Download as pdf or txt
Download as pdf or txt
You are on page 1of 16

Separation and Purification Technology 257 (2021) 117907

Contents lists available at ScienceDirect

Separation and Purification Technology


journal homepage: www.elsevier.com/locate/seppur

Superstructure modeling and stochastic optimization of side-stream


extractive distillation processes for the industrial separation of benzene/
cyclohexane/cyclohexene
Hao Lyu a, Shihan Li b, Chengtian Cui a, c, Xingong Yu b, Jinsheng Sun a, *
a
School of Chemical Engineering and Technology, Tianjin University, Tianjin 300072, PR China
b
Henan Shouheng New Material Co. LTD, No.1 Industrial Area of Circular Economy, Xiangcheng, Xuchang 461700, PR China
c
Institute of Intelligent Manufacturing, Nanjing Tech University, Nanjing 211816, PR China

A R T I C L E I N F O A B S T R A C T

Keywords: The non-sharp side-stream extractive distillation (SSED) differs from the sharp one in that it partially, instead of
Benzene/cyclohexane/cyclohexene separation entirely, collects intermediate components through the side stream. Compared to the sharp one, the non-sharp
Adaptive selection of feasible flowsheets SSED is preferred in some extractive distillation (ED) systems with high boiling-point solvents. However, the
Superstructure-differential evolution
number of candidate configurations increases sharply when a design considers both the sharp and non-sharp
optimization
Non-sharp side-stream extractive distillation
SSEDs. Correspondingly, it results in a challenge to optimization, particularly in systems with multiple side-
stream options. In this study, we propose an adaptive superstructure-differential evolution (DE) method which
optimizes the sequences and parameters simultaneously. As a case study, the industrial benzene/cyclohexane/
cyclohexene separation is optimized, which has two extractive distillation columns and 30 feasible configura­
tions with different SSED options. Compared to the conventional superstructure-DE approach, the improved
method, which selects the feasible flowsheet instead of predetermining a random one, performs better in the
optimization. The less computational cost and better solutions demonstrate its superiority. Moreover, the opti­
mized 3-column and 4-column SSED configurations apply non-sharp side-stream options, which proves that the
non-sharp SSED options are more competitive for this ternary system. In summary, this study highlights the
significance to consider both the sharp and non-sharp SSED configurations and provides a systematic approach to
find high-quality configurations from an increased number of candidates.

operable alternative to reduce the remixing effect in ED systems [15].


Salvador et al. [16] found that the side-stream extractive distillation
1. Introduction (SSED) configuration, compared to the conventional extractive distilla­
tion (CED), could save 13.45%, 12.9%, and 12.45% of the TAC for
Extractive distillation (ED) is an extensively used alternative to ethanol dehydration, heptane/toluene separation with aniline as the
conventional distillation columns to separate mixtures with low relative solvent, and acetone/methanol separation with water as the solvent,
volatilities or azeotropes. It is usually more economical than other respectively. Wang et al. [17] demonstrated that the SSED configura­
methods such as pressure-swing distillation and azeotropic distillation tion, compared to the CED configuration, could save 28.67% of the TAC
[1–4]. However, its energy efficiency is still low, due to the continuous for the separation of acetonitrile/benzene/methanol with chloroben­
solvent regeneration and the remixing effect of intermediate zene as the solvent.
components. However, these works only considered the sharp SSED which entirely
The fully coupled configurations [5–7] are effective methods to collects intermediate components through the side stream. It usually
reduce the remixing effect. They were recommended for ED systems and requires a higher reflux ratio or more theoretical stages of the extractive
found to be economical in some cases [8,9]. However, their superiority distillation column (EDC), which leads to an increased cost. Besides,
was not widely accepted due to manufacturing and control difficulties without intermediate components, the elevated bottom temperature of
[10–12]. Agrawal and Fidkowski [13,14] and Agrawal [6] proposed the the EDC also increases the cost, in particular when the more expensive
liquid-only side-stream column, which could be a simpler and more

* Corresponding author.
E-mail address: jssun2006@vip.163.com (J. Sun).

https://doi.org/10.1016/j.seppur.2020.117907
Received 19 February 2020; Received in revised form 11 September 2020; Accepted 16 October 2020
Available online 22 October 2020
1383-5866/© 2020 Elsevier B.V. All rights reserved.
H. Lyu et al. Separation and Purification Technology 257 (2021) 117907

Abbreviations PSO particle swarming optimization


SRC solvent recovery column
ANE cyclohexane SSED Side-stream extractive distillation
BEN benzene
CED conventional extractive distillation Nomenclature and units
DE differential evolution AOC annualized operating cost, US$/a
DMAC dimethylacetamide BPB payback period, year
ED extractive distillation CAP capital investment, US$
EDC extractive distillation column F feed flow rate, kg/h
ENE cyclohexene NT number of theoretical stages
GA genetic algorithm P pressure, kPa
LMTD logarithmic mean temperature difference RR reflux ratio
LP low pressure S solvent flowrate, kg/h
MP middle pressure T temperature, ◦ C
MTD minimum temperature difference TAC total annualized cost, US$/a

steam is required. Therefore, the sharp SSED will be less economical SSED configurations. Since a side-stream column has more design
when the reduced cost of the solvent recovery column (SRC) could not specifications and fewer degrees of freedom than a CED configuration,
compensate for the increased cost of the EDC. Brito et al. [9] reported the probability of finding a feasible individual that matches the pre­
similar phenomena in the thermally coupled distillation configurations, determined flowsheets by stochastic approaches is relatively low.
which indicated that it is not economical to obtain the pure solvent from Therefore, a large part of the population is a convergence failure or an
the bottom of EDC in the cases. Consequently, the non-sharp SSED, in infeasible solution, which does no contribution to evolution but requires
which the side stream partially collects intermediate components and intensive computational cost. In this study, an adaptive superstructure-
the bottom stream still feeds the next column, could be an alternative. DE approach is proposed, which automatically selects the feasible
Since it considers both sharp and non-sharp SSEDs, the number of flowsheet from all possible candidates to design multi-component SSED
feasible candidate flowsheets increases rapidly, which leads to a higher configurations efficiently. As a case study, the industrial benzene/
computational cost for optimization. Only one distillation sequence is cyclohexane/cyclohexane (BEN/ANE/ENE) separation, a ternary ED
feasible in binary systems unless an intermediate solvent is used [18], process with two continuous EDCs and several potential heat integration
while both direct and indirect distillation sequences become feasible in options, is optimized. The result could prove the superior performance
ternary ED systems [19,20]. The different sequences lead to an increase of the optimization framework and the necessity to consider the non-
in the number of candidate configurations. Particularly, when the sharp SSED option as well as the sharp one.
ternary ED systems have two EDCs, such as benzene/cyclohexane/ In adipic acid and caprolactam plants, the BEN/ANE/ENE separation
toluene [21], tetrahydrofuran/ethanol/water [22], and acetonitrile/ accounts for ~70% of the overall hot utility consumption and produces
benzene/methanol [17], the number of feasible candidates increases to quite a lot of waste heat. Since the total adipic acid capacity has reached
several times larger than that of a binary system, which makes it a great 2,655,000 t/a in China by January 2020 (Supplementary Material S1),
challenge to obtain the optimal design. Moreover, since the performance the consumed steam for the separation reaches 8,750,000 t/a, although
of heat integration options is system-specific for ED systems [23–26], it some primary heat integration measures are already in use. According to
requires to compare the configurations with different heat integration Modla and Lang [37], the consumed hot steam will lead to ~880,000 t/a
options, which will further increase the number of candidates. CO2 emission. Although the data emphasized the great significance for
Apart from comparing a large number of candidate configurations, it further improvement on the separation, it attracted little attention in the
is also a hard task to optimize the structural and operating parameters of previous works. Moreover, the number of feasible configurations for this
each configuration. Since the number of decision variables of a SSED process is quite large, when the direct and indirect distillation se­
configuration is large and the influence of them is usually not monotonic quences, the sharp and non-sharp SSEDs, as well as the corresponding
or independent [27–29], it becomes necessary to optimize the decision heat integration options are all taken into consideration. The challenge
variables simultaneously, which is beyond the capability of heuristic of this optimization problem is enough to prove the superior perfor­
methods based on the rule of thumb. Both stochastic and deterministic mance of the improved superstructure-DE method.
algorithms can be used to solve the optimization problem [30]. This paper is organized as follows: Section 2 introduces the con­
Compared to the deterministic methods, the stochastic ones have better ventional industrial process and the design basis. Section 3 provides
performance in dealing with integer variables and nonconvex objective information about the benchmark and some other CED configurations.
functions with many local optimal solutions. Moreover, the stochastic Section 4 takes a glance into the various SSED configurations and in­
methods are easier to be initialized and more robust against convergence troduces the fundamentals of the improved superstructure-DE method.
failures, which is more difficult for deterministic methods due to the The Comparison of the adaptive superstructure-DE method with some
calculation of derivatives. Therefore, the stochastic algorithms, such as other known approaches demonstrates its superior performance. With
genetic algorithm (GA), differential evolution (DE), particle swarm the improved method to optimize the various SSED configurations
optimization, and their variants [31–35], are gradually accepted in simultaneously, it obtains the 3-column and 4-column optimal designs.
recent years, despite the disadvantageous slow descent in the neigh­ The results could help to demonstrate the necessity to consider both non-
borhood of an optimum and the expensive computational cost. sharp and sharp SSEDs and provide a systematic approach to design
When it requires to optimize dozens of candidates separately, the multi-component SSED configurations with enormous different variants
computational cost of the stochastic optimization becomes unaccept­ efficiently. Finally, Section 5 draws the conclusion.
able. Consequently, Lashkajani et al. [36] proposed a method, which
combines stochastic algorithms and superstructure optimization, to
optimize operating parameters, structural parameters, and sequences
simultaneously. However, the hybrid approach will be less efficient for

2
H. Lyu et al. Separation and Purification Technology 257 (2021) 117907

Fig. 1. The residual curve maps and x-y diagrams of the ternary systems (in mole). (a) residual curves map (b) x-y diagrams for the binary mixtures without DMAC.
(c) x-y diagrams for each binary mixture with equivalent molar DMAC.

2. Process and problem description molecular weights, and very close boiling points. The boiling points of
BEN, ENE and ANE at 1 atm are 80.1 ◦ C, 83.0 ◦ C and 80.7 ◦ C, respec­
2.1. Physical properties of BEN/ENE/ANE ternary system tively. Their residual curve maps represent two azeotropic points and a
distillation boundary (Fig. 1a), which excludes conventional distillation
BEN, ANE, and ENE are all C6s with similar molecular structures, from candidate separations. Besides, the x-y diagrams of the binary

3
H. Lyu et al. Separation and Purification Technology 257 (2021) 117907

Fig. 2. The 4-column ED process for BEN/ANE/ENE separation.

Table 1
The design specifications for BEN/ANE/ENE separation.
Specifications Values

minimum temperature reboilers and 15 ◦ C


difference condensers
pressure drop condensers 5 kPa
theoretical stage 0.3 kPa
design specifications (in BEN separating column top: BEN/ENE ≤ 0.001
mass fraction) (C1) bottom: ENE/
BEN ≤ 0.005
ANE column (C3) top: ENE ≤ 0.007
DMAC ≤ 20 ppm
bottom: ANE/
ENE ≤ 0.00125
solvent regeneration top: trace DMAC
(C2 & C4) bottom:
DMAC ≥ 0.9999

systems show that they are pressure-insensitive with very low relative
volatilities, which means that pressure-swing distillation is infeasible
(Fig. 1b). In industrial practice, ED is used to separate the ternary
mixture with dimethylacetamide (DMAC, CH3CON(CH3)2) as the sol­
vent, of which the boiling point at 1 atm is 166.1 ◦ C. After the equivalent
molar DMAC is added, the relative volatility of each binary system is
enlarged (Fig. 1c), which indicates that ED is feasible to separate the
mixture.
Fig. 3. The architecture of the DE algorithm combined with Aspen Plus.
2.2. The industrial 4-column process
to estimate the missing parameters, due to the lack of available phase
In adipic acid/caprolactam plants, the BEN/ANE/ENE separation equilibrium data between some pairs within the range of the operating
serves as the downstream treatment of the benzene hydrogenation unit. pressure (Supplementary Material S2). The design specifications are
The 4-column ED process uses DMAC as the solvent to collect ENE for shown in Table 1. The minimum temperature difference (MTD), as well
further procedures, remove ANE, and recycle BEN back to the hydro­ as the pressure drop for heat exchangers and trays, are defined as con­
genator (Fig. 2). Due to the large temperature difference along the col­ stant for simplifications. The light ends are mainly methylcyclopentane
umns, it is difficult for the reboilers of the process to reuse the waste heat (the boiling point is 71.8 ◦ C at 1 atm), with trace water, nitrogen, and
from the overhead vapors and the hot solvent. Consequently, the hot light hydrocarbon. In this study, the light ends are treated as methyl­
solvent from the bottom of the solvent recovery columns (SRC) is first cyclopentane for simplification. The temperatures of the inlet and outlet
transferred to the downstream unit to heat a reboiler, of which the streams of the reboiler in the downstream unit, which is integrated to
temperature is within the range of 75 to 85 ◦ C, and cooled by cooling recover the waste heat, are 66.8 and 76.6 ◦ C, respectively, according to
water before recycled to the EDCs. the simulation results in Supplementary Material S3.

2.3. Design specifications 2.4. Economic evaluation

The process simulation in this study adopts NRTL thermodynamic TAC is used to evaluate the economic performance of the candidates.
model. The binary parameters are obtained from the Aspen built-in or The calculation is mainly based on Luyben [38,39], with the details in
the corresponding literature. UNIFAC group contribution method is used Supplementary Material S4. Since the separation of the ternary system is

4
H. Lyu et al. Separation and Purification Technology 257 (2021) 117907

Fig. 4. The 4-column CED configuration (benchmark).

variables are dependent and calculated by Aspen Plus based on the


Table 2
design specifications.
Cost analysis for the CED configurations.
4-column 3-column 3. The benchmark and conventional alternatives
configuration configuration

column shells and internals/(×106 US$) 6.07 7.40 3.1. The industrial 4-column process (Benchmark)
heat exchangers/(×106 3.88 3.63
US$) The original 4-column CED configuration, which is designed by the
exhausting operation 0.06 0.11 rule of thumb, is far from an economical solution. For a fair comparison,
(×106 US$/a) it is optimized by the DE algorithm and selected as the benchmark
heat utility requirement/ LP steam 31.9 34.5 (Fig. 4). The composition data of the streams in the flowsheets in Fig. 4
103 kW MP steam 17.2 16.9
and the other figures are all in mass fraction. The regenerated hot sol­
cold utility requirement/ cooling 34.4 37.3
vent is partially heat integrated with the downstream process to recover
103 kW water
the waste heat. Since this option reduces the total hot utility require­
annualized operating cost/(×106 US 11.57 12.12
$/a)
ment, the reduced costs will be accounted for in the economic evalua­
cost reduction by heat integration/ − 4.01 − 3.58
tion. In detail, the TAC is calculated by subtracting the reduced costs
(×106 US$/a) from the sum of the annualized operating and capital cost. The cost of
total annualized cost/ 10.93 12.31 the benchmark is listed in Table 2.
(×106 US$/a)
3.2. The 3-column CED configurations
*Detailed data could be found in the supplementary material S7.
In this ternary ED system with heavy solvent, there is another
integrated with its downstream unit to reuse the waste heat, the profit feasible distillation sequence, which separates ANE in the first EDC,
from the waste heat recovery will be taken into account. collects ENE in the second EDC, and recovers DMAC from BEN in the last
column (Fig. 5). This direct sequence only requires 3 columns, which
2.5. Simultaneous optimization seems to be more attractive in most ternary ED systems at first glance.
The 3-column CED configuration is optimized by the DE algorithm, and
It is difficult for heuristic methods to optimize more than 20 inde­ the results are shown in Table 2. The TAC of the 3-column scheme is
pendent variables simultaneously. Consequently, the differential evo­ 12.6% higher than the 4-column CED configuration. This means that the
lution (DE), a competitive stochastic algorithm, is used in this study. The direct distillation sequence is not always the better choice where it re­
optimization is implemented by combining an optimizer (Matlab, Py­ quires a much higher reflux ratio or capital cost to separate the lightest
thon, VBA, GPROMs, etc.) with Aspen Plus through a win32.com component from the other components.
interface (Fig. 3). The control factors of DE are determined by a self-
adaptive strategy [34]. The pseudo-code and the declaration of vari­
ables are provided in Supplementary Material S5 & S6. The other

5
H. Lyu et al. Separation and Purification Technology 257 (2021) 117907

Fig. 5. The 3-column CED configuration.

Fig. 6. The 4-column CED configuration (solvent feeds C1 top stage).

3.3. 4-Column CED configuration with heat integration options operated at elevated pressure, it could be integrated with the down­
stream process to recover the latent heat of the overhead vapor (Fig. 7).
Besides the benchmark, the DE algorithm can also find a 4-column Since this integration requires a much larger heat exchange area than
CED configuration that feeds the solvent to the C1 top (Fig. 6). In this cooling the overhead vapor by water directly, a 6.1% increase in the
configuration, there is no section in C1 to remove DMAC. As a result, C1 corresponding capital cost is observed (Table 3). In return, it recovers
top feeds C3 with significantly higher DMAC content. Although no sig­ ~50% more waste heat for the downstream process than other 4-column
nificant TAC reduction is observed with this configuration, the boiling CED configurations. The TAC of this configuration is 8.75 million US$/a,
point of the C1 top product increases, which indicates that the latent which is 19.9% less than the 4-column CED configuration. This result
heat of the overhead vapor could be recovered. When the C1 top is indicates that the heat integration option should be considered in further

6
H. Lyu et al. Separation and Purification Technology 257 (2021) 117907

Fig. 7. The 4-column heat-integrated CED configuration.

considered to be a reason for the low energy efficiency. In the EDCs, the
Table 3 content of intermediate components in the liquid phase increases stage
Cost analysis for 4-column CED configuration with the solvent feeding C1 top by stage from the feed location towards the bottom. It reaches a peak
stage. value in the middle of the stripping sections, where the lighter compo­
Without heat Heat nents are nearly completely separated. While below the peak stage, it
integration integrated begins to decrease, resulting in a lower content at the bottom. Then the
column shells and internals/(×106 US$) 5.81 6.14 bottom product is fed to the following column for regeneration or sep­
heat exchangers/(×106 US 3.93 4.19 aration, and the content of intermediate components increases from the
$) feed location to the top. In this path, the non-monotonically change in
exhausting operation (×106 0.06 0.05 the content of intermediate components should be considered as a
US$/a) remixing effect.
heat utility requirement/ LP steam 31.9 34.4 To reduce the remixing effect, SSED configurations are applied.
103 kW MP steam 17.9 14.8
However, being much more complex than a binary system, the SSED has
cold utility requirement/ cooling 34.4 24.2 several different schemes. In 3-column configurations, since the top
103 kW water
products do not contain the solvent and have low dew points, the heat
annualized operating cost/(×106 US$/a) 11.74 11.47
integration option is not feasible. The possible configurations can be
cost reduction by heat integration/(×106 − 3.97 − 6.18 distinguished based on their different side-stream connection ways and
US$/a)
coded by a vector [SC1, SC2] as shown in Fig. 9. The value of SC1 can be
total annualized cost/(×106 11.07 8.75
selected from [0, 1, 2, 3]: 0 represents the case that C1 has no side
US$/a)
stream, 1, 2 and 3 represent the cases that the bottom of the side-stream
*Detailed data could be found in the supplementary material S7. column C1 feeds C2, C3 and the solvent cooler, respectively. The value
of SC2 could be selected from [0, 1, 2]: 0 represents the case that C2 has
optimization. no side stream, 1 and 2 represent the cases that the bottom of the side-
stream column C2 feeds C3 and the solvent cooler, respectively. In this
4. The SSED configurations and the superstructure-DE way, 12 (4 × 3) different 3-column configurations are found.
optimization Similarly, 9 (3 × 3) different configurations are feasible for 4-column
flowsheets (Fig. 10). A vector [SC1, SC3] codes and distinguishes the
4.1. The remixing effect and SSED configurations different configurations based on their different side-stream connection
ways. The value of SC1 can be selected from [0, 1, 2]: 0 represents the
According to the liquid composition profiles along the columns, the case that C1 has no side stream, 1 and 2 represent the cases that the
remixing effect of the intermediate components in EDCs is observed in bottom of the side-stream column C1 feeds C2 and solvent cooler,
both the direct and indirect distillation sequences (Fig. 8), which is respectively. The value of SC3 can be selected from [0, 1, 2]: 0 represents
the case that C3 has no side stream, 1 and 2 represent the cases that the

7
H. Lyu et al. Separation and Purification Technology 257 (2021) 117907

Fig. 8. Liquid composition profiles for CED configurations. 4-column CED configuration (b) 3-column CED configuration.

8
H. Lyu et al. Separation and Purification Technology 257 (2021) 117907

Fig. 9. The possible 3-column configurations.

9
H. Lyu et al. Separation and Purification Technology 257 (2021) 117907

Fig. 10. The possible 4-column configurations.

10
H. Lyu et al. Separation and Purification Technology 257 (2021) 117907

Fig. 11. The superstructure for 3-column and 4-column flowsheets.

bottom of the side-stream column C3 feeds C4 and the solvent cooler, it can adaptively choose the appropriate schemes. In a sharp SSED col­
respectively. Moreover, heat integration options are available in 4-col­ umn, a feasible solution vector X should match 3 design specifications
umn configurations. The configurations that fully couple the overhead DS1, DS2, and DS3, where DS1, DS2, and DS3 are the purity requirements
vapor of the first EDC with the downstream unit are still the same as the of the top, side-stream, and bottom product, respectively. The multiple
basic ones. However, the boiling point of the overhead vapor is not al­ constraints lead to a very small feasible solution space V and make many
ways high enough to ensure that it can be cooled to liquid after heat tentative solutions X to be infeasible. If the vector X only matches 2
exchanging with the downstream unit. Consequently, it will require an design specifications, for example, DS1 and DS2, the X could be a feasible
extra condenser to cool and liquefy the top product, which is a partially solution of the non-sharp SSED column. Similarly, the vector X that
heat integrated design. Since the fully and partially heat integrated matches DS1 and DS3 may be a feasible solution of the CED column. If the
designs are both considered, each configuration has two extended optimization approach matches the feasible structure SC to the vector X,
structures. For example, the configuration [0,1,0] and [0,1,1], where the fewer solution vectors X will be excluded due to the mismatch. In this
third element of the vector could be 0 or 1, corresponding to the original way, the optimization will be more efficient and robust.
configuration and the partially heat integrated one, respectively. By adding the calculation rules for structure selection, we proposed
Therefore, the number of 4-column candidates increases to 18 an adaptive superstructure-DE method, of which the computing archi­
(3 × 3 × 2). tecture is shown in Fig. 12. During the sequential solving process, it
selects the feasible scheme based on the results of the front modules.
Both the outer code and the embedded Fortran code in Aspen Plus can
4.2. Improved superstructure-DE optimization for SSED configurations implement the scheme selection. The latter one is superior because of
the higher computational speed, and Supplementary Material S5 pro­
It is inefficient to optimize the enormous number of candidates vides the pseudo-code. In this way, the optimization of the superstruc­
separately. Consequently, the superstructure optimization is used to ture becomes simpler, which does not need to deal with the
help optimize the different candidates, together with their structural and superstructure code for scheme selection.
operating parameters simultaneously [36]. Since the dozens of SSED To demonstrate the superior performance of the improved
configurations with the direct distillation sequence are similar in their superstructure-DE framework, we optimize the 3-column configurations
main structures, except for a slight difference in the connection way of by different optimization methods and compare the results (Table.4). All
the specific streams, these configurations can be integrated as a corre­ computational tasks are run on an Intel Core i7-9700 CPU (3.00 GHz,
sponding superstructure with two splitters (Fig. 11). Similarly, the 16.0 GB RAM). The adaptive superstructure-DE approach performs best
configurations with the indirect distillation sequence can be merged into among the different methods. Although it takes a little more time to find
a superstructure. All the possible configurations could be obtained by a a high-quality solution than optimizing a single configuration ([1,1] in
corresponding vector [SC1, SC2], where SCi is the parameters of the ith Fig. 9), notably, the latter one will require ~300 h (25 h × 12 candi­
splitter. Based on the vector, it connects the side streams to corre­ dates) to optimize all feasible configurations separately. Besides, the
sponding destinations and can calculate various configurations in a su­ adaptive superstructure-DE approach can find significantly better solu­
perstructure simulation. tions than the conventional superstructure-DE approach within the close
However, the conventional superstructure-DE, which predefines the calculation time. In summary, the comparison demonstrates the superior
schemes, is not efficient to deal with the SSED configurations. In the performance of the adaptive superstructure-DE method, which can be an
conventional superstructure-DE optimization, the vector of independent improvement of the conventional approach. Therefore, the 4-column
variables is defined as [SC, X], where SC is the superstructure code and X SSED configurations are also optimized by the improved approach,
is the operating and structural parameters. Since the side-stream EDC and the result is shown in Fig. 14 and Table 5.
needs to match more constraints than the conventional EDC, the feasible The optimized 3-column SSED configuration (Fig. 13) has two non-
solution space is smaller, resulting in a lower probability of finding sharp side streams, which saves 12.6% of the TAC compared to the 3-col­
convergent and feasible results by the stochastic search. The tentative umn CED configuration. The reduced remixing effect is the main reason
solution X will be excluded unless the appropriate SC is selected, for the reduced cost. As shown in Fig. 15, the remixing effect from the
although sometimes the excluded X can be a high-quality solution if the side stream of C1 to the upper feed location of C2 and that from the side
appropriate SC is matched. A large number of nonconvergent or infea­ stream of C2 to the upper feed location of C3 are very small, because the
sible results could not promote the evolution of the population but side streams and corresponding feed locations have equivalent inter­
consume intensive computational resources. mediate component contents. When drawing the side streams from C1
Since many infeasible solutions [SC, X] will be feasible solutions and C2, the flow rates of their bottom streams are reduced, which lowers
when a different SC is matched, the optimization will be more efficient if

11
H. Lyu et al. Separation and Purification Technology 257 (2021) 117907

Fig. 12. The adaptive superstructure-DE framework to solve the 3-column and 4-column superstructures. (a) 3-column SSED (b) 4-column SSED.

12
H. Lyu et al. Separation and Purification Technology 257 (2021) 117907

Table 4 Table 5
The performance for different optimization approaches. Cost analysis for the SSED configurations.
Optimization approach Trial number of computational results 3-column 4-column
iterations time configuration configuration

DE optimizes the [1,1] 1 500 26.4 h 11.10 column shells and internals/(×106 US$) 6.97 6.05
configuration heat exchangers/(×106 3.10 4.02
2 500 24.2 h 10.84 US$)
3 500 24.0 h 11.26
exhausting operation 0.09 0.05
DE optimizes the [3,2] 1 500 23.5 h 12.79
configuration (×106 US$/a)
2 500 25.4 h 12.13 heat utility requirement/ LP steam 0.0 18.0
3 500 25.9 h 12.36 103 kW MP steam 41.5 27.8
conventional 1 500 26.9 h 11.96 cold utility requirement/ cooling 31.1 22.6
superstructure-DE 103 kW water
2 500 28.6 h 12.19
annualized operating cost/(×106 US 10.14 10.84
3 500 27.5 h 11.70
$/a)
adaptive superstructure- 1 500 26.6 h 11.12
DE cost reduction by heat integration/ − 2.82 − 5.72
2 500 30.1 h 11.00 (×106 US$/a)
3 500 28.8 h 10.76 total annualized cost/ 10.76 8.53
(×106 US$/a)
*The detailed values of the variables of each trial is in Supplementary Material
S8. *Detailed data could be found in the supplementary material S7.

Fig. 13. The 3-column SSED configuration.

the remixing effect. Especially for C2, since most BEN is collected hand, the heat duty of the C3 reboiler, which always uses MP steam,
through the side stream, the remixing effect is almost eliminated. decreases rapidly when the remixing effect is almost eliminated. The
Meanwhile, when drawing the side stream, the temperature of the C1 reduced cost of C3 is much higher than the increased cost of C2 to draw
bottom increases. If more intermediate components are collected the side stream, and it shows an economical performance to draw the
through the side stream, the elevated bottom temperature will require side stream at a very high flow rate. However, since it requires a much
more expensive hot utility. Since the heat duty of the C1 reboiler is much higher reflux ratio and more stages to collect pure DMAC from C2 bot­
larger than others, the usage of the higher-quality hot utility will cause a tom, it is still more economical for C2 to apply a non-sharp side-stream
great increase in its operating cost, which could not be covered by the configuration instead of a sharp one.
reduced cost of C2. Therefore, the sharp SSED is not economical in this Different from the 3-column SSED configuration, the optimized 4-
case. Different from C1, most of the intermediate component in C2 is column one has only one side stream (Fig. 14). When C1 draws a
collected through the side stream, resulting in the great increase of the sharp side stream, the increase in solvent content at the bottom will lead
temperature of the C2 bottom. However, since the reboiler duty is much to an elevated reboiler temperature, which means that more expensive
lower than those of C1 and C3, the increase of its operating cost will not hot utilities are required. Since the reboiler duty of C1 is much larger
be so high when MP steam is used instead of LP steam. On the other than that of C2, the sharply increased cost of C1 could not be covered by

13
H. Lyu et al. Separation and Purification Technology 257 (2021) 117907

Fig. 14. The 4-column SSED configuration.

the reduced cost of C2, making the sharp SSED less competitive. Simi­ optimize a single configuration, and the results are much better than
larly, the sharp SSED is not economical for C3. Moreover, since the those obtained by the conventional superstructure-DE method.
temperature of the C3 reboiler is close to that of the feed location in C4 Using the adaptive superstructure-DE method, the 3-column and 4-
and the use temperature of LP steam and the temperature, even a slight column SSED configurations are optimized, with 12 and 18 different
temperature increase will significantly increase the cost. As a result, C3 feasible candidates being considered, respectively. The optimized re­
does not draw the side stream and the 4-column SSED configuration has sults do not apply any sharp side-stream options. The 3-column
only one side stream. In this configuration, the side stream only collects configuration has two non-sharp side-stream options, while the 4-col­
a small portion of intermediate components, and there is still a signifi­ umn configuration has only one non-sharp side-stream option, of
cant remixing effect, which leads to the TAC of the 4-column SSED which the second EDC does not draw a side stream. This result shows the
configuration close to the heat integrated 4-column CED configuration. necessity of considering different SSED options and highlights the sig­
nificance of an efficient optimization tool. Although the optimization
5. Conclusion efficiency has been improved greatly, the adaptive superstructure-DE
approach still takes more than 1 day to obtain the result, due to the
In this paper, the sharp and non-sharp SSED configurations are both inefficient random searching. Better searching and solving strategies
considered as the alternatives to the CED ones for the industrial BEN/ have to be developed in further work.
ANE/ENE separation with DMAC as the solvent. The optimization
problem involves 30 different configurations, which have different
distillation sequences, side-stream column connection ways, and heat Declaration of Competing Interest
integration options, together with more than 20 independent variables
that need to be optimized simultaneously. To solve the complex opti­ The authors declare that they have no known competing financial
mization efficiently, we propose an improved superstructure-DE interests or personal relationships that could have appeared to influence
method, which can adaptively select the feasible configuration for the work reported in this paper.
different structural and operating parameters. The adaptive
superstructure-DE method and the conventional superstructure-DE one Acknowledgment
are used to optimize the 3-column configuration, and the results
demonstrate the superior performance of the adaptive method. It can The authors appreciate Benbin Li, Yongfeng Xu, Guoyao Jing in the
obtain high-quality solutions from all possible candidates within the China Pingmei Shenma Group Nylon Technology Co., Ltd. for providing
equivalent computational time required for the DE algorithm to basic feedstock composition information of this research. This research
did not receive any specific grant from funding agencies in the public,

14
H. Lyu et al. Separation and Purification Technology 257 (2021) 117907

Fig. 15. Liquid composition profiles for SSED configurations. (a) 3-column SSED configuration (b) 4-column SSED configuration.

commercial, or not-for-profit sectors. The authors also appreciate References


Dongxiao Zhang, for providing help for code debug.
[1] R. Muñoz, J.B. Montón, M.C. Burguet, et al., Separation of isobutyl alcohol and
isobutyl acetate by extractive distillation and pressure-swing distillation:
Appendix A. Supplementary material simulation and optimization, Sep. Purif. Technol. 50 (2) (2006) 175–183, https://
doi.org/10.1016/j.seppur.2005.11.022.
Supplementary data to this article can be found online at https://doi. [2] W.L. Luyben, Comparison of extractive distillation and pressure-swing distillation
for acetone-methanol separation, Comput. Chem. Eng 47 (8) (2008) 2696–2707,
org/10.1016/j.seppur.2020.117907. https://doi.org/10.1021/ie701695u.
[3] W.L. Luyben, I.L. Chien, Design and Control of Distillation Systems for Separating
Azeotropes, John Wiley & Sons, Inc., 2010. https://doi.org/10.1002/
9780470575802.

15
H. Lyu et al. Separation and Purification Technology 257 (2021) 117907

[4] W.L. Luyben, Comparison of extractive distillation and pressure-swing distillation [22] Y. Zhao, T. Zhao, H. Jia, et al., Optimization of the composition of mixed solvent
for acetone/chloroform separation, Comput. Chem. Eng. 50 (8) (2013) 1–7, for economic extractive distillation process in view of the separation of
https://doi.org/10.1016/j.compchemeng.2012.10.014. tetrahydrofuran/ethanol/water ternary azeotrope, J. Chem. Technol. Biotechnol.
[5] G. Dunnebier, C.C. Pantelides, Optimal design of thermally coupled distillation 92 (9) (2017) 2433–2444, https://doi.org/10.1002/jctb.5254.
columns, Ind. Eng. Chem. Res. 38 (1) (1999) 162–176, https://doi.org/10.1021/ [23] A.A. Kiss, S.J.F. Landaeta, C.A.I. Ferreira, Towards energy efficient distillation
ie9802919. technologies – Making the right choice, Energy 47 (1) (2012) 531–542, https://doi.
[6] R. Agrawal, Thermally coupled distillation with reduced number of intercolumn org/10.1016/j.energy.2012.09.038.
vapor transfers, AIChE J. 46 (11) (2000) 2198–2210, https://doi.org/10.1002/ [24] H. Luo, K. Liang, W. Li, et al., Comparison of pressure-swing distillation and
aic.690461112. extractive distillation methods for isopropyl alcohol/diisopropyl ether separation,
[7] J.A. Caballero, I.E. Grossmann, Optimal synthesis of thermally coupled distillation Ind. Eng. Chem. Res. 53 (2014) 15167–15182, https://doi.org/10.1021/
sequences using a novel MILP approach, Comput. Chem. Eng. 61 (2014) 118–135, ie502735g.
https://doi.org/10.1016/j.compchemeng.2013.10.015. [25] H. Luo, C.S. Bildea, A.A. Kiss, Novel heat-pump-assisted extractive distillation for
[8] L. Sun, Q. Wang, L. Li, et al., Design and control of extractive dividing wall column bioethanol purification, Ind. Eng. Chem. Res. 54 (2015) 2208–2213, https://doi.
for separating benzene/cyclohexane mixtures, Ind. Eng. Chem. Res. 53 (19) (2014) org/10.1021/ie504459c.
8120–8131, https://doi.org/10.1021/ie500291a. [26] X. You, I. Rodriguez-Donis, V. Gerbaud, Reducing process cost and CO 2 emissions
[9] K.D. Brito, G.M. Cordeiro, M.F. Figueirêdo, et al., Economic evaluation of energy for extractive distillation by double-effect heat integration and mechanical heat
saving alternatives in extractive distillation process, Comput. Chem. Eng. 93 pump, Appl. Energy 166 (2016) 128–140, https://doi.org/10.1016/j.
(2016) 185–196, https://doi.org/10.1016/j.compchemeng.2016.06.013. apenergy.2016.01.028.
[10] A.A. Kiss, D.J.P.C. Suszwalak, Enhanced bioethanol dehydration in extractive [27] M.F.D. Figueiredo, K.D. Brito, W.B. Ramos, et al., Effect of solvent content on the
dividing-wall columns, Sep. Purif. Technol. 86 (15) (2012) 70–78, https://doi.org/ separation and the energy consumption of extractive distillation columns, Chem.
10.1016/j.seppur.2011.10.022. Eng. Commun. 202 (9) (2015) 1191–1199, https://doi.org/10.1080/
[11] M. Xia, B. Yu, Q. Wang, et al., Design and control of extractive dividing-wall 00986445.2014.900053.
column for separating methylal–methanol mixture, Ind. Eng. Chem. Res. 51 (49) [28] W.L. Luyben, Distillation column pressure selection, Sep. Purif. Technol. 168
(2016) 16016–16033, https://doi.org/10.1021/ie3015395. (2016) 62–67, https://doi.org/10.1016/j.seppur.2016.05.015.
[12] M.B. Franke, Design of dividing-wall columns by mixed-integer nonlinear [29] C. Cui, J. Sun, Rigorous design and simultaneous optimization of extractive
programming optimization, Chem. Ind. Tech. 89 (2) (2017) 582–597, https://doi. distillation systems considering the effect of column pressures, Chem. Eng. Process.
org/10.1002/cite.201700005. 139 (2019) 68–77, https://doi.org/10.1016/j.cep.2019.04.001.
[13] R. Agrawal, Z.T. Fidkowski, More operable arrangements of fully thermally [30] M.B. Franke, Mixed-integer optimization of distillation sequences with Aspen Plus:
coupled distillation columns, AIChE J. 44 (11) (1998) 2565–2568, https://doi.org/ a practical approach, Comput. Chem. Eng. 131 (5) (2019) 1–7, https://doi.org/
10.1002/aic.690441124. 10.1016/j.compchemeng.2019.106583.
[14] R. Agrawal, Z.T. Fidkowski, New thermally coupled schemes for ternary [31] G. Modla, Energy saving methods for the separation of a minimum boiling point
distillation, AIChE J. 45 (3) (1999) 485–496, https://doi.org/10.1002/ azeotrope using an intermediate solvent, Energy 50 (1) (2013) 103–109, https://
aic.690450306. doi.org/10.1016/j.energy.2012.12.011.
[15] A. Yang, T. Shi, S. Sun, et al., Dynamic controllability investigation of an energy- [32] M. Vázquez-Ojeda, J.G. Segovia-Hernández, S. Hernández, et al., Design and
saving double side-stream ternary extractive distillation process, Sep. Purif. optimization of an ethanol dehydration process using stochastic methods, Sep.
Technol. 225 (2019) 41–53, https://doi.org/10.1016/j.seppur.2019.05.063. Purif. Technol. 105 (2013) 90–97, https://doi.org/10.1016/j.seppur.2012.12.002.
[16] T.A. Salvador, M.H. Nancy, H. Juergen, et al., Design of an energy-efficient side- [33] C. Cui, Z. Xi, S. Liu, et al., An enumeration-based synthesis framework for multi-
stream extractive distillation system, Comput. Chem. Eng 102 (12) (2017) 17–25, effect distillation processes, Chem. Eng. Res. Des. 144 (2019) 216–227, https://doi.
https://doi.org/10.1016/j.compchemeng.2016.12.001. org/10.1016/j.cherd.2019.02.018.
[17] C. Wang, C. Guang, Y. Cui, et al., Compared novel thermally coupled extractive [34] C. Cui, X. Zhang, J. Sun, Design and optimization of energy-efficient liquid-only
distillation sequences for separating multi-azeotropic mixture of acetonitrile/ side-stream distillation configurations using a stochastic algorithm, Chem. Eng.
benzene/methanol, Chem. Eng. Res. Des. 136 (2018) 513–528, https://doi.org/ Res. Des. 145 (2019) 48–52, https://doi.org/10.1016/j.cherd.2019.03.001.
10.1016/j.cherd.2018.06.017. [35] N.V.D. Long, M.A. Qyyum, K. Qadeer, et al., Particle swarm optimization
[18] W.L. Luyben, Improved design of an extractive distillation system with an methodology for optimal distillation retrofit, J. Chem. Eng. Jpn. 52 (4) (2019)
intermediate-boiling solvent, Sep. Purif. Technol. 156 (2015) 336–347, https:// 333–341, https://doi.org/10.1252/jcej.18we164.
doi.org/10.1016/j.seppur.2015.10.020. [36] K.H. Lashkajani, B. Ghorbani, M. Amidpour, et al., Superstructure optimization of
[19] L.V. Ivanova, A.V. Timoshenko, V.S. Timofeev, Synthesis of flowsheets for the olefin separation system by harmony search and genetic algorithms, Energy 99
extractive distillation of azeotropic mixtures, Theor. Found. Chem. Eng. 39 (1) (2016) 288–303, https://doi.org/10.1016/j.energy.2016.01.045.
(2005) 16–23, https://doi.org/10.1007/s11236-005-0022-7. [37] G. Modla, P. Lang, Vapor compression for batch distillation: comparison of
[20] A.V. Timoshenko, E.A. Anokhina, A.V. Morgunov, et al., Application of the different working fluids, Ind. Eng. Chem. Res. 54 (3) (2015) 1081–1092, https://
partially thermally coupled distillation flowsheets for the extractive distillation of doi.org/10.1021/ie504023q.
ternary azeotropic mixtures, Chem. Eng. Res. Des. 104 (2015) 139–155, https:// [38] W.L. Luyben, C.C. Yu, Reactive Distillation Design and Control, John Wiley & Sons,
doi.org/10.1016/j.cherd.2015.07.007. Inc, 2008, https://doi.org/10.1002/9780470377741.
[21] W.L. Luyben, Control comparison of conventional and thermally coupled ternary [39] W.L. Luyben, Principles and Case Studies of Simultaneous Design, John Wiley &
extractive distillation processes, Chem. Eng. Res. Des. 106 (2016) 253–262, Sons, 2011 https://doi.org/10.1002/9781118001653.
https://doi.org/10.1016/j.cherd.2015.11.021.

16

You might also like