See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/257562498

Design and optimization of an ethanol dehydration process using stochastic

methods

Article  in  Separation and Purification Technology · February 2013

DOI: 10.1016/j.seppur.2012.12.002

CITATIONS READS

47 207

5 authors, including:

María Vázquez-Ojeda Juan Gabriel Segovia-Hernandez

Universidad de Guanajuato Universidad de Guanajuato
10 PUBLICATIONS   89 CITATIONS    194 PUBLICATIONS   1,655 CITATIONS   

SEE PROFILE SEE PROFILE

Arturo Hernandez-Aguirre Anton Alexandru Kiss

Centro de Investigación en Matemáticas (CIMAT) The University of Manchester
198 PUBLICATIONS   1,513 CITATIONS    213 PUBLICATIONS   3,739 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Automated Circuit Design View project

Modelling Dependence between Random Variables View project

All content following this page was uploaded by Anton Alexandru Kiss on 23 January 2019.

The user has requested enhancement of the downloaded file.

1 Design and optimization of an ethanol dehydration process using
2 stochastic methods
3
4 María Vázquez-Ojeda1, Juan Gabriel Segovia-Hernández1,*, Salvador Hernández1,
5 Arturo Hernández-Aguirre2, Anton Alexandru Kiss3
1
6 División de Ciencias Naturales y Exactas, Universidad de Guanajuato.
7 Noria Alta s/n, Cp. 36050, Guanajuato, Gto., México.
2
8 CIMAT, A.C., Departamento de Ciencias Computacionales, Callejón de Jalisco s/n, 36240, Mineral de
9 Valenciana, Guanajuato, Gto., México.
3
10 Alumnus of University of Amsterdam, The Netherlands.
11
12 Abstract
13 Due to the increasing demand for renewable fuels that are economically attractive, as well as part of the
14 quest for energy alternatives to replace carbon-based fuels, the purification of ethanol plays a key role.
15 This paper presents the design and optimization of a dehydration process for ethanol, using two separa-
16 tion sequences: a conventional arrangement and an alternative arrangement based on liquid-liquid
17 extraction. Both sequences were optimized using a stochastic global optimization algorithm (differential
18 evolution) implemented in Mathworks Matlab and coupled to rigorous process simulations carried out in
19 Aspen Plus. The economic feasibility of the two configurations was studied by changing the ethanol-
20 water composition in the analyzed feed stream. The results clearly demonstrate that significant savings
21 are possible by extraction when the ethanol content in the feed stream exceeds 10% mol (22 wt%).
22
23 Keywords: energy savings, process optimization, bioethanol dehydration, L-L extraction, distillation
24
25
26
27
28
29 * Author to whom all correspondence should be addressed, e-mail: [email protected], Phone: +52
30 (473) 732-0006 ext 8142.
1 1. Introduction
2 Ethanol is by far the most promising sustainable biofuel, with major advantages over all other fuel
3 alternatives (such as hydrogen) – as it can be easily integrated in the existing fuel systems as a 5-85%
4 mixture with gasoline that does not need any modification of the current engines. Brazil and United
5 States are major users and producers of bioethanol, and as such both countries together were responsible
6 for 88% of the world's ethanol fuel production in 2010. Remarkable, bioethanol is an environmentally-
7 friendly fuel with less greenhouse gases emissions than gasoline, but with similar energy power [1].
8 The bioethanol production at industrial scale relies on several processes, such as: corn-to-ethanol,
9 sugarcane-to-ethanol, basic and integrated lignocellulosic biomass-to-ethanol. However, according to
10 Pimentel [2], the corn and other food crops should be used for food as priority and not for ethanol
11 production. The ethanol production is claimed to increase the degradation of the environment as the corn
12 production causes more soil erosion than any other crop. Moreover, it can take more than 29% of energy
13 to produce one gallon of ethanol as compared to the energy content of a gallon of ethanol. According to
14 these and also other factors, it was concluded that ethanol production from subsidized U.S. corn is in
15 fact not a renewable energy source. Nigam and Singh [3] performed a review of all the available
16 literature up to date concerning liquid biofuels. Although biofuels seem not to be yet an economical
17 alternative to large-scale biofuel supply, there is still an urgent need to perform extensive research in
18 order to introduce more efficient processes by developing the technology, reducing the energy
19 requirements, decreasing emissions and production costs, and ultimately establishing biofuels as an
20 alternative for the future. Anhydrous ethanol is widely used in the chemical industry as a raw material in
21 chemical synthesis of esters and ethers, and as solvent in production of cosmetics, sprays, perfumery,
22 paints, medicines and food, among others. The most popular processes used in ethanol dehydration are:
23 heterogeneous azeotropic distillation using solvents such as benzene, pentane, iso-octane and
24 cyclohexane; extractive distillation with solvents and salts as entrainers; adsorption with molecular
25 sieves; and, processes that use pervaporation membranes [1]. The large-scale production is still
26 dominated by the extractive and azeotropic distillation despite recent advances in pervaporation and
27 adsorption with molecular sieves [4]. The large-scale production of bioethanol fuel requires energy
28 demanding distillation steps to concentrate the diluted streams from the fermentation step and to
29 overcome the azeotropic behavior of the ethanol-water mixture. The conventional separation sequence
30 consists of three distillation columns performing several tasks with high energy penalties: first a column
31 for the pre-concentration of ethanol, second a column for extractive distillation and a third column for
32 solvent recovery (Fig. 1). Considering the high costs of the ethanol dehydration, an optimization of the
1 process is performed – where the designs of distillation and extractive distillation are usually
2 characterized as being of large size problems, since the significant number of strongly nonlinear
3 equations. Note that typically ethanol is pre-concentrated using a conventional distillation column to
4 values close to the azeotrope of ethanol and water (about 95 wt% ethanol), followed by an extractive
5 distillation column where the desired purity (over 99.8 wt% according to the standards) is achieved for
6 ethanol, and a solvent recovery column.
7 Huang et al. [5] provide a review of separation process and technologies related to biorefining including
8 pre-extraction of hemicellulose and other value-added chemicals, detoxification of fermentation
9 hydrolyzates, and ethanol product separation and dehydration. With respect to the azeotropic nature of
10 ethanol–water mixture they conclude that desired processes with low energy consumption are the
11 extractive distillation with ionic liquids and hyperbranched polymers, adsorption with molecular sieve
12 and bio-based adsorbents. New configurations have been studied for the separation process of
13 bioethanol, such as arrangements based on thermally coupled columns and column sections
14 recombination to obtain sequences with significant savings in the capital cost compared to the classic
15 arrangement proposed in the literature [6]. It is also possible to reduce costs by energy integration,
16 studies carried out in extractive distillation column using a partial condenser (the steam is fed to the
17 second column) thus reducing the usage of steam and cooling water [7]. The alternative proposed here is
18 to replace the distillation column used conventionally by a liquid-liquid extraction column, followed by
19 a distillation column to recover the solvent employed. In this work we designed and optimized the
20 process of bioethanol dehydration using the conventional process (Fig. 1) as well as the alternative
21 separation process based on liquid-liquid extraction (Fig. 2). Moreover, different solvents [8-9] were
22 also evaluated in the liquid-liquid extraction column, while ethylene glycol was used as extractive agent
23 in the extractive distillation (ED). The design and optimization was carried out using, as a design tool, a
24 differential evolution algorithm with restrictions coupled with the process simulator Aspen Plus, for the
25 evaluation of the objective function, ensuring that all results obtained are rigorous. In this context,
26 stochastic optimization methods are playing an important role because they are generally robust
27 numerical tools that present a reasonable computational effort in the optimization of multivariable
28 functions; they are also applicable to unknown structure problems, requiring only calculations of the
29 objective function, and can be used with all models without problem reformulation [10, 11].
30
1 2. Approach and methodology
2 Recently, the bioethanol dehydration has been studied for concentrations of 5-10% wt of ethanol from
3 the fermentation step, by extractive and azeotropic distillation in dividing-wall columns that are able to
4 concentrate and dehydrate bioethanol in a single unit [4,12]. These sequences were optimized using
5 sequential quadratic programming (SQP) method. The results show energy savings around 10 and 20%
6 [4,12]. Ahmetovic et al. [13] presented the optimization of a plant producing corn-based ethanol, aimed
7 to reduce the energy requirements, the fresh water consumption and the discharge of wastewater. The
8 optimization was performed using mathematical programming techniques to optimize the energy
9 requirements and the network for the process water. Čuček et al. [14] carried out the integration of
10 different technologies, raw materials and energy for the dry milling process to produce ethanol from
11 corn and stover. The optimization was performed using mathematical programming techniques using the
12 process simulator MIPSYN. In terms of extractive distillation several authors have studied the vapor-
13 liquid equilibrium in a ternary mixture of ethanol-water and a solvent. Wang et al. [15] measured the
14 vapor pressure in water-ethanol mixtures, water-methanol and ethanol-methanol in the presence of an
15 ionic liquid. García-Herreros et al. [16] have recently reported the optimization of the extractive
16 distillation of ethanol using glycerol as an alternative solvent – the problem being formulated as mixed
17 integer nonlinear programming (MINLP). That work determined the number of stages in the columns
18 and the feed stages as discrete variables along with continuous variables, design variables and balance
19 variables of the model. The optimization problem was achieved by combining stochastic and
20 deterministic algorithms. Some of these methods can provide useful results; however, it would be
21 desirable to incorporate rigorous models in the synthesis procedures, in order to increase their industrial
22 relevance and scope of application, particularly for nonideal mixtures. These rigorous MINLP synthesis
23 models exhibit significant computational difficulties, such as the introduction of equations that can
24 become singular, the solution of many redundant equations, and the requirement of good initialization
25 points. The high levels of nonlinearities and nonconvexities in the MESH equations, thermodynamic
26 equilibrium equations, and convergence difficulties are common problems. These difficulties translate
27 into high computational times and the requirement of good initial guesses and bounds on the variables to
28 achieve convergence. Also, these approaches just deal with one objective, total annual cost in most of
29 the cases. Although Messac et al. [17] present good alternatives for the solution of this problem with a
30 multiobjective perspective using non-linear programming optimizers, clear limitations in the case of
31 solving bi-objective problems are shown by Martínez et al. [18]. On the other hand, stochastic
32 optimizers deal with multi-modal and non-convex problems in a very effective way to solve these
1 limitations. Stochastic optimization algorithms are capable of solving, robustly and efficiently;, the
2 challenging multi-modal optimization problems, and they appear to be a suitable alternative for the
3 design and optimization of complex separation schemes [19]. Several heuristic techniques for global
4 optimization mimicking biological evolution have emerged in the literature. Among the most
5 representative algorithms, we can mention the genetic algorithms (GA) [20], particle swarm
6 optimization (PSO) [21], and a new class of evolutionary methods called differential evolution (DE)
7 algorithms [22]. For different theoretical and practical problems, comparative studies have shown that
8 the performance of DE algorithms is clearly better than that of the other two algorithms (GA and PSO);
9 see, for example, Ali and Torn [23], Xu and Li [24], Panduco and Brizuela [25] and Sacco et al. [26].
10 Differential evolution is a stochastic global optimization algorithm developed in 1995, whose operations
11 and search principles are common to evolutionary algorithms [27]. DE works with a population of
12 vectors or individuals which undergo reproduction and mutation operations during several iterations or
13 generations. At the end of every generation the individuals with best fitness value are chosen to populate
14 the next generation. The first step of the DE algorithm is to create the initial population (called parents)
15 and to evaluate the fitness function value of all members (calculated on the function or problem being
16 optimized). Then, the main loop performs the following steps:
17 • For each parent a “mutant” is produced by applying the reproduction and mutation operators;
18 • The mutant fitness function is computed;
19 • The fitness value of the mutant and the parent are compared, and the best of the two (with better
20 fitness value) is copied into a temporary file;
21 • Once all parents are visited, the new population is created by replacing the current parents with the
22 members of the temporary file;
23 • The process is iterated until the stopping criteria is met. At termination the best solution is found in
24 temporary file.
25 The creation of a ‘mutant’ encompasses the following operations: for every dimension d of the problem,
26 three individuals are chosen at random (with no repetition), and a difference vector, hence the name
27 differential, is found by subtracting the second random from the first random vector. This difference is
28 added to the third random and some value K is obtained. Whether this value is inserted into the current
29 dimension d is decided via the crossover probability, CR. If the random number is smaller than CR then
30 insert the value K into the current dimension d of the mutant being generated. Otherwise, insert into the
31 current dimension d of the mutant a copy of the value at dimension d of the current parent. The mutation
32 and reproduction operations accomplish two goals: first, to maintain the diversity and exploration of the
1 population, and second, achieve convergence to the optimal through adaptive mutations. Upon
2 completion DE provides the best solution vector.
3 Numerous investigations have been performed using the differential evolution algorithm to solve
4 different problems in diverse areas of chemical engineering; some resent works include those presented
5 by Yerramsetty and Murty [28], Kheawhom [29], Bonilla-Petriciolet et al. [30], Kumar et al. [31], Vakili
6 et al. [32], Bhattacharya et al. [33], among many others. Their studies demonstrate how the differential
7 evolution algorithm enables to optimize different kinds of problems robustly and efficiently.
8 In this work we use the conventional ethylene glycol solvent for the extractive distillation. As described
9 hereafter, the design and optimization of conventional separation sequence (CSS) and the optional
10 separation sequence (OSS-I) were investigated using the differential evolution algorithm coupled to
11 Aspen Plus process simulator (Fig. 3). Once obtaining the best designs for each sequence, the energy
12 requirements and the total annual cost were determined.
13
14 3. Results and discussion
15 Four feeds are considered by changing the molar composition ethanol/water with a feed flowrate of
16 45.36 kmol/h (equivalent to a production rate of 0.34-2.60 ktpy bioethanol), at 25 °C and 1 atm. Table 1
17 shows the molar and mass compositions of the feed stream. Ethylene glycol is used as extractive
18 distillation agent, at same conditions. Additionally, three extraction solvents were evaluated for the
19 optional design (OSS-I): octanoic acid, octanol and iso-octanol (ethylhexanol). Although other solvents
20 were proposed by Offeman et al. [8-9], the three solvents chosen in this work showed the best results.
21 Note that the design objective is specified for the ethanol mass fraction of 0.999 or more, while for the
22 water, the chosen solvent, and ethylene glycol the objective is a mass fraction of 0.99 or more.
23
24 3.1. Optimization strategy
25 In the conventional separation sequence the goal is to minimize the total operating cost, meaning the
26 cost of steam plus the cooling water for each column. The minimization of the goal is subject to the
27 requirements of purity and recovery of ethanol, water and ethylene glycol – as follows:
28
29 min(CU) = f (SI, RRI, RDI, SFI, FS, SII, RRII, RDII, SFS, SFII, SIII, RRIII, RDIII, SFIII)
30 Subject to (1)
31 y m ≥ xm
32
1 where CU is the cost of utilities i.e. the cost of steam plus the cost of the cooling water of the sequence, S
2 is the number of stages in each column, RR and RD are reflux and flow of distillate in each of the three
3 columns that integrate the sequence , SF is the feed streams of each of the columns, FS is the flow of
4 solvent in the extractive distillation column, SFS is the solvent feed stream to the column, finally ym and
5 xm are the vectors of the purities and recoveries obtained and required of the m components respectively.
6 A total of 14 variables (7 continuous and 7 integer variables) were manipulated.
7 Just as in the conventional sequence, in the optional separation sequence (OSS-I), the goal is to
8 minimize the total operating cost. The minimization of the goal is subject to the requirements of purity
9 and recovery of ethanol, water, ethylene glycol and the solvent used in the liquid-liquid extraction.
10
11 min(CU) = f (C, SEX, FA, SI, RRI, RDI, SFI, FS, SII, RRII, RDII, SFS, SFII, SIII, RRIII, RDIII, SFIII)
12 Subject to (2)
13 y m ≥ xm
14
15 where CU is the cost of utilities of the sequence, C is the acid or alcohol chosen and FA flow of this acid
16 or alcohol, SEX is the number of steps to perform the liquid-liquid extraction, S is the number of stages in
17 each column, RR and RD are reflux and flow of distillate in each of the three columns, SF is the feed
18 streams of each of the columns, FS is the flow of solvent into the column extractive distillation, SFS is the
19 solvent feed stream to the column, finally ym and xm are the vectors of the purities and recoveries
20 obtained and required of the m components respectively. A total of 17 variables (8 continuous and 9
21 integer variables) were manipulated. Table 2 shows the integer variables and search range, while Table 3
22 shows the search ranges of continuous variables of the sequences studied. The search ranges are taken
23 based on practical considerations for equipment design as is referenced by Couper et al. [34], among
24 others.
25 The optimization of the proposed sequences CSS and OSS-I is performed using the differential
26 evolution algorithm (DE) coupled with constraints AspenONE Aspen Plus. This connection allows
27 obtaining rigorous optimal designs meeting the requirements of purity and recovery. The algorithm used
28 in this work is differential evolution. Fig. 3 shows the flow diagram of the algorithm coupled to the
29 Aspen Plus process simulator, while Fig. 4 shows the connection of Mathworks Matlab with AspenTech
30 Aspen Plus via MS Excel, including the flow of data between these programs. During the evolution of
31 the DE, the vector values of decision variables (Vx) are sent from Matlab to Microsoft Excel using DDE
32 (dynamic data exchange) by COM technology. These values are attributed in Excel to the corresponding
1 process variables (Vp) and then sent to Aspen Plus by a similar interface. Note that using the COM
2 technology it is possible to add code such that the applications behave as an Object Linking and
3 Embedding (OLE) automation server. After running the rigorous simulation, Aspen Plus returns to MS
4 Excel the vector of results (Vr). Finally, Excel returns the objective function (FOB) value to Matlab for
5 the DE procedure. Note that the CPU time is high for each optimization step due to the long
6 convergence time required by the Aspen Plus simulator that ensures rigorous process simulation results.
7 For the optimization of the sequences, a population of 100 individuals and 300 generations – amounting
8 30,000 function evaluations in total – were used with the parameters of crossover and mutation of 0.80
9 and 0.40 respectively [27]. The optimization was carried out on a Dell computer with Intel Core i7CPU
10 930 processor at 2.80 GHz, 6.00 GB of RAM, Windows 7 Ultimate, Matlab v7.8.0.347 (R2009a) and
11 Aspen One Aspen Plus V7.0.
12
13 3.2 Process comparison
14 The total annual costs (TAC) were calculated using the method of Guthrie (1968). For this work, a series
15 of general equations were used for the purchase cost of equipment and cost of utilities based on the
16 method of Guthrie presented by Turton et al. [35]. The installation cost is also updated by using the
17 CEPCI index of 556.4 (Jun 2010). The capital cost – purchase plus installation cost – is annualized over
18 a period, which is often referred to as the plant lifetime:
19
20 Annual capital cost = Capital cost / Plant life time (3)
21 TAC = Annual operating cost + Annual capital cost (4)
22
23 The operating costs were calculated based only on the utility costs (steam 3.17 $/GJ when the bottom
24 temperature is less than 150 °C and cooling water 0.16 $/GJ – values taken from Turton et al. [35]),
25 while the plant lifetime was considered ten years, with an operating time of 8400 hrs/yr.
26 The design objectives specified for the purity of ethanol, purity of water, solvent selected in the liquid-
27 liquid extraction stage and flowrate of ethylene glycol were achieved. Tables 4 and 5 show the purities
28 in all streams in mass fraction, while Tables 6 and 7 show the design parameters of the separation
29 sequences. Note that all sequences meet the target design specifications for both recovery and purity (see
30 the product streams). Analyzing the feed stream FI, it can be observed a decrease in the total investment
31 cost (18%) and in the total annual cost (14%) in the optional sequence (OSS-I) and for this arrangement,
32 lower energy requirements were obtained using octanol as solvent in the liquid-liquid extraction column.
1 The energy requirements of the CSS and OSS-I are presented in terms of utilities: steam and cooling
2 water in the reboilers and condensers, respectively. Analyzing the feed stream FII (see Table 6), it can
3 be observed a decrease in total investment cost and in the total operating cost of 34 and 32%
4 respectively (OSS-I). For this arrangement, lower energy requirements were obtained using octanol as
5 solvent in the liquid-liquid extraction column.
6
7 Although there are no energy savings for the optional sequence when working with the feed stream FI
8 and FII, there is a significant reduction in the cost of equipment, as well as the total annual cost when
9 compared to the conventional sequence. However, by decreasing the fraction of ethanol in the feed
10 stream, there is a clear opposite behavior for the FIII. The sequence OSS-I has higher costs for both
11 capital investment and annual operating cost. For the feed stream FIV the same behaviour is observed,
12 the arrangement OSS-I has higher costs with respect to the conventional sequence.
13 Not surprisingly the designs found have many stages and low reflux to minimize energy costs. By
14 reducing the reflux, the energy consumption decreases at the cost of increasing the number of stages in
15 the design. An alternative to be considered in the optimization process is to add a penalty function to
16 limit the maximum number of stages in the column when the change in the objective function is not
17 significant. In this way, it is possible to find solutions with minimal energy consumption with a smaller
18 or similar size to the processes implemented in the industry.
19 The results obtained for cases FIII and FIV are consistent with the experimental results presented by
20 Offeman et al. [8-9] for compositions with presence of significant amounts of ethanol in the ethanol /
21 water mixture, obtained in stage laboratory for liquid-liquid extraction in one step using different
22 extractants (which we tested in this work to a liquid-liquid extraction multistage). The results generated
23 indicate that the viability of the new process, compared with traditional dehydration of ethanol in two
24 distillation systems, depend of the extractant used in the liquid-liquid extraction step.
25
26 4. Conclusion
27 Stochastic global optimization methods – such as differential evolution algorithm – were successfully
28 used for optimizing the separation of ethanol and water, to obtain optimal process designs using the
29 rigorous Aspen Plus process simulator. The conventional sequence was optimized and a novel optional
30 sequence using liquid-liquid extraction systems was proposed and investigated.
31 Furthermore, the ethanol dehydration was studied for four different concentrations from the fermentation
32 step, ranging from 5 to 32 wt% ethanol. The results show considerable savings in total annual cost (32 %
1 approximately) for the proposed alternative systems, for a feed stream with 10% mol (22% wt) of
2 ethanol. When the ethanol in feed stream has a lower concentration, the optimized SSC sequence is
3 actually a better economic alternative. Until an improved extracting agent is found for the liquid–liquid
4 extraction column, this alternative process is limited when the concentrations of ethanol are 5-12% and
5 32 wt% in the feed stream, by its ability to trap the product of interest and it is directly related to the
6 content of the product of interest in the feed stream.
7 Finally the optional sequence optimizes the solvent employed in the liquid-liquid extraction column,
8 thus allowing to optimize not only its design and operating variables but also the use of different
9 compounds to perform the separation The same can be applied to the extractive distillation column,
10 where it is possible to analyze the behavior of different solvents.
11
12 References
13 [1] I.D. Gil, J.M. Gómez, G. Rodríguez, Control of an extractive distillation process to dehydrate
14 ethanol using glycerol as entrainer, Computers and Chemical Engineering 39 (2012) 129-142.
15 [2] D. Pimentel, Ethanol Fuels: Energy balance, economics, and environmental impacts are negative,
16 Natural Resources Research 12 (2003) 127-134.
17 [3] P.S. Nigam, A. Singh, Production of liquid biofuels from renewable resources, Progress in
18 Energy and Combustion Science 37 (2011) 52-68.
19 [4] A.A. Kiss, D.J-P.C. Suszwalak, Enhanced bioethanol dehydration by extractive and azeotropic
20 distillation in dividing-wall columns, Separation and Purification Technology 86 (2012) 70-78.
21 [5] H-J. Huanga, S. Ramaswamya, U.W. Tschirner, B.V. Ramarao, A review of separation
22 technologies in current and future biorefineries, Separation and Purification Technology, 62
23 (2008) 1-21.
24 [6] M. Errico, B.G. Rong Synthesis of new separation processes for bioethanol production by
25 extractive distillation, Separation and Purification Technology 96 (2012) 58-67.
26 [7] M. Taylor, P.C. Wankat Increasing the energy efficiency of extractive distillation, Separation
27 Science and Technology 39 (2005) 1-17.
28 [8] R.D. Offeman, S.K. Stephenson, D. Franqui-Espiet, J.L. Cline, G.H. Robertson, W.J. Orts,
29 Extraction of ethanol with higher alcohol solvents and their toxicity to yeast, Separation and
30 Purification Technology 63 (2008) 444–451.
1 [9] R.D. Offeman, S.K. Stephenson, D. Franqui-Espiet, J.L. Cline, G.H. Robertson, W.J. Orts,
2 Extraction of ethanol with higher carboxylic acid solvents and their toxicity to yeast. Separation
3 and Purification Technology 72 (2010) 180-185.
4 [10] Y.S. Teh, G.P. Rangaiah, Tabu search for global optimization of continuous functions with
5 application to phase equilibrium calculations, Computers and Chemical Engineering 27 (2003)
6 1665-1679.
7 [11] C. Gutiérrez–Antonio, A. Briones-Ramírez, Pareto front of ideal petlyuk sequences using a
8 multiobjective genetic algorithm with constraints, Computers and Chemical Engineering 33
9 (2009) 454-464.
10 [12] A.A. Kiss, R.M. Ignat, Innovative single step bioethanol dehydration in an extractive dividing-
11 wall column, Separation and Purification Technology 98 (2012) 290-297.
12 [13] E. Ahmetovic, M. Martín, I.E. Grossmann, Optimization of energy and water consumption in
13 corn-based ethanol plants, Industrial and Engineering Chemistry Research 49 (2010) 7972–7982.
14 [14] L. Čuček, M. Martín, I.E. Grossmann, K. Kravanja, Energy, water and process technologies
15 integration for the simultaneous production of ethanol and food from the entire corn plant,
16 Computers and Chemical Engineering 35 (2011) 1547–1557.
17 [15] J.F. Wang, C.X. Li, Z.H. Wang, Z.J. Li, Y.B. Jiang, Vapor pressure measurement for water,
18 methanol, ethanol, and their binary mixtures in the presence of an ionic liquid 1-ethyl-3-
19 methylimidazolium dimethylphosphate, Fluid Phase Equilibria 255(2007) 186-192.
20 [16] P. García-Herreros, J.M. Gómez, I.D. Gil, G. Rodríguez, Optimization of the design and
21 operation of an extractive distillation system for the production of fuel grade ethanol using
22 glycerol as entrainer, Industrial and Engineering Chemistry 50 (2011) 3977–3985.
23 [17] A. Messac, A. Ismail-Yahaya, C.A. Mattson, The normalized normal constraint method for
24 generating the pareto frontier, Structural and Multidisciplinary Optimization 25 (2003) 86–98.
25 [18] M. Martínez, J. Sanchis, X. Blasco, Global and well-distributed pareto frontier by modified
26 normalized normal constraint methods for bicriterion problems, Structural and Multidisciplinary
27 Optimization 34 (2007) 197-209.
28 [19] M. Martínez-Iranzo, J.M. Herrero, J. Sanchis, X. Blasco, S. García-Nieto, Applied pareto multi-
29 objective optimization by stochastic solvers, Engineering Applications of Artificial Intelligence,
30 22 (2009) 455-465.
31 [20] D.E. Golberg, Genetic algorithms in search, optimization, and machine learning, Addison-
32 Wesley, USA, 1989.
1 [21] J. Kennedy, R. Eberhart, Particle swarm optimization, Proc. IEEE International Conference on
2 Neural Networks 4 (1995) 1942-1948.
3 [22] R.M. Storn, K.V. Price, Differential evolutions a simple and efficient heuristic for global
4 optimization, Journal of Global Optimization 11 (1997) 341-359.
5 [23] M.M. Ali, A. Torn, Population set based global optimization algorithms: some modifications and
6 numerical studies, Computers and Operations Research 31 (2004) 1703.
7 [24] X. Xu, Y. Li, Comparison between particle swarm optimization, differential evolution and multi-
8 parents crossover, In Proceedings of the IEEE International Conference on Computational
9 Intelligence and Security (2007) 124-127, doi:10.1109/cis.2007.37.
10 [25] M.A. Panduco, C.A. Brizuela, A comparison of genetic algorithms, particle swarm optimization
11 and the differential evolution method for the design of scannable circular antenna arrays,
12 Progress in Electromagnetics Research B, (13) 2009 171-186.
13 [26] W.F. Sacco, N. Henderson, A.C. Rios-Coelho, M.M. Ali, C.M.N.A. Pereira, Differential
14 evolution algorithms applied to nuclear reactor core design, Annals of Nuclear Energy 36 (2009)
15 1093-1099.
16 [27] K.V. Price, R.M. Storn, J. Lampinen, Differential evolution: a practical approach to global
17 optimization, Springer, Berlin, 2005.
18 [28] K.M. Yerramsetty, C.V.S. Murty, Synthesis of cost-optimal heat exchanger networks using
19 differential evolution. Computers and Chemical Engineering 32 (2008) 1861–1876.
20 [29] S. Kheawhom, Efficient constraint handling scheme for differential evolutionary algorithm in
21 solving chemical engineering optimization problem, Journal of Industrial and Engineering
22 Chemistry 16 (2010) 620–628.
23 [30] A. Bonilla-Petriciolet, G.P. Rangaiahb, J.G. Segovia-Hernández, Evaluation of stochastic global
24 optimization methods for modeling vapor–liquid equilibrium data, Fluid Phase Equilibria 287
25 (2010) 111–125.
26 [31] S. Kumar, D. Datta, B.V. Babu, Estimation of equilibrium parameters using differential
27 evolution in reactive extraction of propionic acid by tri-n-butyl phosphate, Chemical Engineering
28 and Processing: Process Intensification 50 (2011) 614–622.
29 [32] R. Vakili, P. Setoodeh, E. Pourazad, D. Iranshahi, M.R. Rahimpour, Utilizing differential
30 evolution (DE) technique to optimize operating conditions of an integrated thermally coupled
31 direct DME synthesis reactor. Chemical Engineering Journal 168 (2011) 321–332.
1 [33] S.S. Bhattacharya, V.K. Garlapati, R. Banerjee, Optimization of laccase production using
2 response surface methodology coupled with differential evolution, New Biotechnology 28 (2011)
3 31-39.
4 [34] J.R. Couper, W.R. Penney, J.R. Fair, Chemical process equipment: selection and design,
5 Elsevier, Second Edition, USA, 2005.
6 [35] R. Turton, R.C. Bailie, W.B. Whiting, J.A. Shaeiwitz, Analysis, synthesis and design of chemical
7 process, Second edition, Prentice Hall, USA, 2004.
8
1 Table 1. Feeds of ethanol-water mixture.
Mol flow [kmol/h] FI FII FIII FIV
Ethanol 6.80 4.54 2.27 0.91
Water 38.56 40.82 43.09 44.45
Molar composition (%)
Ethanol 15 10 5 2
Water 85 90 95 98
Mass composition (%)
Ethanol 31 22 12 5
Water 69 78 88 95
2
3
4
5 Table 2. Search range for integer variables
Sequence Variable Range of search
CSS (FI-FIV)
S 5-100
SF/SFS 2-100
OSS (FI-FIV)
C 1-3
SEX/S 5-100
SF/SFS 2-100
6
1
2 Table 3. Search range for continuous variables
Sequence Variable Range search (low limit-upper limit)
CSS (feed stream) FI FII FIII FIV
RR 0.1-36
RDI [kmol/h] 7.26-10.89 4.66-6.99 2.33-3.50 0.93-1.40
RDII
[kmol/h] 5.40-8.10 3.60-5.40 1.80-2.70 0.72-1.08
RDIII kmol/h] 0.73-1.59 0.86-1.28 0.43-0.64 0.17-0.26
FS [kmol/h] 6.80-13.61 8.44-12.66 4.22-6.33 1.69-2.53
OSS (feed stream) FI FII FIII FIV
RR 0.1-36
RDI [kmol/h] 11.79-18.60 7.89-12.34 4.54-6.82 2.54-4.13
RDII
[kmol/h] 5.39-8.08 3.59-5.39 1.80-2.69 0.72-1.08
RDIII kmol/h] 6.21-10.44 4.45-6.94 2.73-4.14 1.72-3.05
FS [kmol/h] 3.27-5.44 2.54-3.92 1.27-2.18 1.09-2.18
FA [kmol/h] 3.63-5.44 3.63-5.99 3.97-5.96 3.63-6.26
3
4
5
6
7
8 Table 4. Mass fraction for all streams for conventional separation sequence (CSS).
STREAM FEED I DIST-CI WATER MAKEUP SOLV ETHANOL BOTT-CII DIST-CII
SOLV
Purity [wt%] FI
Ethanol 0.3110 0.0000 0.9188 0.0000 0.0000 1.0000 0.0077 0.1178
Water 0.6890 1.0000 0.0812 0.0000 0.0040 0.0000 0.0611 0.8822
Ethylene glycol (solv) 0.0000 0.0000 0.0000 1.0000 0.9960 0.0000 0.9312 0.0000
FII
Ethanol 0.2213 0.9346 0.0000 0.0000 0.0000 0.9990 0.0063 0.1409
Water 0.7787 0.0654 1.0000 0.0000 0.0049 0.0010 0.0266 0.4922
Ethylene glycol (solv) 0.0000 0.0000 0.0000 1.0000 0.9951 0.0000 0.9671 0.3669
FIII
Ethanol 0.1186 0.9008 0.0000 0.0000 0.0000 1.0000 0.0052 0.1102
Water 0.8814 0.0992 1.0000 0.0000 0.0092 0.0000 0.0389 0.6362
Ethylene glycol (solv) 0.0000 0.0000 0.0000 1.0000 0.9908 0.0000 0.9558 0.2536
FIV
Ethanol 0.0496 0.9123 0.0000 0.0000 0.0000 0.9996 0.0052 0.1329
Water 0.9504 0.0877 1.0000 0.0000 0.0089 0.0004 0.0343 0.6572
Ethylene glycol (solv) 0.0000 0.0000 0.0000 1.0000 0.9911 0.0000 0.9605 0.2099
9
1
2 Table 5. Mass fraction for all streams for optional separation sequence (OSS-I).
STREAM FEED MAKEUP ACID/ WATER BOTT- DIST- MAKEUP SOLV ETHANOL BOTT- DIST-
I ACID/ ALCOHOL EX CI SOLV CII CIII
ALCOHOL
Purity [wt%] FI
Ethanol 0.3110 0.0000 0.0007 0.0000 0.2811 0.6824 0.0000 0.0000 1.0000 0.0028 0.0082
Water 0.6890 0.0000 0.0089 0.9981 0.1286 0.2997 0.0000 0.0068 0.0000 0.3251 0.9258
Ethylene glycol solv) 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 1.0000 0.9932 0.0000 0.6527 0.0099
Octanol 0.0000 1.0000 0.9904 0.0019 0.5904 0.0179 0.0000 0.0000 0.0000 0.0194 0.0561
FII
Ethanol 0.2213 0.0000 0.0003 0.0001 0.2468 0.6784 0.0000 0.0000 0.9999 0.0067 0.0196
Water 0.7787 0.0000 0.0068 0.9980 0.1145 0.3031 0.0000 0.0035 0.0001 0.3078 0.8997
Ethylene glycol solv) 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 1.0000 0.9964 0.0000 0.6667 0.0256
Octanol 0.0000 1.0000 0.9930 0.0019 0.6387 0.0185 0.0000 0.0001 0.0000 0.0188 0.0551
FIII
Ethanol 0.1186 0.0000 0.0007 0.0009 0.1405 0.6197 0.0000 0.0000 0.9994 0.0019 0.0059
Water 0.8814 0.0000 0.0049 0.9973 0.0802 0.3384 0.0000 0.0091 0.0006 0.2836 0.8824
Ethylene glycol solv) 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 1.0000 0.9908 0.0000 0.6793 0.0000
Iso-octanol 0.0000 1.0000 0.9944 0.0019 0.7793 0.0419 0.0000 0.0001 0.0000 0.0352 0.1117
FIV
Ethanol 0.0496 0.0000 0.0000 0.0008 0.0626 0.5288 0.0000 0.0000 1.0000 0.0016 0.0044
Water 0.9504 0.0000 0.0094 0.9952 0.0634 0.4656 0.0000 0.0004 0.0000 0.3474 0.9920
Ethylene glycol solv) 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 1.0000 0.9951 0.0000 0.6468 0.0000
Octanoic acid 0.0000 1.0000 0.9906 0.0040 0.8739 0.0056 0.0000 0.0045 0.0000 0.0042 0.0035
1 Table 6. Design parameters for the separation sequences for feed I and II.
CSS (FI) OSS-I (FI) CSS (FII) OSS-I (FII)
Column CI CII CIII CI CII CIII EX CI CII CIII CI CII CIII EX
Number of stage 59 98 9 9 94 20 21 79 69 81 21 78 16 48
Stage of feed 21 79 7 5 86 3 25 59 25 16 70 14

Specifications
Distillate rate [kmol/h] 8.34 6.73 1.52 14.49 6.77 7.64 5.35 4.47 0.88 9.75 4.41 5.27
Bottoms rate [kmol/h] 37.02 8.42 6.90 5.33 12.17 4.53 40.00 9.32 8.44 4.31 8.52 3.25
Reflux ratio 1.62 0.81 0.70 0.11 2.06 0.1 3.12 0.28 0.34 0.11 2.10 0.1

Feed stream
Acid/Alcohol flowrate 5.06* 4.16*
[kmol/h]
Solvent flowrate [kmol/h] 6.80 4.45 8.44 3.25
Stage of feed solvent 19 22 35 34

Product streams
Ethanol flowrate [kmol/h] 6.73 6.77 4.46 4.49
Ethanol purity [wt%] 100 100 99.9 100
Water flowrate [kmol/h] 37.02 30.60 40 35.44
Water purity [wt%] 100 99.97 100 99.8
Acid/Alcohol flowrate 4.99 4.11
[kmol/h]
Acid/Alcohol purity [wt%] 99.04 99.3
Solvent flowrate [kmol/h] 6.80 4.42 8.30 3.21
Solvent purity [wt%] 99.6 99.32 99.5 99.6

Energy Requirements
Reboiler duty [kW] 312 171 44 260 245 113 312 117 25 188 165 79
Condenser duty [kW] -240 - -30 -182 -224 -97 -241 -62 -15 - - -67
132 123 150
Total reboiler duty [kW] 527 618 454 432
Total cooling duty [kW] -402 -503 -319 -341

Economic evaluation
Total operating cost 80 99 70 69
[k$/year]
Total investment cost [k$] 8,863 7,276 11,237 7,384
Total annual cost [k$/year] 966 827 1,194 808

Key performance indicators

Energy requirements 1700 1981 2205 2084
[kWh/ton Ethanol]
2 octanol*, iso-octanol **, octanoic acid***
1 Table 7. Design parameters for the separation sequences for feed III and IV.
CSS (FIII) OSS-I (FIII) CSS (FIV) OSS-I (FIV)
Column CI CII CIII CI CII CIII EX CI CII CIII CI CII CIII EX
Number of stage 50 77 26 10 98 34 35 80 86 13 17 95 54 97
Stage of feed 20 58 19 4 70 31 35 63 4 2 76 46

Specifications
Distillate rate 2.91 2.23 0.59 5.43 2.24 3.12 1.13 0.89 0.20 2.91 0.89 2.02
[kmol/h]
Bottoms rate 42.45 5.23 4.64 4.53 5.37 2.25 44.23 2.04 1.85 4.29 3.11 1.09
[kmol/h]
Reflux ratio 1.11 0.41 0.30 0.12 1.96 0.18 1.46 0.31 0.42 0.11 2.27 0.10

Feed stream
Acid/Alcohol 4.43** 4.01***
flowrate [kmol/h]
Solvent flowrate 4.55 2.18 1.81 1.09
[kmol/h]
Stage of feed 10 13 27 6
solvent

Product streams
Ethanol flowrate 2.23 2.24 0.89 0.89
[kmol/h]
Ethanol purity 100 99.94 100 100
[wt%]
Water flowrate 42.45 39.80 44.23 42.13
[kmol/h]
Water purity [wt%] 100 99.72 100 99.52
Acid/Alcohol 4.37 3.99
flowrate [kmol/h]
Acid/Alcohol purity 99.45 99.06
[wt%]
Solvent flowrate 4.50 2.18 1.79 1.09
[kmol/h]
Solvent purity 99.08 99.08 99.11 99.50
[wt%]

Energy
Requirements
Reboiler duty [kW] 138 62 15 132 81 50 101 24 5 86 36 30
Condenser duty -68 -34 -10 -69 -72 -43 -31 -13 -3 -37 -32 -25
[kW]
Total reboiler duty 215 263 130 152
[kW]
Total cooling duty -111 -184 -47 -94
[kW]

Economic
evaluation
Total operating cost 33 42 19 24
[k$/year]
Total Investment 6,662 8,226 8,421 12,566
Cost [k$]
Total annual cost 700 865 862 1,281
[k$/year]

Key performance
indicators
Energy 2093 2549 3171 3707
requirements
[kWh/ton Ethanol]
2 octanol*, iso-octanol **, octanoic acid ***
1
2 Figure 1. Conventional separation sequence (CSS).
3
4
5
6

7
8 Figure 2. Optional separation sequence (OSS-I).
 Generation of initial population and
DE Initial design Codification of variables evaluation of objective and constraints function using
Aspen Plus

Generate subpopulations of N mutant

no The parent and the mutant are individuals by mutation and crossover
Maximum
compared and the best individual operations.
Generate new number of
is selected to populate the next Then evaluate objective and
population generation
reached generation constraint functions using
Aspen Plus

yes

Best point that has been explored

over all the generations
End
1
2 Figure 3. Flowchart for the differential evolution algorithm.
3
4
5
6
7
8
9
10
11
12
13

14
15 Figure 4. Connection of Mathworks Matlab with AspenTech Aspen Plus via MS Excel

View publication stats