Fuel 103 (2013) 269–277

Contents lists available at SciVerse ScienceDirect

Fuel
journal homepage: www.elsevier.com/locate/fuel

Multi-objective optimization of a bio-diesel production process

Shivom Sharma, G.P. Rangaiah ⇑
Department of Chemical and Biomolecular Engineering, National University of Singapore, Singapore 117576, Singapore

h i g h l i g h t s

" Waste cooking oil is used to produce bio-diesel, which promotes sustainability.
" Studied bio-diesel process is close to industrial practice (namely, Lurgi process).
" Complete bio-diesel plant is optimized for economic and environmental objectives.
" Evolutionary algorithm provides alternate optimal designs for bio-diesel process.
" One plant design is chosen and optimized in case of variation in feed flow rate.

a r t i c l e i n f o a b s t r a c t

Article history: Bio-diesel is a mixture of fatty acid methyl esters, and is produced by trans-esterification of renewable
Received 1 March 2012 feed-stock. This study develops an economically attractive and environmentally acceptable bio-diesel
Received in revised form 15 May 2012 production process using multi-objective optimization. It considers bio-diesel production using waste
Accepted 21 May 2012
cooking oil as the feed-stock, which can reduce bio-diesel cost. Both the pre-treatment and trans-ester-
Available online 8 June 2012
ification sections of this process are simulated in a process simulator followed by optimization for multi-
ple objectives using multi-objective differential evolution with taboo list. Profit, fixed capital investment
Keywords:
and organic waste from the biodiesel process are considered as objectives for optimization. Effect of var-
Bio-diesel
Waste cooking oil
iation in feed flow rate on process performance is also studied. The results show that the amount of
Multi-objective optimization organic waste can change significantly with only small variations in the economic objectives, and they
Differential evolution provide alternative process designs with different environmental impacts, for implementation.
Ó 2012 Elsevier Ltd. All rights reserved.

1. Introduction treated in the presence of an acid catalyst (i.e., esterification),

and later bio-diesel is produced from treated oil using an alkali cat-
Vegetable oils are mixtures of triglycerides, which can be con- alyst (i.e., trans-esterification). Heterogeneous catalyzed and
verted into a mixture of fatty acid methyl esters (FAMEs) by reac- supercritical processes are other processes to produce bio-diesel
tion with methanol (i.e., trans-esterification). The product, FAME is [2,3]. Heterogeneous catalyzed process does not produce aqueous
the bio-diesel. The traditional bio-diesel processes use homoge- waste since catalyst can be easily separated from the reaction
nous acid or alkali as catalyst for trans-esterification. Raw material mixture.
cost is the major contributor to the cost of bio-diesel. Waste cook- Zhang et al. [4] proposed four different processes, namely,
ing oils are cheaper than pure vegetable oils, but the former con- alkali-catalyzed process using pure vegetable oil, alkali-catalyzed
tain high amount of free fatty acid (FFA). The trans-esterification process using waste cooking oil, acid-catalyzed process using
is more efficient and faster with an alkali catalyst compared to waste cooking oil and acid-catalyzed process using hexane extrac-
an acid catalyst, but it requires absence of FFA in the feed. FFA pro- tion, to produce bio-diesel. Their study also reports operational and
duces soap and water by reacting with an alkali catalyst; hence, design details for these process flow-sheets. In another work,
alkali-catalyzed process cannot directly be used to produce Zhang et al. [5] conducted an economic evaluation of the above
bio-diesel from waste cooking oils. Canakci and Van Gerpen [1] four processes, and found that acid-catalyzed process using waste
have combined the advantages of alkali- and acid-catalyzed pro- cooking oil is more economical than others. Hass et al. [6] devel-
cesses into a single process, where waste vegetable oil is first oped a process model to estimate capital and operating costs of
bio-diesel production using an alkali catalyst. West et al. [7] ana-
lyzed four alternative processes: acid catalyzed, alkali catalyzed,
⇑ Corresponding author. heterogeneous acid catalyzed and supercritical, for producing
E-mail address: [email protected] (G.P. Rangaiah). bio-diesel from waste cooking oils, and also conducted economic

0016-2361/$ - see front matter Ó 2012 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.fuel.2012.05.035
270 S. Sharma, G.P. Rangaiah / Fuel 103 (2013) 269–277

analysis of these processes. Their study concluded that heteroge- 2.1. Pre-treatment of waste cooking oil
neous acid-catalyzed process is superior to the other three
processes, but it is still in the development phase. Recently, Zhang Zhang et al. [4] have presented a process to reduce FFA content
et al. [8] presented a process scheme for producing bio-diesel in waste cooking oil. In their process, FFA is reacted with methanol
from pure vegetable oil, and developed a plant-wide control in the presence of acid catalyst. A liquid–liquid extraction (LLE)
structure. column is used to remove the catalyst from the reactor effluent
Several researchers have optimized the bio-diesel process for using fresh glycerol. The extract phase from LLE has glycerol, cata-
single objective. Ghadge and Raheman [9] used response surface lyst and methanol while the raffinate phase has bio-diesel, oil and
methodology to study the effect of methanol quantity, acid concen- methanol. Subsequently, two component splitters are used in the
tration and reaction time on FFA content reduction, and found that study of Zhang et al. [4] to remove bio-diesel and oil from the ex-
methanol quantity is the most prominent factor to reduce FFA con- tract phase, and water from the raffinate phase. Methanol is sepa-
tent. Myint and El-Halwagi [10] optimized four alternate alkali- rated from glycerol using a distillation column, and recycled to the
catalyzed process flow-sheets, with different separation sequences, reactor. The treated oil with some FAME and methanol goes to the
for bio-diesel production. They concluded that the best flow-sheet downstream process where it is converted to bio-diesel. A similar
is with the bio-diesel and glycerol separation first, followed by flow-sheet without glycerol-methanol separation has been used
methanol recovery and finally water washing. Nicola et al. [11] by Garcia et al. [15]; the focus of their study is on the prediction
optimized two process variants of alkaline trans-esterification of of bio-diesel properties.
refined vegetable oil for two objectives: minimum energy con- As a component splitter is not a physical unit, the flow-sheet in
sumption and better product quality, simultaneously. This process Zhang et al. [4] is not suitable for economic evaluation of the pro-
was simulated using ASPEN Plus, and a multi-objective genetic cess. Additionally, it is found from process simulation that a three-
algorithm was used to optimize it. However, optimization of bio- phase separator is sufficient, instead of a LLE column, for removing
diesel production process using waste cooking oil has not been the catalyst from the reactor effluent. Hence, in this study, bio-die-
studied for multiple objectives. Waste cooking oils have significant sel process flow-sheet of Zhang et al. [4] is modified to avoid the
impact on the environment if they are disposed without re-use, component splitter and to use the three-phase separator. Waste
and so their use to produce bio-diesel is attractive for both eco- cooking oil enters the esterification reactor, where FFAs react with
nomic and environmental reasons. methanol in the presence of sulfuric acid (see the upper half of
In this study, bio-diesel production process using waste cooking Fig. 1). The esterification reactor is maintained at 60 °C and
oil is simulated in Aspen Hysys; waste cooking oil is first treated in 400 kPa; the reactant streams are pressurized to reactor pressure,
the presence of sulfuric acid (i.e., pre-treatment), followed by bio- and low pressure steam is used to maintain the reactor tempera-
diesel production from the treated oil using sodium hydroxide. ture. The effluent from CSTR0 preheats the waste cooking oil feed.
Pre-treatment of waste cooking oil reduces FFA content below 1% After that, fresh glycerol is mixed with the effluent before it enters
by weight. The present study optimizes the design of bio-diesel the three-phase separator at around 40 °C. Two liquid phases are
process for three important objectives: maximum profit, minimum formed in the three-phase separator; the light phase contains
fixed capital investment and minimum organic waste. Effect of var- bio-diesel, oil, methanol and water while the heavy phase contains
iation in waste cooking oil flow rate is also explored. Evolutionary glycerol, catalyst, methanol and water. Two distillation columns
multi-objective optimization techniques have been popular for (i.e., G-C1 and BD-C1) are used to recover methanol from both out-
studying the trade-offs between conflicting objectives in many let streams of the three-phase separator. Number of stages and
chemical engineering applications [12]. In this work, multi-objec- operating pressures of these columns are shown in Fig. 1. Columns
tive differential evolution with taboo list (MODE-TL) is used to are under vacuum to avoid decomposition of glycerol and biodiesel
generate the Pareto-optimal front. Taboo list (TL) concept of taboo at high temperatures. Morais et al. [16] have reported thermal
search has been incorporated with MODE to avoid revisiting the decomposition temperature of 250 °C and 150 °C for pure FAME
search space, which can reduce the number of objective function and glycerol respectively. Presence of methanol with FAME and
evaluations to obtain the global optimum [13]. glycerol gives additional temperature margin to avoid deteriora-
Next section of this article develops bio-diesel production pro- tion of FAME and glycerol. The recovered methanol from both
cess using waste cooking oil. Section 3 discusses process simulation, the columns is mixed, pressurized and recycled back to the ester-
esterification and trans-esterification kinetics, and thermodynamic ification reactor. The bottom stream from BD-C1 column contains
model used. Section 4 describes multi-objective optimization bio-diesel, oil and methanol; this stream is pressurized to about
(MOO) problem formulation for bio-diesel production, followed 4 bars and sent to the trans-esterification section. The bottom
by a brief description of MODE-TL algorithm. Section 5 discusses stream from G-C1 column has mainly glycerol, acid and water.
the obtained results for different optimization cases. In the last
section, main conclusions of this study are summarized. 2.2. Bio-diesel production from treated waste cooking oil

The alkali-catalyzed trans-esterification is performed at about

2. Process development 60 °C. Pressure depends upon the type of reactor used: ambient
pressure for batch reactor [17] and 4 bars for CSTR and PFR [18].
For this study, bio-diesel plant capacity is assumed to be Trans-esterification is mass-transfer controlled, and so excess meth-
20,000 metric tons per annum, based on potential waste cooking anol improves oil to bio-diesel conversion [19]. Molar flow ratio of
oil availability in a city of 5 million people such as Singapore [14]. methanol to triglycerides of more than 6 was used by Noureddini
This process can be divided into two sections: (1) pre-treatment and Zhu [19]. Recently, Zhang et al. [8] developed a process flow-
of waste cooking oil using H2SO4 to reduce FFA content, and (2) pro- sheet for bio-diesel from pure vegetable oil. In this work, this flow-
duction of bio-diesel from treated oil using NaOH. Alkali catalyst sheet has been used for bio-diesel production from treated waste
has been chosen for trans-esterification because it can give high cooking oil (see the lower half of Fig. 1). Continuous stirred tank
yield in short time and bio-diesel production using alkali catalyst reactors (CSTRs), distillation columns, three-phase separators, a
is in industrial practice (http://www.lurgi.com/website/Biodiesel. neutralization reactor and a LLE column are the main units in this
57.0.html?&L=1). The two sections of the bio-diesel process using process. Three CSTRs are placed in series, and treated oil mixed with
waste cooking oil are discussed in more details below. methanol and NaOH catalyst enters CSTR1. The effluent mixtures
 S. Sharma, G.P. Rangaiah / Fuel 103 (2013) 269–277 271

Fig. 1. Schematic of bio-diesel production plant using waste cooking oil as feed; see Table 2 for typical stream data in this process.

from CSTR1 and CSTR2 individually pass through the three-phase The recovered methanol in the distillate stream of column BD-C2
separators, where glycerol with some methanol is separated as the is also recycled back to trans-esterification reactors. Since the recy-
light phase. cled methanol should be free of water, water wash column is used
The heavy phase from the three-phase separators (S1-H and after separating methanol from the reaction mixture.
S2-H in Fig. 1) goes to the next CSTR; this phase mainly contains
bio-diesel, oil and methanol. Glycerol streams (i.e., light phases
from three-phase separators 1 and 2) are mixed together, and a 3. Process simulation
distillation column (G-C2) is used to recover unreacted methanol
for recycle. Another distillation column (BD-C2) is used to recover The bio-diesel process using waste cooking oil (Fig. 1) has been
methanol from the effluent of CSTR3. Bottom product from this col- simulated in the Aspen Hysys V-7.2. Vegetable oil is a mixture of
umn contains mainly bio-diesel, and is treated in a neutralization triglycerides of oleic, linoleic, linolenic, palmitic, stearic and other
unit to remove alkali catalyst, followed by a water wash column. acids. Physical properties of different triglycerides present in
272 S. Sharma, G.P. Rangaiah / Fuel 103 (2013) 269–277

vegetable oil are not much different [10]; hence, one of the triglyc- ence has to be maintained at both ends of heaters/reboilers and
erides can be used to represent the vegetable oil. In this work, tri- coolers/condensers (i.e., 25 °C for heating and 5 °C for cooling).
olein (i.e., triglyceride of oleic acid) is considered as the triglyceride Minimum temperature difference at both ends of a heat exchan-
in the waste cooking oil. Physical properties of mono-, di- and tri- ger is constrained to be 10 °C. Initially, design of biodiesel process
olein are taken from Aspen Plus database. Franca et al. [20] have is optimized for two cases; Case A considers trade-off between
studied the phase equilibrium of bio-diesel + glycerol + methanol profit and FCI whereas trade-off between profit and organic waste
system, and found that experimental LLE data were satisfactorily is studied in Case B. After that, three cases of operation optimiza-
predicted by the UNIQUAC model. Hence, in this work, this model tion are performed for a selected process from the design
has been used to predict the physical behavior such as liquid– optimization. The operation optimization, Case C provides the ba-
liquid equilibria. Esterification and trans-esterification kinetics sis for the next two cases (D and E), which consider the effect of
used are respectively from Berrios et al. [21] and Noureddini and variation in waste cooking oil feed rate on the process perfor-
Zhu [19]. mance. Table 1 provides details on these cases considered in this
Pressure in the esterification and trans-esterification reactors study. Constraints are same in all design and operation optimiza-
is fixed at 400 kPa while reactor temperatures are considered tion cases.
as decision variables in the optimization. As mentioned earlier, Fixed capital investment (FCI), also known as total module cost
both glycerol and bio-diesel are sensitive to high temperature; (CTM), is calculated using Eqs. (1)–(3) below and related data from
hence, all distillation columns are under vacuum and conse- Turton et al. [22] for each equipment in Fig. 1. Equipment purchase
quently lower temperatures to avoid product deterioration. Neu- cost (Cp) is calculated using Eq. (1) and cost data in [22]. If the re-
tralization reactor and water wash column are at atmospheric quired capacity of any process equipment is larger than the capac-
pressure. Information for different feed/reactant streams is taken ity range given in [22], then multiple units of that equipment, with
from [4] and [8]. Fig. 1 presents important information for all equal capacity, are assumed. Although this increases the capital
feed/reactant streams in the bio-diesel process. Feed waste cook- cost, use of multiple units for costing is reasonable due to equip-
ing oil contains FFA (0.06 mol fraction) and tri-olein (0.94 mol ment availability. Further, this does not affect FCI much in this
fraction). 2500 kg of waste cooking oil per hour is processed to study since only some optimized solutions require two units of
produce bio-diesel. three-phase separators. If the distillation column diameter is less
than 0.9 m, then the column is taken to be filled with ceramic
packing; otherwise, trays are assumed in the column [7]. Bare
4. Multi-objective optimization problem formulation module cost (CBM) is calculated using Eq. (2); FBM is calculated
based on the correlations and data given in [22]. In Fig. 1, if the
Bio-diesel process is optimized for profit, fixed capital invest- fluid inside an equipment contains acid, then stainless steel (SS)
ment (FCI) and organic waste (i.e., methanol, glycerol, tri/di/ is used as material of construction (MOC). Accordingly, CSTR0, 3-
mono-olein and oleic acid in the waste stream leaving the water phase separator-0, BD-C1, G-C1, BD-C2, G-C2, neut. reactor, P-1
wash column) from the process in Fig. 1. Profit, which includes rev- are made of SS, whereas CS-shell/SS-tube is used as MOC for HE-
enue and cost of manufacturing (COM), is used as one objective 1, condenser and reboiler of BD-C1. MOC for the remaining equip-
function (Eq. (5) given later). FCI and organic waste represent ments is carbon steel. CTM is 1.18 times bare module cost of all
investment in the plant and potential environmental impact of equipments (Eq. (3)). Chemical Engineering Plant Cost Index, CEPCI
the plant respectively. The decision variables for optimization are of 600 is used to account for inflation.
temperature and volume of each of the four CSTRs; these are the
logðC p Þ ¼ K 1 þ K 2 logðCapacityÞ þ K 3 ½logðCapacityÞ2 ð1Þ
crucial process variables for conversion of feed to bio-diesel. The
temperature range for each reactor is selected around the sug- X
gested operating temperature whereas volume range for each reac- C BM ¼ C p F BM ð2Þ
all equipments
tor is chosen based on the required residence time.
Nicola et al. [11] have combined purities of biodiesel and glyc-
FCI or C TM ¼ 1:18C BM ð3Þ
erol, and used it as one objective. The present study treats prod-
uct purities as constraints as these have to satisfy the prevailing Cost of manufacture (COM) is given by Eq. (4) [22], and profit is
requirements (Table 1). To prevent product deterioration, upper the difference between the revenue (earned by selling the prod-
limits are imposed on the temperatures at the bottom of all dis- ucts: bio-diesel and glycerol) and COM (Eq. (5)).
tillation columns [16]. Further, a minimum temperature differ-

Table 1
Different optimization cases studied for bio-diesel production process.

Objective functions in different cases Decision variables and their Constraints

bounds
Design cases PurityBD P 0.99;
A. Max. Profit, Min. FCI 50 6 TCSTR0 6 60 °C; PurityG P 0.95;
50 6 TCSTR1/2/3 6 70 °C; TBD-C1 6 250 °C;
B. Max. Profit, Min. Organic waste (Waste cooking oil feed rate = 2500 kg/h in both these 2 6 VCSTR0 6 3 m3; TBD-C2 6 250 °C;
design cases) 5 6 VCSTR1/2/3 6 10 m3 TG-C1 6 155 °C;
TG-C2 6 155 °C;
Operation cases 50 6 TCSTR0 6 60 °C; DTboth ends P 5 °C for coolers and
condensers;
C. Max. Profit, Min. Organic waste (waste cooking oil feed rate = 2500 kg/h) 50 6 TCSTR1/2/3 6 70 °C DTboth ends P 25 °C for heaters and
reboilers;
D. Max. Profit, Min. Organic waste (2700 kg/h, i.e., 10% increase in waste cooking oil feed DTboth ends P 10 °C for heaters
rate) exchangers.
E. Max. Profit, Min. Organic waste (2000 kg/h, i.e., 20% decrease in waste cooking oil feed
rate)
 S. Sharma, G.P. Rangaiah / Fuel 103 (2013) 269–277 273

COM ¼ 0:28 FCI þ 2:73 ðoperating laborÞ þ 1:23 ðutilities initial population is generated by mutation and crossover on three
þ cost of raw materialÞ ð4Þ randomly selected individuals from the population. After that, the
trial vector is checked for violation of decision variable bounds; if a
bound on any decision variable is violated, then it is randomly re-
Profit ¼ Revenue  COM ð5Þ
initialized within the bounds on that decision variable. Taboo
Costs of waste cooking oil (=0.39 $/kg), methanol (=0.28 $/kg), check is implemented in the evaluation step of trial vector; if the
glycerol (=1.1 and 1.15 $/kg for 95 and 99 wt.%, respectively), NaOH trial individual is near to any individual in the TL by a specified dis-
(=0.75 $/kg for 37 wt.%), HCl (=0.92 $/kg), and steam at 106 bar tance, then it is rejected without calculating objectives and con-
(=0.032 $/kg) are obtained from our industry contacts. Cost of sulfu- straints. Accepted trial individual is stored in the child
ric acid (=0.071 $/kg) is obtained from http://www.ICIS.com (access population, and also added to the TL. After generating the child
date: August, 2011) while cost of bio-diesel (=1.464 $/kg) is taken population, non-dominated sorting followed by crowding distance
from http://www.biofueloasis.com (access date: August, 2011). Fi- calculation of combined population (i.e., parent/current and child)
nally, costs of (process, chilled and cooling) water, (low, medium is used to select individuals for the next generation [23]. MODE-TL
and high pressure) steam and electricity are taken from [22]. algorithm handles inequality constraints by constrained-domi-
In many process optimization problems, evaluation of objec- nance approach of Deb et al. [24]. This algorithm is implemented
tives and constraints at each trial solution requires simulation of using Excel worksheets (to calculate the objective functions: profit,
the entire process, and so it is often computationally expensive. FCI and organic waste, constraints and linking between cells) and
Use of taboo list (TL) avoids the revisit of search space by keeping Visual Basic for Applications (to implement the algorithm steps
a record of recently visited points. So, multi-objective differential and also to interface the bio-diesel process simulation in Aspen
evolution (MODE) with taboo list (TL) can avoid unnecessary func- Hysys Version 7.2 with the Excel based MODE-TL algorithm).
tion evaluations [13]. Fig. 2 shows the flowchart of MODE with TL
(MODE-TL) algorithm. First, population is initialized randomly in-
5. Results and discussion
side the bounds on decision variables. Values of objectives and
constraints are calculated for each initial individual, and the TL is
Values of MODE-TL algorithm parameters used for all optimiza-
randomly filled using the initial population. In each generation
tion cases are: population size (NP) = 100, crossover probability
(i.e., iteration), a trial vector for each target vector in the current/
(Cr) = 0.3, mutation rate (F) = 0.5 and maximum number of gener-
ations (MNGs) = 100. Size of TL is fixed at half of NP (=50), and ta-
Start Set values of Cr, F, NP, MNG boo radius (TR) of 0.01 is chosen for all cases, to decide acceptance/
rejection of new trial individuals. Euclidean distance between the
Randomly initialize population, and evaluate values of objective newly generated trial individual and each individual in the TL is
functions and constraints of all individuals in the population
calculated for normalized values (between 0 and 1) of decision
Randomly select 50% initial variables. These values of algorithm parameters are selected based
individuals and store themin taboo list on our experience on solving many benchmark functions using
MODE-TL.
Set generation no., G = 1

Set individual no., n = 1

5.1. Design optimization

Generate a new mutant individual 5.1.1. Trade-off between profit and FCI
and then a trial individual The Pareto-optimal front obtained for optimizing profit and FCI
simultaneously is shown in Fig. 3a. As mentioned earlier, fixed
Check the trial individual for violation of decision amount of waste cooking oil is converted into bio-diesel; therefore,
variable bounds; if there is any violation, randomly re-
both objectives in the obtained Pareto-optimal front are varying
initialize that decision variable inside the bounds
over small ranges (Fig. 3a). West et al. [7] reported FCI of 1.1 mil-
Perform taboo check to reject the trial lion $ (CEPCI = 394) for a bio-diesel plant of capacity 8000 met-
individuals near to those in taboo list ric tons/annum and waste cooking oil as the feed-stock. Projected
FCI for a plant capacity of 20,000 metric tons/annum, using the
Evaluate values of objective functions and constraints six-tenth rule [22] and when CEPCI = 600, is 2.88 million $, which
of the accepted trial individual, and update taboo list
is comparable to the FCI range in Fig. 3a (i.e., 2.86–2.94 million $).
The trend in the Pareto-optimal front can be visually correlated
Store the trial solution in child population
to the decision variables: TCSTR1 and TCSTR2 (Fig. 3a, c and d). Larger
Yes amount of bio-diesel is produced at higher TCSTR1 and TCSTR2, which
n=n+1 Is n < NP?
results in increased revenue and profit. Light and heavy phases in
No the 3-phase separator-1 and separator-2 have a smaller difference
Combine parent and child populations in densities at high temperature, which increases required size of
these separators. Consequently, FCI of the bio-diesel production
Non-dominating sorting of combined population process increases with increase in the reactor temperature. Tem-
and calculate crowding distance, if required
perature of CSTR0 is also increasing in a small range with increas-
ing profit, while there is no particular trend in the temperature of
Selection of the population for next generation
CSTR3 (Fig. 3b and e). Further, smaller reactors can reduce FCI of
Yes process; hence, volume of each reactor is nearly constant at its
G=G+1 Is G < MNG? lower bound (i.e., esterification: 2 m3 and trans-esterification:
No 5 m3). Fig. 3f shows variation in CSTR0 volume while variations
Stop in the remaining reactor volumes are not shown, for brevity.
Each solution on the Pareto-optimal front is equally good from
Fig. 2. Flowchart of the MODE-TL algorithm. the point of specified objectives, and decision maker can select one
274 S. Sharma, G.P. Rangaiah / Fuel 103 (2013) 269–277

Fig. 3. Selected results for simultaneous maximization of profit and minimization of FCI.

Table 2
Important data of selected streams in Fig. 1, corresponding to the optimal solution ‘‘+’’ in Fig. 3a; total molar flow is in kmol/h and total mass flow is in kg/h.

Stream R0-out S0-L S0-H G-Acid BD-T MR-1A MR-1B

Temperature (°C) 59.25 42.77 42.77 150 244.8 23.98 25.38
Pressure (kPa) 400 200 200 15.6 15.72 14.86 14.86
Total molar flow 12.56 11.22 3.94 2.982 2.983 0.962 8.237
Total mass flow 2885 2832 293.5 264.4 2503 29.18 328.2
Mole fractions
Tri-olein 0.223 0.252 0 0 0.923 0 0.008
Methanol 0.568 0.560 0.216 0.003 0.006 0.878 0.760
Methyl oleate 0.024 0.027 0 0 0.057 0 0.015
Glycerol 0 0 0.661 0.874 0 0 0
Water 0.167 0.161 0.068 0.050 0.007 0.122 0.216
H2SO4 0.018 0 0.055 0.073 0 0 0
Stream F-1 S1-F S1-H F-2 S2-F S2-H F-3 S3-F MR-2 BD-C BD G-pdt Waste
Temperature (°C) 64.33 51.59 51.59 70 56.24 56.24 70 50.47 43.53 226.4 63.78 146.1 64.61
Pressure (kPa) 400 400 400 400 400 400 400 400 400 62 101.2 19.25 110
Total molar flow 31.36 31.36 24.99 2.56 27.55 26.09 0.37 26.46 22.59 8.91 8.53 2.794 2.85
Total mass flow 3413 3413 3074 82.1 3156 3083 11.99 3095 723.4 2533 2495 250.7 87.6
Mole fractions
Tri-olein 0.088 0.010 0.012 0 0.002 0.002 0 0 0 0.001 0 0 0.004
Di-olein 0 0.005 0.006 0 0.001 0.001 0 0 0 0.001 0 0 0.003
Mono-olein 0 0.002 0.003 0 0.001 0.001 0 0 0 0 0 0 0
Methanol 0.896 0.673 0.681 0.989 0.670 0.669 0.989 0.668 0.999 0.014 0.012 0.018 0.009
Methyl oleate 0.005 0.229 0.287 0 0.300 0.317 0 0.318 0 0.946 0.985 0.001 0.008
Glycerol 0 0.072 0 0 0.016 0 0 0.003 0 0.008 0 0.955 0.027
Water 0.001 0.001 0.001 0.001 0.001 0.001 0.001 0.001 0.001 0 0.003 0.007 0.858

solution based on his/her experience and/or requirement. One cor- 5.1.2. Trade-off between profit and organic waste
ner solution on the Pareto-optimal front, shown as ‘‘+’’ in Fig. 3a, The Pareto-optimal front obtained between profit and organic
has been selected for discussion and for further study. Table 2 pre- waste is shown in Fig. 4a. It can be divided into two parts; initially,
sents key information for important streams corresponding to this profit (i.e., 17.795–17.815 million $) varies linearly with the organ-
optimal solution. ic waste, and, finally, organic waste increases faster for a very small
 S. Sharma, G.P. Rangaiah / Fuel 103 (2013) 269–277 275

improvement in the profit (17.82 million $). The initial improve- dominated solutions between these two solutions (‘‘O’’ and ‘‘h’’).
ment in profit is due to increase in TCSTR0 and TCSTR2, and decrease TCSTR3 is constant at 50 °C in the first part, and marginal increase in
in VCSTR1 (Fig. 4a, b, d and g), while the remaining decision variables it can be seen in the second part, which is required to convert un-re-
are either constant or scattered. High values of TCSTR0 and TCSTR2 in- acted feed. As a 3-phase separator is not used after CSTR3, un-reacted
crease conversion of waste cooking oil in CSTR0 and CSTR2, which feed and glycerol appear in the waste stream. Volume of CSTR3 is
improves the profit. constant near its lower bound (5 m3), and is not shown in Fig. 4 for
Temperature of CSTR1 is nearly constant at its upper bound of brevity.
70 °C (Fig. 4c). Variations in TCSTR0, TCSTR2, TCSTR3, VCSTR0, VCSTR1 and
VCSTR2 are contributing to the second part of the obtained Pareto- 5.2. Operation optimization
optimal front (see Fig. 4a, b, d–h). Decrease in reactor volumes im-
proves the profit by decreasing FCI, while amount of organic waste In operation optimization, only operation variables are consid-
increases due to reduction in residence time. On the other hand, ered as decision variables while design variables related to equip-
partly to compensate the effects of decrease in reactor volumes, high ment sizes are fixed based on the design optimization (Section
value of reactor temperature improves the profit and also decreases 5.1.1). A bio-diesel process design shown as ‘‘+’’ in Fig. 3a has been
the organic waste. For example, profit and organic waste corre- selected for studying trade-offs between profit and organic waste,
sponding to solution ‘‘O’’ in Fig. 4a are 1.78  107 USD/annum and for variations in waste cooking oil feed rate. In all operation opti-
64,519 kg/annum respectively. Now, values of different reactor mization cases, ranges of operation variables are same as those
volumes for this solution are replaced by those corresponding to used in the design optimization (see Table 1).
solution ‘‘h’’ (relatively smaller values; Fig. 4a); organic waste
corresponding to the modified set of decision variables is 68,867 5.2.1. Trade-off between profit and organic waste: base case – waste
kg/annum with some improvement in profit. Organic waste corre- cooking oil feed rate = 2500 kg/h
sponding to solution ‘‘h’’ is slightly lower at 67,118 kg/annum due The selected design is optimized for the nominal value of waste
to reactor temperatures higher than those of the modified solution. cooking oil feed rate (i.e., 2500 kg/h). Although effect of operation
Thus, different combinations of decision variables are giving non- variables on the same objectives and processing capacity has been

Fig. 4. Selected results for simultaneous profit maximization and organic waste minimization.
276 S. Sharma, G.P. Rangaiah / Fuel 103 (2013) 269–277

studied in the design optimization case (Section 5.1.2), this optimi- compared to ‘‘base’’ case. Temperatures of CSTR1 and CSTR2 are
zation case is required for comparing different operation optimiza- nearly same at 70 °C in both the cases, which means more un-re-
tion cases. The obtained Pareto-optimal front for the ‘‘base’’ case of acted feed enters CSTR3 in the increased feed rate case compared
operation optimization is not shown as variation in profit value to ‘‘base’’ case. When larger amount of un-reacted feed enters
(18.826  106 USD/annum) is very small. Organic waste varies in CSTR3, high temperature of CSTR3 favors both the objectives. Com-
a narrow range (66.4  103–67.0  103 kg/annum) compared to pared to base case, it can be observed that temperature of CSTR3 is
the design optimization case, which means design variables (i.e., higher and increasing with profit (Fig. 5c), which is required to
size of reactors) have larger effect on the amount of organic waste convert un-reacted feed into bio-diesel.
from the process (see Fig. 4a). Temperatures of CSTR0, CSTR1 and
CSTR2 are nearly constant, close to their respective upper bound, 5.2.3. Trade-off between profit and organic waste: 20% decrease in
while TCSTR3 is nearly constant at its lower bound (not shown for waste cooking oil processing rate to 2000 kg/h
brevity). In conclusion, all decision variables in operation optimiza- Fig. 5d shows the obtained Pareto-optimal front between profit
tion favor both objectives similarly, which result in their constant and organic waste, for 20% decrease in waste cooking oil feed rate
values near to one of their bounds. compared to the ‘‘base’’ case. As expected profit has decreased and
organic waste is lower by 8–25% compared to the base case,
5.2.2. Trade-off between profit and organic waste: 10% increase in although the latter is varying over a wide range. So, this requires
waste cooking oil feed rate to 2750 kg/h further analysis of results obtained in both the cases. Now, profit
In this case, 10% additional waste cooking oil is processed com- is about 22% lower than that in the base case, which is due to larger
pared to ‘‘base’’ case of operation. The obtained Pareto-optimal magnitude of constant terms in the profit calculation. Revenue ob-
front is shown in Fig. 5a. Similar to ‘‘base’’ case operation optimiza- tained in this case is nearly 20% lower than that in the ‘‘base’’ case,
tion, objective values vary over small ranges. Increase in feed rate which means that waste cooking oil to bio-diesel conversion is the
of plant does not increase FCI and operating labor as number of same in both the cases. Similar to the ‘‘base’’ case, temperatures of
processing units is fixed. Consequently, profit for 10% increase in CSTR0 and CSTR1 are constant, near to their respective upper bound
waste cooking oil feed rate, is slightly higher (by 10.9%) compared (not shown for brevity). TCSTR2 is varying between its lower and
to ‘‘base’’ case. In this optimization for increased feed rate, temper- upper bounds, while TCSTR3 is constant near its lower bound except
atures of CSTR1 and CSTR2 are constant near their upper bound of a high temperature corresponding to the large amount of organic
70 °C (not shown for brevity). Variation in temperature of CSTR0 waste (see solution ‘‘+’’ in Fig. 5f). Variation in temperature of
with profit has opposite trend to the Pareto-optimal front (see CSTR2 shows opposite trend to the Pareto-optimal front (see
Fig. 5a and b). Lower value of TCSTR0 leads to lower conversion of Fig. 5d and e).
oleic acid into bio-diesel in the pre-treatment reactor, which in- In the decreased feed rate case, one solution shown as ‘‘+’’ in
creases organic waste from the process. Temperature of CSTR3 is Fig. 5d is further away from the remaining non-dominated solu-
also increasing with profit, to its upper bound (Fig. 5a and c). In this tions; organic waste has increased significantly (57  103 kg/an-
case of increased feed rate, organic waste has increased by 9–13% num) for this solution. Temperature of CSTR2 is lower for

Fig. 5. Selected results for simultaneous profit maximization and organic waste minimization: 10% increase in waste cooking oil feed rate (plots a–c on the left side), 20%
decrease in waste cooking oil feed rate (plots d–f on the right side).
 S. Sharma, G.P. Rangaiah / Fuel 103 (2013) 269–277 277

Table 3
Comparison of three optimal solutions chosen for different feed rates (one solution from ‘‘base’’ case and solutions shown as ‘‘’’ in Fig. 5a and d).

Objectives and decision variables Decreased feed rate (2000 kg/h) Normal feed rate (2500 kg/h) Increased feed rate (2750 kg/h)
Profit (million $) 13.84 17.83 19.79
Organic waste (kg/annum) 51,300 66,430 73,917
TCSTR0 (°C) 59.85 59.97 59.55
TCSTR1 (°C) 69.17 69.96 69.87
TCSTR2 (°C) 65.56 69.99 69.84
TCSTR3 (°C) 50.40 50.74 67.59

solution ‘‘+’’ than that for the remaining non-dominated solutions. [2] Freedman B, Butterfield RO, Pryde EH. Trans-esterification kinetics of soybean
oil. J Am Oil Chem Soc 1986;63:1375–80.
At lower TCSTR2, more un-reacted feed enters CSTR3, which is con-
[3] Lotero E, Liu YJ, Lopez DE, Suwannakarn K, Bruce DA, Goodwin JG. Synthesis of
verted into bio-diesel at the higher temperature of CSTR3. Organic bio-diesel via acid catalysis. Ind Eng Chem Res 2005;44:5353–63.
waste (i.e., feed and glycerol) from CSTR3 goes to the downstream [4] Zhang Y, Dube MA, McLean DD, Kates M. Bio-diesel production from waste
units as it is not removed using a three-phase separator. Hence, or- cooking oil-1: Process design and technological assessment. Bioresour Technol
2003;89:1–16.
ganic waste has increased suddenly, while profit has also increased [5] Zhang Y, Dube MA, McLean DD, Kates M. Bio-diesel production from waste
as larger amount of bio-diesel is produced at high value of TCSTR3. cooking oil-2: Economic assessment and sensitivity analysis. Bioresour
One corner solution on the Pareto-optima front of ‘‘base’’ case Technol 2003;90:229–40.
[6] Hass MJ, McAloon AJ, Yee WC, Foglia TA. A process model to estimate bio-
and two corner solutions on the Pareto-optimal fronts, shown as diesel production costs. Bioresour Technol 2006;97:671–8.
‘‘’’ in Fig. 5a and d, have been selected for comparison. Table 3 [7] West AH, Posarac D, Ellis N. Assessment of four bio-diesel production process
presents values of objectives and decision variables corresponding using Hysys. Bioresour Technol 2008;99:6587–601.
[8] Zhang C, Rangaiah GP, Kariwala V. Design and plant wide control of a bio-
to these selected solutions. Temperatures of CSTR0 and CSTR1 are diesel plant. In: Rangaiah GP, Kariwala V, editors. Plant-wide control: recent
constant, near to their respective upper bound (i.e., 60 and 70 °C) developments and applications. Wiley; 2012.
in all three cases. TCSTR2 increases with the feed rate increase from [9] Ghadge SV, Raheman H. Process optimization for bio-diesel production from
mahua (Madhuca Indica) oil using response surface methodology. Bioresour
2000 to 2500 kg/h; it is limited by its upper bound for feed rate of Technol 2006;97:379–84.
2750 kg/h. Temperature of CSTR3 is near to its lower bound for the [10] Myint LL, El-Halwagi MM. Process analysis and optimization of bio-diesel
feed rates of 2000 and 2500 kg/h, and increases for feed rate of production from soybean oil. Clean Technol Environ Policy 2009;11:263–76.
[11] Nicola GD, Moglie M, Pacetti M, Santori G. Bioenergy II: modeling and multi-
2750 kg/h as larger amount of feed has to be converted into bio-
objective optimization of different bio-diesel production processes. Int J Chem
diesel in CSTR3. React Eng 2010;12:8.
[12] Masuduzzaman, Rangaiah GP. Multi-objective optimization applications in
chemical engineering. In: Rangaiah GP, editor. Multi-objective optimization:
6. Conclusions techniques and applications in chemical engineering. World Scientific; 2009.
[13] Srinivas M, Rangaiah GP. Differential evolution with taboo list for solving
In this study, bio-diesel production process is simulated in As- nonlinear and mixed-integer nonlinear programming problems. Ind Eng Chem
Res 2007;46:7126–35.
pen Hysys; this process uses waste cooking oil as the feed, which [14] Chua CBH, Lee HM, Low JSC. Life cycle emissions and energy study of bio-diesel
facilitates its better utilization and promotes sustainability. Mul- derived from waste cooking oil and diesel in Singapore. Int J Life Cycle Assess
ti-objective differential evolution with taboo list is used to obtain 2010;15:417–23.
[15] Garcia M, Gonzalo A, Sanchez JL, Arauzo J, Pena JA. Prediction of normalized
trade-off solutions among different objectives. Two bi-objective
bio-diesel properties by simulation of multiple feed-stocks blends. Bioresour
design optimization problems are solved to study the effect of Technol 2010;101:4431–9.
important design and operation variables on the performance of [16] Morais S, Couto S, Martins AA, Mata TM. Designing eco-efficient bio-diesel
production processes from waste vegetable oils. In: 20th European symposium
this process. First, trade-off between profit and FCI is explored in
on computer aided, process engineering; 2010. p. 253–8.
design optimization, and it is found that objectives do not vary [17] Van Gerpen J, Shanks B, Pruszko R, Clements D, Knothe G. Bio-diesel
much due to the fixed amount of feed rate. Next, process design production, technology. NREL/SR-510-36244, 2004. <www.nrel.gov/docs/
is optimized for maximization of profit and minimization of organ- fy04osti/36244.pdf>.
[18] Conneman J, Fischer J. Bio-diesel in Europe 1998: bio-diesel processing
ic waste simultaneously, which examines environmental impact of technology. Int liquid bio-fuels congress, Brazil; 1998.
the process design. Following this, bio-diesel operation of a se- [19] Noureddini H, Zhu D. Kinetics of trans-esterification of soybean oil. Biocatal
lected design in the presence of feed rate variation has been suc- Articles 1997;74(11):1457–63.
[20] Franca BB, Pinto FM, Pessoa FLP, Uller AMC. Liquid–liquid equilibria for castor
cessfully optimized for profit maximization and organic waste oil bio-diesel + glycerol + alcohol. J Chem Eng Data 2009;54:2359–64.
minimization simultaneously. In short, the study demonstrates [21] Berrios M, Siles J, Martin MA, Martin A. A kinetic study of the esterification of
the potential of MOO for bio-diesel process design and operation. free fatty acid (FFA) in sunflower oil. Fuel 2007;86(15):2383–8.
[22] Turton R, Bailie RC, Whiting WB, Shaeiwitz JA. Analysis synthesis and design of
chemical processes. 3rd ed. Prentice Hall; 2009.
References [23] Deb K. Multi-objective optimization using evolutionary algorithms. John
Wiley; 2001.
[1] Canakci M, Van Gerpen G. Bio-diesel production from oils and fats with high [24] Deb K, Pratap A, Agarwal S, Meyarivan T. A fast and elitist multi-objective
free fatty acids. Trans ASAE 2001;44:1429–36. genetic algorithm: NSGA-II. IEEE Trans Evol Comput 2002;6:182–97.