Applied Energy 102 (2013) 283–292

Contents lists available at SciVerse ScienceDirect

Applied Energy
journal homepage: www.elsevier.com/locate/apenergy

Transesterification of waste cooking oil by heteropoly acid (HPA) catalyst:

Optimization and kinetic model
Amin Talebian-Kiakalaieh a, Nor Aishah Saidina Amin a,⇑, Alireza Zarei a, Iman Noshadi b
a
Chemical Reaction Engineering Group (CREG), Faculty of Chemical Engineering, Universiti Teknologi Malaysia (UTM), 81310 Skudai, Johor, Malaysia
b
Department of Chemical, Materials & Biomolecular Engineering, University of Connecticut, Storrs, CT 06269, United States

h i g h l i g h t s

" Biodiesel was produced from waste cooking oil using heteropoly acid catalyst.
" Optimization and modeling of the process was performed.
" Kinetic parameters were determined for the triglyceride reaction.
" Catalyst could be reused up to four times without significant loss of activity.
" Biodiesel conformed with ASTM D6751 standards.

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

Article history: Transesterification of waste cooking oil with heterogeneous (heteropoly acid) catalyst and methanol has
Received 20 February 2012 been investigated. Response Surface Methodology (RSM) and Artificial Neural Network (ANN) were
Received in revised form 25 May 2012 employed to study the relationship between process variables and free fatty acid conversion and for pre-
Accepted 1 July 2012
dicting the optimal parameters. The highest conversion was 88.6% at optimum condition being 14 h,
Available online 29 August 2012
65 °C, 70:1 and 10 wt% for reaction time, reaction temperature, methanol to oil molar ratio and catalyst
loading, respectively. The RSM and ANN could accurately predict the experimental results, with
Keywords:
R2 = 0.9987 and R2 = 0.985, respectively. Kinetics studies were investigated to describe the system. The
Biodiesel
Heterogeneous
reaction followed first-order kinetics with the calculated activation energy, Ea = 53.99 kJ/mol while the
Transesterification pre-exponential factor, A = 2.9  107 min1. These findings can help improve an environmentally friendly
Waste cooking oil biodiesel process that conforms to ASTM D6751 standards.
Optimization Ó 2012 Elsevier Ltd. All rights reserved.
Kinetic

1. Introduction some modification before the oils can be used in diesel engines.
While dilution, thermal cracking (pyrolysis), and micro-emulsifica-
Worldwide interest in biodiesel production has increased dra- tion are various methods used to modify vegetable oil, transesteri-
matically recently due to relentless environmental concerns, steep fication is the most common conversion process [12]. The
hike in fuels price and increasing demand for fossil fuel consump- utilization of various types of vegetable oil has raised concerns as
tion [1]. Biodiesel is environmentally friendly and can eliminate or vegetable oil is an important component of food chain material
decrease engine emissions such as unburned hydrocarbons (68%), [13]. In addition, vegetable oils are more expensive than non-edi-
particles (40%), carbon monoxide (44%), sulfur oxide (100%), and ble oils. Therefore, researchers have focused on non-edible or
polycyclic aromatic hydrocarbons (80–90%) [2,3]. In addition, it is waste cooking oils as feed stocks to decrease biodiesel production
biodegradable, and technically feasible. Biodiesel, which can be costs by 60–90% [14,15]. The three-step and overall chemical reac-
used as alternative fossil fuels, is a mono alkyl ester of fatty acids tions for biodiesel production are described in the following equa-
produced from vegetable oil or animal fats [4–9]. More than 300 tions, respectively:
types of fatty acid sources such as animal fats or plant lipids can 8 k1
be used for biodiesel production [10,11]. Vegetable oils require >
> TG þ CH3 OH $ DG þ R1 COOCH3
>
> k4
>
>
< k2
DG þ CH3 OH $ MG þ R2 COOCH3 ð1Þ
>
> k5
⇑ Corresponding author. Tel.: +60 7 553 5579; fax: +60 7 558 8166. >
>
>
> k
E-mail addresses: [email protected] (A. Talebian-Kiakalaieh), nor- : MG þ CH3 OH $3 GL þ R3 COOCH3
[email protected] (N.A.S. Amin). k6

0306-2619/$ - see front matter Ó 2012 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.apenergy.2012.07.018
284 A. Talebian-Kiakalaieh et al. / Applied Energy 102 (2013) 283–292

k7
TG þ 3CH3 OH $ GL þ 3RCOOCH3 ð2Þ 2. Methods
k8

2.1. Materials
where TG, DG, MG, CH3OH, GL, RCOOCH3 are triglyceride, diglyce-
rides, monoglycerides, methanol, glycerol, and fatty acid methyl es- Methanol (purity > 99.8% v/v), heteropoly acid H3PW12O406H2O
ter (FAME), respectively. (PW12), purity > 98.5%) were obtained from Sigma Aldrich (Malay-
The process optimization required costly equipments and time sia). Waste cooking oil was obtained from local restaurants in Uni-
consuming tests. Therefore, the implementation of modeling tools versiti Teknologi Malaysia (UTM). The free fatty Acid (FFA) was
such as Response Surface Methodology (RSM) and Artificial Neural determined by gas chromatograph mass spectroscopy (GC/MS)
Network (ANN) can reduce a large number of laboratory tests and using NukolTM (15 m  0.53 mm, 0.5 lm) column and Flame Ioniza-
associated costs. ANN is a suitable method to solve various prob- tion Detector (FID) detector with helium as the carrier gas. The
lems in science and engineering fields such as robotics, forecasting, oven temperature was programmed at 110 °C for 30 s, and ramped
power systems, signal processing, medicine, psychological sci- to 220 °C at a rate of 10 °C/min for 8 min. In addition, the injector
ences, wind speed prediction, photovoltaic systems, and particu- and detector temperatures were 220 °C and 250 °C respectively
larly in renewable energy [16–24], especially in areas where (Table 1).
conventional methods fail. RSM is a combination of statistical
and mathematical techniques and is suitable for analyzing the ef- 2.2. Transesterification reaction
fects of several reaction parameters (independent variables) on
the response [25]. This technique is effective for improving, devel- The transesterification reaction was performed in a 500 ml
oping and optimizing complicated systems [26]. There are various three-neck flask. For each run, the methanol and catalyst mixture
reports on successful optimization of RSM in different processes was preheated to the desired temperature. Next, WCO that was
such as biodiesel [27–30]. separately heated to a certain temperature was added to the mix-
Among solid acid catalysts, heteropoly acids (HPAs) are quite ture and stirred at 300 rpm mixing intensity. After the reaction was
active for heterogeneous reactions. The advantages of HPAs in- completed, the mixture was then heated to 90 °C for 3 h to allow
clude: (1) insensitive to FFA content (2) able to catalyze esterifica- for the evaporation of excess methanol. The reaction product was
tion and transesterification reactions simultaneously (3) easier to comprised of two layers; the upper layer was methyl ester or bio-
separate (4) requires no washing for product (FAME) (5) gives diesel while glycerin was in the lower layer. In addition, the cata-
higher yield (6) requires much lower catalyst loading compared lyst was easily separated with a simple filtration process and could
to other processes (7) environmentally friendly (8) eliminates or be reused many times. Since there was no catalyst in the final
decreases corrosion and (9) can be reused several times [31]. There product, no washing was needed [36]. Thus, the reaction did not
are few reports on transesterification of triglyceride catalyzed by produce any waste water and the process can be regarded as envi-
HPAs in a batch reactor. Narasimharao et al. [32] used insoluble ronment-friendly. Finally, the quality of the product was compared
Cs salt of heteropoly acids for transesterification of triglyceride to ASTM D6751. ISO 3675(1998) and ASTM D445 methods were
while Noshadi et al. [33] used HPA catalyst in transesterification used to determine the density at 15 °C and kinematic viscosity at
reaction of WCO by reactive distillation process. Therefore, finding 40 °C respectively. In addition, the water content, flash point, cloud
kinetic parameters such as reaction rate constant (k), activation en- point and pour point were determined with ASTM D2709, ASTM
ergy (Ea), and frequency factor (A) for transesterification reaction D93, ASTM D5773, and ASTM D2500, respectively (Table 2).
using HPAs is very important for process simulation and The conversion of WCO is calculated by the following equation:
commercialization.  
Brahmkhatri and Patel [34] used 12-Tungstophosphoric acid AV final
Conversion ð%Þ ¼ 1  100% ð3Þ
anchored to MCM-41 catalyst for biodiesel production by esterifi- AV initial
cation of palmitic acid with methanol. They evaluated the effect
where AVinitial and AVfinal are the acid values of feed and product,
of molar ratio of acid to alcohol, catalyst loading and reaction tem-
respectively.
perature on reaction conversion. Their studies revealed high cata-
lyst activity and first- order kinetic. The reaction rate constant (k)
2.3. Experimental design
for 40, 60, and 80 °C were 2.3  103, 6.4  103, and 12.2  103 -
min1, respectively. In addition, the activation energy and fre-
Investigating a five level four factor Central Composite Design
quency factor at 40 °C in 4 h reaction time were 38 kJ mol1 and
(CCD) required 30 experiments in this study. The fractional facto-
9.0 min1, respectively.
rial design included six center points, eight axial points, and 16 fac-
Birla et al. [35] investigated biodiesel production from waste
torial points. The ranges and the design levels of each variable are
cooking oil in the presence of calcined snail shell as a heteroge-
given in Table 3. The performance of the process was evaluated by
neous catalyst. The effects of different reaction parameters such
analyzing the transesterification reaction conversion as the re-
as catalyst concentration, molar ratio of methanol to oil, time
and reaction temperature were evaluated. They reported a 99.5%
conversion with 87.28% yield. In addition, the activation energy Table 1
(Ea) was 79 kJ/mol and the frequency factor (A) was 2.89  109 - The chemical and physical properties of WCO.
min1 for first-order kinetics. FFA composition (wt%)
In the present study, heterogeneous transesterification of waste
Palmitic 26.5
recycled cooking oil using heteropoly acid H3PW12O406H2O Oleic 21.35
(PW12) catalyst was investigated. RSM and ANN were applied to Stearic 10.9
determine the optimum condition among variables such as reac- Linoleic 1.7
tion temperature, reaction time, methanol to oil molar ratio and Physical properties
Water content 1%
catalyst loading. A kinetic model was suggested and the kinetic
Color Golden yellow
parameters were determined by fitting the model with the exper- Acid value 10.2 (mg KOH/g of oil)
imental results.
 A. Talebian-Kiakalaieh et al. / Applied Energy 102 (2013) 283–292 285

Table 2 Conversion ð%Þ ¼ 87:98 þ 1:50  A þ 2:34  B þ 4:26  C

Properties of biodiesel compared with ASTM standards.
þ 1:93  D  5:58  A2  2:39  B2  3:89
Biodiesel ASTM D6751
Acid value (mg KOH/g of oil) 0.493 60.5
 C 2  2:66  D2  0:64  A  B  0:41
Kinematic viscosity (at 40 °C St (m2/sec)  106 5.24 1.9–6  A  C þ 0:55  A  D  1:13  B  C
Density (kg/m3) 867.2 860–894
Flash point (°C) 168 100–170 þ 0:23  B  D  0:98  C  D: ð7Þ
Cloud point (°C) 1 3 to 15
Pour point (°C) 4.4 5 to 10 where the reaction temperature (A), reaction time (B), methanol to
oil molar ratio (C), and catalyst amount (D).
The model F-value of 950.55 implies that it is significant. There
sponse. The quadratic equation for optimization is given in the fol- is only a 0.01% chance that a ‘‘Model F-Value’’ this large can occur
lowing equation: due to noise. The value of ‘‘Prob > F’’ less than 0.05 indicates the
model terms are significant. In this study A, B, C, D, A2, B2, C2, D2,
X
4 X
4 XX
4
g ¼ b0 þ bi vi þ bii v2i þ bii vi vj þ e ð4Þ AB, AC, AD, BC, CD are significant model terms. The ‘‘lack of fit F-va-
i¼1 i¼1 i¼1 j¼iþ1 lue’’ of 0.27 implies the lack of fit is not significant relative to pure
error. Non-significant lack of fit is good. The ‘‘predicted_R2’’ of
where g is the response, b0 is constant coefficient, bi, bii, and bij are 0.9978 is in reasonable agreement with the ‘‘Adjusted _R2’’ of
linear, quadratic and second order interaction coefficient, respec- 0.9967. The suitable coefficient of determination (R2 = 0.9987) in-
tively. Xi and Xj are independent variables where e is the error. fers that the quadratic model is highly significant and sufficient
The R-square and Adjust R-Square were determined by the follow- to represent the actual relationship between the variables and re-
ing equations: sponses (Table 5). In addition, the ANOVA results reveal that the
SSResidual molar ratio of methanol to oil (C) is the most important variable
R2 ¼ 1  ð5Þ in this study since the Prob > F (<0.0001) is the lowest while the
SSmodel þ SSResidual
F-value (2936.52) was the highest. Fig. 1 illustrates that the pre-
SSResidual =DF Residual dicted values are much closer to the experimental values inferring
R2adj ¼ 1  ð6Þ that the model has been successfully developed in capturing the
ðSSModel þ SSResidual Þ=ðDF Model þ DF Residual Þ
correlation between reaction independent variables and FAME
where SS and DF are the sum of square and degree of freedom, conversion.
respectively.
Design Expert (6.0.7, State ease, Inc., USA) was used for regres-
sion (ANOVA) and graphical analysis and MATLAB toolbox 3.2. Artificial neural network (ANN)
(7.11.0.584 (R2010b)) was used for ANN optimization. R-Square
is the most important factor in examining the variability of inde- The selection of an optimum amount of connection types, learn-
pendent variables. Meanwhile, the F-test is for checking the statis- ing algorithms, transfer function of input and hidden layers, and
tical significance of the model [37]. The number of experiments, particularly the number of hidden neurons are the most important
experimental conditions and results are summarized in Table 4. factors for successful application of ANN. In this study, back prop-
An additional 16 runs which included (molar ratio of methanol agation (BP) was selected for the learning algorithm and Leven-
to oil (30:1, 40:1, 50:1, 60:1, 70:1, 80:1), catalyst amount (5, 7.5, berg–Marquardt (LM) for training. The applications of BP learning
10, 12.5 wt%), reaction temperature (50, 55, 60, 65, 70 °C), and algorithms are recommended by a majority of researchers
reaction time (2, 4, 6, 8, 10, 12, 14, 16 h) were performed. A total [34,38,39] because they are very powerful in function optimization
of 120 samples at different time intervals were collected to illus- and modeling. The transfer function types for hidden layer and out-
trate the effect of each variable on reaction conversion and to dem- put layer are very important because they have a significant influ-
onstrate the RSM and ANN modeling results. ence on the neural networks learning rate or overall performance
of the neural network. In this study, linear and Hyperbolic Tangent
Sigmoid (tansig) functions were selected for input and output hid-
3. Results and discussion
den layer, respectively based upon the wide application of these
functions by various researchers [40–42].
3.1. Response surface methodology (RSM)
Two methods can be used to determine the optimum number of
hidden neurons. The first is the lowest amount of Mean Square Er-
Fitting the data using different types of models such as linear,
ror (MSE) for application of various hidden neurons and the second
two factorial, quadratic, and cubic revealed that the transesterifica-
is by trial and error. Firstly, the amount of MSE for 1–20 hidden
tion of WCO and methanol in the presence of HPA catalyst has been
neurons was determined. According to Fig. 2a, the red1 boxes illus-
properly explained by the quadratic polynomial model. The best
trate the (6, 7, 10, 12, 13, 15, 16, 17, 20) number of hidden neurons
predicted FFA conversion models based on coded factors are shown
which have lower MSE values because the black line is much too
in the following equation:
close to (0) amount of MSE. The blue boxes and red line at the middle
of each blue box indicate around 50% of data and the mean value for
each number of hidden neurons, respectively. In addition, the red
pluses indicate data which can be ignored in the modeling process.
Table 3
Experimental level coded and range of independent parameters. However, to find the optimal number, the best number of hidden
neuron obtained from Fig. 2a on the well fitting or higher amount
Independent parameters Coded levels of parameters
of R-Square were tested. Fig. 2b indicated that the optimum number
2 1 0 +1 +2 of hidden neurons was 12 with R2 = 0.985. However, small amounts
Temperature (°C) A 55 60 65 70 75 of hidden neurons can limit process modeling while much higher
Time (h) B 6 10 14 18 22
Molar ratio (mol/mol) C 30 50 70 90 110 1
Catalyst (wt%) D 5 7.5 10 12.5 15 For interpretation of color in Figs. 1–6, the reader is referred to the web version of
this article.
286 A. Talebian-Kiakalaieh et al. / Applied Energy 102 (2013) 283–292

Table 4
Central composite design of four independent parameters with experimental, RSM, and ANN response values.

Run Tem (°C) Time (h) Molar ratio (mol/mol) Catalyst amount (wt%) Experimental Conversion (%) RSM Conversion (%) ANN Conversion (%)
1 60 10 50 7.5 61 61.36 61.07
2 70 10 50 7.5 64.91 65.05 64.91
3 60 18 50 7.5 68.73 68.78 70.58
4 70 18 50 7.5 70.15 70.24 70.21
5 60 10 90 7.5 74.5 74.62 74.54
6 70 10 90 7.5 76.83 76.98 76.81
7 60 18 90 7.5 77.71 77.84 77.96
8 70 18 90 7.5 78 77.65 78.09
9 60 10 50 12.5 65.1 65.3 65.14
10 70 10 50 12.5 71.5 71.5 71.52
11 60 18 50 12.5 74 73.98 74
12 70 18 50 12.5 77.9 77.62 83.1
13 60 10 90 12.5 74.9 74.95 74.88
14 70 10 90 12.5 79.7 79.5 79.7
15 60 18 90 12.5 79.4 79.1 79.41
16 70 18 90 12.5 81 81.1 80.88
17 55 14 70 10 62.8 62.65 62.84
18 75 14 70 10 68.5 68.66 68.53
19 65 6 70 10 74 73.73 72.8
20 65 22 70 10 82.8 83.07 82.12
21 65 14 30 10 64 63.87 64.05
22 65 14 110 10 80.8 80.93 80.89
23 65 14 70 5 73.7 73.5 73.74
24 65 14 70 15 81 81.21 80.99
25 65 14 70 10 87.2 87.98 88.4
26 65 14 70 10 88.4 87.98 88.4
27 65 14 70 10 88 87.98 88.4
28 65 14 70 10 88.2 87.98 88.4
29 65 14 70 10 88.6 87.98 88.4
30 65 14 70 10 87.5 87.98 88.4
2 2
R = 0.9987 R = 0.985

Table 5
The results for ANOVA test.

Source Sum of square Degree of freedom Mean square F-value Prob > F
Model 1978.03 14 141.29 950.55 <0.0001 Significant
A 54.15 1 54.15 364.31 <0.0001 Significant
B 130.90 1 130.90 880.66 <0.0001 Significant
C 436.48 1 436.48 2936.52 <0.0001 Significant
D 89.20 1 89.20 600.15 <0.0001 Significant
A2 854.63 1 854.63 5749.75 <0.0001 Significant
B2 157.26 1 157.26 1058.02 <0.0001 Significant
C2 416.01 1 416.01 2798.80 <0.0001 Significant
D2 193.63 1 193.63 1302.71 <0.0001 Significant
AB 6.54 1 6.54 44.00 <0.0001 Significant
AC 2.73 1 2.73 18.37 0.0006 Significant
AD 4.79 1 4.79 32.19 <0.0001 Significant
BC 20.45 1 20.45 137.60 <0.0001 Significant
BD 0.88 1 0.88 5.91 0.0280 Significant
CD 15.50 1 15.50 104.31 <0.0001 Significant
Residual 2.23 15 0.15
Lack of fit 0.78 10 0.078 0.27 0.9630 Not significant
Pure error 1.45 5 0.29
Cor total 1980.26 29

amounts can create noise in the database, increasing the calculation rate (Lr) of 0.5 and number of cycles (NC) set at 1000. The MSE and
time, obviously. Thus, the 4–12–1 topology was selected for the R-square were determined by the following equations:
FAME conversion (Fig. 3).
1 Xn
All experimental data were divided into three groups: training MSE ¼ ðy  yai Þ2
i¼1 pi
ð8Þ
(70%), validation (15%), and testing data (15%). The reaction time, n
reaction temperature, molar ratio of methanol to oil, and catalyst P
amount were used as input layers, while, the FAME conversion ðModel  predictioni  Experimental  v aluei Þ2
R2 ¼ 1  P i¼1n
ð9Þ
was used as the output layer. The optimum network parameters ðAv erage  Experimental  v aluei  Experimental  v aluei Þ2
i¼1n
such as number of cycles (NC), learning rate (Lr), weights (W)
and Bias (b) values for input, output and hidden layers were exam- where n is the amount of experimental data, and ypi and yai are
ined based on the maximum coefficient of determination (R2) and predicted and actual output values, respectively.
the minimum mean square error (MSE) was obtained by trial and Comparison between experimental and predicted values for
error. The training was done for experimental data with a learning training, validity and testing data, illustrated that ANN has high
 A. Talebian-Kiakalaieh et al. / Applied Energy 102 (2013) 283–292 287

3.3. Effect of methanol to oil molar ratio on transesterification reaction

According to the ANOVA test result, the methanol to oil molar

ratio is one of the most important parameters affecting the conver-
sion of methyl ester. Based on the stoichiometric transesterifica-
tion reaction, three moles of methanol react with one mole of
WCO to produce three moles of methyl ester and one mole of glyc-
erol [43]. However, in practical terms, increasing the molar ratio of
methanol to oil can increase the reaction conversion and yield
since the transesterification reaction is reversible and the excess
alcohol shifts the equilibrium to the product side [44]. Fig. 4a illus-
trates that conversion increased with time as the methanol to oil
ratio increased from 30:1 to 70:1. The maximum conversion was
88.6% when the molar ratio of methanol to oil was 70:1 mol/mol
at 14 h. However, the conversion was slightly reduced for molar ra-
tios beyond 70:1 mol/mol since the glycerol separations became
Fig. 1. The RSM plot of predicted versus actual conversion (%).
more difficult due to emulsion of biodiesel and glycerol. There
potential for predicting known and unknown data. The predicted are various reports on reduced reaction conversion and yield at
response values for ANN are summarized in the last column of Ta- higher molar ratio. Birla et al. [35] mentioned that the maximum
ble 4. Overall, the performance of both ANN and RSM was good. conversion of 95% was obtained at the molar ratio of 6.03:1 but
However, the RSM model was slightly better in fitting the experi- for higher molar ratio for instance at 9.65:1 the conversion was re-
mental data compared to ANN, but ANN predicted slightly higher duced to 86.02%. Meanwhile, Jain et al. [45] reported a 90.6% yield
conversion (88.40%) compared to RSM (87.98%) at the optimum for biodiesel at an optimum molar ratio of methanol to oil of 3:7
reaction parameters. but the methyl ester yield decreased beyond this. In addition,

Fig. 2. Optimum number of hidden neurons (a) based on lowest MSE and (b) based on highest R-square amounts.
288 A. Talebian-Kiakalaieh et al. / Applied Energy 102 (2013) 283–292

Fig. 3. The multilayer full feed forward neural network topology for the estimation of FAME conversion.

Xuejun et al. [46] and Hak-joo et al. [47] reported similar trends for led to a negative effect on transesterification conversion [53,54].
the yield. Meanwhile, Birla et al. [35] reported that for temperatures higher
than the methanol boiling point of 65 °C, the solvent vaporized,
3.4. Effect of catalyst loading on transesterification reaction thus, methanol remained in the vapor phase in the reactor and
was less available in reaction environment.
The effect of catalyst concentration on transesterification was
investigated by varying the catalyst loading from 5 to 12.5 wt% 3.6. Effect of reaction time on transesterification reaction
based on the weight of WCO (Fig. 4b). The reaction conversion in-
creased as the catalyst amount was increased up to 10 wt%. Fig. 4b Fig. 4e indicates a dramatic increase in the conversion as time
illustrates that reaction conversion increased from 19% to 73% for increased from 2 to 14 h at optimum reaction conditions of
5 wt% catalyst and also it increased from 40% to 88.6% for 10 wt% 10 wt% catalyst loading, 70:1 M ratio of methanol to oil and 65 °C
at 14 h reaction time. However, the conversion decreased with ex- reaction temperature. The maximum conversion reached 88.6% at
cess loading due to increased viscosity of the reaction mixture. Sri- 14 h reaction time and it remained constant at longer reaction
vastava and Prasad [48] and Zhang et al. [49] reported similar times. Jain et al. [45] reported that in esterification reaction, a high-
effect on conversion. Meanwhile, Taufiq-Yap et al. [50] reported er initial rate led to increased methyl ester yield before the rate de-
that when the catalyst dosage was too much, more biodiesel prod- creased before remaining constant. However, in transesterification
ucts were absorbed by catalysts, and as a result, final conversion reaction the methyl ester yield increased dramatically with time
decreased. Thus, the optimum amount of catalyst was 10 wt% at until no further conversion of triglyceride to methyl ester took
14 h reaction time. A three dimensional plot (Fig. 4c) represents place after the optimum reaction time. The 14 h reaction time is
the conversion as a function of catalyst loading and molar ratio reasonable since the reaction temperature is at a relatively low
at 14 h and 65 °C. Moreover, Fig. 4c clearly depicts that higher temperature of 65 °C. Some researchers increased the reaction
amounts of molar ratio and catalyst weight beyond 70:1 mol/mol temperature to decrease the reaction time such as Tan et al. [55]
and 10 wt%, respectively, have a negative influence on reaction who increased temperature to higher than 300 °C to reduce the
conversion. Higher molar ratio deactivated the catalyst and re- reaction time to around 30 min. Higher temperature and pressure
versed the reaction to the reactant side. Mass transfer effect of so- require too much energy which is uneconomical for biodiesel
lid catalyst could also be a factor in the decreasing conversion [50]. plants. Meanwhile, Fig. 4f indicates the 3D surface plot for interac-
Srilatha et al. [51] reported a similar trend for the effect of molar tion of reaction temperature and time at 70:1 M ratio of methanol
ratio and catalyst on reaction conversion. to oil and 10 wt% catalyst loading. The results indicated that the
highest conversion of 88.6% was obtained at 65 °C reaction temper-
3.5. Effect of reaction temperature on transesterification reaction ature and 14 h reaction time.

The reaction was carried out at five different temperatures 3.7. Optimization of process variables
ranging from 50 °C to 70 °C in order to find the influence of tem-
perature on reaction conversion (Fig. 4d). The conversion increased The transesterification of WCO was optimized by RSM and ANN
dramatically as the temperature increased between 50 °C and for obtaining the highest FAME conversion. The predicted conver-
65 °C. In fact, the conversion ramped from 20% to 60% at 50 °C sions were 87.98% and 88.40% by RSM and ANN, respectively at
and from 31% to 88.6% at 65 °C. The results are in good agreement optimum condition of 10 wt% catalyst amount, 70:1 M ratio of
with other studies such as Wei et al. [52] who obtained the high methanol to oil, 65 °C reaction temperature, and 14 h reaction
yield (>95%) at the optimum temperature of 65 °C. However, the time. However, the experimental conversion value was 88.68%.
conversion decreased for temperatures beyond 65 °C which can Thus, RSM and ANN could accurately predict and model the transe-
be attributed to chemical reactions. During the frying process, sterification of WCO with only 0.7% and 0.23% error, respectively.
three types of chemical reactions, hydrolytic, oxidative, and In addition, the results of 16 more experimental runs were in a
thermolytic, occurred. These reactions created some undesirable good agreement with RSM and ANN results. In fact, RSM and
components such as some polymerized triglycerides and FFA that ANN could adequately simulate and model the reaction and could
 A. Talebian-Kiakalaieh et al. / Applied Energy 102 (2013) 283–292 289

(a) (b)

(c) (d)

(e) (f)

Fig. 4. Effect of different reaction parameters on reaction conversion. (a) Effect of ratio of methanol-oil on conversion (catalyst = 10 wt%, reaction temperature = 65 °C). (b)
Effect of catalyst amount on conversion of biodiesel (molar ratio of methanol to oil = 70:1, reaction temperature = 65 °C). (c) Interaction of methanol-oil ratio versus catalyst
loading (reaction time = 14 h, reaction temperature = 65 °C). (d) Effect of reaction temperature on conversion of biodiesel (molar ratio of methanol to oil = 70:1, catalyst
amount = 10 wt%). (e) Effect of reaction time on conversion of biodiesel (molar ratio of methanol to oil = 70:1, reaction temperature = 65 °C, catalyst weight = 10 wt%). (f)
Interaction of reaction time versus temperature (catalyst amount = 10 wt%, molar ratio of methanol-oil = 70:1).

help to eliminate huge and expensive equipment and time con- compounds (methyl ester) from the surface. Next, the catalyst
suming tests for process optimization. was washed with methanol to eliminate any polar compounds
(glycerol). Consequently, the catalyst was dried by air at room tem-
perature and used for further reactions [33,53,56]. The results re-
3.8. Catalyst recycling and reusability
vealed that the heteropoly acid solid catalyst in this study can be
reused up to four times without any serious reduction in reaction
Low price is the main advantage of conventional homogeneous
conversion, but at the fifth time, the conversion decreased to 70%
catalysts in biodiesel production. However, the reusability of heter-
(Fig. 5).
ogeneous catalyst can dramatically reduce production costs and
improve its economy. In this work, reusability studies were carried
out at optimum reaction conditions of 14 h reaction time, 65 °C 4. Kinetic modeling
temperature, 70:1 mol/mol molar ratio, and 10 wt% catalyst load-
ing. The catalyst was simply separated from the final product Kinetic studies of WCO transesterification have been carried out
and then soaked in hexane overnight to remove any non-polar at the optimum conditions of 70:1 M ratio of methanol to oil, 65 °C
290 A. Talebian-Kiakalaieh et al. / Applied Energy 102 (2013) 283–292

Fig. 5. Number of times for catalyst reusability.

reaction temperature, 10 wt% catalyst amount and 14 h reaction

temperature.
The reaction rate according to the following equation is:
dC A
 ¼ k7 C XA C YB  k8 C W Z
c CD ð10Þ
dt
where CA, CB, CC, and CD were the concentrations of triglycerides,
methanol, glycerol, and methyl ester, respectively. X, Y, W, and Z re-
fer to their reaction order. k7 and k8 are the kinetic constants for the Fig. 6. (a) The linear relationship between time and Ln(1  X). (b) Ln k’ versus 1/
forward and backward reactions, respectively. However, the molar T103 According to Arrhenius equation.

ratio of methanol to oil was much higher than other components,

thus, k7 C YB can be constant. In addition, k7 is much larger than k8
dC TG
[57]. ra ¼  ¼ k7 C TG C 3ROH ð16Þ
dt
Therefore, Eq. (10) has been reduced to:
where CTG, C 3ROH and k7 are concentrations of triglyceride, methanol,
dC A 0
 ¼ k C nA ð11Þ and reaction rate constant, respectively. Eq. (16) followed a second
dt
order reaction rate law. However, due to high methanol to oil molar
0
where k0 is modified rate constant and k ¼ k7 C YB ratio, the change in methanol concentration can be considered as
C A ¼ C A0 ð1  XÞ ð12Þ constant during reaction. Thus, the reaction is assumed to obey
pseudo-first order reaction [35,49,58,59]. The rate of reaction is ex-
where X and CA0 are conversion of fatty acid methyl ester (FAME) pressed the following equation:
biodiesel and the initial concentration of oil or triglycerides, respec-
tively. Therefore, the overall reaction rate can be expressed as the dC TG 0
r a ¼  ¼ k C TG ð17Þ
following equation: dt
0
dX k
0
00
with k ¼ k7  C 3ROH .
¼ ½C A0 ð1  XÞn ¼ k ½C A0 ð1  XÞn ð13Þ After fitting the data in Fig. 6a, Ln (1  X) versus time (t), the lin-
dt C A0
ear relationship supported the main hypothesis as pseudo-first or-
where k00 = k0 /CA0 and n = 1. The above equation is integrated to ob- der of reaction. Different reaction rate constants (k0 ) for various
tain the following equation: temperatures are listed in Table 6.
lnð1  XÞ ¼ k t
0
ð14Þ Fig. 6b depicts the (Ln k0 ) versus (1/T  103) based on Arrhenius
equation. The slope and intercept of this graph indicated the values
where if n – 1 the overall equation is integrated to obtain the fol- of activation energy (Ea) and pre-exponential factor (A) from the
lowing equation: Arrhenius equation (Eq. (18)):
00
ð1  XÞð1nÞ ¼ 1 þ ðn  1Þk C nA0 t ð15Þ 0 Ea
Ln k ¼  þ Ln A ð18Þ
Reaction rate constant (k) is a function of temperature accord- RT
ing to the Arrhenius equation (Eq. (18)). Therefore, finding the The values of 53.99 kJ/mol and 2.9  107 were reported for acti-
reaction conversion at different temperatures (50, 55, 60, 65, and vation energy (Ea) and pre-exponential factor (A), respectively.
70 °C) and time intervals (2, 4, 6, 8, 10, 12, 14, and 16 h) is neces- Freedman et al. [58] reported a range of 33.6–84 kJ/mol for (Ea),
sary (Fig. 4d). Thereafter, substituting these conversions into Eqs. the value of (Ea) obtained in this study was within the range. In
(14) and (15) could determine the reaction rate constant. Fig. 6a addition, the activation energy obtained was much lower than that
illustrates that the transesterification reaction results fitted-well of 79 kJ/mol achieved by Birla et al. [35]. In addition, it is important
with the first order kinetic equation (n = 1, Eq. (14)). On the other to recognize that the reaction is truly governed by kinetics or dif-
hand, if second-order transesterification reaction was assumed, the fusion when the maximum capacity of catalyst was used. Bond
rate of reaction, ra can be expressed in the following equation: [60] reported that the diffusion limited reaction and truly kinetic
 A. Talebian-Kiakalaieh et al. / Applied Energy 102 (2013) 283–292 291

Table 6 [10] Atapour M, Kariminia. Characterization and transesterification of Iranian bitter

First-order reaction rate constants at different temperatures. almond oil for biodiesel production. Appl Energy 2011;88:2377–81.
[11] Qui F, Li Y, Yang D, Li X, Sun P. Biodiesel production from mixed soybean oil
Temperature °C Reaction rate constant (k0 (min1)) and rapeseed oil. Appl Energy 2011;88:2050–5.
[12] Li Q, Yan Y. Production of biodiesel catalyzed by immobilized Pseudomonas
50 0.059
Cepacia lipase from sapium sebiferum oil in micro-aqueous phase. Appl Energy
55 0.067
2010;87:3148–54.
60 0.091 [13] Srinivasan S. The food v fuel debate: a nuanced view of incentive structures.
65 0.144 Renew Energy 2009;34(4):950–4.
70 0.1062 [14] Zhang Y, Dube M A, McLean DD, Kates M. Biodiesel production from waste
cooking oil: 2. Economic assessment and sensitivity analysis. Bioresour Tech
2003;90(3):229–40.
[15] Shivakumar, Pai PS, Shrinivasa Rao BR. Artificial neural network based
limited reaction have as low as 10–15 kJ/mol and higher than prediction of performance and emission characteristics of a variable
25 kJ/mol activation energies, respectively. Therefore, the compression ratio CI engine using WCO as a biodiesel at different injection
timings. Appl Energy 2011;88:2344–54.
53.99 kJ/mol activation energy obtained in the present study dem- [16] Casilari E, Jurado A, Pansard G, Dı ´ az-Estrella A. Modeling aggregate
onstrated that the reaction rate is truly governed by kinetics. The heterogeneous ATM source using neural networks. Electron Lett
biodiesel production at 65 °C reaction temperature and 14 h reac- 1996;32:363–5.
[17] Guo B, Li DK, Cheng CM, Lu ZA, Shen YT. Simulation of biomass gasification
tion time is acceptable and reasonable compared to other studies with a hybrid neural network model. Bioresour Technol 2001;76(2):77–83.
such as Freedman et al. [58] who reported 20 h and 77 °C for bio- [18] Zhao J, Xu M, Li M, Wang B, Liu S. Design and optimization of an Atkinson cycle
diesel production or Song et al. [61]who obtained kinetic parame- engine with the artificial neural network method. Appl Energy
2012;92:492–502.
ters (n = 2.2 and Ea = 32.62 kJ/mol) for esterification reaction of [19] Ismail HM, Ng HN, Queck CW, Gan S. Artificial neural networks modeling of
oleic acid at 220 °C, 6.0 Mpa. These reasons may be enough to engine-out response for a light-duty diesel engine fuelled with biopdiesel
encourage more researchers to focus on HPA solid acid catalyst blends. Appl Energy 2012;92:769–77.
[20] Matallanas E, Castillo-Cagigal M, Gutierrez A, Monasterio-Huellin F, Caamano-
for the commercialization or industrialization of biodiesel Martin E, Masa D, et al. Neyral network controller for active demand-side
production. management with PV energy in the residential sector. Appl Energy
2012;92:90–7.
[21] Li N, Xia L, Shiming D, Xu X, Chan MY. Dynamic modeling and control of a
5. Conclusions direct expansion air conditioning system using artificial neural network. Appl
Energy 2012;91:290–300.
[22] Yap WK, Karri V. ANN virtual sensor for emissions prediction and control. Appl
The highest experimental conversion was 88.6% at the optimum Energy 2011;88:4505–16.
condition of 14 h reaction time, 70:1 M ratio of methanol to oil, [23] Uresh MVJJ, Reddy KS, Kolar AK. ANN-GA based optimization of a high ash
65 °C reaction temperature, and 10 wt% catalyst loading. RSM coal-fired supercritical power plant. Appl Energy 2011;88:4867–73.
[24] Sipocz N, Tobiesen FA, Assadi M. The use of artificial neural network models
(R2 = 0.9987) was slightly better than ANN (R2 = 0.985) in data fit- for CO2 capture plants. Appl Energy 2011;88:2368–76.
ting and estimation capabilities. However, ANN prediction for reac- [25] Draper NR, John JA. Response-surface design for quantitative and qualitative
tion conversion at optimum amount of variables was 88.40% which variables. Technometrics 1998;30(4):423–8.
[26] Bas D, Boyacı IH. Modeling and optimization II: comparison of estimation
was better than 87.98% achieved by RSM. The HPA catalyst exhib- capabilities of response surface methodology with artificial neural networks in
ited good activity and stability up to four cycles. The ANOVA re- a biochemical reaction. J Food Eng 2007;78:846–54.
sults revealed that the molar ratio of methanol to oil was the [27] Banerjee R, Bhattacharyya BC. Optimization of multiple inducers effect on
protease biosynthesis by Rhizopusoryzae. Bioprocess Eng 1992;7:225–8.
most important variable in this study. The kinetic model for the [28] Murthy MSRC, Swaminathan T, Rakshit, Kosugi Y. Statistical optimization of
transesterification of WCO to biodiesel was established. The results lipase catalyzed hydrolysis of methyloleate by response surface methodology.
obtained were the reaction order n = 1, activation energy Bioprocess Eng 2000;22:35–9.
[29] Gonzalez-Fernandes C, Molinuevo-Salces B, Garcia-Gonzalez MC. Evaluation of
Ea = 53.99 kJ/mol, and pre-exponential factor A = 2.9  107 min1.
anaerobic codigestion of microalgal biomass and swine manure via response
surface methodology. Appl Energy 2011;88:3448–53.
[30] Melvin Jose DF, Edwin Raj R, Durgo Prasad B, Robert Kennedy Z, Mohhamed
Acknowledgments Ibrahim A. A multi-variant approach to optimize process parameters for
biodiesel extraction from rubber seed oil. Appl Energy 2011;88:2056–63.
The authors would like to express their sincere gratitude to the [31] Sharma YC, Singh B, Upadhyay SN. Advancements in development and
characterization of biodiesel. A review. Fuel 2008;87:2355–73.
Ministry of Higher Education (MOHE), Malaysia for supporting the
[32] Narasimharao K, Brown DR, Lee AF, Newman AD, Siril PF, Tavener SJ, et al.
project under the Fundamental Research Grant Scheme (FRGS) Structure-activity relations in Cs-doped heteropolyacids catalysts for biodiesel
Votes 78402 and 78681. production. J Catal 2007;248:226–34.
[33] Noshadi I, Amin NAS, Parnas RS. Continuous production of biodiesel from
waste cooking oil in a reactive distillation column catalyzed by solid
References heteropolyacid: optimization using response surface methodology (RSM).
Fuel 2012;94:156–64.
[34] Brahmkhatri V, Patel A. Biodiesel production by esterification of free fatty
[1] Liu CZ, Wang F, Stiles AR, Guo C. Ionic liquids fro biodiesel production:
acids over 12-Tungstophosphoric acid anchored to MCM-41. Ind Eng Chem Res
opportunities and challenges. Appl Energy 2012;92:406–14.
2011;50:6620–8.
[2] Tyson KS, McCormick RL. Biodiesel handling and use guide. 3rd ed. Springfield;
[35] Birla A, Singh B, Upadhyay SN, Sharma YC. Kinetics studies of synthesis of
2006 [VA.National Technical, Information Service.DOE/GO-102006-2358].
biodiesel from waste frying oil using a heterogeneous catalyst derived from snail
[3] Wu X, Leung DYC. Optimization of biodiesel production from camelina oil
shell. Bioresour Technol. http://dx.doi.org/10.1016/j.biortech.2011.11.065.
using orthogonal experiment. Appl Energy 2011;88:3615–24.
[36] Fenghua C, Yang C, Fengying Z, Jing L, Jianghua W, Xiaohong W, et al. Biodiesel
[4] Balat M, Balat H. Progress in biodiesel processing. Appl Energy
production from high acid value waste frying oil catalyzed by super acid
2010;87:1815–35.
heteropolyacid. Biotech Bioeng 2008;101(1):93–100.
[5] Gao CF, Zhai Y, Ding Y, Wu QY. Application of sweet sorghum for biodiesel
[37] Montgomery DC. Design and analysis of experiments, student solutions
production by heterotrophic microalga Chlorella protothecoides. Appl Energy
manual. 5th ed. New York: John Wiley & Sons; 2011.
2010;87:756–61.
[38] Rajendra M, Jena PC, Raheman H. Prediction of optimized pretreatment
[6] Lin L, Zhou CS, Saritporn V, Shen XQ, Dong MD. Opportunities and challenges
process parameters for biodiesel production using ANN and GA. Fuel
for biodiesel fuel. Appl Energy 2011;88:1020–31.
2009;88:868–75.
[7] Wen ZZ, Yu XH, Tu ST, Yan JY, Erik D. Synthesis of biodiesel from vegetable oil
[39] Canakci M, Erdil A, Arcaklioglu A. Performance and exhaust emissions of a
with methanol catalyzed by Li-doped magnesium oxide catalysts. Appl Energy
biodiesel engine. Appl Energy 2006;83:594–605.
2010;87:743–8.
[40] Yuste AJ, Dorado MP. Neural network approach to simulate biodiesel
[8] Bhatti HN, Hanif MA, Qasim M, Ata-Ur-Rehman. Biodiesel production from
production from waste olive oil. Energy Fuels 2006;20:399–402.
waste tallow. Fuel 2008;87:2961–6.
[41] Ying Y, Shao P, Jiang S, Sun P. Artificial neural network analysis of immobilized
[9] Bhatti HN, Hanif MA, Faruq U, Sheikh MA. Acid and base catalyzed
lipase catalyzed synthesis of biodiesel from rapeseed soapstock. Comput
transesterification of animal fats to biodiesel. Iran J Chem Chem Eng
Computing Technol Agric II 2009;2:1239–49.
2008;27(4):41–8.
292 A. Talebian-Kiakalaieh et al. / Applied Energy 102 (2013) 283–292

[42] Basri M, Abdrahman RNZR, Ebrahimpour A, Salleh AB, Gunawan ER, [52] Wei Z, Xu C, Li B. Application of waste eggshell as low-cost solid catalyst for
Abdrahman MB. Comparison of estimation capabilities of response surface biodiesel production. Bioresour Technol 2009;100:2883–5.
methodology (RSM) with artificial neural network (ANN) in lipase-catalyzed [53] Cao F, Chen Y, Zhai F, Li J, Wang J, Wang X, et al. Biodiesel production from high
synthesis of palm-based wax ester. BMC Biotechnol 2007;7:53. acid value waste frying oil catalyzed by superacid heteropolyacid. Biotech
[43] Meher LC, VidyaSagar D, Naik SN. Technical aspects of biodiesel production by Bioeng 2008;101:93–100.
transesterification – a review. Renew Sust Energy Rev 2006;10:248–68. [54] Mittelbach M, Enzelsberger H. Transesterification of heated rapeseed oil for
[44] Enweremada CC, Mbarawa MM. Technical aspects of production and analysis extending diesel fuel. J Am Oil Chem Soc 1999;76:545–50.
of biodiesel from used cooking oil – a review. Renew Sust Energy Rev [55] Tan KT, Gui MM, Lee KT, Mohamed AR. An optimized study of methanol and
2009;13:2205–24. ethanol in supercritical alcohol technology for biodiesel production. J Supercrit
[45] Jain S, Sharma MP, Rajvanshi S. Acid base catalyzed transesterification kinetics Fluids 2010;53:82–7.
of waste cooking oil. Fuel Process Technol 2011;92:32–8. [56] Kulkarni MG, Gopinath R, Meher LC, Dalai AK. Solid acid catalyzed biodiesel
[46] Xuejun L, Huayang H, Yujun W, Shenlin Z. Transesterification of soybean oil to production by simultaneous esterification and transesterification. Royal Soc
biodiesel using SrO as a solid base catalyst. Catal Commun 2007;8:1107–11. Chem (Green Chemistry) 2001;8:1056–62.
[47] Hak-Joo K, Bo-Seung K, Min-Ju K, Young MP, Deog-Keun K, Jin-Suk L, et al. [57] Tesser R, Di Serio M, Guida M, Nastasi M, Santacesaria E. Kinetics of oleic acid
Transesterification of vegetable oil to biodiesel using heterogeneous base esterification with methanol in the presence of triglycerides. Ind Eng Chem Res
catalyst. Catal Today 2004;93–95:315–20. 2005;44:7978–82.
[48] Srivastava A, Prasad R. Triglycerides-based diesel fuels. Reniew Sust Energy [58] Freedman B, Pryde EH, Mounts TL. Variables affecting the yields of fatty esters
Rev 2000;4:111–33. from transesterified vegetable oils. JAOCS 1984;61:1638–43.
[49] Zhang L, Sheng B, Xin Z, Liu Q, Sun S. Kinetics of transesterification of palm oil [59] Singh AK, Fernando SD. Reaction kinetics of soybean oil transesterification
and dimethyl carbonate for biodiesel production at the catalysis of using heterogeneous metal oxide catalyst. Chem Eng Technol
heterogeneous base catalyst. Bioresour Technol 2010;101:8144–50. 2007;30:1716–20.
[50] Taufiq-Yap YH, Lee HV, Hussein MZ, Yunus R. Calcium-based mixed oxide [60] Bond GC. Heterogeneous catalysis: principles and applications, Oxford
catalysts for methanolysis of Jatrophacurcas oil to biodiesel. Biomass chemistry series; 1974 [chapter 3, p. 49].
Bioenergy 2011;35:827–34. [61] Song C, Qi Y, Deng T, Hou X, Qin Z. Kinetic model for the esterification of oleic
[51] Srilatha K, Lingaiah N, Devi BLAP, Prasad RBN, Venkateswar S, Prasad PSS. acid catalyzed by zinc acetate in subcritical methanol. Renew Energy
Esterification of free fatty acids for biodiesel production over heteropoly 2010;35:625–8.
tungstate supported on niobia catalyst. Appl Catal A: Gen 2009;365:28e33.