Professional Documents
Culture Documents
To cite this version : Shen, Weifeng and Dong, Lichun and Wei,
Shun'an and Li, Jie and Benyounes, Hassiba and You, Xinqiang and
Gerbaud, Vincent Systematic design of an extractive distillation for
maximum-boiling azeotropes with heavy entrainers. (2015) AIChE
Journal, vol. 61 (n° 11). pp. 3898-3910. ISSN 0001-1541
DOI 10.1002/aic.14908
Extractive distillation is one of the most attractive approaches for separating azeotropic mixtures. Few contributions
have been reported to design an extractive distillation for separating maximum-boiling azeotropes and no systematic
approaches for entrainer screening have been presented. A systematic approach to design of two-column extractive dis-
tillation for separating azeotropes with heavy entrainers has been proposed. A thermodynamic feasibility analysis for
azeotropes with potential heavy entrainers was first conducted. Then, five important properties are selected for entrainer
evaluation. Fuzzy logic and develop membership functions to calculate attribute values of selected properties have been
used. An overall indicator for entrainer evaluation is proposed and a ranking list is generated. Finally, the top five
entrainers from the ranking list have been selected and use process optimization techniques to further evaluate selected
entrainers and generate an optimal design. The capability of the proposed method is illustrated using the separation of
acetone–chloroform azeotropes with five potential entrainers.
Figure 3. Thermodynamic features of 1.0–2 mixtures with respect to the separation of a maximum boiling azeo-
trope with a heavy entrainer.
Entrainer property analysis
As discussed before, the effectiveness of an extractive dis-
tillation process largely relies on an efficient (i.e., economic)
extractive entrainer. The performance of an entrainer is often
evaluated based on its relative volatility,13,39–41 solubility
power,13,39–41 boiling point,13 molecular weight, and melting
point.39 Inspired from these studies, we summarize that a
potential entrainer should be evaluated over properties in three
categories: (1) pure entrainer properties such as boiling point, Figure 4. A triangular membership function.
vapor pressure, molar volume, melting point, and critical prop-
erties; (2) process-related properties including relative volatil-
ity, solubility power, phase stability criterion,42 and (3) process flow sheet that the selectivity alone.13 In addition,
Sustainability-related criteria such as LC50, environmental from Eq. 3, it is concluded that a small value of the activity
waste, impact, health, and safety. From an economic point of coefficient c1
A;S indicating the strong interactions between
view, relative volatility and solubility power are the two most component A and an entrainer, can result in a large capacity or
important properties for the evaluation of an entrainer per- solubility power for the entrainer.
formance.13,39–41,43,44 The third important property is boiling
point.13,44 The fourth important properties are molecular Entrainer evaluation and ranking through fuzzy logic
weight and melting point.40 The molar volume of the (heavy) To implement the aforementioned properties, Kossack
entrainer is interesting mainly in the case of batch extractive et al.13 imposed some strict limits on each property to screen
distillation where it is accumulated in the reboiler and hence out undesired entrainers. To screen the entrainers whose prop-
the volume of the entrainer fed decreases the amount of binary erties satisfy all imposed limits, they are evaluated with the
charge to be processed in one batch.45 In continuous extractive so-called rectification body method the minimum energy
distillation, it is interesting only from the point of view of its demand of each related process. Then they used rigorous
storage. Therefore, we select boiling point, melting point, mixed-integer nonlinear programming (MINLP) to assess each
vapor pressure, and molecular weight from the first category. preliminary design performance with a rigorous MESH (Mass
In the second group, we select relative volatility and entrainer balance equations, Equilibrium equations, Summation of mole
power, which are defined in the following. Phase stability and fraction equations, and Heat or enthalpy balance equations)
sustainability-related criteria from the third group are not con- model. This requires a huge computational effort.
sidered in this article. In the above work, some good entrainers may be eliminated
The relative volatility, denoted as aA,B, is defined as the when are out of the bounds set on the property values. How-
ratio of vapor-liquid equilibrium (VLE) distribution constant. ever, bearing in mind that property prediction models are inac-
That is curate, some of those candidate should be kept within some
tolerance. To achieve this target, we apply a fuzzy logic to
KA yA xB cA P0A represent entrainer property specifications.
aA;B ¼ ¼ ¼ (1)
KB yB xA cB P0B Determining appropriate fuzzy membership functions is a
key issue in fuzzy logic. The membership function fully
where cA and cB denote activity coefficients of components A defines a fuzzy set and provides a measurement of the degree
and B in a liquid phase, respectively; P0A and P0B stand for satu- of similarity of an element (e.g., a property) to a fuzzy set.
rated vapor pressures of A and B, respectively. The membership function maps elements into numerical val-
Although aA;B varies with compositions,22 it is custom to ues in the interval [0, 1]. While a value of zero implies that the
use the infinite dilution ratio as follows element does not belong to the set, a value of unity means that
c1 0 the element fully belongs to the set.48 The membership func-
1 A;S PA
aA;B ¼ 1 (2) tion can be any form such as triangular, trapezoidal,
cB;S P0B
piecewise-linear, Gaussian, and bell-shaped, which can be
where c1 chosen by the user according to the characters of target objec-
A;S is the activity coefficient for a trace of species A in
an entrainer, and c1 tive or designed using machine learning methods (e.g., artifi-
B;S is the activity coefficient for a trace of
cial neural networks, genetic algorithms, etc.). Figure 4
species B in an entrainer. The fugacity of pure compound liq-
illustrates a typical triangular member function, which is
uid in a reference state was supposed to be equal to the vapor
described using Eq. 4 given below
pressure. Note that aA,B can also be used to estimate the selec-
tivity ratio of an entrainer.46 Therefore, a higher value of a1
8
A;B >
> 0 if x a
indicates that fewer entrainers are required in an extractive x2a
>
>
if a x b
>
>
process, which can reduce the energy in the extractive column
>
b2a
<
and the total recovery cost in the entrainer recovery column. lA ðxÞ ¼ (4)
c2x
if b x c
>
The solubility power criterion also called entrainer capacity >
>
> c2b
13
S40
>
p , is defined by Kossack et al. as follows
>
>
0 if x c
:
1 MWA
SP ¼ (3) where a, b, and c represent the x coordinates of three vertices
c1
A;S MWS
of lA(x) in a fuzzy set A. a is a lower boundary, and c is an
Jork et al.47 stated that an entrainer with a low capacity upper boundary where membership degree is zero. b repre-
often possesses a high selectivity. It is found that the entrainer sents the center value, where the membership degree is 1. Usu-
capacity is in a better agreement with the TAC of optimized ally for each degree a linguistic description is used, for
example, bad, acceptable, good, and very good. Next, we pro- Regarding the molecular weight of an entrainer; it is corre-
pose fuzzy membership functions for those five properties. lated with other physical properties such as density and boiling
For the relative volatility under infinite dilution, If a1
A;B < 1, point, which are known to increase along with carbon num-
the entrainer is not suitable because in the case of a maximum ber.52 Based on the work from Schwarz and Nieuwoudt53 and
azeotrope separation with a heavy entrainer we are concerned Takanohashi et al.,54 we set molecular weight upper limits as
with, it indicates that there exists a limit for the entrainer flow 180 g/mol for alkane series, and 240 g/mol for naphthenic and
rate to recover acetone as distillate.18,24 It is considered to be a aromatic hydrocarbons. The membership functions of molecu-
bad entrainer and l(a1 1
A;B ) 5 0. Conversely, for 1 < aA;B < 2, an lar weight are developed as follows
entrainer is treated as acceptable, and the function is given a gain MWs
of 0.3.40,49 The remaining values 2 < a1
(
A;B < 4 means that an 12 if MWs 180
entrainer is good, and the function is gained by an evaluation lðMWs Þ ¼ 180
(9)
value of 0.7.40,49 If a1
A;B > 4, the entrainer is considered as very 0 if MWs 180
good. The membership functions of relative volatility defined in ðfor chain hydrocarbonsÞ
Eq. 5
( MWs
if a1 12 if MWs 240
8
0 A;B 1 240
lðMWs Þ ¼
>
>
>
> 1 (10)
> 0:3ðaA;B 21Þ if 1 < a1 2
>
0 if MWs 240
>
>
> A;B
221
<
lða1 ðfor naphthenic and aromatic hydrocarbonsÞ
A;B Þ ¼ 1
(5)
> 0:7ðaA;B 22Þ if 2 < a1 4
>
>
> A;B Now, we create an entrainer evaluation set U with five
422
>
>
members U1, U2,. . ., U5, which are for relative volatility, solu-
>
>
if a1
>
A;B > 4
:1
bility power, boiling point, melting point, and molecular
weight, respectively. Each member describes the degrees of a
For the solubility power (Sp), an entrainer can be considered
as very good if Sp is greater than 0.5. When Sp is less than property performance for an entrainer used in an extractive
0.02, it indicates a poor capacity entrainer, and therefore, the distillation.
evaluation value would be 0.40 Similarly, the function is eval-
U ¼ U1 ; U2 ; U3 ; U4; U5 (11)
uated by different degrees of 0.3 and 0.7, when Sp is between
0.02–0.135 and 0.135–0.5, respectively. The membership Then, each property attribute value is weighted and the
function of solubility power is defined as follows weighted attribute values are summed together to develop a
total score indicator, Tscore, as given in Eq. 12
0 if 0 < Sp 0:02
8
TScore ¼ w1 lða1
>
A;B Þ1w2 lðSp Þ1w3 lðMWs Þ
>
>
0:3ðSp 20:02Þ
>
(12)
>
>
>
> if 0:02 < Sp 0:135 1w4 lðTb Þ1w5 lðTm Þ
0:13520:02
<
lðSp Þ ¼ (6)
> 0:7ðSp 20:135Þ where w1, w2, w3, w4, and w5 denote weight factors for relative
if 0:135 < Sp 0:5
>
>
>
>
>
>
> 0:520:135 volatility, solubility power, molecular weight, boiling point,
1 if Sp > 0:5 and melting point, respectively. A higher total score (Tscore)
:
for an entrainer represents better performance.
To facilitate the recycling of a heavy entrainer from compo-
The weight factors assigned to each property should be cho-
nents with highest boiling point in the regeneration column, its
sen to suit the specific problem and evidently affect the final
boiling point is usually 30–150 K higher than that of any com-
solution.55 Based on studies in the literature13,38,39,43,55,56 and
ponent in the mixture to reduce more energy in vaporization
our own experience, we conclude that the relative volatility,
of a boiling component.49,50 The following membership func-
and solubility power are more important than the other proper-
tion of boiling point is established in Eq. 7
ties, and boiling point is more important than molecular
if Tb Tbmax 130
8
> 0 weight, and melting point. Therefore, the weight factors for
>
>
<
Tbmax 11502Tb those five properties should satisfy the following relationship,
max
lðTb Þ ¼ if Tbmax 130 < Tb Tb 1150
>
> 150230 w1 > w4 (13a)
>
0 if Tb Tbmax 1150
:
w2 > w4 (13b)
(7) w4 > w3 (13c)
where, Tbmaxis the highest boiling point among components in w4 > w5 (13d)
the mixture to be separated. w1 1w2 1w3 1w4 1w5 ¼ 1 (14)
Regarding the entrainer melting point, it is better for an
entrainer not to freeze at room temperature, which is taken
equal to 303 K. The melting point of an entrainer is considered Process optimization
to be excellent if Tm 273 K.51 The following membership In the literature, optimal solutions for extractive distillation
function of melting point is established in Eq. 8 are first sought using sensitivity analysis on process variables,
8 including reflux ratio, ratio of entrainer flow rate to feed flow
1 if Tm 273
>
>
> rate, distillate flow rate, and the number of trays in rectifying,
3032x extractive, and stripping sections.13,14,57,58 However, the sensi-
<
lðTm Þ ¼ if 273 Tm 303 (8)
>
>
> 3032273 tivity analysis approach may not consider effects arising from
0 if Tm 303 correlated variables, and hence, is likely to fail to obtain the
Table 1. Parameters for TAC Calculation Our objective is to minimize process energy consumption in
the reboiler and condenser of both the extractive and regenera-
Parameters Values
tion columns, which is the same as that of You et al.15 We use
Condensers OF to represent process energy consumption, which is calcu-
Heat transfer coefficient 0.852 kW/(K m2)
Differential temperature 28.8 K lated as follows
Capital cost 9367.8 (area in m2)0.65 Qr1 1m Qc1 1Qr2 1m Qc2
Reboilers OF ¼ (22)
Heat transfer coefficient 0.568 kW/(K m ) 2 FD1 1k FD2
Differential temperature 34.8K
Capital cost 9367.8 (area in m2)0.65 where k represents product price factor; m represents energy
Column vessel capital cost ((M&S)/280)101.9D1.066L0.802 price difference factor for water in the condenser and heating
Tray capital cost ((M&S)/280)4.7D1.55 L fluid in the reboiler. Note that some other objective functions
Energy cost 7920 (13.65*Qr10.4895*Qc)
Payback period 3 years
in the literature could also be used such as TAC,1,60 and heat
duty ratio.57
The complete process optimization model named as CPOM
optimal solution. Optimization techniques like linear program- is presented as follows
ming, nonlinear programing (NLP), mixed-integer linear pro-
Qr1 1m Qc1 1Qr2 1m Qc2
gramming, and MINLP have been developed to overcome this ðCPOMÞ Minimize OF ¼
drawback.59 FD1 1k FD2
Before we present our proposed formulation, we first define subject to: Eqs. 15–21 and other constraints implemented in
all variables in the model. The variables for the extractive col- Aspen plusV process simulator
R
umn are defined as follows: the entrainer flow rate is noted as We select top five entrainers from the ranking list. For each
FE, the reflux ratio is defined as R1, xA,D1 as the purity of dis- selected entrainer, the entire process optimization procedure
tillate A, xB,W1 as the bottom impurity of B in the extractive can be described as follows, First, Aspen plus simulator built-
column, total number of stages is represented by Next, mixture in SQP method is used to solve model CPOM under a given
feed stage is NFF, entrainer feed stage is denoted as NFE, Qr1 column structure. The process is run with the RadFrac model
and Qc1 are defined as heat duty of the reboiler and condenser, in Aspen Plus in which the original UNIversal Functional
respectively. The variables for the entrainer regeneration col- Activity Coefficient (UNIFAC) model is used to calculate the
umn are defined as follows: the reflux ratio is denoted as R2, activity coefficient in VLE. Second, a sensitivity analysis is
total number of stages is defined as Nreg, entrainer feed stage performed to find optimal discrete values of the feed tray loca-
is NR, xB,D2 and xE,W2 are used to denote the purity of distillate tions and total number of stages, while SQP is ran for each set
B and bottom purity of entrainer E in the entrainer regenera- of discrete variable values within bounds:
tion column, Qr2 and Qc2 represent heat duty of the regenera-
tion column, respectively. First of all, the purity of distillate A min Next Next max Next (23)
in the extractive column must satisfy some minimum purity min NFF NFF max NFF (24)
requirement (denoted as minA_D1). min NFE NFE max NFE (25)
xA ;D1 minA D1 (15) min Nreg Nreg max Nreg (26)
Similarly, the bottom impurity of component B must not min NRF NRF max NRF (27)
exceed some requirement (maxB_W1)
The final optimization is found through minimizing OF
xB;W1 maxB W1 (16) value. Finally, the TAC is calculated to compare the separa-
Similar constraints can be written in the entrainer regenera- tion sequences and evaluate selected entrainers. The TAC
tion column as follows, based on Douglas’ cost formulas61 is defined as follows15,62
acetone–chloroform (A–B) system with any of these five the entrainer feed generates an extractive separatrix that is
entrainers (E1–E5) gives rise to a ternary mixture A–B–E, also curved and closer to the B–E side than the RCM stable
which belongs to 1.0-2 class. separatrix, thus increasing the feasible region in which A is
the unstable node. In a brief, the extractive distillation
Thermodynamic feasibility insight improves the recovery yield of component A as the first distil-
Following the proposed design method in Figure 2, we first late cut.18,19,27,28
conduct thermodynamic insight analysis for the process feasi-
bility combining the analysis of residue curve maps and vola-
tility order (i.e., 1.2).
Results from entrainer property analysis
The azeotropic behaviors of the acetone–chloroform (A–B) Physicochemical properties of acetone–chloroform (A–B)
mixture with heavy entrainer DMSO (E1), chlorobenzene (E2), mixture such as molecular weight (MWs), boiling point (Tb),
EG (E3), o-xylene (E4), or benzene (E5) are shown in Table 2. and melting point (Tm) are given in Table 3. The saturated
From Table 2, it can be observed that ternary mixtures A–B–E vapor pressure for acetone–chloroform azeotrope is calculated
demonstrate the same singular properties and they all belong using the correlation given below
to the same 1.02 Serafimov’s class with a reported occurrence lnP ¼ a1b=T1c lnT1d T e (29)
of 8.5%.35
Figure 5 depicts the thermodynamic features for the ternary where the values for T, a, b, c, d, and e are presented in Table
mixture acetone–chloroform–chlorobenzene. Similar princi- 3. The results for saturated vapor pressure are also given in
ples can also be generated for the rest ternary mixtures includ- Table 3. From Table 3, it can be observed that acetone and
ing acetone–chloroform–DMSO, acetone–chloroform–EG, chloroform have very close saturated vapor pressure and boil-
acetone–chloroform–Benzene, and acetone–chloroform–o- ing point.
xylene. In Figure 5, the univolatility curve aAB 5 1 reaches the The original UNIFAC model is used to compute activity
binary side B–E when using these heavy entrainers. Based on coefficients of VLE evaluate the five thermodynamic proper-
the feasibility analysis presented in Figure 3, both acetone and ties at atmospheric pressure that we selected. The results are
chloroform can be recovered as products.18,24 For each of the presented in Table 4. From Table 4, it is observed that the five
selected heavy entrainers, the univolatility curve aAB 5 1 entrainers have rather different values of these properties
reaches the entrainer–chloroform edge. Therefore, there is no which will translate into different total score that are now
maximum entrainer flow rate to recover acetone which is our discussed.
goal.
Laroche et al.23 acknowledged that the acetone–chloroform
separation could be performed by azeotropic distillation as
Results from entrainer evaluation and ranking through
well in which the entrainer is directly added to the mixture fuzzy logic
because acetone and chloroform are both unstable nodes of the We use the proposed membership functions (i.e., Eqs. 5–10)
residue curve map. Lang et al.27,28 showed that it is better to to calculate property attribute values and total scores for E1–
use extractive distillation for the separation of acetone–chloro- E5. The weighting factor values assigned for E1–E5 from Eqs.
form (A–B) mixture rather than azeotropic distillation because 13 and 14 are given as follows,
1
W ¼ w1 ; w2 ; w3; w4; w5 ¼ f0:4; 0:35; 0:05; 0:15; 0:05g tive volatility (aA;B ) and gets by far the highest attribute value
1
(30) l(aA;B ) for that property which reaches 0.9606. Conversely, the
attribute value for the solubility power (SP) and the boiling
Note that the assignment in Eq. 30 reflects that relative vol- points are the smallest. Hence, the total score for EG (E3)
atility, solubility power, and boiling point are considered as reaches 0.5556. It remains the highest value of all five
the main selection criteria to screening entrainers based on entrainers because of the highest weight factor is assigned for
computer-aided molecular design (CAMD) approaches for an relative volatility in total score calculation. For DMSO (E1)
extractive distillation process.13,40,46 1
that has the second best attribute value l(aA;B ) and the best
To validate the rationality of assigned weighting factor val- attribute value l(SP), the total score reaches 0.5229 because
ues that should meet the requirement in Eqs. 13 and 14, a sen- the attribute values for boiling point and melting points are
sitive study is conducted using five scenarios with different low. Finally, we observe that the lowest total score is 0.3621
assigned weight factors. Table 5 gives the weighting factor for benzene, closely followed by 0.3918 for chlorobenzene
values for each scenario. Consequently, the total scores of E1– due to the fact that benzene boiling point is too low and the
E5 for each scenario are obtained in Table 6. Although we
relative volatility attribute is low whereas chlorobenzene has a
change the assigned weighting factor values, resulting in a lit-
very low attribute value for the relative volatility. The total
tle bit change of total scores for E1–E5, the same ranking is
scores of E1–E5 are ranked in the following order:
obtained. In other words, if the assigned weight factors change
EG > DMSO > o-xylene > chlorobenzene > benzene.
a little bit and satisfy the conditions in Eqs. 13 and 14, the
ranking for E1–E5 will not change. The possible reason is that
Process optimization
the attribute values for those five properties of E1–E5 are dis-
tributed between 0 and 1, which are small values. The total In this part, we run an optimization based on the total
scores for E1–E5 do not change significantly with those small energy consumption per distillate product of the process
attribute values when changing assigned weighting factor val- including both the extractive distillation column and the
ues within a reasonable region restricted by Eqs. 13 and 14. entrainer regeneration column for each five entrainers. We
The computational results are presented in Table 7. As it rank them according to the TAC of each process and compare
can be seen from Tables 4 and 7, EG (E3) has the highest rela- them the ranking proposed by the fuzzy logic-based total
score. The operating conditions are kept constant for all
entrainers and are given in Table 8. Starting from a 100 kmol/
Table 5. The Values of Weight Factors for E1–E5 in Five h equimolar feed acetone–chloroform, the distillate flow rate
Scenarios form both columns are fixed to 50 kmol/h with at least 0.995
Cases W1 W2 W3 W4 W5 purity. Both columns operate at 1.1 atm with no pressure drop.
Feed temperature for the extractive column is set equal to
1 0.40 0.35 0.05 0.15 0.05
2 0.30 0.30 0.15 0.15 0.10 320 K, whereas the regeneration column is fed at boiling point
3 0.30 0.40 0.05 0.15 0.10 by the bottom stream of the extractive column. The lower and
4 0.40 0.40 0.00 0.15 0.05 upper bounds of the operating variables are estimated using
5 0.30 0.35 0.10 0.15 0.10 the sensitivity analysis, and the results are given in Table 9.
reflux ratio (0.203 for extractive column and 0.369 for regen-
Table 8. Operating Conditions for Process Optimization
eration column). This is typically related to the fact that the
Extractive Regeneration relative volatility of EG is the largest one among E1–E5 (Table
Specifications Column Column 4). However, the entrainer feed flow rate is the lowest for
Feed flow rate (kmol/h) 100 – DMSO (i.e., 164.9 kmol/h) followed by the value with EG
Distillate flow rate (kmol/h) 50 50 (i.e., 221.5 kmol/h). The possible reason is that DMSO has a
Column pressure (atm) 1.1 1.1 much higher solution power value than others, which leads to
Feed temperature (K) 320 320 a lower feed flow rate of DMSO. From Table 10, it can also be
Main feed composition –
xA, mole fraction 0.5 observed that the duties for condensers and reboilers in the
xB, mole fraction 0.5 extractive distillation column and in the regeneration column
xE, mole fraction 0 – are slightly lower for EG than for DMSO and are much
Entrainer composition – smaller than for the other entrainers (i.e., E2, E4–E5). These
xA, mole fraction 0
xB, mole fraction 0
results are correlated to the reflux ratio values and the
xE, mole fraction 1 – entrainer feed flow rate: DMSO with the lowest entrainer flow
Controlled variables constraint rate requires almost the same boiler duty with that of EG
Top product 0.995 acetone 0.995 chloroform because the separation requires a higher reflux ratio for
Bottom product 0.001 acetone 0.9999 entrainer DMSO than for EG.
Similarly the heat duty for E2 (chlorobenzene) is very large
During optimization process, it is observed that for each because the E2 feed flow rate of 1659.2 kmol/h to the extrac-
case, there is a nonmonotonic effect on each manipulated vari- tive column entrainer flow rate is by far the greatest of all
able as was acknowledged by many authors.1,12 For instance, entrainers and a large reflux ratio of 9.803 for the regeneration
R1, R2, and entrainer feed flow rate will be affected by the feed
stages NFF, NFE, and NRF. For a given entrainer flow rate, there
is an optimal reflux ratio R1 that yields the maximum acetone Table 9. Lower and Upper Bounds of Operating Variables
purity. The higher the entrainer flow rate, the higher the ace- During Process Optimization
tone purity can be obtained but then the OF and TAC figures
Operating Variables Lower Limit Upper Limit
also increase. The impurities of chloroform and entrainer are
affected by the entrainer flow rate and reflux ratio R1. Increas- Extractive column
Total number of stage, Next 15 50
ing the entrainer feed flow rate decreases the chloroform Feed tray, NFF 9 50
impurity but increases the entrainer impurity. Entrainer feed tray, NFE 3 25
The final optimization results for E1–E5 are given in Table Reflux ratio, R1 0.2 15
10. Figure 6 explicitly illustrates total scores and optimal TAC Entrainer feed (kmol/h),FE 40 350
Regeneration column
values for E1–E5.
Total number of stage, Nreg 10 54
From Table 10 and Figure 6, we observe that the process Feed tray, NRF 3 30
TAC for the separation of the acetone–chloroform mixture Reflux ratio, R2 0.3 30
with E1–E5 are ranked as E3 (EG) < E1 (DMSO) < E4 (o-
xylene) < E2 (chlorobenzene) < E5 (benzene). Furthermore,
this is consistent with the total score ranking based on thermo-
dynamic properties of the entrainers and mixture A–B.
The TAC value of $630,000 using E1 (DMSO) and
$610,000 using E3 (EG) is much smaller that the TAC for the
three other entrainer candidates. The process with DMSO
supremacy verifies the recommendation of Luyben to choose
that entrainer to achieve 0.995 pure products with the same
main feed and distillate flow rates.32 The DMSO design fig-
ures are very close to those reported by Luyben32 thus validat-
ing our design.
It is worth mentioning that Luyben1 concluded that EG,
chlorobenzene, and o-xylene are not suitable to separate the
acetone–chloroform mixture with the desired separation speci-
fications. From Table 10, we show that all entrainers can
achieve the separation with the purity targeted by Luyben.1 E3 Figure 6. Total scores and TAC values for E1-E5.
(EG) is a very good candidate as the process TAC is similar to [Color figure can be viewed in the online issue, which is
DMSO’s one. Indeed, EG (E3) allows one to use the lowest available at wileyonlinelibrary.com.]
Table 10. Comparison of Optimal Design Results with E1–E5
DMSO Chlorobenzene EG o-Xylene Benzene
E1 E2 E3 E4 E5
Extractive column
Total number of stage, Next 23 45 30 37 44
Feed tray, NFF 11 14 9 17 21
Entrainer feed tray, NFE 4 9 3 5 20
Reflux ratio, R1 0.899 5.350 0.203 1.225 10.247
Entrainer feed (kmol/h), FE 164.9 1659.2 221.5 354.9 270.7
Ar (m2) 78.4 458.6 72.8 141.1 255.99
Ac (m2) 65.5 105.8 22.6 76.8 187.06
Column diameter (m) 1.01 3.09 0.869 1.51 2.09
Tray capital cost(106$) 0.018 0.209 0.019 0.056 0.112
Vessel capital cost(106$) 0.169 1.010 0.183 0.398 0.653
Condenser and reboiler cost (106$) 0.302 0.697 0.223 0.391 0.625
Condenser duty(MW) 0.776 2.596 0.497 0.910 4.59
Reboiler duty(MW) 1.550 9.065 1.438 2.789 5.06
Regeneration column
Total number of stage, Nreg 12 33 10 16 43
Feed tray, NRF 5 8 4 6 18
Reflux ratio, R2 0.756 9.803 0.369 2.413 26.100
Ar (m2) 56.7 220.2 64.9 88.9 559.03
Ac (m2) 61.0 138.8 21.5 117.6 381.54
Column diameter (m) 0.83 2.35 0.780 1.34 3.16
Tray capital cost(106$) 0.006 0.098 0.004 0.018 0.207
Vessel capital cost(106$) 0.072 0.575 0.056 0.162 0.995
Condenser and reboiler cost (106 $) 0.265 0.544 0.210 0.381 1.018
Condenser duty(MW) 0.723 4.010 0.565 1.393 11.02
Reboiler duty(MW) 1.12 4.353 1.283 1.757 11.05
Heat exchanger, HX1
Exchanger duty (MW) 1.080 6.720 1.583 2.156 0.399
TAC (106$/year) 0.63 2.64 0.61 1.07 3.02
column is needed to achieve the target product specifications. separation of acetone–chloroform system. The optimal design
This is because the acetone–chlorobenzene system does not of the two-column extractive distillation process for DMSO is
have a large relative volatility (i.e., 1.2527 in Table 4), which similar to the one published by Luyben.24 The one with EG
leads to difficulty in attaining the desired acetone purity. (E3) is illustrated in Figure 7. The liquid composition and tem-
When E5 (benzene) is used as the entrainer, a large reflux perature profiles in the extractive column and the regeneration
ratio (i.e., 26.1) of the regeneration column is needed as the column with EG (E3) are given in Figure 8. It is noticed that
boiling point of benzene is closest to that of chloroform among the goal of the extractive section is fulfilled in preventing the
E1–E5. Notice that for benzene the entrainer feed and main entrainer EG (E3) from going significantly into the rectifying
feed trays locations are one tray apart. Typically, this indicates section. With high attribute value of entrainer evaluated
that the azeotropic distillation process, also feasible, is likely a through fuzzy logic, only very small value of reflux ratios are
better design than the extractive distillation process. The required for both extractive and regeneration columns. From
inherent problem of using benzene as the entrainer should be Figures 8a, c, the flat composition profiles cannot be gener-
related to the ease of attaining higher purity levels in extrac- ated. Hence, a large number of stages for each section are
tive distillation with DMSO as entrainer with only modest required to achieve the specified targets.
increases in R and entrainer flow.
Based on the above discussion, one can conclude that Conclusions
DMSO (E1) and EG (E3) are the two best entrainers for the
In this article, we proposed a systematic approach to the
design of an extractive distillation for the separation of
maximum-boiling azeotropes with heavy entrainers, consider-
ing also the entrainer regeneration column. The proposed
method was illustrated using the separation of acetone/chlo-
roform azeotropes with five heavy entrainers (DMSO,
chlorobenzene, EG, o-xylene, and benzene). The thermody-
namic insights suggested that (1) acetone–chloroform (A–B)
mixture with all those heavy entrainers E belong to the same
1.0-2 Serafimov’s class ternary diagram; (2) acetone is chosen
as a distillate product with a larger feasible region than that of
chloroform as the extractive separatrix is curved and closer to
the B–E side than the RCM stable separatrix. (iii) a prescreen-
ing of the entrainer performance can be made and the ranking
obtained is confirmed by the optimization of the full process.
Figure 7. Optimal design of steady-state two-column The prescreening uses fuzzy logic attribute values related to
extractive distillation process using EG. properties: the relative volatility under infinite dilution, the
Figure 8. Liquid composition and temperature profiles for acetone-chloroform mixture with EG (E3).
(a) Liquid composition profiles in the extractive column; (b) Temperature profile in the extractive column; (c) Liquid composition
profiles in the regeneration column; (d) Temperature profiles in the regeneration column. [Color figure can be viewed in the online
issue, which is available at wileyonlinelibrary.com.]
solubility power of the entrainer vs. the product, the entrainer Notation
molecular weight and boiling and melting points. The opti- Variables
mized results are discussed in relation with the properties to
explain the correlation between the entrainer prescreening FE = entrainer flow rate, kmol/h
ranking and the key operating parameters that correspond to K= vapor-liquid equilibrium distribution constant
MW = molecular weight, g/mol
an optimal design. Next = total number of stages of extractive column
Both the prescreening ranking and the optimized process NFF = mixture feed stage in extractive column
results demonstrated that EG and DMSO are the best NFE = entrainer feed stage in extractive column
entrainers for the separation of acetone–chloroform system Nreg = total number of stages of regeneration column
NRF = feed stage in regeneration column
using extractive distillation followed by o-xylene, chloroben- P= pressure, kPa
zene, and benzene. Finally, optimal design of steady-state P8 = saturated vapor pressure, kPa
two-column extractive distillation process using EG are Qc = heat duty of the condenser, MW
presented. Qr = heat duty of the reboiler, MW
R= reflux ratio
The proposed method can incorporate additional properties SP = solubility power
such as phase stability and sustainability-related properties for Tb = boiling temperature, K
entrainer screening. It can also be extended to screening more Tscore = total score indicator
potential entrainer candidates for other ternary Serafimov’s Tm = melting temperature, K
U= entrainer evaluation set
classes using CAMD. W= weight factor
k= product price factor
m= energy price difference factor
x= mole fraction in liquid phase
Acknowledgments xA,D = distillate purity of A, mole fraction
xB,W = bottom impurity of B, mole fraction
This work is financially supported by the National Natural xB,D = distillate purity of B, mole fraction
Science Foundation of China (21206174, 21176270). xE,W = bottom purity of entrainer, mole fraction
y= mole fraction in vapor phase 11. Lee FM, Pahl RH. Entrainer screening study and conceptual extrac-
a= relative volatility tive distillation process to produce anhydrous ethanol from fermenta-
c= activity coefficient tion broth. Ind Eng Chem Process Des Dev. 1985;24(1):168–172.
c1 = activity coefficient at infinite dilution 12. Knapp JP, Doherty MF. Minimum entrainer flow for extractive dis-
lA (X) = membership function tillation: a bifurcation theoretic approach. AIChE J. 1994;40(2):243–
268.
Abbreviations 13. Kossack S, Kraemer K, Gani R, Marquardt W. A systematic synthe-
A= lower boiling point component in azeotropic mixtures sis framework for extractive distillation processes. Chem Eng Res.
B= higher boiling point component in azeotropic mixtures 2008;86(7):781–792.
CAMD = computer-aided molecular design 14. Luyben W. Effect of entrainer on controllability in extractive distil-
CPOM = complete process optimization model lation. Ind Chem Eng Res. 2008;47(13):4425–4439.
DMSO = dimethylsulfoxid 15. You X, Rodriguez-Donis I, Gerbaud V. Improved design and effi-
E= entrainer ciency of the extractive distillation process for acetone–methanol
EG = ethylene glycol with water. Ind Eng Chem Res. 2015;54:491–501.
HX1 = heater exchanger for entrainer recycle stream 16. Wahnschafft OM, Westerberg AW. The product composition regions
MILNP = mixed-integer nonlinear programming of azeotropic distillation columns: II. Separability in two-feed col-
NLP = nonlinear programing umns and entrainer selection. Ind Eng Chem Res. 1993;32(6):1108–
OF = objective function 1120.
SQP = Sequential Quadratic Programming 17. Thomas B, Karl Hans S. Knowledge integrating system for the selec-
TAC = total annualized cost tion of entrainers for extractive and azeotropic distillation. Comput
maxB_W1 = maximum impurity requirement of B at bottom of extrac- Chem Eng. 1994;18(Suppl.1):25–29.
tive column 18. Rodrıguez-Donis I, Gerbaud V, Joulia X. Thermodynamic insights
max_FE = the upper limit of entrainer flow rate, kmol/h on the feasibility of homogeneous batch extractive distillation. 1.
max_Next = the upper limit of total number of stages of extractive Azeotropic mixtures with heavy entrainer. Ind Chem Eng Res. 2009;
column 48(7):3544–3559.
max_NFF = the upper limit of mixture feed stage 19. Rodrıguez-Donis I, Gerbaud V, Joulia X. Thermodynamic insights
max_NFE = the upper limit of entrainer feed stage on the feasibility of homogeneous batch extractive distillation, 2.
max_Nreg = the upper limit of total number of stages of regeneration Low-relative-volatility binary mixtures with a heavy entrainer. Ind
column Chem Eng Res. 2009;48(7):3560–3572.
max_NRF = the upper limit of feed stage in regeneration column 20. Rodrıguez-Donis I, Gerbaud V, Joulia X. Thermodynamic insight on
min_FE = the lower limit of entrainer flow rate, kmol/h extractive distillation with entrainer forming new azeotropes. In:
min_Next = the lower limit of total number of stages of extractive Proceedings of Distillation & Absorption, Eindhoven, Holland, Sep-
column tember 12–15, 2010;431–436.
min_NFF = the lower limit of mixture feed stage 21. Rodrıguez-Donis I, erbaud V, Joulia X. Thermodynamic insights on
min_NFE = the lower limit of entrainer feed stage the feasibility of homogeneous batch extractive distillation, 3. Azeo-
min_Nreg = the lower limit of total number of stages of regeneration
tropic mixtures with light boiling entrainer. Ind Chem Eng Res.
column
2012;51(2):4643–4660.
min_NRF = the lower limit of feed stage in regeneration column
22. Rodrıguez-Donis I, Gerbaud V, Joulia X. Thermodynamic insights
minA_D1 = minimum purity requirement of A at top of extractive
on the feasibility of homogeneous batch extractive distillation, 4.
column
minB_D2 = minimum purity requirement of B at top of regeneration Azeotropic mixtures with intermediate boiling entrainer. Ind Chem
column Eng Res. 2012;51(18):6489–6501.
minE_W2 = minimum purity requirement of E at top of regeneration 23. Laroche L, Bekiaris N, Andersen HW, Morari M. Homogeneous aze-
column otropic distillation: comparing entrainers. Can J Chem Eng. 1991;69:
RCM = residu curve map 1302–1319.
VLE = vapor-liquid equilibrium 24. Shen WF, Benyounes H, Gerbaud V. Extension of thermodynamic
insights on batch extractive distillation to continuous operation. 1.
Azeotropic mixtures with a heavy entrainer. Ind Eng Chem Res.
Literature Cited 2013;52:4606–4622.
1. Luyben WL, Chien IL. Design and Control of Distillation Systems 25. Shen WF, Gerbaud V. Extension of thermodynamic insights on
for Separating Azeotropes, London: Wiley Online Library, 2010. batch extractive distillation to continuous operation. 2. Azeotropic
2. Bernot C, Doherty MF, Malone MF. Patterns of composition change mixtures with a light entrainer. Ind Eng Chem Res. 2013;52:4623–
in multicomponent batch distillation. Chem Eng Sci. 1990;45(5): 4637.
1207–1221. 26. Gerbaud V, Rodriguez-Donis I. Azeotropic distillation, In: G orak A,
3. Knapp JP, Doherty MF. A new pressure-swing-distillation process Olujic Z, editors. Distillation Book, vol. II, Chapter 6, Amsterdam:
for separating homogeneous azeotropic mixtures. Ind Chem Eng Res. Elsevier, 2014:201–245.
1992;31(1):346–357. 27. Lang P, Modla G, Benadda B, Lelkes Z. Homoazeotropic distillation
4. Phimister JR, Seider WD. Semicontinuous, pressure-swing distilla- of maximum azeotropes in a batch rectifying column with continu-
tion. Ind Chem Eng Res. 2000;39(1):122–130. ous entrainer feeding. I. Feasibility studies. Comput Chem Eng.
5. Modla G, Lang P. Feasibility of new pressure-swing batch distilla- 2000;24:1665–1671.
tion methods. Chem Eng Sci. 2008;63(11):2856–2874. 28. Lang P, Modla G, Kotai B, Lelkes Z, Moszkowicz P. Homoazeo-
6. Luyben W. Comparison of extractive distillation and pressure-swing tropic distillation of maximum azeotropes in a batch rectifying col-
distillation for acetone/chloroform separation. Comput Chem Eng. umn with continuous entrainer feeding. II. Rigorous simulation
2013;50:1–7. results. Comput Chem Eng. 2000;24:1429–1435.
7. Doherty MF, Malone MF. Conceptual Design of Distillation Sys- 29. Van Dongen DB, Doherty MF. On the dynamics of distillation proc-
tems, New York: McGraw-Hill, 2001. esses VI. Batch distillation. Chem Eng Sci. 1985;40:2087–2093.
8. Lladosa E, Monton J B, Cruz Burguet M. Separation of di-npropyl 30. Bernot C, Doherty MF, Malone MF. Feasibility and separation
ether and n-propyl alcohol by extractive distillation and pressure sequencing in multicomponent batch distillation. Chem Eng Sci.
swing distillation: computer simulation and economic optimization. 1991;46:1311–1326.
Chem Eng Proc. 2011;50:126621274. 31. Dussel R, Stichlmair J. Separation of azeotropic mixtures by batch
9. Modla G, Lang P. Removal and recovery of organic solvents from distillation using an entrainer. Comput Chem Eng. 1995;19:S113–
aqueous waste mixtures by extractive and pressure swing distillation. S118.
Ind Eng Chem Res. 2012;51:11473211481 32. Luyben W. Control of the Maximum-boiling acetone/chloroform
10. Munoz R, Monton JB, Burguet MC, De la Torre J. Separation of azeotropic distillation system. Ind Chem Eng Res. 2008;47(16):
isobutyl alcohol and isobutyl acetate by extractive distillation and 6140–6149.
pressure-swing distillation: simulation and optimization. Sep Purif 33. Seader JD, Henley, Ernest J, Seader JD. Separation Process Princi-
Technol. 2006;50:1752183. ples, New York: Wiley, 1998.
34. Gerbaud V, Rodriguez-Donis I, Extractive Distillation. Distillation: 50. Rodrıguez A, Serrano L, Moral A, Perez A, Jimenez L. Use of high-
Equipment and Processes, Chapter 6, Oxford: Elsevier, 2014;02: boiling point organic solvents for pulping oil palm empty fruit
201–246 bunches. Bioresour Technol. 2008;99(6):1743–1749.
35. Hilmen E, Kiva V, Skogestad S. Topology of ternary VLE diagrams: 51. Smallwood IM. Front matter. Handbook of Organic Solvent Proper-
elementary cells. AIChE J. 2003;48:752–759. ties. Butterworth-Heinemann: Oxford, 1996.
36. Shen WF, Benyounes H, Gerbaud V. A Review of extractive distilla- 52. Katritzky AR, Mu L, Lobanov V. Correlation of boiling points with
tion advanced operation polices and strategies. Rev Chem Eng.
molecular structure. 1. A training set of 298 diverse organics and a
2015;31(1):13–26.
test set of 9 simple inorganics. J Phys Chem. 1996;100(24):10400–
37. Laroche L, Bekiaris N, Andersen HW, Morari M. The curious
10407.
behavior of homogeneous azeotropic distillation – implications for
entrainer selection. AIChE J. 1992;38:1309–1328. 53. Schwarz CE, Nieuwoudt I. Phase equilibrium of propane and alka-
38. Ostergaad JJ. Fuzzy Logic Control of a Heat Exchange Process. nes: Part I. Experimental procedures, dotriacontane equilibrium and
Fuzzy Automata and Decision Processes, Amsterdam: North-Hol- EOS modelling. J Supercrit Fluids. 2003;27(2):133–144.
land, 1977. 54. Takanohashi T, Iino M, Nakamura K. Evaluation of association of
39. Tiverios PG, Van Brunt V, Extractive distillation solvent characteri- solvent-soluble molecules of bituminous coal by computer simula-
zation and shortcut design procedure for methylcyclohexane2toluene tion. Energy Fuels. 1994;8(2):395–398.
mixtures. Ind Eng Chem Res. 2000;39(6):1614–1623. 55. Van Dyk B, Nieuwoudt I. Design of entrainers for extractive distilla-
40. Song J, Song H-H, Computer-aided molecular design of environmen- tion. Ind Eng Chem Res. 2000;39(5):1423–1429.
tally friendly entrainers for separation processes. Chem Eng Technol. 56. Hostrup M, Harper PM, Gani R. Design of environmentally benign
2008;31:177–187. processes: integration of solvent design and separation process syn-
41. Momoh SO. Assessing the accuracy of selectivity as a basis for sol- thesis. Comput Chem Eng. 1999;23(10):1395–1414.
vent screening in extractive distillation processes. Sep Sci Technol.
57. De Figueir^edo MF, Guedes BP, De Araujo JMM, Vasconcelos LGS,
1991;26(5):729–742.
42. Pretel EJ, Lopez A, Bottini SB, Esteban A. Computer-aided molecu- Brito RP. Optimal design of extractive distillation column- A sys-
lar design of entrainers for separation processes. AIChE J. 1994; tematic procedure using a process simulator. Chem Eng Res Des.
40(8):1349–1360. 2011;89(3):341–346.
43. Odele O, Macchietto S. Computer aided molecular design: a novel 58. Langston P, Hilal N, Shingfield S. Simulation and optimization of
method for optimal solvent selection. Fluid Phase Equilib. 1993;82: extractive distillation with water as entrainer. Chem Eng Process:
47–54. Process Intensif. 2005;44(3):345–351.
44. Newhouse T, Baran PS, Hoffmann RW. The economies of synthesis. 59. Kossack S, Kraemer K, Gani R, Marquardt W. A systematic synthe-
Chem Soc Rev. 2009;38(11):3010–3021. sis framework for extractive distillation processes. In: Proceedings
45. Lang P, Yatim H, Moszkowicz P, Otterbein M. Batch Extractive of European Congress of Chemical Engineering, Copenhagen. 2007;
Distillation under Constant Reflux Ratio. Comput Chem Eng. 1994; 16–20.
18:1057–1069. 60. Arifin S, Chien IL. Design and control of an isopropyl alcohol dehy-
46. Gmehling J, M€ollmann C. Synthesis of distillation processes using dration process via extractive distillation using dimethyl sulfoxide as
thermodynamic models and the Dortmund data bank. Ind Eng Chem an entrainer. Ind Eng Chem Res. 2008;47:790–803.
Res. 1998;37:3112–3123.
61. Douglas JM. Conceptual Design of Chemical Processes, New York:
47. Jork C, Kristen C, Pieraccini D, Stark A, Chiappe C, Beste Y-A,
McGraw-Hill, 1988.
Arlt W. Tailor-made ionic liquids. J Chem Therm. 2005;37:537–558.
48. Sim~oes MG, Friedhofer M. An implementation methodology of a 62. Luyben WL. Distillation design and control using aspenTM simula-
fuzzy based decision support algorithm. Int J Knowledge Based tion. Wiley Online Library, 2006.
Intell Eng Syst. 1997;1(4):267–275. 63. Marshall, Swift. Average M & S index for year 2011. Chemical
49. O’ Neill A, Khan U, Nirmalraj PN, Boland J, Coleman JN. Gra- Engineering’s PLANT COST INDEX database. 2012;119:84.
phene dispersion and exfoliation in low boiling point solvents.
J Phys Chem. 2011;115(13):5422–5428.