Available online at www.sciencedirect.

com

ScienceDirect
Available online at www.sciencedirect.com
Procedia Computer Science 00 (2018) 000–000

ScienceDirect www.elsevier.com/locate/procedia

Procedia Computer Science 120 (2017) 245–252

9th
9thInternational
InternationalConference
Conferenceon
onTheory
Theoryand
andApplication
ApplicationofofSoft
SoftComputing,
Computing,Computing
Computingwith
with
Words and Perception, ICSCCW 2017,
2017, 24-25
22-23 August
August 2017,
2017, Budapest,
Budapest, Hungary
Hungary

Fuzzy logic model for prediction of cold filter plugging point of

biodiesel from various feedstock
Filiz Al-Shanableha*, Ali Evcilb, Mahmut Ahsen Savaşc
a,b,c
Mechanical Engineering Department, Near East University, Nicosia, 99138, North Cyprus

Abstract

The present study investigates the potential of Fuzzy Logic model to predict the cold filter plugging point (CFPP) of biodiesel fuel
from its fatty acid (FA) composition. CFPP defines the operability for diesel fuel which is strongly influenced by the FA of
feedstock. Prediction of CFPP based on the FA of feedstock can reduce the experimental effort to produce a biodiesel suitable for
a regional climate. To generate fuzzy prediction model, data sets containing FA composition and CFPP temperatures of 60 biodiesel
samples traced from literature. To check the accuracy of the model developed, waste/refined frying oil (WFO/RFO) and
waste/refined canola oil (RCO/WCO) were converted to biodiesel fuel by transesterification then, their CFPP temperatures were
determined following the EN and ASTM standards. The CFPP temperatures estimated by the multi input single output (MISO)
fuzzy model were in close agreement with the experimental values. The validation results confirm the applicability of the fuzzy
model developed here with its high degree of accuracy and minimum time demand can be rapid alternative to the costly and time
consuming actual experiments and trails.

© 2018 The Authors. Published by Elsevier B.V.

Peer-review under responsibility of the scientific committee of the 9th International Conference on Theory and application of
Soft Computing, Computing with Words and Perception.

Keywords: biodiesel; cold filter plugging point; fatty acid composition; fuzzy logic

1. Introduction

Biodiesel is one of the most popular alternatives among the liquid fuels due to its suitability for the conventional
diesel engines with minimum or no modifications, as well as in blends with petroleum diesel. The usage of biodiesel
have beneficial effects on the environmental pollution, dependency to petroleum exporting countries, as well as
improvement of domestic industry. Despite its positive effects, biodiesel produced should meet the specifications of

Corresponding author: Filiz Al-Shanableh. Tel.; +90-392-2236624; fax: +90-392- 223 6624.
E-mail address: filiz.shanableh@neu.edu.tr

1877-0509 © 2018 The Authors. Published by Elsevier B.V.

Peer-review under responsibility of the scientific committee of the 9th International Conference on Theory and application of
Soft Computing, Computing with Words and Perception.

1877-0509 © 2018 The Authors. Published by Elsevier B.V.

Peer-review under responsibility of the scientific committee of the 9th International Conference on Theory and application of Soft
Computing, Computing with Words and Perception.
10.1016/j.procs.2017.11.235
246 Filiz Al-Shanableh et al. / Procedia Computer Science 120 (2017) 245–252
2 Filiz Al-Shanableh et al./ Procedia Computer Science 00 (2018) 000–000

ASTM D6751 or EN 14214 standards before being commercialized in the market. One of the major technical obstacles
to the practical use of biodiesel fuel is its cold flow properties, which can be critical depending on the climate and
seasonal conditions of a region (Hoekman et al., 2012). Crystal formation and agglomeration in cold weather can lead
to large crystals that restrict or block flow through fuel lines and filters, leading to fuel starvation and subsequent
engine failure. The cold flow properties of biodiesel and all diesel fuels can be described via the cold filter plugging
point (CFPP), cloud point (CP), pour point (PP). Among these parameters, the CFPP is direct and reliable indicator
for low-temperature engine operability. The CFPP is more commonly used to estimate of the lowest temperature at
which a fuel gives trouble-free flow in certain fuel systems (Echim et al. 2012). The CFPP specified in ASTM D6371-
05 and EN 116:1998 directly affects the diesel engine performance in winter. It determines the lowest temperature
where 20 ml of fuel can be drawn through a 45 micron screen in 60 seconds with 200 mm of water (1.96 kPa) vacuum.
Present work concentrates on the influence of the fatty acid (FA) composition of feedstock on the CFPP of biodiesel
fuel produced. Biodiesel is a mixture of fatty acid methyl esters (FAME) with each ester component contributing to
the properties of the fuel and those are identical to that of its parent oil or fat. Prediction of CFPP of biodiesel based
on the FA composition of the feedstock prior to production can lead to a way for a suitable biodiesel for the regional
climate conditions. One of the methods that can used to determine the effects of the FA composition on CFPP
temperature of biodiesel is the Fuzzy Logic Method or simply fuzzy method. Fuzzy method that can be an alternative
to classical statistical or mathematical methods was presented by L.A. Zadeh. Non-random uncertainty does not suit
the use of statistical or mathematical methods, and these methods are inadequate for such cases. Such non-random
uncertainties are identified and modelled as fuzzy (Ata and Dinçer, 2017). Prediction modelling studies such as
multiple regression analysis were proposed to predict the CFPP temperature of biodiesel from different feedstock.
Sarin et al. (2009) proposed CFPP prediction model based on palmitic methyl ester content (P FAME) while Su et al.
(2011) modelled based on the weighted average number of carbon atoms in FAME (Nc) and total unsaturated FA
content (UFAME). Moser (2008), Ramos (2009) and Wang (2011) prediction models was related to the total saturated
FA contents (ΣSats). All of these prediction models achieved around 0.90 of coefficient of determination (R2) however,
for a limited number of experimental data. In these models, it is essential for the user specified data points to fit the
curve in order to obtain an empirical correlation. Nevertheless, the curve fitting is not necessary in soft computing
methods such as fuzzy logic systems or artificial neural networks (ANN). While fuzzy model has not been adapted for
CFPP prediction yet, Al-Shanableh et al. (2016) predicted the CFPP temperatures based on the nine FA components
using the ANN model and the R2 value was found as 0.97 and noted that ANN was superior to the other methods. For
the prediction of CFPP, experimental data of 60 biodiesel fuel samples were traced from the literature with their
corresponding FA compositions of parent feedstock. The values collected were utilized as input and output data for
fuzzy logic model. The generated fuzzy model was used to predict CFPP of four different biodiesel samples that were
produced in this work, i.e., waste/refined frying oil based biodiesel (WFOME/RFOME) and waste/refined canola oil
based biodiesel (RCOME/WCOME).

2. Methods

2.1. Biodiesel Preparation and Determination of CFPP

Four types of feedstock, namely waste frying oil (WFO), refined frying oil (RFO), refined canola oil (RCO) and
waste canola oil (WCO) were converted into biodiesel via a base catalyzed transesterification as prescribed by
Freedman et al. (1986). Nine FA components (C12:0, C14:0, C16:0, C18:0, C18:1, C18:2, C18:3, C20:0 and C20:1)
of those feedstock were determined following the EN ISO 5508 as tabulated in Table 1.

Table 1. FA compositions of biodiesel feedstock used in the current work.

Fatty Acid Composition (wt %)
C12:0 C14:0 C16:0 C18:0 C18:1 C18:2 C18:3 C20:0 C20:1 SFA MUFA PUFA
WFO 1.18 0.10 39.29 4.04 40.42 13.84 0.18 0.0 0.0 44.99 40.56 14.02
RFO 1.02 0.08 38.4 3.52 39.2 16.9 1.02 0.0 0.0 43.35 39.28 17.92
RCO 0.08 0.0 5.63 1.57 62.97 21.34 6.99 0.46 1.04 7.74 64.01 28.33
WCO 0.56 0.0 6.02 2.01 63.22 18.08 4.61 0.49 1.12 9.08 64.3 22.69
 Filiz Al-Shanableh et al. / Procedia Computer Science 120 (2017) 245–252 247
Filiz Al-Shanableh et al./ Procedia Computer Science 00 (2018) 000–000 3

Transesterification was end up with the conversion of WFO to WFOME (Waste Frying Oil Methyl Ester), RFO to
RFOME, RCO to RCOME and WCO to WCOME. Chemical composition and fatty acid profiles of biodiesel samples
were analyzed using the Gas Chromatography– Mass Spectrometry (GC-MS) in accordance with the EN 14103
standard. CFPP of all biodiesel samples produced were measured following the ASTM D6371-05 and found as 14°C,
13°C, -7.5°C and -7°C for WFOME, RFOME, RCOME and WCOME respectively.

2.2. Data Collection

For the prediction of CFPP, experimental data of 60 biodiesel fuel samples were identified from the literature (Al-
Shanableh et al., 2016) with their corresponding FA compositions of parent feedstock and CFPP temperatures.
Literature data was utilized as input and output data for the fuzzy logic model. FA components were grouped to reduce
the number of input variables for the current fuzzy modelling as saturated FA (SFA = C12:0 + C14:0 + C16:0 + C18:0
+ C20:0), monounsaturated FA (MUFA = C16:1 + C18:1 + C20:1) and polyunsaturated FA (PUFA = C18:2 + C18:3).
SFA, MUFA and PUFA used as input data sets while CFPP as output data set. The maximum and minimum values
for input/output data sets in the 60 samples are listed in Table 2. When a feedstock’s FA composition falls within
these ranges then, models developed here could be employed to predict the CFPP of biodiesel produced from this
feedstock.

Table 2. Minimum and maximum values of FA and CFP of literature data.

Min. Max. Mean
SFA 2.80 78.10 22.86
MUFA 2.50 91.90 42.47
PUFA 0.70 85.30 55.60
CFPP (˚C) -13 17 1.22

2.3. Fuzzy Logic Modelling

Prof. Zadeh (1962) developed a fuzzy logic approach to deal with uncertainty and imprecision in decision making
processes for real world applications. The main benefits of a fuzzy logic system are its simplicity and speed,
development, lower cost, and greater ease of maintenance. Fuzzy logic does not require a mathematical model. A
fuzzy system makes its decision based on the inputs in the form of language statements and variables are tested with
if-then rules. A fuzzy process consists of three parts; fuzzification, fuzzy inference system (FIS) and defuzzification
(Ata and Dinçer, 2017).
Fuzzification considers measurement of input variable values and conversion of input term into particular linguistic
terms that can be considerably marked as fuzzy sets. This involves categorizing all the input variables into appropriate
linguistic categories; each categorical assignment is associated with a degree of membership between 0 and 1 that is
determined by fuzzy membership functions. A graphical representation of the variable boundary between fuzzy sets
is defined as membership function (MF). Triangular, trapezoidal, bell shaped MFs are commonly used in engineering
application among which trapezoidal MF is chosen in the current survey.
The core section of a fuzzy logic system is the FIS part, which combines the facts obtained from the fuzzification
with the rule base and conducts the fuzzy reasoning process. Information entering the rule processing unit are in a way
combined with such rules like ‘if … else … and’, stored in rules-based processing unit. Among different FISs, the
Mamdani fuzzy model is one of the most commonly used in fuzzy logic for solving many real-world problems.
Defuzzification is the process to convert fuzzy MFs back to crisp form. When rules are applied to input variables,
the outcome of each rule is still in fuzzy format, meaning that all different outcome possibilities are expressed as
degrees of certainty. There are different defuzzification techniques such as centroid weighted-average method,
bisector of area, centre of gravity, centre of mass, centre of largest area, mean of maxima to calculate the final
solution and here the centroid weighted-average method is used.
A fuzzy model expresses a complex system in the form of fuzzy implications. Mamdani model can be built by
using these implications (linguistic relationships) and observed data. The Mamdani-based fuzzy models use excessive
number of rules for system modelling. Let X be input (regression) matrix and g an output vector defined as
248 Filiz Al-Shanableh et al. / Procedia Computer Science 120 (2017) 245–252
4 Filiz Al-Shanableh et al./ Procedia Computer Science 00 (2018) 000–000

 x11 x12 
x x21 
X  x1 ,...., x2    21
T

   
 
 x n1 xn 2 

g  g1 ,....., g 2 

where upper script T denotes the transpose. In the Mamdani fuzzy model, both the antecedent and consequent are
fuzzy propositions. A general form of linguistic fuzzy if-then rule can be given as;
Ri : if x is Ai then y is Bi, i=1,2,….,K
where Ri is the rule number, Ai and Bi are the fuzzy sets, x is the antecedent variable representing the input in the
fuzzy system, and y is the consequent variable related to the output of the fuzzy system. The MF, μ(x) is defined as
the fuzzy subset a in the universe of discourse, x. The trapezoidal MF can be shown as

0  x  a
x  a
 a xb
b  a

 ( x)  1  b  x  c
d  x
 cxd
d  c
0  d  x

The trapezoidal MF is defined by parameters a, b, c, and d and their corresponding points presented in Fig.1.

Fig. 1. Trapezoidal membership function.

3. Results and Discussions

3.1. Fuzzy Logic Model Developed to Predict CFPP of Biodiesel

Multi Input Single Output (MISO)-Mamdani fuzzy models are developed, trained and tested to predict the CFPP
temperatures of biodiesel samples. The MATLAB R2015a (8.5.0.197613., Mathworks Inc., Natick, USA) Fuzzy
Logic Designer application was used to create the fuzzy prediction model. A three-input, one-output rule-based
Mamdani-type fuzzy logic model is developed. The model is constructed into fuzzy system using input, SFA, MUFA,
PUFA and the output parameter as CFPP. The fuzzy trapezoidal MF for the three input variables are shown in Fig. 2
(a-c), respectively. Similarly, the graph showing fuzzy trapezoidal MF for the output variable can be seen in Fig. 2
(d). Number of MF for each input variables were five while it was seven for the output variable.
 Filiz Al-Shanableh et al. / Procedia Computer Science 120 (2017) 245–252 249

Filiz Al-Shanableh et al./ Procedia Computer Science 00 (2018) 000–000 5

Fig. 2. The fuzzy membership function for three inputs; (a) SFA, (b) MUFA, (c) PUFA and output; (d) CFPP.

Numerical parameters of input and output variables were fuzzificated as linguistic variables. Linguistic variables
for five trapezoidal inputs MF were very low (L1), low (L2), medium (L3), high (L4), very high (L5) and for seven
trapezoidal output MF were very low (D1), low (D2), midlow (D3), medium (D4), midhigh (D5), high (D6), very high
(D7). For SFA corresponding parameters [a b c d] were taken as follows; L1 [-18 -2 6 16], L2 [2 18 22 38], L3 [22 38
42 58], L4 [42 58 62 78], L5 [62 78 80 80], for MUFA; L1 [-21 -2 2 21], L2 [2 21 26 45], L3 [26 45 49 68], L4 [49 65
70 85], L5 [71 76 90 115], for PUFA; L1 [-19 -2 2 19], L2 [2 19 24 41], L3 [24 41 45 62], L4 [45 62 67 84], L5 [67 84
88 105] and for CFPP D1 [-17.5 -13.5 -11.5 -8.5], D2 [-12.5 -8.5 -6 -3.5], D3 [-7.5 -3.5 -2.5 1.5], D4 [-2.5 1.5 2.5 6.5],
D5 [2.5 6.5 7.5 11.5], D6 [7.5 11.5 12.5 16.5], D7 [12.5 16.5 17.5 21.5]. The number of MFs and their parameters were
determined by trial and error, the aim was predicted CFPP temperatures should be as close as to actual counterparts.
The next stage in the fuzzy logic is to construct the if-then rules to represent the relationship between input and
output variables based on the linguistic terms. In this model, 43 rules were written at the beginning and reduce to 24
by changing the boundaries of parameters of MFs for the best fit of the model. Table 3 lists the rules generated by the
fuzzy inference system. The most important input parameter was the SFA which was non-linearly effecting CFPP
temperatures. Up to 40 % of SFA content CFPP showed a decreasing trend however, above 40 % it started to increase.
Generally, the SFA components introduced negative influences on CFPP temperature, i.e. increased th CFPP
temperature. On the other hand MUFA and PUFA has a positive effect on CFPP temperature, i.e. reduced to CFPP
temperature. While MUFA was affecting mostly non-linearly, PUFA influenced only linearly. It can be concluded
that PUFA has a positive influence on CFPP temperature, i.e. higher the PUFA content the lower the CFPP
temperature.
250 Filiz Al-Shanableh et al. / Procedia Computer Science 120 (2017) 245–252
6 Filiz Al-Shanableh et al./ Procedia Computer Science 00 (2018) 000–000

Table 3. The rules generated for prediction of CFPP by FIS.

Rule number If-Then Rules
R:1 If (SFA is L1) and (MUFA is L5) and (PUFA is L2) then (CFPP is D1)
R:2 If (SFA is L1) and (MUFA is L1) and (PUFA is L5) then (CFPP is D1)
R:3 If (SFA is L1) and (MUFA is L3) and (PUFA is L2) then (CFPP is D1)
R:4 If (SFA is L1) and (MUFA is L4) and (PUFA is L2) then (CFPP is D2)
R:5 If (SFA is L1) and (MUFA is L2) and (PUFA is L4) then (CFPP is D2)
R:6 If (SFA is L1) and (MUFA is L2) and (PUFA is L5) then (CFPP is D2)
R:7 If (SFA is L2) and (MUFA is L4) and (PUFA is L1) then (CFPP is D2)
R:8 If (SFA is L1) and (MUFA is L2) and (PUFA is L3) then (CFPP is D3)
R:9 If (SFA is L2) and (MUFA is L2) and (PUFA is L4) then (CFPP is D3)
R:10 If (SFA is L5) and (MUFA is L1) and (PUFA is L1) then (CFPP is D3)
R:11 If (SFA is L2) and (MUFA is L4) and (PUFA is L2) then (CFPP is D3)
R:12 If (SFA is L2) and (MUFA is L2) and (PUFA is L3) then (CFPP is D3)
R:13 If (SFA is L1) and (MUFA is L2) and (PUFA is L4) then (CFPP is D3)
R:14 If (SFA is L2) and (MUFA is L3) and (PUFA is L3) then (CFPP is D4)
R:15 If (SFA is L2) and (MUFA is L3) and (PUFA is L2) then (CFPP is D4)
R:16 If (SFA is L2) and (MUFA is L1) and (PUFA is L4) then (CFPP is D5)
R:17 If (SFA is L3) and (MUFA is L3) and (PUFA is L2) then (CFPP is D5)
R:18 If (SFA is L1) and (MUFA is L5) and (PUFA is L1) then (CFPP is D5)
R:19 If (SFA is L4) and (MUFA is L2) and (PUFA is L1) then (CFPP is D5)
R:20 If (SFA is L3) and (MUFA is L3) and (PUFA is L1) then (CFPP is D6)
R:21 If (SFA is L5) and (MUFA is L2) and (PUFA is L1) then (CFPP is D6)
R:22 If (SFA is L4) and (MUFA is L3) and (PUFA is L1) then (CFPP is D7)
R:23 If (SFA is L2) and (MUFA is L1) and (PUFA is L5) then (CFPP is D7)
R:24 If (SFA is L3) and (MUFA is L3) and (PUFA is L2) then (CFPP is D7)

The last stage called defuzzification was performed to obtain predicted output variables using the rules generated.
In this model, the centre of area method is used for defuzzification. CFPP temperatures of samples were predicted by
providing SFA, MUFA and PUFA compositions in Matlab Fuzzy Logic Designer rule reviewer section. The fuzzy
model developed provides the predicted values of biodiesel samples when proper input data are fed into the model.
Table 4 shows input/output data sets of some biodiesel samples, applied rule for fuzzified data sets and predicted
CFPP values. Linguistic variables of input/output variables are given in the parenthesis.

Table 4 Some examples of the data sets used to model fuzzy model and predicted CFPP values.
Oil/ Fat SFA MUFA PUFA CFPPEXP Rule no. CFPPPRE
Almond 13.6 (L2) 77.1 (L4) 8.4 (L1) -6 (D2) R:7 -6.1
Babassu 76.7 (L5) 14.2 (L2) 1.8 (L1) 10 (D6) R:21 8.1
Beef tallow 50.4 (L4) 40.8 (L3) 4.0 (L1) 14 (D7) R:22 14.9
Borage 13.1 (L2) 17.1 (L2) 64.8 (L4) -4 (D3) R:9 -3.2
Camelina 13.4 (L2) 31.2 (L2) 52.6 (L3) -3 (D3) R:12 -3.1
Canola 7.1 (L1) 62.2 (L4) 30.7 (L2) -7 (D2) R:4 -5.7
Castor 2.0 (L1) 90.3 (L5) 4.6 (L1) 7 (D5) R:18 4.8
Coconut 78.1 (L5) 6.8 (L1) 2.2 (L1) -5 (D3) R:10 -5.4
Coffee 15.0 (L2) 70.0 (L4) 13.5 (L2) -4 (D3) R:11 -2.5
Corn 14.3 (L2) 27.2 (L2) 57.5 (L4) -3 (D3) R:9 -3.5
Cottonseed 26.8 (L2) 1.9 (L1) 71.1 (L4) 5 (D5) R:16 5.4
Evening primrose 8.1 (L1) 6.6 (L1) 85.3 (L5) -10 (D1) R:3 -8.8
Hazelnut 10.1 (L1) 79.5 (L5) 11.0 (L2) -13 (D1) R:1 -12.5
Jatropha 18.4 (L2) 39.1 (L3) 41.8 (L3) 0 (D4) R:14 -1.1
Lesquerella 3.4 (L1) 79.5 (L4) 16.4 (L2) -6 (D2) R:4 -6.2
Palm 49.0 (L3) 41.9 (L3) 8.9 (L1) 12 (D6) R:20 10.7
Peanut 13.8 (L2) 4.2 (L1) 81.7 (L5) 17 (D7) R:23 17.1

The accuracy and predictive capability of the fuzzy logic model developed were evaluated in terms of coefficient
of determination (R2) and root mean square error (RMSE) as follows:

n
( y exp,i  y p ,i ) 2
R2  1 i 1


n
i 1
( y p ,i  y exp,ave ) 2
 Filiz Al-Shanableh et al. / Procedia Computer Science 120 (2017) 245–252 251
Filiz Al-Shanableh et al./ Procedia Computer Science 00 (2018) 000–000 7


n
RMSE  1
n i 1
( y p ,i  y exp,i ) 2

where n is the number of experimental data, yexp,i is the experimental value, yp,i is the predicted value, yexp,ave is
the average experimental value and n is the number of input variables.
R2 and RMSE of the fuzzy model for CFPP prediction were found as 0.986 and 0.871, respectively. The high value
of the R2 was an indication of the good fit for a particular model while low RMSE indicated high precision. An R2
value approaching to unity, implied a good correlation between experimental and predicted values as noted in Fig. 3.
These results showed that, fuzzy logic rules were appropriately selected. In general, the fuzzy logic is correlated with
the human thoughts in generating verbal fuzzy rules.
20
15
10
CFPPpredicted

5
0
-5 R² = 0.986
-10
-15
-15 -5 5 15 25
CFPPexperimental

Fig. 3. The CFPP values predicted by the fuzzy logic model versus experimental counterparts.

3.2. Validation of Fuzzy Logic Model Developed

The prediction ability of the fuzzy logic model generated were tested against those measured CFPP values of four
biodiesel samples produced in this work. In addition, the testing data set were applied to those CFPP prediction models
given in the literature by Moser (2008), Sarin et al. (2009), Su et al. (2011), Ramos et al. (2009), Wang (2011 and Al-
Shanableh et al. (2016). Table 5 provides a performance comparison of the CFPP prediction models and fuzzy logic
model developed in this work.

Table 5. A comparison of the measured and the predicted CFPP temperatures of biodiesel samples.
CFP Method Correlation WFOME RFOME RCOME WCOME RMSE R2
Experimental 14 13 -7.5 -7
Moser 0.438(ΣSats)-8.93 10.8 10.1 -4.4 -4.9 2.849 0.840
Sarin et al. 0.511(PFAME)-7.823 12.3 11.8 -4.9 -4.8 1.996 0.934
Su et al. 18.019(NC)-0.804(UFAME) 10.5 8.8 -20.7 -18.7 9.341 0.698
CFPP Ramos et al. 3.1417LCSF-16.477 2.1 1.1 -10.8 -9.9 8.697 0.176
(°C) 1.7556LCSF-14.772
Wang et al. 27.7 23.3 -6.9 -5.9 8.593 0.753
LCSF= ∑(MPn×Cn)/100
Al-Shanableh et al. ANN based model 10.9 11.1 -6.8 -7.2 1.854 0.958
Proposed Fuzzy
12.9 13.1 -8.2 -7.4 0.684 0.996
Logic Model
PFAME: Palmitic acid methyl ester content (wt %); UFAME: Total unsaturated FA (wt%); ΣSats: total saturated FA content (wt%)
NC: Weighted-average number of carbon atoms; LCSF: The long-chain saturated factor for C16:0-C24:0
Cn: Mass fraction of saturated FA (wt %); MPn : Melting point of saturated FA

The current fuzzy logic model predicted the CFPP temperatures of four biodiesel samples with higher accuracy
than the other models as seen in Table 5. The CFPP temperatures of any biodiesel sample can be estimated by using
“.fis” file of proposed model if the SFA, MUFA and PUFA of its feedstock fall within range which are given in Table
2. The advantages of fuzzy model for predicting CFPP temperature could release experimental effort and time and
offers a choice to estimate the suitable feedstock to produce biodiesel for a particular regional climate.
252 Filiz Al-Shanableh et al. / Procedia Computer Science 120 (2017) 245–252
8 Filiz Al-Shanableh et al./ Procedia Computer Science 00 (2018) 000–000

4. Conclusions

A fuzzy logic model was created to predict CFPP temperature of biodiesel from various feedstock. Input parameters
used in the fuzzy model construction were saturated fatty acids and mono/poly unsaturated fatty acids of the
biodiesel’s feedstock. The fuzzy model developed here was tested by the four biodiesel samples produced also in this
work. CFPP temperatures estimated by the fuzzy logic model were in close agreement with their experimental
counterparts, i.e. R2 value was as 0.987. The prediction performances of the generated model in terms of RMSE was
found as 0.871 that indicated a high accuracy. The literature comparison proves that the developed fuzzy model
appears to be more precise than those published CFPP prediction models before. In this content, an estimation of the
CFPP temperatures of a potential biodiesel fuel from a new feedstock in prior to the experimental work appeared be
valuable and also would shorten the practical effort.

Acknowledgements

A part of this work was supported by the Research Project No: NEU 2010-2-21 of the Near East University
Research Fund.

References

Al-Shanableh, F., Evcil, A., Savaş, M.A., 2016. Prediction of Cold Flow Properties of Biodiesel Fuel Using Artificial Neural Network. Procedia
Comput Sci 102: 273-280. doi:10.1016/j.procs.2016.09.401
Aluclu, I., Dalgic, A., Toprak, Z.F., 2008. A fuzzy logic based model for noise control at industrial work places. Appl Ergon 39: 368–378.
ASTM D6751-09., 2009. Standard Specification for Biodiesel Fuel Blend Stock (B100) for Middle Distillate Fuels.
Ata, S., Dinçer, K., 2017. Fuzzy logic modeling of performance proton exchange membrane fuel cell with spin method coated with carbon nanotube.
Int Jo Hyd En 42 (4): 2626–2635.
Echim, C., Maes, J., Greyt, W.D., 2012. Improvement of cold filter plugging point of biodiesel from alternative feedstocks. Fuel 93: 642–648.
EN 14214:2012. Automotive Fuels - Fatty Acid Methyl Esters (FAME) – Requirements and Test Methods.
Freedman, B., Butterfield, R.O., Pryde, E.H., 1986. Transesterification kinetics of soybean oil. J Am Oil Chem Soc 63 (10): 1375–80.
Hoekman, S.K., Broch, A., Robbins, C., Ceniceros, E., Natarajan, M., 2012. Review of biodiesel composition, properties, and specifications. Renew
Sust Energ Rev 16 (1): 143–169.
Moser BR., 2008. Influence of blending canola, palm, soybean, and sunflower oil methyl esters on fuel properties of biodiesel. Energ and Fuel
22(6): 4301–6.
Ramos, M.J., Fernández, C.M., Casas, A., Rodríguez, L., Pérez, A., 2009. Influence of fatty acid composition of raw materials on biodiesel
properties. Bioresource Technol 100: 261–268.
Sarin, A., Arora, R., Singh, N.P., Sarin, R., Malhotra. R.K., Kundu, K.., 2009. Effect of blends of Palm-Jatropha-Pongamia biodiesels on cloud
point and pour point. Energy 43: 2016-2012. doi: 10.1016/j.energy.2009.08.017
Sharma, Y.C., Singh, Y.C., Upadhyay, S.N., 2007. Advancements in development and characterization of biodiesel: A review. Fuel 87: 2355-2373.
Su, Y.C., Liu, Y.A., 2011. Selection of prediction methods for thermophysical properties for process modeling and product design of biodiesel
manufacturing. Ind. Eng. Chem. Res 50: 6809-6836.
Wang, S., Ma, S., 2011. Improving the cold flow properties of biodiesel from waste cooking oil by surfactants and detergent fractionation. Fuel 90:
1036–1040.