.26 1.93
1-0ctanol n-Heptane 293.15 0.1 4.17 3.88
2-Octanol n-Heptane 293.15 0.1 3.69 3.87
3-Octanol n-Heptane 293.15 0.1 3.42 2.81
4-Octanol n-Heptane 293.15 0.1 3.41 2.81
Decane 1-Propanol 363.15 0.0503 5.01 6.11
1-Propanol Decane 363.15 0.214 2.82 3.49
n-Hexane Ethanol 298.15 0.1 5.92 5.35
Ethanol n -Hexane 298.15 0.1 7.13 8.01
Methanol n-Hexane 333.75 0.138 6.08 6.46
n-Hexane Methanol 333.05 0.054 12.21 13.06
0 NOTE: Of the systems shown here only ethanol-hexane was included in the data base.

Table VI. Experimental and Calculated Activity Coefficients for Alkane-Primary Amine Systems
Component 1 Component 2 T ,K X1 yl(expt1) yl(calcd)
Methylamine n -Nonane 293.2 0 3.55 3.72
n-Nonane Methylamine 293.2 0 7.90 7.97
Ethylamine n -Hexane 293.2 0 2.95 2.62
n-Hexane Ethylamine 293.2 0 2.64 2.80
n-Propylamine n-Hexane 293.2 0 2.60 2.47
n-Hexane n-Propylamine 293.2 0 2.56 2.34
n-Butylamine n-Hexane 333.2 0.128 1.66 1.68
n-Hexane n-Butylamine 333.2 0.104 1.77 1.67
n -Hexylamine n-Hexane 333.2 0.123 1.55 1.49
n-Hexane n-Hexylamine 333.2 0.094 1.49 1.42

parameters are required to represent the interaction between Although no ternary (or higher) systems were included in
COOH and H 2 0 groups for the systems shown. Since data for the data base, predicted multicomponent equilibria are in
systems including formic acid are scarce, it is not certain good agreement with experiment. Table VI1 shows predictions
whether the formic acid-water prediction shown in Figure 1 for the system chloroform-methanol-ethyl acetate. This
is fortuitous. As more data become available, it may be ad- system is unusual, because it exhibits both large positive and
vantageow to include formic acid as a special group in the negative deviations from Raoult's law. Table VI11 shows
same manner as that for methanol. predicted and experimental activity coefficients for the system

454 Ind. Eng. Chem., Process Des. Dev., Vol. 16, No. 4, 1977
Table VII. Experimental and Calculated Activity Coefficients a t 1 Atm for the System Chloroform(1)-Methanol(2)-
Ethyl Acetate(3) (Nagata, 1962)
Y1 Y2 Y3
x1 X2 Exptl UNIFAC Exptl UNIFAC Exptl UNIFAC
0.075 0.077 0.541 0.612 2.32 2.47 0.985 0.999
0.030 0.898 1.82 2.10 1.01 1.01 2.41 2.24
0.608 0.234 1.09 1.08 2.11 2.17 0.570 0.637
0.540 0.075 0.982 0.972 3.50 3.81 0.611 0.588
0.527 0.219 1.00 1.01 2.15 2.26 0.701 0.739
a The average absolute deviation between yl(exptl)and y,(calcd) is 0.011mole fraction.

Table VIII. Calculated Activity Coefficients for the System 1,2-Dichloroethane(l)-l-Propanol(2)-Toluene(3)-


Acetone(4) a t 700 mmHe (Ma et al.. 1969) Fitted with Binarv Data (Wilson Eauationk Predicted bv the UNIFAC Method
Exptl activity coefficients Calcd activity coefficients
x1 x2 13 Y1 Y2 Y3 Y4 Yl Y2 Y3 Y4

0.613 0.232 0.056 1.17 1.84 1.26 0.79 1.15 1.83 1.34 0.71 Wn
1.10 1.96 1.15 0.91 Ub
0.515 0.168 0.097 1.09 2.07 1.32 0.87 1.09 2.09 1.32 0.80 W
1.05 2.16 1.15 0.95 U
0.307 0.335 0.281 1.19 1.60 1.34 0.90 1.26 1.51 1.33 0.82 W
1.16 1.63 1.26 1.01 u
0.307 0.207 0.419 1.09 2.07 1.18 0.99 1.15 1.95 1.18 0.84 W
1.05 2.14 1.11 1.01 u
0.202 0.512 0.251 1.40 1.28 1.56 0.97 1.50 1.22 1.53 0.91 W
1.38 1.29 1.51 1.10 u
0.251 0.267 0.252 1.12 1.67 1.38 0.96 1.15 1.58 1.33 0.88 W
1.10 1.73 1.26 1.02 u
0.064 0.039 0.861 0.98 3.35 1.02 1.43 1.11 3.80 1.01 1.26 W
0.98 3.57 1.01 1.27 u
0.063 0.089 0.072 0.94 1.75 1.78 0.97 0.87 1.54 1.27 0.99 w
0.93 1.90 1.52 1.01 u
Mean Deviations between Experimental and Calculated Vapor-Phase Mole Fractions
Wilson UNIFAC method Raoult’s law
Component fitted predicted predicted
1 0.011 0.009 0.030 Mean deviation = 5,
=
2 0.010 0.006 0.087 ( 2 I):
k=l
h(expt1) - y,.k(calcd)
3 0.004 0.007 0.036 ilr = no. of data points
4 0.014 0.016 0.022
W = Wilson’s equation. U = UNIFAC method.

1,2-dichloroethane-l-propanol-toluene-acetone (Ma et al., was necessary to include liquid-liquid equilibrium data in the
1969). The authors give Wilson parameters based on the data, data base.
and activity coefficients calculated from these parameters are Although the parameters are almost exclusively based on
also shown. The UNIFAC method and the Wilson equation vapor-liquid equilibria, the present version of the UNIFAC
(with the parameters of Ma et al.) give overall mean deviations model can predict liquid phase-splitting as indicated in Figure
in vapor-phase mole fractions of 0.01;by contrast, using 3. For design of a liquid-liquid extraction cascade, one needs
Raoult’s law does not result in satisfactory predictions. to know the relative distribution of the solute(s) between the
Systems Containing Branched Hydrocarbons and two liquid phases. Since the solute mole fractions are often
O t h e r Isomers. Fredenslund e t al. (1975)considered the ef- small, the relative error in the predicted mole fractions-and
fect of isomerization to only a limited extent. Since then much hence in the solute distribution ratios-may be large. This
experience has been gained in this respect and, with the new problem is encountered by all methods for calculating activity
UNIFAC parameters, the phase behavior of systems con- coefficients and the UNIFAC model is no exception. There-
taining isomers may usually be predicted with good results. fore, UNIFAC is usually not suitable for accurate liquid-liquid
Table V shows that secondary and primary alcohols are rep- extraction calculations. However, UNIFAC often predicts
resented very well using the same set of parameters; similarly, phase-splitting with sufficient accuracy for designing heter-
Figure 2 shows good agreement for predicted and calculated ogeneous-azeotropic-distillationcolumns.
activity coefficients for the 2,2,3-trimethylbutane-benzene
system. This favorable result is particularly gratifying because Multicomponent Distillation Design Calculations
no branched alkanes were included in the data base for de- The multicomponent column-design procedure used here
termining the methylene-aromatic hydrocarbon group in- is a modification of the successive approximation method
teraction parameters. described by Naphtali and Sandholm (1971).The distillation
Liquid-Liquid Equilibria. The present set of UNIFAC column configuration is shown in Figure 4.The general design
parameters is, where possible, based on vapor-liquid equi- equations, which include material- and energy balances and
librium data only. For systems such as alkanes with water it Murphree efficiencies coupled with vapor-liquid equilibrium

Ind. Eng. Chem., Process Des. Dev., Vol. 16,No. 4, 1977 455
- UNIFAC Distillate
0 OBSERVED r-bondense-

w Reflux

-
Feed
Fn f n , i

XI
1 2 1
Figure 2. Activity coefficients for the 2,2,3-trimethylbutane(l)-
benzene(2) system at 1 atm (Harrisonand Berg, 1946). This binary
system was not included in the data base.

- UNIFAC
Figure 4. Distillation column configuration.

dence is introduced to save computer time. This introduces


no significant error.
Second virial coefficients are used to calculate vapor-phase
fugacity coefficients. We use the method of Hayden and
O’Connell(l975, see Appendix C) to estimate the virial coef-
ficients for all possible binary interactions a t temperatures
T I and Tz and interpolate the virial coefficients linearly in the
temperature.
X1
(3) Specification of Separation Problem. The user must
A T = 0.0175 specify the number of actual stages, stage efficiencies, feed-
Figure 3. Experimental and calculated vapor phase mole fractions and sidestream locations, feed compositions, flow rates, and
and phase-splitin the system n-heptane(l)-acetonitrile(2) (Werner thermal states, total distillate flow rate, sidestream phase
and Schuberth, 1966). conditions and flow rates, reflux ratio, and column pres-
sure.
Above it was stated that UNIFAC is used to obtain UNI-
relationships, are given in Appendix B. A simplified set of QUAC parameters. Using UNIQUAC parameters rather than
design equations, based on the assumption of constant molal UNIFAC parameters directly is based on the following. (1)
overflow, is also given. The UNIQUAC method, which uses only molecular compo-
The calculation scheme is shown schematically in Figure sitions, is faster than the UNIFAC method, which uses both
5. Input to the calculations is as follows. molecular and group compositions. (2) Tests for systems
(1) Identification of Components to Be Separated. where experimental data are available show that activity
Antoine constants, giving the pure-component vapor pres- coefficients calculated from the UNIQUAC model and esti-
sures, and enthalpies of vaporization must be known for each mated limiting activity coefficients are, on the average, as
component. The latter are not needed for the simplified ver- accurate as those calculated directly using the UNIFAC
sion. method. (3) If experimental information is available for some
(2) Parameters for Determination of Thermodynamic of the constituent binaries, UNIQUAC parameters based on
Properties. If all the necessary UNIFAC group-interaction these data may be incorporated directly into the calcula-
parameters are available, the activity coefficient of every tions.
component i, infinitely dilute in component j for all possible The column calculations can now proceed. The scheme
i-j pairs in the multicomponent system, is calculated using outlined in Figure 5 is that of the simplified version of the
the UNIFAC method. As explained in Appendix A, these program (constant molal overflow). The procedure for the full
calculations are carried out a t two temperatures T I and T2, version is analogous.
typically the saturation temperatures of the least and most The total molar flow rates for each stage are calculated
volatile components. The activity coefficients a t infinite based on the separation problem specification. (In the case
dilution form the basis for estimating the UNIQUAC pa- of the full version, they can only be estimated a t this point).
rameters for the molecular energy interactions. The linearized As an initial estimate, a linear temperature profile is assumed,
temperature dependence, eq 8A, is used rather than the ex- and the phase compositions on each stage are equated to the
ponential form, eq 7A. The calculated infinite-dilution activity nearest feed composition. This gives the zeroth estimate of
coefficients a t the temperatures T I and T2 enable the deter- the independent variables l,,,, u,,,,, and T,, and hence the
mination of Aji (0) and A;i(I).The linear temperature depen- zeroth values of the test functions. The Newton-Raphson

456 Ind. Eng. Chern., Process Des. Dev., Vol. 16,No. 4, 1977
Define mixture to be separated:
Antoine constants Ai Bi Ci
Group counts vk‘i)
Temperature range T1 to T2
--- RAOULT S LAW

*
Calculate activity coefficients at infinite dilution, y?,
at T1 ard Ti using UNIFAC. This also gives ri, q i , and1Ei

to UlilQLlAC eqcation. This gives A ! ? ’ and A!:):


J 1
T.. I A(?’ + A!f).T
J l J1 J1 m \
.c \
0 4 . a
0

~
Specify separation problem: No. of stages, feed stages,
r e f l L x ratio, feed compositions and no. of moles per hour,
amount of b o t t o m s product, sidestreams, and column pressure I 0.3 . P
-I

Calc-late total liquid and vapor molar fiow rates on each


stage: Lr and V,. Estimate initial temperature- and
car.cer,trat
ior. profiles

b 3 5 7 9 11 13 15 17 19 d

.Calculate for each stage n and each component i , using


J.I~I.Q C A C , t h e equilibrium ratio at the last estimated tem- Bottom S t a g e Number TOP
peratdye and concentration i K . . Determine the deriva-
t h e ; o r Zn,: ~ i T hrescect to.!f;uid phase component molar Figure 6. Liquid-phase concentration profiles for 1-propanol and
f l o w rates and temperature toluene with cyclohexane and ethanol in a simple distillation col-
umn.
1
Solve tne ccrrpor.ent naterial balances for each stage with
‘he new values o f t i l e stripping factors C
n,i ‘JnKn,i‘Ln
ample of systems where estimation of activity coefficients is
I - I absolutely essential; the second system was chosen because
of the availability of extensive experimental multicomponent
Are new vapor-liquid equilibrium data (Table VIII); the third system
is an example of extractive distillation.
the previously For the first two systems, feed is introduced on stage 10,
estimated
values? there are 20 stages, and there are no sidestreams. The column
Yes, within
suecified limits configuration for the extractive-distillation example is given
Print resulting distillate and bottoms compositions and in Figure 7.
column temperature- and concentration profiles The System Cyclohexane( 1)-Ethanol(2)-1-Propa-
Figure 5. Procedure for distillation column calculations (simplified nol(3)-Toluene(4). Hydrocarbon-alcohol systems are known
version). to deviate greatly from ideality; minimum-boiling azeotropes
are found in many cases. In this system, no fewer than four
binary azeotropes exist. Predictions of equilibrium ratios in
iteration method, utilizing simultaneous convergence of all the ternary system of components 2,3, and 4, using the UN-
independent variables, is now used to obtain the true values IFAC method, indicate that no ternary azeotropes exist in this
of l,,,, u,,,~,and T,,. Simultaneous convergence is more efficient system.
than a sequential convergence method when the volatilities For this system we choose as feed composition x 1 = 0.1, x p
of the components depend both on temperature and compo- = 0.4,x 3 = 0.4, x 4 = 0.1. The reflux ratio is 2.25, and 60 mol
sitions. out of 100 mol in the feed are specified as distillate. The main
Convergence is normally attained within 4-9 iterations. The results of the column calculations are shown in Table IX. The
computer time per iteration is proportional to the number of toluene-ethanol azeotrope has a marked effect on the results.
stages. In the simplified version (constant molal overflow), Although toluene is the heaviest-boiling component, i t goes
four components and 20 stages typically require 0.4 s per it- into the distillate almost quantitatively. The relative content
eration on an IBM 370/165 computer using a Fortran G or of toluene and ethanol in the distillate corresponds closely to
Watfiv compiler. About 50%of this time is used for algebraic that of the azeotrope. It boils a t 77”C, 20°C lower than the
manipulations, and 50%is used for evaluating activity coef- normal boiling point of 1-propanol, which forms 99% of the
ficients. The full Naphtali-Sandholm procedure requires the bottoms product. A similar calculation, where the UNIQUAC
same number of iterations and typically 2.0 s per iteration for model is replaced by the Wilson equation gives essentially the
the above example. same results. Raoult’s law, on the other hand, gives completely
Similar calculations, where the Wilson equation is used erroneous results. The calculated propanol and toluene con-
instead of the UNIQUAC equation, require slightly less centration profiles are shown in Figure 6.
computer time. The parameters in the Wilson equation (or Computations performed with the full version of the column
any other two-parameter expression for activity coefficients) design procedure indicate that the assumptions of constant
may be calculated from the UNIFAC parameters in the same molal overflow and ideal vapor phase are justified in this
manner as that used to determine UNIQUAC parameters. case.
Results of Distillation Calculations (Simplified Ver- The System 1,2-Dichloroethane( 1)-1-Propanol(2)-
sion). Distillation-column design calculations assuming Toluene(3)-Acetone(4). Predicted activity coefficients using
constant molal overflow and ideal vapor phase were carried the UNIFAC method are given in Table VIII, which also
out for three different systems: cyclohexane(1)-ethanol(2) shows experimental activity coefficients.
-l-propanol(3)-toluene(4); 1,2-dichloroethane(l)-l-propa- Essential results of the column calculations are given in
nol(2)-toluene(3)-acetone(4); and methylcyclohexane(1)- Table IX together with those based on Raoult’s law and those
toluene(2)-phenol(3). The first system was chosen because using the Wilson equation instead of the UNIQUAC equa-
it exhibits large deviations from ideality and is thus an ex- tion.

Ind. Eng. Chem., Process Des. Dev., Vol. 16,No. 4,1977 457
c d=:2.21- mles/h I I

?12=76.74
1
=o
-
Feed stage 12
moleslh
2
;o
21
stages ZC'4P:TER TIYE:
x -1.0 p = l atn 5 . 4 s e c a n t-le 191: ? 7 0 / 1 C 5

ST3RA3E SEQLIRE

.%e+ stage 7
_.c
Fy23.26 mles/h
xl=0.5
x 2--0 ' 5
x,=o

cb =W.:f no:espb

T h e f r a m e d nunoe?s a r e c a l c u l i t e 3 FeiUltS S t a s e YU7ber


Figure 7. Column configuration and liquid phase concentration profiles for extractive distillation of methylcyclohexane(1)and toluene(2) with
phenol(3).

Table IX. Component Flow Rates of Distillate and Bottoms. Comparison of Calculations Based on UNIFAC. Raoult's
Law and Wilson's Equation. Simplified Version; 20 Stages; Feed Stage: 10; Pressure: 1 atm; No Sidestreams
UNIQUAC Raoult's law WilsonC
Feed," Distillate," Distillate, Bottoms, Distillate, Bottoms, Distillate, Bottoms,
mol of mol Refluxa molof mol of mol of mol of mol of mol of
comp. i total ratio comp. i comp. i comp. i comp. i comp. i comp. i
Cyclohexane(l)-Ethanol(2)-l-Propanol(3)-Toluene(4)
(1): 10 60 (1):10.0
2.25 (1): 0.0 (1): 9.7 (1): 0.3 (1):10.0 (1): 0.0
(2):40 (2):39.7 (2): 0.3 (2):39.9 (2): 0.1 (2):39.9 (2): 0.1
(3):40 (3): 0.3 (3):39.7 (3):10.3 (3):29.7 (3): 0.7 (3):39.3
(4):10 (4): 9.9 (4): 0.1 (4): 0.1 (4): 9.9 (4): 9.3 (4): 0.7
1,2-Dichloroethane(l)-l-Propanol(
2)-Toluene(3)-Acetone(4)
(1):45 50 4 (1): 34.2 (1): 10.8 (1):42.3 (1): 2.7 (1): 35.5 (I): 9.5
(2):45 (2):10.8 (2):34.2 (2): 2.6 (2):42.4 (2): 9.6 (2):35.3
(3): 5 (3): 0.0 (3): 5.0 (3): 0.0 (3): 5.0 (3): 0.0 (3): 5.0
(4): 5 (4): 5.0 (4): 0.0 (4): 5.0 (4): 0.0 (4): 4.9 (4): 0.1
a Specified by user. Parameters estimated with UNIFAC method. c Parameters fitted with binary data.

Table X. Component Flow Rates of Distillate and to-stage solution of a n extractive-distillation process for the
Bottoms. Comparison of Calculations Based on UNIFAC above two components using phenol as a solvent. Using UN-
and Raoult's Law. Full Version." Ethanol( 1)- IFAC coupled with the simplified version of the distillation
I-Propanol(2)'-Water(3)-Acetic Acid(4) program, we obtain the same results as those presented by
Mol in Mol in bottoms Smith. The process and resulting concentration profiles are
Component distillate product Method shown in Figure 7.The solution of this problem required seven
iterations and 3.4 s on the IBM 370/165 computer.
1 24.8 0.2
2 14.3 10.7 Raoult's Results of Distillation Calculations (Full Version). If
3 10.4 14.5 law excess enthalpy effects cannot be neglected, if stage efficien-
4 0.4 24.6 cies are to be included, or if the vapor phase is strongly noni-
deal, it is necessary to use the full Naphtali-Sandholm pro-
1 22.4 2.6 Full cedure. (Slight vapor phase nonidealities can be included in
2 9.2 Naphtali-
15.7 the simplified version by inserting the vapor-phase fugacity
3 18.4 Sandholm
6.6
4 0.0 25.0 with coefficient in the stripping factor (see Appendix B,eq 7B)and
UNIFAC subsequently neglecting derivatives of the test functions with
UReflux ratio: 2.0;distillate: 50 molh; number of stages: 15;feed respect to vapor-phase composition). Systems containing
stage: number 7;feed: (1):25 mol/h; (2):25 mol/h; (3):25 mol/h; organic acids are examples of systems where the full Naph-
(4):25 mol/h. tali-Sandholm procedure must be used. Liquid-phase noni-
dealities are, as before, accounted for via the UNIFAC cor-
relation, and vapor-phase nonidealities, including organic-acid
In the example given here, the feed contains mainly com- dimerization, are determined as explained in Appendix C. The
ponents 1 and 2, and the reflux ratio is 4. Separation between results shown in Tables X and X I are for the case of a simple
these two components is poor due t o the 1,2-dichloroethane- distillation column with one feed stream at stage 7 , no side-
1-propanol azeotrope (boiling point 68 "C) going into the streams, and 15 stages. Figure 8 shows results for a more
distillate. complex column.
The UNIQUAC results agree well with those using the The System Ethanol(l)-l-Propano1(2)-Water(3)-
Wilson equation. However, Raoult's law predicts a different Acetic Acid(4). For a n even feed of 25 mol/h of each com-
(and erroneous) distribution ratio for components 1and 2. ponent to a column as specified in Table X, one might expect
The System Methylcyclohexane( l)-Toluene(2)-Phe- a split between 1-propanol and water, such that most of the
nol(3). Methylcyclohexane and toluene are difficult t o sepa- 1-propanol goes into the distillate, most of the water into the
rate by simple distillation. Smith (1963) presents a stage- bottoms product, Raoult's law gives this result. However, as

458 Ind. Eng. Chem., Process Des. Dev., Vol. 16, No. 4, 1977
Table XI.Component Flow Rates of Distillate and
Bottoms. Comparison of Calculations Based on UNIFAC
and Raoult’s Law. Full Version.n Benzene( 1)-1-
Propanol(2)-Toluene(3)-Acetic Acid(4)
I\I
Compo-
nent Distillate Bottom Method
I 1 24.95 0.05 Raoult’s
2 9.77 0.23 law 1 stages 1
3
4
5.39
9.89
4.61
45.11
lf7i = 1
yi = 1
Constant
1 n n , =O.7
I Sidestream (Saturated
stage 9 ,
Liquij:

I1 1 25.00 0.00
2 9.43 0.57 molal flow
3 10.00 0.00 (oi =1
4 5.57 49.43 yi from UNIFAC
F,:5L mcles/h
I11 1 25.00 0.00 Constant
2 6.95 3.05 molal flow
3 9.99 0.01 Pi f 1
4 8.06 46.94 yi from UNIFAC
IV 1 25.00 0.00 Full
2 6.96 3.04 Naphtali-
3 9.99 0.01 Sandholm
4 8.05 46.95 Pi f 1
yi from UNIFAC X i 4 ’ J T E F , TI:.iE: 17.4 s e c on t h e IBK :70/155
Reflux ratio: 2.0; distillate: 50 mol/h; number of stages: 15;
feed stage: number 7; feed: (1): 25 mol/h; (2): 10 mol/h; (3): 10 STORAGE R E Q U I R E M E N T : e 8 K Bytes.
mol/h; (4): 55 mol/h.
-he framed n u n t e r s a r e c a l c d i a t e d results.
Figure 8. Distillation results for the system ethanol(1)-1-propa-
the UNIFAC correlation correctly predicts, water has a much nol(2)-water(3)-aceticacid(4).
higher activity in acetic acid than 1-propanol has, and hence
most of the 1-propanol goes into the bottoms product. In these
calculations, we have neglected the effect of the relatively slow Omissions in the group-parameter table clearly indicate
esterification of the alcohols with acetic acid. where new experimental data are needed. As new and better
Figure 8 shows a summary of results for a complex column, phase-equilibrium data become available, the UNIFAC
where Murphree stage efficiencies of 0.7 were included. This method, appropriately revised, will increase its accuracy of
required five iterations and 17.4 s computer time on the IBM predictions and, more important, increase its range of appli-
370/165 computer. cability. However, even a t this incomplete state of develop-
The System Benzene(l)-l-Propano1(2)-Toluene(3)- ment, it is evident that UNIFAC may be useful in at least some
Acetic Acid (4).Table XI shows results for the case of an of those commonly encountered industrial situations where
uneven feed: 25 mol/h benzene, 10 mol/h each of 1-propanol required data are lacking and where, for commercial reasons,
and toluene, and 55 mol/h of acetic acid. These results show a reasonable design is needed quickly. In chemical engineering
the effects of successively including corrections for liquid- process design there is always a need for good data, seasoned
phase nonideality (II), vapor-phase nonideality (III), and judgment, and sound experience. UNIFAC is no substitute
excess enthalpy effects (IV). I t is essential to include correc- for these requirements but it can contribute toward meeting
tions for both liquid and vapor-phase nonideality, whereas the their demands.
inclusion of enthalpy balances has only limited effect on the A description and print-out of the UNIFAC programs for
calculations. For most of the systems we have studied, the estimating activity coefficients and the computer programs
inclusion of enthalpy balances has little effect on the product which interface UNIFAC with distillation-column design is
compositions. They may, however, be of much importance in given by Fredenslund et al. (1977).
the determination of internal column flow rates.
Acknowledgment
Conclusions The authors thank Mr. Vagn Hansen for carrying out the
This work has indicated that UNIFAC provides a powerful “full version” of the Naphtali-Sandholm column calculations,
tool for the chemical design engineer. As outlined and illus- Mr. Thomas Anderson for his assistance in reducing to prac-
trated here, it is possible to construct a completely comput- tice the correlation indicated in Appendix C, the National
erized procedure for process design of a multicomponent Science Foundation and Statens teknisk-videnskabelige
distillation column, including estimation of phase equilibria. Forskningsraad for partial financial support, and Professor
The UNIFAC method for calculating activity coefficients Ulfert Onken (University of Dortmund) for his active interest
interfaces easily with iteration techniques for distillation- in this work.
column computations.
While the engineering utility of UNIFAC has been dem- Appendix A
onstrated, its utility inevitably rests on the quantity and T h e UNIQUAC a n d UNIFAC Models. Equations giving
quality of UNIFAC group-interaction parameters. These, in the activity coefficients as functions of composition and
turn, depend on the quantity and quality of experimental temperature are here stated very briefly, both according to the
phase-equilibrium data. The essence of UNIFAC is that it molecular UNIQUAC model (Abrams and Prausnitz, 1975)
interprets and correlates such data to enable predictions of and according to the group-contribution method, the UNI-
phase equilibria in previously studied mixtures under new FAC model (Fredenslund et al., 1975). Both models have a
conditions and in mixtures that have not been studied a t all combinatorial contribution to the activity coefficients, es-
experimentally. sentially due to differences in size and shape of the molecules,

Ind. Eng. Chem., Process Des. Dev., Vol. 16, No. 4, 1977 459
and a residual contribution, essentially due to energetic in- group-composition variable, e k , is now the group fraction of
teractions group k in pure fluid i.
In yi = In y i c + In y i R
(1A)
111. Enthalpy of Mixing. One may obtain enthalpies of
mixing from the UNIQUAC and UNIFAC models. From
combinatorial residual classical thermodynamics, it follows that
I. Combinatorial Part. The combinatorial contribution
is the same in the two models

t
li = - (ri
2
- q i ) - (ri - 1); t = 10 Substituting eq l A , 2A, and 6A into eq 13A we obtain

Pure-component parameters ri and qi are, respectively,


measures of molecular van der Waals volumes and molecular The parameter T,, is given by eq 8A.
surface areas. They are calculated as the sum of the group- Equation 14A provides only a crude estimate when, as done
volume and group-area parameters, R k and Qk here, parameters (U,, - U,,) and (ulL- u,,) are determined
from vapor-liquid equilibrium data and are assumed to be
r; = Uk (i)Rk; qi = Uk (’)& (44
independent of temperature.
k k
where V k ( I ) , always an integer, is the number of groups of type Appendix B
h in molecule i. Group parameters R k and Qk are obtained
from van der Waals group volumes and surface areas, vk and Equations for Multicomponent Distillation Design. The
Ah, given by Bondi (1968) distillation column configuration is shown in Figure 4. Dis-
tillation columns with up to 50 (actual) stages, up to ten
Rk = Vk/15.17; Qk = Akl(2.5 X lo9) (5A) components, a partial condenser, and any number of feed- and
sidestreams may be considered. In our “operating column
11. Residual Part. A. UNIQUAC Equation. analysis” (King, 1971, p 520), the user must specify: (a) the
number of actual stages, (b) stage efficiencies, (c) feed- and
(6.4) sidestream locations, (d) feed compositions, flow rates, and
thermal states, (e) distillate flow rate, (f) sidestream phase
conditions and flow rates, (g) reflux ratio, and (h) column
(7-4) pressure. The nomenclature for an arbitrary stage n ,which
includes the possibility of feed- and sidestreams, is given
where u,, is a characteristic energy for the j-i interaction. For below.
a small temperature range, the effect of temperature on T,, can
be approximated by a linear form
T,, = A,, (O) + A,, ( l ) * T (8.4)
where A,,(O) and A,, ( l )are coefficients related to (u,, - u,,).
B. UNIFAC Method (Group-Contribution).

In y,R = Uk(’)[lnr k - In r k “ ) ] (9.4)


k
all groups
I’k is the group residual activity coefficient, and r k (I)is the
residual activity coefficient of group k in a reference solution
containing only molecules of type i.

E Um(i)X, Ln vfl- 1
8, = ’- Q m X m (1W
EQ,X,’ x m = CC U k ( i ) x i hn ln,i un- I ,i Hn- 1
n i k
Subscript n: flow from stage n , n = 1 , 2 , . . . , N ; subscript i:
X , is the fraction of group m in the mixture. component i, i = 1 , 2 , . . . M ; H = vapor phase enthalpy; h =
liquid phase enthalpy; H I = feed enthalpy; V = total vapor
Qnm = exp[-(a,,/T)I *( 12A)
flow; u = component vapor flow; L = total liquid flow; 1 =
Parameter anm characterizes the interaction between groups component liquid flow; F = total feed; f = component feed;
n and m. For each group-group interaction, there are two SL= liquid sidestream; Sv = vapor sidestream.
parameters: anm # amn. No ternary (or higher) parameters General Relations w i t h Enthalpy Balances. For stage
are needed to describe multicomponent equilibria. n one obtains the following set of dependent relationships
Equations 10A-12A also hold for In rk ( i ) , except that the (test functions F k ( f l , i ) which
) must be satisfied.

480 Ind. Eng. Chem., Process Des. Dev., Vol. 16, No. 4, 1977
Component Balances. (Total: N X M relations) The assumption of equality of heats of vaporization and zero
excess enthalpy is tantamount to the assumption of “constant
molal overflow”. I t is not necessary to assume ideal vapor
phase a t this point. However, in the solution of eq 9B it is
convenient to assume that {,i does not depend upon the vapor
~,,i - ~,-1,j - In+l,j - fn,j =0 (1B) phase composition.
Newton-Raphson Iteration. Solving eq 4B or 9B means
Enthalpy Balances. (Total: N relations)
finding the set of values of the independent variables, x , which
makes the set of test functions become equal to zero
F(x) = 0
H , - Hn-l - hn+l- Hf,, = 0 (2B) In Newton-Raphson iterations, a new set of values of the test
functions, FcneW),are generated from a previous estimate in
Equilibrium Conditions with Murphree Stage Ef- the following fashion
ficiencies, v,,,. (Total: N X M relations).
F(new)(x(new))
= F(old)(X(old))
F.?(n,i)= vn,iKn,lVnL,i/Ln - on,,
+ (1 - vn,i)un-1,iVn/Vn-1 = 0 (3B)

This equation is used to estimate x(new). When (x(new) - x(Old))


is sufficiently small, the correct set of values of x has been
The above relationships comprise a vector of test functions
found, and the iteration stops.
The variations between subsequent iterations are limited
as follows: (1) negative component molar flow rates are
equated to zero; (2) component flow rates exceeding L, are
equated to L,; (3) the maximum change in component molar
flow rates is O.25Ln, and (4)the maximum change in the
+
which contains N X (2M 1) elements, and which may be temperature a t each stage, T,, is 10K. The limits of 0.2515,
solved for equally many unknowns and 10K are chosen arbitrarily.

(t}
The Jacobi matrix, (&Flax),is here very large, often of the
order 200 X 200. The evaluation is greatly facilitated by the
x= fact that the conditions on stage n only is influenced directly
+
by the conditions on stages n 1and n - 1. As a result, the
Jacobian becomes block-tridiagonal in structure, which per-
mits rapid solution by block elimination.
where the vector 1 contains all the elements l,,,, v all elements The derivatives of test functions with respect to tempera-
and T all elements T,. ture are found analytically, those with respect to component
Once all I,,,, u ~ ,and
~ , Tn’sare known, the product compo- flow rates numerically.
sitions and product flow rates and the concentration- and
temperature profiles in the column follow readily.
Simplified Version (Constant Molal Overflow). If one Appendix C. Determination of Pure-Component
assumes Properties and Vapor-Phase Nonideality
Pure-Component Reference Fugacity. The UNIFAC
q n I = 1, Pn,i = 1, and AHnM= 0, all n and i, and correlation applies to vapor-liquid equilibria a t normal
s l w = S 2 w .. . . = u M v a P
pressures. The pure-component reference fugacity is calcu-
eq 1B and 3B may be combined as follows lated from eq 3 in the text.
Antoine constants are used to calculate P,”,and values for
pis are obtained from the correlation by Hayden and O’Con-
ne11 (197~4,which includes components deviating strongly
from ideality due to chemical interactions.
Vapor-Phase Nonideality. A t normal pressures, vapor-
phase fugacity coefficients of pure components and compo-
nents in a mixture may be calculated from second virial
coefficients, B,,

where P,,,s is the pure-component vapor pressure of compo-


nent i on stage n .
One other set of independent relations is in this case given
by the mass balance on each stage
F5(n)= Ln - ln,1 - ln,2 - . . . - l n ,=
~ 0 (8B) M is the number of components in the system.
Equations 6B and 8B comprise N X ( M + I ) relations In Hayden and O’Connell’s generalized correlation for
predicting second virial coefficients, the overall second virial
coefficient is considered as a sum of contributions form free
pairs of molecules, metastable pairs, physically bound pairs.
and chemically bound pairs
containing equally many independent variables

(10B)
and equations are presented for each contribution. For asso-

Ind. Eng. Chem., Process Des. Dev., Vol. 16, No. 4 , 1977 461
ciating components such as carboxylic acids, the dimerization un,i = vapor-phase molar flow rate of component i from stage
equilibrium constant, K p , is obtained from n
u; = molar liquid volume of component i
K = - B b o u n d -k Bmetastable -k Bchern. xi = liquid-phase mole fraction of component i
P (4C) Xk = group fraction of group k
RT yi = vapor-phase mole fraction of component i
z = lattice coordination number, here equal to 10
The Hayden-O'Connell generalized correlation for Bi; is ap-
plicable to most systems where the UNIFAC method is Greek Letters
presently used. Input to the correlation consists of: (a) pure- y I = activity coefficient of component i
component information: critical temperature and pressure, rk = activity coefficient of group k
Thompson's mean radius of gyration, dipole moment, and an rk( i ) = activity coefficient of group k in pure component i
association factor, which depends only upon the type of as- pi = fugacity coefficient of component i
sociating group (hydroxyl, amine, carboxylic acid, etc.); (b) @i = segment fraction of component i
mixture information: solvation factor; (c) the system tem- U k (i) = number of groups of kind k in molecular species i

perature. Hayden and O'Connell give the input parameters, 0; = area fraction of component i
(3k = area fraction of group k
including association and solvation factors, for a large number T,> = UNIQUAC parameter, see eq 7A
of systems. qnrn= UNIFAC parameter defined in eq 12A
For systems containing carboxylic acids, we have chosen to = stripping factor of component i on stage n
neglect Bfreeand use the values of K p given by Tamir and
Wisniak (1975) and Prausnitz (1969, p 138) to represent the L i t e r a t u r e Cited
vapor-phase nonideality. Abrams. D. S., Prausnitz, J. M., AlChEJ., 21, 116 (1975).
Bondi, A., "Physical Properties of Molecular Crystals, Liquids, and Glasses",
Wiley, New York. N.Y., 1968.
Christiansen, L. J., Fredenslund, A , , AIChEJ., 21, 49 (1975).
Nomenclature Derr, E. L., Deal, C. H., I. Chem. E. Symp. Ser. No. 32, 3, 40 (1969).
Fredenslund, A., Jones, R. L., Prausnitz, J. M., AIChEJ., 21, 1086 (1975).
anm = UNIFAC binary interaction parameter (related to Fredenslund, A., Gmehling. J.. Rasrnussen, P., "Vapor-Liquid Equilibria using
qnrn 1 UNIFAC", Elsevier, Amsterdam, 1977.
A;; = UNIQUAC parameter, see eq 8A Gmehling, J., Onken, U., "Vapor-Liquid EquilibriumData Collection", DECHEMA
Bi - second virial coefficient for i-j interaction Chemistry Data Series, Frankfurt am Main, 1977; see also Onken, U.,
Gmehling, J., Chem. Ing. Tech. (1977).
fid pure-component reference fugacity of component i Guggenheim, E. A., "Mixtures", Clarendon Press, Oxford, 1952.
=
fn,i molar feed rate of component i on stage n Harrison, J. M.. Berg, L., Ind. Eng. Chem., 38, 117 (1946).
F, = molar feed rate on stage n Hayden. J. G., O'Connell, J. P., lnd. Eng. Chem., Process Des. Dev., 14, 209
H, = enthalpy of vapor from stage n (1975).
Ito, I., Yoshida, F., J. Chem. Eng. Data, 8, 315 (1963).
h, = enthalpy of liquid from stage n King, C. J., "Separation Processes", McGraw-Hill. New York, N.Y., 1971.
AHM = enthalpy of mixing Langmuir. I., "The Distribution and Orientation of Molecules", Third Colloid
M v a p = enthalpy of vaporization Symposium Monograph, The Chemical Catalog Co., Inc., New York, N.Y.,
1925.
K,,i = equilibrium ratio of component i on stage n Ma, K. T., McDermott, C., Ellis, S. R. M., 1. Cbem. E. Symp. Ser. No. 32, 3, 89
1,,i = liquid molar flow rate of component i from stage n (1969).
1, = pure-component constant defined in eq 3A Nagata, I., J. Chem. Eng. Data, 7, 307 (1962).
L , = total liquid molar flow rate from stage n Naphtali, L. M.,Sandholm, D. P., AlChEJ., 17, 148 (1971).
Neider, J. A., Mead, R., Comput. J., 7, 308 (1965).
Pis = pure-component vapor pressure of component i Prausnitz, J. M., "Molecular Thermodynamics of Fluid-Phase Equilibria",
qi = pure-component area parameter of component i Prentice-Hall, Englewood Cliffs, N.J., 1969.
Qk = group area parameter for group k Ronc, M., Ratcliff, G. A., Can. J. Chem. Eng., 53, 329 (1975).
ri = pure-component volume parameter of component i Sebastiani, E., Lacquaniti, L., Chem. Eng. Sci., 22, 1155 (1967).
Smith, B. D., "Design of Equilibrium Stage Processes", pp 424-428,
R k = group volume parameter of group k McGraw-Hill, New York, N.Y., 1963.
R = gas constant Tamir. A,, Wisniak, J., Chem. Eng. Sci., 30, 335 (1975).
SnV = vapor-phase sidestream from stage n Werner, G., Schuberth, H., J. Prakt. Chem., 31, 5 (1966).
S , L = liquid-phase sidestream from stage n Wilson, G. M., Deal, C. H., Ind. Eng. Chem., Fundam., I , 20 (1962).
Wichterle, I., Linek, J., Hala, E., "Vapor-Liquid Equilibrium Data Bibliography",
T = temperature Elsevier, Amsterdam, 1973; Supplement I, 1976.
u,; = UNIQUAC binary interaction parameter (related to
~ , and
i A .i) Receiced for reuieu: November 5. 1976
V , = totai vapor molar flow rate from stage n Accepted April 8, 1977

462 Ind. Eng. Chem.. Process Des. Dev., Vol. 16, No. 4, 1977

You might also like