You are on page 1of 17

Evolutionary Design of Optimum Distillation

Column Sequence
Ramin Bozorgmehry Boozarjomehry,1 Ali Pourahmadi Laleh1 * and William Y. Svrcek2
1. Department of Chemical & Petroleum Engineering, Sharif University of Technology, Tehran, Iran
2. Department of Chemical & Petroleum Engineering, The University of Calgary, Calgary, Alberta, T2N 1N4, Canada

Synthesis of the optimum distillation column sequence (DCS), which incorporates a huge search space composed of both conventional and
complex arrangements, is a highly complicated combinatorial problem in the field of chemical process design and optimisation. In this study, a
novel procedure for the synthesis of optimum DCS proposed by Boozarjomehry et al. [Boozarjomehry et al., 87, 477–492 (2009)] is expanded to
include the complex distillation arrangements. The method is based on evolutionary algorithms, and the total annual cost (TAC) is the main criterion
used to screen alternatives. Efficient procedure has been proposed for encoding mechanism to include and classify various complex arrangements
together with conventional distillation columns. All columns existing in each DCS alternative are designed using the most recommended short-cut
methods to estimate the TAC of the DCS. Four standard benchmark case studies are carried out to clearly demonstrate the excellent performance
of the proposed method. The produced results for these problems indicate that the proposed method outperforms the other existing approaches
in terms of flexibility, accuracy and comprehensiveness.

Keywords: process synthesis, distillation column sequence, optimisation, genetic algorithms

INTRODUCTION columns (sharp separations). Therefore, non-sharp separations as


well as thermally coupled configurations have been excluded from

T
he optimal synthesis of separation systems is one of the
their search area. Leboreiro and Acevedo (2004) have developed a
most important problems in the field of chemical process
GA framework for optimising some separation processes. In their
synthesis. This well-known problem can be addressed as:
work, commercial software has been used for simulation of the
‘Given a feed stream of known conditions (i.e. composition, flow
predefined separation alternatives, and capabilities of GA have
rate, temperature and pressure), what is the best process that sep-
been shown while both heat integration and complex arrange-
arates the desired products from the feed at minimum total annual
ments are involved in the studied cases. However, this work has
cost (TAC)’. In order to deal with this problem, various approaches
mostly focused on obtaining the optimum column conditions,
have been presented such as:
such as the optimum number of trays and the best location of the
1. Parametric studies (Tedder and Rudd, 1978). feed tray, in some predefined separation sequences. Therefore, for
2. Evolutionary strategies (Stephanopoulos and Westerberg, implementation of the approach for each case study, the promising
1976). sequences of interest and their important optimisation variables
3. Conceptual approaches (Gadkari and Govind, 1988; Glinos should be specified at first. These preliminary preparations should
and Malone, 1989; Shah and Kokossis, 2001). be made by some subjective engineering insights and judgments.
4. Statistical methods (Bezzina et al., 1989). Consequently, it is evident that the proposed method is absolutely
5. Algorithmic methods (Floquet et al., 1994; Garrard and Fraga,
1998).
∗ Author to whom correspondence may be addressed.
As in the current work, some researches have employed genetic E-mail address: apourahm@ucalgary.ca
algorithms (GA) in order to design distillation-based separation Can. J. Chem. Eng. 90:956–972, 2012
processes. Wang et al. (1998) used an adjusted GA to synthesise © 2011 Canadian Society for Chemical Engineering
DOI 10.1002/cjce.20589
the optimum heat integrated distillation sequence. Their con- Published online 6 June 2011 in Wiley Online Library
sidered arrangements, however, consisted of simple distillation (wileyonlinelibrary.com).

| 956 | THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING | | VOLUME 90, AUGUST 2012 |
case-sensitive and also cannot effectively be extended for the erly to result in a plausible strategy for dealing with the problem
multi-component mixtures due to the huge number of separation which is to be solved.
alternatives involved. A similar study with the same shortcom- The rest of the study is organised as follows: The Complex
ings as associated with the work of Leboreiro and Acevedo (2004) Arrangements Involved Section presents the complex distilla-
has been carried out by Calzon-McConville et al. (2006). They tion arrangements included in this study and the motivations
have performed some optimisation case studies on fully or par- associated with them. The Problem Statement Section describes
tially thermally coupled arrangements and used a commercial important features of the under investigation problem and its
software package for simulation purposes. In their work, the formulation. The Genetic Algorithm Implemented for Design of
separation alternatives have been restricted to seven predefined Optimal DCS Section provides the implemented method for selec-
sequences composed of simple columns and side-column arrange- tion of the optimum DCS and consists of adjustments made to
ments. Moreover, only optimisation of interconnecting flow rates the conventional GA in order to enhance its performance. The
of side-column arrangements has been considered. Thus, the use performance of the proposed method is tested on four stan-
of some strong optimisation strategies, such as GA, has not been dard benchmark problems in the Results and Discussion Section.
necessary, and a much simpler method, that is, direct search, has Finally, the Conclusion Section provides the most significant
been used instead. achievements and conclusions of this study.
In this study, a novel method for synthesis of the optimum con-
ventional DCS based on the GA, proposed by Boozarjomehry et al.
COMPLEX ARRANGEMENTS INVOLVED
(2009), is expanded to include the complex distillation arrange-
ments. GA, as a stochastic global optimisation technique, imitates The previous studies performed on the optimum design of DCS
the process of natural selection (survival of the fittest) and evo- are usually restricted to simple distillation columns which have
lution (Holland, 1975; Goldberg, 1989). This method generates the following three main characteristics (Smith, 2005):
new points in the search space by applying operators to current
1. They have one feed stream and two product streams.
points and moving toward more optimal regions in the search
2. They use one reboiler and one condenser.
space (Haupt and Haupt, 1998). The GA has the inherent ability
3. They perform sharp separations, that is, there is no com-
of simple representation to encode highly complicated structures
ponent between key components if the components in the
and uses simple transformations to improve such structures dur-
mixture are sorted according to their relative volatilities.
ing the optimisation of highly complex objective functions. Hence,
GA has been widely applied to combinatorial optimisation prob- If the characteristics of simple distillation columns are viewed
lems. In fact, the idea behind GA method can be utilised in various as some restrictions, by relaxation of these restrictions some
ways, and in any way, the involved parameters should be set prop- new arrangements can be produced. These arrangements are

Figure 1. The new arrangements produced by relaxing the characteristics of simple distillation columns.

| VOLUME 90, AUGUST 2012 | | THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING | 957 |
divided into two main groups: (1) conventional columns with sharp separations are again taken into account while exploring
non-sharp separations and (2) complex arrangements. The the feasible arrangements in this work.
most well-known complex configurations consist of side-stream The prefractionator arrangement, due to eliminating the mix-
column, side-stripper and side-rectifier arrangements, prefraction- ing losses of middle product and reducing the mixing losses of
ator arrangement and Petlyuk column. Figure 1 represents all different mixtures at the feed tray, has more thermodynamic effi-
the well-known ‘non-simple’ configurations while separating a ciency than simple column sequences for the same separation task
ternary mixture. (Smith, 2005). Thermally coupled arrangements, that is, Petlyuk
Giridhar and Agrawal (2010) recommended that a good search and side-column arrangements are also satisfied with the ther-
space aimed to find the optimum DCS should be complete and modynamic benefits mentioned for prefractionator arrangement.
compact so that includes all potentially useful configurations Thermally coupled arrangements use thermal coupling to provide
while excluding the configurations dominated by other more use- some of the necessary heat transfer. In consequence, some of con-
ful configurations. These characteristics are taken into account in densers or reboilers are replaced by thermal coupling. Although
the current work, and as a huge extension of previous studies, these arrangements require less heat duties, more of the heat
all the promising complex arrangements as well as conventional transfer is accomplished at extreme levels. Then, depended on
configurations (with both sharp and non-sharp separations) are the conditions, the use of thermally coupled arrangements may be
included while synthesising the optimum DCS. The involved advantageous or disadvantageous. However, all of the thermally
incentives and details are presented below. coupled arrangements are promising for both operating and cap-
As proposed by Doherty and Malone (2001) and confirmed ital cost savings (Smith, 2005). Considering the above facts and
by the authors’ preceding work (Boozarjomehry et al., 2009), comments, all of the mentioned arrangements are added into the
although using non-sharp separations increases the number of search space of the optimisation problem.
distillation columns in the sequence, it is possible that non-sharp Side-stream column arrangement is attractive when the feed
separations lead to more economical DCS’s. Therefore, the non- is dominated by the middle product (typically more than 50%)

Figure 2. Transformation of complex arrangements into their thermodynamically equivalent conventional configurations.

| 958 | THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING | | VOLUME 90, AUGUST 2012 |
and one of the other products is present in only minor quanti- Note, these arrangements are constructed in complex form. Hence,
ties, typically less than 5% (Smith, 2005). In view of the fact that when economically evaluating, the position of each distillation
there are only three distillation sections in side-stream column, section (after design) must be considered on its actual complex
pure products cannot be achieved. Furthermore, side-column arrangement. The procedure for sizing conventional distillation
arrangements tend to side-stream column arrangements as the columns has been detailed in the preceding study (Boozarjomehry
number of trays and heat duty of reboiler (for side-stripper) et al., 2009).
or condenser (for side-rectifier) approach to zero. This condi-
tions is reached when the composition of feed approaches to
the mentioned particular range. In overall, since the side-column GENETIC ALGORITHM IMPLEMENTED FOR
arrangements have already been included, adding the side-stream DESIGN OF OPTIMAL DCS
column arrangement to the search space is not necessary and is Figure 3 presents a block flow diagram of the steps required to
neglected. implement a GA. During the implementation, various methods
Boozarjomehry et al. (2009) have shown that as the number can be used in the selection of parents, crossover and mutation
of mixture components increases, the number of conventional steps. In this study, a natural GA with binary chromosome is used
DCS alternatives grows very sharply. For instance, there are to find the optimum DCS. Among the five major steps of natural
approximately 1092 alternatives available for separation of a GA (i.e. encoding mechanism, fitness evaluation, parent selection,
10-component mixture by conventional DCS. Therefore, it is evi-
dent that by adding the significant complex arrangements into
the search space, the combinatorial aspect of the problem is
intensified extremely, and there exists an astronomic number of
separation alternatives to be dealt with in the current study.

PROBLEM STATEMENT
In the optimisation problem of finding the best DCS, these vari-
ous separation alternatives should be compared against each other
based on a specific performance criterion. An appropriate perfor-
mance criterion to rank various DCS’s is TAC. Both the annualised
capital costs and the operating costs are taken into account in the
TAC of a typical distillation column sequence (DCS), as repre-
sented by Equation (1):


NE

TAC = CCF × Ccapi + Cop (1)


i=1

where, Ccapi is the installed capital cost, Cop is the total annual
operating cost, CCF is the capital charge factor which annualises
the capital cost and NE is the number of pieces of equipment (e.g.
columns, pumps, heaters, etc.) in the DCS. In this study, the CCF
is assumed to be 0.33/year as recommended by Douglas (1988);
and Doherty and Malone (2001).
The total installed cost of the DCS equipment, including
distillation columns, condensers, reboilers, reflux drums and
intermediate heat exchangers, is determined using the correla-
tions developed by Guthrie (1974) and updated using the Marshal
and Swift Index.
The total annual operating cost of the DCS (Cop ) is mainly com-
posed of the utility and energy used for reboilers, condensers,
pumps, etc. The cooling water is used as cold utility and multiple
levels of steam are used as hot utilities. Furthermore, if a furnace
is used as the reboiler or intermediate heater, it is assumed that
the furnace is using natural gas as its fuel. In this study, the utility
costs are calculated using the correlations proposed by Doherty
and Malone (2001).
In order to calculate the cost of distillation columns a consistent
sizing procedure is required. For this purpose, the most recom-
mended short-cut method of Winn–Underwood–Erbar–Maddox
with some rational settings was used. In the case of com-
plex arrangements, these arrangements were transformed into
their thermodynamically equivalent configurations of conven-
tional columns as proposed by Carlberg and Westerberg (1989a,
1989b). Figure 2 shows two instances for such transformations. Figure 3. The typical steps of genetic algorithm.

| VOLUME 90, AUGUST 2012 | | THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING | 959 |
crossover and mutation) the first two are problem dependant able. Hence, if there is n different values for a specific variable,
while the last three are independent of the problem. However, the length of the binary string representing this variable in the
in some cases the performance of the algorithm depends heavily chromosome is E{log2 (n)} + 1, where E(x) is a function whose
on the interaction of these five major steps. The present study value is the greatest integer variable which is less than or equal
falls into this category, namely, this interaction results in a simple to x.
method with excellent performance. A brief overview of the con- Remark 2. In order for GA to perform accurately, each chromo-
stituting stages of GA’s and their implementation was described some must specify one and only one DCS, and each DCS must
by Boozarjomehry et al. (2009). be specified by one and only one chromosome. That is, a one-to-
one correspondence between chromosomes and DCS’s should be
satisfied.
Encoding Mechanism Taking into account the above considerations, the following pro-
The phenotypes in the optimum design of DCS’s are the order cedure is proposed for decoding the chromosomes and specifying
of separations and the type of distillation columns. The order of the corresponding DCS.
separations can be completely captured and explained by deter- According to the number of mixture components and conse-
mination of key components for each column. After determining quently, the number of alternatives for the light key component,
the key components, the type of distillation column is specified. the relevant number of the first bits of chromosome is decoded.
In the proposed method, as outlined in Figure 4, the key com- After determining light key component, the number of alterna-
ponents and the type of distillation columns are determined by tives for the heavy key component is known and by decoding the
GA. Therefore, due to discrete nature of phenotypes, GA performs relevant number of next bits, heavy key component is specified.
efficiently if the genotypes are represented by binary strings as After determining the key components, there are two cases:
recommended by the schema theorem (Holland, 1975). Before
explaining the procedure of decoding chromosomes, two impor- 1. The key components are adjacent: In this case, the separa-
tant remarks are worth highlighting: tion is accomplished through a simple distillation column
Remark 1. In general, for each integer variable, there is a valid (sharp separation) apparently. Hence, decoding the next bits
interval in which the variable can change. Each variable in its of chromosome is not required.
corresponding interval can be represented by the least number 2. The key components are not adjacent: In this case, the type
of bits that can enumerate various legitimate values of the vari- of distillation column must be determined by decoding the

Figure 4. The outline of encoding strategy.

| 960 | THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING | | VOLUME 90, AUGUST 2012 |
next bits of chromosome, and various cases are involved. The rectifier has already been used for a lighter intermediate
separation may be accomplished through a conventional dis- key component).
tillation column or through a prefractionator arrangement or
a Petlyuk column or one of the various arrangements of side- The number of required chromosome bits for a n-component
strippers and side-rectifiers. These various configurations are mixture cannot be fixed at first, and proactive values (higher num-
classified at first. The prefractionator arrangement and Pet- bers) should be generally used to keep the method away from
lyuk column, due to their similarity, are classified as Class ‘shutdown’ because of chromosome bit shortage. Therefore, some
1. The conventional distillation column together with the of the last bits of chromosomes are not usually used when speci-
various arrangements of side-strippers and side-rectifiers are fying corresponding DCS’s. These remainder bits, having no effect
classified as Class 2. The reason why the conventional col- on the specified DCS, are called ‘Trailing’ bits. In order for the GA
umn is not classified separately is explained later.Since the to perform efficiently and robustly (Remark 2), all trailing bits are
Class 2 contains much more arrangements compared with set to ‘0’. The procedure for decoding is explained by the following
the Class 1, the chance of selection from Class 2 is adjusted Illustrations.
to be three times more than the chance of selection from
Class 1. Thus, in order to specify the class of separation, two Illustration 1
bits of chromosome are decoded. If the result is 0, Class 1 Assume that when separating a five-component mixture, the DCS
is chosen and if the result is 1, 2 or 3, Class 2 is chosen. corresponding to the following chromosome is to be specified:
If Class 2 is chosen, in order to satisfy the one-to-one cor-
respondence between chromosomes and DCS’s, number 1 is
encoded in the two decoded bits. Instead of the applied 3/1
ratio, higher ratios such as 7/1 or 15/1 would also be con-
sidered by decoding 3 or 4 bits of chromosome. This would
seem reasonable, particularly for mixtures containing a large
number of components. However, the number of valuable The main feed stream is consisted of five components A, B, C,
mixture components is usually less than 10, and the use of D and E (sorted in descending manner according to their volatili-
3/1 ratio by decoding only 2 bits of chromosome is more ties). This stream is fed to the first distillation configuration. There
rational from a practical perspective. After determining the are four alternatives for the light key of the first distillation config-
separation class, the type of distillation column is specified uration. Thus, the first couple of the chromosome bits ([00]) are
explicitly: decoded. The result is 0 and A is selected as light key component.
(a) Determining distillation column type from Class 1: In consequence, there are four alternatives for heavy key compo-
There are two alternatives. If the next bit is 0, the pre- nent and by decoding the second couple of the chromosome bits
fractionator arrangement is used and if the next bit is ([11]), component E is selected as heavy key component. Since
1, the Petlyuk column is used. the key components are not adjacent, type of distillation column
(b) Determining distillation column type from Class 2: is specified by decoding next chromosome bits. The third couple
This case is more complicated than the previous one. of bits are decoded to specify the class of separation. The result is
Primarily, it must be specified which of intermedi- 3 and the type of distillation column is selected from the Class 2.
ate components are intermediate key components and Considering Remark 2, number 1 is encoded in the two decoded
which of them are distributed components. There are bits. Three components, B, C and D, enter the column as inter-
two extreme cases: if all the intermediate components mediate components and thus, the destiny of each of them must
are determined as key components, the main frac- be determined. The next bit (corresponding to the 4th pickup) is
tion of each of them will be separated through the 1, and, hence, B is an intermediate key component, and since the
products of side-columns and if none of them are deter- 5th pickup bit is 0, the main fraction of B is separated through the
mined as intermediate key components, they will be product of a side-stripper. Similarly, since the 6th pickup bit is 1, C
distributed between the distillate and bottom products is an intermediate key component, but the 7th bit pickup is 1 and
of a conventional column. The second extreme case is thus, the main fraction of C is separated through the product of
corresponding to a conventional distillation column. a side-rectifier. Moreover, if the heavier intermediate components
Therefore, if the conventional column were classified (D) are chosen as intermediate key components (which is the case
separately at first, this arrangement would be specified here as the 8th pickup bit is 1), side-rectifiers will be used neces-
by two different chromosome structures that is not rea- sarily for these components. Assuming that the product purities of
sonable in view of Remark 2.For each of intermediate the first distillation configuration are acceptable, the reminder bits
components, one bit of chromosome is decoded. If the are not decoded, and the DCS of interest is specified as shown in
result is 0, that component will be distributed between Figure 5a. The updated structure of the considered chromosome
the products and if the result is 1, the main fraction will be:
of that component is separated through the product
of a side-stripper or a side-rectifier (to be specified).
For each of intermediate key components, the type of
side-column is specified immediately. If the next bit is
0, the main fraction of considered component will be
separated through the product of a side-stripper. If the
next bit is 1, the main fraction of considered component
will be separated through the product of a side-rectifier Illustration 2
and, also, for the heavier intermediate key components, As the second illustration, assume that when separating a four-
a side-rectifier will be used necessarily (since a side- component mixture, it is desired to specify the DCS corresponding

| VOLUME 90, AUGUST 2012 | | THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING | 961 |
Fitness Function
One of the main issues with the use of the GA is the fitness function
selected to rank the chromosomes. In Problem Statement Section,
it was shown that the TAC can be used to compare the performance
of the DCS configurations. The best DCS arrangement is the one
with the lowest TAC. Hence, the fitness function of GA should be
related to TAC in a reverse manner. Considering Equation (1), the
following equation is used as the fitness function:
 

NE

f = − CCF × Ccapi + Cop (2)


i=1

Solution Methodology
The modified Genitor algorithm is used in this study. As described
by Boozarjomehry et al. (2009), this algorithm provides a superior
performance, in its accuracy and convergence, when compared to
the original Genitor algorithm (Whitley, 1989). In the modified
Genitor algorithm, the selection probability of ith individual from
the mating pool, Pi , is calculated using Equation (3):
 √  
E (
−1+ 1+8n)
−i √ 
Figure 5. The two specified DCS’s for illustration of encoding
2 −1 + 1 + 8n
Pi = i = 0, 1, 2 . . . E (3)
mechanism. n 2

where, n is the population size, E(x) is a function returning the


to the following chromosome: largest integer value which is smaller than x. Using Equation (3)
in the Genitor algorithm modifies the distribution of available
individuals in the mating pool and leads to significant advantages
(Boozarjomehry, 1997).
In addition to the above modification, supervised crossover and
mutation operators were used in the reproduction step to shift the
The main feed stream is consisted of four components A, B, C crossover and mutation points out of the trailing bits region in
and D. For the first distillation configuration, there are three alter- case the crossover or mutation point lies in the region of trailing
natives for selection of the light key component. Thus, the first bits (Boozarjomehry et al., 2009).
couple of the chromosome bits ([00]) are decoded. The result is
0 and A is selected as the light key component. In consequence, RESULTS AND DISCUSSION
there are three alternatives for the heavy key component and by
decoding the second couple of the chromosome bits ([11]), com- In order to illustrate the performance of the presented method,
ponent D is selected as the heavy key component. Since the key four standard benchmark case studies are carried out. The guide-
components are not adjacent, type of distillation column is spec- lines proposed by Haupt and Haupt (1998) have been used for
ified by decoding next chromosome bits. The third couple of bits specifying the GA parameters in these case studies.
are decoded to specify the class of separation. The result is 0 and
the type of distillation column is selected from the Class 1, that is,
a prefractionator arrangement or a Petlyuk column will be used. Table 1. Problem specifications for Example 1
The next bit (corresponding to the 4th pickup) is 1, and the Pet-
lyuk column is hence selected. It is assumed that the purities of Components Mole fractions
the overhead and bottom products of Petlyuk column are accept-
able and the intermediate product stream is dominated by B and Propane 0.05
iso-Butane 0.15
C. Thus, excess purification of the overhead and bottom products
n-Butane 0.25
is not required, and a simple distillation column must be used
iso-Pentane 0.20
for separation of the intermediate product components. Conse- n-Pentane 0.35
quently, the last eight bits are not decoded and in order to obtain Feed flow rate 907.2 kmol/h
the updated structure of chromosome, number 0 is encoded in all Thermodynamic condition of feed Saturated liquid at 1750 kPa
the trailing bits:

Table 2. GA parameters used in optimisation of Example 1

Chromosome length = 36 bits Number of selected parents = 20


Population size = 50 Mutation rate = 20%
Figure 5b shows the corresponding DCS as specified above. Convergence criterion = 95% individual similarity

| 962 | THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING | | VOLUME 90, AUGUST 2012 |
Example 1
The first problem has widely been studied in the literature (e.g.
Malone et al., 1985; Gadkari and Govind, 1988; Glinos and Mal-
one, 1989; Porter and Momoh, 1991; Floquet et al., 1994; Jobson,
1997; Shah and Kokossis, 2001; Rong et al., 2001). The feed spec-
ifications are presented in Table 1. The Peng–Robinson equation
of state was used as the thermodynamic model for vapor–liquid
equilibrium calculations for this hydrocarbon mixture. The oper-
ating pressure was assumed to be in the range of 4–2000 kPa. Table
2 provides the parameters used in the GA to solve this problem.
Figure 6 shows the change in the average and best fitness
by generations. The following chromosome which contains a
description of the optimum DCS was resulted:
Figure 6. The variations of the average fitness and the best fitness versus
succeeding generations for Example 1. [1001000000000000000000000000]

Figure 7. The two rival DCS’s for Example 1. (a) The optimum DCS. (b) The suboptimal DCS.

Table 3. The important characteristics of the columns of Figure 7a for Example 1

Parameter Column 1 Column 2 Column 3 Column 4

P (kPa) 530 724 154 1361


Rmin 1.40 6.45 9.59 4.64
Nmin 15.2 41.5 42.8 17.3
R 1.82 9.03 14.4 6.03
N 30 81 83 37
Ttop (◦ C) 45.3 45.6 40.6 40.5
Tbot (◦ C) 91.8 65.4 49.1 80.6
Qcond (kJ/h) 2.235 × 107 3.219 × 107 6.735 × 107 0.4414 × 107
Qreboil (kJ/h) 2.378 × 107 3.231 × 107 6.738 × 107 0.4530 × 107
D (kmol/h) 408.0 181.9 182.3 45.8
W (kmol/h) 499.2 226.1 316.8 136.1
Eov (%) 68.5 75 71 73
Height (m) 30.8 69.8 75.2 35
Diameter (m) 2.72 2.92 4.76 1.30
TAC ($/year) 0.673 × 106 0.912 × 106 1.308 × 106 0.264 × 106

Component A B C D E

Ultimate product purity (%) 98.52 99.00 99.30 98.51 99.71

| VOLUME 90, AUGUST 2012 | | THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING | 963 |
Figure 7a presents the schematic of the DCS corresponding to the difference is only some 1.0%, the DCS shown in Figure 7a is
the optimum chromosome. The important characteristics of the better than the one shown in Figure 7b. This minor difference has
columns and the purity of the ultimate products are presented in also been confirmed in other studies (Malone et al., 1985; Gadkari
Table 3. The TAC corresponding to this DCS is 3.159 × 106 $/year. and Govind, 1988).
This DCS has also been reported as the optimum simple DCS Shah and Kokossis (2001) after specifying the DCS of Figure 7b
by most other researchers (Malone et al., 1985; Gadkari and as the best simple DCS, reported some complex DCS’s as better
Govind, 1988; Glinos and Malone, 1989; Porter and Momoh, 1991; configurations (Figure 8). In their study, a conceptual formu-
Rong et al., 2001). However, there is a slightly different configu- lation, based on thermodynamic and engineering insights, was
ration (shown in Figure 7b) reported as the optimum by some developed for synthesis of the optimum DCS, and ‘conceptual loss’
of the researchers (Floquet et al., 1994; Jobson, 1997; Shah and was used as a criterion for comparison of various DCS’s.
Kokossis, 2001). The major difference between these two DCS According to the values of conceptual losses assigned to vari-
arrangements is caused by the second and fourth columns whose ous DCS’s in Figure 8, the design (I) was reported as the optimum
most important characteristics are summarised in Table 4. The DCS, and designs (II) and (III) were reported as suboptimal ones.
TAC of the DCS shown in Figure 7b is 3.192 × 106 $/year. Although But, in fact, neither these designs nor the other complex DCS’s are

Figure 8. The best configurations for separating the mixture of Example 1 as reported by Shah and Kokossis (2001).

| 964 | THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING | | VOLUME 90, AUGUST 2012 |
more economical than the best simple DCS which was introduced
Table 4. The most important characteristics of columns 2 and 4 of as the optimum configuration for separating this light hydrocar-
Figure 7b for Example 1
bon mixture. The most important characteristics of the columns
of Figure 8 are presented in Table 5. The TAC’s of the designs (I),
Parameter Column 2 Column 4
(II) and (III) are 4.096 × 106 , 3.371 × 106 and 5.331 × 106 $/year,
P (kPa) 1361 530 respectively. Hence, these designs are not more economical than
Rmin 9.00 8.12 the best simple DCS represented in Figure 7a. In fact, none of
R 12.6 11.4 the complex DCS’s are preferable for this case. As explained by
N 37 78 Rong et al. (2001), for this light hydrocarbon mixture, the oper-
Ttop (◦ C) 40.5 40.2 ating pressures of complex configurations are generally much
Tbot (◦ C) 88.7 52.8 higher than those of conventional DCS’s. This fact results in a
Qcond (kJ/h) 0.8535 × 107 3.078 × 107 significant decrease in the relative volatilities and a significant
Qreboil (kJ/h) 0.8735 × 107 3.081 × 107
increase in the operating temperatures. Thus, the values of Rmin ,
Eov (%) 74 74
Height (m) 34.4 68
N, and, hence, capital and operating costs of reboilers, condensers
Diameter (m) 1.84 2.88 and distillation towers are increased. Therefore, for this example,
TAC ($/year) 0.408 × 106 0.802 × 106 the use of complex arrangements has no economic advantage.

Table 5. Important characteristics of columns of Figure 8

Parameter Column 1 Column 2 Column 3 Column 4

Design (I)
P (kPa) 530 1361 1361 154
Rmin 1.40 59.00 — 9.59
R 1.82 82.60 — 14.39
N 30 125 33 83
Ttop (◦ C) 45.3 40.5 — 40.6
Tbot (◦ C) 91.8 94.1 80.6 49.1
Qcond (kJ/h) 2.235 × 107 5.246 × 107 — 6.735 × 107
Qreboil (kJ/h) 2.378 × 107 3.449 × 107 1.819 × 107 6.738 × 107
D (kmol/h) 408.0 45.8 — 182.3
W (kmol/h) 499.2 226.1 136.1 316.8
Eov (%) 68.5 77.3 73.6 70.9
Height (m) 30.8 102.2 29.4 75.2
Diameter (m) 2.72 3.59 2.39 4.76
TAC ($/year) 0.673 × 106 1.741 × 106 0.373 × 106 1.308 × 106
Design (II)
P (kPa) 724 724 1361 154
Rmin 11.17 — 4.64 9.59
R 16.75 — 6.03 14.39
N 72 45 37 83
Ttop (◦ C) 45.6 — 40.5 40.6
Tbot (◦ C) 105.7 65.4 80.6 49.1
Qcond (kJ/h) 5.696 × 107 — 0.441 × 107 6.735 × 107
Qreboil (kJ/h) 2.215 × 107 3.635 × 107 0.453 × 107 6.738 × 107
D (kmol/h) 181.9 — 45.8 182.3
W (kmol/h) 499.2 226.1 136.1 316.8
Eov (%) 73.0 74.9 73.6 70.9
Height (m) 64.4 39.0 35.0 75.2
Diameter (m) 3.67 3.02 1.30 4.76
TAC ($/year) 1.308 × 106 0.480 × 106 0.264 × 106 1.319 × 106
Design (III)
P (kPa) 530 530 1361 1361
Rmin 1.40 11.57 59.00 —
R 1.82 17.35 82.60 —
N 106 58 125 33
Ttop (◦ C) 45.3 86.5 40.5 —
Tbot (◦ C) 95.1 — 94.1 80.6
Qcond (kJ/h) 2.235 × 107 7.047 × 107 5.246 × 107 —
Qreboil (kJ/h) 9.426 × 107 — 3.449 × 107 1.819 × 107
D (kmol/h) 408.0 182.3 45.8 —
W (kmol/h) 316.8 — 226.1 136.1
Eov (%) 73.6 75.7 77.3 73.6
Height (m) 91.4 49.2 102.2 29.4
Diameter (m) 5.14 4.34 3.59 2.39
TAC ($/year) 2.277 × 106 0.940 × 106 1.741 × 106 0.373 × 106

| VOLUME 90, AUGUST 2012 | | THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING | 965 |
Rong et al. (2001) economically evaluated 13 thermally coupled
DCS’s (composed of some side-column arrangements) and the 14
simple DCS’s. In their study, also, the simple DCS shown in Fig-
ure 7a was reported as the best case among the considered cases.
Thus, it can be concluded that the method developed by Shah and
Kokossis (2001) is not efficient for synthesising the optimum DCS
from an economic perspective.

Example 2
To study the effect of corrosive nature of components on the pro-
duced optimum DCS, it was assumed that the first component of
the mixture used in the previous example would be corrosive. The
corrosive nature of the first component forced the method to use
more corrosion-resisting material of construction (e.g. stainless Figure 9. The variations of the average fitness and the best fitness versus
steel) for those columns in which the concentration of this com- succeeding generations for Example 2.
ponent was not negligible. Therefore, higher capital costs were
calculated for those DCS’s in which the corrosive component
passed through more columns.
The GA parameters for this case were set as Example 1. Figure 9
shows the trend of the average and best fitness. The optimum
chromosome produced by the GA is in the following:

[0001010100000000000000000000]

The TAC for the optimum DCS configuration is 3.652 ×


106 $/year. Figure 10 and Table 6 provide the DCS arrangement
and important characteristics corresponding to the optimum geno-
type.
As shown, in this example, a conventional DCS with a
non-sharp separation was reported as the optimum. Note, the
optimality of a conventional DCS with non-sharp separations is
in contrast with the common perceptions. None of the complex
DCS’s or the simple ones are better than the presented DCS. The
reason why complex arrangements are not present in the optimum
DCS can be explained as per Example 1. Furthermore, the results
of economic evaluation for all the simple DCS’s indicated that the
Figure 10. The optimum DCS for Example 2.
TAC of the best simple DCS (i.e. [A/BCDE, BC/DE, B/C, D/E]) is

Table 6. The important characteristics of the columns of Figure 10 for Example 2

Parameter Column 1 Column 2 Column 3 Column 4 Column 5

P (kPa) 1070 1372 429 530 154


Rmin 6.06 1.72 1.65 9.32 9.59
Nmin 12.2 16.8 14.4 38.0 42.8
R 7.89 2.23 2.15 13.0 14.4
N 27 33 29 81 83
Ttop (◦ C) 49.0 40.2 41.2 40.2 40.6
Tbot (◦ C) 102.4 81.1 82.9 52.8 49.1
Qcond (kJ/h) 1.043 × 107 0.1984 × 107 2.077 × 107 2.810 × 107 6.735 × 107
Qreboil (kJ/h) 1.103 × 107 0.2038 × 107 2.173 × 107 2.813 × 107 6.738 × 107
D (kmol/h) 73.8 45.4 334.2 109.3 182.3
W (kmol/h) 833.4 28.8 499.2 224.9 316.8
Eov (%) 71 73.5 67.8 73.9 70.9
Height (m) 27.2 31.4 30.2 70.4 75.2
Diameter (m) 2.12 0.83 2.65 2.75 4.76
TAC ($/year) 0.732 × 106 0.277 × 106 0.581 × 106 0.757 × 106 1.305 × 106

Component A B (column 2) B (column 4) C D E

Ultimate product purity (%) 99.69 95.30 98.76 99.35 98.51 99.71

| 966 | THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING | | VOLUME 90, AUGUST 2012 |
Table 7. Problem specifications for Example 3

Components Mole fractions

Ethanol 0.25
iso-Propanol 0.15
n-Propanol 0.35
iso-Butanol 0.10
n-Butanol 0.15
Feed flow rate 500.4 kmol/h
Thermodynamic condition of feed Saturated liquid at 101.325 kPa

Table 8. GA parameters used in optimisation of Example 3

Figure 11. The variations of the average fitness and the best fitness
Chromosome length = 36 bits Number of selected parents = 24
versus succeeding generations for Example 3.
Population size = 50 Mutation rate = 15%
Convergence criterion = 95% individual similarity

A/B, C/DE, D/E] with estimated TAC of 4.193 × 106 $/year, which
3.817 × 106 $/year, which is 4.53% more than the optimum cost is 2.2% more than the optimum cost proposed here.
(Boozarjomehry et al., 2009). As the first analysis, the prefractionator arrangement in Fig-
ure 12 was replaced by a Petlyuk column. With this replacement,
Example 3 the TAC for DCS was increased to be 4.216 × 106 $/year (2.8%
A mixture of alcohols whose characteristics are shown in Table more than the optimum cost). The columns 3 and 4 from the two
7 was studied as the third benchmark problem. The operating sequences are the same but the economic parameters of columns
pressure range was between 4 and 2000 kPa. Table 8 provides the 1 and 2 were changed such that the cost of the discussed DCS
GA parameters used to solve this problem. became higher than that of the optimum scenario. The most
Figure 11 shows the trends in the average and best fitness and important characteristics of these two columns which form a Pet-
Figure 12 presents a schematic of the generated optimum DCS. lyuk arrangement are presented in Table 10. As Table 10 indicates,
The chromosome corresponding to the optimum DCS was the the TAC of the first column of Petlyuk arrangement is less than
following: that of the prefractionator arrangement. However, due to the huge
increase in the reflux ratio of the second column, the diameter of
[0101000000000000000000000000] column and heat duties of condenser and reboiler increase. Con-
sequently, the TAC for Petlyuk arrangement becomes higher than
Table 9 presents the important parameters of the columns in that for the prefractionator arrangement.
the DCS. The TAC for the optimum DCS is 4.102 × 106 $/year. In the study performed by Rong et al. (2001), the TAC for one
According to the preceding study (Boozarjomehry et al., 2009), of the complex configurations (Figure 13) was reported to be
the best conventional DCS for the present problem is [AB/CDE, less than that of all the simple DCS’s and the other considered

Table 9. The important characteristics of the columns of Figure 12 for Example 3

Parameter Column 1 Column 2 Column 3 Column 4

P (kPa) 101 101 101 101


Rmin 0.92 1.25 8.95 5.71
Nmin 18.8 44.6 65.2 29.7
R 1.10 2.16 13.42 7.99
N 39 81 126 56
Nfeed 20 16 & 57 70 26
Ttop (◦ C) 84.3 79.8 78.2 107.9
Tbot (◦ C) 104.1 113.4 82.2 117.8
Qcond (kJ/h) 2.034 × 107 2.425 × 107 6.959 × 107 1.773 × 107
Qreboil (kJ/h) 2.068 × 107 2.441 × 107 6.960 × 107 1.775 × 107
D (kmol/h) 248.7 200.0 124.8 50.5
M (kmol/h) — 174.9 — —
W (kmol/h) 251.7 125.5 75.2 74.9
Eov (%) 47.9 54.5 61.0 58.1
Height (m) 53.6 92.0 128.6 62.6
Diameter (m) 2.14 2.44 3.82 2.08
TAC ($/year) 0.617 × 106 0.975 × 106 1.858 × 106 0.652 × 106

Component A B C D E

Ultimate product purity (%) 99.70 98.84 99.60 98.00 99.67

| VOLUME 90, AUGUST 2012 | | THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING | 967 |
Table 12. Problem specifications for Example 4

Components Mole fractions

n-Heptane 0.10
n-Octane 0.10
n-Nonane 0.10
n-Decane 0.10
n-Undecane 0.10
n-Dodecane 0.10
n-Tridecane 0.10
n-Tetradecane 0.10
n-Pentadecane 0.10
n-Hexadecane 0.10
Feed flow rate 1000.0 kmol/h
Thermodynamic condition of feed Saturated liquid at 101.325 kPa

Figure 12. The optimum DCS for Example 3.

Table 10. The most important characteristics of the discussed Petlyuk


arrangement in Example 3

Parameter Column 1 Column 2

P (kPa) 101 101


Rmin — 2.27
R — 4.25
N 39 80
Ttop (◦ C) 84.3 79.8
Tbot (◦ C) 104.1 113.4
Qcond (kJ/h) — 4.035 × 107
Qreboil (kJ/h) — 4.085 × 107
Eov (%) 47.9 54.5
Height (m) 54.2 90.8
Diameter (m) 2.14 3.14 Figure 13. The favourable DCS for Example 3 reported by Rong et al.
TAC ($/year) 0.343 × 106 1.363 × 106 (2001).

Table 11. Important characteristics of the favourable DCS of Rong et al. (2001) for Example 3

Parameter Column 1 Column 2 Column 3 Column 4

P (kPa) 101 101 101 101


Rmin 11.94 — 2.28 2.52
R 17.90 — 2.96 3.28
N 171 60 30 27
Nfeed 92 1 30 27
Ttop (◦ C) 78.2 — 97.2 107.9
Tbot (◦ C) 117.8 82.3 — —
Qcond (kJ/h) 9.124 × 107 — 2.717 × 107 0.8474 × 107
Qreboil (kJ/h) 6.666 × 107 6.074 × 107 — —
D (kmol/h) 124.8 — 174.0 50.8
W (kmol/h) 74.9 75.8 — —
Eov (%) 58.1 60.9 56.4 58.1
Height (m) 182.6 61.8 35.4 31.2
Diameter (m) 3.97 3.60 2.33 1.40
TAC ($/year) 2.831 × 106 0.957 × 106 0.358 × 106 0.178 × 106

| 968 | THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING | | VOLUME 90, AUGUST 2012 |
developed a simulated annealing-based method in order to solve
Table 13. GA parameters used in optimisation of Example 4
the problem. Floquet et al. (1994) assumed only a specific recov-
ery of key components; hence desired purity levels of products
Chromosome length = 80 bits Number of selected parents = 96
Population size = 160 Mutation rate = 10% were not necessarily satisfied. This is, however, in contrast with
Convergence criterion = 95% individual similarity practice that the objective of a DCS is to produce the products with
the specified purities. Moreover, the search area of Floquet et al.
(1994) was restricted to simple columns and columns with one
side stream. The current GA method, on the other hand, accounts
for the required realistic objective (i.e. high purity products), and
the search area is much broader than that of Floquet et al. (1994).
Hence, the result obtained in this study cannot be compared with
that obtained in the original study of Floquet et al. (1994). How-
ever, as noted above, the scalability of the GA-based method would
be demonstrated while dealing with this severely combinatorial
problem.
Table 12 provides the mixture characteristics. The parameters
used in the GA to solve this problem are presented in Table 13.
Figure 14 clearly shows the excellent convergence through only
nine generations. Note, due to the large search space dimension, a
large population size was selected. Eighty-bit chromosomes were
Figure 14. The variations of the average fitness and the best fitness used for this 10-component mixture and the chromosome corre-
versus succeeding generations for Example 4.
sponding to the optimum case was:
12 complex ones. Table 11 presents the most important charac-
[1101011000001010110100010000. . .00]
teristics of columns of this sequence. The TAC for this DCS is
4.325 × 106 $/year (5.4% more than the optimum case). Although
Figure 15 shows the schematic of the optimum DCS, while Table
this sequence is more economical than most of configurations
14 contains the significant column parameters. The TAC for this
consisted of side-columns and simple columns, the DCS shown
DCS is 13.084 × 106 $/year.
in Figure 12 is better than it. Note, Rong et al. (2001) did not
Some sensitivity analyses performed on the optimum DCS are
include many of complex arrangements such as prefractionator
outlined in the following:
and Petlyuk configurations in their study.
• The use of side-rectifier instead of Column 2 (side-stripper)
Example 4 increases the TAC up to 14.437 × 106 $/year.
As the last example, a large-scale multi-component mixture was • The use of 4 extra side-stripper in Column 1 instead
studied to verify the scalability of the proposed method. The of Columns5, 6, 8, 9 and 10, increases the TAC up to
problem was originally studied by Floquet et al. (1994). They 20.174 × 106 $/year.

Figure 15. The optimum DCS for Example 4.

| VOLUME 90, AUGUST 2012 | | THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING | 969 |
|
970
|
Table 14. The important characteristics of the columns of Figure 15 for Example 4

Parameter Column 1 Column 2 Column 3 Column 4 Column 5 Column 6 Column 7 Column 8 Column 9 Column 10

P (kPa) 101 101 101 101 101 101 101 101 101 101
Rmin 17.81 — — — 4.84 1.46 5.67 2.17 1.25 1.93
R 24.93 — — — 6.29 1.89 7.94 2.82 1.62 2.78

THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING


N 117 17 3 21 52 36 68 34 37 85

|
Nfeed 87 1 1 1 20 18 34 17 19 22 & 59
Ttop (◦ C) 98.7 — — — 204.8 140.2 270.4 125.8 180.6 173.7
Tbot (◦ C) 277.1 255.6 220.9 158.9 235.4 190.5 284.8 150.5 206.2 215.3
Qcond (kJ/h) 8.2834 × 107 — — — 1.3959 × 107 2.1325 × 107 4.5371 × 107 1.3211 × 107 1.3817 × 107 1.4198 × 107
Qreboil (kJ/h) 3.3465 × 107 1.6336 × 107 0.3577 × 107 5.0587 × 107 1.4269 × 107 2.3364 × 107 4.5466 × 107 1.3373 × 107 1.4152 × 107 1.4355 × 107
D (kmol/h) 100.0 — — — 41.5 198.1 100.0 99.5 128.0 94.7
M (kmol/h) — — — — — — — — — 87.7
W (kmol/h) 200.0 100.0 140.6 459.4 99.1 261.3 100.0 98.6 133.3 79.0
Eov (%) 59.4 60.7 30.0 61.8 66.7 64.3 68.7 64.2 59.7 66.4
Height (m) 122.6 19.2 8.4 22.8 51.2 38.0 63.8 36.2 41.6 81.2
Diameter (m) 5.78 2.82 1.30 3.73 2.60 3.26 4.86 2.43 2.58 2.69
TAC ($/year) 3.525 × 106 0.832 × 106 0.159 × 106 1.214 × 106 0.936 × 106 1.040 × 106 2.939 × 106 0.572 × 106 0.779 × 106 1.088 × 106

Component A B C D E F G H I J

Ultimate product purity (%) 99.50 99.00 99.02 99.31 98.99 99.03 99.39 99.00 99.00 99.50

|
VOLUME 90, AUGUST 2012
|
• The use of 7 side-rectifier in Column 1 instead of Columns 2, be taken into account while seeking the optimum distillation
3, 4, 5, 6, 8, 9 and 10, increases the TAC up to 16.181 × 106 $/ configuration.
year.
• Replacement of the prefractionator arrangement (Columns 9
and 10) with a Petlyuk column increases the TAC up to NOMENCLATURE
13.096 × 106 $/year. Ccap total installed capital cost ($)
• Replacement of the prefractionator arrangement (Columns 9 CCF capital charge factor
and 10) with a side-stripper arrangement increases the TAC up Cop total annual operating cost ($/year)
to 13.405 × 106 $/year. D distillate product flow rate (kmol/h)
• Replacement of the prefractionator arrangement (Columns 9 Diameter diameter of column (m)
and 10) with a side-rectifier arrangement increases the TAC up E(x) the greatest integer which is less than or equal to x
to 13.343 × 106 $/year. Eov overall column efficiency (%)
• The use of conventional columns for separation of this mixture f fitness function
is much more expensive. For instance, the TAC for direct and Height height of column (m)
indirect simple DCS are 17.609 × 106 and 21.751 × 106 $/year, M intermediate product flow rate (kmol/h)
respectively, which are 34.6% and 66.2% more than the N number of theoretical trays
optimum cost. The preceding study of Boozarjomehry et al. Nfeed feed stage number accounting from top
(2009) indicated that the TAC of the best conventional DCS is Nmin minimum number of theoretical trays
15.696 × 106 $/year. This value is 20.0% higher than the opti- P operating pressure of column (kPa)
mum cost presented in this study. Pi probability of selection for ith individual from mating
pool
It is finally worth highlighting the fact that as a significant
Qcond heat duty of condenser (kJ/h)
characteristic of the proposed method, GA has exerted its intel-
Qreboil heat duty of reboiler (kJ/h)
ligence to find the optimum DCS surrounded by a gigantic
R operating reflux ratio
number of similar but non-optimum DCS alternatives in the last
Rmin minimum reflux ratio
example.
TAC total annual cost ($/year)
Tbot column bottom temperature (◦ C)
Ttop column top temperature (◦ C)
CONCLUSIONS W bottom product flow rate (kmol/h)
A novel method, based on the GA, has been presented for the
optimum design of DCS’s for the separation of multi-component
mixtures. In this work, distillation columns are not restricted to Subscripts
simple columns, and the search space expands to include the bot bottom of column
conventional distillation columns and complex columns such as cap capital
prefractionator arrangement, Petlyuk column, and various config- cond condenser
urations of side-strippers and side-rectifiers. Except for azeotropic op operating
mixtures, the method can be used for both ideal and non-ideal ov overall
mixtures since a wide variety of thermodynamic models has reboil reboiler
been incorporated into the software package. Moreover, economic top top of column
effects of equipment material of construction and all necessary
unit operations, such as preheating and precooling, are taken into
account when estimating the TAC for a DCS. For cases in which REFERENCES
it is not possible to use cooling water and steam as cooling and Bezzina, M., L. Pibouleau and S. Domenech, “A Statistical
heating media, economic evaluation is appropriately based on the Approach to the Synthesis of Separation Sequences,” Comput.
use of refrigeration loops and fired heaters. The excellent per- Chem. Eng. 13, 651–660 (1989).
formance of the method in terms of its flexibility, accuracy and Boozarjomehry, R. B., “Application of Artificial Intelligence in
speed has been shown by its application for design of the opti- Feedback Linearisation,” Ph.D. Dissertation, The University
mum DCS’s for four benchmark problems. It was realised that of Calgary, Calgary, Alberta, Canada (1997).
the method is capable of producing the exact optimum DCS in Boozarjomehry, R. B., A. Pourahmadi Laleh and W. Y. Svrcek,
a short time. For example, on a Pentium D (3.20 GHz) and 2.0 “Automatic Design of Conventional Distillation Column
GB of RAM PC, the average run-time required to obtain the opti- Sequence by Genetic Algorithm,” Can. J. Chem. Eng. 87(3),
mum DCS for the 5- and 10-component mixtures was 10 and 477–492 (2009).
30 min, respectively. The required chromosome bits for 5- and Calzon-McConville, C. J., M. B. Rosales-Zamora, J. G.
10-component mixtures were 36 and 80, respectively. Thus, the Segovia-Hernández, S. Hernández and V. Rico-Ramı́rez,
number of required chromosome bits for a n-component mixture “Design and Optimisation of Thermally Coupled Distillation
can be fixed at about 8 × n. The synthesised optimum DCS’s for Schemes for the Separation of Multicomponent Mixtures,”
the studied cases did indicate that both non-sharp separations Ind. Eng. Chem. Res. 45, 724–732 (2006).
and complex arrangements, due to their economic advantages, Carlberg, N. A. and A. W. Westerberg, “Temperature-Heat
should be included in the search space while synthesising the Diagrams for Complex Columns. 2. Underwood’s Method for
optimum DCS. Furthermore, the results obtained for the bench- Side Strippers and Enrichers,” Ind. Eng. Chem. Res. 28,
mark case studies demonstrated the strength and capability of GA 1379–1386 (1989a).
in solving complicated combinatorial optimisation problems. In Carlberg, N. A. and A. W. Westerberg, “Temperature-Heat
the future work, heat integration of the separation process would Diagrams for Complex Columns. 3. Underwood’s Metbthod

| VOLUME 90, AUGUST 2012 | | THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING | 971 |
for the Petlyuk Configuration,” Ind. Eng. Chem. Res. 28, Algorithms and Their Applications,” J. D. Schaffer, Ed.,
1386–1397 (1989b). Morgan Kaufmann Publishers Inc, San Francisco, CA (1989),
Doherty, M. F. and M. F. Malone, “Conceptual Design of pp. 116–123.
Distillation Systems,” McGraw-Hill, New York, NY (2001).
Douglas, J. M., “Conceptual Design of Chemical Processes,”
McGraw-Hill, New York, NY (1988). Manuscript received November 23, 2010; revised manuscript
Floquet, P., L. Pibouleau and S. Domenech, “Separation received March 15, 2011; accepted for publication March 24, 2011.
Sequence Synthesis: How to Use Simulated Annealing
Procedure?” Comput. Chem. Eng. 18, 1141–1148 (1994).
Gadkari, P. B. and R. Govind, “Analytical Screening Criterion for
Sequencing of Distillation Columns,” Comput. Chem. Eng.
12, 1199–1213 (1988).
Garrard, A. and E. S. Fraga, “Mass Exchange Network Synthesis
Using Genetic Algorithms,” Comput. Chem. Eng. 22,
1837–1850 (1998).
Giridhar, A. and R. Agrawal, “Synthesis of Distillation
Configurations: I. Characteristics of a Good Search Space,”
Comput. Chem. Eng. 34, 73–83 (2010).
Glinos, K. and M. F. Malone, “Net Work Consumption in
Distillation—Short-Cut Evaluation and Applications to
Synthesis,” Comput. Chem. Eng. 13, 295–305 (1989).
Goldberg, D. E., “Genetic Algorithms in Search, Optimisation,
and Machine Learning,” Addison-Wesley Reading, MA
(1989).
Guthrie, K. M., “Process Plant Estimating, Evaluation, and
Control,” Craftsman Books, Solano Beach, CA (1974).
Haupt, R. L. and S. E. Haupt, “Practical Genetic Algorithms,”
John Wiley & Sons, Inc., New York, NY (1998).
Holland, J. H., “Adaptation in Natural and Artificial Systems,”
The University of Michigan Press, Ann Arbor, MI (1975).
Jobson, M., “Short-Cut Evaluation of Distillation Sequences,”
Comput. Chem. Eng. 21, S553–S557 (1997).
Leboreiro, J. and J. Acevedo, “Processes Synthesis and Design of
Distillation Sequences Using Modular Simulators: A Genetic
Algorithm Framework,” Comput. Chem. Eng. 28, 1223–1236
(2004).
Malone, M. F., K. Glinos, F. E. Marquez and J. M. Douglas,
“Simple, Analytical Criteria for the Sequencing of Distillation
Columns,” AIChE J. 31, 683–689 (1985).
Porter, K. E. and S. O. Momoh, “Finding the Optimum Sequence
of Distillation Columns—An Equation to Replace the Rules of
Thumb (Heuristics),” Chem. Eng. J. 46, 97–108 (1991).
Rong, B. G., A. Kraslawski and L. Nyström, “Design and
Synthesis of Multicomponent Thermally Coupled Distillation
Flowsheets,” Comput. Chem. Eng. 25, 807–820 (2001).
Shah, P. B. and A. C. Kokossis, “Knowledge Based Models for the
Analysis of Complex Separation Processes,” Comput. Chem.
Eng. 25, 867–878 (2001).
Smith, R., “Chemical Process Design and Integration,” John
Wiley & Sons, Inc., Chichester (2005).
Stephanopoulos, G. and A. W. Westerberg, “Studies in Process
Synthesis—II. Evolutionary Synthesis of Optimal Process
Flowsheets,” Chem. Eng. Sci. 31, 195–210 (1976).
Tedder, D. W. and D. F. Rudd, “Parametric Studies in Industrial
Distillation, Part I. Design Comparisons,” AIChE J. 24,
303–315 (1978).
Wang, K., Y. Qian, Y. Yuan and P. Yao, “Synthesis and
Optimisation of Heat Integrated Distillation Systems Using an
Improved Genetic Algorithm,” Comput. Chem. Eng. 23,
125–136 (1998).
Whitley, D., “The Genitor Algorithm and Selection Pressure:
Why Rank-Based Allocation of Reproduction is Best,” in:
“Proceeding of the Third International Conference on Genetic

| 972 | THE CANADIAN JOURNAL OF CHEMICAL ENGINEERING | | VOLUME 90, AUGUST 2012 |

You might also like