SYNTHESIS OF DISTILLATIONBASED
SEPARATION SYSTEMS
Arthur W. Westerberg and
Oliver Wahnschafft*
Department of Chemical Engineering and the
Engineering Design Research Center
Carnegie Mellon University
Pittsburgh, Pennsylvania
I. Introduction
11. The Richness of the Solution Space
111. Assessing the Behavior of a Mixture
A. Azeotropic Behavior: InfiniteDilution KValues
B. Liquidniquid Behavior: InfiniteDilution Activity
Coefficients
IV. Separating Nearly Ideal Systems
A. Analysis
B. System Synthesis for Nearly Ideal Systems
V. Separating Highly Nonideal Mixtures
A. Azeotropic Separation: Example 1
VI. Synthesis Discussion
A. AnalysisDriven Synthesis
B. Impact of Number of Species on Representation,
Analysis, and Synthesis Methods
C. In Summary
VII. Preanalysis Methods
A. EquilibriumPhase Behavior
B. Distillation Column Behavior
VIII. Synthesis Method for Nonideal Mixtures
A. Azeotropic Separation: Example 2
B. Azeotropic Separation: Example 3
IX. More Advanced Preanalysis Methods
A. Species Behavior
B. Limiting Simple Distillation Column Behavior
C. Extractive Distillation
X. Postanalysis Methods: Column Design Calculations
Acknowledgments
References
64
66
69
69
73
75
75
80
90
91
94
94
94
97
98
98
105
107
108
121
131
131
140
157
166
167
167
*Currently as AspenTech, Cambridge, Massachusetts.
63
ADVANCES IN CHEMICAL ENGINEERING, VOL. 23
Copyright 0 1996 by Academic Press, Inc.
All rights of reproduclion in any form reserved.
63
ARTHUR W. WESTERBERG AND OLIVER W A H N S C H A F n
This tutoricrl piper is ( I reiien. of recent advances in the synthesis
idecil mid notiideal distillLitionbcrsed separation systenis. We start
hp .shon.ing that the space of alterriati\~eseparatiori processes is enorr i m i s . We tiiscws sinrple nrethods to c ~ l m ci . nti.xttrre
~ ~ ~ either as nearly
ideril or as iionidecil, in rshich case it displuys azeotropic and possibly
liquid/liqiiid behavior.
For nearly ideal mixtures, iiisights based on marginal vapor frows
perttiit the dei~elopitieiitc$ n siinple screening criterion computed using
on/! relutive tdrrtilities arid cortiporieiit,feed flowrates to ,find the better
colirinii .seqiiences ,from m r m g the ninny possible. This criterion expluiiis set~eralof the traditional heuristics.
We clsk how one cuii iiiiviit altertinfive structures to separute izo~iicienl riri.xtirre.s. We present and illustrate an approach with three exatiiples: separatirig nbutanol and stater; sepcircrtirrg acetone, chloro,fi)ri?i. mid heiicene; mid separating npentcine, acetone, methanol, and
\inter. We ,find that these processes ciht*ny.s contciin recycles because
1t.e ure uiiahle to obtain the sharp separutions possible for ideal
c!f
in ixtures.
Nest. ri’e esplore more ad\*anced inethods to assess the behai’ior of
cwiiples inirtirres. We discuss niw algorithms to jirid all azeotropes ,for
ci mixtiire: ii’e ul.so discuss the im~hleiiiof.finding the regions for liquid
liquid behtwior.
Esuinple problems ure included to highlight the need to estimate the
errtire set of products that ccin be reciched.for a given feed when using
( I purticvilar h p e o f sel~arutioiiunit. We shot11 that readily coniputed
distillrticrn ciirves and pinch point cim~esulloit. 11s to identifi the eritire
rc.cichnble region f.r simple ririrl e.rtructib,e distillation for ternary
niixtirres. This nnu!\
proi*esthat jinite r e j i u often permits iricreased
sryrircition; n’e ( m i corripute e.ractlp how jiir we can cross socalled
“distillation houiiclaries. For estracthle distillation, we illustrate how
to j i i d tiiinirtiuin .snl~,ent
rates, minitnuin repus rcitios, and, iirtere.stincgly,
~?i~.~itttiiiir
rejliir ratios.
1. Introduction
The goal of this paper is to discuss a methodology for the prelitninary design
of separation processes for liquid mixtures using distillation and extractionbased technologies. We define the preliminary design step to be the one in which
we discover the alternative overall system structures that might be reasonable
for solving the problem. Choosing which alternative to use among those pro
SYNTHESIS OF DISTILLATIONBASED SEPARATION SYSTEMS
65
posed in this step and adjusting the operating conditions to their best values is
a followup step not covered by the preliminary design step; it will not be
covered here to any significant extent. If the vaporlliquid equilibrium behavior
of the species is reasonably ideal, we shall, however, present a simple method
to screen among the distillationbased alternatives to find the likely better ones.
The selection of the better sequences for relatively ideal mixtures is the subject
of several past reviews (Westerberg, 1980, 1985; Nishida et al., 1981; Hlavacek,
1978; Hendry et al., 1973).
We use the word “discover” in the previous paragraph very carefully. In
artificial intelligence the discovery step is the one in which we identify the
building blocks we have to use when solving a problem. For well behaved
mixtures, discovery is simple: we can quickly sketch likely alternatives built
from such building blocks. It is the search among those alternativesof which
there can be a large numberthat is the problem. Often, especially when the
species display azeotropic behavior, the problem is to discover which types of
separation steps can be useda process that requires us to conjecture a method
and then to carry out significant computer calculations (e.g., column or flash
unit simulations) to find if the conjectured method is useful. Proposing any
solution for these harder problems, much less enumerating a number of alternatives, is a difficult task.
The following illustrates the type of problem we would like to be able to
solve:
Design a separation process to split a (liquid) mixture of 25% methanol, 40%
water, and 35% ethanol into the three relatively pure products of methanol,
water, and ethanol.
This particular problem is not a simple one to solve because water and ethanol
form an azeotrope. In this paper, we concentrate first on selecting the better
alternative processes for ideal (or near ideal) mixtures, and then we present a
prototypical method for generating alternatives for nonideal mixtures such as
the one above.
Distillation processes are large consumers of hot and cold utilities. It is often
useful to consider their heat integration (Rathore et al., 1974; Andrecovich and
Westerberg, 1985) where the heat expelled from one column supplies part or
all of the heat needed by another. While important, we shall not consider such
issues here.
This paper assumes the reader is familiar with the standard textbook presentation for staged processes such as that in McCabe and Smith (1976). More
extensive texts on distillation include King (1980), Henley and Seader (198 1),
and Holland (1981). We shall build on the standard background to develop
needed insights for designing separation systems and to understand less conventional singleunit configurations which are often a part of such designs.
66
ARTHUR W . WESTERBERG AND OLIVER WAHNSCHAFFT
II. The Richness of the Solution Space
As a tirst step in design, we must be aware that many more alternatives may
be available for solving a problem than at first seem likely (Westerberg, 1985),
even for wellbehaved mixtures. To confirm this statement, we show that many
alternative structures are available first for a single distillation step and then for
a system of steps. The intent of this section is simply to expand our thinking
about this problem.
We start by sketching a simple twoproduct distillation column, as shown in
Fig. 1. This column is the unit operation most likely to be considered as a
building block for these problems. The twoproduct distillation column splits its
feed into two essentially disjoint product sets. The distillate contains species A,
B. and C (plus, of course. a small amount of D and generally only traces of the
other species) while the bottom product contains predominantly D and E. Heat
is injected at the bottom into the reboiler, the hottest point in the column, and
removed from the condenser at the top, the coldest point in the column. Liquid
flows downward against a flow of vapor upward, with the more volatile species
enriching as one moves up the column.
Typically, we put a specification onto such a column of the following form:
recover 99.8% of the species C in the distillate while recovering 99% of species
D in the bottoms. From a practical point of view, then, the light and heavy keys
for this split are C and D, respectively. In this case, 0.2% of C will make it to
the bottom product and will be the lightest species to make it to the bottom in
more than trace amounts. Similarly, 1 % of D will make it to the top along with
at most a trace of E.
Columns can be run differently from the one shown above. For example, we
may remove heat from or add it to trays within a column. We can add heat by
removing part of the liquid from a tray. vaporizing it, and returning t.he vapor
ABC
DE
FIG. 1. A simple twoproduct column.
67
6
t
Intercooler
E
lnterheater
FIG. 2. Columns with interheating andor intercooling.
just below the tray from which it was removed; or we can remove heat by
withdrawing some vapor, condensing it, and placing it back into the column.
Such a configuration is called an intercooled and/or interheated column, as
shown in Fig. 2. When there are three or more species in the feed, we can alter
the set of product compositions a column can produce by using interheating or
intercooling, as we shall note later.
Another commonly used configuration is a column with a side enricher or a
side stripper. As illustrated in Fig. 3, the column with a side stripper can separate
B
ABC
FIG. 3. Column with a side stripper.
68
ARTHUR W. WESTERBERG AND OLIVER WAHNSCHAFFT
three species. each to a desired purity. This column has one condenser and two
reboilers. At first glance, it looks as if it is the equivalent of one and onehalf
columns; but it is really more like two.
We next consider separating a mixture of ethane, propane, nbutane, and npentane. Two evident alternative solutions are shown in Fig. 4. Three more,
equally “evident” solutions exist. as well as many more less obvious solutions.
There is, however, a problem with these two (and the remaining three) sequences. The top of the column in which ethane is a product is very cold. The
column is pressurized to elevate its temperature. but. even so, it cannot be
brought up to anywhere near ambient temperatures. Thus, we will have to cool
the condenser using refrigeration, which is very expensive.
Is there any way we might reduce the amount of refrigeration required for
the sequence on the left of Fig. 4? One possibility is to recycle some of the 12pentane produced back to the top of the first column, thus providing part or all
of the liquid required for reflux instead of condensing ethane. The normal boiling point for npentane is about 310 K, which means it can be liquefied easily
at ambient temperatures and very slightly elevated pressures. Running the column at several atmospheres pressure can reduce but cannot eliminate the loss
of npentane with the ethane. If this loss cannot be allowed, we can add a few
trays above where wpentane is fed into the column and use a much reduced
amount of ethane as reflux, significantly reducing rather than eliminating the
need for refrigeration. Economics will, of course, indicate whether such designs
are a good idea.
fi
ethane
ethane
“‘“1
I
I
feed

propane
nbutane
npentane
Fici. 3 . Two alternativc separation sequences for separating a mixture of four species.
SYNTHESIS OF DISTILLATIONBASED SEPARATION SYSTEMS
69
111. Assessing the Behavior of a Mixture
It does little good to assume ideal vapor/liquid equilibrium behavior for a
mixture that is highly nonideal. The design proposed assuming ideal behavior
has little to do with the design actually required. There is, in fact, no reason
even to assume it represents a lower cost bound. For example, suppose we wish
to separate water from toluene. Assuming ideal behavior would lead us to design
a distillation column. But in fact, toluene and water “hate” each other, and can
be separated by using a simple decanter. Therefore, the first step in designing a
separation process is often to assess the vapor/liquid/liquid equilibrium behavior
of the species to be separated. We start with an approach that is useful for this
activity, given today’s plethora of excellent computerbased physicalproperties
packages. Having experimental data is the only assured way to know the behavior, and many such data are currently available. Horsley (1973), for example,
has compiled a listing of known azeotropes.
Assume we can have a reliable flash computation available to us and that the
physical property options to be used for the species can be reasonably selected
(a step for which help from an expert consultant within the company may be
necessary). Then the following is a first step we can use to discover if the
mixture displays highly nonideal behavior.
A. AZEOTROPIC
BEHAVIOR:INFINITEDILUTION
KVALUES
We will examine a quick method based on performing two flash computations
to determine the existence of azeotropic behavior for binary mixtures. The hydrogenbonding classes for the species in a mixture are also a clue that the
mixture might exhibit liquidliquid behavior. Indeed, we have used these classes
to find mixtures that display nonideal behavior as illustrative examples.
1. InjniteDilution KValues
We can predict azeotropic behavior as follows from infinitedilution Kvalues.
Using a flowsheeting system, we perform a bubblepoint calculation for each
species in the mixture. Assuming a mixture contains the species A, B, C, and
D, we wish to compute the infinitedilution Kvalues for three of the species in
the remaining one. For example, we perform a flash calculation where A is
dominant and B, C , and D are in trace amounts, using something like a feed
composition of 0.99999, 0.000003333, 0.000003333, 0.000003334. It does not
70
ARTHUR W. WESTERBERG AND OLIVER WAHNSCHAFFT
matter what type of flash computations we do: bubble point, dew point, or a
50/50 split. We do these at the pressure intended for the separation device to
be considered and repeat for all species. Then from each we tabulate the Kvalues for the trace species. The Kvalue for the abundant species is always
unity at infinite dilution. The ratio of the vapor composition to the liquid composition for each of the trace species gives its infinitedilution Kvalue. As we
shall discuss later when considering liquidfliquid behavior, these same flash
computations should also allow us to retrieve infinitedilution activity
coefficients.
Given infinitedilution Kvalues, we want next to examine each species pair
where one is plentiful and the other is in trace amount. As an example, Fig. 5
shows how Kvalues vary for a binary mixture of acetone and chloroform versus
composition. We see that the Kvalue for a drop of chloroform (far left) is less
than unity. The vapor composition yc is less than that for the liquid, xc. The K value for a drop of acetone in chloroform is also less than unity. The mixture
displays a maximumboiling azeotrope.
An interpretation of these Kvalues is as follows:
For KY2 = 2
1
.XI 1
in 2
Maximumboiling azeotrope:
G2< 1 A G, < 1
Minimumboiling azeotrope: KT2 > 1 A
g ,> 1
64
62
60
oc
58
56
0
Acetone
0.2
0.4
0.6
0.8
1
Chloroform
Fw. 5. Liquid and vapor mole fractions v> T at equilibrium for acetone and chloroform. Pressure i < 1 arm.
SYNTHESIS OF DISTILLATIONBASED SEPARATION SYSTEMS
71
Example. Table I (each row comes from one of the above flash computations)
lists infinitedilution Kvalues that were computed for a mixture of acetone,
chloroform, and benzene. The infinitedilution Kvalue for acetone in chloroform
is 0.6 and for chloroform in acetone is 0.4. Both are less than 1; a maximumboiling azeotrope must exist. For chloroform and benzene, the values are 1.5
and 0.4. These values do not suggest the existence of an azeotrope. We assume
normal behavior. A similar conclusion is obtained for acetone and benzene
where the Kvalues are 3.0 and 0.7, respectively.
2. HydrogenBonding Guidelines
Berg (1969), in a paper on selecting agents for extractive distillation, classifies species into hydrogenbonding classes. The deviation from ideality is then
predicted depending on the classes represented in the mixture. Quoting from
Berg, the classes are as follows:
Class I: Liquids capable of forming 3dimensional networks of strong hydrogen bondse.g., water, glycol, glycerol, amino alcohols, hydroxylamine,
hydroxyacids, polyphenols, amides, etc.
Class 11: Other liquids composed of molecules containing both active hydrogen atoms and donor atoms (oxygen, nitrogen, and fluorine)e.g., alcohols,
acids, phenols, primary and secondary amines, oximes, nitro compounds with
alphahydrogen atoms, nitriles with alpha hydrogen atoms, ammonia, hydrazine, HF, HCN (plus nitromethane, acetonitrile even though these form 3dimensional networks; they have weaker bonds than OH and NH bonds in
class I)
Class 111: Liquids composed of molecules containing donor atoms but no
active hydrogen atomse.g., ethers, ketones, aldehydes, esters, tertiary amines
(including pyridine type), nitro compounds and nitriles without alphahydrogen
atoms
Class IV: Liquids composed of molecules containing active hydrogen atoms
but no donor atomse.g., chlorinated hydrocarbons with two or three chlorines
per carbon (CHCl,, CH2C12, CH3CHC12, CH2ClCH2Cl, CH2ClCHClCH2C1, CH2C14HC12)
TABLE I
INFINITEDILUTION
KVALUESFOR MIXTURE
K"
In:
Acetone
Chloroform
Benzene
Acetone
Chloroform
Benzene
1.o
0.6
3.0
0.4 (rnax)
0.7 (normal)
0.4 (normal)
1.o
1.o
1.5.
72
iZRTHUR W. WESTERBERG AND OLIVER WAHNSCHAFFT
Class V: All other liquidsi.e., liquids having no hydrogenbonding capabi1itiese.g.. hydrocarbons. CS2, sulfides, mercaptans, halohydrocarbons not
in class 1V. nonmetallic elements such as iodine, phosphorus, sulfur.
Quoting again. their expected deviation from Raoult's Law are shown in
Table TI.
We can consider our example again. The classes for these species are as
follow~s:
Acetone:
Benzene:
Chloroform:
class I11
class v
class I V
The binarypair behaviors suggested by this article are as follows:
Acetone. benzene
Acetone. chloroform
Chloroform. benrene
III+V
llI+IV
IV+V
d: Quasiideal
b: Always  behavior
d: Quasiideal
As before, we see that the problem in this mixture is the acetone, chloroform
pair. As we have already seen. they have a maximum boiling azeotrope.
TABLE 11
EXPECTEDDEVIATlONs I;ROM RAOLT'SL 4 W
(FROM BERG. 1969)
a. Alw)ays + deviations. frequently limited solubility (min boiling aieotropes if any). Hydrogen
bonds broken only.
h. Always  debiations (tendency for niax boiling
a7eorropes). Hydrogren bonds formed only.
c. Uually + deviations; some very complicated situations. Some will give maximum azeotropes
(from negative deviations). Hydrogen bonds both
formed and broken. but dissocation of class I or
I1 liquid is more important effect.
d. Quasiideal systems. always + deviations or
ideal. Minimum azeotropes only if any. No hydrogen bonds involved.
SYNTHESIS OF DISTILLATIONBASED SEPARATION SYSTEMS
73
B . LIQUIDLIQUID
BEHAVIOR:
INFINITEDILUTION
ACTIVITY
COEFFICIENTS
Mixtures may also form two or more liquid phases at equilibrium. For example, a 50/50 mol% liquid mixture of toluene in water will partition into a
waterrich liquid phase and a toluenerich liquid phase. We just used infinitedilution Kvalues as a means to predict azeotropic behavior. We can argue that
we should use infinitedilution liquid activity coefficients to alert us to the potential for liquidhquid behavior. We do so as follows.
For a liquid mixture at constant temperature and pressure, an equilibrium
state is one that minimizes the total Gibbs free energy for the system (Smith
and Van Ness, 1987). Figure 6 shows a plot of the total Gibbs free energy of
a binary mixture where we form the total from three terms: one that molefraction averages the Gibbs free energy for the two pure species, one that computes the ideal Gibbs free energy of mixing, and one that estimates the excess
Gibbs free energy. We can model this last term by using an empirical relationship such as the Margules equation. We have rescaled the ordinate for this plot
by dividing all terms by RT, where R is the universal gas constant and T is the
absolute temperature.
Suppose the rescaled Gibbs free energy, Gl/RT, for species 1 (left side
Fig. 6) is 0 while for pure species 2 it is 0.5. Then, if the Gibbs free energy
simply mixed with no effect of mixing nor any effect from nonidealities, the
line joining 0 on the right to 0.5 on the left would give us the mixture Gibbs
free energy:
G,"
G1
2 = x,RT
RT
G2
+ x2RT
where x1 + x2 = 1
However, even for an ideal mixture, there is an effect on the Gibbs free energy
from the entropy of mixing, namely,
o  " m t L ? u ! ~ q o ? 0
0
0
0
0
0
0
0
0
mole fraction
FIG. 6. Gibbs free energy for binary mixture.
74
ARTHUR W. WESTERBERG A N D OLIVER WAHNSCHAFFT
Gm
ix
RT

xI In x,
+ x 2 In x2
Note it always has this exact shapethat of an upwardopening “convex” curve
that passes through zero at 0 and 1. If the mixture were ideal, this curve plus
that averaging the pure species Gibbs free energies would assume this upwardopening convex shape.
The following Margules equation is one form in which we can approximately
represent nonideal behavior, allowing us to estimate the excess Gibbs free energy as a function of composition:
We would find. when taking compositions to their limiting values, that
A,I = In(y;);
A 1 2 = In(yq)
where A , ? and A?, are constants for this equation and y
: is the infinitedilution
activity coefficient of species i in the other species. This curve can assume all
sorts of shapes as it is cubic in mole fraction. It too must be zero at both ends.
With a downwardopening concave shape as illustrated here, it starts by canceling only a part of the effect of ideal mixing; then it more than cancels this
effect and takes over in making the G,,,/RT curve switch from convexupward
to concavedownward. It is this switch in shape that indicates liquidliquid behavior. The mixing term approaches its endpoints with an infinite slope so the
G,,,,/RTcurve always starts out in the downward direction, no matter what model
we use to estimate the nonideal behavior.
If our total curve switches to a concavedownward appearance anywhere
along it. as it does here between approximately 10% and 90% B in A, any
mixture with compositions between these two points will break into two liquid
phases at equilibrium. Suppose we compute the total Gibbs free energy for a
50/50 mixture on the C,,,/RT curve. We can get a lower total Gibbs free energy
by breaking the mixture into two mixtures, one at approximately 10% and the
other at approximately 90%. Their total Gibbs free energy is along a straight
line connecting their individual Gibbs free energies. The lowest possible value
would be along a support line that just touches the total curve from below. The
value on this line is below that predicted for the mixture; thus the system can
reduce its total Gibbs free energy by breaking into these two phases.
By carrying out numerical studies, we find that the Margules equation predicts the onset of liquidliquid behavior if either of the following is (approximately) true:
SYNTHESIS OF DISTILLATIONBASED SEPARATION SYSTEMS
75
For example, if yTin2is 0.001, then we need to worry about 1iquidAiquid behavior if yTinlis greater than about 9(0.001)”3 = 0.9.
We can propose to use this guideline to alert us to the potential for liquid
liquid behavior. For example, we might consider the need to check more thoroughly for 1iquidAiquid behavior if we replace the 9 by a 6 and either of these
test passes.
We used infinitedilution activity coefficients of 10 and 20 to create Fig. 6.
Both are greater than 9, so we should expect the Margules equations to predict
liquidhiquid behavior. Water and toluene have infinitedilution activity coefficients in the thousands. They really dislike each other and break into relatively
pure phases. If we examine the total Gibbs free energy curve, we gain the
impression that the curve is totally convexupward; however, there is a slight
downward move at the extremes because of the infinite downward slope of the
mixing term at the extreme compositions. The two liquid phases are almost, but
not quite pure.
1. Relating InfiniteDilution KValues and Activity Coefficients
If we have evaluated infinitedilution Kvalues to test for the existence of
azeotropes, we can use those numbers to get a quick estimate of the corresponding activity coefficients by noting that
If the mixture is at 1 atm, T is the normal boiling point for the plentiful species j .
IV. Separating Nearly Ideal Systems
A . ANALYSIS
Here, we review some techniques that are useful for analyzing nearly ideally
behaving distillation columnsi.e., for predicting how these columns might perform, given certain specifications. The first topic will be minimum reflux calculations so we can determine the required internal flowrates in a column. However, to really understand this topic, we should first examine the concept of a
pinch point in a column. Using our understanding of a pinch point on a McCabeThiele diagram, we shall see that a pinch point occurs when the compositions passing each other between two trays (thus satisfying the operating line
equations) are also in equilibrium with each other (King, 1980).
76
ARTHUR W. WESTERBERG AND OLIVER WAHNSCHAFFT
1. Pinch
We follow the development in Terranova and Westerberg (1989) to explain
a pinch point. The material balance and equilibrium relationships for the top
section of a column are as shown in Fig. 7.
The species material balance is
V?, = LY,+
DxD,,
We can express equilibrium as follows:
where arkis the re1cirii.e v o l a t i l i ~of species i relative to an arbitrarily selected
key species ke.g.. the heaviest species in the mixture or the most plentiful (or
the lightest, etc.). Note that
is a ?~iole7fractionL1iieraged
re1atit.e volatility. It will lie somewhere between
the largest and the smallest relative volatility for the mixture.
Using Raoult's law
we can estimate relative volatility as
The ratio of Kvalues is an exact definition for relative volatility. The ratio
of vapor pressures is an approximate one that assumes Raoult's law holds,
thereby assuming that the Kvalues are not compositiondependent. The advan
FIG. 7 . Top section of a column showing flows and compositions.
SYNTHESIS OF DISTILLATIONBASED SEPARATION SYSTEMS
77
tage to relative volatilities are that they are much less temperature and pressuresensitive than are Kvalues.
A pinch point occurs when the vapor and liquid compositions for the
operating relationships are also in equilibrium with each other. For such a situation, the operating point of streams flowing against each other between stages
is also on the equilibrium surface for the problem. To step through a pinch point
requires an infinite number of stages. Substituting equilibrium into the species
material balance equation for species i, we get
which, when solved for xi, gives
We multiply both sides by the relative volatility (Yik and sum over all species to
€9
If we are given
all the relative volatilities,
the mole fractions and total flow for the distillate product, and
the vapor and liquid flows in the column at the pinch point (i.e., the reflux
ratio defined at the pinch point),
this equation is a single equation in the one unknown elk. It has to be solved
numerically, using something like Newton’s method. Once we have &, we can
use the previous equation to solve for each of the mole fractions xi at the pinch
point.
If the relative volatilities are composition, temperature, and pressuredependent, we can use this composition as input to a bubble point computation.
When solved, we will have new estimates for the relative volatilities. Then we
can iterate the computation until all the numbers are consistent.
We can also compute the condenser duty with a heat balance around this part
of the column:
78
ARTHUR W. WESTERBERG AND OLIVER WAHNSCHAFFT
where H,., h,, and h,) are molar enthalpies for the given mixtures and Q, is the
condenser heat duty. Note that there is nothing approximate about this computation. We can compute the molar enthalpies given the temperature, pressure,
and stream composition using a rigorous physical property package and obtain
as accurate a number as the property computations allow.
2. Uridencood's Method
It is possible to deriveroughlythe
equations underlying the Underwood
method (Underwood, 1946) from the above. The variable R represents the reflux
ratio defined in terms of the liquid flow Lif the pinch point relative to the distillate
top product flow; i.e.,
R = UD.
We again write the equation giting rl at the pinch point, and then do some
rearrangements and variable transformations.
XI
=
a__
ih
ffL

x D ~

QIID

L
alkv/D  L

ffk

xD I
ffl~(Rmln
k 1 )

ah
Summing over all species, we get
Adding 1
=
X IxD., to both sides gives
Finally. multiplying both sides by D gives the form
 Rmin
SYNTHESIS OF DISTILLATIONBASED SEPARATION SYSTEMS
79
which is one of the familiar equations from the Underwood method. A similar
relationship exists for the bottom of the column (note the minus sign):
We can now write a relationship between the vapor flows in the top and
bottom of the column:
v=B+
(1  q)F
Then assuming the roots for the top and the bottom are both the same and
using this equation, we derive the remaining Underwood equation:
The assumption that the 9 values are the same is quite an assumption. It takes
a good deal of arguing to make that plausible. We shall not go into that here,
however.
Use of Underwood’s Method. Table I11 presents an example that illustrates a
computation we might do to compute the minimum reflux flows for a column.
In this example, species C distributes between the top and bottom product in
the column. Underwood’s method permits us to compute how it distributes. The
approach for using Underwood’s equations to compute minimum reflux is as
follows:
Write Eq. (3) for each of the roots C#J lying between the species that appear
in both the top and bottom productshere B, C, and D. So we write it
twice. There will be a root between CYBE and aCEand between aCEand
a D E . Call them dBCand &, (for the relative volatilities, E is arbitrarily
selected to be the key species).
PROBLEM FOR
Species
A
B (W
C
D (hk)
E
TABLE 111
ILLUSTRATING
UNDERWOOD’S
METHOD
Relative
volatility
Feed
(kmoUs)
Top product
(kmolls)
4
3
1
1
1
1
1
1
0.98
2
1.5
1
?
0.03
0
80
ARTHUR W. WESTERBERG AND OLIVER WAHNSCHAFFT
4.1
4 
+
+Bc
3 . 1
3 
+
~ B C
2 . 1
+
2  4ec
1.5 . 1
1.5 
1 . 1
+
1 
+BC
~ B C
= ( 1  1)4 = 0
4.1
4
 CbCD
3 . I
+
3
C
D
'#'
+
2 . 1
2 
4CD
+
1.5
.1

1 . 1
+

4CD
+CD
= ( 1  1)4 = 0
which gives +ec = 1.673825 and '#'cD = 2.395209.
Write Eq. ( I ) twice, one for each root, and solve these two equations for
the two unknowns dc ( = X ~ , ~and
D )R,,,,. (Solved for these two variables,
the equations will be linear.) Once dc is known, D can be computed and
then the mole fractions for the top product:
4

4 . 1
1.673825
. 0.98
2 4
+ 3 3 1.673825
+ 2  1.673825
+
which gives V,,,,
=
5.664 kinolls and dc
=
1.5 . 0.03
 1.673825
1 . 0
 Vmm
I  1.673825
+
1.5
0.324 kmol/s.
B. SYSTEMSYNTHESIS
FOR NEARLY
IDEAL SYSTEMS
We shall present our ideas in this section by example. Consider the following
separations problem, for which the species should all behave relatively ideally
when in B mixture.
1. Example: Separation of Five Alcohols
Tmble IV gi\,es the species jlows f . r a fivespecies alcohol mixture. Design a
,system oj"distillation coliiriins to separrrte them.
TABLE IV
FIVEALCOHOI
SEPARATION
Species
i
twbutanol
I pentanol
I hexanol
1 heptanol
I octanol
,fCi)
(kmolh)
5
10
70
SO
15
SYNTHESIS OF DISTILLATIONBASED SEPARATION SYSTEMS
81
We first need to assess how many different sequences we might actually
invent for this problem. Using only simple columns, we can construct the alternative sequences shown in Fig. 8, where each column does a fairly sharp
split between adjacent key species. As we can see, there are 14 sequences. The
third sequence has two binary separations at the end for it in this “tree” of
alternatives. Note that both are required, so only one sequence results in the
counting.
Thompson and King (1972) developed a formula for predicting the number
of simple sequences for such a problem:
8!
 [2(5  l)]! No. Seq. = [2(N  l)]! = 14
[N  1]!N!
[5  1]!5!
4!5!
where N is the number of species in the original mixture. We note that for a
problem where the vaporAiquid equilibrium behavior of the species is relatively
ideal, discovery of the alternative simple sequences is straightforward.The number of sequences grows to over 290,000 for a tenspecies mixture. Separating a
tenspecies mixture into ten relatively pure single species products is a rather
large separation problem. However, if the analysis is simple, this number of
alternative sequences is not too large a problem to be investigated using a computer. We would want to do it efficiently, none the less.
2. Separation Selection Using Marginal Costs
In this section we present a very simple method based on marginal cost (Modi
and Westerberg, 1992) to compare the different sequences. As it is a very apWCDE
<
CDE
 DIE
C D
2
cm
 wc
4
0 0
DIE
CD
AWCDE
1
5
6
7
m
A O C I D € ~  m ~ABlC
~
NBCD
<
8
9
DIE
WCD
0ciD
C D
BIG
10
11
BIG
13
14
AA
FIG. 8. All 14 “simple” separation sequences for separating a mixture of five species.
82
ARTHUR W. WESTERBERG A N D OLIVER WAHNSCHAFFT
proximate method, it should be used only to remove sequences from consideration. leaving one with a (hopefully) much reduced set to investigate with more
accurate methods.
We start by noting two heuristics presented a number of years ago by Hendry
et a/. ( 1 973) that we should consider before proceeding very far into the selection of the sequences.
Dangerous Species Heuristic: Remove dangerous arid corrosive
species early.
We would wish to handle dangerous (deadly if released, toxic, carcinogenic,
explosive) and corrosive species in as few pieces of equipment as possible. (Best
of all, avoid handling them completely by not designing a process with such
species present.) They should be removed first. If such a species is either the
most or the least volatile. then we can remove it in the first column. Otherwise,
it will take at least two columns (a split just above and a split just belowin
either order) to remove it.
Final Product Heuristic: Produce final products as distillate products,
iiot as b o t t o m products.
All mixtures will contain contaminants, such as heavy organics, which tend
to discolor the yield if they are removed from the process with final products.
That is. heavy species will exit a column with the bottoms product, contaniinatinp it. Customers will pay more for a cleaner looking product (even when it
really may not matter). However, this heuristic should not be followed blindly.
Many times it is not costeffective to follow this heuristic; at other times it may
be impossible to abide by itfor example, when the heaviest species is the
product to be sold. Moreover, it is possible to clean up a product in other ways:
e.g.. by vaporizing and then condensing the material, leaving the contaminant
as a heavy residual. or by passing the product through a bed of activated carbon.
Assuming we have considered these two heuristics, we need to select among
the tnany alternative sequences that may remain for a problem. One criterion is
to select the one with the lowest annual cost in units like $/year (&/year,
DM/year. etc.). Annual cost is the total of the costs per year to operate the plant
and the annualized cost of the investment required to build the plant. Investments
are measured in dollars. Annualizing investments means that we convert an
investment cost in dollars ($) into an equivalent expense in dollars per year
($/yr).
A very simple method for annualization is to divide investment cost by the
nutnber of years over which the company wants that investment paid back by
the earnings of the process; e.g., we divide by 3 years if we want the investment
hack in three years. Many other ways exist, such as establishing a set of equal
SYNTHESIS OF DISTILLATIONBASED SEPARATION SYSTEMS
83
yearly payments over a specified lifetime of the process, say 15 years, that would
have the same present value as the investment. Present value requires us to set
an acceptable interest rate over the inflation rate, say 1015% per year.
We propose finding the sequence with lowest costs by using marginal costs.
We compute marginal costs by devising a base cost that all sequences have as
a minimum, then estimating only the added costs that distinguish one sequence
from another. One appealing base cost is the cost to carry out all the needed
separationssuch as A/B, B/C, CAI, and DW for our earlier exampleas if
they were done with no other species present. Then the incremental cost of a
task is the cost to carry out those separations with the other species present, a
cost that differs from sequence to sequence. For nonideal systems, the cost may
actually go down when other species are present. However, for ideal systems,
the costs increase.
Returning to our previous fivespecies example, the sequence { A/BCDE,
B/CDE, C/DE, D/E} differs from the sequence (AB/CDE, A/B, CDE, CAI} as
shown in Table V. The entries in the second and third columns indicate the
extra species present for each binary separation when accomplished using these
two different sequences. The binary separation B/C is done with species D and
E present in the first sequence, while it is done with species A, D, and E for
the second.
The particular costrelated quantity we shall consider here is the marginal
vaporflow rather than the actual annual cost. It is a quantity we can more readily
estimate. In fact, for nearly ideal systems, we shall show a very easy way to
approximate it.
We define the symbol
AV(i4,list) = Marginal vapor flow for column having i a n d j as light
and heavy key species, respectively. The list contains species other than
i and j which are present in the feed.
For example, AV(B/C,ADE) is for the task AB/CDE: This column is splitting
B from C, and the other species present are A, D, and E.
TABLE V
EXTRASPECIESAS A FUNCTION
OF THE
COLUMNSEQUENCE
USED
Binary
Separation
Sequence
1
Sequence
2
AIB
BIC
CID
DE
CDE
DE
E
none
none
ADE
none
C
84
ARTHUR W. WESTERBERG AND OLIVER WAHNSCHAFFT
Marginal vapor flow is the added vapor flow required in the column because
the other speciesi.e., those on the listare present. The vapor flow in a column is an indicator of the cost of purchasing and operating the column. A
difficult separation will have a large vapor flow because it will require a large
reflux ratio. Also, refluxed material has to be vaporized and condensed, which
directly affects the utility costs for operating the column. Therefore, it makes
sense to try to minimize the total of the vapor flows for a system of columns.
Formally, we can define marginal vapor flow as
where the last term is the vapor flow in a column to split i f r o m j with no other
species present (the list is empty j.
We should note that all sequences to separate a mixture will have the same
set of binary splits. For example, each alternative sequence for the separation
of ABCDE into five singlespecies pure products will have a split between A
and B, another between B and C, etc. The difference among the alternative
sequences is the presence or absence of other species when carrying out each
of these binary splits. The total of the vapor flows for a sequence is the base
set of vapor flows V(i/j). where i and j are A/B,B/C, C/D, and D/E, plus its
marginal vapor flows. Thus, the difference in marginal vapor flows is the difference in total vapor flows among the sequences. The sequence with the
minimum marginal vapor flows is the sequence with the minimum total vapor
flows,
How can we estimate a marginal vapor flow for a column? One approach is
to estimate the minimum reflux required using any method that is appropriate.
If the separation is among species that are acting nearly ideally, we can use
Underwood’s method.
3. FiivAlcohol Example Continued
For our fivealcohol example, let species A be nbutanol, B be 1pentanol,
etc. Now. let US consider the column that accomplishes the separation AB/CDE.
To estimate the internal vapor ffows in the column, we will have to assume
recoveries for the species. Here. we shall assume that 99% of the key species
go to their respective products, while everything lighter than the light key goes
to the distillate and everything heavier than the heavy key ends up in the
bottoms.
Applying Underwood’s method gives us a minimum vapor flow of 72.5
kmolh for a column accomplishing the separation of ABICDE. Without species
A. D, and E present, the minimum vapor flow is computed to be 44.5 kmol/h.
The marginal vapor rate is therefore 38.0 kmol/h.
SYNTHESIS OF DISTILLATIONBASED SEPARATION SYSTEMS
85
Let us look more closely at the Underwood equations to see if we can quickly
compute an approximate answer. The minimum vapor flow for the split
ABKDE is given by
v .=
+
QAC

accdc
~ B C ~ B+
~ A C ~ A
min
6BC
UBC

6BC
aCC

6BC
but it is also related to the bottoms vapor flow as follows:
Vmin = Vmin
 
+ (1
 q)F
accbc
~ B C ~ B
aBC


6BC
aCC

aDcbD
aDC

6BC

6BC
~ E C ~ E
ffEC
+ (1

 4)F
6BC
Let us assume the root +BC does not move all that much if we add other species.
Then these two equations would suggest that V,,, is increased by an amount
~ A C ~ A
aAC

6BC
because species A is present and by an amount

aDCbD

%C

6BC
~ E C ~ E
aEC

6BC
because species D and E are present. The extra species are those lighter than
the light key (species A) and heavier than the heavy key (species D and E).
They will be essentially fully recovered in their respective products, so we
should be able to substitute the feed flows for the product flows for each of
them. Also in the terms for the species D and E, the denominator of the expressions is negative, so the terms are positive.
We can thus estimate the added vapor flow for each as the absolute value
for the appropriate term, i.e.,
I
aik

6lk,hk
I
where species i is in the feed when separating species Ik from hk.
Figure 9 is a sketch of the relative sizes of the numerator and denominator
for terms such as this for our example problem.
The amount of added vapor flow because A is present is proportional to aAC
divided by aAC c $ ~We
~ . do not need a particularly accurate value for cbBc
86
ARTHUR W. WESTERBERG A N D OLIVER WAHNSCHAFFT
a
AV(BIC,D) z
DC
'0c%c
O d
D
FIG. 9. Relative sire of numerators and denominators in term to estimate marginal vapor flow:
( a ) terms relative to each other: (b) ratios.
to get a reasonable value for this term. A similar argument holds for the term
when D is present and. although not shown. when E is present, too. We could,
for example, let @Bc be the average of the two surrounding relative volatilities.
Our final term to approximate the increase in internal vapor flow for a column
is thus
We see that the term in front of the flow for the species in its respective product
is much larger for a light species than for a heavy one. Thus, Figure 9 alerts us
to the heuristic that the vapor flow is more sensitive to the presence of extra
light species than the presence of extra heavy ones. This heuristic is often stated
as follows:
SYNTHESIS OF DISTILLATIONBASED SEPARATION SYSTEMS
87
Direct Sequence Heuristic: Prefer removing the most volatile species j h t .
Repeated application of this heuristic gives what is called the direct sequence.
For example, the direct sequence for our problem is ABCDE, B/CDE, CDE,
and D/E. (The indirect sequencethe
other one with a special nameis
ABCDE, ABCD, AB/C, Am.)
4. FiveAlcohol Example Continued Further
The calculations were done using vapor pressure data available in Reid et al.
(1987) for each of the species; Table VI gives the results. The temperature
selected for evaluating the vapor pressure is the bubble point at 1 atm for the
feed mixture, i.e., at 434.21 K. Using the relative volatilities and feed flows
shown in Table VI, we can estimate the marginal vapor rates shown in Table
VII using the equation
For example, with C present for the Am split, we get
1.0000 * 20 kmol/h
1.0000 
(3.3199
= 12.9 kmolh
1.7735
To read Table VII, the marginal rate for AB/CDE, AV(B/C, ADE), is computed by adding the terms for the B/C split (second row) with A, D, and E
TABLE VI
AND RELATIVEVOLATILIES
FOR
VAFQR PRESSURE
EXAMPLEPROBLEM
Species
Vapor
pressure
(bars)
Relative
volatility
(kmolih)
Isobutanol
Ipentanol
1hexanol
1heptanol
1octanol
3.8279
2.0449
1.1530
0.6407
0.3542
3.3199
1.7735
1.oooo
0.5557
0.3072
5
10
20
50
15
f(0
88
ARTHL!R W. WESTERBERG AND OLIVER WAHNSCHAFFI
B
A
A/B
B/C
ClD
DIE
****
*.,. ,
8.6
6.5
5.7
q:***
a%*
17.8
13.2
C
D
E
17.9
14.0
33.4
2.1
4.3
9.8
****
****
****
d*X*
35.1
****.
"Key species are listed along the left side. extra species across
top.
added: namely, 8.6 + 33.4 + 4.3 = 46.3 kmol/h. This number should be compared to the more exact number. 38.0 kmolh computed earlier. We should not
expect it to be any more accurate than these numbers indicate.
We note there are two rather large numbers: one for which D is present with
the B/C split and the other for which C is present with the D/E split. No doubt
these should be avoidedas they can be if we split A B C D E at the start. Then
the ''co5t" to have A present for the AB/C split (8.6) compared to that of having
C present in the A/BC split ( 12.9) suggests we should probably split C off next.
That fixes the 5equence as
ABCDE. AB/C, A/B. D/E
The only other sequence that avoids these two high costs is A/BCDE followed
by BC/DE.
We can search over all possible sequences quite quickly as follows. We sum
up the marginal costs for each split possible in the problem and organize them
in the following way.
AIBCDE
ABCIDE
ABCIYE
28.9
BICDE
BClDE
BCDIE
37.7
27.6
48.4
C/DE
CDE
9.8
35.2
BICD
BCID
33.4
17.8
AIBC
ABIC
12.9
86
34. I
54 1
AIBCD
XBICD
ABC/D
26.9
42.0
19.0
To develop a solution using the least cost, we start with the total feed and
compare the costs for the first splits. Here, the costs range from a low of 28.9
to a high of 54.1 kmolh. so we select A/BCDE first. We now must complete
the sequence: i.e., separate BCDE. The leastcost next step is 27.6 corresponding
to BC/DE. which gives us a total of 56.5 kmolh. That completes this sequence
SYNTHESIS OF DISTILLATIONBASED SEPARATION SYSTEMS
89
because the marginal costs for the two remaining splits, B/C and D/E, are zero
(there are no extra species present for them). Note that this is the second of the
two solutions we spotted above. Can we find a cheaper solution? Observe that
neither of the other first splits for BCDE is preferable: each has a highercost
first step (and each requires another split, with an added cost, to finish them).
We try ABCDE next at a firststep cost of 34.1. The AB/C split adds 8.6
for a cost of 42.7. This sequence is the first one we found intuitively above,
and it is better. No sequence starting with AB/CDE (firststep cost of 46.3) or
ABCDE (54.1) can possibly be better. We are done.
The approach we just used to completely search this space is a form of branch
and bound. We branched off the least expensive first steps and slowly eliminated
the need to look at others as they each had a firststep cost (bound) that was
too high to win.
Because errors in these numbers can occur, we might want to look at anything
within 20% of the best with a more accurate analysis. If none exists, we are
done with a relatively cheap analysis. If some exist, we could use Underwood’s
method (i.e., find the roots) to estimate the minimum reflux ratios as we did
above for one of the columns.
5. Distillation Heuristics
The marginal cost approach explains (at least partially) many of the commonly published heuristics used to select the better sequences. We have already
seen how it explains the direct sequence heuristic, which is stated as follows:
Direct Sequence Heuristic: All other aspects of the problem being equal,
remove the most volatile species Jirst.
For all other things to be equal, the amounts of the species must be the same
and the relative volatility between all adjacent pairs of species must also be the
same. For example, in the case of a fourspecies feed having equal amounts of
each species (e.g., 1 kmol/h each) and relative volatilities of aAD= l.23 =
1.728, aBD
= 1.22 = 1.44, CXCD = 1.2’ = 1.2, and CYDD = 1.2’ = 1 , “all other
aspects of the problem [would be] equal.” For each binary split (e.g., A/B or
B/C), the McCabeThiele diagram would be drawn with the equilibrium curve
based on a relative volatility of 1.2.
In carrying out the above search, we become aware that different sequences
are associated with different numbers of extra species contributions. For example, in the sequence
(ABCDE, B/CDE, CDE, and D/E}
CDE, DE, and E show up as six instances of extra species, while in the sequence
{ABC/DE, A/BC, B/C, D/E}
90
ARTHUR W . WESTERBERG A N D OLIVER WAHNSCHAFFT
only four instances, ABE and C. show up. The more we split the problem into
halves. the fewer the number of extra species contributions we will add into the
objective for the above search. We can use this observation to justify partially
the following commonly stated heuristic:
50/50 Split Heuristic: Separute the mixture into roughly equal
amounts of products.
Using marginal vapor flows, we can also explain the following heuristic fairly
straightforwardly. as the approximation for the added vapor flow for a species
in any mixture in which it appears is proportional to its flow in the feed:
Major Species Heuristic: Have the rnujor species in as ,few splits as possible.
One other heuristic very commonly stated is
Save the difJicidt splits jor lust.
In other words. do the easy splitsjrst. A split is easy if the relative volatility
between the two key species is large. The (plausible) argument is that the hard
splits should be done when no other species are present. Since the marginal
vapor flow computation neither supports nor rejects this heuristic, we might
draw the conclusion that this heuristic is not valid for the problem as we have
posed it above. But there is some justification for this heuristic when we consider
the energy integration of columns (e.g., using the heat expelled from the condenser of one column as the heat input into the reboiler of another).
All of these heuristics depend on amounts and relative volatilities for the
species in the mixture. Based on precisely these quantities, the simple
computation
would seem to quantify their relative importance.
V. Separating Highly Nonideal Mixtures
We shall now look at the synthesis of separation systems for liquid mixtures
of \pecks that display highly nonideal behavior. We reference a small sampling
SYNTHESIS OF DISTILLATIONBASED SEPARATION SYSTEMS
91
of the work on this topic: Ewell and Welch (1949, the classical text by Hoffman
(1964), early work on classifying behavior by Berg (1969), a scheme for classifying vaporliquid behavior for ternary mixtures by Matsuyama (1 973, many
papers on homogeneous continuous azeotropic distillation by Doherty and coworkers (Doherty and Perkins, (1978); Doherty, (1985); Doherty and Caldarola,
(1985); Levy et al., (1985); Van Dongen and Doherty, (1985); Knight and Doherty, (1989); Julka and Doherty, (1990); Foucher et al., (1991); Fidkowski et
al., (1993); work by Petlyuk (1978), Stichlmair et al., (1989), an expert system
described by Barnicki and Fair (1990), work on multiple steady states by Bekiaris et al., (1993), work by Bossen et al., (1993), and our own work. Poellmann and Blass (1994) presented a review and new analysis methods for azeotropic mixtures.
Distillation remains a likely option for separating a complicated liquid mixture except for the fact that we cannot readily predict what the products will be
if we do distill a given mixture. Extractive distillation is often a viable alternative. If the species display liquid/liquid behavior (e.g., hydrocarbons and water), we can consider simple decantation or liquid/liquid extraction, too.
In general, we will try simple distillation; but, when we do, we often discover
that significant amounts of almost all of the species show up in either or both
of the distillate and bottoms products, no matter how we run the column. The
inability to effect sharp splits gives rise to the recycling of streams within the
separation process itselfsomething we did not require earlier when we looked
at the separation of ideally behaving mixtures.
A. AZEOTROPIC
SEPARATION:
EXAMPLE1
Consider the separation of water from nbutanol. The phase behavior for this
mixture is quite complex. There is a minimumboiling azeotrope formed as well
as a liquidliquid phase separation when the liquid is cooled enough. Figure 10
is a sketch of the general shape of the phase behavior for this system. (It is not
an accurately drawn phase diagram.)
Several textbooks and reference books (e.g., the third edition of Perry’s
Handbook, 1950) use this example to illustrate azeotropic distillation. They
show the solution sketched in Fig. 11 for a feed whose composition is about
60% nbutanol. If we analyze this configuration, we see that it separates the
mixture. The same textbooks suggest that for a feed below the azeotropic composition, one should use the same configuration but put the feed into the decanter. The question that occurs immediately is: How was this configuration
selected? Was it a trialanderror procedure, or is there some way to find it
directly?
92
ARTHUR W. WESTERBERG A N D OLIVER WAHNSCHAFFT
t
12*
u
ep
0
c
1oo
2 phase
liquid
*
80
0
composition (mole fra tion
nbutanol)
1.o
FIG. 10. Vapor/liquid equilibrium diagram for tibutanollwater.
decanter
feed
water
FIG. 1 1 . Configuration presented in handbooks for separating nbutanol and water.
93
SYNTHESIS OF DISTILLATIONBASED SEPARATION SYSTEMS
As will be true in virtually all synthesis problems, the secret to finding a
solution is to find and use the right representation. For this problem, a convenient
representation is shown in Fig. 12. We show a line that has the key compositions
marked along it for the phase diagram for the system.
Starting with a feed at about 60% nbutanol, we see that we can distill it into
two products: one close to the azeotropic composition as the distillate product
and one, nearly pure nbutanol, as the bottoms product. The bottoms product is
a desired product; however, we must still separate the top. We find that the top
azeotrope lies in the region where it partitions into two liquid phases if we cool
it. We therefore cool the azeotrope and put it into a decanter. Next, we decide
to distill the waterrich phase (bottom phase in the decanter) into nearly pure
water and azeotrope, again getting the azeotrope as a distillate product. We
already know what to do with the azeotrope: cool it and feed it to the decanter.
The only remaining stream is the upper decanter product, the one that is richer
in nbutanol.
We could distill this upper product, getting nbutanol and azeotrope. However, the first column is already able to do this task, so we can choose to feed
it into that column. We see that, with minor modifications, we have just invented the structure shown in the handbooks. In a similar manner, we should
be able to invent two structures that can handle a feed to the waterrich side
of the azeotrope. In one, the feed enters the decanter and in the other, the
feed enters the second column (an alternative not mentioned in the handbooks).
LL
boundary
a=4)wope
LL
boundary
pure
pure
nbutanol
water

composition
+
column 1
mixer, decanter
feed
pure
nbutanol
column 2
Pure
water
FIG. 12. Convenient diagram to synthesize a separation process for the nbutanollwater system.
94
ARTHUR W. WESTERBERG AND OLIVER WAHNSCHAFFT
VI. Synthesis Discussion
‘4. ANALYSISDRIVER
SYNTHESIS
When the species of interest display highly nonideal behavior, the synthesis
of separation processes falls into the class of synthesis problems we might call
ana1y~i.scliiverrsyithesis. We use that term because the real work in this synthesis problem is to establish enough information (i.e., carry out complex presyitliesis cirzcilysis computations) to begin the synthesis process. We have to find
out about the behavior of the mixtures of the species (e.g., are there azeotropes)
before we can even propose the types of equipment we should use to carry out
a separation. Then, we must also worry about how these species will behave in
any equipment we suggest. Finally. we must carry out a postsynthesis analysis
in which we design the equipment to evaluate (e.g., determine costs for) the
different alternatives. At a minimum, then, column design requires us to set the
column pressure and compute the reflux ratio, the number of stages, and the
column diameter. For nonideal behavior, these computations can be very
difficult.
We thus see a presynthesis analysis that characterizes the behavior of the
species and a postsynthesis analysis that requires us to design equipment. Both
require that we compute equilibrium phase behavior (vaporAiquid, liquidhquid,
vapor/liquid/liquid. etc.)
B. IMPACT OF NUMBER OF
SYNTHESIS
METHODS
SPECIES ON REPRESENTATION, ANALYSIS,
AND
The synthesis of distillationbased separation systems is strongly supported
for two and threespecies mixtures. There is much less support for mixtures of
four species. and even less for tive. There are several reasons for this, some of
which are fundamental.
The presynthesis analysis allows us to understand the topology of the phase
behavior of the species when mixed. With that understanding, we find that we
can conjecture reasonable process alternatives. LJnfortunately, we are limited to
a threedimensional world when it comes to human visualization. Thus we can
appeal to visualization of complex topologies only when we can show them in
two and, with difficulty. three dimensions. The triangular composition diagram
is an excellent way to visualize complex behavior among species, but it is specific to threespecies mixtures. When we go to four species, we stretch the ability
of people to see what is going on; when we go to five, we are approaching
impossibility.
SYNTHESIS OF DISTILLATIONBASED SEPARATION SYSTEMS
95
Certainly, computers can work well at higher dimensions. Unfortunately,
however, when we go to higher dimensions, what were lines often become
planes, then volumes, and so forth. Whereas we can search for exotic behavior
along a line with a “computational sledge hammer,” it currently takes enormous
effort to search a plane thoroughly for complex behavior. For volumes and
higher dimensional spaces, then, we must use simpler and less complete searching at this time. This is the problem we face when analyzing systems with five
or more species.
Another issue is the way we can use the degrees of freedom for a column.
We argue that things change qualitatively when going from systems of three to
four to five and higher species. In Fig. 13, we show what we might call the
“natural” degrees of freedom for a columnthe ones we generally pick when
we wish to compute the performance of a column. These specifications lead to
the easiest column calculation to converge. Using our intuition about columns,
we argue that, if we specify
the
the
the
the
the
the
feed to a column (flowrate, composition, pressure, and temperature),
column operating pressure (P),
number of trays in the top (ntop),
number of trays in the bottom (nbot),
reflux ratio (R), and
distillate product flow rate (D),
then the column will operate as expected. That is, a typical simulation of the
column would tell us the top and bottom product composition, as well as the
traybytray temperatures, compositions, etc. Assuming our intuition is correct,
we note that, once we specify completely the feed to a column, a column provides us with five more degrees of freedom. We shall now discuss how we
might use these to analyze a column in other ways, for example, in designing
the column.
.
feed
FIG. 13. “Natural” degrees of freedom for a column.
96
ARTHUR W. WESTERBERG AND OLIVER WAHNSCHAFFT
Of the five degrees of freedom, a designer typically fixes the column pressure
or the reboiler. Thus, we shall assume
that P is not at our disposal. That leaves us with four degrees of freedom. Can
we use these four degrees of freedom to fix the recoveries of four species leaving
the column in the top distillate product'? For example, can we design the column
so that 99% of species A, 9S% of species B , 1% of species C, and 0.5% of
species D will leave in the distillate?
Let us carry out the following computational experiment involving the solving of a column model.
First we shall solve a column using the natural specificationsthose listed
above. Generally. such a model is easy to solve unless the species are highly
nonideal in their behavior. We observe the recoveries for the two key species,
tinding that 85.1'7r of the light key leaves in the distillate while 74.3% of the
heavy key leaves in the bottom product. We now change what we intend to
compute and what we intend to specify as fixed for the model calculation. First,
we decide t o fix the recovery of the light key and ask the model to compute
the required reflux ratio. We then alter the light key recovery to 85.0% (very
slightly away from the value obtained above) and resolve the model, keeping
all the other fixed variables (pressure. reflux ratio, number of trays top and
bottom) at their current values. If the simulator only permits the natural
specification to be made. then we can only change the reflux ratio iteratively
until the recovery changes to the desired value. The reflux ratio should decrease slightly. allowing more of the heavy species to exit in the top product. Since the distillate flowrate is fixed, this will reduce the recovery of the
light key.
Being more daring. we now change the recovery to 80% and then to 90%.
In both cases there is no difficulty in converging to an answer. We discover
that this trade is a practical one. We revert back to a recovery of 85.0%. Experience with many simulations suggests that, while i t is a practical one, it is
very difficult to converge with such a specification given at the start of the
calculation.
Next we fix the recovery of the heavy key in the bottom product at 74S%
and ask the model to adjust the distillate top product flowrate accordingly. For
this calculation we still want the light key recovery to stay at 85.0%. so the
biinulator adjusts the values for both the reflux ratio and the distillate top product
flowrate, giving us our desired slightly altered key species recoveries. It has no
difficult4 in converging. We find that the larger changes also converge readily.
So we can trade light and heavy key recovery specifications for distillate total
Aou and reflux ratio. provided we d o it with a modicum of care.
We now ask if we can specify the recovery of a third species. The variable
we trade is the number of trays in the top section of the column. The number
10 set the temperature of the condenser
SYNTHESIS OF DISTILLATIONBASED SEPARATION SYSTEMS
97
of trays is a discrete quantity so, for a traybytray model, we should not be
able to fix the recovery exactly for the third species while holding the recoveries
of the keys fixed at 85.0% and 74.5%. However, a collocation model (Cho and
Joseph, 1983; Stewart et al., 1984; Seferlis and Hrymak, 1994; Huss and Westerberg, 1994) allows us to model a column approximately such that the number
of trays is a continuous variable. We assume the availability of such a model.
With this model, we find that the changes we can accomplish in the recovery
of the third species have to be very small. For example, we might ask that the
amount of a species heavier than the heavy key have its recovery in the distillate
top product decrease from about 1% to about 0.5% by adjusting the number of
trays in the top section. Such a change will cause changes in the distillate flowrate, the reflux rate, and the number of trays in the top section. If we try to
increase the recovery to 3%, the model fails. We return it to a 1% recovery.
Further testing convinces us the trade is theoretically possible, but the range for
changing the specification is quite small.
Finally, we ask if we can trade the number of trays in the bottom section for
the recovery of a fourth species. Our intuition tells us we could be in numerical
trouble here. In principle, we can make the trade; however, very small changes
lead to very large changes in the number of stages and even more often, to
computational failure. The fourth specification is theoretically possible but computationally nearly impossible.
,
What are the implications of this experiment? First, from a theoretical point
of view, we can specify the recoveries of four species at most for a column.
From a practical point of view, the first two are relatively easy (if we take care
in doing it) to specify, adding a third specification is difficult, and a fourth
virtually impossible. We cannot, as a result, specify completely the top product
of a column for more than four species in theory and more than three in practice.
c. IN SUMMARY
Any method that we develop for three species starts to run into the practical
limitations we described above when we try to extend them to four species. We
have difficulties in presenting them to humans for visualization; moreover, difficult searches explode in size, no longer running along lines, but extending over
planes, volumes, and higher dimensional spaces. Finally, we cannot specify
product compositions for more than four species separations in theory, and we
have real, practical difficulty with more than three.
For these good reasons, the literature explores methods to handle three species mixtures thoroughly and hesitates to extend them to mixtures with four and
more species.
98
ARTHUR W. WESTERBERG A N D OLIVER WAHNSCHAFFl
VII. Preanalysis Methods
There are two classes of analysis methods that we shall explore. The first
class allows us to understand the phase behavior of the species in the problem.
The second allows us to analyze the behavior or design of a piece of separation
equipment.
We need to understand the behavior of the species in our problem. As we
noted earlier, it is not very useful to design a separation process assuming ideal
behavior when such behavior does not exist. Such a design might not represent
even a lower bound on the cost of the process we might actually develop. For
example. toluene and water dislike each other so much that they readily separate
into two liquid phases, each with little of the other species present. That is, they
behave nonideally. If the compositions for these two phases meet our specifications, we can separate these two species by a relatively inexpensive decantation procedure, which is much more effective than designing a column predicated on ideal behavior.
Some of the analysis methods we are about to discuss rely on computing
residue curves, so we first need to understand what such curves are and how to
compute them.
A. EQUILIBRIUMPHASE
BEHAVIOR
Earlier in this paper we discussed a simple method to detect if an azeotrope
will exist between two species, A and B. The method requires us to perform
two bubblepoint computations, one in which we have a trace of A in B and
one in which we have a trace of B in A. We used the infinitedilution Kvalues
we computed in these bubblepoint computations for the trace species to reveal
where an azeotrope exists. If both infinitedilution Kvalues are less than unity,
there must be a maximumboiling azeotrope between these species. (There could
even be more than a single azeotrope between the species, such as two maximumboiling azeotropes separated by a minimumboiling azeotrope. We assume
this is rare, but there must be at least one.) If the Kvalues are both greater than
1. by similar arguments there is at least one minimumboiling azeotrope. Otherwise. we suggested, we could assume nonazeotropic behavior.
As we also discussed earlier, we can assess whether liquidliquid behavior
is likely by examining infinitedilution activity coefficients.
In this section we wish to look in more detail at the nonideal behavior of
mixtures of a given set of species. We shall start by examining residue and
distillation trajectory plots that show the vaporfliquid behavior of mixtures when
SYNTHESIS OF DISTILLATIONBASED SEPARATION SYSTEMS
99
they are being distilled. Such plots are very informative in that they depict the
VLE behavior in a manner that is useful for synthesizing separation processes
for such mixtures. We shall then consider methods by which we might discover
all the azeotropes for a given set of species. Finally we shall discuss methods
for determining the multipleliquid phase behavior for a set of species.
1. Residue and Distillation Curves
We look first at residue curves, which correspond to batch distillation, and
then to distillation curves, which correspond to separation trajectories in columns
operating under total reflux conditions.
a. Residue Curves. A residue curve (Hoffman, 1964; Doherty and Perkins,
1978) traces the composition of the liquid in a batch still in composition space
versus time. Along this curve the temperature always increases, the composition
of the heaviest species increases, and the composition of the most volatile species decreases.
Consider the batch still shown in Fig. 14, which contains a mixture that we
boil away with time. What is the composition of the liquid in the still versus
time? A dynamic material balance for the species i is
dxiM
  x . dM
+M=dt
dt
dt
y;v
We note that
dM
dt
=
v
allowing us to write
M
dx.
= V(x;  y;)
dt
Defining a dimensionless time as 6 = t / ( M N ) , this equation becomes the fol
FIG. 14. Batch still.
100
ARTHUR W. WESTERBERG AND OLIVER WAHNSCHAFFT
lowing simple relationships:
where setting J', = K,r, assumes that the vapor is in equilibrium with the liquid
composition.
These equations can now be integrated versus dimensionless time. For a
threespecies mixture. we can plot the resulting compositions as a parametric
function of f3 on a triangular diagram. Note. the x,'s sum to unity so one of the
equations is not independent. The last mole fraction can either be obtained by
integrating the above equations or by integrating all but the last of these equations and computing the last mole fraction so the sum of mole fractions is unity.
b. Distillation c'itnjeA. We can also compute a composition trajectory directly
for a column by stepping from one tray to the next for a column operating at
total reflux, as Fig. 15 illustrates. The material balance equations for this column
are
and
1
n
n' Ln
V
Yn+1 n+l
Flc;. 15. Total reflux column {no feed or product)
SYNTHESIS OF DISTILLATIONBASED SEPARATION SYSTEMS
101
therefore,
Yn+l,r
=
~ n . 1
Equilibrium gives
~ n + ~ .=
r Kn+l,rXn+l.r
Propagation from one point to the next therefore combines these last two equations to give
Xn.1
= Kn+l.r~n+l,r
If we step down the column, we would know x,,,~and would have to compute
a dew point to find the composition x , + , , ~Given
.
that, we compute its dew point
to step down to tray n + 2, etc. Stepping up the column requires that we
compute a sequence of bubble points in a similar manner. Temperature increases
as we step down a column, which is the same direction temperature moves when
time increases for the residue curve computations.
2. Sketching Residue Curve Plots f o r Ternary Systems
Figure 16 shows a whole set of residue curve trajectories that might be computed for a stillpot containing initially a mixture of water, ethanol, and glycol.
glycol
7 
water
7 
2I
min boiling
azeotrope
FIG. 16. Trajectories for compositions in stillpot.
ethanol
101
ARTHUR W. WESTERBERG AND OLIVER WAHNSCHAFFT
(Alternatively. we could sketch distillation curve trajectories, which look very
similar.) All of the residue curve trajectories move from the lowest temperature
in the diagramthe ethanol and water azeotropeand end at the highest temperature in the diagrampure glycol. At any point in a residue plot, the equilibrium vapor composition is moving in the direction that is tangent to the curve,
and in the opposite direction the trajectory is moving with time.
There are examples where the trajectories break the diagram into regions.
Such 3 structure appears when there is more than one local minimum and/or
maximum temperature in the diagram. We see such behavior in Fig. 17. The
lower two species again have a minimumboiling azeotrope between them; the
third (top) species is the most volatile one. To guess the topological behavior
for this diagram, we first place temperatures for the normal boiling point for the
pure species and for the azeotropes onto the diagram. We then place arrows
around the edges to indicate the directions for increasing temperatures, as shown.
From just these few arrows, we observe that there are two local maximum
temperatures in this diagram, one in the lower right and one in the lower left.
Let us assume that there can be at most one ternary azeotrope and then attempt
to sketch in trajectories within the diagram to expose its general structure.
Occasionally we can posit more than one structure possible, based on knowing only the temperatures for the pure species and the binary azeotropes. Since
such structures can occur, care must be taken to see that all possible structures
are discovered. Based on topological arguments, Zharov and Serafimov ( 1975;
see also Serafimov. 1987) developed an equation among the number of “nodes”
and “saddles“ appearing in a residue curve map. Independently, Doherty and
120
170
15s
FIG. 17. Directions of increasing temperature along edges of triangular composition diagram.
103
SYNTHESIS OF DISTILLATIONBASED SEPARATION SYSTEMS
Perkins (1979) developed a special version of this equation for threespecies
mixtures. Foucher et al. (1991) present detailed rules to sketch and to check the
consistency of data on ternary diagrams. One version of this equation for three
species mixtures is
4(N3  S,)
+ 2(N2  5'2) + ( N ,  5'1)
=
1
where Ni is the number of nodes and Si is the number of saddles involving
exactly i species on a ternary diagram.
A node is any point where all temperature trajectories enter or all leave,
whereas a saddle is a point where Some trajectories enter while others leave.
The pure species in the lower left and right corners have all trajectories entering;
the top pure species has all leaving. As each involves one species, each is a
singlespecies node. We cannot yet classify the binary azeotrope along the lower
edge. Its type will depend on whether trajectories enter it or leave it from the
interior of the composition triangle. If they enter, then some trajectories will
enter while those along the lower edge leave, making it a twospecies or binary
saddle. If they leave, all the trajectories leave, making it a twospecies node.
Points strictly inside the diagram are threespecies points.
A recent review article by Fien and Liu (1994) describes how to apply this
formula in some detail. Applying it for the ethanolwaterglycol example in
Fig. 16, we see that 4(0  0) + 2(1  0) (1  2) = 1 satisfies this equation.
If there can be only one ternary node or saddle, only N3 = S3 = 0 can satisfy
this equation and the sketch must be unique.
For Fig. 17, we might ask if the azeotrope at 155" is a node or a saddle, or
could it be either based on the information given. Substituting into the formula
for both options, we get
+
If a node:
4(N,  S,)
+ 2(1  0) + (3  0) = 1
or 4(N3  S,) = 4
If a saddle: 4(N,  S3)
+ 2(0  1) + (3  0) = 1
or 4(N3  Sj)
=
0
The former would allow a ternary saddle (i,e., N3 = 0, S3 = 1) to exist while
the latter permits only N3 = S3 = 0i.e., no ternary node nor saddleif our
assumption is valid that there is at most one ternary node and/or saddle.
Figure 18 shows the diagram with the required ternary saddle where the
binary azeotrope is a node. It has four distillation regions (labeled I to IV)
separated by distillation boundaries, each of which has its own set of maximum
and minimum temperatures within it. For example, region I has a maximum
temperature of 170" and a minimum of 120". It is this property of having its
own unique minimum temperature (from which all trajectories emanate) and
maximum temperature (at which all terminate) that characterizes a region.
I 04
ARTHUR W. WESTERBERG AND OLIVER WAHNSCHAFFT
120
170
155
F I G 18. Triangular coinpobition diagram where binary azeotrope is a node.
Amuning the binary azeotrope is a saddle (as shown in Fig. 191, we note
that there is a trajectory that starts at the upper vertex and passes through the
minimumboiling azeotrope on the lower edge. This trajectory is a distillation
boundary that splits the diagram into two distinct distillation regions, labeled 1
and 11. Each region has the same minimum temperature but a different maximum
temperature within it.
120
170
155
FIG. 19. Triangular composition diagram where binary azeotrope is a saddle
SYNTHESIS OF DISTILLATIONBASED SEPARATION SYSTEMS
105
The only way we can select between these two options is to obtain more
information, either by computing the trajectories using a physical properties
model we believe or by obtaining experimental data for the problem.
B. DISTILLATION
COLUMNBEHAVIOR
1. Limiting Behavior
There are two limits at which we can examine the behavior of a distillation
column. The first is at total reflux (i.e., with an infinite reflux ratio, which is
often called infinite reflux conditions). The other extreme is to operate at miniinurn reflux. In this section we shall limit our discussion to the total reflux case;
in later sections we shall look at operating columns at finite reflux (ratio) conditions. Intuitively, we tend to expect that a column will give its maximum
separation when run at infinite reflux. While this is true for ideally behaving
species, it does not have to be true when separating nonideally behaving species.
Thus, we need to look carefully at running columns all the way from minimum
to total reflux conditions.
2. Reachable Products for Total Reflux
One of the steps in developing alternative structures for a separation process
is to discover the possible products for a proposed technology. There is no
general method to accomplish this task for mixtures displaying complex behavior, even for distillation. For nearly ideally behaving mixtures, identifying possible products is a trivial task, sufficiently so that it is seldom recognized as a
required step in posing solutions. For the special case of separating ternary
mixtures using distillation columns that produce two products, Wahnschafft
(1992) and Wahnschafft et al. (1992) show that all possible products for azeotropic mixtures can be determined with a relatively simple analysis which involves residue and distillation curves and pinch point trajectories.
To trace out these curves is to compute a sequence of flash calculations (a
relatively easy task compared to computing distillation column performance).
What is remarkable is the fact that we really need to use residue curves (those
curves produced by solving ordinary differential equations) for a part of this
analysis. Until now, we have assumed that these curves corresponded only to
the time behavior of a batch still. Here, however, they become a necessary
ingredient in the analysis of a staged distillation column behavior instead of a
convenient approximation.
To understand the reachableproduct problem, let us imagine separating an
equimolar ideally behaving threespecies mixture of species A, B, and C (A
I06
ARTHUR W. WESTERBERG A N D OLIVER WAHNSCHAFFT
being most volatile, C the least) in a two product column. We illustrate these
ideas in Fig. 20. Assume the column has a very large number of trays and that
it will be run at very high reflux.
We first remove only one drop of distillate top product D, letting the rest of
the feed exit from the bottoms product of the column. The distillate will be pure
A, while the bottoms will be virtually all of the feed. We show these two where
the distillatetofeed ratio D/F is zero. We then draw off 1 8 of the feed in the
top and 99% in the bottom. We will continue to remove pure A in the top and
the rest of the feed in the bottom. Material balance dictates that the feed must
lie on the straight line joining the distillate and bottoms compositions. The bottoms moves directly away from the feed toward the BC edge as we remove
more and more A. When we reach the point where we are removing onethird
of the feed in the top, we will remove essentially all the A in the top and all
the B and C in the bottoms. If we remove SO% of the feed from the top, it will
be all of the A and half of the B, with the remaining half of the B leaving in
the bottoms together with all the C. When the distillate is twothirds of the feed,
we have all the A and B in the top and all the C in the bottoms. At 80%, we
take all the A and B and some of the C out the top, with the bottoms being the
rest of the C.
We can then backmix whatever is taken off in either product with the other
to reach any compositions that lie between the product compositions. Thus, we
can reach any product in the shaded bowtie region on this figure.
At total reflux, the column itself (without the backmixing) can only reach
products that lie both on the same distillatioil arnv (not residue curve) and on
a straight line passing through the feed so as to satisfy the overall column species
FIG. 20. Distillate and botloms products versus D/F for an ideally behaving ternary mixture
SYNTHESIS OF DISTILLATIONBASED SEPARATION SYSTEMS
107
material balances. The limiting distillation curve for nearly ideal mixtures is that
passing through the feed.
For this limit, think of removing one drop of top product, with the rest leaving
in the bottoms while operating at near total reflux conditions. The top distillate
product (one drop) will lie anywhere along the distillation curve passing through
the feed, while the bottoms will be the feed. A similar argument holds for
removing one drop in the bottoms at (near) total reflux conditions. Figure 21
shows the reachable region for such a column without the backmixing that we
allowed before.
For nearly ideally behaving mixtures, the above completes the analysis
needed to identify the reachable region at total reflux. For nonideally behaving
mixtures, the shape of the distillation curves can lead to very interesting reachableproduct regions. An Sshaped curve, for example, can lead to two disjoint
reachable regions, a situation we shall examine later. The rule to remember for
total reflux is that the column can produce any distillate product D and bottoms
product B where
the compositions for D and B lie on the same distillation curve, and
the composition for the feed F lies on a straight line between the compositions for D and B.
VIII. Synthesis Method for Nonideal Mixtures
We now consider the separation of a nonideal mixture where the species do
not display liquidliquid behavior. Almost certainly, the technology of choice
C
m
I\
B
distillation
A
FIG. 21. Reachable region for a twoproduct column operating at total reflux.
I08
ARTHUR W. WESTERBERG A N D OLIVER WAHNSCHAFFT
will be distillation. In the following example, we examine the need to discover
reachable interesting products when feeding a mixture into a distillation column.
By selecting different interesting product sets, we generate design alternatives.
As with the nbutanollwater separation example. we shall also see the need for
recycle in completing the design. Here we explore two different reasons for
using recycle: (1 ) to separate a mixture produced later in the process that is
similar to an earlier one (i.e., discovering a r e c u r ~ i \ ~so/irtion)
e
and ( 2 ) to adjust
the feeds to a column to allow it to produce two interesting products instead of
one (i.e.. rnirirzg to get better separation, an apparent contradiction).
A. AZEOTROPIC
SEPARATION: EXAMPLE
2
Consider the separation of the mixture of acetone, chloroform. and benzene
shown in Fig. 22. How do we generate alternative structures systematically for
solving this problem'?
We start by sketching the "binary separation tasks" for this problem. as
shown in Fig. 23. As we suggest separations, we can keep track of which portions of these tasks they accomplish. We use the arrows on Fig. 23 to show the
minimum separation that must be accomplished between each of the pairs of
species if the product specifications are to be met. The feed composition is also
illustrated. To interpret this diagram. suppose the feed is 36% acetone and 24%
chloroform. yielding a
Benzenefree composition =
36
~
36
+ 24
 60% acetone
We can see this on the sketch for the acetonekhloroform pair. The chloroform
product has to be at least 99% chloroform so it can contain at most 1% acetone,
36% Acetone
24 % Chloroform
40% Benzene
FIG. 2 2 . Specifications for the separation of an acetone/chloroform/benztne mixture.
SYNTHESIS OF DISTILLATIONBASED SEPARATION SYSTEMS
Acetone
Acetone
109
Benzene
Feed
Chloroform
Benzene
Chloroform
FIG 23. Binary separation tasks and relative concentrations of the feed mixture.
while the acetone can have no more than 0.1% chloroform in it for it to be at
least 99.9% pure.
We can map these tasks on the edges of our ternary composition diagram,
as shown in Fig. 24. Note that a line projected from the pure acetone corner
through the feed to the opposite edge has a constant ratio of benzene and chloroform all along it. Thus, the projected point on the benzenekhloroform edge
benzene
feed (acetone
free basis)
acetone
4
t
b
i sipation
task for acetone
and chloroform
h
w
' \
feed (benzene
free basis)
b
chloroform
FIG. 24. Mapping separation task onto edges of triangular composition diagram.
110
AKTHUK W. WESTERBEKG A N D OLIVER WAHNSCHAFFT
is the composition of the feed on an acetonefree basis. Similar projections from
the chloroform and benzene corners give the feed compositions on a chloroformfree and benzenefree (should be 60% acetone) basis, respectively. The mapping
of the tasks is the result of similar projections. Note that the task along the
bottom is to separate acetone from chloroform. Near the chloroform edge, we
see a small gap. which says the chloroform product on that basis can be as much
as 1% acetone.
We next need to know how these species will behave. First, their normal
boiling points are 565°C. 61.2"C. and 80.1"C for the acetone, chloroform, and
benzene. respectively. Thus, acetone is the most volatile, while benzene is the
least. The lowest temperature. 56.5"C. is hot enough for cooling to be done
using cooling water (i.e., above 2530°C). so it makes sense to consider running
the columns at a pressure of 1 atm.
We predicted their behavior earlier using infinitedilution Kvalues, with the
results at 1 atm shown in Table VIII. Only the acetone and chloroform appear
to display azeotropic behavior. With this information and that for pure species
boiling points at the pressure of interest. we can sketch the ternary diagram for
this mixture. We can also use a computer code to generate it, which was done
for Fig. 25. We see that there is one maximumboiling azeotrope between acetone and chloroform.
Two features appear on the residue curve diagram in Fig. 25:
( I ) A distillation boundary exits. We deduce this when attempting to explain
the azeotropic behavior determined using infinitedilution Kvalues.
( 2 ) The boundary is curved. This, too. can be partially predicted by noting
that the infinitedilution Kvalues for acetone and chloroform in lots of benzene
indicate that acetone is more volatile. Therefore, chloroform acts like an intermediate species in the benzenerich end of the diagram. The residue curves start
out aiming at chloroform from benzene.
What if we are dealing with a computer that does not draw or read these
nice graphs. or what if we are dealing with more than three species? How might
we proceed to find this behavior? We can carry out column simulations for a
TABLE VIII
INFINIT~DII.I!TION
KVAI.UESAND THEIRINTERPRETATION
K
In:
Acetone
Chloroform
Benzene
L
Acetone
Chloroform
Benzene
1 .o
0.6
3.0
0.4(max)
1 .O
1 .S
0.7 (normal)
0.5 (normal)
1 .o
SYNTHESIS OF DISTILLATIONBASED SEPARATION SYSTEMS
111
Benzene 80.1 "C
.+
'
Acetone
56.5"C
I
I
I
I
1
1
1
0.8
0.6
0.4

1
I
I
Chloroform
0.2
61.2%
XAcetone
FIG. 25. Residue curve diagram for acetonekhlorofordbenzene mixture. A residue curve
boundary passes from the maximumboiling azeotrope between acetone and chloroform to pure
benzene.
column having lots of trays and a very high reflux ratio. The simulations can
be carried out versus the amount of distillate product drawn off relative to the
amount of feed, i.e., versus D/F. We shall assume for the moment that a column
with lots of trays and a large reflux ratio will give us the maximum separations
possible. That is true for ideally behaving mixtures, but (as we noted earlier and
shall discuss later) it does not have to be true for azeotropic mixtures. Nonetheless, we proceed with that assumption.
Figure 26 is a sketch of the results obtained by carrying out these rigorous
simulations. (It should be noted that some of these simulations can be very
difficult to converge.) For small D/F, the top product of our column is pure
acetone, the most volatile species in the region of the feed. At D/F = 0, the
bottoms product is, in fact, the feed. As we take more and more overhead, the
bottoms product moves on a trajectory away from the feed composition in the
direction opposite the pure acetone vertex, continuing either until all the acetone
is removed or until the bottoms product hits a distillation boundary. If we do
not hit a distillation boundary, we should continue to get pure acetone until
D/F = 0.36, the fraction of the feed that is acetone. However, at D/F values
112
ARTHUR W. WESTERBERG AND OLIVER WAHNSCHAFFT
Benzene
AcetoneI
0 to 0.31
D/F
0.6
Chloroform
FIG. 26. Trajectories for distillate and bottoms products versus D/F for column
above 0.31, the rigorously simulated column will no longer yield pure acetone
in the top. We surmise we have hit a distillation boundary at the bottom of the
column as the bottoms still contains acetone. and we now find chloroform showing up in the distillate.
Increasing D/F further causes more and more chloroform to appear in the
tops while all of the benzene remains in the bottoms; it becomes purer and purer
in benzene as we proceed. When D/F reaches 0.60, the bottoms is 40% of the
feed, and it is essentially pure benzene. The feed was 40% benzene. Thus, the
top must be essentially pure acetone and chloroform, which it is. Increasing
D/F further forces some of the benzene to exit with the top product. At D/F =
1, the top product is the feed. The limiting composition for the bottoms is pure
benzene, the point where the temperature is highest in the distillation region in
which the column is operating.
We have. with this set of computations, discovered a portion of the residue
curve boundary for this mixture. It is the portion that is most relevant for our
feed mixture. This analysis also gives us a first estimate for the set of reachable
products using distillation for this mixture. (The analysis is not complete. as we
shall see later, because it does not discover what we might be able to reach with
smaller reflux ratios.)
it is possible to show these results without using a triangular composition
plot. In this form now to be proposed. we can imagine using a computer to
detect the behavior we have just described.
113
SYNTHESIS OF DISTILLATIONBASED SEPARATION SYSTEMS
First, we formally define the idea of a binary separation range for each of
the binary pairs. This range is the length of the vectors on the plots we used
earlier to indicate the part of a task that a column accomplishes. An equation
to compute such a range for each value of D/F used in simulating the column
is the following:
separation rangejj =
I(J)
h
+fi
distillate
(J)
h
+fi
bottoms
1
If both species go entirely to opposite products, the separation range is unity.
If both are entirely in one product, there is no separation and the separation
range has a limiting value of zero. We are looking for the maximum points in
this plot, Fig. 27. (There are other useful representations we could plot, such as
the molar flowrates for each of the species leaving in the distillate or bottoms
or the splits on the speciese.g., 50% going into the distillate. Each of these
gives a slightly different view, but all are aimed at indicating interesting products
among the many possible.)
1
0.9
0.8
a5
0.7
0.5
0.4
0.3
0
0.1
0.2
0.4
0.3
0.5
0.6
0.7
D:F
FIG. 27. Separation ranges obtained by rigorous distillation column simulation.
114
ARTHUR W. WESTERBERG A N D OLIVER WAHNSCHAFF7
We would see the same interesting points we discovered before in these data:
At D/F of 0.31, we maximize the separation of acetone to chloroform.
At D/F of 0.6, both the acetonebenzene and benzenekhloroform ranges
maximize at unityimplying that all the benzene is in one product and
acetone and chloroform are in the other.
At just below D/F of 0.6, we find a point where the amount of acetone in
the bottoms product is so low that acetone can exit with the chloroform in
a subsequent column while leaving the chloroform product contaminated
with no more than 1% acetone, which is pure enough to meet specs.
We show these separations on a triangular composition plot in Fig. 28. The
three interesting splits are as follows:
The direct split: A top product of pure acetone, the bottoms being a mixture
of all three species in significant amounts.
The indirect split: A bottom product of pure benzene, the top being only
acetone and chloroform(two interesting products).
The intermediate split: A bottom product which is a mixture of benzene
and chloroform with a trace of acetone. The ratio of acetone to chloroform
Benztne gO.1 "C
'
56.5"C
0.8
0.6+
0.4
XAcetone
0.2
61.2"C
FIG. 28. The three most interesting splits for the acetonekhloroformknzene problem.
SYNTHESIS OF DISTILLATIONBASED SEPARATION SYSTEMS
115
in this bottom product is just that required for the chloroform product1% acetone relative to the chloroform. The top is a mixture of acetone and
chloroform.
I . Indirect Split: Alternative I
The indirect split produces two interesting products in one column. However,
we can quickly see that removing pure benzene as the bottom product in the
first column is not a useful starting point. It leaves us with a binary mixture of
acetone and chloroform to separate. A second column will give us a top product
of acetone and a bottom product which is the maximumboiling azeotrope
formed by acetone and chloroform. We will need some way to break this azeotrope, which, as we shall discover, can be done using benzene. Thus removing
benzene first is, in fact, counterproductive. Anticipating better success by leaving
benzene in, we shall rule out considering this option.
2. Mixing to Get Two Desired Products from One Column: Alternative 2
We might be able to use mixing to get two interesting products from a single
separation. We note we can mix benzene with the original feed to move the
material balance line so it permits acetone to be the top product and a fairly
acetonefree benzenekhloroform mixture to be the bottom product. Doing so
means we can get a solution to our problem that involves only two columns.
The material balance lines for the mixing task and the two columns are shown
in the triangular diagram in Fig. 29, along with the corresponding process
flowsheet.
First, benzene is mixed with the feed, yielding the mix point M. This mixture
feeds column 1, yielding a pure acetone top product and a bottoms of virtually
all the benzene and chloroform plus a trace of acetone. This bottom product is
fed to column 2, where it is separated into benzene product and chloroform
product. Part of the benzene product is then recycled to mix with the original
feed.
Recycling material to alter a column feed so a single separator can produce
two desired products can often be a feature of these processes. In the next
alternative solution, we shall again propose a recycle, but this time it allows us
to separate an intermediate product.
3. Direct Split: Alternative 3
The next alternative flowsheet we shall consider starts with the direct split,
where we take pure acetone off as product from the first column. The material
116
ARTHUR W. WESTERBERG A N D OLIVER WAHNSCHAFFT
Benzene 80.1 "C
Fic;. 29. Flowzheet resulting from use of recycle mixing to improve separations.
balancc lines for the columns and mix points required are shown in the triangular
plot in Fig. 30.
The first column produces nearly pure acetone from the top and a mixture
of all three species in the bottom. This mixture is near the distillation boundary
and occurs when D/F is about 0.31. as we saw earlier. We feed the bottom
product to a second column. which separates benzene (bottom product) from
the acetone and chloroform in its feed (top product). We find we can draw a
material balance line through the mixture fed to column 2 , which connects benzene t o a mixture of acetone and chloroform in the other (righthand side) region
heryiuse of the ciir\ature of the rlistillcitioii hoirtidun. This curvature is often
important in devising separation schemes.
The top product from the second column. D,, is separated into pure chloroform (top, Dj) and a mixture of acetone and chloroform very near the maximuniboiling azeotrope (bottom, B 3 ) . We now seem to have an impasse. as we have
an azeotrope to separate. just as we did for alternative 1. This time, however.
the impasse is not "impassable."
We examine the separation ranges covered by the three columns already
proposed. Figure 31 shows these ranges. To read this figure, examine the three
ranges shown for column 1. The range covered by column 1 for the acetone/
SYNTHESIS OF DISTILLATIONBASED SEPARATION SYSTEMS
117
Benzene 80.1 "C
B2
o
Intermediate and
product compositions Intermediate and
product compositions converged with recycle
56.5 "CD i
f
XAcetone
FIG. 30. Threecolumn alternative that starts with direct split.
Acetone
Acetone
Benzene
Benzene
Chloroform
Feed
1%
0.1 O
Chloroform
FIG. 31. Separation ranges covered by three columns for second flowsheet alternative. Column
numbers are shown on their respective ranges.
118
ARTHUR W. WESTERBERG A N D OLIVER WAHNSCHAFFI
chloroform binary mixture is from a top product of pure acetone to a bottom
product that is about 15% acetone in chloroformthe mixture on the distillation
boundary. We can discover the “15%” by projecting from pure benzene through
the bottom product to the acetonekhloroform edge. This projection gives the
composition of the mixture on a benzenefree basis.
Projecting from the chloroform vertex through the top and bottom products
to the acetonebenzene edge gives the range covered for the acetonebenzene
split by column 1. The binary range is from pure acetone (top) to a mixture of
about 1 1 % acetone in benzene (bottom).
Projecting from the acetone vertex to the benzenekhloroform edge shows
that column 1 does not affect the ratio of benzene to chloroform. It is the same
in all products as it is in the feed (for the pure acetone, this is a limiting condition). We show the range for column 1 for this binary mixture as being unchanged from the feed.
Figure 31 also shows the results for columns 2 and 3. Examining all these
separations, we see that the entire ranges for all binary pairs are covered with
the structure proposed so far. We have. in a sense, already solved the separation
problem. We should therefore consider feeding the azeotropic mixture coming
off the bottom of column 3 back into this process to separate it. We propose
putting it back into column 1, as all three columns are needed to cover all the
ranges that this feed requires for it to be separated. Figure 30 shows the flowsheet for this option with this recycle.
Feeding the azeotrope back shifts the composition of the total feed to column
I to a point between F and B3. We need to check that the resulting flowsheet
will function as proposed. It does: the point M is the result of carrying out a
rigorous simulation for the flowsheet, including the recycle. In the upper left of
Fig. 30, we show the range covered by the acetonekhloroform split by the
original structure (open 1 o o p i . e . . having no recycle) and by the final structure
(with the recycle).
Finding that an intermediate column product can be recycled because a structure already exists to separate it is like discovering a recursive solution to our
separation problem. This reason for recycling is different from the one for the
previous alternative; it is a necessary part of many of these flowsheets.
4. ltirrrriiediure Split: Altenintiiv 4
The last alternative we consider is to start with the intermediate split, where
we remove a mixture of benzene. chloroform and a trace of acetone in the
bottom product of the first column. (See Fig. 3 2 . ) Without a systematic approach. we would very likely miss this alternative. We separate the bottom
product from the tirst column into benzene and chloroform in a second column.
The top product from the first column is a mixture of acetone and chloroform,
SYNTHESIS OF DISTILLATIONBASED SEPARATION SYSTEMS
119
FIG. 32. Alternative threecolumn flowsheet. Again, note mixing of the azeotropic mixture, B3,
with the feed.
which we separate in a third column into acetone and the azeotrope between
the acetone and chloroform. Using reasoning similar to that for the third alternative, we propose recycling this azeotrope back to the first column.
5. A General Approach to Creating Alternatives
Figure 33 summarizes the steps we take to generate alternatives for these
complex separation processes. We can illustrate the approach by reviewing the
steps we took for the current problem.
Starting with a feed mixture and product specifications, we analyze the feed
and products to discover the binary splits required. We then place the feed onto
a stack of streams to be processed. The stack is not empty, so we select the
feed stream. It is not a product. We determine its binary concentrations and pass
through the next two steps downward because no structure already exists that
could serve as a partial structure to separate it. We select a separation method
using our knowledge about the problem; here, we pick distillation. The “v”
shown in this box means that several different options might be proposed here,
with the different solutions branching along different paths. As we are considering ordinary distillation, we do not need to pick a separation agent.
120
ARTHUR W. WESTERBERG AND OLIVER WAHNSCHAFFT
8
Determine
Q
4
Flowsheet
inW0dUCe
splitters to
account for
mixing goals
determine its
binary
"secondary"
recycle
I
"prim+lIy"
recycle remove
mixing goal
""we
extending
recycle"
0
mixing goal
and new
separation
choose
separation
agent
used 7
FIG. 33. Algorithm to create alternatives.
1
0
searchfor
reachable
1
interesting
SYNTHESIS OF DISTILLATIONBASED SEPARATION SYSTEMS
121
Simulations for differing D/F ratios sweep out the reachable separations. The
interesting products among these are acetone, pure benzene, an acetone and
chloroform mixture with no benzene, and a bottom mixture of benzene, chloroform, and a trace of acetone. We need to consider each of these as starting
points for alternatives.
No separating agent was used, so we consider using mixing to create separation tasks that might allow two interesting products to occur in the same next
step. We do see such a possibilitymixing something (which we selected to be
benzene) with the feed to move it to a material balance line between acetone as
a top product and the benzene/chloroform mixture with a trace of acetone as
the bottom product. The need for something to mix with the feed is placed on
the list as a new separation task. Note: Anything above the material balance line
that can move the feed to that line is a legal stream for mixing. It just happens
that benzene is a fairly obvious choice here.
We remove the feed from the stack and continue with the current option. We
have a top product of acetone. Since acetone is a product, we cycle back from
the step that tests if it is and remove it from the stack. We also have the benzene/
chloroform mixture with a trace of acetone to process.
We make this mixture the stream to consider and proceed through the steps
with it. The alternative flowsheets arise when we return to steps where we
created alternatives, such as the one in the lower right where we have a number
of interesting products not yet considered.
There are three recycle decision steps shown in the middle of the diagram.
An instance of a primary recycle is the use of another stream (we used benzene)
to move the feed to the material balance line between two interesting products
so that both may be produced by one separator. Benzene, being a product, would
be discovered in the step before last (introduce splitter to account for mixing
goals remaining) rather than here, however. The recycling of the azeotropic
mixture of chloroform and acetone back to the first column is an example of a
secondary recycle. We have not seen the third case for recyclea range extending recycle. There are times when a downstream separator does not have
some of the species in its feed which, if they were there, it would separate to
some extent. Thus we might still propose recycling a stream if the flowsheet
proposed so far almost covers all the splits needed to process that stream.
We now consider a more complex example where IiquidAiquid extraction
and extractive distillation are among the processes to appear in the solution.
B. AZEOTROPIC
SEPARATION:
EXAMPLE
3
We wish to devise a separation process based on distillation, liquidniquid
extraction, and extractive distillation for the mixture of solvents shown in Fig.
122
ARTHUR W. WESTERBERG A N D OLIVER WAHNSCHAFFT
3
0
1.42% Water
Fic;. 34. Specifications for feed and products for Example 3.
34. Solvent recovery systems are often very complex systems to design, as they
need to separate very different molecules.
We start by computing the infinitedilution binary data shown in Table IX.
The upper values are Kvalues at the boiling point of the more plentiful species.
We will also need liquid activity coefficients if we wish to consider extraction
processes: the lower values in each entry are infinitedilution binary liquid activity coefficients at ambient conditions.
We now need to think of all the reasonable ways we might separate this very
complex mixture. We shall use insights from the above data as well as any
insights we have as chemical engineers. This step is a knowledgeintensive one.
Since pentane and water exhibit immiscibility, we might consider decunrution
as the first step. If i t worked, it would be an inexpensive one to carry out. But
a rigorous threephase equilibrium calculation predicts that, in the presence of
acetone and methanol, the small water fraction in the feed does not form a
second liquid phase: so we reject this idea. The calculation also reveals that the
feed mixture is almost at the azeotropic composition for the pentane/methanol
binary pair.
~!WISITEDII.ITION
TABLE IX
PAIKWISE KVALUESAND ACTiViTY COEFFICIENTS
FOR EXAMPLE3
c, \ c,
Pentane
Acetone
Methanol
Water
Prntane
I .0
1 .0
7.9
4.7
29.6
14.4
8106
3213
3.0 (min)
6.6
1 .0
5.9 (min)
23.1
1.3 (min)
2.0
71.4 (het)
I537
1.05 (min)
7.4
0.4 (ok)
1.6
1 .0
Acetone
Methanol
Water
1 .o
2.4
7.0
38.5
11.5
I .o
1 .o
7.8
22
1 .o
SYNTHESIS OF DISTILLATIONBASED SEPARATION SYSTEMS
123
Next, we consider distillation; but, with the exception of water and methanol,
all other pairs exhibit azeotropic behavior. Also, water is present in very a small
amount here. We should be trying to get rid of the pentane first. Thus, distillation
does not seem an appropriate first step.
If distillation is rejected, we might consider extractive distillation. The K values of acetone and methanol at infinitedilution (38.5 and 7.8, respectively)
in water indicate that water could be used as an extractive distillation agent for
the separation of these species. However, adding water would almost certainly
introduce a second phase with the pentane.
If used, we would create two liquid phases in view of the calculations we
did above to see if decantation is a good first step.
If water will force us to have a second phase, we might then consider using
liqui&Ziquid extraction. We shall see that this is a good suggestion. Before
evaluating it, however, we need to review some important ideas associated with
liquidliquid extraction.
Liquidliquid extraction is typically used to remove small amounts of heavier
species mixed with a large amount of a light species. Conventional distillation
would require a large amount of the lighter species to be condensed overhead
and would tend to be uneconomic. An example would be to remove a small
amount of ethanol from a lot of diethylethyl ether. As a bulk separation method,
liquidhiquid extraction is also suitable for isolating fractions of species with
similar molecular structure from other species, as is the case when separating
alkanes from aromatics. The extraction is done, for example, by finding a solvent
such as acrylonitrile that likes the aromatics and forms a separate liquid phase
with the alkanes. Extraction is usually performed at temperatures well below the
boiling points of the species involved, which makes activity coefficients, y, a
better means of evaluation than vaporliquid Kvalues.
The composition of species distributing between two liquid phases, I and 11,
is determined by equating their respective fugacities:
where xi are mole fractions, yi are activity coefficients, andfi are standardstate
fugacities. Assuming that the same standardstate fugacities are used for both
phases (for example, pure liquid i at the temperature and pressure of the mixture), we see that the ratio of mole fractions is the inverse of the ratio of activity
coefficients; i.e.,
To separate two species, we need to know how well the two phases differentiate between them in this ratio. Separability factors, $$, are frequently used
to indicate potential for separation:
123
ARTHUR W . WESTERBERG A N D OLIVER WAHNSCHAFFT
Returning to our example solventseparation problem, let us consider using
liquidliquid extraction to remove pentane from methanol a n d o r acetone. A
suitable solvent is one that is immiscible with the bulk species, pentane. If at
all possible, we would like the solvent to be present in the mixture so we do
not have to introduce any other species into our separations problem. For our
example problem, water is present. Noting the infinitedilution Kvalues for water and pentane, we see that water will be highly immiscible with pentane.
Aside from being immiscible with the pentane, the solvent, water, has to
effect a reasonably different distribution of the species to be separated, methanol
and/or acetone and pentane.
We consider methanol first. For a quick estimate of the separation factors
that water can produce. we consider the limiting selectivity that would be obtained if the methanol and pentane were infinitely dilute in both the extract phase
(waterrich) and the raffinate (pentanerich) phases if we were to use liquid
liquid extraction:
Selectivity is excellent.
Although selectivity between acetone and pentane (using water and pentane
as the extract and raffinate. respectively) is not quite as high (= 1850), it still
suggests that we could separate these species, too.
We use rigorous simulation to determine feasible separations using water as
a solvent. For a theoretical tenstage liquidliquid extraction process, we find
that rather little water is needed to recover virtually all methanol from the pentane. At higher solvent flowrates the waterrich extract contains more and more
acetone. but it cannot produce a complete separation of acetone and pentane.
Thus. we select the solvent flow at which the methanolpentane separation is
sufficiently sharp. Figure 35 gives the separation selected.
As shown. the raffinate stream, F , leaving such an extractor essentially
contains pentane and acetone plus a trace of water. This pentanerich mixture
can be separated in a distillation column, producing pentane as bottoms and a
distillate which is limited to the minimumboiling azeotrope between pentane
and acetone. (The infinitedilution Kvalues of 7.9 and 3.0 indicate that such an
azeotrope exists.) Figure 36 gives the flowrates and relative concentrations between the major species of the streams entering and leaving the distillation
column (labeled Di2) that carries out this separation for the raffinate (pentanerich) stream.
,,
SYNTHESIS OF DISTILLATIONBASED SEPARATION SYSTEMS
125
F11
P 17.4168
A I.0054
P
A
M
W
EX1
FO
: Pentane
: Acetone
: Methanol
: Water
7r
FIG. 35. Liquidiliquid extraction step to recover methanol.
It is evident that we have produced one of our desired products here, nearly
pure pentane. We now have to separate the nearazeotropic stream, F2,.An ideal
situation occurs if we can recycle it back to the liquidliquid extractor. If its
composition is close to that of the original feed to the process, we could simply
mix it with the feed. Our two separation devices then would process this feed,
giving us a pure pentane product and, by material balance, a second stream
which is the feed but with this same amount of pentane missing.
How might we decide if the stream F2, coming from the top of the second
column is close enough in composition of the extractor feed to be recycled back
and processed with it? Two characteristics of the stream FZLits composition
and its flowratemust be important in this decision. If the compositions of the
two streams are really close, we assume there should be no problem, no matter
the flowrate. On the other hand, if there is very little flow in F2,relative to the
feed (for example, only one drop), then we should almost always be able to
F11
A 1.0054
7I
I
F21
0121
FIG. 36. Distillation column to remove acetone from pentane.
126
ARTHUR W. WESTERBERG AND OLIVER WAHNSCHAFFT
recycle it. no matter its composition. The total flowrate for the distillate stream
F21is 16% of that of the original total feed to the process. It is modest, but not
negligible. It contains essentially only pentane and acetone, which are in the
ratio of 3.3 to 1 , while the feed has these same species in the ratio of 6.2 to 1.
We can perform approximate material balance calculations to see what happens to the process if we were to recycle the distillate and process it with the
feed. We model each separator as a set of constant split factors for each species.
For example. we see that 97.94% of the pentane, 35.00% of the acetone. and
none of the methanol leave in the top stream from the extractor. Water enters
as the extraction agent and also as a small part of the feed; 0.63% of the total
water entering leaves with this top stream. We capture these results in the first
three rows of numbers in Table X. We denote molar flow for species k in stream
j leaving unit i by pj,(k], and the fraction of the flow of species k in the feed to
unit i leaving in streamj by 5,Jkl.
Examining the distillation column. we find that 19.00% of the pentane entering it leaves in its distillate product. Similarly. 99.98% of the acetone and
I052 of the water entering leave in the distillate. We assume these numbers do
not change even if we recycle the distillate product from the second column.
We capture these numbers in Table X, too.
The following equation gives the material balance for the total flow &(k)
of a species k entering into the extractor when we recycle stream F21.
P b m = PoAh
+
POdk)
+
5lI(k)52l(~)F&W
For example. the total flow of pentane into the extractor equals the pentane
in the original feeds plus that which recycles, which is 0.9794 . 0.1900 times
the total flow of total pentane into the extractor. Solving for the total flow of
species k into the extractor, we get
TABLE X
BAL.WCESFOR FIRST
LJXITS 'TO ESTIMATEIMPACT
REC'YCi.INC DISTILLATE
P R O D ~ T T BACK TO EXTRACTOR
AFTPHOXIMATE hf.ATEHI.AL
Total original feed. p,,l + p,,.
l o p product from extractor
Fraction of feed in top product. t l l
Dislillatc product from column
Frrrction column feed in distillate. tzl
1
Fllth)t21(k)
New feed (original feed times above factor)
Composition original feed
Composition of extractor input with recycle
1
OF
Pentane
Acetone
Methanol
Water
17.7827
17.4168
0.979
3.309
0.1900
1.229
1.8729
I ,0054
0.3.50
I .0052
0.9998
1.538
2.6842
0
0
I
3.654
0.0230
0.0063
0.0023
o. I on0
I.0006
21.9
0.659
0.670
4.42
0.106
0.136
2.68
O.I(K)
0.082
3.64
0. I35
0.112
~
127
SYNTHESIS OF DISTILLATIONBASED SEPARATION SYSTEMS
which, when applied to pentane, gives
Total flow of pentane into the extractor
=
17.7827/(1

0.1861)
=
21.9
In a similar manner, we can compute the total flows for the other species, getting
the results shown in the third line from the bottom of Table X. Finally, we
compute the compositions for the original feed and for the feed after adding the
recycle so we can compare them. We see that these compositions are surprisingly close to each other. The net result of recycling here is to increase the flow
of the feed to the extractor unit by approximately 20% but with little impact on
its composition. We therefore elect to recycle the distillate.
We now return to consider the extract stream FI2,from our first separation
step. It still contains all species in nonnegligible fractions. We propose that this
mixture be distilled. We check if distillation is reasonable by using rigorous
simulation. While we do not show the details of these runs here, they were
carried out by varying D/F ratios over a range of values as we did when we
were investigating the acetone/chloroform/benzene example earlier. These runs
indicate two interesting product sets. One allows for the complete removal of
water. However, because water is a candidate separating agent for the acetone/
methanol split, we put it aside and consider the other optionnamely, to recover
all the pentane.
Figure 37 shows the results from carrying out a rigorous simulation for this
option. Because pentane forms azeotropes with acetone and methanol, these
species appear in both products. Noting that although F31 is a much smaller
stream than FZ1,it has similar compositions, we decide to recycle it back to the
extractor feed also. The bottom product of column DI3 consists of acetone,
methanol, and waterwith no pentune.
Let us examine what we have accomplished with the liquid/liquid extraction
step followed by the two distillation columns. The “upper” column, DI2, produces essentially a pure pentane product, which we remove. The “lower” column, DI3, produces a product with no pentane. We recycle all the other streams
M 2.6842
A 1.8675
F32
FIG. 37. Distillation column to remove pentane from methanol and water.
128
ARTHUR W. WESTERBERG AND OLlVER WAHNSCHAFFT
99.886 %
I
1.7541
0.0013
d
W 2.75
W 3.612
A 1.7557
2.6536
F42
F32
FIG. 38. Extractive distillation column to separate methanol from acetone.
so these three units, collectively, are there to remove pentane from the original
feed.
So what do we do with the mixture coming off the bottom of the lower
column, DI3? Looking at our earlier data, we see that water boils at a higher
temperature than the two remaining species and does not form an azeotrope with
either of them. The Kvalues of acetone and methanol at infinitedilution in water
(38.5 and 7.8, respectively) suggest that water could be used as an extractive
distillation agent for the separation of these species. In such a column, acetone,
being decidedly more volatile, is recovered as a pure distillate product. It should
be noted that, because of the tangent pinch between acetone and water, this
column might best be operated below ambient pressure to exploit the improved
vaporliquid equilibrium near this tangent pinch. The results shown in Fig. 38
are from simulating such a column.
Finally, the bottom product of the extractive distillation column, ED4, can
be separated in a simple column since it contains the nonazeotropic species
methanol and water only. This column is shown is Fig. 39, again based on a
rigorous simulation.
The process produces all the desired pure species products. The water product
from our last column, DI5, also has to provide the water used as the extractive
agent in the liquidliquid extraction step and in the extractive distillation step.
The complete set of steps is shown in the flowsheet in Fig. 40.
We now need to put in the proposed recycle flows to see if the structure
discovered is possible when they are present. Rigorous simulation gives the
results shown in Fig. 41. It is noteworthy that, despite the complexity of the
nonideal mixture behavior, it has been possible to obtain all species as highly
pure singlespecies products using only water which is already present as the
separating agent.
129
SYNTHESIS OF DISTILLATIONBASED SEPARATION SYSTEMS
"0
L;;3n
f
A"EE
w 0.0001
W 6.3607
e
I
FIG. 39. Watdmethanol separation.
\
pr.EJ
w 0.0001
l
FIG. 40. Total process flowsheet before adding recycle streams.
I 30
ARTHUR W . WESTERBERG AND OLIVER WAHNSCHAFFT
I
F21
F11
b 012
I
P 17.7827
EX1
4
F31
F22
F64
k[
W 0.3333
r
FO
EX : LiquidLiquid Extraction
ED : Extractive Distillation
DI : Distillation
FS : Splitter
P : Pentane
A : Acetone
M : Methanol
W :Water
F42
FIG. 11, Total flowsheet after including recycles
We note that the liquidliquid extraction step accomplishes separations across
three relatively azeotropic compositions: namely, that between acetone and pentane, that between methanol and pentane, and that between water and pentane.
It should also be noted that. since the extractor does not have to perform sharp
ceparations. it requires an overall optimization to determine how many stages
should be used. In principle, a single stage or, in other words, a simple decanter
hhould suffice to make the process feasible.
The comparison between the simulation results for the original steps while
they were being proposed and the final process with the recycles included shows
that it is indeed possible to maintain the separation functionality of a sequence,
and even to slightly improve the achievable puritiesone of the essential premises of the sequential synthesis approach being advocated here.
Alternatives to this process are generated by returning to those steps where
we made decisions and checking to see if alternatives decisions might exist. For
example. we could look for a different extraction agent in the IiquidAiquid extractor. We eliminated one obvious alternative when we discovered that adding
more water to the liquidliquid extractor did not allow us to remove all the
acetone from the top product stream even though there is a good separation
factor suggesting this alternative.
Sargent (1994) presents a related approach to the synthesis of distillation
processes. His goal is to generate a superstructure of interconnected columns
from ;I “state/task” network. The superstructure contains all the process alter
SYNTHESIS OF DISTILLATIONBASED SEPARATION SYSTEMS
131
natives as substructures. For ideally behaving species, he reproduces the superstructure proposed earlier by Sargent and Gaminibandara (1976).
For nonideal systems, Sargent first determines all the distillation regions. He
then labels the pure species and all the azeotropes as pseudospecies in the
system, giving the label A to the most volatile, B to the next, and so forth. For
example, he would label the diagram in Fig. 25 for acetone, chloroform, and
benzene with A for acetone, B for chloroform, C for the maximumboiling
azeotrope between acetone and chloroform, and D for benzene. The region on
the left side of the distillation boundary has the species A, C and D, while the
region to the right has the species B, C, and D. He then uses the “bowtie’’
analysis we discussed earlier (and which we shall discuss in more detail in
Section 1X.B) to identify the reachable products for each of the regions. For
example, in the left region, a first column can produce A, AC, CD, and D as
productswhere CD is a product along the distillation boundary. The product
CD can be split in a second column. If the distillation boundary were straight,
the distillate would be C (azeotrope) and the bottoms would be D (benzene).
Here, however, the boundary is curved, indicating a mixture of B and C (azeotrope and chloroform) as the distillate product and D as the bottoms.
Sargent gives heuristic rules to suggest where to place recycle streams. Also,
when two different columns give rise to the same product, he merges the products in the superstructure. By eliminating parts of the superstructure to form a
substructure that produces all the products of interest, one generates a design
alternative. Dropping different parts generates different alternatives.
To find the best substructure for a given feed, the idea is to optimize this
superstructure to find the optimal substructure. Two potential difficulties arise:
( 1 ) It is difficult enough to solve the model equations for a single column displaying azeotropic behavior, much less optimize a superstructure containing several such columns with numerous recycles; (2) the superstructure almost certainly has local optima, requiring the use of much more costly approaches to
find global optima. Although we shall overcome these restrictions in the next
few years, they are currently serious difficulties.
IX. More Advanced Preanalysis Methods
We reexamine how to assess species and equipment behavior when dealing
with highly nonideal mixturesthis time in much more detail.
BEHAVIOR
A. SPECIES
In this section, we discuss advanced methods of finding azeotropes and determining if a mixture displays liquid/liquid behavior.
I31
ARTHUR W . WESTERBERG A N D OLIVER WAHNSCHAFFT
I . Fhciirig A:rotropes
If a mixture forms an azeotropic composition, the vapor and liquid compositions at equilibrium are identical to each other. The following equations define
this situation:
y, = K J ,
K,
=
for i = 1,2. ...,
iiC
K,(T,P, all x,, all yI)
”C
2.r;
I=
= 1
I
y, = s,
,for i
=
1,2, ..., iiC  1
There are 3 1 1 ~+ 2 variables (xi. J,. and Ki for all species, temperature T, and
pressure P ) in these 3nC + 1 equations. If we fix pressure, the model is completely fixed. Solving, we will determine the bubblepoint temperature for the
azeotrope as well as its composition. We note. therefore, that an azeotropic
composition for a mixture is. as we already knew, pressuredependent.
Suppose we have a mixture of water, pyridine. and toluene. We set the pressure to I atm and attempt to solve the above equations. Lacking any further
insight. we set all the vapor and liquid compositions equal to 0.33333 and then
attempt a solution using a Newtonbased method. The problem with finding
azeotropes becomes immediately evident. There are six solutions to these equations: the three binary azeotropes and the three pure species. There is no ternary
azeotrope. To find all azeotropes for a mixture, we must find all solutions to the
above equations. Finding multiple solutions to a set of highly nonlinear equations like these is usually a very difficult task.
We can attack this problem in two ways: ( 1 ) try to find a numerical procedure
that will find all roots to a set of nonlinear equations, or (2) try to develop a
method that uses physical insights to find all the roots.
We know of no numerical procedures that will guarantee finding all the
solutions to an arbitrary set of nonlinear equations. “Continuation” methods
are often capable of finding more than one solution if several exist. Fidkowski
P r cil. ( 1993) propose using such a method along with discovering bifurcation
points to compute all the azeotropic compositions for a mixture. Their homotopy
function
SYNTHESIS OF DISTILLATIONBASED SEPARATION SYSTEMS
133
causes the Kvalues for the species to move from those Raoult’s law predicts
to their nonideal values as the homotopy parameter t moves from 0 to 1. Except
for very closeboiling species, there will be no azeotropes assuming Raoult’s
law. Thus, we know that only the pure species will satisfy these equations when
the continuation parameter is 0. Their algorithm starts by solving these equations
for t = 0 at each of the pure species. As the continuation parameter increases,
these equations can become singular at one or more of the pure species’ starting
points. If they do, the singularity indicates the solution trajectory plotted versus
the continuation parameter branches, one branch staying at the pure species
(always a solution) and the other heading to the composition of a binary azeotrope along one of the adjacent binary edges of the composition diagram. An
eigenvector analysis of the local linearized behavior of these equations indicates
the direction for the bifurcating solution. The temperature along these branches
increases for a maximumboiling azeotrope and decreases for a minimumboiling azeotrope. The solution along one of these branches may itself become
singular, indicating a further bifurcation. Fidkowski and colleagues conjectureand their computational experience suggeststhat one will discover all azeotropes in the system by tracing all these solution trajectories starting from all pure
species.
Wahnschafft (1994) has developed and tested extensively an interesting alternative method of finding all the azeotropes. The slight change that Wahnschafft makes has a significant impact on how to find azeotropes. He replaces
nC
the third and last of the equations above (i.e., c y ,
=
1 and yi = xi for i
=
i= I
1,2, ..., nc  1) with
Kj = 1
for i
=
1,2, ..., nc
This alternative set of equations is generally satisfied only at azeotropic points
involving all nc species. Assuming for the moment that each nc species system
has only one azeotropei.e., that a binary mixture has only one binary azeotrope, a ternary system has at most one ternary azeotrope, and so forththe
advantage of the above formulation is that there are no longer multiple solutions
to the same set of equations, but rather a different set of equations to find each
of the azeotropic points. In particular, the pure species generally do not satisfy
the azeotrope condition for any set of K j = 1 involving more than just one
species. The latter could happen only if an azeotrope occurred exactly at a pure
species pointwe would be talking about the special situation of a tangent
pinch.
In principle, we could use any rootfinding method for solving problems
involving the equations K, = 1 for combinations of species representing different
tuples, such as all binary combinations, all ternary combinations, quaternary,
134
ARTHUR W. WESTERBERG A N D OLIVER WAHNSCHAFFT
and so on. up to the one system of equation involving all nc species. However,
the rootfinding methods may still suffer from convergence problems. Wahnschafft found that it is more robust to devise a continuation method that can
lead up to the solution at the azeotropic points one wants to find.
Wahnschafft based his continuation method on a relaxation of the azeotropy
condition for a given iiC species system. At an azeotropic point, all nc Kvalues
equal unity, which is a special situation of the condition that they are equal. In
the surroundings of an azeotrope involving rzc species, the Kvalues are no
longer equal to unity, but subsets of them will be equal to each other; in other
words, the relative volatility between species pairs will still be unity at certain
points close to an nc species azeotrope.
Consider Fig. 42, where we plot three trajectories for a ternary system that
involves three binary and one ternary azeotropes. At a binary azeotrope, the
relative volatility between the two species involved, i.e., their Kvalue ratio,
equals unity. We can seek out the trajectory emanating from this azeotrope into
the three species region along which the relative volatility of these two species
remains equal (i.e., along which their Kvalues remain equal). The Kvalue for
the third species will generally be different. However, we find that, owing to
the existence of the ternary azeotrope, there will be one specific composition
along this trajectory where the Kvalue of the third species becomes the same
as the other two; moreover that situation occurs as all three Kvalues become
equal to unity. We can find the same point by tracing along another curve
starting at either of the other binary azeotropes.
The trajectories along which two (or more) species have the same volatility have been called isovolatility cunies. Based on the ability to trace out
such curves in the composition space, we can outline a simple algorithm to
robustly determine all azeotropic points that are predicted for a multispecies
mixture:
(1)
Fix the pressure at the value desired.
C
FIG. 42. Trajectories having subsets of all Kvalues equal to unity.
SYNTHESIS OF DISTILLATIONBASED SEPARATION SYSTEMS
135
(2) Compute the infinitedilution Kvalues for all pairs of species, as we did
earlier to check for the existence of binary azeotropes. For this example,
we would find that all three binary pairs display an azeotrope.
(3) Where a binary azeotrope exists (e.g., along the AB axis), use a Newtonbased method to find the point along that axis where the two Kvalues
are unity.
(4) At that azeotrope, compute the infinitedilution Kvalue for one of the
missing species.
( 5 ) Increase the composition for that missing specieshere, species Cmonitoring its Kvalue. Solve the equations, keeping the Kvalues for the
other two species equal to each other. Monitor the Kvalue of the newly
introduced species (here, species C ) to determine if its value gets close
to unity (e.g., by checking for changing from values larger than 1 to
values less than 1 or vice versa, and by checking the derivative of the
Kvalue along the curve to see if it passes through a minimum or maximum). In either of these cases, we have approached a ternary azeotrope.
Switch to solving the equations for the azeotrope (where all Kvalues are
unity) using a Newton based method. If the trajectory hits the side of the
composition space, stop searching along it.
(6) Given a ternary azeotrope and a fourth species in the mixture, compute
its infinitedilution Kvalue at this ternary azeotrope. Increase the amount
of the fourth species, keeping the Kvalues for the other three equal to
each other. Stop where the fourth Kvalue passes near 1 or shows a
minimum or maximum along the trajectory being traced out. Solve directly for the fourspecies azeotrope. Stop searching this trajectory when
it hits a side of the composition space.
(7) Repeat the second step above to find the remaining binary azeotropes. If
more than a single binary, ternary, or higherorder azeotrope is suspected,
march away from each of lowerorder azeotropes toward it, as is done
in the third and fourth steps above.
(8) And so forth.
The algorithm stops looking for azeotropes involving n + 1 species when there
are no azeotropes involving n species. When applying this algorithm, we have
to solve the equivalent of several flash unit calculations for each trajectory we
trace. This algorithm very efficiently finds all azeotropes predicted by the physical property models being used for a multicomponent mixture. It does not require an eigenvalue/eigenvector analysis to spot bifurcations. Finally, it can be
made to take advantage of the topological constraint that Zharov and Serafimov
(1975) developed (see Section VILA. 1.c).
Doherty and Perkins (1979) examined the possibility of the existence of ternary azeotropes if there are no binary azeotropes. They show that the infinite
136
ARTHUR W. WESTERBERG AND OLIVER WAHNSCHAFFT
dilution Kvalue for one of other species must be exactly unity at a pure species
node, which is equivalent to saying that the isovolatility trajectory starts at a
point which is a pure species and a binary azeotrope at the same time. Obviously
such situations will be extremely unlikely. but the proposed algorithm would
not present any difficulty in moving to a ternary azeotrope for this case.
We are aware of no papers that prove this type of result for azeotropes
involving more than three species; but, assuming such results do exist. the above
algorithms should find all azeotropes.
a. Exmiple: Ethaiiol arid Water. We first detect the existence of the azeotrope
by carrying out two bubblepoint calculations. each at nearinfinite dilutionthe first with molar compositions of 0.9999 and 0.0001 and the second with
0.0001 and 0.9999 of ethanol in water. The desired Kvalue for each calculation
is the ratio of the vapor composition. y, to the liquid composition, .I.for the
trace species; we get 1.29 (the Kvalue of water for a trace of water i n ethanolj
and 14.6 (the Kvalue of ethanol for a trace of ethanol in water), respectively.
As discussed earlier, these numbers indicate a minimumboiling azeotrope exists. very likely for an ethanolrich composition.
We next search over the range of compositions from 0 to 1 for ethanol in
water. computing a bubble point for each. At low ethanol composition, ethanol
is more volatile. For a mixture that is nearly pure ethanol, the reverse is true.
We are looking for the point where the Kvalue for ethanol changes from above
unity to below. Near this point, we switch from a bubblepoint computation to
that for the equations above where we set the two Kvalues to unity. A Newtonbased convergence algorithm converges easily to the desired azeotropic composition,
= 0.868 (using Unifac) if we start the calculation close to the
answer.
b. E.mrip/e: Ethanol, Water; arid Toluene. For a ternary mixture, we first discover one or more of the binary azeotropes, as we propose to use one of these
azeotropes to start the search for a ternary azeotrope. We just discussed finding
the ethanol/water azeotrope: we can also quickly discover the ethanoUtoluene
azeotrope .
In looking for a water/toluene azeotrope, we run into the problem that these
two species generally form two liquid phases, a waterrich one and a toluenerich one. We cannot ignore this twoliquid phase behavior, lest we get nonsensical answers. An azeotrope occurs when the composition for the combined
liquid phases equals that for the vapor phase. Having Kvalues for a species set
to unity, as we discussed above, means that the composition for that species for
the combined liquid phases matches that for the vapor phase. Whether allowing
[or two liquid phases or not. water and toluene form a binary azeotrope for a
SYNTHESIS OF DISTILLATIONBASED SEPARATION SYSTEMS
137
Ethanol 78.5 OC
p = 1 bar
Toluene 110.6 O C
84.0 C
Water 100.0 QC
FIG. 43. Ternary diagram for water, ethanol, and toluene.
mixture that is somewhat more than 50% water. When allowing for two liquid
phases, the temperature is, as shown below, at 84°C. If we do not permit twoliquid phase behavior, the temperature is much lower, namely, 65.6"C, a result
that does not match at all with experimental data.
Figure 43 shows these azeotropes as well as a ternary one we now wish to
find. Starting at the ethanol/water binary azeotrope, the infinitedilution Kvalue
for toluene is 2.78. Allowing for two liquid phases, the above algorithm locates
the ternary azeotrope without difficulty. If we do not allow for two liquid phases.
computations indicate there is no ternary azeotrope.
2. Discovering LiquidLiquid(/Liquid) Behavior
Wasylkiewicz et al. (1993) present a method to find regions where mixtures
partition in two or more liquid phases. They base it on the Gibbs tangent plane
test (Michelsen, 1982, 1993) to decide if a current composition resides in a
single or a multipleliquid phase region.
We shall first illustrate, using Fig. 44, why a mixture will split into two or
more liquid phases by examining the shape of the Gibbs free energy for a binary
138
ARTHUR W. WESTERBERG AND OLIVER WAHNSCHAFFT
I
x2
XO
composition, x
FIG. 44. G i b b free energy for binary mixture that breaks into two liquid phases at any compohition between .v, and .vI.
mixture versus composition. We examine in particular the composition so. Its
singlephase Gibbs free energy is above the tangent plane that supports the Gibbs
free energy function at points .xi and x. The single phase can reduce its Gibbs
free energy by splitting into two phases having the two compositions xi and x2.
The total Gibbs free energy for the two phases is
a value that lies, as illustrated, on the tangent plane below G(xo).
Based on this insight, we place a plane that is tangent to G(x) at the composition of interest. If that plane lies entirely below G(x) for all x, then the given
mixture will remain as a singleliquid phase at equilibrium; otherwise. it will
split into multipleliquid phases. Figure 45 illustrates for the point xo. The tan
x,
composition, x
iit
FIG. 45. Tangent plane supporting G(x) at x,,. If G(XJlies above this plane anywhere, the mixture
.v,, will break into multiple liquid phases.
SYNTHESIS OF DISTILLATIONBASED SEPARATION SYSTEMS
139
water
toluene
npropanol
FIG. 46. Stationary points in the modified Gibbs free energy function over the composition
space for multispecies mixtures. Diagram is approximate.
gent plane that supports G(x) at xo lies partially above G(x) on the righthand
side of the figure. This mixture lies between compositions x, and x2 on the
previous figure and will split into these two liquid phases. But how do we carry
out this test? It would appear we have to cover the composition space with test
points and hope that no regions are missed.
Wasylkiewicz et al. (1993) presented a method to carry out this test for
general multispecies mixtures. They pick a point to test and create a tangent
plane for it. The distance between the tangent plane and the Gibbs free energy
surface defines a nonlinear surface above the composition triangle. The starting
point is on a ridge in this surface. These authors propose tracing all ridges
starting from this point to find all extreme point .along these ridges in the surface
(as opposed to searching the entire composition space). These paths can bifurcate. Figure 46 illustrates a typical path that their algorithm will trace. Only the
extreme points need to be checked to discover if the tangent plane goes above
the Gibbs free energy surface, thus indicating liquidliquid behavior. The extreme points are good guesses for the phase compositions when they detect
liquid/liquid behavior. Ridgefollowing and bifurcation detection involve evaluating eigenvectors and eigenvalues.
To test a single composition is a considerable amount of work. To develop
a phase diagram for a mixture, we have to place compositions strategically over
all of the composition space, detecting tieregions (as lines, triangles, etc.). Fortunately, a tieregion covers all the compositions in it which no longer have to
be explored. Figure 47, based on a figure in Wasylluewicz et al. (1993), illustrates a completed phase diagram.
McDonald and Floudas (1994) and Michelsen (1994) also present methods
to find the phase conditions for a given mixture. Both methods discover the
number of phases and the compositions of these phases. McDonald and Floudas
globally minimize the Gibbs free energy of the mixture using a computer pack
1 40
ARTHUR W. WESTERBERG AND OLIVER WAHNSCHAFFT
water
toluene
npropanol
FIG. 47. Resulting liquidfliquid phase diagram. Diagram is approximate.
age they call GLOPEQ. Michelsen formulates the problem as a twolevel problem. The outer level computes phase fugacity coefficients for all the possible
phases. They show how to formulate the inner problem as a convex continuousvariable minimization problem.
B . LIMITING
SIMPLE
DISTILLATION
COLUMN
BEHAVIOR
We shall determine the limiting behavior for simple distillation columns by
discovering all the products of such a column, regardless of the number of stages
it has or the reflux ratio used in operating it. We present and extend here the
ideas advanced in Wahnschafft (1992) and Wahnschafft et al. (1992).
I . Reuchuhlu Regions
To develop the alternative process configurations needed to separate a given
feed mixture into a set of specified products, we need to know just what distillate
and bottoms product compositions we can reach when using a conventional
distillation column. We shall start by examining this problem for ideally behaving mixtures. We shall then look at the much harder problem in which the
mixtures do not behave ideally.
The Material Balatice Constraint. The following constraint holds for any
number of species (see Treybal, 1968):
LI.
Material Balance Constraint: For a ccmvetitional siiiglrfeed, twoproduct
distillatioti calirmti operating at steady state. the feed cottiposition niiist
lie oii ci straight lirie between the compositioris for the distillate mid
bottoms prodiicts.
141
SYNTHESIS OF DISTILLATIONBASED SEPARATION SYSTEMS
b. InJnite ReJux. There are two convenient ways to imagine having infinite
reflux conditions in a column. First we can consider having a real column where
we turn off the feed and the product flows while maintaining a finite flow for
the internal liquid and vapor, L and V. We get an infinite reflux ratio by having
a zero denominator in the equation
The second extreme we can imagine is to maintain finite flows for the feed and
products but increase the internal flows for L and V to infinite values. This
second case cannot really occur, as we would need a column with an infinite
diameter. It is a limiting case. Both ways to think of infinite reflux are useful.
In the latter case the column is still thought of as producing its products.
At infinite reflux we can add a second constraint that must hold for the
compositions of the two products:
Infinite Re$ux Constraint: The compositions for the distillate and bottoms
products must both lie on the same distillation curve.
This constraint follows from the definition of a distillation curve. Each point
along a distillation curve represents both the vapor and the liquid compositions
just above (or below) any tray, including those at the ends of the column where
products would normally be withdrawn. This constraint, together with the material balance constraint, completely defines the reachable products for a column
at total reflux.
Figure 48 illustrates some possible pairs of product compositions we might
obtain for a given feed. It is for a relatively ideal threespecies mixture, H A L
H
A
r
h
distillation
F~
I
distillates
D
. B
L
FIG. 48. Examples of reachable products for a column operating at infinite reflux.
142
ARTHUR W. WESTERBERG AND OLIVER WAHNSCHAFFT
The line marked with a I is one of two extreme situations we can see for this
example. It has a bottoms composition on the distillation curve passing through
the feed and a distillate composition which is exactly the feed. It is the limit of
maintaining a finite feed while reducing the bottoms product flow to zero. If we
take one drop of bottoms and the rest as distillate product, the distillate will
have the same composition as the feed. The bottoms product is along the distillation curve at a point corresponding to the number of trays in the bottom of
the column. Allowing for fractional trays, we can reach any point along this
line.
Line 3 illustrates the other extremewhich is to connect a bottoms composition lying on the IH edge to one lying on the LI edge. There is a limiting
distillation curve that passes from L to I to H on which these two points lie.
Line 2 connects two points lying on a distillation curve that is between these
other two extreme distillation curves for this example.
By plotting all such points, we create the shaded reachable region shown on
Fig. 49. The straight edge from F to the LI edge occurs because any point along
that edge can be a distillate if pure H is the bottoms product. All distillation
curves passing through that edge will ultimately reach H, given an infinite number 'of trays.
If we bypass some of the feed and then mix what we bypass with either the
distillate product or with the bottoms product, we can fill all the bowtie area
between the straight line passing from L through F to the IH edge and the line
passing from H through F to the LI edge. The point marked a in Fig. 49 illuswhite space that can be
filled by bypassing and
istillation curve
F
I
L
FIG. 19. All reachable products for a simple distillation column separating a nearly ideal mixture
with the given feed composition F.
SYNTHESIS OF DISTILLATIONBASED SEPARATION SYSTEMS
143
trates. This composition is reached by producing the two products at the extreme
ends of the line as these are on the same distillation curve. Then by mixing the
bypassed feed with the bottoms, we reach point a.
c. Finite Reflux (Petlyuk, 1978; Petlyuk et al., 1981; Wuhnschufft, 1992; Wuhnschufft et al., 1992; Poellmunn and Bluss, 1994). To understand what we can
reach by reducing the reflux, we can write a material balance around the top of
a column (as shown in Fig. 50):
Vn+l = L,
+D
This material balance says that the composition for V,+, will lie between the
compositions for L,, and D on the straight line connecting them.
We can imagine being on a composition diagram and stepping down the
column starting at the top tray. Figure 51 illustrates the case when we have a
total condenser and withdraw a liquid distillate product. (We leave it to the
reader to discover that essentially the same analysis holds if we withdraw a
vapor distillate product. If the distillate is a twophase equilibrium mixture, the
point D iies on the straight line joining the distillate’s vapor and liquid compositions.) At the top of the column for a total condenser, the vapor leaving the
top tray, V , , condenses to form the distillate, D, and then refluxes back to the
column, Lo. Therefore, they all have the same composition. We step along a
distillation curve from that point to the composition for L1, which is in equilibrium with that for V1. It is exactly one stage along this curve. From the argument
above, the composition for V, must lie on the straight line connecting L1 to D.
We step along the distillation curve, passing through V , one stage to Lz. We
find V, along the straight line connecting L2 to D.
We can keep on stepping along in this manner until we run into a situation
where the compositions start to repeatthe characteristic of a pinch point for
the column. Here, we see V,l leading to L,. For a pinch situation, we find both
V, and V,,, along the same straight line back to D from L,, and these points
coincide. Then L,,, coincides with L,, ad infinitum. Note that we are talking
FIG. 50. Flow in the top of a column.
144
ARTHUR W . WESTERBERG A N D OLIVER W A H N S C H A F n
vl
x
\\
distillation curves
FIG. 51. Stepping down
a
b,
I
residuecurve
column from the top.
about the same pinch points we discussed earlier. Thus we have already talked
about how we might find such points by carrying out a computation similar to
a flash computation.
There is a lot of interesting geometry in this diagram. First, we can look at
where V,,, is placed along the line relative to the positioning of L,, and D.From
the level rule. we write
,
L,, _
  = R,,
D
b
where n and b are the lengths shown in the diagram. This equation says that
the reflux ratio defined in terms of the liquid flow leaving the rzth stage divided
by the distillate top product flowrate is the ratio of the line lengths n to h. If
the reflux ratio does not change much, then this ratio does not change, and V k + ,
is always about the same fraction of the distance along the line from Lk back
to D. Figure 51 shows all the vapor compositions (V,. V,. and finally V,,) positioned in this manner.
Next we remember the equations that detine a residue curve:
These equations have an interesting geometric interpretation. They say that the
vapor composition V,, in equilibrium with the liquid composition L,, lies along
SYNTHESIS OF DISTILLATIONBASED SEPARATION SYSTEMS
145
a line that is tangent to the residue curve passing through the composition L,?.
If that line also passes through D, then V,,, is on the same line. It is possible
to prove also that V, and V,,, must be coincident and therefore so are L,, and
L,,,. The proof follows from observing that if V, and V,,, are on the same
line back to D, then L,, and L, must also be coincident. If L,, and L, are
not coincident, then there must be at least two points where residue curves are
tangent to that line. The proof takes some effort, but ultimately it demonstrates
that these points are coincident.
Finally, from the results of these observations, we can prove that a pinch
point is any point where a line emanating from the distillate composition is
tangent to a residue curve. The vapor composition in equilibrium is on that
same line. The reflux ratio to reach that pinch point is the ratio A.
Figure 52 illustrates stepping to a pinch point from a liquid distillate composition on a triangular diagram. This diagram is also rich in geometric interpretation. Starting at the distillate composition with a particular reflux ratio for
the column, we step down the column as shown, finally ending at a pinch point.
If we choose a slightly larger reflux ratio, we reach another pinch point farther
FIG. 52. All compositions that can be reached for both finite and infinite reflux conditions from
the distillate D and bottoms B compositions by stepping away from them using a traybytray
calculation.
146
ARTHUR W. WESTERBERG A N D OLIVER WAHNSCHAFFT
away from D. In this manner, we can construct a curve made up of pinch points.
Which pinch point we will reach in a column depends on the reflux ratio; thus,
this curve is parametric in the reflux ratio. We saw this parametric behavior
earlier when we discussed pinch points just before developing insights into Underwood’s method for estimating minimum reflux.
Any point between the distillation curve passing through the distillate and its
corresponding pinch point curve is reachable from the distillate using traybytray computations (allowing fractional trays). Here, for example, we could reach
the bottoms product labeled B by choosing a reflux ratio that leads to a curve
that traverses at first very close to the distillation curve through D until the traybytray computations bend inward toward the pinch curve. We stop at B by
choosing a finite number of trays (it takes an infinite number to reach the pinch
curve) along that curve. We define the gray region as the reachable region for
D. We can repeat a similar construction for the bottoms product. producing a
pinch point curve and then a reachable region for B.
If the shaded regions do not overlap, no conventional column can produce
these two products at the same time. We need to argue that the converse is true;
namely, if the regions overlap, a column exists that can produce these two
products. We would have no trouble making that statement for this particular
column as B is reachable from D directly using traybytray computations. We
would feed the bottom tray in this column (very likely not the best way to run
a column to get these two products).
There is a point f * that resides on both pinch point curves in Fig. 52. We
can make the following argument concerning P *:
By construction, f * is simultaneously on both pinch point curves.
As it is on the pinch point curve for D, the residue curve passing through
P* is tangent to the line from P* to D.
As it is on the pinch point curve for B , the residue curve passing through
P* is tangent to the line from P* to B.
Thus the line must be the same straight line (both pass through P* and
both have the same slope).
By overall material balance for the column, the feed point F is on a straight
line between D and B.
Thus the point f * must be on that line.
Thus the point P* must be a pinch point for F , too.
We therefore conclude that f * is a pinch point for all three points: D, B and F.
For a reasonable topology for the residue map, we next can discover that the
points between P* and F may not be products from any column for which F is
a feed. The construction in Fig. 53 illustrates. This figure corresponds either to
a total condenser with a liquid top product or to a partial condenser with a vapor
top product. If the top product is twophase. the distillation curve passes through
SYNTHESIS OF DISTILLATIONBASED SEPARATION SYSTEMS
147
FIG. 53. Topology of the point P', the intersection point for a liquid distillate and bottom
product pinch point curves.
both the vapor and liquid compositions for the distillatethey are in equilibrium
with each otherwhile the distillate pinch point curve passes through D.
First, the column feed F must lie on the line between the products D and B.
Both points F , and F2 are possible feeds that could lead to the product D
and B.
If the feed is F I , a point lying between D and P*, then the pinch curve
through D lies entirely to one side of the point F , and curves back to cross at
P*. The corresponding distillation curve lies even farther away to the same side
(the pinch point curve turns more sharply than the distillation curve). The reachable region for D excludes all points between D and P*. Thus the points between
F , and P* are unreachable by the top part of the column. As D could move to
be coincident with F 1 , points beyond F , are not excluded.
148
ARTHUR W. WESTERBERG AND OLIVER WAHNSCHAFFT
If the feed is Fz. the pinch curve for B excludes all points between B and
P* by a similar argument. Again. the points between F2 and P* are unreachable.
We have proved our assertion.
We now should understand the meaning of a reasonable topology. It is where
the distillation curves turn in only one direction over the region of interest, a
property they will have for ideally behaving species. We shall look in a moment
at a case where the topology is more complex.
The region between F and P* is a function only of F. the feed to the column,
as we can find P* by finding the pinch point curve for F from the arguments
above. We have subtly turned our attention from D and B to F. Thus, we can
map out this unreachable region by knowing only the column feed.
Excluding the option of bypassing and remixing any of the feed with either
of the products, the diagram for an ideally behaving set of species in Fig. 54
maps out the entire set of reachable products for a given feedfor any reflux
conditions from minimum to infinite. We can reach the lightly shaded region
using total reflux. We can extend the region with the dark gray areas by using
finite reflux.
FIG. 51. Reachable products for a column separating ideally behaving species.
SYNTHESIS OF DISTILLATIONBASED SEPARATION SYSTEMS
149
i. Sshaped residue and distillation curves. For nonideally behaving species,
some of the residue and distillation curves take on a more complex shape. We
shall examine the situation when some of them are Sshaped as in Fig. 55. SShaped residue curves appear to the left of the maximumboiling azeotrope
along the lower edge. They are also present in the residue diagram for acetone,
chloroform, and benzene (Fig. 25), appearing just to the right of the maximumboiling azeotrope along the lower edge. With little difficulty, we can demonstrate
that this shape is quite common. We now examine its implications for finding
reachable products for a column.
Figure 56 shows a region with Sshaped distillation curves, where the feed
lies somewhere near the inflection point for one of these curves. We note some
very interesting differences from what we have discussed so far. We know from
earlier discussion that we can reach any products at total reflux when both lie
at opposite ends of a straight line passing through F while also lying on the
same distillation curve. B I and D ,are such a pair. Interestingly, B2 corresponds
to two different distillate products, 0; and 0;'.
We see that the total reflux products are outside the region bounded by the lines passing to the minimum and
maximum temperatures for the region (total reflux products were wholly inside
this region before).
The shaded regions between the two Sshaped bounding curves are the reachable product regions for total reflux for the feed F. To see how we construct
these two bounding curves, consider the two points marked a lying simultaneously on the same straight line passing through F and on the same distillation
curve. The one to the left and below the feed lies where the straight line just
brushes a distillation curve. If we rotate this straight line counterclockwise, it
will no longer intersect this distillation curve. Thus, a column at total reflux
cannot reach points to the left of the upper point a where this straight line
intersects this same distillation curve. We create one segment of these bounding
140
130
135
Max 140
FIG. 55. Composition diagram featuring Sshaped distillatiadresidue curves.
1 so
ARTHUR W. WESTERBERG AND OLIVER WAHNSCHAFFT
distillation
curves 1
bow t i region bounded bv
temperatures for region
/ //
Fit;. 56. Reachable products (shaded are3 between two Sshaped bounding curves) at total refiux
f i x S\haped distillation curves.
lines by finding the trajectory of points where a straight line passing through
the feed just brushes a distillation curve and the other segment by locating the
corresponding point on the other side of the feed where this line intersects the
same distillation curve.
We find the same complexity for mapping out reachable products for finite
refluxsee Fig. 57. We ask if we can distill the feed F into the distillate product
D and bottoms product 5.The shaded regions indicate the reachable regions for
these two products. They overlap, suggesting we can reach them. We see there
are two feed pinch points. P; and Pi, along the straight line joining B , F , and
D. whereas we saw only one before. In a manner similar to that used earlier.
we can develop an argument that says that, if B lies between Pi and F , then D
cannot lie between Pi and F. Similarly, if D lies between P i and F , then B
cannot lie between P ; and F. These segments between P; and F and between
Pi and F are reachable (in contrast to earlier findings), but they are not simultaneously reachable in the same column.
SYNTHESIS OF DISTILLATIONBASED SEPARATION SYSTEMS
151
‘ 0
FIG. 57. Reachable products for finite reflux for an Sshaped region.
We see that the Sshaped region makes the process of discovering reachable
regions more difficult (but not impossible).
ii. Crossiizg residue curve boundaries (Wahnschafft, 1992; Wahnschafft, et
al., 1992). Nikolaev et al. (1979) and Van Dongen (1983), among others, demonstrate by column simulations that one can cross residue curve boundaries
when operating a column at finite reflux. Consider Fig. 58, which shows the
distillation curves for acetone, chloroform, and benzene. We are going to be
very particular and note that this plot features a distillation curve boundary as
opposed to residue curve boundary. No column operating at total reflux can
produce a distillate and bottoms product on opposite sides of this boundary
because the two products must reside on the same distillation curve. However,
this restriction does not hold for finite reflux. As we have shown above, we can
step to a product by carrying out platebyplate computations from any com
152
ARTHUR W. WESTERBERG AND OLIVER WAHNSCHAFFT
Benzene
Acetone
Chloroform
FIG. 5X. Bifurcation of the pinch curve trajectory when a product is in another distillation region.
position residing between the distillation curve passing through a product and
the pinch curve emanating from that product. We propose here a bottoms product in the righthandside region and a distillate product in the lefthandside
region. with the feed between as shown. Is this possible?
Examining Fig. 58. we see that there are two pinch point curves corresponding to B (i.e., to the stripping section of the column). Mentally treating the
distillation curves shown as residue curves (they will be similar in shape), we
see that a pinch point occurs where a straight line passing through the bottoms
product composition is just tangent to a residue curve. The pinch curve passing
through B, crossing into the lefthandside region, and ending at the acetone
node satisfies this requirement. However, so does a totally disjoint pinch curve
starting at the maximumboiling azeotrope, heading upward into the righthandside region, and ending at the chloroform node. For a small enough reboil ratio,
we will quickly move from the distillation curve passing through B toward the
pinch curve emanating from B. At total reflux, we will stay on the distillation
curve. and, for an infinite number of stages, we will end up at the chloroform
node which is on the disjoint pinch curve. Somewhere in between, we must
jump from the one pinch curve to the other.
SYNTHESIS OF DISTILLATIONBASED SEPARATION SYSTEMS
153
Figure 59 illustrates the process of stepping up the column away from the
bottoms product end. This figure reminds us of the geometry involved. First,
we note that L, must lie on a straight line between B and V,,,. Next, we
remember that the residue curve passing through L,, points at the vapor composition in equilibrium with the composition corresponding to L,,. We consider
three cases. (1) If the residue curve passing through the liquid composition
points as shown, the composition in equilibrium with L,, will be down and to
the right of L,. In this case, the column compositions will move to the right as
we step away from B, causing us to move toward the disjoint pinch trajectory.
(2) If the tangent to the residue curve points straight at B, then L, is a pinch
point, and we would stop moving. (3) If it points to the left as we move away,
the trajectory moves to the left. Thus we move to the disjoint pinch trajectory
only if the composition of the liquid steps across the lower pinch curve, i.e.,
passes from case (3) through case (2) to case (1).
Associated with every pinch point is a vapor composition in equilibrium with
it, a composition we already have from computing the pinch point itself. We
are reminded by Fig. 59 of the geometry for computing the reboil ratio to reach
a given pinch point. It is the ratio of the distance from B to L, (b on the figure)
divided by the distance from L, to V,,, (a on the figure). We can label each
pinch point on both trajectories with a corresponding reboil ratio. The point B
itself corresponds to a reboil ratio of zero. The reboil ratio increases to infinity
along the pinch trajectory emanating from B as it passes to the pure acetone
node. To see that it goes to infinity, we note that the distance b gets larger as

f3z Vn+,/B =
FIG. 59. Stepping up the column from the bottoms product.
I54
ARTHUR W. WESTERBERG A N D OLIVER WAHNSCHAFFT
we move away from B while the distance a must go to zero as we approach
acetone.
Using a similar argument. we recognize that the reboil ratio for the disjoint
pinch trajectory passes from infinity at the azeotrope to a finite but positive
number and back to infinity as we approach the chloroform node. If we keep
the reboil ratio less that the smallest reboil ratio along the disjoint trajectory,
we cannot reach this trajectory. So, again, we see that we need to use small
reboil ratios to cross a boundary.
iii. Bottonis conzpositions that permit crossing of a boundan. We are now
in a position to map out those bottoms compositions that have a reachable
products region that crosses the distillation boundary. Place a bottoms composition somewhere in the lower right of the composition diagram Fig. 58 for
acetone. chloroform and benzene. If it is close enough to the chloroform node,
two pinch point trajectories will again appear, but they will be qualitatively
different from before. The one starting at the bottoms composition will end at
the chloroform node. The second will move from the acetone node and end at
the azeotrope. In this case, the reachable region for B will stay entirely in the
righthandside region, lying between the distillation curve passing through B to
chloroform and the pinch point curve emanating from B and passing to
chloroform.
There must be one or more compositions somewhere between this one and
the one shown Fig. 58, where the trajectories switch from the one shape to the
other. We can imagine the two trajectories just touching and then trading
branches. Let’s explore where this will happen.
Examine Fig. 60. We show the residue curves that have the same Sshape as
those in the lower left part of the righthandside region for acetone. chloroform,
and benzene. Each curve is convex to the left at the top (curves toward the left)
and convex to the right at the bottom, and each switches the direction it curves
at its inflection point. Point n is such an inflection point on one of these residue
curves. Draw a straight line through point CI such that it has the same slope as
the residue curve. Above the inflection point, this straight line is entirely to the
right of the residue curve, and below it is entirely to the left.
Next. pick a bottoms product. B. that lies on this straight line above the
inflection point. The pinch point curve emanating from B will move to the left
initially because of the curvature of the local residue curves. It will move left
until it encounters point N which, by construction. is a pinch point for B. It cannot cross this residue curve. however, because any residue curve an infinitesimal
bit to its left can have no pinch point with B. On the other hand, a curve just
t o the right will have two pinch points with B, one just before CI and onc just
after. The pinch point trajectory thus “reflects” off this residue curve. After
encountering point a, it heads to the lower right and ultimately to the chloroform
node.
SYNTHESIS OF DISTILLATIONBASED SEPARATION SYSTEMS
155
pinch
curve
FIG. 60. The pinch point trajectory emanating from the bottoms product “reflects” off a pinch
point that occurs at an inflection point of a residue curve.
Thus, any point B that lies on a line passing through an inflection point of
a residue curve with the same slope as the residue curve cannot have a pinch
point trajectory emanating from it that crosses the distillation boundary. We
draw a family of lines through every such inflection point. To the left of all
these lines are the bottoms products, B, that can be bottoms products with a top
product in the lefthandside region. Points on and to the right of these lines are
not candidates.
Finally, we note with a similar set of arguments that no product in the lefthandside region can have a pinch point curve that crosses the distillation boundary, thereby justifying the statement appearing in the literature that one cannot
cross such boundaries from the “convex” side.
2. Thoughts on the Geometry for Two and Four or More Species
All the above insights are for three species. How do these insights extend to
mixtures with two species and with four or more species. If we extrapolate the
twodimensional (planar) triangular diagram for three species down to two, we
get a composition line. Each edge of a triangular diagram is such a line. For a
fourspecies mixture, we need a threedimensional tetrahedron such that each of
its four sides is a threespecies triangular diagram. Five species require four
dimensions, one more than we can comfortably visualize, so we almost certainly
must abandon visualization for five or more species. But we can attempt to
describe the geometry of expressing equilibrium, distillation, and residue curves;
operating ‘‘lines”; and, finally, pinch point trajectories.
I56
ARTHUR W. WESTERBERG AND OLIVER WAHNSCHAFFT
Vaporliquid equilibrium is a mapping from a liquid composition to a vapor
composition. It can be done by including tie lines from one to the other for all
compositions. On a line, such a mapping is very difficult to visualize, so we
typically use a second dimension where we can plot vapor composition versus
liquid composition as in a McCabeThiele plot or as in a temperature versus
composition diagram. For three or four species, showing tie lines is fairly direct.
The important point is that equilibrium is not a line (as we might think because
of our familiarity with McCabeThiele plots) but a mapping.
Distillation and residue lines that we have plotted for three species on a planar
triangular diagram remain as lines for all dimensional spaces. Each corresponds
to a trajectory of composition points. A distillation curve corresponds to the
curve we pass through the liquid compositions that occur on the trays of a
distillation column operating at total reflux. A residue curve is the trajectory we
get when we solve the differential equations
ds,
  x ,  y1
d8
Each curve is "pinned" down by requiring it to pass through a particular
composition.
For a fixed retlux ratio for a column, the operating line remains a line in
higherdimension composition space, as it too is a line passing through the
sequence of liquid compositions appearing on the trays as we move up or down
a column.
One line we have to think more carefully about is the pinch point line as we
move to other dimensions. Each point on the line is the end point of a operating
line for a distillation column operating with a fixed distillate (or bottoms) product, varying parametrically with the reflux ratio we use to define that operating
line. Thus, a pinch point curve emanating from a fixed distillate composition is
a sequence of points. It remains a line. It may bifurcate. as we have shown
above, but it remains a line.
The compositions from which we can reach a distillate or bottoms product
are those lying along a distillation curve if we operate at total reflux. If we
operate with a finite reflux ratio, these trajectories move off this line and toward
the pinch point trajectory emanating from that product, ending on it if we have
an infinite number of stages. Thus, the reachable compositions should look like
a ribbon moving through higherdimension space. Almost certainly, this ribbon
has bulges in it, but it should be a bounded twodimensional surface between
these two lines. However, this geometry is for reaching a particular product
coinposi tion.
Thinking about all reachable products for a given feed is more difficult problem, particularly for four or more species. We still require the material balance
constraint to hold. Thus, the points must line up on a straight line in whatever
SYNTHESIS OF DISTILLATIONBASED SEPARATION SYSTEMS
157
dimension space we examine. At total reflux, the points must also lie on the
same distillation line. On a triangular diagram, all distillation lines lie in a planar
surface, so we have little difficulty in seeing that these two constraints map out
a twodimensional portion of that diagram. But if four species are represented
in a threedimensional tetrahedron, do these same two constraints map out a
threedimensional portion or only a twodimensional portion of that tetrahedron?
Even if we answer this question, are we prepared to discover the shape of this
space for each problem we face, and would we find it useful if we did? It was
for this reason that we have proposed a different strategy to discover the reachable products for separating a mixture of four and more species: namely, the
plotting of separation ranges as in Fig. 27. This representation does not provide
the complete picture for three species, but it does give us one approach for such
mixtures that neither grows combinatorially with the number of species nor
defies our ability to visualize the results.
C. EXTRACTIVE
DISTILLATION
Many researchers have contributed to the literature on extractive distillation,
including Benedict and Rubin (1949, Hoffman (1969, Tanaka and Yamada
(1965), Berg and Yeh (1985), Hunek et al. (1989), Pham et al. (1989), Ryan
and Doherty (1989), Pham and Doherty (1990), Wahnschafft (1992), and Wahnschafft and Westerberg (1993). We shall base much of our discussion here on
the last two references.
1. Diference Point
Suppose we would like to separate water (normal boiling point 100°C) from
isopropanol (nbp 82°C). Isopropanol and water form a minimumboiling azeotrope (nbp 80°C) at about 72% isopropanol and cannot be separated into pure
products by ordinary distillation. We discover by computing infinitedilution Kvalues that, in the presence of sufficient ethylene glycol (nbp 197"C), isopropanol is more volatile than water, and the two can be separated completely. We
need to operate a distillation column such that, whenever isopropanol and water
are together in the column, there is sufficient ethylene glycol present to make
the isopropanol more volatile. Suppose we feed relatively pure ethylene glycol
(the extractive solvent E) as a separate feed a few trays from the top of a column,
as shown in Fig. 61. Being a highboiling material, it will largely head straight
down the column in the liquid phase. The feed (where A is the isopropanol and
B is the water) enters several trays below the solvent. The section of the column
between the two feeds washes out B. Just below the solvent feed, it must be
essentially completely removed, or else it will end up in the distillate product.
1 S8
ARTHUR W. WESTERBERG A N D OLIVER WAHNSCHAFFT
r

A

E+B
FK;. 61. Typical extractile distillation column
The bottom section of the column is to remove species A. Species A heads
down the column but is removed more and more until, just before the bottom
tray. it i h more or less completely depleted. The top few trays above the solvent
teed are to separate A from E. As ethylene glycol is a very highboiling material
relative to isopropanol, two trays will remove it.
We can determine the reachable products for this type of column by extending the concepts developed for an ordinary column. We again develop our insights for a threespecies mixture using a ternary composition diagram.
We start by writing a material balance for the section of the column between
the two feeds. as shown in Fig. 62.
=
i,,+ D

S
This equations looks just like a material balance for a normal column, except
that we have replaced D by D  S.
Let us define A as a difference point (Hoffman. 1964)
A = D  S
FIG. 62. Intermediate section of extractive distillation column
SYNTHESIS OF DISTILLATIONBASED SEPARATION SYSTEMS
159
and let it play the role of D in a normal column. Thus, A is the constant difference that we will see between the total (and species) vapor and liquid flows
in this section of the column. For a normal column, the operating line must pass
through the composition point for D. For the intermediate section of this column,
the operating line must pass through the composition point for A, given by
Let us look again at the case of separating isopropanol from water using
ethylene glycol. If we assume the solvent is essentially pure ethylene glycol and
the top product essentially pure isopropanol, then
XA,E =
OD  1.0s  . S
D  S
DS’
XA,A =
1.OD  OS  . D
D  S
DS’
xA.B =
0
For D > S, the first composition is negative and the second is greater than
1. The third (for water), being 0, says that this composition lies on the edge of
the triangle opposite the node for pure water (B). A negative composition for
the solvent E and a composition greater than unity for species A says the A
point lies outside the normal composition triangle. Figure 63 shows where A
will lie for the case of pure solvent feed and pure distillate top product. As
noted, when D > S, X A , E is negative which must be past the N B edge. Composition x ~ is ,greater
~ than unity, which must be above the vertex for A. In
\
D>S, A>O
\
\
\
\
\
FIG. 63. Location of A point and corresponding material balance line for extractive distillation.
160
ARTHUR W. WESTERBERG A N D OLIVER WAHNSCHAFFT
other words, the point must lie along the A/E edge extending above and to the
left of the vertex for pure A.
If D = S, the A point is at infinity along this line. As S starts to exceed D,
A jumps to negative infinity along this same edge. It moves toward the point E
as S gets larger and larger relative to D. This makes sense. When S = 0, the A
point should be at the distillate composition, here pure A. Adding solvent moves
it upward along the line, extending the A/E edge. It should ultimately approach
pure solvent as S becomes very large relative to D.
If S. = D  S is positive, the material balance constraint says the vapor
composition for f r , l + l is a mixture of the liquid composition for
and the
composition for A. Thus the point for
will lie between i, and A. This puts
fJ,jLlabove and to the left of i,,in Fig. 63. If A = D  S is negative, then the
same material balance equation says L, is a mixture of the vapor composition
and the composition for A, which now lies below and to the right (as
for
we have sketched it in Fig. 63). In this case,
is again above that for i,
along the line passing through A. Therefore, while A jumps from plus infinity
to minus infinity, the relative positions for
and instay the same, and their
positions move smoothly even as A jumps.
We stated above that isopropanol (A) is more volatile than water (B) in the
presence of lots of ethylene glycol (E). Figure 64 is a sketch of how the residue
curves appear for this mixture on a triangular composition diagram. Cover the
upper left part of the diagram, leaving only the node for E showingi.e., where
we circle the node for E. This part of the diagram looks just like that for an
ideal mixture where A is the most volatile species and B the intermediate. As
c,,+l
c,,+,
A
\
1
to A
Ffci 64. Stepping up the middle section of an extractive distillation column
161
SYNTHESIS OF DISTILLATIONBASED SEPARATION SYSTEMS
one passes along a residue curve toward species A, the system becomes
“aware” that A is not the lightest species around, but rather the minimumboiling azeotrope is. The curves turn and head for it instead, terminating at A.
They pick up a distinct S shape to them, as shown.
Let us assume we need to separate an azeotropic mixture of A and B. Since
a conventional distillation column cannot carry out this separation, we plan to
use solvent E as an extractive agent. We perform an analysis to see if it will
to the column is the sum of the azeotropic mixture
work. The total feed, Ftotalr
and the solvent E. Its composition will lie along the line joining the compositions
for the azeotrope and E. We want relatively pure A for the distillate (possibly
contaminated with a small amount of B); so we sketch in about where we wish
to find our desired distillate product composition. The bottoms product will be
a mixture of species B and E. We place the products on our diagram, noting
that both must lie on a straight line passing through the total feed composition.
If we sketch the compositions that can reach the distillate product in a conventional column, we find them limited to being near the AE edge. To see this,
follow the residue curve passing through the distillate product toward higher
and higher temperatures. The pinch point trajectory emanating from the distillate
product is a bit to the inside of this residue curve, but never too far from it
here. All compositions that can reach the distillate are bounded by this pinch
point trajectory and the distillation curves (close to the residue curve) passing
through the distillate composition point. Similarly, we see that the compositions
that can reach the bottoms are roughly those between the bottoms and the azeotrope. Not surprisingly, the two regions do not intersect, and thus we know we
cannot separate the total feed mixture (azeotrope mixed with solvent E) in a
conventional singlefeed column.
Our intention is to use an extractive distillation column. The extractive agent
is to be fed separately near the top of the column. We wish to see if this section
has a composition trajectory that steps between a pair of compositions that the
top and bottom sections can each reach from their respective product
compositions.
Let us set the solvent feed initially equal to that of the distillate, which puts
the A point at infinity. We choose this point initially only because it is easy to
sketch the lines passing through it: they are parallel to the AE edge.
and
infor the column section between the two feeds will lie on such a line, as
shown. Let us then locate
given in.The slope of the residue curve passing
through i, points at
We see that
is thus to the right of
We sketch
in a possible point. We note then that we move up the column when the tray
number decreases from n+ 1 to n. Thus,
moves toward the A E edge relative
to V,,,. This is the exact direction in which we must move to connect the
compositions reachable from the bottoms product to those reachable by the
distillate. (If there were no extractive agent present, the A point would be located
c,,+l
cn.
c,z,
c,,
cn+].
c,,
162
ARTHUR W. WESTERBERG AND OLIVER WAHNSCHAFFT
at pure A. For this case. G,?, and i,,would be on a line passing through the
node for A instead, and we would find ourselves moving toward the node for
B rather than toward the A/E edge as we move up the column. That is incompatible with our intentions for this column.) With solvent fed as in an extractive
column, we move from compositions we can reach from the bottoms product
toward those along the A/E edge. from which we can step to the desired distillate
product just a few trays above the solvent feed.
We moved toward the NE edge because of the relative slopes of the material
balance line passing through the A point and the residue curve passing through
i!,.
Imagine that we have an i,f
that lies much closer to the node for E, where
the relative slopes are the reverse. Here, the column does not function as desired.
The limit point for the desired behavior is where the slope of the line passing
through A and the slope of the residue curve coincide. That is precisely a pirich
point for A. With A at infinity, this occurs where a residue curve has a slope
parallel to the A/E edge. We have plotted a trajectory of pinch points for the A
point. It starts at the node for B and ends at the node for E here. Above this
line. we move from left to right (the desired direction) in the section of the
column between the feeds; below, we move in the other direction.
The S shape of the residue curves gives us another pinch point curve for A.
It starts at the azeotrope and ends at the node for A . Above this curve, movement
is again in the wrong direction. Thus. we have only the region between these
two curves in which we can correctly operate the section of the column between
the two feeds.
What does it mean to have two pinch point curves for A? To operate as we
wish, we need to keep the composition of the liquid on the trays between these
two curves for the section of the column between the feeds. We shall now show
that one of the curves dictates a minimum reflux ratio and the other a maximum
reflux ratio that must be used to operate the column for a fixed solvent ratio.
We shall then show that there is a minimum solvent ratio we can useone
where the pinch point trajectories just touch. We will also show that larger ratios
move the pinch curves apart, making it easier to effect the separation.
2. The Itnpcict of Rejlux Ratio fos Fixed Solvent Ratio
The reflux ratio, R, for a column is the ratio of the liquid we reflux back to
the column at the top stage relative to the distillate product flow; i.e., R =
U D . where L is the liquid flow in the top section of our column. Suppose we
operate our column with a fixed solventtodistillate ratio; i.e., R,v = S/D =
constant. If R, and D are fixed, the location of the A point is fixed because
A
=
D

S = D  RsD = ( 1  Rs)D
With 2, fixed, we can find its pinch point trajectories. as we show in Fig. 65.
When i,,sits exactly on such a trajectory, the compositions for G,,+, and
c,,
SYNTHESIS OF DISTILLATIONBASED SEPARATION SYSTEMS
163
FIG.65. Geometry to explain minimum and maximum reflux ratios for an extractive distillation
column.
coincide because the line through the A point and the residue curve passing
through i,,
have the same slope at such a pinch point. The former points at
while the latter points at ?,,. We show the point i, sitting on the lower
curve and a second case in which
sits precisely on the upper curve.
Let us first examine the point L,, sitting on the lower curve. We remember
that a material balance for the section of the column between the feeds is
=
?,,+I
in+ D

S
=
i,,+ A
where we show the A point in Fig. 65 for the case in which A is positive (Rs
< 1). Using the lever rule, we can write
0
i
L
==_
A
+ qSS  RD + qsRsD
D

S
( 1  Rs)D


R + qsRs
( 1  Rs)
where L is the liquid flow above the solvent feed and qs characterizes the thermal
condition of the solvent (typically the solvent will be subcooled with a qs greater
than unity). Thus the column reflux ratio is given by
I#
ARTHUR W . WESTERBERG A N D OLIVER WAHNSCHAFFT
We can determine R for each pinch point by computing the distances CI and
h. i,,cannot actually sit on the pinch curve as shown because that requires a
column with an infinite number of trays. We must place it between the two
pinch curves at a point where b is smaller, which, by the above, requires R to
decrease. assuming that the distance 0 does not change much with such a move.
We argue. therefore, that the reflux ratio for a real column must be less than
any R we see along this pinch point trajectory. This curve therefore defines a
t m ~ i i i z i i mallowed reflux ratio for operating the column.
From the geometry we see here, the maximum value of R along the pinch
point trajectory is likely where the distance b is a maximum (but it may not be
if N changes as we move along the trajectory).
The upper pinch curve produces the above equations, only this time in terms
of the prime variables we show on Fig. 65. This time we must increase b,
increasing R, to keep the liquid compositions inside the region between the two
pinch curves. We must have R larger than the least value produced by all the
points along this pinch curve.
We see that we have both an upper bound and a lower bound on the column
refiux ratio. Does having an upper bound make intuitive sense? We have set the
solvent flow proportional to the distillate product flow: i.e., S = RsD. As we
increase ihe reflux ratio R, the ratio of solvent feed flow. RsD, to reflux flow,
RD. decreases. This decreases the solvent concentration throughout the column.
thus reducing its impact on the liquid activity coefficients that we are using to
separate A from B. With an infinite reflux ratio, the solvent flow reduces to
x r o . and we have a normal column operating at total reflux which we know
cannot separate A from B.
A similar set of arguments gives us the same results as above when the
solvent rate exceeds the distillate rate, i.e., when the A point is to the lower
right in Fig. 65. Again. the lower curve defines a maximum reflux ratio, and
the upper a minimum.
There are other restrictions on the minimum value for the column reflux
ratio. Both the bottom section (below the bottom feed) and top section above
the solvent feed must be able to reach compositions that join with the trajectory traced by the middle section for a given reflux ratio. These must not
restrict R to be outside the limits we just discovered, or else the column cannot operate.
3. The linpucr qf Solvent Ratio
In the previous section, we fixed the solvent ratio. Here
kary. We noted above that, as the solvent ratio increases,
away from the distillate composition toward infinity, jumps
and then moves toward the solvent composition, always
we shall allow it to
the A point moves
to negative infinity,
along the line that
SYNTHESIS OF DISTILLATIONBASED SEPARATION SYSTEMS
165
passes through these two compositions. As we decrease the solvent ratio, the
two pinch curves shown in Fig. 64 move closer together until they just coincide.
Decreasing the solvent ratio below this value precludes the existence of a path
for the composition trajectory for the column section between the feeds. Thus
the column cannot function. Increasing the solvent ratio moves the two pinch
point curves away from each other, giving an even larger region for the composition trajectory to pass for the section between the feeds.
We argue that the two pinch trajectories coincide at an inflection point for
one (or more) of the residue curves. See Fig. 66. The line marked I in this
figure has the same slope as the residue curve at its inflection point. For this
line, A would be quite close to the pure A node (implying a small amount of
solvent). If we move A away from A, there are two points, one to each side of
the inflection point, which would point at the same A point (see the two lines
marked 2). Only at the inflection point is there a single pinch point for the
corresponding A point.
To find the minimum solvent rate, we draw all lines passing through all
inflection points for the residue curves and project them to the line on which
the A point must exist, i.e., the line passing through the composition points for
the distillate and the solvent. The A point must be no closer to the distillate
composition point than the most distant such intersection point to keep the A
\
\
\
to A
to node
for B
\
\
residue
I
\
to node
for E
FIG. 66. Locating the A point that gives the minimum solvent flow.
166
ARTHUR W WESTERBERG AND OLIVER WAHNSCHAFIT
pinch points for the corresponding residue curve from coinciding. This construction defines the minimum solvent ratio needed for the column.
X. Postanalysis Methods: Column Design Calculations
Column design represents the third major analysis activity in analysisdriven
synthesis. However, our discussion here is short as column design is not the
theme of this paper.
Once one has proposed alternative configurations for systems of separation
devices to effect a desired separation. one must then design these devices so the
various alternatives may be compared. For a distillation column, the first set of
design decisions is to choose the number of trays. the feed tray location, and
the reflux ratio at which to operate it. For a binary separation, the McCabeThiele diagram (or the concepts behind it) is an indispensable aid in making
these decisions.
An approach to setting the reflux ratio often involves computing a minimum
reflux ratio, a topic we have discussed several times in this article with respect
to pinch points. Considerable work has been done on computing minimum reflux
ratios for columns. Koehler (1991) reviews this work. As a rule of thumb, one
sets the reflux ratio to be 20100% or so above the minimum, a number that
experience shows trades off the number of trays with column diameter in a near
costoptimal way.
With the reflux ratio fixed. one can step off the number of trays for a binary
column using a McCabeThiele diagram or. if one wishes to account for heat
effects, a PonchonSavant diagram (Treybal. 1968). establishing the number of
trays needed and where to feed the column.
For more than two species and a reflux ratio set to 1.2 times the minimum.
;I rule of thumb is to compute the total number of trays required for a total
reflux column to produce the separation desired and then double this number as
a first guess (Douglas. 1988). The next decision is select the tray on which to
feed the column. For multispecies columns. the placement is not obvious. Typically, one must search by placing it on any one of a range of trays using a
traybytray simulation. thereby discovering which tray location requires the
least reflux to effect the desired separation.
Using collocation models. which we mentioned earlier (Huss and Westerberg,
1994). one can formulate the design problem as a continuousvariable optimiLation problem with present worth as the objective function. The optimization
problem will select the number of trays needed in both the top and bottom
sections along with the reflux ratioall of which are continuous variables in
wch ;t model.
SYNTHESIS OF DISTILLATIONBASED SEPARATION SYSTEMS
167
Acknowledgments
The National Science Foundation, through its grant to the Engineering Design Research Center,
and Eastman Chemical Co., through its support of the Computer Aided Process Design Consortium
at Carnegie Mellon University, provided support for this work.
References
Andrecovich, M. J., and Westerberg, A. W. “A Simple Synthesis Method Based on Utility Bounding
for Heat Integrated Distillation Sequences,” AIChE J. 31, p. 363 (1985).
Barnicki, S. D. and Fair, J. R. “Separation System Synthesis: A KnowledgeBased Approach. 1.
Liquid Mixture Separations,” Ind. Eng. Chem. Res. 29, 421432 (1990).
Bekiaris, N., Meski, G. A., Radu, C. M., and Morari, M. “Multiple Steady States in Homogeneous
Azeotropic Distillation Columns,” Ind. Eng. Chem. Res. 29, 421432 (1993).
Benedict, M., and Rubin, L. C. “Extractive and Azeotropic Distillation. 1. Theoretical Aspects.”
Tmns. Am. Inst. Chem. Eng., 41, 353370 (1945).
Berg, L. “Selecting the Agent for Distillation,” Chem. Eng. Prog. 65, (9), 5257 (1969).
Berg, L.. and Yeh, A. 1985. “The Unusual Behavior of Extractive DistillationReversing the
Volatility of the AcetoneIsopropyl Ether System,” AIChE J. 31, (3), 504506 (1985).
Bossen, B. S., Jorgensen, S. B., and Gani, R. “Simulation, Design, and Analysis of Azeotropic
Distillation Operations,” Ind. Eng. Chem. Res., 32, 620633 (1993).
Cho, Y. S., and Joseph, B. “ReducedOrder SteadyState and Dynamic Models for Separation
Processes,” AIChE J, 29, 261269. 270276 (1983).
Doherty, M. F. “The Presynthesis Problem for Homogeneous Azeotropic Distillations has a Unique
Explicit Solution,” Chem. Eng. Sci. 40, 18851889 (1985).
Doherty, M. F., and Caldarola, G. A. “Design and Synthesis of Homogeneous Azeotropic Distillations. 3. The Sequencing of Columns for Azeotropic and Extractive Distillations,” Znd. Eng.
Chem. Fundam. 24,474485 (1985).
Doherty, M. F., and Perkins, J. D. “On the Dynamics of Distillation Processes. I. The Simple
Distillation of Multicomponent NonReacting Homogeneous Liquid Mixtures,” Chem. Eng.
Sci. 33, 281301 (1978).
Doherty, M. F., and Perkins, J.D. “On the Dynamics of Distillation Processes. III. The Topological
Structure of Ternary Residue Curve Maps,” Chem. Eng. Sci. 34, 14011414 (1979).
Douglas, J. M. “Conceptual Design of Chemical Processes.” McGrawHill, New York, 1988.
Ewell, R. H., and Welch, L. M. “Rectification in Ternary Systems Containing Binary Azeotropes,”
Ind. Eng. Chem. 37, 12241231 (1945).
Fidkowski, Z. T., Malone, M. F., and Doherty, M. F. “Computing Azeotropes in Multicomponent
Mixtures,” Comput. Chem. Eng. 17(12), 11411 155 (1993).
Fien, G. A. F., and Liu, Y. A. “Heuristic Synthesis and Shortcut Design of Separation Processes
Using Residue Curve Maps: A Review,” Ind. Eng. Chem. Res. 33, 25052522 (1994).
Foucher, E. R., Doherty, M. F., and Malone, M. F. “Automatic Screening of Entrainers for Homogeneous Azeotropic Distillation,” Ind. Eng. Chem. Res. 30, 760772 (1991).
Hendry, J. E., Rudd, D. F., and Seader, J. D. “Synthesis in the Design of Chemical Processes,”
AIChE J. 19(1), 115 (1973).
Henley, E. J., and Seader, J. D. “EquilibriumState Separation Operations in Chemical Engineering,” Wiley, New York, 1981.
I68
ARTHUR W. WESTERBERG AND OLIVER WAHNSCHAFFT
Hlav;icek. V. "S>ntheais in the Design of Chemical Processe\." Compirt. Cherfi. E q . 2, 6775
(197x1.
Hoffman E. J . ".4rentropic and El;tracti\e Distillation." Wiley (Interscience). New York, 1964.
Holland. C. D. "Fundamentals of Multicomponent Distillation." McGrawHill. New York. 1981.
Horsley. L. H. "A~eotropic Data 111." Adv. Chem. Ser. No. 116. American Chemical Society,
Wa\hington. DC. 1973.
Huneh. J.. Gal. S.. Posel. F.. and Gavic. P. "Separation of an Areotropic Mixture by Reverse
Eztractive Di\tillation." AlChE J . 35(7). 12071210 (1989).
Hu\s. R. S.. and Westerberg, A . W. "Collocation Methods for Distillation Design." Paper 1 3 1 ~ .
A n n u a l AIChE Meeting. San Fran
Julka. V.. :ind I h h e r t ~. M .F. "Geometric Behavior and Miniinurn Flows for Nonideal Multicompnnent Distillation." U w n . E q . Sci. 45, 18011 X22 (1990).
King. C J . "Separation Processes." 2nd ed. McGraaHill, New York, 1980.
Knight. J . R.. and Doherty. M. F. "Optimal Design and Synthesis of Homogeneous Aleotropic
Di\tillation Sequence\." h i d . Errs. Cl7rm. Rrs. 28, 565572 ( 1989).
Knehler. J. W.."Strukttirsynthese und minimaler Energiebedarf Nichtidealer Bektifikationen."
Ph.D. Di\certation. Technical University of Munich (1991).
L;michc.. l... Andenen. H. W.. and Morari. M. "Homogeneous Azeotropic Distillation: Comparing
Entrctiners." C m . J . C/wni. En,?. 69, 13011 3 19 ( 1991).
Levy. S. G.. Van Dongen. D. 5.. and Doherty. M. F. "Design and Synthesis of Homogeneous
Azeotropic Distillations. 2. Minimum Reflux Calculations for Nonideal and Azeotropic Column\," Irid €fig. C/iem. Firtidtrm. 24, 4 6 3 1 7 3 ( 1985).
bla~su);ma. H. "Synthesi5 of Azeotropic Distillation Systems.'' Paper presented at the JapanU.S.
Joint Seminar. Kyoto. Japan (1975).
McCabe. W ,L... and Smith. J. C. "Unit Operations of Chcmical Engineering." 3rd ed. McGrawHill. Ne\r Yorh. 1976.
McDonald. C.M.. and Floudas. C. A. "Global Solutions for the Phase and Chemical Equilibrium
Problem." Paper 2 2 0 ~ AIChE
.
Annual Meeting. San Francisco (1994).
%4ichelren,M. L. .The Isothermal Flash Problem. Part 1: Stability." Fluid Pliase Eqirilib. 9, 119
(10x2).
Michelwn. kl. L. "Phabe Equilibrium Calculations. What is Easy and What is Difficult." Corfiput.
L'ircwr. En,y. 17t.716). 1 3 1 4 3 9 ( 1993).
klichelwi. M. L. .'Calculation of Mukiphahe Equilibrium." Conipirt. Cheni. Etig. 18(7). 545550
(1901,.
klodi. A. K. and \\'esterberg. A. W . "Di\tillation Colitnm Sequencing Using Marginal Price," Irrtl.
L'rig. Chc.rn. Hrs. 31, X39848 ( 1999).
Nikulue\. N. S.. Kiva. \'. N..Mnzzhukhin. A. S.. Seratinio\*.L..A,. and Golohorodkin, S. 1. "Utiliiation of Functional Operators for Determining the Regions of Continuous Rectification."
Tfrfwr.poi/f/d Chrrff. Eflg. (Etfg/. Tftirf.V/.J 13, 4 1 8 1 2 3 ( 1979).
Ni\hida, N., Stephanopuulo\. G . and Westerberg. .4. W . "A Review o f Process Synthes
./. 27. 371 ( 1981 ).
Perr?. J. H. "Chemical Engineers' Handbook." 3rd ell. McGmwHill. New York, 1950.
Petl>uh. I. €3. "Rectification of Zcotropic. Areotropic. and Continuous Mixture\ in Simple and
CompleA Intinite Columns with Finite Reflux." 7lwor. Fortnti. C'Iim. Errg. (EngI. T r t u d ) 12,
h71678 (1978).
Petlyuh. F. B.. Seralimo\. L. A,. A\et'yan. V. S.. and Vinogradova. E. I . "Trajectories of Reversible
Rectitication when One of the Components Completely Disappears in Each Section." Tlreor.
f'fIir/fd. Chrtfl. Elf.q. rfll,y/. Tr~frf\/.
J 15. 185192 (IC)811.
Ph:iiii. H. 5 . .and Dohert). M. F. "Design and S)nthcsi\ of Heterogeneous Areotropic Distillations.
Ill.C h m n Sequences." Chrnr. Erin. P i . 45. 18311 854 (1990).
SYNTHESIS OF DISTILLATIONBASED SEPARATION SYSTEMS
169
Pham. H. N., Ryan. P. J., and Doherty, M. F. “Design and Minimum Reflux for Heterogeneous
Azeotropic Distillation Columns,” AIChE J. 35,(10), 15851591 (1989).
Poellmann. P., and Blass, E. “Best Products of Homogeneous Azeotropic Distillations,” Gas &pan
Pur$ 8(4), 194228 (1994).
Rathore, R. N. S., Vanwormer, K. A,, and Powers, G. J. “Synthesis Strategies for Multicomponent
Separation Systems with Energy Integration,” AIChE J. 20,491 (1974).
Reid, R. C., Prausnitz, J. M., and Poling, B. E. “The Properties of Gases and Liquids,” 4th ed.
McGrawHill, New York, 1987.
Ryan, P. J., and Doherty, M. F. “DesigdOptimization of Ternary Heterogeneous Azeotropic Distillation Sequences,’’ AIChE J. 35,(10), 15921601 (1989).
Sargent, R. W. S. H. “A Functional Approach to Process Synthesis and its Application to Distillation
Systems.” Tech. Rep. Centre for Process Systems Engineering, Imperial College, London, 1994.
Sargent, R. W. S. H., and Gaminibandara, K. “Optimum Design of Plate Distillation Columns.”
in “Optimization in Action” (L. W. C. Dixon, ed.), 267314. Academic Press, London,
1976.
Seferlis, P., and Hrymak, A. N. Optimization of Distillation Units Using Collocation Models. AIChE
J.. 40, 813825 (1994).
Serafimov, L. A. “Thermodynamic Topological Analysis and the Separation of Multicomponent
Polyazeotropic Mixtures,’’ Theor. Found. Chem. Eng. (Engl. Trans/.J 21,4454 (1987) (translated from Teor. Om. Khim. Tekhnol. 21(1), 7485, (1987)).
Smith. J. M., and Van Ness, H. C. “Introduction to Chemical Engineering Thermodynamics,” 4th
ed. McGrawHill. New York, 1987.
Stewart, W. E.. Levien, K. L.. and Morari, M. “Collocation Methods in Distillation,” in “Foundations ComputerAided Process Design (FOCAPD’83)” (A. W. Westerberg and H. H. Chien.
eds.). pp. 539569. Cache Corp., Ann Arbor, MI, 1984.
Stichlmair, J., Fair, J. R., and Bravo, J. L. “Separation of Azeotropic Mixtures via Enhanced Distillation,” Chern. Eng. Prog. 85,(1). 6369 (1989).
Tanaka, S.. and Yamada, J. “Graphical Solution of Operating Region in Extractive Distillation,”
Kagaku Kogaku (Abr. E. Engl.) 3,(l), 4043, (1965).
Terranova. B. E., and Westerberg, A. W. “TemperatureHeat Diagrams for Complex Columns, I .
Intercooled/Interheated Distillation Columns,” Ind. Eng. Chern. Res. 28, 13741379 (1989).
Thompson, R. W., and King, C. J. “Systematic Synthesis of Separation Schemes,” AICI7E J. 18,
941948 (1972).
Treybal, R.E. MassTransfer Operations, 2nd ed. p. 297. McGrawHill, New York, 1968.
Underwood, A. J. V. “Fractional Distillation of Multicomponent MixturesCalculation of Minimum Reflux Ratio,” J. fnsr. Petrol. 32, 614 (1946).
Van Dongen, D. B. “Distillation of Azeotropic Mixtures. The Application of SimpleDistillation
Theory to the Design of Continuous Processes,” Ph.D. Dissertation, University of Massachusetts, Amherst ( 1 983).
Van Dongen, D. B., and Doherty, M. F. “Design and Synthesis of Homogeneous Azeotropic Distillations. 1. Problem Formulation for a Single Column,” Ind. Eng. Clzem. Fundam. 24, 454463 (1985).
Wahnschafft, 0. M. “Synthesis of Separation Systems for Azeotropic Mixtures with an Emphasis
on DistillationBased Methods,” Ph.D. Dissertation, University of Munich. Munich, Germany
(1992).
Wahnschafft. 0. M. “A Simple and Robust Continuation Method for Determining All Azeotropes
Predicted by a Multicomponent VaporLiquid Equilibrium Model,” in preparation (1994).
Wahnschafft, 0. M., and Westerberg, A. W. “The Product Composition Regions Azeotropic Distillation Columns. 2. Separability in TwoFeed Columns and Entrainer Selection.” Ind. Eng.
Chem. Res. 32, 11081 120 (1993).
I70
ARTHUR W. WESTERBERG AND OLIVER WAHNSCHAFFT
Wahnschafft, 0. M.. Koehler, J.. Blasa. E.. and Westerberg, A. W . “The Product Composition
Region5 of Single Feed Areotropic Distillation Columns,“ tnd. Eng. Chrm. Rrs. 21, 13452362 (1992).
Waaylkieaic7, S.. Sridher, L.. Malone, M. F., and Doherty, M. F. “Gibbs Tangent Plane Anaylsis
for Complex Liquid Mixtures: A Global Algorithm.” Paper IS I a, Annual Meeting AIChE.
St. Louis. MO (1993).
Westerberg. A. W. “A Review of Process Synthesis. in Computer Applications to Chemical Engineering Process Design and Simulation,“ ACS S y n p Ser. 124, 5487 (1980).
Westerberg. A. W. “The Synthesis of DistillationBased Separation Systems,” Compur. Cheni. En,?.
9,(S), 421479 (1985).
Zharov. W. T., and Serafimov, L. A. “Physicochemical Fundamentals of Distillations and Rectitications (in Russian).” Khimiya. Leningrad, 1975.