Accepted Manuscript

Development of A Reaction/Distillation Matrix for Systematic

Generation of Sequences in A Single Two Component
Reaction-Separation Case Study

Mahya Nezhadfard , Leila S. Emami , Norollah Kasiri ,

Mohammad H. Khanof , Amirhossein Khalili-Garakani ,
Javad Ivakpour

PII: S0098-1354(18)30665-3
DOI: 10.1016/j.compchemeng.2018.06.025
Reference: CACE 6151

To appear in: Computers and Chemical Engineering

Received date: 2 August 2017

Revised date: 24 June 2018
Accepted date: 30 June 2018

Please cite this article as: Mahya Nezhadfard , Leila S. Emami , Norollah Kasiri ,
Mohammad H. Khanof , Amirhossein Khalili-Garakani , Javad Ivakpour , Development of A
Reaction/Distillation Matrix for Systematic Generation of Sequences in A Single Two Com-
ponent Reaction-Separation Case Study, Computers and Chemical Engineering (2018), doi:
10.1016/j.compchemeng.2018.06.025

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service
to our customers we are providing this early version of the manuscript. The manuscript will undergo
copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please
note that during the production process errors may be discovered which could affect the content, and
all legal disclaimers that apply to the journal pertain.
 ACCEPTED MANUSCRIPT

Highlights
 A systematic approach for synthesis of reaction-distillation processes is proposed.
 The algorithm is expansion of previously proposed “Separation Matrix”.
 As a case study, 28 different sequences for n-Butane isomerization process are analyzed
using the proposed algorithm.

T
IP
CR
US
AN
M
ED
PT
CE
AC
 ACCEPTED MANUSCRIPT

Development of A Reaction/Distillation Matrix for Systematic Generation of

Sequences in A Single Two Component Reaction-Separation Case Study

Mahya Nezhadfarda, Leila S. Emamia, Norollah Kasiri*a, Mohammad H. Khanofa,

Amirhossein Khalili-Garakania,b, Javad Ivakpourc

a
Computer Aided Process Engineering (CAPE) laboratory, School of Chemical Oil and Gas

T
Engineering, Iran University Of Science And Technology, Narmak, Tehran 1684613114,

IP
Iran.

CR
b
Chemical & Process Engineering Department, Niroo Research Institute, Tehran 1468613113

c
US
, Iran
AN
Petroleum Refining Technology Division, Research Institute of Petroleum Industry, Tehran

1485733111, Iran.
M

*
Corresponding author E-mail: [email protected]
ED

Abstract
PT

Studies on simultaneous synthesis of the reaction-separation processes are limited. The

initial step is choosing a method to generate all possible arrangements. In this work, the
CE

Separation Matrix algorithm previously developed by the authors has been expanded to
include the reaction part of the process to form a Reaction/Distillation matrix.
AC

In order to achieve a comprehensive algorithm, initially all feasible matrices have been
analyzed to find common points among non-synthesizable matrices. The ability of new
matrix to include a broad selection of flowsheets has then been investigated. A single two-
component reaction, is assumed to prevail within a four-component mixture, the fourth
being the initially nonexistent reaction product. A systematic approach is presented to
determine a large selection of sequences for a four-component system including three
distillation columns (simple or complex) and a reactor. A practical case study of the
 ACCEPTED MANUSCRIPT

isomerization of n-Butane has been analyzed using the procedure developed here.

Keywords: Process synthesis, Reaction-separation systems, Reaction/Distillation Matrix,

Probable sequence generation

1. Introduction

T
Since the introduction of the term “process synthesis”, the process systems engineering

IP
(PSE) community has experienced considerable advancement (Cremaschi, 2015). Over the
recent decades, process synthesis has had a prominent impact on the design, development

CR
and operation of (petro) chemical processes. Since the 1970s significant achievements have
been made in the general area of process synthesis for chemical processes, and numerous

US
synthesis frameworks, strategies and tools have been developed and proposed (Barnicki &
Siirola, 2004; Douglas, 1985; Nishida et al., 1981; Westerberg, 2004; Yuan et al., 2013).
AN
Over the recent years important studies have been carried out on process synthesis.
Papalexandri and Pistikopoulos proposed a generalized framework for process synthesis
M

alternatives based on fundamentals of heat/mass transfer principles (Papalexandri &

Pistikopoulos, 1996). Bertran et al. presented a systematic framework for novel and
ED

sustainable design-synthesis of processing routes along with the associated computer-aided

methods and tools (Bertran et al., 2016). Tula et al. presented an integrated computer-aided
PT

software-tool that searches the design space for hybrid/intensified more sustainable
process options (Tula et al., 2017). Kokossis et al. introduced a new framework to support
CE

synthesis of complex engineering problems using a paradigm that combines optimization

with ontological knowledge modeling (Kokossis et al., 2016). Demirel et al. developed a
AC

novel and systematic process design and intensification method. They replaced the
traditional and classical unit operation-based flowsheets and superstructures process units
representation, and proposed a new fundamental building blocks representation (Demirel
et al., 2017).

In most chemical processes, reactions are associated with a series of separation stages.
Reaction-separation processes are common processing cases in chemical industries and
usually are highly sensitive to change or disturbance in design and operational variables. The
 ACCEPTED MANUSCRIPT

design of the reactor-separator-recycle (RSR) systems may be looked upon as the

foundation of most chemical plants, which other process sections are to support (Kramer,
2012; Omtveit et al., 1993). It should be noted that this study only deals with the reaction-
separation systems in which the reactor and the distillation columns are improvised
individually and reactive distillation systems are not within the scope of this study.

The successful selection of reactor, flow configuration and operational parameters of a RSR
system, determines advancement and rationality of a whole chemical engineering process.

T
More robust methods are needed in the synthesis of these systems. Previous researches on

IP
process integration have generally considered reaction and separation as processes that
occur sequentially in a flowsheet. Each of the trio, the reactor synthesis, the separation and

CR
the recycle problems may be posed as Mixed Integer Programming problems, but obviously
the corresponding formulation of the entire system is much larger and more complex. Thus,

US
various methods are developed considering the parts of these systems. Over the last three
decades various procedures for reactor networks (RNs) synthesis consisting of complex
AN
reactions have been proposed by researchers which can be classified in three major groups:

The first group involves graphic methods which are called attainable region (AR). This
M

method was proposed by Horn (Horn, 1965). In this method it is tried to find the best
combination of ideal reactors using graphic methods in the concentration space. This
ED

approach was used later in other problems such as optimal control, combined reaction and
separation by several researchers (Glasser et al., 1987; Hopley et al., 1996; Kauchali et al.,
PT

2002; Khumalo et al., 2006; Rooney et al., 2000).

The second group of RNs synthesis is based on definition of superstructures. Many

CE

important work were carried out, the first being the work of Chitra and Govind (Chitra &
Govind, 1985). Their model involved two plug flow reactors with a recycle stream, using
AC

recycle ratio as the optimization variable. Superstructure optimization involving feasible

configuration of ideal reactors along with all possible connections often leading to a
nonlinear programming (NLP) or mixed integer nonlinear programming (MINLP) (Achenie &
Biegler, 1990; Kokossis & Christodoulos, 1990). Recently Živkovid and Nikačevid proposed a
novel method for reaction synthesis based on superstructure optimization. Their method
establishes strong interconnection between process intensification (PI) principles and
process system engineering (PSE) techniques, which are used for problem formulation and
 ACCEPTED MANUSCRIPT

optimization. In their method, different process options are generated using PI principles
and PSE techniques to determine the optimal process route among those options using
optimization methods and algorithms (ŽivkoviL & NikaLeviL, 2016).

Although various attempts have been done to increase efficiency and effectiveness of both
groups still drawbacks and disadvantages can be observed in both groups of methods. The
problem of AR method is that considering three dimensions (three constituents in the
problem) is really difficult and often impossible graphically and the method is also quite

T
time consuming. Methods based on superstructures confront the question of whether all

IP
potential feasible configurations are taken into account in the formulation. To overcome
this problem complex structures are considered. As the structure becomes more complex

CR
the number of variables increases and the solution becomes more complicated. Another
problem about superstructures is using NLP and MINLP models for optimization in which

US
reaching the best solution is not guaranteed. Therefore, to reduce complexity of the RNs
synthesis problems and guarantee attaining an optimum solution, the third group of
AN
methods was proposed. In these methods instead of solving all variables simultaneously,
continuous variables are handled by classic mathematical methods, while integer variables
M

are treated with stochastic methods. In this way the complexity of the problem is reduced
and probability of reaching optimum solution is increased (Soltani & Shafiei, 2015).
ED

Many other synthesis algorithms have also been developed for the solution of distillation
column sequences. Synthesis algorithms can be used for systematically generating
PT

sequences in optimization procedures. Thompson and King have examined such

configurations (King, 1987; Thompson & King, 1972). Heuristics for the synthesis have been
CE

proposed based on physical insights by researchers (Perkins, 1989; Rathore et al., 1974;
Seader & Westerberg, 1977). Since the superstructure representation in the process
AC

synthesis area presents a systematic manner for the simultaneous consideration,

assessment and optimization of alternatives, it has indisputably become a desirable and
eligible approach for researchers.

The superstructure concept has previously been assessed by many authors, seeking to
embed all possible configurations. An early superstructure was proposed by Sargent and
Gaminibandara (Sargent, 1998), allowing the possible existence of non-sharp splits. They
also proposed an extended space to include satellite configurations not embedded in the
 ACCEPTED MANUSCRIPT

Sargent and Gaminibandara superstructure. A systematic rule-based method was also

provided in which individual basic configurations might be created by beginning at the
product end in a network and marching back towards the feed end. In doing so some
decisions need to be made regarding reboilers, condensers and transfer streams (Agrawal,
1996, 2003). The exhaustive space was published for multi component feeds, generated
through a combination of individual split possibilities (Fidkowski, 2006; Rong et al., 2003). A
state task network formulation was proposed by Sargent (Sargent, 1998) and extended by
Yeomans and Grossmann (Yeomans & Grossmann, 1999). Caballero and Grossmann have

T
proposed a mathematical generalized disjunctive model for synthesis of multicomponent

IP
distillation networks (Caballero & Grossmann, 2004). After reviewing several alternatives for

CR
the separation of four-component mixtures, Agrawal and Giridhar have found that the heat
duty of basic configurations is always lower in comparison to the non-basic ones (Giridhar &

US
Agrawal, 2010a). Therefore, searching for a configuration with low heat duty only through
basic configurations seems to be adequate.
AN
Despite earlier efforts, the formulation of the search space has not enjoyed as much
success. Thus to find optimal configurations, paying enough attention to the formulation of
M

the search space along with mathematical optimization methods is necessary. A proper and
optimum search space as stated by Agrawal ought to be complete and compact. Complete
ED

in not missing any potentially useful configuration, and compact in not containing a large
number of configurations usually excluded in comparisons with others (Giridhar & Agrawal,
2010b). So, search space reduction methods are crucial to reduce the number of the
PT

candidates and analysis time. Khalili-Garakani et al. presented a new search space reduction
CE

method based on exergy analysis (Khalili-Garakani et al., 2016). By using the State Task
Network approach, a new algorithm for the synthesis of distillation column sequences in
nonsharp separations with possible multicomponent products is presented by Ivakpour and
AC

Kasiri (Ivakpour & Kasiri, 2009). To generate different distillation column sequences, the
Separation Matrix (SM) was introduced which makes synthesis of complete sequences
easier and synthesis of reduced sequences possible. Shah and Agrawal also presented a
method to generate the search space using the concept of the separation matrix for
sequences with maximum n-1 columns (Shah & Agrawal, 2010).
 ACCEPTED MANUSCRIPT

Although many systematic methods for process design have been developed over the last
25 years, only a few contributions address the simultaneous design and optimization of
integrated reaction and separation processes. If these methods are developed, interaction
between optimal conditions can be considered. Linke and Kokossis presented stochastic
optimization technology for the synthesis of integrated reaction and separation process
superstructures (Linke & Kokossis, 2003b). They used shortcut simulation models for the
separation tasks. In order to carry out process performance assessment, Recker and
Marquardt developed a nonlinear programming approach for the conceptual design of

T
integrated reaction-separation systems using shortcut models (Recker & Marquardt, 2012).

IP
Marcoulaki et al. presented an approach for designing reaction-separation networks using

CR
stochastic optimization methods (Marcoulaki et al., 2001). Recker et al. presented a
systematic and optimization-based approach for the design of integrated reaction-

US
separation processes. In their work shortcut methods are utilized to screen alternative
flowsheet structures (Recker et al., 2015). Linke and Kokossis presented a new framework
AN
for the synthesis of chemical processes involving reaction and separation. Their
methodology utilizes superstructures that are generated from combinations of generic
synthesis units to capture the possible processing alternatives and provide a venue for
M

optimization (Linke & Kokossis, 2003a). Over the early stages of process development,
selection of the best design case from the candidates has to be made with defective data. A
ED

two-stage stochastic programming is one approach for optimization under these conditions
of uncertainty. In two-stage stochastic problems, the design parameters ought to be fixed
PT

while the operating parameters may be adapted for the realization of the uncertain
parameters (Steimel & Engell, 2015). Most of the previous attempts to solve the reactor-
CE

separator-recycle synthesis problems have employed mathematical programming to do so

as a complex MINLP problem (Recker et al., 2014). Also a characteristic equation (CE) to
AC

maximize total yield of a RSR system is presented (Wang et al., 2006).

So far, simplified performance models provide the main body of the available mathematical
programming models for distillation columns, including short-cut models, linear mass
balance equations and aggregated models. While some of these methods can provide useful
results in terms of preliminary designs or bounds for process synthesis, it is clear that in
order to increase their industrial relevance and scope of application, for non-ideal mixtures
in particular, direct incorporation of rigorous models in the design procedures is inevitable.
 ACCEPTED MANUSCRIPT

Rigorous MINLP synthesis models exhibit significant computational difficulties such as the
requirement of a good initialization point, the introduction of equations that can become
singular and the solution of many redundant equations. In summary, these difficulties
mount to the requirement of good initial guesses, high computational times and bounds on
the design variables to achieve model convergence (Cabrera-Ruiz et al., 2011).

The Reaction/Distillation matrix algorithm presented in the present work is the developed
version of Separation Matrix previously presented by the authors (Ivakpour & Kasiri, 2009).

T
It broadens the applications of the Separation Matrix to the processes including chemical

IP
reaction. The advantage of the method presented in this paper compared to the
abovementioned methods which use shortcut models for doing the calculations of reaction

CR
or distillation or both, which are not exact and include some simplifications, is that it can be
connected to commercial software like ASPEN PLUS or HYSYS to perform the simulation and
optimization of the processes rigorously.
US
Incorporation of the reaction in the separation matrix totally changed its rules and caused to
AN
creation of a completely new algorithm which is introduced in the following sections. The
presented Reaction/Distillation matrix algorithm deals with matrices with certain rules and
M

principles. Therefore it is easily programmable in all programming environments capable of

working with matrices such as MATLAB, C sharp, etc.
ED

Integrated processes including reaction and separation segments are far more complicated
compared to pure separation process system designs. Many system design procedures and
PT

algorithms do not claim attaining the best overall solution even in processes solely limited
to separation trains. In many cases such as heuristic based methods contradictory rules
CE

generate design complications. The more complicated the process becomes including a
chain of reactions in possibly more than one reactor, separation of many none ideal
AC

components in several separation equipments and involvement of a lot of energy usage,

stringent and delicate search of the solution space to ensure location of an appropriate
solution is of utter most importance. The complication of this practice makes it more and
more of a mathematical nature. Thus only the more mathematical approaches stand a
better chance of providing proper means of locating better process sequences. The
Separation Matrix approach is probably the most mathematically oriented algorithm with
potential capabilities of doing so.
 ACCEPTED MANUSCRIPT

Despite earlier studies on the Separation Matrix, the present work is the first attempt to
incorporate reactions into the matrix. Since the current study is the first attempt in
developing such algorithms, a simple system has been chosen for assessment purposes. But
it doesn’t mean that this matrix is unable to extend to more complex reaction conditions.
Generalization of the Reaction/Distillation Matrix to more complex reactions and the
reactions with more than two components is subject of the future work of the authors. The
proposed algorithm in this paper promises the improvement of synthesis and optimization
algorithms of reaction-separation systems using exact and rigorous simulation models.

T
IP
In developing optimum solutions for mixtures containing more than three components, the
first concern is defining a search space including all the alternative configurations to extract

CR
a good (ideally the best) configuration with simultaneous savings in energy and/or capital
costs without unnecessary increase in the operational and control complexities and costs.

US
As a first step, providing an algorithm to create all possible sequences, for system including
a reactor and distillation columns is necessary. In this study we will try to further develop
AN
the Separation Matrix obtained from Kasiri and Ivakpour’s work, and achieve an algorithm
for the systematic generation of this space.
M

2. Theory
ED

In this section, the development process for the Reaction/Distillation matrix is fully
explained. Since the Reaction/Distillation matrix is an extended version of the separation
PT

matrix developed by Kasiri and Ivakpour (Ivakpour & Kasiri, 2009), understanding the
separation matrix algorithm is a prerequisite for understanding the Reaction/Distillation
CE

matrix algorithm. Therefore, in section 2.1, the separation matrix algorithm is completely
explained. Incorporating reaction into the separation matrix and extending it to
AC

Reaction/Distillation matrix for generation of all the possible sequences of a reaction-

separation system is carried out in the present work. As mentioned earlier, in this study only
the reaction-separation systems in which the reactor and the distillation columns are
improvised individually are dealt with and reactive distillation systems are not within the
scope of this study. In section 2.2, by explaining the changes that the reaction caused in the
matrix, the development stages of the Reaction/Distillation matrix are fully described.
 ACCEPTED MANUSCRIPT

Finally in section 2.3 the algorithm of Reaction/Distillation matrix derived from rules
governing the matrices obtained from the previous section is presented.

2.1. Separation Matrix Algorithm

In the Kasiri and Ivakpour’s work, synthesis of different distillation column configurations is
accomplished based on the introduction of an upper triangular matrix called Separation
Matrix. The Separation Matrix (SM) is another representation of STN and contains the

T
sequence specifications (i.e., the number of distillation columns, interconnections between

IP
them, and bypass possibilities). For comparison, the general forms of the separation matrix
representation and the complete state task network have been illustrated in Figure 1. In the

CR
state task network, the nodes indicate the mixtures and the lines connecting two nodes
represent the separation tasks which must be performed in the distillation column sections.

US
In the SM, mixtures collected from a condenser are demonstrated by {I}, the mixtures
collected from the reboilers by {II}, and the mixtures produced as a side stream of
AN
distillation columns by {O}. The Greek letter φ is used to determine the possibility of having
three available options (φ={I, II, O}). According to the assumptions used in STN
M

representation, each specific stream occurs only once and also each distillation column is
only equipped with one reboiler and one condenser. So, the products containing the lightest
ED

final product must be extracted from the condenser while the products containing the
heaviest final product must be collected from the reboiler (Ivakpour & Kasiri, 2009). In
Figure 1, this is illustrated for the general separation matrix of a four component mixture, in
PT

which each stream containing P1 must be produced by the condenser ({I}) while each
containing P4 must be collected at the reboiler ({II}). Other products may be given by the
CE

condensers, reboilers, or as side streams (φ={I, II, O}). For detailed description of a
sequence, specification of all φ options must be provided.
AC

After determination of each φ option, the required modification must be performed in the
separation matrix. Therefore, the location of each mixture in the SM demonstrates its
constituents by itself. Hence, the matrix is required to contain only the φ options of each
stream that indicate the structure of the sequence. The required modifications in the SM
are dependent on the φ option chosen. As stated by the state task representation, when a
mixture is to be collected from a condenser (e.g. * +), all mixtures on the same
 ACCEPTED MANUSCRIPT

diagonal but in the previous columns ought to be eliminated from the SM (Figure 2a).
Similarly, when a mixture is to be collected from a reboiler (e.g. * +), all mixtures
containing it as the lightest part must be eliminated from the SM. Mixtures to be eliminated
in this way are those belonging to the same row as the mixture in question in the preceding
columns (Figure 2b). When a mixture is collected as a side stream (e.g., * +), no
elimination would be required (Figure 2c). After each φ-option selection, the required
modifications must be made in the SM. Because the φ-option selection at each column only
alters the previous separation matrix columns, this procedure is performed from the last

T
column to the former ones. The complete SM will be created by definition of all * +s. The

IP
same operation may be performed to generate the separation matrices for all complete

CR
sequences by taking into account all possible φ-option choices. It is clear that as all possible
choices for the φ-option are investigated for each φ, that all possible sequences can be

US
generated by this algorithm. In the following, determination of the relevant sequence from
a specified SM is explained. At the first step, the positions of the top products for each
AN
distillation column are determined. Having done so, the corresponding feed containing the
investigated product may be determined. According to the SM specifications, streams with
similar light parts like the feed stream and the investigated products, are located on the
M

same row. Each feed is separated into two products in the distillation columns. Thus, the
next product of the distillation column and the investigated feed must have similar heavy
ED

parts. If this product is collected from the reboiler ( * +), it will be the bottom product.
Otherwise, it is a side-stream product and the procedure must be repeated while the
PT

bottom product is obtained. The sequence structure and the interconnections between
distillation columns are specified using interconnecting streams. An example demonstrating
CE

the procedure for obtaining of a structure of a sequence from a SM is given in Figure 3.

AC

Figure1. Sequence representation: a) STN and b) SM representation of a four-component system

 ACCEPTED MANUSCRIPT

T
IP
CR
US
Figure 2. Separation matrices for a three product separation (general state with three possibilities: (a) direct,
AN
(b) indirect, and (c) prefractionator
M
ED
PT

Figure 3. Procedure of determination of the sequence structure from the separation matrix.
CE

In this way, the separation matrix is formed. In the current work, addition of a reaction to
AC

SM is studied for synthesis of a reaction-separation system. Because it is the first attempt in

developing separation matrix, a simple system is assumed that includes a single two-
component reaction within a four component mixture, the fourth being the initially
nonexistent reaction product. Inclusion of more complex reactions and the reactions with
more than two components in the separation matrix is subject of future work of the
authors.
 ACCEPTED MANUSCRIPT

2.2. Development of the Reaction/Distillation matrix

Adding reaction to separation matrix described in the previous section is the novelty of this
work. In this section, we describe development of the Reaction/Distillation matrix for a four-
component system in which a two-component reaction takes place. The most fundamental
change in the SM is due to the fact that the appearance of reaction limits the possible
options of -choices. In a four-component system six forward and six reverse two-
component reactions can prevail.

T
It is assumed that the product is not present in the feed. According to this assumption, after

IP
setting a two-component reaction, product is removed from matrix arrays. The resulting

CR
matrix is illustrated in Figure 4 for AB reaction.
In the next step, those arrays of the preliminary matrix in which the reaction may occur will
be determined. In fact, in order to enter the reaction into the matrix, the reactor position,

US
its feed and output stream must be located at matrix arrays. Hereon, the term ‘reactive
array’ is used for this purpose. Certainly, the reaction proceeds in streams in which the
AN
reactant is a key component. At this stage it is assumed that the product must be the
element of a reactive array too. Otherwise, after product generation, a mixture is obtained
M

that does not exist in the default matrix arrays. So a problem would be faced in the
synthesis algorithm. In this study, all of the distillation columns are to be equipped with one
ED

reboiler and one condenser. Therefore, two arrays in each column of the matrix will be
specified with states I and II. By inserting the reaction into the matrix, the -option will
PT

become a function of reactive array’s situation. In accordance with the previous section,
selection of options requires the modifications in the SM. So the -option shouldn’t lead
CE

to the elimination of the reactive array.

AC

Figure 4. The SM after removal the product from the matrix

Also, each specific stream occurs only once in the matrix. In order to achieve a
comprehensive algorithm, initially all feasible SMs for all reactions, including all  options
 ACCEPTED MANUSCRIPT

for each matrix, have been manually drawn and analyzed trying to find common points
between non-synthesizable matrices. However, in this work due to the elimination of the
product character from the matrix, duplicate arrays are created. Therefore the -option set
must be selected so as to omit these arrays. In fact, after setting -options for the nth
column, the situation of duplicate arrays is checked. If they still remain within the matrix,
the duplicate arrays located in previous column must be removed by the selection of proper
-options for n-1th column. For example, stream is repeated in the matrix of Figure 4.
By setting -options in Figure 5a, both similar arrays were omitted and the matrix synthesis

T
is completed. With the same selection for the nth column, due to the elimination of from

IP
the third column, the other at the 2nd column could be left in the matrix as it is not

CR
duplicated anymore. This trend is illustrated in Figure 5b. In Figure 5c, after determining -
options of the last column, duplicate arrays are still present within the matrix. So, the choice
of -option must lead to removing the
column in the present case, in other words, US array in the previous column, the second
.
AN
M
ED
PT
CE
AC

Figure 5. Several options for duplicate arrays

Therefore, presence priority of similar arrays within the matrix is at columns in which
members have the same number of components ( remained in the third column
consisting of members with the two components). But, when the reactive array is located
atn-1th column, repeated appearance of the reactant within the matrix is inevitable. It
should be noted that as the initial feed and the final products (at the nth column) cannot be
 ACCEPTED MANUSCRIPT

removed from the matrix, therefore arrays repeating these at other locations would be
deleted. For example, in Figure 6, the must be eliminated and .

Figure 6. Non-elimination of feed array.

T
Also, if the specified -option set for the ith column leads to the removal of all members of i-

IP
1thcolumn, the matrix cannot be synthesize and this set will be omitted from -option

CR
choices. If the reactive array is located at (1,1), the feed stream enters the reactor first,
before being sent to the distillation columns for separation purposes. In this case, the

US
product character appears at the first column of the matrix and system will have a complete
4-component matrix where -options have no limitations, because any choice of -options
AN
will not eliminate the (1,1) member of the matrix.
As a result possible sequences are similar to the 4-component separation system defined in
the previous works with the addition of a reactor. The next step is followed with the
M

reaction . This reaction, in addition to the initial feed, may occur in stream
(located at (1,2)) and (located at (1,3)).For the reaction (at (1,2)), in addition
ED

to the removal of * + option from and choices(non-elimination of reactive array),

also * +option is possible for while there are no duplicate arrays after determination of
PT

and options(For example in Figure 7). This restriction also applies to the other
reactions.
CE

For the reaction (corresponding to (1,3)), it is not possible to determine sequence

structures using the usual procedure for the separation matrices presented in Figure 8. With
AC

an overview of the matrix-options of other reactions, there are similar cases, thus a general
rule may be deduced through comparison as follows:
 When the ‘x’ product of forward reactions is the heaviest component of the
reactive array, the choice of {O} must be omitted from the φ-option set. In fact,
this product cannot be obtained as a side stream.
In other words it may be stated that:
 ACCEPTED MANUSCRIPT

 While ‘x’ member is a final product of a forward reaction and a reactive array is
selected at the same diagonal but in the previous columns, then this product can
only be collected from the reboiler and as a heavy component of the column.
( )

T
IP
CR
US
Figure 7. Example of possible {O} option for φBC.
AN
M
ED

Figure 8. Non-synthesizable matrices for the reaction IA→IAB.

Also in the case of reverse reactions, those matrices facing difficulty with synthesis have
PT

been investigated and by finding a way to share between them the same principle, a similar
rule as above is deduced.
CE

 When the ‘x’ product of reverse reactions is the lightest component of the
AC

reactive array, the choice of {O} must be omitted from the φ-option set.
In other words it may be stated that:

 While ‘x’ member is a final product of a reverse reaction and the reactive array is
selected in the same row of the previous column, then this product can only be
collected from the condenser and as a light component of the column. ( )
 ACCEPTED MANUSCRIPT

When the reaction product is the lightest or heaviest system component, a complete
sequence cannot be generated if the reactive array located at n-1thcolumn. Also in this case,
according to Figure 9, could only be one option due to the earlier limitations.

th
Figure 9. Generation of the lightest or heaviest system component at n-1 column

T
IP
2.3. Reaction/Distillation matrix algorithm

CR
In general, the following algorithm is used to generate possible sequences for a system
consisting of separation and a two-component reaction:
US
1. A (n×n) matrix is composed and in accordance with the separation matrix algorithm, all
AN
arrays in the first row with and all arrays on the diagonal with are
determined.
M

2. A two-component reaction is selected and the arrays including both components are
determined.
ED

3. The reaction product is removed and the matrix is checked for redundant duplicate
arrays:
- If a duplicate element (except reactant) was in n-1thcolumn and it was similar to one of
PT

the final products (at nth column), it must be eliminated at this stage.
4. The reactive array is selected.
CE

5. The reaction product appears in the matrix after the column of reactive array.
6. The last column is considered ( ).
AC

7. If selection of or option for available in the ith column (according to the rules of the
separation matrix) lead to the removal of the reactive array, they are omitted from -
choices.
8.

a) In forward reactions, if the reactive array is selected at the same diagonal as the final
product x, the possible option for this product is just determined with .
 ACCEPTED MANUSCRIPT

b) In reverse reactions, if the reactive array is selected at the same row as the final
product x, the possible option for this product is just determined with .
9. Other choices for -option in ith column is investigated and eliminations due to them are
applied to the matrix.
10. Previous column (i-1th) is selected and the available -options are determined. At this
point, in addition to restrictions at stage 7, the {O} option is possible for any only if
there is no duplicate array of the same in the matrix. With determination of -
options, omissions due to them are applied in the matrix.

T
11. If the -option set is selected and its mentioned eliminations lead to the removal of all

IP
members of previous column, the matrix cannot be synthesize and this set is omitted

CR
from the -option choices.
12. As long as there is unspecified in the matrix, stage 10 will be applied to the previous
column ( ).
US
13. The process of drawing sequences from each matrix is carried out by separation matrix
AN
algorithm.
14. After synthesis of all cases, return to step 4 and the algorithm for other reactive arrays is
repeated.
M

3. Performance evaluation of Reaction/Distillation matrix using a case study

ED

A simple case study has been evaluated in this section to show the performance of the
PT

proposed algorithm presented in the previous section . First the system and the reaction studied
in the case study is described. Then generation steps of the possible sequences using
CE

Reaction/Distillation matrix algorithm is explained for the considered system. Finally, simulation and
economic evaluation method of the case study is explained.
AC

3.1. Problem Statement and Assumptions

The two-component reaction chosen is the isomerization of n-Butane to isobutane.

Increasing refinery productivity and improving fuel quality demands to engage isobutane as
a key feedstock for producing both oxygenated additives and alkylates. Substantial growth
of isobutane output via isomerization of n-butane poses a real technological challenge
(Urzhuntsev et al., 2014). Feed specifications for case study are provided in Table 1.
 ACCEPTED MANUSCRIPT

The assumptions used are listed as follows:

1. The system includes a two-component isomerization reaction in one reactor and

distillation process in three distillation columns. Distillation sequences are “Complete
Configurations” in which n-1 columns are used to separate an n-component mixture.

2. The system considered is a four-component system including Ethane, Propane,

Isobutane as the product and n-Butane as the reactant.

T
3. Isobutane is not present before the reaction takes place.

IP
4. The columns are to be equipped with one reboiler and one condenser.

CR
Ethane concentration is considered to be 0.92 and concentration of Propane, Isobutane and
n-Butane is 0.99 in the final products. Conversion percentage in the reactor is 66.24%.

US
Table 1. Case study: Feed specification
AN
Component B.P.(°C) Mole fraction Flow rate (Kmol/hr)

Ethane(A) -88.3 0.2 20

M

Propane(B) -42.2 0.4 40

Iso-Butane(C) -11.7 0 0
ED

n-Butane(D) -0.4 0.4 40

Total - 1.0 100.0

PT
CE

3.2. Generation of the possible sequences using Reaction/Distillation matrix algorithm

AC

According to the algorithm presented in section 2.3, first the separation matrix for a four-
component system is developed:

 FABCD I ABC I AB IA 
 II BCD BC B 

 II CD C 
 
 II D 
 ACCEPTED MANUSCRIPT

Figure 10. Separation matrix for a four-component system

According to the volatility order of the components, the reaction considered in our case
study is adjusted with D→C case. The arrays which include both components (reactant
and product) are shown in the Figure 11. The reactive array can be selected among
these three arrays:

 FABCD I ABC I AB IA 
 II BCD BC B 


T
 II CD C 
 

IP
 II D 

CR
Figure 11. Determination of the arrays including both reactant and product elements

US
After elimination of the reaction product from the arrays selected at the previous stage,
the matrix changes to the following form in which the only duplicate element in n-1th
AN
column similar to one of the final products is the reactant which shouldn’t be
eliminated:
M

 FABD I ABC I AB IA 
 II BD BC B 

ED

 II D C 
 
 II D 
PT

Figure 12. Separation matrix after removing the product from the arrays
CE

The reactive array is selected among arrays which have both reactant and product
elements. First, (1,1) array is selected as the reactive array. Product element is added to
AC

the arrays after the reactive array. The obtained matrix is illustrated in Figure 13:

 FABD ABCD I ABC I AB IA 

 II BCD BC B 

 II CD C 
 
 II D 

Figure 13. General form of the Reaction/Distillation matrix for D→C reaction occurring at (1,1) array
 ACCEPTED MANUSCRIPT

The reaction occurring at (1,1) array doesn’t change the matrix and the matrix is exactly
the same as the matrix for four-component system without reaction. Because:

1. None of the φ-options eliminates the reactive array.

2. Since the reaction occurs at the first array of the matrix, product is not
omitted from the remaining arrays, so generation of duplicate arrays is
impossible.

Therefore, when the reaction occurs at (1,1) array there are no limitations for φ-options. In

T
this case, 17 different matrices are generated which are illustrated in the Figure 14.

IP
Now we return to the reactive array selection stage. This time array (2,2) is selected as the

CR
reactive array. After addition of the product to the arrays in the columns after the reactive
array, the general form of the Reaction/Distillation matrix is illustrated in Figure 15.

US
Now, for generation of the matrices, selection among φ-options should be carried out.
Selection must be done in a way that the reactive array is not removed. Therefore:
AN
BC , B  II
M

With eliminating option II from the options of and , 7 cases are eliminated from 17
possible cases with only 10 remaining.
ED

I two cases, determining the O option for φ causes the generation of duplicate arrays. So
these cases should be removed from the possible cases. These two cases are shown in the
PT

Figure 16.

With the elimination of these two cases, 8 cases remain which are depicted in the Figure 17.
CE

The last array remaining as the reactive array is the array (3,3). After addition of the product
AC

in the arrays in the columns after the reactive array, the general form of the
Reaction/Distillation matrix is illustrated in Figure 18.

Since the selected reactive array is at the same row as the reaction product (C), according to
stage 8-b of the Reaction/Distillation matrix algorithm, the possible option for this product
may only be determined with . Therefore the remaining options for the matrix are
the three cases related to B  I , II ,O whose matrices are shown in the Figure 19.
 ACCEPTED MANUSCRIPT

 FABD ABCD   IA   FABD ABCD   IA   FABD ABCD   IA 

 II BCD  I B   II BCD I BC I B   II BCD I BC I B 
  
 II CD I C    II C   II CD OC 
     
 II D   II D   II D 

(1,1)-1 (1,1)-2 (1,1)-3

 FABD ABCD  I AB IA   FABD ABCD I ABC I AB IA   FABD ABCD I ABC I AB I A 

   II B     II B     II B 
  
 II CD IC    II C   II CD OC 

T
     
 II D   II D   II D 

IP
(1,1)-4 (1,1)-5 (1,1)-6

CR
 FABD ABCD  I AB I A   FABD ABCD  I AB I A   FABD ABCD  I AB I A 
 II BCD  O B   II BCD I BC O B   II BCD I BC O B 
  
 II CD I C    II C   II CD OC 



II D 

 US 
II D 



II D 
AN
(1,1)-7 (1,1)-8 (1,1)-9

 FABD ABCD I ABC  IA   FABD ABCD I ABC  IA   FABD ABCD I ABC I AB I A 

  II BC I B    II BC I B    II BC O B 
  
M

  II C   II CD OC    II C 
     
 II D   II D   II D 
ED

(1,1)-10 (1,1)-11 (1,1)-12

 FABD ABCD I ABC I AB I A   FABD ABCD I ABC  IA   FABD ABCD I ABC  IA 

PT

  II BC O B   II BCD O BC I B   II BCD O BC I B 
  
 II CD OC    II C   II CD OC 
     
CE

 II D   II D   II D 

(1,1)-13 (1,1)-14 (1,1)-15

AC

 FABD ABCD I ABC I AB I A   FABD ABCD I ABC I AB I A 

 II BCD O BC O B   II BCD O BC O B 
 
  II C   II CD OC 
   
 II D   II D 

(1,1)-16 (1,1)-17

Figure 14. Generated Reaction/Distillation matrices for D→C reaction occurring at (1,1) array.
 ACCEPTED MANUSCRIPT

 FABD I AB I AB IA 
 II BD BCD BC B 

 II CD C 
 
 II D 

Figure 15. General form of the Reaction/Distillation matrix for D→C reaction occurring at (2,2) array.

 FABD I AB I AB IA   FABD I AB I AB IA 
  O B 
 II BD BCD O BC O B   II BD BCD O BC
  II C   OC 

T
II CD
   
 II D   II D 

IP
CR
Figure 16. Non-synthesizable matrices for D→C reaction occurring at (2,2) array.

 FABD   IA   FABD   IA 
 I B   I B 




II BD BCD 
II CD IC 
II D 

US 



II BD BCD I BC
 II C 
II D 

AN
(2,2)-1 (2,2)-2

 FABD   IA   FABD  I AB IA 
M

 II BD BCD I BC I B   II BD BCD  O B 
 
 II CD OC   II CD IC 
   
ED

 II D   II D 

(2,2)-3 (2,2)-4
PT

 FABD  I AB IA   FABD  I AB IA 
 II BD BCD I BC O B   II BD BCD I BC O B 
 
CE

  II C   II CD OC 
   
 II D   II D 
AC

(2,2)-5 (2,2)-6

 FABD I AB  IA   FABD I AB  IA 
 II BD BCD O BC I B   II BD BCD O BC I B 
 
  II C   II CD OC 
   
 II D   II D 

(2,2)-7 (2,2)-8

Figure 17. Generated Reaction/Distillation matrices for D→C reaction occurring at (2,2) array.
 ACCEPTED MANUSCRIPT

 FABD I AB I AB IA 
 II BD B B 

 II D CD C 
 
 II D 

Figure 18. General form of Reaction/Distillation matrix for D→C reaction occurring at (3,3) array.

 FABD   IA   FABD  I AB IA   FABD  I AB IA 

 II BD  I B     II B   II BD  O B 
  

T
 II D CD IC   II D CD IC   II D CD IC 
     

IP
 II D   II D   II D 
(3,3)-1 (3,3)-2 (3,3)-3

CR
Figure 19. Generated Reaction/Distillation matrices for D→C reaction occurring at (3,3) array.

3.3. Simulation
US
AN
As explained in the previous section, the proposed algorithm presented 28 sequences for
synthesizing this system which are illustrated in Figure 20. Sequence names are shown in
M

the format of (m,n)-i. (m,n) indicates the reactive array in that sequence and i is the
sequence number. Aspen HYSYS V8.6 is used for steady-state simulation of these sequences.
ED

UNIQUAC thermodynamic EOS is applied for the simulations. The specifications used for the
simulation of distillation columns were purities of the final products. Ethane concentration
PT

is considered to be 0.92 and concentrations of Propane, i-Butane and n-Butane were 0.99 in
the final products. Conversion percentage in the reactor was 66.24%. For simulation of the
CE

reaction stage, an equilibrium reactor model was employed.

AC

At the first stage, in order to find the preliminary data required for designing the distillations
columns such as reboiler and condenser pressure, minimum reflux ratio, number of trays
and the feed stage, shortcut method was applied. The pressures of the columns were
optimized at this stage with the objective being to reach as close to atmospheric pressure as
possible. The results of the shortcut method were used as the initial estimates in the
rigorous calculations. At this stage the reflux ratios of the columns were optimized and the
objective was reaching to the specified product purities utilizing the minimum hot and cold
 ACCEPTED MANUSCRIPT

duty in the reboiler and condenser respectively. After doing the rigorous simulation, a
sensitivity analysis is applied on the simulated cases to ensure that comparing the cases in
terms of total annual cost is reasonable and the cases are not far away from their optimal
situation. Therefore, after performing rigorous simulation using the data obtained from
shortcut simulation, some changes are made on the variables of distillation columns to see
how utility and capital costs are affected by these changes.

There are two kind of distillation columns in the sequences simulated in this study: Simple

T
and complex columns. Simple columns have only one feed and two product streams. For

IP
simulation of the simple columns, the initial data is obtained using shortcut distillation
column which gives the number of trays, location of feed tray and an initial guess for reflux

CR
ratio. Then the data obtained from shortcut calculations is used in rigorous simulation. It has
been observed that the simulated rigorous distillation column in this way is almost in its

US
optimal situation. In these columns, decreasing the number of trays increases the heat
duties in the condenser and reboiler of the columns and this increases the utility cost. The
AN
decrease of capital cost due to decrease of number of trays is not so much that can
compensate the increase of utility cost, so the total annual cost increases. Decreasing the
M

number of trays on these columns sometimes increases the reflux ratio as well, which leads
to increase of utility cost (because of increasing the condenser duty) and capital cost
ED

(because of increasing the column diameter). Therefore, further optimization process in not
necessary for simple distillation columns.
PT

But for complex distillation columns which have more than one feed and two product
streams, the shortcut calculation doesn’t give good initial guesses. The default shortcut
CE

distillation column in ASPEN HYSYS has only one feed and two product streams. Therefore,
for simulation of the complex distillation columns, the column is divided to n parts where n
AC

is the number of feed streams entering the column. Then, for each part a shortcut
distillation column is simulated. Then the initial data for the main columns is obtained using
these n shortcut columns. For example, the number of trays in the complex column is the
summation of numbers of trays in the shortcut columns. This procedure has been used by
researchers for initial design of Brugma and Petlyuk sequences (Fidkowski, 2006; Jiménez et
al., 2003; Ramírez-Corona et al., 2010). In this way, an initial guess is obtained for rigorous
simulation of the complex column, which is far away from the optimal situation for this
 ACCEPTED MANUSCRIPT

column. In order to improve the condition of these columns and approaching the optimal
situation, first the number of trays has been decreased to the extent that it doesn’t lead to
significant increase in condenser and reboiler duties. More decrease of the number of trays
has been done considering the trade-off between reduction of capital cost due to reduction
of the column height and the increment of utility cost due to increment of heat duties of
condenser and reboiler and the increment of capital and utility cost due to increase of reflux
ratio which simultaneously increases the column diameter (capital cost) and the condenser
duty (utility cost). Now the obtained number for the trays is almost optimal. It means that

T
after making every change in the number of trays, the total annual cost is calculated and

IP
this is repeated until reaching the least TAC. Then, the different locations for the feed trays

CR
and the side streams have been tried to reach the least reflux ratio for the column. After
obtaining the optimal number of trays and reflux ratio, the final TAC has been calculated for
each sequence.
US
AN
M

(1,1)-2
(1,1)-1
ED
PT
CE

(1,1)-3
(1,1)-4
AC

(1,1)-5 (1,1)-6

Figure 20. Proposed sequences by algorithm for case study

 ACCEPTED MANUSCRIPT

(1,1)-7 (1,1)-8

T
IP
CR
(1,1)-9 (1,1)-10

US
AN

(1,1)-11 (1,1)-12
M
ED
PT

(1,1)-14
(1,1)-13
CE
AC

(1,1)-15 (1,1)-16

Figure 20. Continued

 ACCEPTED MANUSCRIPT

(1,1)-17 (2,2)-1

T
IP
CR
(2,2)-2 (2,2)-3

US
AN

(2,2)-4 (2,2)-5
M
ED
PT

(2,2)-7
(2,2)-6
CE
AC

(2,2)-8
(3,3)-1

Figure 20. Continued

 ACCEPTED MANUSCRIPT

(3,3)-3
(3,3)-2

Figure 20. Continued

T
4. Results and Discussions

IP
CR
The economic evaluation of the distillation processes was performed using Aspen Process
Economic Analyzer. Capital and utility costs of distillation processes for each sequence is

US
also calculated using this tool. Reactor price is calculated using diagrams presented in Peters
(Peters & Timmerhaus, 1991) and were updated using the Marshall and Swift index. Utility
AN
includes steam of the reboiler and cooling water and the refrigerants used in the
condensers of distillation columns. In this study, the desired products in the acceptable
range of specifications are presumed to have the same price in optimization of all sequence
M

structures. The best scheme is introduced by investigating the total annual cost (TAC). The
TAC values have been calculated as follows:
ED

TAC = annual utility cost + Annualized capital cost (1)

PT

( )
Annualized Capital Cost = capital cost (2)(Smith, 2005)
( )
CE

In equation (2) n is the plant lifetime which is considered equal to 10 years and i is fractional
interest rate per year which is considered to be 15% in Iran. In the current work, the capital
AC

cost mainly consists of the installed cost of distillation column shell and trays as well as the
reactor and heat exchanger costs.

Having carried out the assessments, the TAC value lays within 1391071 to 1762553 $/year
with the minimum value belonging to the case (1,1)-7 of Figure 20 as shown in Figure 21.
Costs of all 28 simulated sequences are listed in Table 2.
 ACCEPTED MANUSCRIPT

By examining relative volatility of the components, the obtained result can be justified. In
sequences in which the reactor is placed at the beginning of the process and the reaction
product is a four-component mixture containing A, B, C and D, looking at the relative
volatility of components of this mixture, it can be seen that the most difficult separation is
related to C and D components, because their relative volatility is the closest to one
(αCD=1.35), with the easiest separation being related to B and C components (αBC=2.5).
Considering the heuristic rules proposed for the selection of the sequence for simple
nonintegrated distillation columns which are based on the observations made in many

T
problems, the low cost of four first cases in Table 2 is justifiable. Heuristic rule 1 states that:

IP
“Separations where the relative volatility of the key components is close to unity or that

CR
exhibiting azeotropic behavior should be performed in the absence of non-key components.
In other words, do the most difficult separation last” (Smith, 2005). According to this rule in

US
our case separation of C and D components should be performed at the last stage of the
separations. In the first four cases is Table 2 having least TAC value, separation of C and D
AN
components is done last. However, there are other rules and sometimes they are in conflict
with each other and each one suggests a different sequence. Therefore one cannot justify
the difference in the TAC of all sequences using these rules only. Nevertheless the low-TAC
M

sequences in our case are justifiable with these rules.

ED

As can be seen in the first sequence of Table 2 which has the lowest TAC, the reactor is
placed at the beginning of the process and all the components of the three-component
mixture, including reactant and the other components are entering the reactor. In these
PT

sequences, due to entry of non-reactant components to the reactor, the volume and thus
CE

the cost of reactor is more than the sequences in which one or both non-reactant
components are separated before entering the reactor. However, according to the
assessment performed on all the possible sequences, it is observed that apparently in the
AC

case determined as the best case, increase in the reactor cost due to volume increase is
compensated by other items such as utility costs.

The conventional process used in NGL recovery, whose feed is similar to the feed used in the
case study of this paper, employs a direct configuration, in which first methane is separated
in the demethanizer column, then ethane, propane and butane are separated in the
deethanizer, depropanizer and debutanizer columns, respectively. Separation of i-Butane
 ACCEPTED MANUSCRIPT

and n-butane is done last (Park et al., 2015). The conventional scheme for NGL recovery
plant is illustrated in Figure 22. This direct separation configuration comes up ranked 15th in
the sequences sorted in ascending TAC order in Table 2 as sequence (1,1)-1, but since there
are not any reactions occurring in the LNG recovery plant, the comparison should be made
among (1,1) cases of the case study in which the four-component mixture in the reactor
outlet is separated using three distillation columns. The direct configuration comes up
ranked 10th in the sequences sorted in Table 2 which indicated presentation of 9 better
sequences. It should be noted that there are methane and C5+ components in the NGL

T
recovery plant feed, which are not present in the feed of our case study, but the comparison

IP
could still be valid.

CR
US
AN
M
ED

Figure 21. The proposed optimal sequence for case study.

PT
CE
AC

Figure 22. Conventional scheme of NGL recovery (Park et al., 2015)

 ACCEPTED MANUSCRIPT

Table 2. Cost Table of the simulated sequences

Capital Annualized
Utility Cost TAC
Sequence Cost Capital
Name Cost
(USD/Year) (USD) (USD/Year)
(USD/Year)
(1,1)-7 $332,052 $6,628,750 $1,059,019 $1,391,071

T
(1,1)-4 $390,933 $6,358,420 $1,015,830 $1,406,763

IP
(3,3)-2 $396,249 $6,558,932 $1,047,865 $1,444,114
(3,3)-3 $396,739 $6,596,701 $1,053,899 $1,450,638

CR
(1,1)-12 $391,642 $6,672,870 $1,066,067 $1,457,709
(1,1)-16 $407,351 $6,593,330 $1,053,360 $1,460,711
(1,1)-13 $380,320 $6,930,120 $1,107,166 $1,487,486
(1,1)-15
(2,2)-5
(2,4)-4
$430,710
$411,550
$419,315
US
$6,699,780
$6,831,841
$6,819,901
$1,070,367
$1,091,465
$1,089,557
$1,501,077
$1,503,015
$1,508,872
AN
(1,1)-10 $431,890 $6,822,400 $1,089,957 $1,521,847
(3,3)-1 $439,534 $6,851,883 $1,094,667 $1,534,201
(1,1)-8 $409,701 $7,128,180 $1,138,808 $1,548,509
(1,1)-5 $468,691 $6,778,280 $1,082,908 $1,551,599
M

(1,1)-1 $433,033 $7,049,670 $1,126,266 $1,559,299

(2,2)-6 $349,922 $7,599,691 $1,214,138 $1,564,060
ED

(1,1)-6 $484,550 $6,876,410 $1,098,585 $1,583,135

(2,2)-7 $405,428 $7,439,371 $1,188,525 $1,593,953
(1,1)-11 $420,221 $7,377,790 $1,178,687 $1,598,908
PT

(2,2)-8 $467,562 $7,367,311 $1,177,012 $1,644,574

(1,1)-2 $471,568 $7,343,300 $1,173,176 $1,644,744
(1,1)-9 $340,162 $8,170,600 $1,305,347 $1,645,509
CE

(1,1)-14 $540,964 $6,968,840 $1,113,352 $1,654,316

(1,1)-3 $441,221 $7,669,890 $1,225,353 $1,666,574
(2,2)-1 $518,481 $7,528,677 $1,202,792 $1,721,273
AC

(2,2)-2 $508,555 $7,593,137 $1,213,091 $1,721,646

(1,1)-17 $387,695 $8,516,530 $1,360,613 $1,748,308
(2,2)-3 $547,572 $7,604,967 $1,214,981 $1,762,553

5. Conclusion
 ACCEPTED MANUSCRIPT

Systems involving reaction and separation tasks are a significant part of chemical
engineering processes with huge cost and energy reduction potentials if systematically
approached through simultaneous synthesis of the reaction and distillation sections. As an
initial step of any approach, the generation of all possible and feasible flowsheets must be
attended to. In this paper, a new systematic algorithm is presented to generate a much
broader set of sequences than ever before for a system involving a reactor and several
distillation columns. An attempt has been made to create a unique matrix for a reaction and
separation system by the inclusion of a single two component reaction to the SM already

T
developed and introduced by the authors. The matrix thus developed enables systematic

IP
generation of all potential sequences which could lead to the selection of the optimum. In

CR
this way non-synthesizable matrices are also predictable. Then, an applied case has been
studied as an example to demonstrate the performance of the proposed algorithm. In this

US
case study, the optimal sequence structure is introduced by checking possible schemes. By
the assessments done regarding the separation configuration alternatives for the cases of
AN
the studied case study where the reactor is improvised at the beginning of the process, 9
potentially better separation alternatives for the conventional NGL recovery process could
be suggested. Although the reaction studied in this research is a basic case it is the preface
M

of studying more complex reactions, there is plenty of room for further development of the
procedure to larger processes including more complex reactions. In this way, efforts have
ED

been made on the expansion of this algorithm.

PT
CE
AC
 ACCEPTED MANUSCRIPT

References

Achenie, L. K. E., & Biegler, L. T. (1990). A superstructure based approach to chemical reactor
network synthesis. Computers & Chemical Engineering, 14, 23-40.
Agrawal, R. (1996). Synthesis of Distillation Column Configurations for a Multicomponent Separation.
Industrial & Engineering Chemistry Research, 35, 1059-1071.

T
Agrawal, R. (2003). Synthesis of multicomponent distillation column configurations. AIChE Journal,
49, 379-401.

IP
Barnicki, S. D., & Siirola, J. J. (2004). Process synthesis prospective. Computers & Chemical
Engineering, 28, 441-446.

CR
Bertran, M. O., Frauzem, R., Sanchez-Arcilla, A. S., Zhang, L., Woodley, J. M., & Gani, R. (2016). A
generic methodology for processing route synthesis and design based on superstructure
optimization. Computers and Chemical Engineering.
Caballero, J. A., & Grossmann, I. E. (2004). Design of distillation sequences: from conventional to

2329. US
fully thermally coupled distillation systems. Computers & Chemical Engineering, 28, 2307-

Cabrera-Ruiz, J., Miranda-Galindo, E., Segovia-Hernandez, J. G., Hernandez, S., & Bonilla-Petriciolet,
A. (2011). Evaluation of stochastic global optimization methods in the design of complex
AN
distillation configurations.
Chitra, S. P., & Govind, R. (1985). Synthesis of optimal serial reactor structures for homogeneous
reactions. Part I: Isothermal reactors. AIChE Journal, 31, 177-184.
M

Cremaschi, S. (2015). A perspective on process synthesis: Challenges and prospects. Computers &
Chemical Engineering, 81, 130-137.
Demirel, S. E., Li, J., & Hasan, M. M. F. (2017). Systematic process intensification using building
blocks. Computers and Chemical Engineering, 105, 2-38.
ED

Douglas, J. M. (1985). A hierarchical decision procedure for process synthesis. AIChE Journal, 31,
353-362.
Fidkowski, Z. T. (2006). Distillation configurations and their energy requirements. AIChE Journal, 52,
PT

2098-2106.
Giridhar, A., & Agrawal, R. (2010a). Synthesis of distillation configurations. II: A search formulation
for basic configurations. Computers & Chemical Engineering, 34, 84-95.
CE

Giridhar, A., & Agrawal, R. (2010b). Synthesis of distillation configurations: I. Characteristics of a

good search space. Computers & Chemical Engineering, 34, 73-83.
Glasser, D., Crowe, C., & Hildebrandt, D. (1987). A geometric approach to steady flow reactors: the
attainable region and optimization in concentration space. Industrial & Engineering
AC

Chemistry Research, 26, 1803-1810.

Hopley, F., Glasser, D., & Hildebrandt, D. (1996). Optimal reactor structures for exothermic reversible
reactions with complex kinetics. Chemical Engineering Science, 51, 2399-2407.
Horn, F. (1965). Attainable and non-attainable regions in chemical reaction technique. In CHIMICA &
L INDUSTRIA (Vol. 47, pp. 235-&): POLO VLE ABRUZZI 5, 20131 MILAN, ITALY.
Ivakpour, J., & Kasiri, N. (2009). Synthesis of Distillation Column Sequences for Nonsharp
Separations. Industrial & Engineering Chemistry Research, 48, 8635-8649.
Jiménez, A., Ramírez, N., Castro, A., & Hernández, S. (2003). Design and Energy Performance of
Alternative Schemes to the Petlyuk Distillation System. Chemical Engineering Research and
Design, 81, 518-524.
 ACCEPTED MANUSCRIPT

Kauchali, S., Rooney, W. C., Biegler, L. T., Glasser, D., & Hildebrandt, D. (2002). Linear programming
formulations for attainable region analysis. Chemical Engineering Science, 57, 2015-2028.
Khalili-Garakani, A., Ivakpour, J., & Kasiri, N. (2016). A new search space reduction method based on
exergy analysis for distillation columns synthesis. Energy, 116, Part 1, 795-811.
Khumalo, N., Glasser, D., Hildebrandt, D., Hausberger, B., & Kauchali, S. (2006). The application of
the attainable region analysis to comminution. Chemical Engineering Science, 61, 5969-
5980.
King, C. J. (1987). Separation processes based on reversible chemical complexation. In: Wiley: New
York.
Kokossis, A., Labrador-Darder, C., & Cecelja, F. (2016). Semantically enabled process synthesis and
optimisation. Computers & Chemical Engineering, 93, 64-86.
Kokossis, A. C., & Christodoulos A, F. (1990). Optimization of complex reactor networks—I.

T
Isothermal operation. Chemical Engineering Science, 45, 595-614.
Kramer, K. (2012). Optimization-based synthesis of hybrid separation processes. Technical University

IP
of Aachen, Germany

Linke, P., & Kokossis, A. (2003a). Attainable reaction and separation processes from a

CR
superstructure-based method. AIChE Journal, 49, 1451-1470.
Linke, P., & Kokossis, A. (2003b). On the robust application of stochastic optimisation technology for
the synthesis of reaction/separation systems. Computers & Chemical Engineering, 27, 733-
758.
US
Marcoulaki, E., Linke, P., & Kokossis, A. (2001). Design of separation trains and reaction-separation
networks using stochastic optimization methods. Chemical Engineering Research and Design,
79, 25-32.
AN
Nishida, N., Stephanopoulos, G., & Westerberg, A. W. (1981). A review of process synthesis. AIChE
Journal, 27, 321-351.
Omtveit, T., Valaskova, Z., Wahl, P. E., & Lien, K. M. (1993). Decomposed algorithmic synthesis of
reactor-separation-recycle systems. Computers & Chemical Engineering, 17, S95-S100.
M

Papalexandri, K. P., & Pistikopoulos, E. N. (1996). Generalized modular representation framework for
process synthesis. AIChE Journal, 42, 1010-1032.
Park, J. H., Khan, M. S., Andika, R., Getu, M., Bahadori, A., & Lee, M. (2015). Techno-economic
ED

evaluation of a novel NGL recovery scheme with nine patented schemes for offshore
applications. Journal of Natural Gas Science and Engineering, 27, 2-17.
Perkins, J. D. (1989). Conceptual design of chemical processes J. M. Douglas, McGraw-Hill, New York,
PT

1988. pp. xviii + 601, price £ 32.50. ISBN 0-07-017762-7. Journal of Chemical Technology &
Biotechnology, 46, 249-249.
Peters, M. S., & Timmerhaus, K. D. (1991). Plant design and economics for chemical engineers:
McGraw-Hill.
CE

Ramírez-Corona, N., Jiménez-Gutiérrez, A., Castro-Agüero, A., & Rico-Ramírez, V. (2010). Optimum
design of Petlyuk and divided-wall distillation systems using a shortcut model. Chemical
Engineering Research and Design, 88, 1405-1418.
AC

Rathore, R. N. S., Vanwormer, K. A., & Powers, G. J. (1974). Synthesis of distillation systems with
energy integration. AIChE Journal, 20, 940-950.
Recker, S., & Marquardt, W. (2012). A Nonlinear Programming Approach to Conceptual Design of
Reaction-Separation Systems. In Computer Aided Chemical Engineering (Vol. 30, pp. 592-
596).
Recker, S., Skiborowski, M., Redepenning, C., & Marquardt, W. (2014). Systematic and Optimization-
Based Design of Integrated Reaction-Separation Processes. In J. D. S. Mario R. Eden & P. T.
Gavin (Eds.), Computer Aided Chemical Engineering (Vol. Volume 34, pp. 417-422): Elsevier.
Recker, S., Skiborowski, M., Redepenning, C., & Marquardt, W. (2015). A unifying framework for
optimization-based design of integrated reaction–separation processes. Computers &
Chemical Engineering, 81, 260-271.
 ACCEPTED MANUSCRIPT

Rong, B.-G., Kraslawski, A., & Turunen, I. (2003). Synthesis of Functionally Distinct Thermally Coupled
Configurations for Quaternary Distillations. Industrial & Engineering Chemistry Research, 42,
1204-1214.
Rooney, W. C., Hausberger, B. P., Biegler, L. T., & Glasser, D. (2000). Convex attainable region
projections for reactor network synthesis. Computers & Chemical Engineering, 24, 225-229.
Sargent, R. W. H. (1998). A functional approach to process synthesis and its application to distillation
systems. Computers & Chemical Engineering, 22, 31-45.
Seader, J. D., & Westerberg, A. W. (1977). A combined heuristic and evolutionary strategy for
synthesis of simple separation sequences. AIChE Journal, 23, 951-954.
Shah, V. H., & Agrawal, R. (2010). A matrix method for multicomponent distillation sequences. AIChE
Journal, 56, 1759-1775.
Smith, R. (2005). Chemical Process: Design and Integration: Wiley.

T
Soltani, & Shafiei. (2015). Isothermal Reactor Network Synthesis Using Coupled Non-Dominated
Sorting Genetic Algorithm-II (NSGAII) with Quasi Linear Programming (LP) Method. Iranian

IP
Journal of Chemical Engineering, IAChE, Vol. 12.
Steimel, J., & Engell, S. (2015). Conceptual design and optimization of chemical processes under

CR
uncertainty by two-stage programming. Computers & Chemical Engineering, 81, 200-217.
Thompson, R. W., & King, C. J. (1972). Systematic synthesis of separation schemes. AIChE Journal, 18,
941-948.
Tula, A. K., Babi, D. K., Bottlaender, J., Eden, M. R., & Gani, R. (2017). A computer-aided software-

105, 74-95. US
tool for sustainable process synthesis-intensification. Computers & Chemical Engineering,

Urzhuntsev, G. A., Ovchinnikova, E., Chumachenko, V. A., Yashnik, S. A., I. Zaikovsky, V., & Echevsky,
AN
G. (2014). Isomerization of n-butane over Pd–SO4/ZrO2 catalyst: Prospects for commercial
application (Vol. 238).
Wang, X., Wang, C., Jin, S., & Chen, H. (2006). Optimal operating conditions for reactor-separator-
recycle system. Scientific & Industrial Research, 65, 867-872.
M

Westerberg, A. W. (2004). A retrospective on design and process synthesis. Computers & Chemical
Engineering, 28, 447-458.
Yeomans, H., & Grossmann, I. E. (1999). A systematic modeling framework of superstructure
ED

optimization in process synthesis. Computers & Chemical Engineering, 23, 709-731.

Yuan, Z., Chen, B., & Gani, R. (2013). Applications of process synthesis: Moving from conventional
chemical processes towards biorefinery processes. Computers & Chemical Engineering, 49,
217-229.
PT

ŽivkoviĿ, L. A., & NikaĿeviĿ, N. M. (2016). A method for reactor synthesis based on process
intensification principles and optimization of superstructure consisting of phenomenological
modules. Chemical Engineering Research and Design, 113, 189-205.
CE
AC
 ACCEPTED MANUSCRIPT

List of figure captions

Figure 1. Sequence representation: a) STN and b) SM representation of a four-

component system

Figure 2. Separation matrices for a three product separation (general state with three

T
possibilities: (a) direct, (b) indirect, and (c) prefractionator

IP
Figure 3. Procedure of determination of the sequence structure from the separation

CR
matrix.

Figure 4. The SM after removal the product from the matrix

US
Figure 5. Several options for duplicate arrays
AN
Figure 6. Non-elimination of feed array.

Figure 7. Example of possible {O} option for .

M

Figure 8. Non-synthesizable matrices for the reaction IA→IAB.

ED

Figure 9. Generation of the lightest or heaviest system component at n-1th column

Figure 10.Separation matrix for a four-component system

PT

Figure 11. Determination of the arrays including both reactant and product element
CE

Figure 12. Separation matrix after removing the product from the arrays

Figure 13. General form of reaction/distillation matrix for D→C reaction occurring at
AC

(1,1) array

Figure 14. Generated reaction/distillation matrices for D→C reaction occurring at (1,1)
array.

Figure 15. General form of reaction/distillation matrix for D→C reaction occurring at
(2,2) array.
 ACCEPTED MANUSCRIPT

Figure 16. Non-synthesizable matrices for D→C reaction occurring at (2,2) array.

Figure 17. Generated reaction/distillation matrices for D→C reaction occurring at (2,2)
array.

Figure 18. General form of reaction/distillation matrix for D→C reaction occurring at
(3,3) array.

Figure 19. Generated reaction/distillation matrices for D→C reaction occurring at (3,3)

T
array.

IP
Figure 20. Proposed sequences by algorithm for case study

CR
Figure 21. The proposed optimal sequence for case study.

Figure 22. Conventional scheme of NGL recovery

US
AN
M
ED
PT
CE

Graphical Abstract
AC