Journal Pre-proof

Ultrasound-assisted process optimization and tribological characteristics of biodiesel

from palm-sesame oil via response surface methodology and extreme learning
machine - Cuckoo search

M.A. Mujtaba, H.H. Masjuki, M.A. Kalam, H.C. Ong, M. Gul, M. Farooq, Manzoore
Elahi M. Soudagar, Waqar Ahmed, M.H. Harith, M.N.A.M. Yusoff
PII: S0960-1481(20)30863-6
DOI: https://doi.org/10.1016/j.renene.2020.05.158
Reference: RENE 13647

To appear in: Renewable Energy

Received Date: 17 March 2020

Revised Date: 27 May 2020
Accepted Date: 28 May 2020

Please cite this article as: Mujtaba MA, Masjuki HH, Kalam MA, Ong HC, Gul M, Farooq M,
Soudagar MEM, Ahmed W, Harith MH, Yusoff MNAM, Ultrasound-assisted process optimization and
tribological characteristics of biodiesel from palm-sesame oil via response surface methodology and
extreme learning machine - Cuckoo search, Renewable Energy (2020), doi: https://doi.org/10.1016/
j.renene.2020.05.158.

This is a PDF file of an article that has undergone enhancements after acceptance, such as the addition
of a cover page and metadata, and formatting for readability, but it is not yet the definitive version of
record. This version will undergo additional copyediting, typesetting and review before it is published
in its final form, but we are providing this version to give early visibility of the article. Please note that,
during the production process, errors may be discovered which could affect the content, and all legal
disclaimers that apply to the journal pertain.

© 2020 Published by Elsevier Ltd.

Author Contribution

MA Mujtaba: Conceptualization, Experimentation and writing-original draft preparation,

HH Masjuki: Supervision, MA Kalam: Reviewing and Editing, Hwai Chyuan Ong:
Methodology, M Gul: Formal analysis, M. Farooq: Software and validation, Manzoore
Elahi M. Soudagar: Tribology, Waqar Ahmed: Visualization, MH Harith: SEM,
MNAM Yusoff: Resources.
 1 Ultrasound-assisted process optimization and tribological

2 characteristics of biodiesel from Palm-Sesame oil via Response surface

3 methodology and Extreme Learning Machine - Cuckoo search

4 MA Mujtabaa,b*, HH Masjukia,c, MA Kalama*, HC Onga, M Gula,d, M. Farooqb,

5 Manzoore Elahi M. Soudagara, Waqar Ahmede, MH Haritha, MNAM Yusoffa
6
a Center for Energy Science, Department of Mechanical Engineering, University of Malaya, Kuala
7
Lumpur 50603, Malaysia
8
b Department of Mechanical Engineering, University of Engineering and Technology, New Campus
9
Lahore, Pakistan
10
c Department of Mechanical Engineering, Faculty of Engineering, IIUM, 50728 Kuala Lumpur,
11
Malaysia
12
d Department of Mechanical Engineering, University College of Engineering and Technology,
13
Bahauddin Zakariya University, 60000 Multan, Pakistan
14
e Advanced CFD lab, Department of Mechanical Engineering, University of Malaya, Kuala Lumpur
15
50603, Malaysia
16 *
Corresponding author: kalam@um.edu.my, m.mujtaba@uet.edu.pk

17 Abstract

18 The purpose of this study was the improvement of cold flow and lubricity

19 characteristics of biodiesel produced from the palm-sesame oil blend. Extreme learning

20 machine (ELM) and response surface methodology (RSM) techniques were used to

21 model the production process and the input variables (time, catalyst amount, methanol

22 to oil ratio, and duty cycle) were optimized using cuckoo search algorithm. The mean

23 absolute percentage error (MAPE), coefficient of determination (R2), mean square error

24 (MSE), root mean square error (RMSE), and standard error of prediction (SEP) were

25 calculated to evaluate the performance of RSM and ELM. The results showed that

26 ELM model had better performance in prediction than RSM model. The optimum yield

27 of P50S50 biodiesel obtained was 96.6138 % under operating parameters of time

28 (38.96 min), duty cycle (59.52 %), methanol to oil ratio (60 V/V %) and catalyst

29 amount (0.70 wt.%). The cold flow characteristics of P50S50 biodiesel are significantly

30 improved like cloud point (7.89 ), pour point (3.80 ), and cold filter plugging point

1
31 (- 1.77 ) with better oxidation stability 6.89 h. The average coefficient of friction

32 P50S50 biodiesel was lower than palm biodiesel (B100) and B10 commercial diesel by

33 2.29% and 12.37% respectively.

34 Keywords

35 Biodiesel, Response surface methodology, extreme learning machine, Alternative fuel,

36 Palm oil, Sesame oil

37

38

39

40

41

42

43

44

45

46

47

48

49

50

51

52

53

54 Corresponding Authors:
1
55 MA Kalam (kalam@um.edu.my)
1
56 MA Mujtaba (m.mujtaba@uet.edu.pk)

2
57 List of abbreviations

58 ANOVA = Analysis of variance, COF = coefficient of friction, CPO = Crude palm oil,

59 CSO = Crude sesame oil, CV = Calorific value, GC = Gas chromatography, FP = Flash

60 point, ASTM = American society for testing and materials, EN = European, FAME =

61 Fatty acid methyl ester, KOH = Potassium hydroxide, KV = Kinematic viscosity, NOx

62 = Oxides of nitrogen, POME = Palm oil methyl esters, , SOME = Sesame oil methyl

63 esters, USA = united states of America, WSD = wear scar diameter, RSM = Response

64 surface methodology, ELM=Extreme learning machine

65 1 Introduction

66 World energy consumption is gradually rising due to rapid growth in population.

67 Energy consumption is mainly contributed by fossil fuels (diesel, petrol, natural gas,

68 nuclear, and coal). Asia is leading with 41% energy consumption up to 2040 among

69 other countries around the globe [1]. There is a serious concern related to energy

70 demand and environmental pollution. Mostly, the transport sector consumed fossil

71 fuels, which are the main cause of greenhouse gas emissions. By 2030, the transport

72 sector would account for 50% of greenhouse gas emissions [2]. Fossil fuels will be

73 replaced with renewable biofuels to eradicate environmental pollution (carbon

74 emissions) as well as to fulfill the transport sector energy demand. Environmental

75 health is mainly affected by carbon emissions emitted by the transport sector [3].

76 Biodiesel is a significant potential biofuel to substitute crude diesel due to its similar

77 physicochemical properties. Its utilization is safe for engines because biodiesel has a

78 lower ignition delay time, higher combustion efficiency, provide better lubricity

79 without sulfur content, etc. [4],[5].

80 Recently, the main problem with biodiesel commercialization and industrialization is

81 its bad oxidative stability and poor cold flow properties [6]. Biodiesel synthesis from

82 high saturated fatty acid oils limits its usage in cold weather conditions due to its poor

3
 83 cold flow characteristics [7]. Biodiesel handling in terms of transportation or long-term

84 storage is another crucial issue, biodiesel shows poor oxidative stability due to high

85 unsaturation [8]. United States (US) is the highest producer of biodiesel in 2018

86 followed by Brazil, Indonesia, Germany, and Argentina [2]. Soybean and corn are

87 predominantly feedstocks for biodiesel used in the USA, rapeseed as a feedstock in

88 Europe, and Asia’s most common feedstock is palm oil [9]. Indonesia and Malaysia use

89 their crude palm oil as a potential feedstock for producing for biodiesel [10]. Indonesia

90 is the major and leading producer of palm-based biodiesel (approx. 12 billion liters)

91 followed by Malaysia (1.5 billion liters). Currently, B20 is implemented in Indonesia

92 and B10 in Malaysia. Malaysia has the second-highest production of crude palm oil

93 [11]. Palm contains 35 to 55% oil content. The high percentage of saturated fatty acids,

94 which resulted in better oxidative stability as compared to other feedstocks used for

95 biodiesel production. Palm oil-based biodiesel has very poor cold flow features like

96 cold filter plugging (12 ), pour point (15 ), and cloud point (16 ) due to low

97 unsaturated fatty acids [12]. Most of the researchers improved the cold flow properties

98 by adding synthetic antioxidants but now many researchers are blending the oils before

99 transesterification to improve the cold flow characteristics. In a low-temperature

100 climate region, the problems associated with biodiesel are poor cold flow properties,

101 incomplete combustion, starting problem in engine operation, and clogging of filters,

102 etc. [13]. The cold flow properties and oxidative stability of palm-biodiesel can also be

103 enhanced by blending palm oil with other suitable feedstock (like sesame oil) before

104 production. Mujtaba et al. [14] reported that sesame oil has very unique

105 physicochemical properties among other feedstocks. Sesame oil exhibits very

106 promising cold flow properties as well as oxidative stability due to high unsaturation

107 and presence of natural antioxidants. Similar results were reported by other researchers.

108 Pullen et al. [8] investigated the factors affecting the oxidative stability of biodiesel.

4
109 Oxidative stability of twelve biodiesel samples was measured and only sesame oil

110 methyl esters (6.25 h) meet the limit of oxidative stability defined by EN 14214 (min 6

111 h) among all other FAME samples. Sesame oil methyl esters exhibits better oxidative

112 stability among other feedstocks instead of having more than 85 % unsaturated fatty

113 acids due to naturally existing antioxidants and tocopherols like sesamolin, sesamin,

114 sesamol and Vitamin E.

115 These natural antioxidants and tocopherols of sesame oil also help in resisting oxidative

116 rancidity when having exposure to air for a long time [14]. The sesame oil methyl

117 esters has common fatty acids like Palmitic acid C16:0 (7 to 9 %), Stearic acid C18:0 (4

118 to 5 %), Oleic acid C18:1 (40 to 50 %), Linoleic acid C18:2 (35 to 40 %) and Arachidic

119 acid C20:0 (0.4 to 1 %) [15]. Sesame oil has a very low acid value in comparison to

120 palm oil. High acid value oil requires the esterification process to bring acid value

121 below 2 mg KOH/g before the transesterification process, which increased the overall

122 cost of biodiesel due to extra process cost and time. Mixing of oil is an alternate way to

123 enhance the physicochemical properties of feedstock oils before the transesterification

124 process with less expense.

125 Ultrasound-assisted transesterification is a very promising technique for the production

126 of biodiesel due to its cost-effectiveness and energy-efficient. Less energy consumed

127 by ultrasound-assisted transesterification as compared to conventional

128 transesterification [16]. Ultrasound produces cavitation (bubbles formation due to

129 variation in pressure) which enhances the chemical reaction of the transesterification

130 process by providing mechanical and activation energies. Martinez and Gude [17]

131 reported that specific ultrasonic energy (pulse sonication or continuous sonication) had

132 a significant influence on the yield of biodiesel. They reported the maximum yield 98%

133 of biodiesel obtained from waste cooking oil for pulse sonication (5s on and 1s off) and

134 93% of biodiesel yield for continuous sonication. Pulse sonication improves mass

5
135 transfer and enhances energy efficiency. Pulse sonication can be used to convert any

136 feedstock oil to biodiesel [18]. The duty cycle is generally expressed as a percentage,

137 which corresponds to the ratio of the pulse duration to the total cycle time [19]. Pulse

138 duration is “On” time and “Off” time of ultrasound sonicator. Subhedar and Gogate

139 [20] revealed that the ultrasonic duty cycle is the most important parameter to be

140 considered when assessing the energy requirements and process economies to obtain

141 maximum yields. An optimum duty cycle is necessary during biodiesel production to

142 ensure that sonicator operates efficiently and protects the ultrasound probe tip from

143 erosion [17]. At an industrial level, ultrasound is the most promising technique in terms

144 of energy-efficient, less catalyst consumption, time-efficient, and cost-effective.

145 The transesterification process involved many parameters (time, methanol to oil ratio,

146 catalyst concentration, temperature, amplitude, duty cycle, etc.) that directly affect the

147 reaction and biodiesel yield. The response surface methodology (RSM) was utilized to

148 visualize the effect of independent input process parameters on output variables. This

149 tool assists the researchers by designing a less number of experiments to get optimum

150 results [21]. RSM tool is used by various researchers to optimize biodiesel production

151 process parameters by varying input independent process variables with a limited

152 number of experiments to save cost, effort, material, and time [22]. Dwivedi and

153 Sharma [23] used RSM based on Box Behnken design to optimize the process

154 parameters for the production of biodiesel from Pongamia oil. Niju et al. [24] optimized

155 the process parameters (heterogeneous catalyst (Malleus malleus shells) concentration,

156 methanol-to-oil ratio, and reaction time) to maximize the biodiesel yield from waste

157 cooking oil to 93.81% using RSM based on Box-Behnken design. Box-Behnken design

158 for each process variable restricted to three levels to prevent the extreme values of

159 process parameters that reduce the probability of failure. Box-Behnken based design of

6
160 RSM is most preferable, cost effective and efficient than Central Composite based

161 design [25],[26].

162 The neural network of the Extreme Learning Machine (ELM) is feed-forward with a

163 single hidden layer. In extreme learning machines, the hidden layers do not need tuning

164 and the hidden layer parameters can be fixed. The parameters of the hidden layer in

165 extreme learning machines are initialized randomly and the Moore-Penrose generalized

166 inverse is used to determine the output weights analytically with very fast learning

167 speed. Therefore, ELM has superior generalization capability than conventional

168 artificial neural networks [27],[28]. Wong et al. [29] reported that ELM superior than

169 other modeling in accuracy and time. Cuckoo search with the lévy flight is a powerful

170 meta-heuristic optimization that is based on parasitism combining with the Lévy flight

171 algorithm. Silitonga et al. [30] used ELM with the cuckoo search algorithm to predict

172 and optimize the process parameters to achieve maximum yield of 96.19% from Ceiba

173 pentandra oil using microwave-assisted transesterification process.

174 The aim of this study was to optimize the process parameters of ultrasound-assisted

175 transesterification in order to produce palm-sesame biodiesel using RSM tool based on

176 the Box-Behnken design and ELM modelling coupled with the cuckoo search

177 algorithm. Palm and sesame oils were blended in appropriate proportion before

178 transesterification to improve the cold flow characteristics of palm biodiesel. Finally,

179 tribological characteristics of the palm-sesame biodiesel were investigated to determine

180 its lubricity in terms of wear scar diameter and coefficient of friction.

181 2 Materials and methods

182 2.1 Materials and chemicals

183 Crude palm olein (CPO) was taken from Sime Darby Plantation Berhad (Jomalina

184 Refinery), Malaysia. Crude sesame oil (CSO) was sourced from the local market of

185 Lahore, Pakistan. Methanol with purity 99.9% from Friendemann Schmidt, ACS,

7
186 potassium hydroxide (KOH) pellets of AR grade from Friendemann Schmidt, and

187 Whatman filter papers from Filtres Fioroni).

188 2.2 Experimental Methodology

189 2.2.1 Palm-sesame oil mixtures and selection of best blend

190 Five CPCSO mixtures were prepared by blending CPO and CSO at different

191 proportions: 1) 90:10 wt.%, 2) 80:20 wt.%, 3) 70:30 wt.%, 4) 60:40 wt.%, 5) 50:50

192 wt.% and labeled as P90S10, P80S20, P70S30, P60S40, and P50S50, respectively,

193 whereas P represents Palm Olein and S represents Sesame. These blends with different

194 proportions were prepared in a double-jacketed reactor at a stirring speed of 1500 rpm

195 and 60 temperature for 2 hours to get a homogenous mixture.

196 Four physicochemical properties were selected and measured for the respective five

197 blends to choose the best one for optimization and presented in Table 1.

198 Table 1: Physicochemical characteristics of crude palm, sesame oils, and its oil
199 mixtures

Property Crude oils Crude palm and sesame oil blends

CPOa CPaO CSOa CSO P90S10a P80S20a P70S30a P60S40a P50S50a
[31] [32]
Kinematic 38.82 38.1 33.11 31.51 38.43 38.24 38.16 37.93 37.20
viscosity
at 40
(mm2/s)
Density at 914.4 920 922.7 900 915.2 916.0 916.9 917.6 918.4
15
(kg/m3)
Acid 4.16 18.5 1.96 0.42 3.67 3.37 3.07 2.97 2.78
value (mg
KOH /g)
Calorific 38.884 37.5 38.917 38.900 38.086 38.127 38.261 38.558 38.794
value
(MJ/kg)

8
 CPO: crude Palm olein; CPaO: crude Palm oil; CSO: crude Sesame oil; P90S10: Palm-Sesame oil
blend (90:10 wt.%); P80S20: Palm-Sesame oil blend (80:20 wt.%); P70S30: Palm-Sesame oil
blend (70:30 wt.%); P60S40: Palm-Sesame oil blend (60:40 wt.%) and P50S50: Palm-Sesame oil
blend (50:50 wt.%).
a
properties measured in this research

200

201 According to the physicochemical characteristics of different oil blends presented in

202 Table 1. P50S50 and P60S40 oil blends were selected among other oil blends for

203 biodiesel production. Ultrasound technique is used for the production of biodiesel from

204 P50S50 and P60S40 oil blends under the following operating conditions: Methanol

205 40%, catalyst 1 wt.%, time 30 min, amplitude 40% and duty cycle 70%.

206 Physicochemical characteristics and fatty acid compositions of P50S50 and P60S40

207 methyl esters were measured to choose one best oil blend for biodiesel synthesis.

208 Based on the results, the P50S50 oil blend was chosen for this study. It gave us the best

209 substitution in acid value, kinematic viscosity (KV), and calorific value (CV) among

210 other oil blends. The P50S50 oil blend and P50S50 biodiesel have the lowest kinematic

211 viscosities at 40 and acid values. The P50S50 oil blend and P50S50 biodiesel have

212 higher densities at 15 and calorific values. Moreover, the P50S50 oil blend has a

213 lower saturated fatty acid composition among other oils as shown in Table 2. Lower

214 saturated fatty acid composition leads towards double and triple bonds which results in

215 poor oxidative stability and more prone to rancidity of oil. Regardless of having high

216 unsaturation, crude sesame oil exhibits very good oxidative stability because of

217 naturally existing antioxidants and tocopherols [14]. Due to this unique property, the

218 P50S50 oil blend will produce biodiesel with better oxidative stability and lower cold

219 flow characteristics.

220 Table 2: Fatty acid compositions of palm, sesame, and palm-sesame blends biodiesels

Chemical name Carbon structure Palm-sesame blend Palm Sesame

9
 biodiesel biodiesel biodiesel
P60S40a P50S50a POMEa SOMEa
Myristic acid C14:0 0.64 0.48 0.92 -
Palmitic acid C16:0 29.85 24.59 38.96 9.71
Stearic acid C18:0 4.31 4.45 4.05 4.90
Oleic acid C18:1 43.42 42.48 44.95 39.88
Linoleic acid C18:2 20.87 26.92 10.54 43.63
Linolenic acid C18:3 0.43 0.49 0.37 0.56
Arachidic acid C20:0 0.47 0.57 - 0.89
Total saturated fatty acids 35.27 30.09 43.94 15.5
Total unsaturated fatty acids 64.62 69.89 55.86 84.07
P60S40: Palm-Sesame oil blend (60:40 wt.%), P50S50: Palm-Sesame oil blend (50:50 wt.%),
POME: palm oil methyl esters and SOME: sesame oil methyl esters.
a
Fatty acids composition measured by GC equipment (Model: Agilent 7890A, USA) with the
operating condition as described by [22]
221

222 2.3 Experimental setup for transesterification process and tribological studies

223 Ultrasound-assisted transesterification was performed using ultrasound equipment of

224 QSONICA (Q500 Sonicator), which had 20 kHz frequency with a maximum rated

225 power of 500 W. The tapered microtip with 1/2” (inch) diameter probe was used for

226 this study.

227 Tribological characteristics of P50S50 biodiesel were investigated by using an

228 automatic 4-ball (FBT-3: DUCOM brand) tribo friction and wear testing machine.

229 2.4 Transesterification process

230 POME and SOME were prepared using the ultrasound-assisted transesterification

231 process with the following operational parameters: methanol to oil ratio (40%), KOH

232 catalyst amount (1 wt.%), time (30 min), amplitude (40%) and duty cycle (70%).

233 While Palm-Sesame blend biodiesel (P50S50) was also prepared by using the

234 ultrasound-assisted transesterification process in a 500 ml batch reactor to get the

10
235 optimum set of process parameters. For this, the measured amount of homogenized

236 P50S50 oil blend was added to the reactor. The calculated amount of KOH pellets and

237 methanol mixed until pellets dissolved completely in methanol. The blend of methanol

238 and KOH catalyst was charged to the reactor containing the P50S50 oil blend. The

239 ultrasound unit amplitude was fixed to 40% for all batch experiments. The following

240 parameters were varied to analyze their effect on P50S50 biodiesel yield: methanol to

241 oil ratio (30% - 60%), reaction time (10 min – 40 min), KOH catalyst amount (0.5

242 wt.% - 1.5 wt.%) and duty cycle (50% - 70%). When the transesterification reaction

243 completed then reactor mixture was shifted to funnel for the separation of biodiesel and

244 glycerine with impurities. This mixture was left for 8h settling time in the funnel.

245 Lower layer separated from separating funnel by an opening stopcock. The upper layer

246 containing catalyst and methanol content was washed with warm water (40 ) until the

247 clear water layer was found. Hot bubble washed biodiesel was shifted to the round

248 bottom flask (500 ml) for further purification of biodiesel by removing water and

249 methanol content using rotary evaporator equipment at 70 with rotational speed 150

250 rpm for 30 min. Lastly, evaporated biodiesel was filtered by using the Whatman filter

251 paper to get rid of the remaining trace amount of catalyst and stored in a vacuum

252 chamber. The biodiesel yield was obtained from the transesterification process using

253 Equation 1 [30]:

254 Biodiesel yield % = 100 (1)

255 2.5 Design of Experiments

256 2.5.1 RSM design based on Box-Behnken design

257 Response surface methodology (RSM) based on the Box-Behnken design in Design-

258 Expert version 10.0 software was used to design the experiments for the

259 transesterification process of the P50S50 oil blend. Design-Expert version 10.0

260 software was used to design the experiment for optimizing the P50S50 methyl esters

11
261 yield. Four input process parameters (time (X1), catalyst concentration (X2), methanol

262 to oil ratio (X3), and duty cycle (X4) were chosen to study the effect of each parameter

263 (input independent variable) on P50S50 methyl esters yield (output dependent

264 variable). Each input process variable has three coded values included minus 1, center

265 point and plus 1 are given in Table 3 and the total number of experiments is 30. The

266 quadratic polynomial equation was fetched using response surface regression analysis

267 on experimental data in the Box-Behnken experimental design tool. Equation. co-

268 relates the interaction between response variable (P50S50 yield) and input process

269 variables (time, catalyst concentration, methanol to oil ratio, and duty cycle). This

270 equation was used to predict the response (yield) by varying different input process

271 parameters.

'
272 ! = "# + ∑'&() "& & + ∑'&() "&& *
& + ∑',(&-) . . "&, &, (2)
&()

273 In the above Equation 2 Y predicted the yield of P50S50 methyl esters. Xi is the input

274 independent parameter, Co and Ci are the intercept and regression 1st-order coefficient

275 of RSM model, respectively. Cii is the regression quadratic coefficient of the model for

276 the ith factor. Cij is the regression coefficient among ith and jth input parameters, and

277 the number of input parameters represented by k.

278 Table 3: Independent input process variables used for the optimization of P50S50
279 biodiesel yield

Input process variables Units Coded factors Coded process variables levels
-1 Level Center +1 Level
Time min X1 10 25 40
Catalyst concentration Wt.% X2 0.5 1 1.5
Methanol to oil ratio % X3 30 45 60
Duty cycle % X4 50 60 70
280

12
281 2.5.2 ELM modeling

282 ELM is a mathematical model with one single hidden layer. In ELM, the parameters of input

283 weights and hidden bias are initialized randomly, while the Moore-Penrose generalized

284 inverse was used to determine the output weights. The generalization of a single hidden layer

285 feed-forward neural networks in ELM using the function as described in Equation 3 [33]:

286 /0 1 = ∑0&() 2& 3 4& , 6& , 1 , 1 ∈ 8 9 , :& ∈ 8 9 (3)

287 Where wi and bi are the hidden notes learning parameters. ith hidden note and the output node

288 are connected by weight βi. While G(wi,bi,x ) gives output value of the ith hidden note for the

289 input x and calculated by Equation 4.

290 3 4& , 6& , 1 = ;<∑9,(& 4&, 1, + 6& =, 6& ∈ 8 (4)

291 Where 4&, = >4&) , 4&* , … , 4&9 @A is the weight vector which connects the input layer and ith

292 is the hidden node to i with input to j. Also, 6& is the bias of the ith the hidden node :& , 1 =

293 >1) , 1* , … , 19 @A is the inner product of a vector :& in 8 9 .

294 Using Equation 4 can find G (4&, , 6& , for RBF hidden note with activation function

295 ; 1 : 8 → 8 (e.g., Gaussian) in Equation 5 as follows:

A
; D6& E∑9,(&<1,
*
296 3 4& , 6& , 1 = − 4&, = G , 6 ∈ 8 - (5)

297 4& and 6& represent the center and impact factor of ith RBF node, R+ represents a set of all

298 positive real values. A particular case of SLFN that has RBF nodes in its hidden layer forms

299 the RBF network. For N, arbitrary distinct samples (1& , H& ∈ 8 9 1 8 I where n x 1 input

300 vector is represented by xi and m x 1 target vector is represented by ti. If an SLFN with L

301 hidden nodes approximates N samples with zero error, then it implies there exist 2&, 4& and 6&

302 such that.

303 /& 1 = ∑0&() 2& 3 4& , 6& , 1 , J = 1, … K (6)

304 Equation 7 may expressed compactly as:

305 L2 = M (7)

13
306 Where

3 4) , 6) , 1) … 3 40 , 60 , 1)
N, 6O, 1P= = Q
L<4 S
3 4) , 6) , 1R … 3 40 , 60 , 1R RT0
307 (8)

308 U = 4) , … , 40 ; 6O = 6) , … , 60 ; 1P = 1) , … . , 10
With 4

2)A H)A
2= W ⋮ Y :Z[ M = W ⋮ Y
20A H0A
309 (9)
0TI RTI

310 H is the hidden layer output matrix of SLFN with ith column of H being the ith hidden node’s

311 output with respect to inputs x1, …,xn in the Equation 8 and 9.

312 2.5.3 Cuckoo search algorithm

313 Cuckoo search algorithm is an optimization algorithm that is based on brood parasitism

314 of the cuckoo species by laying their eggs in randomly chosen bird's nest, if the host birds

315 discover the egg (using the probability of pa ∈ [0, 1]), they will either remove the egg or just

316 leave the nest and build a new nest.

317 For generating the new solutions, a randomization process with Lévy Flights is performed

318 which can be seen as follows:

= + ] ^ _`ab c
\-) \
319 & & (10)

320 where α > 0 is the step size which is associated with the level of the problem being worked

321 on. While _`ab c states the position equation function of Lévy Flights, which form of an

322 equation is as follows.

323 _`ab~e = H f
, 4ℎ`h` 1 < c ≤ 3 (11)

324 2.5.4 Random sub-sampling cross-validation

325 In this study, 30 datasets were used for Random sub-sampling cross-validation. Thus, 24

326 datasets were selected for training and 6 datasets for testing. The method was run for ten

327 times, the mean average percentage error (MAPE) and standard deviation (SD) based on

328 Equation (12 and 13) were used to analyze the results.

14
 lmno = 100 × q ∑9&(#
rs rt
q
)
9 rt
329 (12)

uv = E
∑z
w{| Tw T̅
y

9 )
330 (13)

331 The number of point represent by n, and the predicted value is represented by yp, while the

332 predicted value and actual presented by ye and yavg , respectively. The mean value of x1;…;

333 xn represent by 1̅ .

334 2.5.5 Statistical analysis

335 The performance of RSM model and ELM was estimated in terms of coefficient of

336 determination (R2), mean square error (MSE), the °°root mean square error (RMSE), and

337 standard error of prediction (SEP) as given in Equations (14) - (17) according to Ishola et al.

338 [34] and Ighose et al. [35].

∑z rs rt y
8 * = 1 − ∑zw{|
rt r}~• y
339 (14)
w{|

y
∑z
340 luo = w{| r€ r•
9
(15)

8luo = E9 ∑9&() b€ − b•
) *
341 (16)

× 100
‚ƒ„…
r}~•
342 SEP = (17)

343 Where, n is the number of points, while ye , yp and yavg represent actual value, predicted

344 value, and an average of the actual values respectively.

345 2.6 Physicochemical properties of P50S50, palm and sesame biodiesels

346 The physicochemical properties of biodiesel samples were measured using different

347 types of equipment such as SVM 3000-automatic (density and viscosity), Automatic

348 NTL Normalab NTE 450 (cloud point, pour point and cold filter plugging point),

349 C2000 basic calorimeter-automatic (calorific value), Metrohm 873 Rancimat (oxidative

350 stability) and Pensky-martens automatic NPM 440 (flashpoint). The physicochemical

351 properties of palm, sesame, P50S50 methyl esters were compared with standard

15
352 specifications of ASTM D-6751 and EN-14214 as shown in Table 8. The measured

353 physicochemical properties of biodiesel and biodiesel blends were compared with

354 petroleum diesel.

355 3 Results and discussion

356 3.1 Optimization of process parameters of ultrasound-assisted transesterification by

357 RSM

358 In this study, the P50S50ME yield was maximized by optimizing the independent

359 process variables, time, catalyst concentration, methanol to oil ratio, and duty cycle.

360 The quadratic regression model was suggested after a regression analysis performed on

361 Box-Behnken experimental design results. The P50S50ME yield results for 30

362 experimental runs, obtained by using the quadratic regression model equation, are

363 shown in Table 4.

364 Table 4: Experimental design for optimization of ultrasound-assisted transesterification

365 process parameters of the P50S50 oil blend
No Time, Catalyst Methanol Duty Experimental RSM ELM
X1 concentration, to oil cycle, P50S50 Predicted Predicted
(min) X2 ((w/w) %) ratio, X3 X4 (%) biodiesel P50S50 P50S50
((v/v) %) yield (%) biodiesel biodiesel
yield (%) yield (%)
1 40 0.5 45 60 92.74 92.92 92.71
2 10 0.5 45 60 88.04 87.86 88.04
3 25 1 60 50 94.47 94.48 94.47
4 10 1 30 60 89.94 89.85 89.57
5 40 1.5 45 60 90.71 90.9 90.71
6 25 1 45 60 93.47 93.67 93.77
7 25 1.5 60 60 89.44 89.47 89.44
8 10 1 45 70 91.14 91.31 91.14
9 25 0.5 45 70 90.17 90.33 90.17
10 25 1 60 70 90.97 90.81 90.97
11 40 1 45 70 94.17 94.06 94.16
12 25 1 45 60 93.94 93.67 93.77
13 10 1 45 50 91.91 92.01 91.91
14 25 1.5 30 60 90.82 90.98 90.82
15 25 0.5 45 50 90.8 90.86 90.80
16 40 1 60 60 93.51 93.61 93.51

16
 17 25 1 45 60 93.61 93.67 93.77
18 40 1 30 60 92.31 92.13 92.31
19 25 0.5 30 60 87.35 87.3 87.35
20 25 1 30 50 89.46 89.63 89.46
21 25 1 45 60 93.51 93.67 93.77
22 10 1 60 60 91.08 91.27 91.08
23 25 1 45 60 93.57 93.67 93.77
24 25 1.5 45 70 91.38 91.33 91.38
25 10 1.5 45 60 91.51 91.33 91.51
26 25 1 30 70 92.77 92.77 92.77
27 25 0.5 60 60 91.87 91.7 91.87
28 25 1 45 60 93.91 93.67 93.77
29 25 1.5 45 50 91.47 91.32 91.47
30 40 1 45 50 94.07 93.89 94.07
366

367 The P50S50 methyl esters yield is calculated by the quadratic model Equation 18 as

368 given below.

369 !†‡#„‡# = 93.67 + 1.16 ) + 0.36 * + 0.72 Œ − 0.13 • − 1.37 ) * + 0.015 ) Œ +

370 0.22 ) • − 1.48 * Œ + 0.13 * • − 1.70 Œ • − 0.53 )

*
− 2.38 *
*
− 1.42 Œ
*
−

371 0.32 •
*
(18)

372 Here, YP50S50 shows the P50S50 biodiesel yield, and X1, X2, X3, and X4 exhibit the

373 time, catalyst amount, methanol to oil ratio, and duty cycle. The relationship between

374 the predicted yield of P50S50 methyl esters and the experimental yield of P50S50ME

375 expressed in Figure 1.

17
376
377 Figure 1: Actual yield vs predicted yield of P50S50 methyl esters
378 A statistical tool ANOVA (analysis of variance) is used to investigate the importance of

379 the RSM (quadratic) model to optimize the P50S50 biodiesel’s yield. ANOVA results

380 described in Table 5.

381 Table 5: ANOVA results for a quadratic regression model

Source Sum of Degree of Mean Square F value P-value

Squares freedom Prob > F
Model 99.66 14 7.12 157.64 < 0.0001 Significant
X1-Time 16.08 1 16.08 356.04 < 0.0001
X2-Catalyst 1.58 1 1.58 35.08 < 0.0001
X3-Methanol 6.29 1 6.29 139.36 < 0.0001
X4-Duty cycle 0.21 1 0.21 4.61 0.0486
X1X2 7.56 1 7.56 167.47 < 0.0001
X1X3 0.164 1 0.164 0.020 0.8896
X1X4 0.19 1 0.19 4.19 0.0586
X2X3 8.70 1 8.70 192.72 < 0.0001
X2X4 0.073 1 0.073 1.61 0.2232
X3X4 11.59 1 11.59 256.75 < 0.0001
X12 1.93 1 1.93 42.72 < 0.0001

18
 X22 38.98 1 38.98 863.16 < 0.0001
X32 13.88 1 13.88 307.45 < 0.0001
X42 0.72 1 0.72 15.96 0.0012
Residual 0.68 15 0.045
Lack of Fit 0.47 10 0.047 1.12 0.4810 Not Significant
Pure Error 0.21 5 0.042
Corrected Total 100.34 29

Std. Dev. 0.21 R2 0.9932

C.V. % 0.23 Adj R2 0.9869
Mean 91.80 Pred R2 0.9701
Adeq Precision 47.789
382

383 The quadratic model F value was 157.64 and the p-value was also (< 0.0001), which

384 specified that the quadratic regression model was “significant”. Less than 0.0500 values

385 of “Prob > F” indicates that model terms are significant and but if “Prob > F” are

386 greater than 0.10 than model terms will be insignificant. Table 5 showed the analysis

387 of variance (ANOVA) of the quadratic regression model that was developed for

388 maximizing the biodiesel yield. In this model X1, X2, X3, X4, X1X2, X2X3, X3X4, X12,

389 X22, X32, X42 model terms are the most significant parameters that affect the yield of

390 biodiesel. The lack of fit “F value” is 1.12 which indicated that lack of fit is

391 insignificant relative to the pure error. The value R2 is 0.9932 which indicates that

392 99.32% deviation in P50S50 methyl esters yield was due to independent input process

393 variables chosen for this model. High R2 value ensures a better estimation between

394 experimental data and the quadratic model.

395 3.2 Combined effect and interaction between process variables on P50S50 biodiesel

396 yield

397 3.2.1 Effect of each parameter on P50S50 biodiesel yield

398 According to ANOVA results presented in Table 5, most critical parameters were time

399 and methanol to oil ratio followed by catalyst concentration and duty cycle.

19
400 By increasing the reaction time, which helps in diffusion between oil, catalyst KOH,

401 and methanol and results in higher reaction rate and methyl esters yield [22],[36].

402 The P50S50 methyl esters yield is slightly increased with the increment of KOH

403 concentration from 0.5 to 0.8 (wt.%) and then further increment in concentration from

404 0.8 to 1.5 (wt.%) results in the reduction of biodiesel yield. The excess amount of

405 catalyst KOH increased the viscosity of biodiesel due to emulsification and leads to the

406 formation of gels [37].

407 The duty cycle has very less impact on biodiesel yield. The biodiesel yield was

408 gradually decreased with increment in duty cycle % due to a reduction in reaction rate

409 caused by improper mixing of the reaction mixture and lower cavitation phenomena. A

410 similar trend was reported by different researchers [38],[39].3-D response surface

411 graphs were also plotted to analyze the combined effect of process variables on P50S50

412 methyl esters yield.

413 3.2.2 Interaction between two process variables on P50S50 biodiesel yield

414 3-D surface graphs are plotted and presented in Figure 2 to investigate the combined

415 interaction effect of process variables on P50S50 methyl esters yield.

416 The combined effect of reaction time (10 to 40 min) and catalyst amount (0.5 to 1.5

417 (wt. %)) on P50S50 methyl esters yield is visualized from Figure 2 (a). The maximum

418 94.54% biodiesel yield achieved at time 40 min and catalyst amount 0.874 (wt.%). In

419 the case of catalyst concentration, biodiesel yield increased up to 0.87 (wt.%)

420 concentration but further increase in catalyst amount causes a decrease in biodiesel

421 yield from 94.54% to 90.84%. A higher dosage of KOH concentration reversed the

422 transesterification process which decreased the conversion of oil to biodiesel and

423 assisted in the formation of soap. The purification process of biodiesel becomes very

424 complicated due to the presence of soap with glycerol and biodiesel. The lowest yield

425 (89%) at the highest concentration of catalyst (KOH) and the lowest reaction time

20
426 indicate that the reaction time is insufficient to attain optimum biodiesel yield. It can be

427 observed that adequate reaction time is required to obtain a homogeneous mixture of

428 catalyst (KOH), methanol, and palm-sesame oil to form products and achieve optimum

429 yield [34].

430 The combined effect of reaction time (10 to 40 min) and methanol to oil ratio (30 to 60

431 (V/V) %) on biodiesel yield presented in Figure 2 (b). The maximum output yield of

432 P50S50 biodiesel (94.65%) obtained at a reaction time 40 (min) with methanol to oil

433 ratio of 53.108 (V/V) %. The biodiesel’s yield increases with increment in reaction

434 time up to 40 min because adequate time is supplied for completing the

435 transesterification reaction. In the case of methanol to oil molar ratio, P50S50

436 biodiesel’s yield increased up to 53 (V/V) % then further increase in methanol resulted

437 in a slight reduction of biodiesel yield from 94.65 % to 94.22%. Excess dosage of

438 methanol will dilute the reaction mixture and reduce catalyst effectiveness during the

439 transesterification process [35]. If the methanol-to-oil molar ratio exceeds the optimum

440 limit, biodiesel separation becomes difficult due to emulsion formation and the overall

441 cost will be increased [22],[40],[41].

442 The combined effect of reaction time (10 to 40 min) and duty cycle (50 to 70 %) on

443 P50S50 biodiesel yield is shown in Figure 2 (c). Biodiesel’s yield is increasing

444 gradually with an increase in reaction time. The duty cycle is a significant parameter to

445 optimize the energy consumption during the ultrasound-assisted transesterification

446 process [17]. The duty cycle is varied from 50 to 70 %, biodiesel yield is increased up

447 to 57.44 % then further increase in duty cycle resulted in the reduction of biodiesel

448 yield from 94.52 to 94.01%. The higher the duty cycle, the longer the cavitation effect

449 arises in fixed reaction time. It was observed from the 3D graph, increasing the duty

450 cycle increases reaction progress due to increased cavitation activity until optimum

451 value [42]. Further increase in duty cycle beyond optimum value resulted in lower

21
452 biodiesel yield. A similar trend was reported by Maddikeri et al. [43] and Joshi et al.

453 [44] with optimum values of duty cycles 60% and 70% respectively.

454 The combined effect of catalyst amount and methanol to oil ratio is presented in Figure

455 2 (d). Maximum biodiesel yield of 94.75% P50S50 was obtained at catalyst

456 concentration 0.80 (wt. %) and methanol to oil ratio 53.46 (V/V) %. Biodiesel yield

457 enhanced with increment in methanol to oil ratio up to 53.46 (V/V) % beyond this limit

458 yield will start to decrease. The high percentage of methanol to oil ratio leads to the

459 emulsification of biodiesel and glycerol. Less amount of catalyst and methanol leads to

460 lower biodiesel yield due to the possibility of reversible reaction and incomplete

461 conversion of feedstock oil to biodiesel. Beyond the optimal concentration of catalyst,

462 the biodiesel yield decreased significantly due to saturation of catalyst particles in the

463 reaction mixture, formation of soap (saponification process), and increased reaction

464 mixture viscosity [45]. A similar trend was reported by Ong et al. [46] and Dharma et

465 al. [47].

466 The combined effect of catalyst concentration and duty cycle shown in Figure 2 (e).

467 The maximum 94.55% biodiesel yield was attained at a concentration of KOH 0.86

468 (wt.%) and duty cycle 57.15 %. There is a slight decrement in yield by increasing the

469 duty cycle % beyond the optimum limits. Higher duty cycle produces too many

470 consecutive cavitation activities which lead to cushioning effect resulting in less

471 cavitation activity which decreases the yield of biodiesel with even higher catalyst

472 dose. A similar result was reported by other researchers [42],[43],[44]. Biodiesel yield

473 is decreased by up to 90.78% with a further increase in the concentration of catalyst

474 from 0.86 to 1.5%. Excess dosage of the KOH catalyst helped in gel-forming due to

475 emulsification resulting in a reduced amount of methyl ester content in the final product

476 [42]. The duty cycle effect is very less in the reduction of yield as compared to catalyst

477 concentration. The duty cycle is a more important parameter in the optimization of

22
478 energy consumption of ultrasound-assisted transesterification. Catalyst is the most

479 important variable, which should be optimized to obtain the maximum yield.

480 Combined effect of methanol to oil molar ratio and duty cycle have been plotted in

481 Figure 2 (f). There is an inverse relationship between duty cycle and methanol to oil

482 ratio. Biodiesel yield enhanced by increasing methanol to oil ratio up to 59% but

483 further increase resulted in a small decrease of biodiesel’s yield. Biodiesel yield

484 decreased from 95.27 to 92.23% gradually with increment in duty cycle from 50 to

485 70%. Biodiesel yield decreased, due to poor cavitation activity, beyond the optimal

486 value of the duty cycle. The duty cycle is an important parameter that should be

487 optimized for maximum biodiesel yield and energy savings. Maddikeri et al. [39]

488 obtained a maximum yield of Karanja methyl esters with a 60% duty cycle. The

489 optimum value of the duty cycle varied according to feedstock, catalyst, and alcohol

490 used for obtaining the maximum yield.

491

492

493

494

23
495

496 Figure 2: 3-D graphs from RSM representing interaction effects between process
497 variables: a) time and catalyst concentration, b) time and methanol, c) time and duty
498 cycle, d) catalyst and methanol, e) catalyst and duty cycle and f) methanol and duty
499 cycle on P50S50 biodiesel yield

24
500 3.3 ELM modelling

501 Figure 3 shows the values of the correlation coefficient (R) determined by the ELM.

502 From Figure 3, it is shown that the R values for the training (R=0.99972), testing

503 (R=0.99357), and overall datasets (R=0.99815) are close to 1. These values indicated

504 that a good correlation developed between the experimental values and the predicted

505 values generated by ELM.

506
507 Figure 3: The value of coefficient determination (R) for ELM
508 3.4 Random sub-sampling cross-validation for the ELM model

509 We have total 30 datasets; 24 dataset were imported for training and 6 datasets for

510 testing. The procedure was used in this study is the data running ten times, as shown in

511 Table 6. The values of MAPE and SD are < 0.5 ( i.e 0.101315 and 0.02179

25
512 respectively), these results proof that ELM model will predict yield of P50S50ME more

513 accurately.

514

515 Table 6: Results of the random sub-sampling cross-validation

Repetition MAPE
R-1 0.065651
R- 2 0.119798
R- 3 0.109659
R- 4 0.129578
R- 5 0.059996
R- 6 0.095333
R- 7 0.106328
R- 8 0.094248
R- 9 0.116168
R- 10 0.116391
Average 0.101315
SD 0.02179
516

517 3.5 Statistical analysis

518 The statistical analysis of the performances between RSM and ELM is mentioned in

519 Table 7. From Table 7, it is shown that both ELM and RSM models have R2 values that are

520 close to 1, indicating good fit of the models. The MSE, RMSE and MAPE

521 value was found greater for RSM (0.022657, 0.1505, 0.146952) compared to ELM

522 (0.013582, 0.1168, 0.059996), respectively. The SEP for RSM and ELM was found to be

523 0.1639 and 0.1272, respectively. Based on the statistical analysis results, it is concluded that

524 ELM has better performance in terms of prediction accuracy as compared to RSM.

525 Table 7: Statistical analysis

Statistical RSM ELM

analysis
R2 0.9935 0.9963
MSE 0.022657 0.013582
RMSE 0.1505 0.1168
MAPE 0.146952 0.059996
SEP 0.1639 0.1272
526

26
527 3.6 Optimization of transesterification process parameters using Cuckoo search

528 optimization algorithm

529 The optimum of P50S50 biodiesel yield determined from the ELM coupling with

530 cuckoo search algorithm via levy flight model is 96.6138%, under the condition as

531 follows: (1) time: 38.96 (min), (2) KOH catalyst concentration: 0.70 %(w/w), (3)

532 methanol to oil ratio: 60 %, and (4) duty cycle: 59.52 %. The experiments were

533 conducted in triplicate using optimum ultrasound assisted transesterification input

534 process variables to validate the optimum predicted P50S50 yield. ELM coupled with

535 the cuckoo search algorithm predicted a higher yield of 96.138%. The average

536 experimental yield of P50S50 biodiesel is 95.89% with a standard deviation of 0.0026,

537 which depicts a good agreement between experimental yield and predicted yield by

538 ELM coupled with the cuckoo search algorithm.

539 3.7 Physicochemical characteristics of P50S50, POME and SOME methyl esters

540 The physicochemical properties of palm methyl esters, sesame methyl esters, and palm-

541 sesame blend (50:50) methyl esters were estimated in accordance with biodiesel

542 standard methods ASTM D6751, EN 14214, and MS 2008:2008 presented in Table 8.

543 In this study, sesame oil was blended with palm oil to enhance its cold flow properties

544 and lubricity characteristics. The high percentage of unsaturation causes improvement

545 in cold flow properties like sesame oil methyl esters has 84% (unsaturated fatty acids)

546 and palm oil biodiesel has 55% (unsaturated fatty acids). P50S50 biodiesel has very

547 good cold flow characteristics like pour point (3.80 ), cloud point (7.94 ), and cold

548 filter plugging point (-1.77 ). P50S50 biodiesel is more viable as biodiesel in

549 comparison to palm-biodiesel due to its better cold flow properties.

550

551 Table 8: Physicochemical properties of POME, SOME and P50S50 biodiesels

552
553

27
 Property Test Biodiesel
Unit Method Diesel standard Biodiesel and biodiesel blends
specifications
ASTM EN POME SOME
[46] POME SOME P50S50
D6751 14214 [12] [32]
Kinematic ASTM
(mm2/s) 2.87 1.9-6.0 3.5-5.0 4.48 4.5 4.37 4.47 4.42
viscosity at 40 D445
Density at 15 ASTM 860-
(kg/m3) 839.4 870-890 876.1 864.4 884.7 867 880
D4052 900
Acid value (mg ASTM
0.15 < 0.50 0.50 0.40 0.24 0.35 0.12 0.37
KOH /g) D664
Calorific value ASTM
(MJ/kg) 45.67 - 35 39.56 38.5 39.71 40.1 41.24
D240
Flash point ASTM
78.50 130 > 101 >120 135 >150 180 >150
D93
Cloud point ASTM
2.0 - - 15.50 16 0.12 -5 7.94
D2500
Pour point ASTM
2.0 - - 12 15 -4.69 -9 3.80
D97
Cold filter ASTM
0.0 - - 2.12 12 -5.73 - -1.77
plugging point D6371
Oxidative EN
h 13.20 3h 6h 13.40 10.3 6.32 >6 6.89
stability 14112
Copper strip ASTM
- 1a 3 max - 1a 1a 1a - 1a
corrosion D130
Cetane number ASTM
- 48.50 47 51 59.15 54.6 47.65 56.32 53.37
D4737
FAME content a wt.% - - - 96.5 > 98 - > 98 > 98 > 98
Linolenic acid -
methyl ester wt.% - - 1-15 0.37 - 0.56 - 0.49
content a
POME: Palm olein methyl esters, SOME: Sesame oil methyl esters, P50S50: Palm-Sesame blend methyl
esters (50:50 wt.%)
a
FAME content and Linolenic acid content of methyl esters were calculated using gas chromatography

554
555 According to ASTM and EN biodiesel standards, the oxidative stability of biodiesel

556 should be 3 h and 6 h respectively. Sesame oil is a viable feedstock to blend with palm

557 oil to get better cold flow characteristics as well as to maintain its oxidative stability

558 within a range of biodiesel standards. P50S50 biodiesel had 6.89 h oxidative stability,

559 which fulfilled both biodiesel standard limits. P50S50 biodiesel had 30 % saturated

560 fatty acids, which is better to achieve high oxidative stability.

561 Other physicochemical properties of P50S50 methyl esters are also better than POME

562 and SOME like lower acid value, lower kinematic viscosity, higher calorific value, and

563 better cold flow properties. FAME content of P50S50 methyl esters is more than 98

28
564 (wt.%) which is above the value specified by EN 14214 (96.5 wt.%). Physicochemical

565 properties of P50S50 methyl esters meet the minimum criteria set by biodiesel

566 standards ASTM D6751 and EN14214 so it is concluded that P50S50 biodiesel is a

567 potential fuel in the replacement of crude diesel.

568 3.8 Lubricity and wear characteristics

569 Friction and wear behavior of B10 (Diesel), B100 (Palm biodiesel) and B100 (P50S50

570 biodiesel) were estimated according to ASTM D4172 standard. The experimental

571 conditions were speed 1200 rpm, temperature 75 , load 40 kg, and duration of

572 experiment 600 seconds. Figure 4 (a) represents the coefficient of friction of tested

573 fuels with respect to time variation at 1200 rpm. In the early stage of the experiment,

574 instability of the coefficient of friction (COF) can be seen up to 90 seconds that is

575 named as run-in period, after that a steady-state condition is achieved, as shown in

576 Figure 4 (a), which is due to the presence of ester content [48]. During the start of the

577 experiment, the COF was very high due to the absence of a lubricating film between

578 metallic contact and later on steady-state friction trend starts to stabilize after the

579 formation of the lubricating film between mating surface [49]. The run-in period for

580 pure P50S50 biodiesel is lowest due to the absorption of ester molecules existing in the

581 biodiesel sample. P50S50 biodiesel has high unsaturation (presence of double bonds)

582 compared to palm biodiesel. The presence of ester content in the fuel sample leads to

583 the conversion of run in the period to steady-state condition quickly [50]. B10 (Diesel)

584 has a higher COF among tested fuels due to the presence of high Sulphur content. Fuel

585 lubricity is a very critical parameter related to the durability of diesel engine

586 components. Most diesel engine components are self-lubricated with diesel fuel, such

587 as (fuel injectors, fuel pumps, etc.) [51].

588 Wear scar diameter (WSD) of tested fuels that are measured by SEM micrographs of

589 metal balls. Figure 4 (b) exhibits the WSD and average COF of B10 (Diesel), B100

29
590 (Palm biodiesel) and B100 (P50S50 biodiesel). B10 (diesel) had high value of WSD

591 0.948 mm than B100 (Palm biodiesel) 0.869 mm WSD and B100 (P50S50 biodiesel)

592 0.746 mm. Biodiesel P50S50 exhibited less WSD value than other test fuels due to high

593 unsaturated fatty acids that formed the lubricating film between metal contact surfaces.

594 The average COF of P50S50 biodiesel 0.085 is lowest followed by B100 (Palm

595 biodiesel) 0.087 and B10 (diesel) 0.097. The higher percentage of oxygen content, high

596 percentage of unsaturation in fatty acids, and long carbon chain of fatty acids in

597 biodiesel enhanced the lubricity of fuel and reduced the wear and friction between

598 metallic contact surfaces.

599

600
601

602 Figure 4: a) Coefficient of friction of B10 (Diesel), B100 (Palm biodiesel) and B100
603 (P50S50 biodiesel) at 1200 rpm, temperature 75 , time for test 600 seconds and load
604 40 kg. b) wear scar diameter (WSD) and the average coefficient of friction (COF) of
605 B10 (Diesel), B100 (Palm biodiesel) and B100 (P50S50 biodiesel) at 1200 rpm. c),d)

30
606 and e) SEM micrographs (worn surfaces) of stationary steel balls for B10 (Diesel),
607 B100 (Palm biodiesel) and B100 (P50S50 biodiesel) at 1200 rpm.
608
609 SEM micrographs of worn surface of stationary steel balls under 40 kg load of B10

610 (Diesel), B100 (Palm biodiesel) and B100 (P50S50 biodiesel) tested fuels presented in

611 Figure 4 (c), (d) and (e) respectively. Abrasive wear can be seen in the form of

612 scratches and extrusion of small metal can be visualized in the form of adhesive wear.

613 Adhesive wear can be seen in the case of B10 (diesel) due to the detachment of

614 particles from the worn surface. Deep grooves and scratches can be seen in the case of

615 B10 (diesel) due to a lack of lubricating film between rubbing surfaces. The removal of

616 material in the case of B10 (diesel) is more than biodiesel tested fuel specimens. Pure

617 biodiesel tested fuels showed less wear compared to B10 (diesel) due to a high

618 concentration of oleic and linoleic acids that acts as a lubricity enhancer [52].

619 4 Conclusion

620 In this study, ultrasound-assisted transesterification was used to produce the biodiesel

621 from palm-sesame blend oil. P50S50 biodiesel yield was maximized with optimization

622 of transesterification process variables using RSM based on Box-Behnken design and

623 ELM coupling with a cuckoo search optimization algorithm. From the results, it could

624 be concluded that ELM performed better compared to RSM in predicting the P50S50

625 yield. Maximum optimal P50S50 yield of 95.89% was obtained using ELM coupling

626 with cuckoo search optimization algorithm using reaction time (38.96 min), duty cycle

627 (59.52%), methanol-to-oil ratio (60%), and catalyst amount (0.70 wt.%). The P50S50

628 biodiesel had better cloud point (7.89 °C), pour point (3.80 °C) and cold filter plugging

629 point (1.77 °C) with better oxidative stability 6.89 h. Physicochemical properties of the

630 P50S50 methyl esters complied with the ranges specified for biodiesel (ASTM D6751

631 and EN 14214). The average COF of P50S50 biodiesel was 12.37 % lower than B10

632 (diesel) and WSD of P50S50 was also 21.29% lower than B10 (diesel). The wear and

31
633 friction characteristics of P50S50 biodiesel were also reduced in comparison to other

634 tested fuels because of high unsaturated fatty acids. Moreover, physicochemical

635 properties and lubricity of P50S50 biodiesel are improved in comparison to B100 (Palm

636 biodiesel) and B10 (Commercial diesel).

637

638 Acknowledgments

639 The authors would like to thank the Faculty of Engineering at the University of

640 Malaya, Malaysia for their support through the research grant no FP142-2019A under

641 the Fundamental Research Grant Scheme (FRGS) from Ministry of Higher Education,

642 Malaysia; and Higher Education Commission, Islamabad, Pakistan for financial

643 support [Grant No. 5-1/HRD/UESTPI (Batch-VI)/4954/2018/HEC].

644 References

645 [1] EIA, Annual Energy Outlook 2019, 2019.

646 [2] REN21, Renewables 2018 Global Status Report, (Paris: REN21 Secretariat), 2018.

647 [3] V. Uusitalo, M. Leino, H. Kasurinen, L. Linnanen, Transportation biofuel efficiencies

648 from cultivated feedstock in the boreal climate zone: Case Finland, Biomass and Bioenergy
649 99 (2017) 79-89.

650 [4] M.A. Wakil, M.A. Kalam, H.H. Masjuki, A.E. Atabani, I.M. Rizwanul Fattah, Influence
651 of biodiesel blending on physicochemical properties and importance of mathematical model
652 for predicting the properties of biodiesel blend, Energy Conversion and Management 94
653 (2015) 51-67.

654 [5] H.C. Ong, H.H. Masjuki, T.M.I. Mahlia, A.S. Silitonga, W.T. Chong, T. Yusaf, Engine
655 performance and emissions using Jatropha curcas, Ceiba pentandra and Calophyllum
656 inophyllum biodiesel in a CI diesel engine, Energy 69 (2014) 427-445.

657 [6] T.M.I. Mahlia, Z.A.H.S. Syazmi, M. Mofijur, A.E.P. Abas, M.R. Bilad, H.C. Ong, A.S.
658 Silitonga, Patent landscape review on biodiesel production: Technology updates, Renewable
659 and Sustainable Energy Reviews 118 (2020) 109526.

660 [7] J.F. Sierra-Cantor, C.A. Guerrero-Fajardo, Methods for improving the cold flow
661 properties of biodiesel with high saturated fatty acids content: a review, Renewable and
662 Sustainable Energy Reviews 72 (2017) 774-790.

32
663 [8] J. Pullen, K. Saeed, Experimental study of the factors affecting the oxidation stability of
664 biodiesel FAME fuels, Fuel Processing Technology 125 (2014) 223-235.

665 [9] A.S. Silitonga, H.H. Masjuki, T.M.I. Mahlia, H.C. Ong, W.T. Chong, M.H. Boosroh,
666 Overview properties of biodiesel diesel blends from edible and non-edible feedstock,
667 Renewable and Sustainable Energy Reviews 22 (2013) 346-360.

668 [10] M. Gul, H.H. Masjuki, M.A. Kalam, N.W.M. Zulkifli, M.A. Mujtaba, A Review: Role
669 of Fatty Acids Composition in Characterizing Potential Feedstock for Sustainable Green
670 Lubricants by Advance Transesterification Process and its Global as Well as Pakistani
671 Prospective, BioEnergy Research 13(1) (2020) 1-22.

672 [11] C.H. Neoh, C.Y. Lam, C.K. Lim, A. Yahya, Z. Ibrahim, Decolorization of palm oil mill
673 effluent using growing cultures of Curvularia clavata, Environmental Science and Pollution
674 Research 21(6) (2014) 4397-4408.

675 [12] A.E. Atabani, A.S. Silitonga, I.A. Badruddin, T. Mahlia, H. Masjuki, S. Mekhilef, A
676 comprehensive review on biodiesel as an alternative energy resource and its characteristics,
677 Renewable and sustainable energy reviews 16(4) (2012) 2070-2093.

678 [13] G. Dwivedi, M.P. Sharma, Impact of cold flow properties of biodiesel on engine
679 performance, Renewable and Sustainable Energy Reviews 31 (2014) 650-656.

680 [14] M.A. Mujtaba, H. Muk Cho, H.H. Masjuki, M.A. Kalam, H.C. Ong, M. Gul, M.H.
681 Harith, M.N.A.M. Yusoff, Critical review on sesame seed oil and its methyl ester on cold
682 flow and oxidation stability, Energy Reports 6 (2020) 40-54.

683 [15] A. Karmakar, S. Karmakar, S. Mukherjee, Properties of various plants and animals
684 feedstocks for biodiesel production, Bioresour Technol 101(19) (2010) 7201-10.

685 [16] S.X. Tan, S. Lim, H.C. Ong, Y.L. Pang, State of the art review on development of
686 ultrasound-assisted catalytic transesterification process for biodiesel production, Fuel 235
687 (2019) 886-907.

688 [17] E. Martinez-Guerra, V.G. Gude, Determining optimum pulse mode for ultrasound
689 enhanced biodiesel production, Journal of Industrial and Engineering Chemistry 35 (2016)
690 14-19.

691 [18] E. Martinez-Guerra, V.G. Gude, Transesterification of used vegetable oil catalyzed by
692 barium oxide under simultaneous microwave and ultrasound irradiations, Energy Conversion
693 and Management 88 (2014) 633-640.

694 [19] Z. Pan, W. Qu, H. Ma, G.G. Atungulu, T.H. McHugh, Continuous and pulsed
695 ultrasound-assisted extractions of antioxidants from pomegranate peel, Ultrasonics
696 Sonochemistry 18(5) (2011) 1249-1257.

697 [20] P.B. Subhedar, P.R. Gogate, Ultrasound assisted intensification of biodiesel production
698 using enzymatic interesterification, Ultrasonics Sonochemistry 29 (2016) 67-75.

699 [21] C. Muthukumaran, R. Praniesh, P. Navamani, R. Swathi, G. Sharmila, N.M. Kumar,

700 Process optimization and kinetic modeling of biodiesel production using non-edible Madhuca
701 indica oil, Fuel 195 (2017) 217-225.

33
702 [22] J. Milano, H.C. Ong, H.H. Masjuki, A.S. Silitonga, W.-H. Chen, F. Kusumo, S. Dharma,
703 A.H. Sebayang, Optimization of biodiesel production by microwave irradiation-assisted
704 transesterification for waste cooking oil-Calophyllum inophyllum oil via response surface
705 methodology, Energy Conversion and Management 158 (2018) 400-415.

706 [23] G. Dwivedi, M.P. Sharma, Application of Box–Behnken design in optimization of

707 biodiesel yield from Pongamia oil and its stability analysis, Fuel 145 (2015) 256-262.

708 [24] S. Niju, R. Rabia, K. Sumithra Devi, M. Naveen Kumar, M. Balajii, Modified Malleus
709 malleus Shells for Biodiesel Production from Waste Cooking Oil: An Optimization Study
710 Using Box–Behnken Design, Waste and Biomass Valorization 11(3) (2020) 793-806.

711 [25] S.C. Ferreira, R. Bruns, H. Ferreira, G. Matos, J. David, G. Brandao, E.P. da Silva, L.
712 Portugal, P. Dos Reis, A. Souza, Box-Behnken design: an alternative for the optimization of
713 analytical methods, Analytica chimica acta 597(2) (2007) 179-186.

714 [26] P.U. Okoye, S. Wang, W.A. Khanday, S. Li, T. Tang, L. Zhang, Box-Behnken
715 optimization of glycerol transesterification reaction to glycerol carbonate over calcined oil
716 palm fuel ash derived catalyst, Renewable Energy 146 (2020) 2676-2687.

717 [27] G. Huang, G.-B. Huang, S. Song, K. You, Trends in extreme learning machines: A
718 review, Neural Networks 61 (2015) 32-48.

719 [28] D. Wang, P. Wang, Y. Ji, An oscillation bound of the generalization performance of
720 extreme learning machine and corresponding analysis, Neurocomputing 151 (2015) 883-890.

721 [29] K.I. Wong, P.K. Wong, C.S. Cheung, C.M. Vong, Modeling and optimization of
722 biodiesel engine performance using advanced machine learning methods, Energy 55 (2013)
723 519-528.

724 [30] A.S. Silitonga, A.H. Shamsuddin, T.M.I. Mahlia, J. Milano, F. Kusumo, J. Siswantoro,
725 S. Dharma, A.H. Sebayang, H.H. Masjuki, H.C. Ong, Biodiesel synthesis from Ceiba
726 pentandra oil by microwave irradiation-assisted transesterification: ELM modeling and
727 optimization, Renewable Energy 146 (2020) 1278-1291.

728 [31] S. Yusup, M. Khan, Basic properties of crude rubber seed oil and crude palm oil blend
729 as a potential feedstock for biodiesel production with enhanced cold flow characteristics,
730 Biomass and Bioenergy 34(10) (2010) 1523-1526.

731 [32] A. Sarve, S.S. Sonawane, M.N. Varma, Ultrasound assisted biodiesel production from
732 sesame (Sesamum indicum L.) oil using barium hydroxide as a heterogeneous catalyst:
733 Comparative assessment of prediction abilities between response surface methodology
734 (RSM) and artificial neural network (ANN), Ultrason Sonochem 26 (2015) 218-228.

735 [33] G.-B. Huang, D.H. Wang, Y. Lan, Extreme learning machines: a survey, International
736 journal of machine learning and cybernetics 2(2) (2011) 107-122.

737 [34] N.B. Ishola, A.A. Okeleye, A.S. Osunleke, E.J.N.C. Betiku, Applications, Process
738 modeling and optimization of sorrel biodiesel synthesis using barium hydroxide as a base
739 heterogeneous catalyst: appraisal of response surface methodology, neural network and
740 neuro-fuzzy system, 31(9) (2019) 4929-4943.

34
741 [35] B.O. Ighose, I.A. Adeleke, M. Damos, H.A. Junaid, K.E. Okpalaeke, E.J.E.C. Betiku,
742 Management, Optimization of biodiesel production from Thevetia peruviana seed oil by
743 adaptive neuro-fuzzy inference system coupled with genetic algorithm and response surface
744 methodology, 132 (2017) 231-240.

745 [36] F. Gunawan, A. Kurniawan, I. Gunawan, Y.-H. Ju, A. Ayucitra, F.E. Soetaredjo, S.
746 Ismadji, Synthesis of biodiesel from vegetable oils wastewater sludge by in-situ subcritical
747 methanol transesterification: Process evaluation and optimization, biomass and bioenergy 69
748 (2014) 28-38.

749 [37] P.D. Patil, S. Deng, Optimization of biodiesel production from edible and non-edible
750 vegetable oils, Fuel 88(7) (2009) 1302-1306.

751 [38] S. Joshi, P. Gogate, S. Kumar, Intensification of ultrasound assisted esterification of

752 karanja oil for production of biodiesel with optimization using response surface methodology,
753 Chem. Eng. Process. Process Intensif. 124 (2018) 186-198.

754 [39] G.L. Maddikeri, A.B. Pandit, P.R. Gogate, Ultrasound assisted interesterification of
755 waste cooking oil and methyl acetate for biodiesel and triacetin production, Fuel processing
756 technology 116 (2013) 241-249.

757 [40] M.J. Goff, N.S. Bauer, S. Lopes, W.R. Sutterlin, G.J. Suppes, Acid catalyzed
758 alcoholysis of soybean oil, Journal of the American Oil Chemists' Society 81(4) (2004) 415-
759 420.

760 [41] H. Hamze, M. Akia, F. Yazdani, Optimization of biodiesel production from the waste
761 cooking oil using response surface methodology, Process Safety and Environmental
762 Protection 94 (2015) 1-10.

763 [42] S.S. Kashyap, P.R. Gogate, S.M. Joshi, Ultrasound assisted synthesis of biodiesel from
764 karanja oil by interesterification: Intensification studies and optimization using RSM,
765 Ultrasonics Sonochemistry 50 (2019) 36-45.

766 [43] G.L. Maddikeri, A.B. Pandit, P.R.J.F.P.T. Gogate, Ultrasound assisted interesterification
767 of waste cooking oil and methyl acetate for biodiesel and triacetin production, 116 (2013)
768 241-249.

769 [44] S. Joshi, P. Gogate, S.S.J.C.E.P.P.I. Kumar, Intensification of ultrasound assisted

770 esterification of karanja oil for production of biodiesel with optimization using response
771 surface methodology, 124 (2018) 186-198.

772 [45] M.G. Nayak, A.P. Vyas, Optimization of microwave-assisted biodiesel production from
773 Papaya oil using response surface methodology, Renewable Energy 138 (2019) 18-28.

774 [46] H.C. Ong, J. Milano, A.S. Silitonga, M.H. Hassan, A.H. Shamsuddin, C.-T. Wang, T.M.
775 Indra Mahlia, J. Siswantoro, F. Kusumo, J. Sutrisno, Biodiesel production from Calophyllum
776 inophyllum-Ceiba pentandra oil mixture: Optimization and characterization, Journal of
777 Cleaner Production 219 (2019) 183-198.

778 [47] S. Dharma, H.H. Masjuki, H.C. Ong, A.H. Sebayang, A.S. Silitonga, F. Kusumo, T.M.I.
779 Mahlia, Optimization of biodiesel production process for mixed Jatropha curcas–Ceiba
780 pentandra biodiesel using response surface methodology, Energy Conversion and
781 Management 115 (2016) 178-190.

35
782 [48] T. Konishi, E. Klaus, J. Duda, Wear characteristics of aluminum-silicon alloy under
783 lubricated sliding conditions, Tribology Transactions 39(4) (1996) 811-818.

784 [49] M. Habibullah, H.H. Masjuki, M.A. Kalam, N.W.M. Zulkifli, B.M. Masum, A. Arslan,
785 M. Gulzar, Friction and wear characteristics of Calophyllum inophyllum biodiesel, Industrial
786 Crops and Products 76 (2015) 188-197.

787 [50] T. Konishi, E.E. Klaus, J.L. Duda, Wear Characteristics of Aluminum-Silicon Alloy
788 under Lubricated Sliding Conditions, Tribology Transactions 39(4) (1996) 811-818.

789 [51] R.H. Barbour, D.J. Rickeard, N.G. Elliott, Understanding Diesel Lubricity, SAE
790 International, 2000.

791 [52] D.P. Geller, J.W.J.F. Goodrum, Effects of specific fatty acid methyl esters on diesel fuel
792 lubricity, 83(17-18) (2004) 2351-2356.

793

794 Annexure A

795 Table A1: Energy consumption in ultrasound transesterification

Pulse mode Time on Time off Power Consumed Saved energy Duty cycle Saved energy
(on/off) (s) (s) (W) energy (kJ) % output (%)
(kJ)
5,5 750 750 500 375 375 50 50
5,3 938 562 500 469 281 62.5 37.5
5,2 1071 429 500 535 215 71.4 28.5
5,1 1250 250 500 625 125 83.3 16.6
796

797 It can be observed from the Table A1 that maximum energy output of 50% is saved with

798 50% duty cycle. Maximum energy 625 kJ is consumed with pulse sonication (5s on and 1s

799 on). For this study, 50% to 70% duty cycle was selected for production of biodiesel. Under

800 same operating conditions: reaction time (25 min), methanol to oil ratio (60%) and catalyst

801 concentration (1 Wt%), maximum energy savings of 50% are possible while maintaining

802 maximum a biodiesel yield of 94.47% with 50% duty cycle (5,5). Using 70% duty cycle (5,2)

803 the maximum biodiesel yield of 90.97% was obtained under the same operating conditions

804 with maximum energy savings of 28.5%.

805

806

36
807

808

37
Highlights

• Biodiesel was produced from Palm-sesame oil blend using ultrasound technique

• Cold flow and tribological characteristics of palm-sesame biodiesel were improved

• The process variables were optimized by RSM and ELM with cuckoo search

algorithm

• ELM-CS model predicted better than RSM based on Box Behnken Design
 Declaration of Interest Statement

I on the behalf of the all authors, declare that there is no conflict of interest and all the

elements of the submission are also in compliance with the journal publishing ethics. By

submitting this revised manuscript, the authors agreed that the copyright for their article

should be transferred to this journal if the article is accepted for publication. The work

contained within the research paper is our original contribution and has not been published

anywhere.

Muhammad Mujtaba Abbas

Centre for Energy Sciences

Faculty of Engineering
University of Malaya
50603, Kuala Lumpur, Malaysia
E-mail: m.mujtaba@uet.edu.pk
Corresponding Author