Article

pubs.acs.org/IECR

Globally Optimal Networks for Multipressure Distillation of

Homogeneous Azeotropic Mixtures
Paul G. Ghougassian and Vasilios Manousiouthakis*
Department of Chemical & Biomolecular Engineering Department, University of California, Los Angeles, California 90095,
United States

ABSTRACT: In this article, a methodology for the globally optimal synthesis of a network of vapor−liquid equilibrium
flash separators that can operate at multiple pressures and separate an azeotropic mixture is presented. The objective
function minimized is the total flow entering the network flashes. The proposed synthesis methodology employs the
infinite-dimensional state-space (IDEAS) conceptual framework, which is shown to be applicable to the problem under
consideration. The resulting infinite linear programming (ILP) IDEAS formulation is shown to have several properties that
allow its simplification. The approximate solution of this IDEAS ILP is pursued through the solution of a number of finite-
dimensional linear programs (FLPs) of ever increasing size, whose optimum values form a sequence that converges to the
ILP’s infimum. The proposed optimal design methodology is general in nature and can be used to separate any number of
pressure-sensitive azeotropic mixtures, with or without use of an entrainer. The method is demonstrated on a first case
study involving the dual-pressure separation of a methyl acetate/methanol binary mixture, which exhibits a minimum-
boiling azeotrope, without using an entrainer, and a second case study involving the dual-pressure separation of a ternary
mixture of water, methanol, and acetone that also exhibits a minimum-boiling azeotrope for the methanol/acetone binary
mixture, again without using an entrainer. The IDEAS-generated globally optimal design is shown to be 31.54% better than
an optimized, dual-pressure, traditional, two-column design for the binary mixture (case 1) and 15.15% better for the
ternary mixture (case 2).

1. INTRODUCTION trial-and-error simulations at specific design parame-

Knowledge regarding the sensitivity of azeotropes to
pressure effects dates back to the 19th century. 1−3 Azeo-
tropic mixtures are typically separated by either homoge-
neous or heterogeneous azeotropic distillation, in batch 4
or continuous 5 configuration. The separation process
requires the addition of an entrainer and/or the use of pres-
sure swing distillation (PSD), which exploits the depend-
ence of the azeotrope on pressure.5 The main advantage
of the PSD process is that it does not require an entrainer,
that is, a substance that facilitates the considered separa-
tion but requires additional separation steps downstream
and might itself be a toxic chemical or degrade into by-
products harmful to the environment. The systematic gen-
eration process synthesis paradigm has been predicted to
expand in the foreseeable future, 6 with the separation of an
azeotropic mixture listed as an important problem to
address in the 21st century. 7 A large list of pressure-
sensitive binary azeotropes prevalent in the chemical industry that
can be separated through the PSD process has been
tabulated.8 The PSD process involves the operation of
multiple columns at different pressures to bypass pressure-
dependent azeotropic pinch points and recover high-purity
products. A two-column PSD is an especially interesting
process, because its two columns operate at two dif-
ferent pressures and can be readily thermally integrated, by
matching the heat removed by the high-pressure con-
denser with the heat required by the low-pressure boiler
or with the heat required by the feed preheater.9,10 Several
algorithms have been developed to optimize azeotropic
columns based on energy and cost considerations, using
ters.5,11−16 However, these techniques do not guarantee
global optimality and consider only a limited number of
alternatives. Geometric approaches to identify and optimize
azeotropic separation processes through homotopy, arc
length continuation, residue curve maps, distillation boun-
daries, shortcut methods, and distillation lines have also been
proposed, 1 7 − 2 2 but they have visualization limita-
tions and also do not guarantee global optimality. Unusual
and nontraditional variant designs have also emerged.23,24
Recently, researchers have begun using more refined
optimization techniques relying on the formulation of
complex mixed integer nonlinear programs (MINLPs).25−33
Although the MINLP methodology provides improvements,
solutions are locally optimal and only as good as the initially
supplied superstructure, because of the nonlinearity and
nonconvexity that naturally arises from the superstructure-
based MINLP problem formulation. In addition, commer-
cially available globally optimal nonlinear programming
software cannot handle the large number of variables result-
ing from the complexity of distillation column (super)-
structures.
This article demonstrates the use of the infinite-dimensional
state-space (IDEAS) framework in simultaneously synthe-
sizing a network that can separate azeotropic mixtures
Received: February 16, 2012
Revised: June 16, 2012
Accepted: July 7, 2012
Published: July 8, 2012

© 2012 American Chemical Society 11183 dx.doi.org/10.1021/ie300423q | Ind. Eng. Chem. Res. 2012, 51, 11183−11200
Industrial & Engineering Chemistry Research
and minimizing the total flash inlet flow in the aforemen-
tioned network. The IDEAS framework is a generalized
methodology that allows for the generation of globally
optimal flow sheet designs. The IDEAS framework
decomposes a process network into an operator network
(OP), where the unit operations (reactors, distillation
columns, heat exchangers, etc.) occur, and a distribution
network (DN), where the flow operations (mixing, splitting,
recycling, and bypass) occur. The optimal process network
structure is identified through solution of an infinite linear
program (ILP) that is formulated within the IDEAS
framework. The ILP’s solution is approximated by finite-
dimensional linear programs of ever-increasing size. The
solution of these linear programs is guaranteed to be globally
optimal. IDEAS has been successfully applied to numerous
globally optimal process network synthesis problems, such as
mass-exchange network synthesis, 34 complex distillation
network synthesis,35−37 power cycle synthesis,38 reactor
network synthesis,39,40 reactive distillation network syn-
thesis,41 separation network synthesis,42 attainable region
construction,43−46 and batch attainable region construc-
tion.47
In this work, IDEAS is shown to be applicable to the
globally optimal synthesis of networks that employ steady-
state vapor−liquid equilibrium flash separators, operating at
multiple pressures, to separate azeotropic mixtures. The
representation of a theoretical tray of a distillation column
as a combination of two separate devices, namely, a mixer
that can receive vapor and liquid inlets at different
temperatures and an equilibrium (flash) separator that has
as its inlet the mixer’s outlet and as outlets a vapor stream
and a liquid stream that are in equilibrium with one
another, has been suggested. 48 A schematic representation
of this concept is shown in Figure 1. Because the possibility
Figure 1. Mixer + flash separator representation of a distillation
column tray.48
of mixing is incorporated within the DN of the IDEAS
framework and the equilibrium separator can be made
part of the OP of IDEAS, a distillation network can be
represented within the IDEAS framework as a network of
vapor−liquid equilibrium flash separators. Thus, IDEAS can
employ this representation to synthesize distillation net-
works.
The rest of the article is structured as follows: First, the
applicability of IDEAS to the multipressure flash separator
network synthesis problem is established, and the resulting
IDEAS mathematical formulation is presented. Properties
that facilitate the solution of this mathematical formulation
are then presented. This is followed by two case studies in
which the IDEAS framework is used to generate an optimal
network for the separation of a binary mixture (case study 1)
and a ternary mixture (case study 2). The obtained results
are then discussed, and conclusions are drawn.
11184
Article
2. APPLICABILITY AND MATHEMATICAL
FORMULATION OF IDEAS FOR MULTIPRESSURE
FLASH SEPARATOR NETWORK SYNTHESIS
In this work, the isothermal, isobaric flash separator shown
in Figure 2 is considered. The flash separator’s vapor and
Figure 2. Representation of a flash separator.
liquid exit streams are considered to be in phase equilibrium
with one another. Because the main goal of the synthe-
sis task to be undertaken is to establish the feasibility of
separation of an azeotropic mixture using multipressure dis-
tillation and to identify a reasonably sized distillation
network design that can carry out the aforementioned
separation, only mass/component balance and phase equi-
librium relations are incorporated in the flash sepa-
rator’s mathematical model, and the considered optimi-
zation goal is the minimization of the total flow entering
the network’s flashes. This objective function can be
thought of to be directly minimizing the network capital
cost (flash separator total volume) by considering that
each flash separator has the same residence time. It can also
be thought of to indirectly minimize energy consump-
tion, because distillation network energy costs are typically
related to the network’s flows. As long as no constraints are
imposed on the network’s energy consumption, energy
balances need not be incorporated in the flash separator and
distribution network models for the minimum total network
flow to be correctly identified over all feasible networks.
Indeed, by omitting all energy balances, the optimization
problem has a larger feasible region than if the energy
balances were included. Thus, the obtained minimum is less
than or equal to the minimum obtained with the energy
balances included. These two minima are actually equal,
because energy balances can always be written a posteriori
for the network minimizing total flow to identify the net-
work’s heating and cooling needs. The lack of any
specifications on these needs suggests that the network
would be feasible even if energy balances were considered.
On the other hand, the synthesis of an optimally heat-
integrated multipressure distillation network would require
that energy consumption specifications be imposed on the
optimization problem, thus necessitating the use of energy
balances. This latter problem will be the subject of future
research efforts.
It is true that distillation is an inherently nonlinear
process with nonlinear constraints. The beauty of IDEAS
is exactly that it overcomes the inherently nonlinear nature
of the distillation process optimization problem, not
through some kind of Taylor series approximate lineariza-
tion, but rather by exploiting the decomposition and linea-
rity properties that naturally exist in the distillation process
model, based on knowledge of the intensive properties
and design parameters and independently of the extensive
properties.
dx.doi.org/10.1021/ie300423q | Ind. Eng. Chem. Res. 2012, 51, 11183−11200
Industrial & Engineering Chemistry Research Article

The considered equilibrium separator model employs the Ψ1: Rn + 2 → R2,

Gamma−Phi vapor−liquid equilibrium formulation49 and
mass and component balances ⎡P⎤
⎢ ⎥
⎢T ⎥
fkP − xkLF L − ykV F V = 0 ∀ k = 1, ..., n (1) ⎢ L⎥
Ψ1: u1 = ⎢ x1 ⎥ → Ψ1(u1)
n
⎢⋮⎥
⎢ L⎥
∑ xkL = 1 ⎢⎣ xn ⎥⎦
k=1 (2)
⎡ n ⎤
⎢ ∑ xkL − 1 ⎥
n
⎢ k=1 ⎥
∑ ykV =1 ≙⎢ ⎥
k=1 (3) ⎢ n xkLγk({xlL}ln= 1, T ) Pksat(T ) ⎥
⎢∑ − 1⎥
⎣ k=1 P ⎦
ykV ϕk ({ylV }ln= 1, T , P)P − xkLγk({xlL}ln= 1, T ) Pksat(T ) = 0
∀ k = 1, ..., n (4) Ψ2 : Rn + 2 × Rn + 2 → Rn,
⎡ u1 ⎤ ⎛⎡ u1 ⎤⎞
A variety of thermodynamic models can be employed in Ψ2 : ⎢ ⎥ → Ψ2⎜⎢ ⎥⎟
quantifying the functions ϕk({yVl }nl=1,T,P) and γk({xLl }nl=1,T),Psat ⎣u2 ⎦ ⎝⎣ u 2 ⎦⎠
k ,
∀k = 1, ..., n. In the illustrative case studies outlined below, ideal ⎡ x Lγ ({x L}n , T ) P1sat(T ) V ⎤
gas behavior is assumed [ϕk({yVl }nl=1,T,P) = 1, ∀k = 1, ..., n]; the ⎢ f P − x1LF L − 1 1 l l = 1 F ⎥
Wilson equations (eqs 5 and 6) are used to model the nonideal ⎢ 1 P ⎥
liquid-phase coefficients γk({xLl }nl=1,T), ∀k = 1, ..., n, and the ≙ ⎢⎢ ⋮ ⎥
⎥
extended Antoine equation (eq 7) is used to model the vapor
⎢ L L n sat ⎥
k (T), ∀k = 1, ..., n
pressure Psat x γ ({x } , T ) P ( T )
⎢ f P − xnLF L − n n l l = 1 n
F V⎥
⎣ n P ⎦
⎡ n ⎤
ln γk({xl }l = 1, T ) = 1 − ln ∑ xj Λk , j(T )⎥
⎣
⎡ L n
⎦
⎤ ⎢ L
y = [ y1T | y2 T ]T
⎢⎣ j = 1 ⎥⎦
n ⎛ x L Λ (T ) ⎞ = [T y1V ··· ynV P x1L ··· xnL | f1 f2 ··· fn F L F V ]T
− ∑ ⎜⎜ n ⎟
i i,k
∀ k = 1, ..., n
i=1 ⎝ ∑ x L
Λ (T ) ⎟⎠
j=1 j i,j (5) Φ1: Rn + 2 × Rn + 2 → R2n + 2,
Φ1: (u1 , u 2) → y1 = Φ1(u1 , u 2)
⎛ − Ak , j ⎞
V jL ⎡ T ⎤
Λk , j(T ) = L exp⎜ ⎟ ∀ k = 1, ..., n; ∀ j = 1, ..., n ⎢ L ⎥
Vk ⎝ RT ⎠ L n sat
⎢ x1 γ1({xl }l = 1, T ) P1 (T ) ⎥
(6) ⎢ P ⎥
⎢ ⎥
Bk ⎢ ⋮ ⎥
ln[Pksat(T )] = Ak + + Dk ln(T ) + Ek T Fk ⎢ L L n sat ⎥
T + Ck ⎢ xn γn({xl }l = 1, T ) Pn (T ) ⎥
=⎢ ⎥
∀ k = 1, ..., n (7) P
⎢ ⎥
⎢ P ⎥
The equations above lead to the conclusion that the ⎢ L ⎥
aforementioned flash separator model can be employed to ⎢ x1 ⎥
construct the following input−output information map ⎢ ⋮ ⎥
⎢ ⎥
Φ: Rn + 2 × Rn + 2 → R2n + 2 × Rn + 2, ⎢⎣ xnL ⎥⎦

Φ: u → y = Φ(u) = [[Φ1(u1 , u 2)]T | [Φ2(u1 , u 2)]T ]T

Φ2 T : Rn + 2 × Rn + 2 → Rn + 2,
where Φ2 T : [ u1T u 2 T ]T → y2 T = Φ2 T([ u1T u 2 T ]T )
u = [ u1T | u 2 T ]T = [ f1 ··· fn F L F V ]T
= [ P T x1L ··· xnL | f1 f2 ··· fn F L F V ] The engineering importance of the above information maps
can best be understood as follows: Consider that u1 =
[ P T x1L ··· xnL ]T such that Ψ1(u1) = 0 ∧ u1 ≥ 0 is
u ∈ D = {u = (u1 , u 2) ∈ Rn + 2 × Rn + 2: Ψ1(u1)
known. This can be ascertained by first considering
= 0 ∧ Ψ2(u1 , u 2) = 0 ∧ u1 ≥ 0 ∧ u 2 ≥ 0} [ P T x1L ··· xnL− 2 ]T ≥ [0 0 0 ··· 0]T to be known and
11185 dx.doi.org/10.1021/ie300423q | Ind. Eng. Chem. Res. 2012, 51, 11183−11200
Industrial & Engineering Chemistry Research Article

then solving for all q physically meaningful solutions u1,Ψ3(u1) is a linear operator and thus Ψ2(u1,u2) = Ψ3(u1)u2 is
[ xnL− 1 xnL ]T ≥ [0 0]T of the system of equations linear in u2.
At this point, it should be noted for future reference that
n n
xkLγk({xlL}ln= 1, T )Pksat(T ) the incorporation of energy balances in the problem
∑ xkL − 1 = 0, ∑ −1=0 formulation does not change these characteristics of the
P
k=1 k=1
flash separator model. Indeed, given a predictive thermody-
q can be either zero or a positive integer. For each of the q namic model that accounts not only for phase equilibrium
solutions, resulting physically meaningful vectors u1 = but also for molar enthalpy, calculation of all of the vapor/
[ P T x1L ··· xnL ]T{yVk }nk=1 can then be computed through liquid product mole fractions and molar enthalpies requires
only knowledge of the flash P, T, and {xLl } n−2 l=1 and does
eq 4. In turn, this suggests that, for any u1 such that Ψ1(u1) = 0
∧ u1 ≥ 0, one can actually evaluate the image Φ1(u). not require any knowledge of either the feed component
Having defined the maps Ψ1, Ψ2, Φ1, and Φ2, the following flows or the feed enthalpy flows or the total product molar
properties can be readily verified. flows.
IDEAS property 1 is expressed as ∃Φ3: Rn+2 → R2n+2 such It has been established that, for a fixed vector u1, the
that Φ1(u1,u2) = Φ3(u1), ∀(u1,u2) ∈ D. This map is employed flash separator model is defined by an input−
output map Φ3(u1) whose domain is defined as the set u1 =
⎡ T ⎤ [ P T x1L ··· xnL ]T ∈ Rn+2: Ψ 1(u1) = 0 and by an identity
⎢ L ⎥
L n sat
⎢ x1 γ1({xl }l = 1, T ) P1 (T ) ⎥ operator Φ4(u1) whose domain is the null space of Ψ3(u1),
⎢ P ⎥ which, for fixed u1, is a linear operator. To incorporate
⎡T ⎤ ⎢ ⎥ the variation of u1 in the optimization process, so as to
⎢ ⎥ ⎢ ⋮ ⎥ avoid suboptimal solutions, an infinite number of flash
⎢P⎥ ⎢ L ⎥
L n sat
⎢ xn γn({xl }l = 1, T ) Pn (T ) ⎥ separator units each with a fixed value of u1 is considered
⎢x L⎥
Φ3: u1 = ⎢ 1 ⎥ → Φ3(u1) ≙ ⎢ ⎥ such that the union of the considered u1 values is dense
P
⎢⋮⎥ ⎢ ⎥ in the set over which u 1 can vary. A one-to-one
⎢ L⎥ ⎢ P ⎥ correspondence is then created between the infinite
⎢⎣ xn ⎥⎦ ⎢ L ⎥ sequence {u1(i)}∞ i=1 consisting of all possible values of u1
⎢ x1 ⎥ and the infinite sequence {Ψ3[u1(i)]}∞ i=1 of corresponding
⎢ ⋮ ⎥
⎢ ⎥ linear maps used to define the domain of the identity
⎢⎣ ⎥⎦ operator for each flash separator unit. These sequences are
xnL
then used to define the domain and action of a linear
∀ u1 ∈ {u1 ∈ Rn + 2: Ψ1(u1) = 0 ∧ u1 ≥ 0} operator (termed IDEAS OP) that quantifies the effect of all
flash separator units and has its domain and range be
This implies that y1 ≙ Φ1(u1,u2) = Φ3(u1) can be evaluated subsets of infinite-dimensional spaces. This relation gives
based only on knowledge of u1 (intensive properties and rise to the linear OP constraints, which hold true around
design parameters) and independently of u2 (extensive every OP unit.
properties). To account for all possible multipressure distillation flow-
IDEAS property 2 is expressed as ∃Φ4: Rn+2 → R(n+2)×(n+2) sheets, the (OP) operator needs to be interfaced with a
such that Φ2(u1,u2) = Φ4(u1)u2 ∀(u1,u2) ∈ D. Indeed, this distribution network (DN) where all stream splitting and
map is Φ4: u1 → Φ4(u1) ≙ I ∈ R(n+2)×(n+2), and thus, mixing and pressure adjustment occurs, as shown in Figure 3.
y2 ≙ [ f1 f2 ··· fn F L F V ]Φ4: u1 → Φ4(u1) ≙ I ∈ Each of the cross-flow streams in the DN is characterized by
R(n+2)×(n+2), and thus, y2 ≙ [ f1 f2 ··· fn F L F V ]T = Iu2 =
u2. This implies that, for fixed u1, Φ4(u1) is a linear operator
(the identity operator) and y2 ≙ Φ2(u1,u2) = Φ4(u1)u2 is linear
in u2.
IDEAS property 3 is expressed as ∃Ψ3: Rn+2 → Rn×(n+2) such
that Ψ2(u1,u2) = Ψ3(u1)u2 ∀(u1,u2) ∈ D. Indeed, this map is

Ψ3: u1 → Ψ3(u1)
⎡ x1Lγ1({xlL}ln= 1, T ) P1sat(T ) ⎤
⎢ 1 0 ··· 0 −x1L − ⎥
⎢ P ⎥
⎢ L L n sat ⎥
⎢ x 2 γ2({xl }l = 1, T ) P2 (T ) ⎥
⋮ 1 ··· 0 −x 2L −
≙⎢ P ⎥
⎢ ⎥
⎢0 ⋮ ⋱ 0 ⋮ ⋮ ⎥
⎢ L L n sat ⎥
⎢ xn γ ({xl }l = 1, T ) Pn (T ) ⎥
⎢⎣ 0 0 ··· 1 −xnL − n ⎥⎦
P
This implies that Ψ3(u1) can be evaluated based only on Figure 3. IDEAS representation of a (multipressure) flash separator
knowledge of u1 and independently of u2 and that, for fixed network.

11186 dx.doi.org/10.1021/ie300423q | Ind. Eng. Chem. Res. 2012, 51, 11183−11200

Industrial & Engineering Chemistry Research Article

a flow rate variable and fixed destination and origin M ∞

conditions. This information takes the form of the following [zkO(i)]l F O(i) ≤ ∑ zkI(j) F OI(i , j) + ∑ xkL(j) F OL(i , j)
sequence triplets: DN inlet to DN outlet (FOI,zO,zI), DN j=1 j=1
inlet to OP inlet (FPI,zP,zI), OP liquid outlet to DN outlet ∞
(FOL,zO,xL), OP vapor outlet to DN outlet (FOV,zO,yV), OP + ∑ ykV (j) F OV(i , j) ≤ [zkO(i)]u F O(i)
liquid outlet to OP inlet (FPL,zP,xL), and OP vapor outlet to j=1
OP inlet (FPV,zP,yV).
∀ i = 1, ..., N ; ∀ k = 1, ..., n (12)
A linear objective is considered in the proposed IDEAS
formulation. It can be generally presented as
∞ M ∞ ∞ fkP (i) − xkL(i) F L(i) − ykV (i) F V(i) = 0
∑ ∑ αi ,jF PI(i , j) + ∑ ∑ βi ,jF PL(i , j) ∀ i = 1, ..., ∞ ; ∀ k = 1, ..., n (13)
i=1 j=1 i=1 j=1
∞ ∞
M
+ ∑ ∑ γi ,jF PV(i , j) fkP (i) − F P(i) zkP(i) = 0 ⇒ fkP (i) − ∑ zkI(j) F PI(i , j)
i=1 j=1
j=1
∞ ∞
This objective function can be used to realize a wide array
of objectives, through appropriate selection of the cost − ∑ xkL(j) F PL(i , j) − ∑ ykV (j) F PV(i , j) = 0
coefficients αi,j, βi,j, and γ i,j, associated with each of the j=1 j=1

problem’s flow variables. In the case study below, the ∀ i = 1, ..., ∞ ; ∀ k = 1, ..., n (14)
minimization of total flash volume is considered. Under
the assumption that each flash has the same residence time
as any other flash (i.e., assuming that the same time is F I ≥ 0, F O ≥ 0, F P ≥ 0, F L ≥ 0, F V ≥ 0,
needed to reach equilibrium), the common residence F OI ≥ 0, F PI ≥ 0, F OL ≥ 0, F OV ≥ 0,
time shared by all flashes can be factored out. This yields
αi,j = βi,j = γ i,j = 1, which then yields the following objective F PL ≥ 0, F PV ≥ 0, fkP ≥ 0
function
Constraints 8 and 9 represent the DN inlet (splitting)
∞ M ∞ ∞ ∞ ∞
and outlet (mixing) total flow balances, respectively,
∑ ∑ F PI(i , j) + ∑ ∑ F PL(i , j) + ∑ ∑ F PV(i , j) graphically appearing on the left and top of the DN block
i=1 j=1 i=1 j=1 i=1 j=1 in Figure 3. Constraints 10 and 11 correspond to the OP
total liquid and vapor outlet flow balances, respectively,
which represents the total flow entering the network’s
which undergo splitting operations in the DN and
flashes (total network unit inlet flow). The resulting
graphically appear on the bottom of the DN. Constraint
mathematical formulation of IDEAS (IF1) is
12 correspond to the bounds on the quality variables of the
∞ M ∞ ∞ DN outlet and is represented by component balances,
ν ≙ inf ∑ ∑ F PI(i , j) + ∑ ∑ F PL(i , j) which undergo mixing operations in the DN and graphically
i=1 j=1 i=1 j=1 appear on the top of the DN. Constraint 13 represents
∞ ∞ the separator model, and finally, constraint 14 corresponds
+ ∑ ∑ F PV(i , j) to the OP inlet component flow balances, which undergo
i=1 j=1 mixing operations in the DN and graphically appear on the
right of the DN. In addition, all flow variables are
subject to nonnegative.
N ∞ Close examination of this formulation reveals that it gives rise
F I(j) − ∑ F OI(i , j) + ∑ F PI(i , j) = 0 ∀ j = 1, ..., M to an infinite linear program (ILP). The aforementioned line-
i=1 i=1
arity of the infinite program ν stems from the linearity of the
(8) IDEAS OP established earlier, and the linearity of the con-
straints that quantify the IDEAS DN:
M ∞ ∞
(1) The DN total flow mixing and splitting balances
F O(i) − ∑ F OI(i , j) − ∑ F OL(i , j) − ∑ F OV(i , j) = 0 appearing in constraints 8−11 are inherently linear.
j=1 j=1 j=1
(2) The DN component flow mixing balances appearing in
∀ i = 1, ..., N (9) constraints 12 and 14 are linear, because the vapor
(liquid) composition of vapor (liquid) streams enter-
N ∞ ing or leaving the DN at any junction are known and
F L ( j) − ∑ F OL(i , j) − ∑ F PL(i , j) = 0 ∀ j = 1, ..., ∞ fixed.
i=1 i=1
Two propositions are next presented to simplify the structure
(10)
of the above IDEAS formulation by reducing the number of
N ∞ variables considered without compromising optimality.
F V ( j) − ∑ F OV(i , j) + ∑ F PV(i , j) = 0 Proposition 1. The OP inlet and outlet flow variables f Pk (i),
i=1 i=1 F (i), and FV(i), ∀i = 1, ..., ∞, can be eliminated from the
L

IDEAS formulation (IF1) through variable substitution without

∀ j = 1, ..., ∞ (11) compromising optimality.
11187 dx.doi.org/10.1021/ie300423q | Ind. Eng. Chem. Res. 2012, 51, 11183−11200
Industrial & Engineering Chemistry Research Article

Proof. Substitution of f Pk (i), ∀i = 1, ..., ∞, from eq 13 into eq M ∞

14 and subsequent substitution of FL(i) and FV(i), ∀i = 1, ..., ∞, ⇔∑ zkI(j) F PI(i , j) + ∑ xkL(j) F PL(i , j)
from eqs 10 and 11 into the resulting equation yields j=1 j = 1/ j ≠ i

M ∞ ⎡
∑ zkI(j) PI
F ( i , j) + ∑ xkL(j) PL
F ( i , j) ∞ ⎢N
+ ∑ ykV (j) F PV(i , j) − xkL(i)⎢∑ F OL(j , i)
j=1 j=1
j=1
⎢
∞
⎢ j=1
+ ∑ ykV (j) F PV(i , j) − xkL(i) F L(i) − ykV (i) FV(i) j≠i ⎣
j=1 ⎤ ⎡ ⎤
(10),(11)
M ∞ ∞ ⎥ ⎢N ∞ ⎥
⇒ ∑ zkI(j) F PI(i , j)+ ∑ xkL(j) F PL(i , j)
= 0 ==== + ∑ F PL(j , i)⎥ − ykV (i)⎢∑ F OV(j , i) + ∑ F PV(j , i)⎥⎥
j=1 j=1 j=1
⎥ ⎢ j=1
⎥ ⎢ j=1 ⎥
∞ ⎡N j≠i ⎦ ⎣ j≠i ⎦
+ ∑ ykV (j) F PV(i , j) − xkL(i)⎢∑ F OL(j , i)
j=1
⎢⎣ j = 1 =0 ∀ i = 1, ..., ∞ ; ∀ k = 1, ..., n (17)

∞ ⎤ ⎡N Equation 17 no longer contains the variables FPL(i,i) and

FPV(i,i) ∀i = 1, ..., ∞. This is also the case for all other equality
+ ∑ F PL(j , i)⎥ − ykV (i)⎢∑ F OV(j , i)
j=1
⎥⎦ ⎢⎣ j = 1 constraints. Thus, these variables appear only in the objective
function and in their respective non-negativity constraints. In
∞ ⎤ addition, the weight coefficient of each of these variables in the
+ ∑ F PV(j , i)⎥⎥ = 0 ∀ i = 1, ..., ∞ ; ∀ k = 1, ..., n objective function is positive (actually equal to 1). Because ν is
j=1 ⎦ a minimization problem, the value of each of these variables at
(15) the optimum is zero. QED
Based on these two properties, the mathematical formulation
Constraint 15 involves only the flow variables occurring of IDEAS (IF2) becomes
inside the DN, namely, FPI(i,j), FPL(i,j), FPV(i,j), FOL(i,j),
∞ M ∞ ∞
and FOV (i,j), and replaces eqs 10, 11, 13, and 14.
Furthermore, the non-negativity constraints f Pk (i) ≥ 0, ν ≙ inf ∑ ∑ F PI(i , j) + ∑ ∑ F PL(i , j)
i=1 j=1
FL(i) ≥ 0, and FV(i) ≥ 0, ∀i = 1, ..., ∞, are ensured by the i=1 j=1
j≠i
non-negativity of FPL(i,j), FPV(i,j), FOL(i,j), FOV(i,j), xLk (i),
∞ ∞
and yVk (i), ∀i = 1, ..., ∞; ∀j = 1, ..., ∞; ∀k = 1, ..., n. In
addition, the variables f Pk (i), FL(i), and FV(i), ∀i = 1, ..., ∞, + ∑ ∑ F PV(i , j)
do not appear in the objective function. Thus, they can be i=1 j=1
j≠i
omitted. QED
Proposition 2. The self-recycling liquid flows and vapor subject to constraints 8,9,12, and 17
flows FPL(i,i) and FPV(i,i), ∀i = 1, ..., ∞, can be eliminated
from the IDEAS formulation (IF1) without compromising F I ≥ 0, F O ≥ 0, F OI ≥ 0, F PI ≥ 0, F OL ≥ 0,
optimality.
Proof. The variables FPL(i,i) and FPV(i,i), ∀i = 1, ..., ∞, F OV ≥ 0, F PL ≥ 0, F PV ≥ 0 (18)
appear only in equality constraint 15 and in their respective An infinite-dimensional linear program cannot be solved
non-negativity inequalities. Constraint 15 can be rewritten as explicitly. However, its solution can be approximated by a
M ∞ series of finite linear programs of increasing size, whose
∑ zkI(j) PI
F ( i , j) + ∑ xkL(j) F PL(i , j) sequence of optimum values converges to the infinite-
j=1 j=1 dimensional problem’s infimum. In particular, consider the
j≠i aforementioned IDEAS formulation with optimum value
∞ ν, which aims at the synthesis of a multipressure flash
+ ∑ ykV (j) F PV(i , j) + xkL(i)F PL(i , i) + ykL (i) F PV(i , i) distillation network with M inlets and N outlets and allows
j=1 for the possible use of an infinite number of multipressure
j≠i flash separator units. By considering an ever-increasing
⎡ ⎤ number G of multipressure flash separator units, the
⎢N ∞ ⎥ optimum objective function values of the resulting finite
L ⎢ linear programs form a nonincreasing sequence {ν G} ∞
− xk (i) ∑ F OL(j , i) +
⎢
∑ F (j , i) + F (i , i)⎥⎥
PL PL 1 that
converges to ν. The DN of these finite-dimensional for-
j=1
⎢ j=1 ⎥ mulations contains (M + 2G) × (N + G) cross-flow streams.
⎣ j≠i ⎦ Proposition 2 reduces the number of those streams to
⎡ ⎤ (M + 2G) × (N + G) − 2G streams. These cross-flow
⎢N ∞ ⎥ streams are distributed as follows: M × N DN inlet to DN
− yk (i)⎢∑ F OV(j , i) +
V
∑ F (j , i) + F (i , i)⎥⎥ = 0
PV PV
outlet streams, M × G DN inlet to OP inlet streams, G × N
⎢ j=1 OP liquid outlet to DN outlet streams, G × N OP vapor
⎢ j=1 ⎥
⎣ j≠i ⎦ outlet to DN outlet streams, G2 − G OP liquid outlet to OP
inlet streams, and G2 − G OP vapor outlet to OP inlet
∀ i = 1, ..., ∞ ; ∀ k = 1, ..., n (16) streams.
11188 dx.doi.org/10.1021/ie300423q | Ind. Eng. Chem. Res. 2012, 51, 11183−11200
Industrial & Engineering Chemistry Research Article

Next, the proposed IDEAS framework is illustrated on a case captured in Figure 4, which illustrates the T−x−y diagram
study involving the dual-pressure distillative separation of an for the mixture under study at the two pressures considered.
azeotropic mixture.

3. CASE STUDY 1: DUAL-PRESSURE DISTILLATION OF

A BINARY MIXTURE OF METHYL ACETATE (1) AND
METHANOL (2)
In this case study, the dual-pressure (1 and 3 bar) distillative
separation of an equimolar mixture of methyl acetate
(species 1) and methanol (species 2) is considered. Indeed,
mixtures of methyl acetate (1) and methanol (2) exhibit
azeotropic behavior at 65.5% mole fraction of methyl
acetate at 1 bar and 56.1% mole fraction of methyl ace-
tate at 3 bar. This behavior is captured by the Gamma−
Phi vapor−liquid equilibrium model outlined in section 2.
The vapor phase is considered to be an ideal gas, whereas
the liquid-phase activity coefficients are quantified by the
Wilson equations 5 and 6 and the vapor pressure of
the various species is quantified by the Antoine equation7.
The molar flow rate of the stream to be separated is 2 mol/
s. The specifications of the two desired product streams
are 20% methyl acetate mole fraction for the first product
and 80% methyl acetate mole fraction for the second
product.
A traditional two-column PSD design method and the
aforementioned IDEAS design method are compared. To
this end, both design methods were emplyed to minimize
the total network unit inlet flow. The traditional design was
Figure 4. T−x1−y1 diagrams of methyl acetate (1)/methanol (2) at
pursued within the UniSim software platform, whereas the (top) 1 and (bottom) 3 bar.
IDEAS design was carried out with in-house-developed
IDEAS software. Both UniSim and IDEAS employed the
aforementioned Wilson and Antoine equation thermo- Given that UniSim’s coefficients are already built into the
dynamic models. Associated coefficient values are already software, for purposes of a fair comparison between the
built-in UniSim and can also be found in the literature traditional and IDEAS methods, UniSim’s values were also
(Gmehling et al.50 for the Wilson equation coefficients employed in IDEAS.
and Poling et al. 51 for the Antoine equation coefficients). Next, the optimization of a traditional two-column PSD
The values of these coefficients are summarized in Tables 1 design is carried out (section 3.1), followed by the IDEAS
optimization (section 3.2).
and 2. The thermodynamic behavior of these models is
3.1. Optimized Traditional Two-Column PSD Design.
A traditional two-column PSD design and its UniSim
Table 1. Wilson Equation Coefficients from UniSim and representation are shown in parts a and b, respectively, of
Gmehling et al.50 Figure 5. The objective function minimized here (FTotal) is the
total network unit inlet flow, which is FTotal = FT1 + FT2, where
Gmehling Gmehling
coefficient et al. UniSim coefficient et al. UniSim FT1 and FT2 are the sums of vapor and liquid molar flow rates
A11 (cal/mol) 0 0 A22 (cal/mol) 0 0 entering each plate in columns T1 and T2, respectively.
A12 (cal/mol) −31.19 9.117 V1 (cm3/mol) 79.84 80.26 To determine the number of degrees of freedom
A21 (cal/mol) 813.18 776.6 V2 (cm3/mol) 40.73 40.76 associated with the flow structure of the separation system
in Figure 5, the mass and component mass balance equa-
tions (using methyl acetate as the reference species) outside
Table 2. Antoine Equation Coefficients from UniSima and columns T1 and T2 and the inlet and outlet specifications
Poling et al.51 b are considered

k=1 k=2
⎧ F = B1 + B2 ⎫
⎪ ⎪
coefficient Poling et al. UniSim Poling et al. UniSim ⎪ D1 = D2 + B2 ⎪
⎪ ⎪
Ai 14.240 96.52 16.578 59.84 ⎪ F1 = D1 + B1 ⎪
Bi −2662.78 −7050 −3638.27 −6283 ⎨ ⎬, {F , z , xB1 , xB2}
Ci 219.69 0 239.50 0 ⎪ Fz = B1xB1 + B2 xB2 ⎪
Di 0 −12.38 0 −6.379
⎪ ⎪
⎪ D1xD1 = D2xD2 + B2 xB2 ⎪
Ei 0 1.137 × 10−5 0 4.617 × 10−6 ⎪ Fz = D x + B x ⎪
Fi 0 2 0 2 ⎩ 11 1 D1 1 B1 ⎭
a
T (K), P (kPa). bT (°C), P (kPa). ≙ {2, 0.5, 0.2, 0.8}

11189 dx.doi.org/10.1021/ie300423q | Ind. Eng. Chem. Res. 2012, 51, 11183−11200

Industrial & Engineering Chemistry Research Article

Figure 5. (a) Traditional two-column azeotropic separation system. (b) UniSim representation of the azeotropic separation system.

Figure 6. Two views of iso-xD1 lines of FTotal as a function of xD2 in (xD1,xD2) space.

This is a set of 10 equations involving 12 variables. There-
fore, there are two degrees of freedom (which can be chosen as
xD1 and xD2) associated with the traditional design’s flow
structure. These are augmented by four degrees of freedom
associated with column internals, namely, the number of plates
and the feed plate location for each of the two columns,
thus giving rise to six degrees of freedom in total for the
traditional design.
The solution to the minimization problem is carried
out using an exhaustive search in the triangular feasible
11190
region of the (xD1,xD2) two-dimensional space defined
by the physical inequality constraints xAZ1 ≥ xD1 ≥ xD2 ≥
xAZ2 . For every point (xD1 ,xD2 ) in this region, the
aforementioned mass and component mass balances are
solved, and the input−output flow/composition informa-
tion for each of the two columns is determined. Then, the
optimization problem is reduced to two separate single
column optimization problems, each of which involves
identifying the optimum number of plates and optimum
feed plate location and is carried out by systematically
dx.doi.org/10.1021/ie300423q | Ind. Eng. Chem. Res. 2012, 51, 11183−11200
Industrial & Engineering Chemistry Research
varying these integer-valued variables in UniSim. Every
time the number of plates or the location of the feed plate
changes, UniSim automatically adjusts the reflux and reboil
ratio to ensure feasible column operation. Following this
procedure, the globally minimum FTotal value as a function
of xD1 and xD2 is shown in Figure 6 and is found to be
202.93 mol/s at (xD1,xD2) optim = (0.629,0.586).
3.2. IDEAS-Generated Globally Optimal Azeotropic
Separation Design. According to the phase rule52 for binary
mixtures in equilibrium at a fixed pressure, knowledge of
temperature can yield a finite number of corresponding
mole fractions, thus in the context of IDEAS u1 = [ P T ]T.
Using the procedure outlined below, it can be determined
that, for P = 1 bar, if T ≤ 330 K, there exist two xL1 solutions
(q = 2), whereas if T > 330 K, there exists only one solution
xL1 (q = 1). Similarly, for P = 3 bar, if T ≤ 365.7 K, there exist
two xL1 solutions (q = 2), whereas if T > 365.7 K, there exists
only one solution xL1 (q = 1).
The following numerical/graphical procedure captures all
feasible flash separators corresponding to the two operating
pressures considered and a discretization of the feasible
temperature range.
(1) For each pressure, select a temperature T from the
considered temperature grid and then evaluate Psat k , ∀k =
1, 2, from eq 7.
(2) Express first γk, ∀k = 1, 2, and subsequently yVk , ∀k = 1, 2,
as a function of xL1 from eqs 5 and 6 and from eq 4,
respectively
(3) Generate the graph of the function ∑2k=1yVk versus xL1
plot for the chosen P and T and identify all values of xL1
∈ [0, 1]at which ∑2k=1yVk = 1 holds (as illustrated in
Figure 7). Each of these values corresponds to a
Figure 7. ∑2k=1yVk versus xL1 plots: (▲) (P, T) = (1 bar, 328 K), where
q = 2; (●) (P, T) = (1 bar, 333 K), where q = 1.
feasible flash separator with known values P, T, {xLk }2k=1,
and {yVk }2k=1.
(4) Return to step 1 and repeat steps 1−3 until all
temperatures and pressures have been considered.
All flashes created for a given pressure are part of an
ensemble of flash separator units called a pressure universe.
The above procedure is therefore repeated for several dif-
ferent pressure levels, to create as many pressure universes
as desired. At least two pressure universes are required for
11191
Article
the separation of a binary azeotropic mixture. The grid
discretization used for the temperature range can be selec-
ted to be uniform or nonuniform. A uniform grid generates
flashes at equal temperature intervals, whereas a nonuni-
form grid generates more flashes for some temperature
intervals and fewer flashes for others. Both uniform (section
3.2.1) and nonuniform (section 3.2.2) grids are used in the
following development to acquire globally optimal sol-
utions.
Exact product specifications on the network outlet are
achieved by setting the upper and lower bounds of the
outlet composition vector to be the same for each of the
outlet streams 1 and 2 such that [zO(1)]l = zO̅ (1) = [zO(1)]u
and [zO(2)]l = zO̅ (2) = [zO(2)]u, where zO̅ (1) and zO̅ (2) are
vectors of fixed mole fractions for the network outlet
streams 1 and 2 corresponding to zO̅ (1) = [0.8 0.2] and
zO̅ (2) = [0.2 0.8]. The network’s two outlet flows are such
that FO(1) = FO(2) = 1, and the network’s inlet mole
fractions and flow are zI(1) = [0.5 0.5]T and FI(1) = 2 . The
solution process for the IDEAS formulation involves the
sequential solution of finite linear programs of increasing
size G, until the values of the finite optima do not change
significantly (less than 0.25%).
3.2.1. Uniform Temperature Grid Discretization Strategy.
Figure 8 shows the convergence of the IDEAS optimum
Figure 8. IDEAS convergence: νG as a function of G. Optimized
traditional design (squares), three IDEAS approximations (ovals):
0.15 K, 0.075 K, and 0.0375 K.
objective function ν G as G increases. As previously
mentioned, the number of units made available to IDEAS
depends on the level of discretization of the temperature
space. Starting with an initial discretization level of 0.15 K,
the discretization levels chosen are uniform, cover the
entire temperature space, and correspond to 0.15 K/i, ∀T
(G = 193 units for i = 1, G = 386 units for i = 2, and G = 772
for i = 4).
At the smallest temperature discretization (0.0375 K) the
generated flow sheet has a converged total network unit
inlet flow value of 138.9072 mol/s. Because a tempera-
ture discretization of 0.0375 K contains as a subset in its
universe of flashes all flashes generated at a tempera-
ture discretization of 0.075 K and because the total net-
work unit inlet flow between these last two points dif-
fers by only 0.2%, convergence can be declared. This
represents a 31.54% reduction over the optimized two-
column design.
The actual number of units employed at the IDEAS opti-
mum is only a small fraction of the number of units made
dx.doi.org/10.1021/ie300423q | Ind. Eng. Chem. Res. 2012, 51, 11183−11200
Industrial & Engineering Chemistry Research Article

Figure 9. IDEAS optimum for G = 193 available units (uniform discretization). Liquid and vapor states indicate the phase of only the flash outlets
but not the inlets.

Table 3. Flash Outlet Flow Rates and Mole Fractions for IDEAS Design
unit no. xL (methyl acetate) flow destination yV (methyl acetate) flow destination
1 0.5628 13.992 2 0.6047 14.827 9
2 0.5117 7.690 3 0.5792 14.977 1
3 0.4444 4.860 4 0.5454 8.682 2
4 0.3751 3.320 6 0.5082 3.857 3
5 0.2957 0.209 7 0.4590 0.248 4
6 0.2830 0.795, 0.876 7, 8 0.4501 2.071 4
7 0.2035 0.548 outlet 1 0.3844 0.456 5
8 0.1956 0.452 outlet 1 0.3766 0.423 6
9 0.6249 18.449 10 0.5894 13.856 1
10 0.6653 5.263 11 0.6143 17.466 9
11 0.6938 4.536 12 0.6333 4.267 10
12 0.7360 2.729 14 0.6631 3.540 11
13 0.7681 0.206 15 0.6883 0.195 12
14 0.7818 0.750, 0.446 16, outlet 2 0.6997 1.536 12
15 0.8060 0.113 outlet 2 0.7211 0.0926 13
16 0.8167 0.441 outlet 2 0.7312 0.309 13
inlet 0.500 2.00 3 − − −

available to the IDEAS formulation. Indeed, the IDEAS-
generated flow sheet corresponding to 193 available units
(0.15 K) is shown in Figure 9 and consists of a total of 16
units. This particular IDEAS design has an optimum
objective function corresponding to 152.4843 mol/s, which
represents a 24.85% decrease over the optimized two-
column design. Every flow and corresponding quality vector
for that design is detailed in Table 3.
Because all approximating problems are linear programs,
the global optimality of their solution and associated
objective function values {ν G}∞ 1 is guaranteed. Thus, the
11192
IDEAS infimum ν provides the globally optimal solution to
the posed optimal design problem and, hence, represents a
lower bound to the cost of any alternative design employing
the same technologies (flash separators) and operating
pressures. This is confirmed in Figure 8, where the
optimum value of the traditional design (202.93 mol/s) is
seen to be above the IDEAS value of ν G = 138.9072 mol/s
for G = 772 available units.
When a stream portion is indicated as leaving a flash and
entering another flash, its thermal condition is known
at the point of departure but not at the point of arrival.
dx.doi.org/10.1021/ie300423q | Ind. Eng. Chem. Res. 2012, 51, 11183−11200
Industrial & Engineering Chemistry Research Article

The model considers that an appropriate cooling/heating This suggests that although, a very large number of units
apparatus and possibly an appropriate pressure-altering and interconnections is made available for consideration,
apparatus are associated with each DN flow from one flash only a select few participate in the optimal network. In addi-
to another. What is important to realize is that the net- tion, the optimal networks are physically meaningful
works identified by the present formulation can always be and can be realistically constructed. Other interesting simi-
realized a posteriori through the use of the aforemen- larities between the IDEAS optimum networks representa-
tioned apparatuses, and as previously mentioned, the tion and the traditional azeotropic separation structure are
identified minimum total flow is correct over all networks as follows:
regardless of whether the formulation includes energy (1) High-purity methyl acetate leaves the high-pressure
balances. (3 bar) sequence, whereas high-purity methanol leaves
Close examination of the generated optimum IDEAS
the low-pressure (1 bar) sequence, both in liquid
networks reveals that they have a number of properties:
form.
(1) They employ only a small percentage (<10%) of the (2) There exists only one stream connecting the low-pres-
available units. sure sequence to the high-pressure sequence and only
(2) They exhibit a small degree of interconnectedness, one recycle stream connecting the high-pressure sequence
despite the high degree of interconnectedness inherent back to the low-pressure sequence.
in the IDEAS framework. (3) The feed enters the low-pressure sequence at a single
(3) They exhibit countercurrent structures in some parts of unit, whose exit compositions straddle the feed
the network but not in others. composition.
(4) They exhibit only two interconnections between the (4) Vapor streams always flow against the temperature
high- and low-pressure parts of the network. gradient of flashes (hot to cold) belonging to a pressure
universe, whereas liquid streams always flow in the direc-
Table 4. Nonuniform Grid Discretizations of T for the Two tion of the temperature gradient (cold to hot) of flashes
Pressure Levels Considered belonging to a pressure universe.
P (bar) T range (K) T discretization (K) In contrast, interesting differences between the IDEAS repre-
sentation and the traditional azeotropic separation structure are
1 T ≤ 327 0.01875 (i = 16)
as follows:
1 T > 327 0.3 (i = 1)
3 T ≤ 359.6 0.01875 (i = 16) (1) Communication between low- and high-pressure sequen-
3 T > 359.6 0.3 (i = 1) ces occurs at the last stage of each sequence in IDEAS

Figure 10. IDEAS optimum for G = 150 available units (nonuniform discretization). Liquid and vapor states indicate the phase of only the flash
outlets but not the inlets.

11193 dx.doi.org/10.1021/ie300423q | Ind. Eng. Chem. Res. 2012, 51, 11183−11200

Industrial & Engineering Chemistry Research Article

Table 5. Flash Outlet Flow Rates and Mole Fractions for IDEAS Design 2
unit no. xL (methyl acetate) flow destination yV (methyl acetate) flow destination
1 0.5654 8.831 3 0.6060 14.794 15
2 0.5365 3.396 4 0.5915 4.440 1
3 0.5303 3.444 4 0.5884 5.380 1
4 0.4890 5.739 5 0.5679 7.833 2
5 0.4256 4.518 7 0.5357 6.732 4
6 0.3809 0.107 8 0.5114 0.1308 5
7 0.3443 2.211, 0.244 10, 11 0.4902 3.247 5
8 0.3134 0.149 12 0.4708 0.136 5
9 0.2866 0.231 13 0.4526 0.238 6
10 0.2630 1.165 14 0.4353 1.044 7
11 0.2420 0.102 outlet 1 0.4187 0.141 7
12 0.2231 0.0741 outlet 1 0.4025 0.0754 8
13 0.2059 0.127 outlet 1 0.3866 0.103 8
14 0.1902 0.695 outlet 1 0.3711 0.469 9
15 0.6272 9.857 16 0.5907 13.823 1
16 0.6539 6.609 17 0.6071 8.866 15
17 0.6852 5.751 18 0.6272 5.615 16
18 0.7323 2.628 19 0.6603 4.755 17
19 0.7652 0.952, 0.853 20, 21 0.6858 1.322 18
20 0.7919 0.644 outlet 2 0.7084 0.307 18
21 0.8146 0.355 outlet 2 0.7292 0.498 19
inlet 0.500 2.00 5 − − −

Table 6. Wilson Equation Coefficients

coefficient value coefficient value
A11 (cal/mol) 0 A31 (cal/mol) 291.27
A12 (cal/mol) 469.55 A32 (cal/mol) −161.88
A13 (cal/mol) 1448.01 A33 (cal/mol) 0
A21 (cal/mol) 107.33 V1 (cm3/mol) 17.88
A22 (cal/mol) 0 V2 (cm3/mol) 40.76
A23 (cal/mol) 583.11 V3 (cm3/mol) 74.47

Figure 11. Isopurity lines of the total flow entering network flashes as Table 7. Antoine Equation Coefficientsa
a function of pressure.
coefficient k=1 k=2 k=3
Ai 65.93 59.84 71.30
Bi −7228 −6283 −5952
designs, while it occurs at intermediate stages in the Ci 0 0 0
traditional design. Di −7.177 −6.379 −8.531
(2) The high-pressure sequence has no rectifying section. Ei 4.031 × 10−6 4.617 × 10−6 7.824 × 10−6
Analysis of the above flow sheets indicates that the flash inlet Fi 2 2 2
flow quantity is largest for network flashes within close a
T (K), P (kPa).
proximity of the azeotropic pinch point (lower temperatures).
Therefore, a nonuniform temperature grid is introduced in the
next section to allow the generation of more flashes around to
the azeotropic pinch point. This approach considerably reduces discretization strategy and G = 772 units. The IDEAS-
the problem size while still resulting in globally optimal flow generated flow sheet corresponding to this number of
sheets. flashes is shown in Figure 10, with the flow and the
3.2.2. Nonuniform Temperature Grid Discretization corresponding quality vector detailed in Table 5. Therefore,
Strategy. Starting with an initial discretization level of 0.3 employing a nonuniform temperature grid discretization
K, the discretization level chosen is nonuniform, with its can potentially serve as an intelligent allocation of limited
specifics displayed in Table 4 corresponding to G = 150 computational resources while still yielding globally
units. The resulting objective function is equal to 138.6528 optimal designs.
mol/s, which is effectively identical to the converged In the next section, the effects of the operating pressure
globally optimal value of 138.9072 mol/s determined in the of the second universe of flashes (while the operating
previous section using a uniform temperature grid pressure of the first universe of flashes is kept equal to
11194 dx.doi.org/10.1021/ie300423q | Ind. Eng. Chem. Res. 2012, 51, 11183−11200
Industrial & Engineering Chemistry Research Article

given product purity, a design operating at lower pressure

requires more total unit inlet flow than a design operat-
ing at a higher pressure. This intuitive trend manifests itself
in IDEAS designs as well. Figure 11 shows the total unit
inlet flow as a function of pressure at three different
product purities for a uniform discretization of 0.15 K. This
discretization level was selected because it leads to readily
realizable flow sheets and a smaller computational burden
than that needed for smaller discretization levels.
The obtained IDEAS optimization results indicate that,
after an initial sharp decrease in total unit inlet flow at low
pressures, the impact of further increasing pressure is
minimal. This trend repeats itself at all product purities,
albeit at lower product purities the curves begin to level off
at lower pressures (earlier) than for higher product purities.
Therefore, IDEAS can help determine not only optimal
design structures and the optimal total unit inlet flow but
also appropriate levels of operating pressure for a desired
Figure 12. T−x2−y2 diagram of methanol (2)/acetone (3) at (top) 0.2 level of product purity.
and (bottom) 1 bar.
4. CASE STUDY 2: DUAL-PRESSURE DISTILLATION OF
A TERNARY MIXTURE OF WATER (1), METHANOL
Table 8. Design Specifications (Component, Mole Fraction,
(2), AND ACETONE (3)
Flow Rate)
In this case study, the dual-pressure distillative separation
specification set 1 specification set 2 of a mixture of water (species 1), methanol (species 2), and
product 1 water, 0.80, 1.0 mol/s water, 0.80, free acetone (species 3) is considered. The mixture’s molar
product 2 methanol, 0.80, 1.0 mol/s methanol, 0.80, free flow rate and the mole fractions of species 1, 2, and 3 are
product 3 acetone, 0.80, 1.0 mol/s acetone, 0.80, 1.2 mol/s 3 mol/s and 0.3, 0.35, and 0.35, respectively. Binary mix-
tures of methanol (2) and acetone (3) exhibit a minimum-
boiling azeotrope at 79.07% mole fraction of acetone (3)
atmospheric) and product purity on the IDEAS optimum is at 1 bar and 328.5 K and 97.5% mole fraction of acetone (3)
quantified. at 0.2 bar and 288.7 K. Similarly to the previous case
3.3. Effects of Pressure and Purity on the Total Flash study, the mixture’s thermodynamic behavior is captured
Inlet Flow. The IDEAS framework makes it possible to by a Gamma−Phi vapor−liquid equilibrium model. The
systematically and rigorously evaluate the effects that vapor phase is considered to be an ideal gas, whereas the
product purity and pressure have on the optimum objective liquid-phase activity coefficients are quantified by the
function value of the total inlet flash flow. Consider the Wilson equations (eqs 5 and 6), and the vapor pressures
traditional azeotropic separation structures described in of the various species are quantified by the Antoine
section 3.1. For a given pressure, a design specifying lower equation7.
product purity requires less total unit inlet flow than a Similarly to case study 1, a traditional two-column PSD
design specifying higher product purity. Similarly, for a design method and the aforementioned IDEAS design

Figure 13. Optimized traditional two-column setup for specification set 2.

11195 dx.doi.org/10.1021/ie300423q | Ind. Eng. Chem. Res. 2012, 51, 11183−11200

Industrial & Engineering Chemistry Research
method are compared. To this end, both design methods
were used minimize the total network unit inlet flow. The
traditional design was pursued within the UniSim software
platform, whereas the IDEAS design was carried out with
in-house-developed IDEAS software. Both UniSim and
IDEAS employed the aforementioned Wilson and Antoine
equation thermodynamic models. The coefficient values
employed in the two methods are summarized in Tables 6
and 7. The ability of these thermodynamic models to
capture the azeotropic behavior of the methanol (2)/
acetone (3) mixture is shown in Figure 12, which illustrates
the mixture’s T−x−y equilibrium diagram, at the two pres-
sures considered.
Two sets of design specifications were considered, as shown
in Table 8.
For each set of specifications, a traditional two-column
design33 was compared to the IDEAS-generated optimal design.
The traditional design was carried out in UniSim, and the
aforementioned thermodynamic data were used in both the
UniSim and IDEAS designs.
4.1. Optimized Traditional Two-Column Design. A
UniSim representation of the considered two-column de-
sign is shown in Figure 13. This traditional design has been
shown to be suitable to separate the water−methanol−
acetone mixture.33 The objective function to be minimized
is again the same as for the first case study (total net-
work unit inlet flow), FTotal = FT1 + FT2, where FT1 and FT2
are the sums of vapor and liquid molar flow rates entering
each plate in columns T 1 (P = 0.2 bar) and T2 (P = 1 bar).
respectively. The first distillation column in the considered
design has a known feed. Thus, it has four degrees of
freedom,48 two of them integer variables (number of
plates and feed plate location) and the other two
continuous. In this case study, for both specification sets,
the two continuous degrees of freedom are chosen to be the
specified top product flow rate and specified acetone mole
fraction. Once the first column’s degrees of freedom are
specified, the second column’s feed is also specified. Thus
again, the second column has two integer and two con-
tinuous degrees of freedom. For the first set of speci-
fications, the second column must meet four specifica-
tions, namely, the flow rates of both of its products are
known, and the methanol mole fraction in its distillate
and the water mole fraction in its bottom are known. Given
the continuous nature of the specifications and the
integer nature of two of the four degrees of freedom, it
is likely that the considered traditional design might not
be able to meet the first set of specifications, because it
has four integer degrees of freedom and must meet two
continuous specifications. On the other hand, for the
second set of specifications, the first column again has
only two integer degrees of freedom. The second column,
however, must meet two specifications, namely, the meth-
anol mole fraction in its distillate and the water mole
fraction in its bottom are known. By selecting these varia-
bles as the second column’s continuous degrees of free-
dom, the second column has two integer degrees of
freedom. Therefore, the traditional design has no unmet
continuous specifications, and four integer degrees of
freedom that can be used to optimize the design’s total
network flow.
11196
Article
An exhaustive search over all possible values of the four
integer degrees of freedom is carried out through repeated
UniSim simulations. For specification set 1, which must
meet two continuous specifications, there exists no two-
column design capable of delivering the desired specifica-
tions. For specification set 2 (Figure 13), which has no
unmet continuous specifications, the optimized two-column
design yields distillate and bottom streams flows equal to 0.7634
and 1.037 mol/s, respectively. The optimized total network flow
is equal to 44.487 mol/s.
4.2. IDEAS-Generated Globally Optimal Azeotropic
Separation Design. According to the phase rule52 for
ternary mixtures in equilibrium at a fixed pressure, knowl-
edge of temperature, and species 1 mole fraction can yield
a finite number of corresponding mole fractions for the
other two species; thus in the context of IDEAS, u1 can be
chosen as u1 = [ P T x1L ]T. Superscript q, when needed, is
used to indicate multiple xL2 solutions corresponding to a
fixed T,xL1 .
A numerical/graphical procedure similar to the one
outlined in the first case study is employed to capture all
feasible flash separators corresponding to the two operat-
ing pressures considered, by discretizing the feasible
range of both temperature T and the first species’ liquid-
phase mole fraction xL1 . Figure 14 illustrates how feasible
Figure 14. ∑3k=1yVk versus xL2 plots for an xL1 discretization of 0.125 at
T = 336 K, P = 1 bar. Triangles indicate feasible flash separators.
separators are generated, by identifying the xL2 values for
which ∑3k=1yVk = 1 at any given values of xL1 , T, and P.
In this case, a nonuniform grid, with an increa-
sed refinement strategy at low temperatures, to better
capture the change in composition around the azeotropic
pinch point, is used. The grid sizes for T and xL1 are displayed in
Table 9 for different temperature ranges and pressures.
Table 9. Nonuniform Grid Discretizations of T and xL1 for
the Two Pressure Levels Considered
P (bar) T range (°C) T discretization (K) xL1 discretization
1 T ≤ 60 1 0.03125 (1/32)
1 T > 60 2 0.0625 (1/16)
0.2 T ≤ 18 0.5 0.03125 (1/32)
0.2 T > 18 1 0.0625 (1/16)
dx.doi.org/10.1021/ie300423q | Ind. Eng. Chem. Res. 2012, 51, 11183−11200
Industrial & Engineering Chemistry Research Article

(1) For the first set of specifications, which is unattainable by Table A2. Optimal Network Flashes: P = 0.2 bar, T (°C)
the traditional two-column method, IDEAS is able to
T = 16.0, #105, x1 = 0.0312, x2 = 0.1190, y1 = 0.0138, y2 = 0.0976
identify a feasible design. The optimum solution
T = 16.5, #108, x1 = 0.0625, x2 = 0.1630, y1 = 0.0236, y2 = 0.1250
obtained features a total network flow of 43.785
T = 17.0, #113, x1 = 0.0938, x2 = 0.1980, y1 = 0.0315, y2 = 0.1452
mol/s. The identities of the flashes participating in the
T = 17.5, #120, x1 = 0.1250, x2 = 0.2270, y1 = 0.0382, y2 = 0.1617
optimum network, as well as the optimal interconnect-
T = 18.0, #125, x1 = 0.0312, x2 = 0.4550, y1 = 0.0095, y2 = 0.3074
ing flows, are shown in Appendix A and Appendix B,
T = 18.0, #129, x1 = 0.1562, x2 = 0.2520, y1 = 0.0441, y2 = 0.1763
respectively.
T = 18.0, #130, x1 = 0.1875, x2 = 0.2020, y1 = 0.0518, y2 = 0.1432
(2) For the second set of specifications, which is attainable
T = 19.5, #135, x1 = 0.0625, x2 = 0.5530, y1 = 0.0173, y2 = 0.3747
by the traditional two-column method with a minimum
T = 19.5, #138, x1 = 0.2500, x2 = 0.3020, y1 = 0.0597, y2 = 0.2106
total network flow of 44.487 mol/s, the IDEAS optimum
#139,
design features a total network flow of 37.738 mol/s, T = 19.5, x1 = 0.3125, x2 = 0.2270, y1 = 0.0704, y2 = 0.1616
which is 15.15% lower than the traditional design’s T = 20.5, #144, x1 = 0.0625, x2 = 0.6260, y1 = 0.0171, y2 = 0.4342
optimum value. T = 20.5, #147, x1 = 0.2500, x2 = 0.3880, y1 = 0.0589, y2 = 0.2744
T = 20.5, #148, x1 = 0.3125, x2 = 0.3160, y1 = 0.0696, y2 = 0.2271
T = 20.5, #149, x1 = 0.3750, x2 = 0.2490, y1 = 0.0788, y2 = 0.1831
5. CONCLUSIONS T = 20.5, #150, x1 = 0.4375, x2 = 0.1890, y1 = 0.0865, y2 = 0.1431
A methodology is demonstrated for the global minimiza- T = 21.5, #159, x1 = 0.3125, x2 = 0.3830, y1 = 0.0702, y2 = 0.2833
tion of total network flash inlet flow for pressure swing T = 21.5, #160, x1 = 0.3750, x2 = 0.3170, y1 = 0.0797, y2 = 0.2393
distillation (PSD) systems using the IDEAS framework. T = 22.5, #167, x1 = 0.0625, x2 = 0.7350, y1 = 0.0175, y2 = 0.5490
IDEAS yields globally optimal designs that minimize the T = 22.5, #168, x1 = 0.1250, x2 = 0.6570, y1 = 0.0334, y2 = 0.4914
total network flow required to break an azeotrope achiev- T = 22.5, #171, x1 = 0.3125, x2 = 0.4370, y1 = 0.0717, y2 = 0.3359
ing the desired level of separation. This allows for a T = 22.5, #172, x1 = 0.3750, x2 = 0.3700, y1 = 0.0816, y2 = 0.2901
rigorous comparison of alternative designs. The IDEAS T = 23.5, #181, x1 = 0.1250, x2 = 0.7000, y1 = 0.0342, y2 = 0.5474
framework is able to successfully generate optimal
T = 23.5, #185, x1 = 0.3750, x2 = 0.4130, y1 = 0.0842, y2 = 0.3385
distillation networks using flash units as building blocks,
T = 23.5, #186, x1 = 0.4375, x2 = 0.3490, y1 = 0.0933, y2 = 0.2936
without any preconception of a network structure. Only
vapor−liquid equilibrium data information is provided a T = 23.5, #190, x1 = 0.6875, x2 = 0.1280, y1 = 0.1199, y2 = 0.1311
priori to the IDEAS design procedure. Two case studies T = 24.5, #195, x1 = 0.1875, x2 = 0.6610, y1 = 0.0506, y2 = 0.5448
consisting of a binary and a ternary mixture are consi- T = 24.5, #199, x1 = 0.4375, x2 = 0.3840, y1 = 0.0970, y2 = 0.3388
dered, and the obtained IDEAS designs indicate that the T = 24.5, #200, x1 = 0.5000, x2 = 0.3210, y1 = 0.1055, y2 = 0.2930
traditional two-column design can be significantly im- T = 25.5, #206, x1 = 0.0625, x2 = 0.8460, y1 = 0.0191, y2 = 0.7269
proved (by 31.54% for the binary case and by 15.15% for T = 25.5, #207, x1 = 0.1250, x2 = 0.7680, y1 = 0.0366, y2 = 0.6615
the ternary case). IDEAS-generated designs surpass the T = 25.5, #213, x1 = 0.5000, x2 = 0.3490, y1 = 0.1102, y2 = 0.3349
two-column PSD traditional designs, because the flexibility T = 26.5, #219, x1 = 0.1250, x2 = 0.7960, y1 = 0.0380, y2 = 0.7209
of the IDEAS framework allows the consideration of all T = 26.5, #220, x1 = 0.1875, x2 = 0.7210, y1 = 0.0545, y2 = 0.6569
possible flow and unit combinations. Not only can IDEAS T = 26.5, #225, x1 = 0.5000, x2 = 0.3730, y1 = 0.1154, y2 = 0.3768
lead to improvements when compared to optimized T = 27.5, #230, x1 = 0.1250, x2 = 0.8210, y1 = 0.0395, y2 = 0.7825
traditional designs but it can also provide vital informa- T = 27.5, #231, x1 = 0.1875, x2 = 0.7450, y1 = 0.0568, y2 = 0.7145
tion in other areas. Indeed, for the binary mixture, the T = 27.5, #236, x1 = 0.5000, x2 = 0.3940, y1 = 0.1210, y2 = 0.4194
obtained IDEAS designs indicate that, following an initial
T = 28.5, #243, x1 = 0.3750, x2 = 0.5480, y1 = 0.1041, y2 = 0.5790
sharp decrease, the dependence of the optimum objective
T = 29.5, #247, x1 = 0.1250, x2 = 0.8630, y1 = 0.0430, y2 = 0.9117
function value on pressure is minimal, and higher purity
level requirements lead to increased optimum objective T = 29.5, #248, x1 = 0.1875, x2 = 0.7870, y1 = 0.0619, y2 = 0.8373
function values with similar pressure dependence. Also, for T = 29.5, #249, x1 = 0.3125, x2 = 0.6380, y1 = 0.0949, y2 = 0.6963
ternary mixtures, IDEAS can deliver compositions and T = 29.5, #253, x1 = 0.6250, x2 = 0.2970, y1 = 0.1537, y2 = 0.3860
outlet flow rates unattainable using the rigid two-column T = 30.5, #256, x1 = 0.2500, x2 = 0.7290, y1 = 0.0829, y2 = 0.8258
design structure. T = 30.5, #257, x1 = 0.3125, x2 = 0.6550, y1 = 0.0994, y2 = 0.7534

■
T = 30.5, #258, x1 = 0.4375, x2 = 0.5120, y1 = 0.1280, y2 = 0.6171
APPENDIX A T = 31.5, #261, x1 = 0.2500, x2 = 0.7450, y1 = 0.0869, y2 = 0.8891
T = 31.5, #262, x1 = 0.3125, x2 = 0.6700, y1 = 0.1043, y2 = 0.8120
The identities of the flashes participating in the optimum
T = 31.5, #263, x1 = 0.3750, x2 = 0.5970, y1 = 0.1201, y2 = 0.7384
network are listed in Tables A1 and A2.
T = 33.5, #270, x1 = 0.4375, x2 = 0.5500, y1 = 0.1486, y2 = 0.7753
T = 35.5, #273, x1 = 0.5000, x2 = 0.4980, y1 = 0.1808, y2 = 0.8044
Table A1. Optimal Network Flashes: P = 1 bar, T (°C) T = 36.5, #276, x1 = 0.6250, x2 = 0.3640, y1 = 0.2213, y2 = 0.6784
T = 38.5, #279, x1 = 0.7500, x2 = 0.2370, y1 = 0.2766, y2 = 0.5600
T = 69.0, #79, x1 = 0.3125, x2 = 0.6760, y1 = 0.1275, y2 = 0.8319 T = 39.5, #280, x1 = 0.6875, x2 = 0.3120, y1 = 0.2751, y2 = 0.7180
T = 77.0, #98, x1 = 0.6875, x2 = 0.3050, y1 = 0.3167, y2 = 0.6273
T = 77.0, #100, x1 = 0.8750, x2 = 0.0870, y1 = 0.3771, y2 = 0.2261
T
T
T
=
=
=
79.0,
81.0,
85.0,
#101, x1 = 0.8750, x2 = 0.0970, y1 = 0.4083, y2 = 0.2722
#102, x1 = 0.8750, x2 = 0.1060, y1 = 0.4417, y2 = 0.3209
#104, x1 = 0.8750, x2 = 0.1210, y1 = 0.5160, y2 = 0.4249
■ APPENDIX B
The optimal interconnecting flows are listed in Table B1.
11197 dx.doi.org/10.1021/ie300423q | Ind. Eng. Chem. Res. 2012, 51, 11183−11200
Industrial & Engineering Chemistry Research Article

Table B1. Optimal Interconnecting Flows

FP(171,1) = FPL(159,148) = FPL(273,269) = FPV(149,160) = FPL(101,104) = FPL(248,230) = FPV(125,135) = FPV(230,257) =
0.140 0.142 0.090 0.344 0.097 0.029 0.423 0.031
FP(185,1) = FPL(160,149) = FPL(273,270) = FPV(150,149) = FPL(102,104) = FPL(249,195) = FPV(129,138) = FPV(231,263) =
0.800 0.342 0.273 0.156 0.114 0.181 0.246 0.290
FP(186,1) = FPL(168,144) = FPL(276,243) = FPV(159,135) = FPL(104,98) = FPL(253,186) = FPV(130,139) = FPV(243,279) =
0.010 0.050 0.617 0.103 0.280 0.018 0.005 0.962
FP(199,1) = FPL(172,148) = FPL(276,258) = FPV(159,199) = FPL(104,213) = FPL(256,248) = FPV(135,144) = FPV(247,261) =
1.091 0.432 0.371 0.058 0.372 0.132 0.513 0.062
FP(213,1) = FPL(181,144) = FPL(276,263) = FPV(160,125) =
0.958 0.703 0.154 0.236 FPL(104,253) = FPL(257,220) = FPV(135,168) = FPV(248,79) =
0.028 0.174 0.068 0.063
FOL(1,100) = FPL(185,147) = FPL(279,213) = FPV(160,186) =
0.814 1.059 0.273 0.011 FPL(104,267) = FPL(258,195) = FPV(135,245) = FPV(248,256) =
0.040 0.115 0.088 0.188
FOL(1,101) = FPL(186,148) = FPL(279,225) = FPV(160,200) =
0.086 0.018 0.341 0.039 FPL(104,279) = FPL(261,247) = FPV(135,267) = FPV(248,257) =
FOV(1,190) = FPL(190,100) = FPL(279,236) = FPV(167,207) = 1.941 0.042 0.011 0.069
0.092 0.028 1.601 0.198 FPL(104,280) = FPL(262,220) = FPV(138,100) = FPV(248,261) =
FOL(2,167) = FPL(199,147) = FPL(279,245) = FPV(167,220) = 0.259 0.089 0.525 0.217
0.158 0.599 0.087 0.153 FPL(135,125) = FPL(262,231) = FPV(138,147) = FPV(248,262) =
FOV(2,206) = FPL(200,148) = FPL(279,276) = FPV(168,243) = 0.197 0.185 1.599 0.254
0.043 0.074 0.205 0.115 FPL(138,129) = FPL(262,248) = FPV(139,138) = FPV(249,276) =
FOV(2,220) = FPL(213,160) = FPL(280,270) = FPV(171,236) = 0.103 0.150 0.376 0.115
0.080 0.289 0.048 0.180 FPL(138,139) = FPL(263,248) = FPV(144,167) = FPV(253,102) =
FOV(2,230) = FPL(219,167) = FPL(280,273) = FPV(172,225) = 0.291 0.019 0.138 0.028
0.031 0.051 0.430 0.199 FPL(139,102) = FPL(267,200) = FPV(144,181) = FPV(256,270) =
FOV(2,248) = FPL(220,168) = FPV(100,138) = FPV(181,258) = 0.064 0.050 0.477 0.229
0.685 0.104 0.345 0.245
FPL(139,113) = FPL(267,213) = FPV(144,195) = FPV(257,276) =
FOL(3,105) = FPL(220,181) = FPV(101,100) = FPV(185,236) = 0.146 0.014 0.297 0.077
0.011 0.392 0.041 0.382
FPL(139,190) = FPL(269,261) = FPV(147,171) = FPV(258,279) =
FOL(3,108) = FPL(225,159) = FPV(102,101) = FPV(190,150) = 0.032 0.151 0.156 0.990
0.182 0.189 0.026 0.095
FOV(3,113) = FPL(225,172) = FPV(105,108) = FPV(195,243) = FPL(144,135) = FPL(269,262) = FPV(147,185) = FPV(261,269) =
0.191 0.361 0.007 0.228 0.355 0.022 0.994 0.083
FOV(3,120) = FPL(230,206) = FPV(105,130) = FPV(195,258) = FPL(147,138) = FPL(270,249) = FPV(147,199) = FPV(261,273) =
0.612 0.027 0.003 0.365 1.216 0.144 0.713 0.437
FPL(79,248) = FPL(231,181) = FPV(108,120) = FPV(200,101) = FPL(148,149) = FPL(270,257) = FPV(147,213) = FPV(263,276) =
0.021 0.085 0.055 0.015 0.533 0.116 0.584 0.458
FPL(79,256) = FPL(236,185) = FPV(108,139) = FPV(206,219) = FPL(149,104) = FPL(270,262) = FPV(148,147) = FPV(270,98) =
0.173 1.245 0.134 0.054 0.262 0.123 0.406 0.134
FPL(79,262) = FPL(236,199) = FPV(113,129) = FPV(206,247) = FPL(149,120) = FPL(270,263) = FPV(148,159) = FPV(273,280) =
0.024 0.918 0.152 0.025 0.482 0.033 0.199 0.225
FPL(98,258) = FPL(245,171) = FPV(113,139) = FPV(207,231) = FPL(149,150) = FPL(273,79) = FPV(148,172) = FPV(279,104) =
0.127 0.173 0.186 0.188 0.098 0.145 0.274 1.388
FPL(98,276) = FPL(248,207) = FPV(120,138) = FPV(207,249) = FPL(150,102) = FPL(273,261) = FPV(149,148) =
0.287 0.144 0.250 0.152 0.038 0.134 0.746
FPL(100,104) = FPL(248,219) = FPV(120,149) = FPV(219,257) = FPL(159,138) =
1.063 0.042 0.899 0.035 0.086

■ AUTHOR INFORMATION
Corresponding Author
xLk (i) = kth-species equilibrium liquid composition leaving
the ith unit
*E-mail: vasilios@ucla.edu. Tel.: +1 310 206 0300. Fax: +1 310 yVk (i) = kth-species equilibrium vapor composition leaving
206 4107 420. the ith unit
Psat
k (T) = kth-species temperature-dependent saturated vapor
Notes
pressure
The authors declare no competing financial interest.

■
γk({xLl }nl=1,T) = kth-species nonideal liquid activity coefficient
ACKNOWLEDGMENTS Λk,j(T) = Wilson equation temperature-dependent parame-
ter
Financial support for this work through NSF Grant NSF-CBET ϕk({yVl }nl=1, T,P) = kth-species nonideal fugacity coefficient
0829211 is gratefully acknowledged.

■ NOTATION
Thermodynamic Variables
Traditional PSD Design Variables

F = inlet flow rate to the PSD system

Ai,j = Wilson equation interaction parameters between the ith F1 = inlet flow rate to the first distillation column, F1 = F +
and jth species D2
Ak, Bk, Ck, Dk, Ek, Fk = Antoine equation kth-species D1 = flow rate of distillate stream leaving the first distillation
parameters column for the second column
R = universal gas constant D2 = flow rate of distillate stream leaving the second
P = flash-unit pressure distillation column, recycled back to the first column
T = flash-unit temperature B1 = bottom flow rate leaving the first distillation column
Vk = kth-species molar volume B2 = bottom flow rate leaving the second distillation column
11198 dx.doi.org/10.1021/ie300423q | Ind. Eng. Chem. Res. 2012, 51, 11183−11200
Industrial & Engineering Chemistry Research
FT2 = total flow of vapor and liquid inside the second (high-
pressure) distillation column
FT1 = total flow of vapor and liquid inside the first (low-
pressure) distillation column
FTotal = sum of the flow rates entering each plate inside each
column in the PSD system
z = Composition of methyl acetate in the inlet flow rate to
the PSD system
z1 = Composition of methyl acetate in the inlet flow rate to
the first distillation column
xD1 = Composition of methyl acetate in the distillate stream
leaving the first distillation column
xD2 = Composition of methyl acetate in the distillate stream
leaving the second distillation column
xB1 = Composition of methyl acetate in the bottom stream
leaving the first distillation column
xB2 = Composition of methyl acetate in the bottom stream
leaving the second distillation column
xAZ1 = Azeotropic composition of methyl acetate in the low-
pressure column (column T1)
xAZ2 = Azeotropic composition of methyl acetate in the high-
pressure column (column T2)
IDEAS Variables
FI(i) = ith DN inlet stream
FL(i) = ith OP liquid outlet
FO(i) = ith DN outlet stream
FOI(i,j) = jth DN inlet stream to the ith DN outlet
FOL(i,j) = ith DN outlet stream from the jth OP liquid outlet
FOV(i,j) = ith DN outlet stream from the jth OP vapor outlet
FPI(i,j) = ith OP inlet stream from the jth DN network inlet
FPL(i,j) = ith OP inlet stream from the jth OP liquid outlet
FPV(i,j) = ith OP inlet stream from the jth OP vapor outlet
FV(i) = ith OP vapor outlet
G = total number of flashes generated in all pressure
universes for different discretizations
M = number of IDEAS network inlets
N = number of IDEAS network outlets
xLk (i) = kth-species, ith OP liquid outlet composition
yVk (i) = kth-species, ith OP vapor outlet composition
ZIk(i) = kth-species, ith DN inlet stream composition
ZOk (i) = kth-species, ith DN outlet stream composition
zO(i), [zO(i)]l, [zO(i)]u = composition vector of the ith DN
outlet stream and its lower and upper bounds, respectively
■
REFERENCES
(1) Roscoe, H. E. On the Composition of the Aqueous Acids of
Constant Boiling Point. J. Chem. Soc. 1860, 13, 146−164.
(2) Roscoe, H. E. On the Composition of the Aqueous Acids of
Constant Boiling PointSecond Communication. J. Chem. Soc. 1862,
15, 270−276.
(3) Roscoe, H. E.; Ditmar, W. On the Absorption of Hydrochloric
Acid and Ammonia in Water. J. Chem. Soc. 1859, 12, 128−159.
(4) Diwekar, U. Batch Distillation Simulation, Optimal Design, and
Control, 2nd ed.; CRC Press: Boca Raton, FL, 2011; p 236.
(5) Knapp, J.; Doherty, M. A New Pressure-Swing Distillation
Process for Separating Homogeneous Azeotropic Mixtures. Ind. Eng.
Chem. Res. 1992, 31, 346−357.
(6) Barnicki, S.; Siirola, J. Process Synthesis Prospective. Comput.
Chem. Eng. 2004, 28 (4), 441−446.
(7) Noble, R.; Agrawal, R. Separations Research Needs for 21st
Century. Ind. Eng. Chem. Res. 2005, 44 (9), 2887−2892.
(8) Horsley, L. H., Ed. Azeotropic Data III; Advances in Chemistry
Series; American Chemical Society: Washington, DC, 1973; Vol. 116.
11199
Article
(9) Knapp, J.; Doherty, M. Thermal Integration of Homogeneous
Azeotropic Distillation Columns. AIChE J. 1990, 36, 969−984.
(10) Abu Eishah, S. I.; Luyben, W. L. Design and Control of a Two-
Column Azeotropic Distillation System. Ind. Eng. Chem. Process Des.
Dev. 1985, 24, 132−140.
(11) Knight, J.; Doherty, M. Optimal Design and Synthesis of
Homogeneous Azeotropic Distillation Sequences. Ind. Eng. Chem. Res.
1989, 28, 564−572.
(12) Pham, H. N.; Ryan, P.; Doherty, M. F. Design and Minimum
Reflux for Heterogeneous Azeotropic Distillation Columns. AIChE J.
1989, 35 (10), 1585−1591.
(13) Ryan, P.; Doherty, M. F. Design/Optimization of Ternary
Heterogeneous Azeotropic Distillation Sequences. AIChE J. 1989, 35
(10), 1592−1601.
(14) Bossen, B. J.; Jorgensen, S. B.; Gani, R. Simulation, Design and
Analysis of Azeotropic Distillation Operations. Ind. Eng. Chem. Res.
1993, 32, 620−633.
(15) Mortaheb, H. R.; Kosuge, H. Simulation and Optimization of
Heterogeneous Azeotropic Distillation Process with a Rate-Based
Model. Chem. Eng. Process. 2004, 43 (3), 317−326.
(16) Martin, L. L.; Huang, X. Optimized Heat and Power Exchanger
Network in Ethanol−Water Pressure Swing Distillation. In Proceedings
of the AIChE Annual Meeting; American Institute of Chemical
Engineers (AIChE): New York, 2005; Paper 599d.
(17) Fidkowski, Z. T.; Malone, M. F.; Doherty, M. F. Computing
Azeotropes in Multicomponent Mixtures. Comput. Chem. Eng. 1993,
17 (12), 1141−1155.
(18) Pham, H. N.; Doherty, M. F. Design and Synthesis of
Heterogeneous Azeotropic Distillations: II. Residue Curve Maps.
Chem. Eng. Sci. 1990, 45 (7), 1837−1843.
(19) Pham, H. N.; Doherty, M. F. Design and Synthesis of
Heterogeneous Azeotropic Distillations: III. Column Sequences.
Chem. Eng. Sci. 1990, 45 (7), 1845−1854.
(20) Liu, G.; Jobson, M.; Smith, R.; Wahnschaftt, O. Shortcut Design
Method for Columns Separating Azeotropic Mixtures. Ind. Eng. Chem.
Res. 2004, 43 (14), 3908−3923.
(21) Tapp, M.; Holland, S. T.; Hildebrandt, D.; Glasser, D. Column
Profile Maps. 1. Derivation and Interpretation. Ind. Eng. Chem. Res.
2004, 43, 364−374.
(22) Gutierrez-Antonio, C.; Jimenez-Gutierrez, A. Method for the
Design of Azeotropic Distillation Columns. Ind. Eng. Chem. Res. 2007,
46 (20), 6635−6644.
(23) Phimister, J. R.; Seider, W. D. Semicontinuous, Pressure Swing
Distillation. Ind. Eng. Chem. Res. 2000, 39, 122−130.
(24) Modla, G.; Lang, P. Separation of a Ternary Homoazeotropic
Mixture by Pressure Swing Batch Distillation. Hung. J. Ind. Chem.
2008, 36, 89−94.
(25) Brusis, D.; Frey, Th.; Stichlmair, J.; Wagner, I.; Duessel, R.;
Kuppinger, F.-F. MINLP-Optimization of Several Process Structures
for the Separation of Azeotropic Ternary Mixtures. Comput.-Aided
Chem. Eng. 2000, 8, 109−114.
(26) Bauer, M. H.; Stichlmair, J. Design and Economic Optimization
of Azeotropic Distillation Processes Using Mixed-Integer Non-Linear
Programming. Comput. Chem. Eng. 1998, 20 (9), 1271−1286.
(27) Frey, Th.; Bauer, M. H.; Stichlmair, J. MINLP-Optimization of
Complex Column Configurations for Azeotropic Mixtures. Comput.
Chem. Eng. 1997, 21 (1), S217−S222.
(28) Li, C.; Zhang, X. Design of Separation Process of Azeotropic
Mixtures Based on the Green Chemical Principles. J. Clear Prod. 2007,
15 (7), 690−698.
(29) Caballero, J. A.; Grossmann, I. E. An Aggregated MINLP
Optimization Model for Synthesizing Azeotropic Distillation Systems.
Comput. Chem. Eng. 1999, 23 (1), S85−S88.
(30) Kossak, S.; Kraemer, K.; Marquardt, W. Efficient Optimization-
Based Design of Distillation Columns for Homogeneous Azeotropic
Mixtures. Ind. Eng. Chem. Res. 2006, 45 (25), 8492−8502.
(31) Kraemer, K.; Kossak, S.; Marquardt, W. Efficient Optimization-
Based Design of Distillation Processes for Homogeneous Azeotropic
Mixtures. Ind. Eng. Chem. Res. 2009, 48 (14), 6749−6764.
dx.doi.org/10.1021/ie300423q | Ind. Eng. Chem. Res. 2012, 51, 11183−11200
Industrial & Engineering Chemistry Research Article

(32) Linninger, A. A. Industry-wide Energy Saving by Complex

Separation Networks. Comput. Chem. Eng. 2009, 33 (12), 2018−2027.
(33) McCallum, B. R.; Lucia, A. Energy Targeting and Minimum
Energy Distillation Column Sequences. Comput. Chem. Eng. 2010, 34
(6), 931−942.
(34) Wilson, S.; Manousiouthakis, V. I. I DEAS Approach to Process
Network Synthesis: Application to Multi-component MEN. AIChE J.
2000, 46 (12), 2408−2416.
(35) Drake, J. E.; Manousiouthakis, V. I. IDEAS Approach to Process
Network Synthesis: Minimum Plate Area for Complex Distillation
Networks with Fixed Utility Cost. Ind. Eng. Chem. Res. 2002, 41, 4984.
(36) Drake, J. E.; Manousiouthakis, V. I. IDEAS Approach to Process
Network Synthesis: Minimum Utility Cost for Complex Distillation.
Chem. Eng. Sci. 2002, 57, 3095.
(37) Holiastos, K.; Manousiouthakis, V. I. Infinite-Dimensional State-
Space (IDEAS) Approach to Globally Optimal Design of Distillation
Networks Featuring Heat and Power Integration. Society 2004, 43
(24), 7826−7842.
(38) Martin, L. L.; Manousiouthakis, V. I. Globally Optimal Power
Cycle Synthesis via the Infinite Dimensional State Space (IDEAS)
Approach Featuring Minimum Area with Fixed Utility. Chem. Eng. Sci.
2003, 58 (18), 4291−4305.
(39) Zhou, W.; Manousiouthakis, V. I. Non-Ideal Reactor Network
Synthesis through IDEAS: Attainable Region Construction. Chem. Eng.
Sci. 2006, 61 (21), 6936−6945.
(40) Zhou, W.; Manousiouthakis, V. I. Variable Density Fluid
Reactor Network SynthesisConstruction of the Attainable Region
through the IDEAS Framework. Chem. Eng. J. 2007, 129 (1−3), 91−
203.
(41) Burri, J. F.; Manousiouthakis, V. I. Global Optimization of
Reactive Distillation Networks using IDEAS. Comput. Chem. Eng.
2004, 28, 2509−2521.
(42) Justanieah, A.; Manousiouthakis, V. I. IDEAS Approach to the
Synthesis of Globally Optimal Separation Networks: Application to
Chromium Recovery from Wastewater. Adv. Environ. Res. 2003, 7 (2),
547−562.
(43) Burri, J. F.; Wilson, S. D.; Manousiouthakis, V. I. Infinite-
Dimensional State-Space Approach to Reactor Network Synthesis:
Application to Attainable Region Construction. Comput. Chem. Eng.
2002, 26 (6), 849−862.
(44) Manousiouthakis, V. I.; Justanieah, A.; Taylor, L. A. The Shrink
Wrap Algorithm for the Construction of the Attainable Region: An
Application of the IDEAS Framework. Comput. Chem. Eng. 2004, 28
(9), 1563−1575.
(45) Zhou, W.; Manousiouthakis, V. I. On Dimensionality of
Attainable Region Construction for Isothermal Reactor Networks.
Comput. Chem. Eng. 2008, 32 (3), 439−450.
(46) Posada, A.; Manousiouthakis, V. I. Multi-Feed Attainable Region
Construction Using the Shrink-Wrap Algorithm. Chem. Eng. Sci. 2008,
63 (23), 5571−5592.
(47) Davis, B. J.; Taylor, L. A.; Manousiouthakis, V. I. Identification
of the Attainable Region for Batch Reactor Networks. Ind. Eng. Chem.
Res. 2008, 47 (10), 3388−3400.
(48) Gilliland, E. R.; Reed, C. E. Degrees of Freedom in
Multicomponent Absorption and Rectification Columns. Ind. Eng.
Chem. 1942, 34 (5), 551−557.
(49) Smith, J. M.; Van Ness, H. C.; Abbott, M. M. Introduction to
Chemical Engineering Thermodynamics, 7th ed.; The McGraw-Hill
Chemical Engineering Series; McGraw-Hill: New York, 2005; p 545.
(50) Gmehling, J.; Onken, U.; Rarey-Nies, J. R. Vapor−Liquid
Equilibrium Data Collection; Chemistry Data Series; DECHEMA:
Frankfurt am Main, Germany, 1981−1988; Vol. 1, Parts 1a, 1b, 2c,
and 2e.
(51) Poling, B. E.; Prausnitz, J. M.; O’Connell, J. P. The Properties of
Gases and Liquids, 5th ed.; McGraw-Hill: New York, 2001; Appendix A.
(52) Smith, J. M.; Van Ness, H. C.; Abbott, M. M. Introduction to
Chemical Engineering Thermodynamics, 7th ed.; The McGraw-Hill
Chemical Engineering Series; McGraw-Hill: New York, 2005; pp
339−340.

11200 dx.doi.org/10.1021/ie300423q | Ind. Eng. Chem. Res. 2012, 51, 11183−11200