See discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/233784359
Multivariate near infrared spectroscopy models
for predicting the oxidative stability of
biodiesel: Effect of antioxidants addition
Article in Fuel · July 2012
DOI: 10.1016/j.fuel.2012.02.017
CITATIONS
READS
13
99
4 authors, including:
Nuno Canha
Pedro M Felizardo
52 PUBLICATIONS 480 CITATIONS
18 PUBLICATIONS 685 CITATIONS
University of Lisbon
SEE PROFILE
4Tune Engineering
SEE PROFILE
Jose C Menezes
Technical University of Lisbon
93 PUBLICATIONS 1,543 CITATIONS
SEE PROFILE
All content following this page was uploaded by Nuno Canha on 18 January 2017.
The user has requested enhancement of the downloaded file. All in-text references underlined in blue are added to the original document
and are linked to publications on ResearchGate, letting you access and read them immediately.
Fuel 97 (2012) 352–357
Contents lists available at SciVerse ScienceDirect
Fuel
journal homepage: www.elsevier.com/locate/fuel
Multivariate near infrared spectroscopy models for predicting the oxidative
stability of biodiesel: Effect of antioxidants addition
Nuno Canha a, Pedro Felizardo a, José C. Menezes b, M. Joana Neiva Correia a,⇑
a
b
Centre of Chemical Processes, Instituto Superior Técnico, Technical University of Lisbon, Av. Rovisco Pais, 1049-001 Lisbon, Portugal
IBB, Centre for Biological and Chemical Engineering, Instituto Superior Técnico, Technical University of Lisbon, Av. Rovisco Pais, 1049-001 Lisbon, Portugal
a r t i c l e
i n f o
Article history:
Received 7 July 2011
Received in revised form 19 November 2011
Accepted 7 February 2012
Available online 20 February 2012
Keywords:
Biodiesel
Oxidative stability
NIRS
Calibration models
a b s t r a c t
Biodiesel, a mixture of long chain fatty acid esters, is an environmental friendly alternative to fossil fuel.
This fuel is produced by a transesterification reaction between vegetable oils or animal fats and a short
chain alcohol, usually methanol, in the presence of a catalyst. European governments are targeting the
incorporation of 10% of biofuels in transportation fuels by 2020. Therefore, the global market for biodiesel
is expected to have a significant growth in the next 10 years. According to the European legislation, from
the 25 parameters that have to be analyzed to certify biodiesel quality, oxidative stability is of concern,
especially when storing biodiesel for long periods. In fact, that property measures the susceptibility of
biodiesel to oxidative degradation and is strongly dependent on the type of oil used in the production
process and on storage conditions. Thus, EN 14214 establishes a minimum value of 6 h for the oxidative
stability of biodiesel under stressed conditions of a standardized assay. This work reports the use of near
infrared spectroscopy (NIRS), coupled with multivariable classification and calibration techniques, to
determine the oxidative stability of biodiesel with and without antioxidants. Therefore, biodiesel samples
produced from soybean, palm, rapeseed, sunflower and waste frying oils, from mixtures of these oils and
also several of these samples after different storage conditions and storage periods some of them containing antioxidants (induction periods between 0.66 and 17.75 h) were used to develop the calibration models. The model for samples without antioxidants is able to estimate the oxidative stability of unknown
samples with a root mean square error of prediction (RMSEP) of 0.6 h, which is similar to the reference
method error (0.5 h). The introduction of samples containing antioxidants in the calibration/validation
sets led to higher prediction errors (RMSEP = 1.28 h) that may be considered acceptable.
Ó 2012 Elsevier Ltd. All rights reserved.
1. Introduction
It is known that there are several economical, political and
environmental problems associated to the use of fossil fuels and
biodiesel is pointed as an environmental friendly alternative to fossil fuel because it does not contribute to the net atmospheric CO2
level. Additionally, its use as fuel in diesel engines also contributes
to a reduction of the emissions of several pollutants (particulate
matter, carbon monoxide, sulphur and polycyclic aromatic hydrocarbons) but adversely affects the emissions of NOx [1]. When
waste frying oils (WFOs) are used as a raw material for biodiesel,
the process is also considered as an industrial or household waste
treatment [2].
Biodiesel production is commonly carried out through a transesterification reaction between vegetable oils or fats and an alcohol
(usually methanol), in the presence of a catalyst, to produce an
ester and glycerol as byproduct [2]. Presently, basic homogeneous
⇑ Corresponding author. Tel.: +351 21 8417344; fax: +351 21 8417246.
E-mail address: qjnc@ist.utl.pt (M. Joana Neiva Correia).
0016-2361/$ - see front matter Ó 2012 Elsevier Ltd. All rights reserved.
doi:10.1016/j.fuel.2012.02.017
catalysed batch processes are generally in use at industrial level,
but the use of heterogeneous catalysts has several advantages because it allows reducing the operational costs due to the reduction
of catalysts consumption and to simpler purification operations
[3]. According to geographic limitations and oil prices, biodiesel
can be produced from different feedstocks and, as mentioned
above, using different production technologies. As a result, the final
product can have different properties and so, according to EN
14214, there are 25 parameters that have to be analyzed to certify
biodiesel quality and the oxidative stability is one of these parameters [4]. The oxidative stability, which is a measure of how easily
biodiesel suffers oxidative degradation, is one of the major drawbacks for its widespread commercialization [5]. Out of specification
values of the oxidative stability create problems during the fuel
storage and distribution and also affect the performance of the
vehicle because the degradation products led to the formation of
engine deposits that can block the filters and fuel injectors [6,7].
On the other hand, there are also several biodiesel properties that
are directly affected by its degradation, such as the viscosity and
the acid value that increase with the decrease of the induction
N. Canha et al. / Fuel 97 (2012) 352–357
period [6–8]. Apart from the type of feedstock, namely its fatty acid
composition and content of unsaturated chains, there are several
factors that affect the oxidation process of biodiesel during storage
such as presence of air, light, high temperatures, presence of antioxidants and the presence of contaminants in, for example, the
storage tanks because metals may have a catalytic effect on oxidation [8,9]. Therefore, the standard specifications in Europe include
a minimum induction time of 6 h at 110 °C determined using the
Rancimat method [10], which is a simple but time consuming analytical method.
The use of NIRS, associated with multivariate data analysis, to
monitor the quality of vegetable oils and biodiesel is reported in
several papers [11–22]. NIR spectroscopy is a well-known analytical technique based on the absorption of electromagnetic in the region from 780 to 2500 nm (12820–4000 cm–1). The analysis of the
NIR spectra requires the use of chemometric tools, which include
principal components analysis (PCA) to perform the qualitative
analysis of the spectra, or the projection into latent structures
(PLS) regression to develop calibration models between spectral
and analytical data for quantitative analysis [23]. To improve the
correlation between the absorption of NIR radiation and the analytical reference data, specific spectra pre-processing and variable
selection methods may be used [15,19]. Variable selection methods
or interval partial least squares (iPLS) allow the determination of
the spectral region(s) where variations are specifically related to
changes in the analyte concentration.
The advantages of applying NIR spectroscopy to determine the
oxidative stability of biodiesel are of significance. In fact, after
the NIR calibration has been established, it will only be necessary
to acquire the NIR spectrum of the sample, a procedure that takes
less than 2 min, instead of several hours that are necessary to
spend to perform the same determination using the Rancimat
apparatus.
2. Experimental
2.1. Materials and method
Methanol, chromatographic grade (99.5%), was supplied by José
M. Vaz Pereira (Lisbon, Portugal) and sodium methoxide, commercial grade (30% by weight of sodium methoxide in methanol) was
purchased from BASF (Lisbon, Portugal). The antioxidants butylated
hydroxyltoluene (BHT, 99%), butylated hydroxyanisol (BHA, 98%)
and hydroquinone (HQ, 99%) were purchased from Sigma–Aldrich.
2.1.1. Biodiesel samples
Industrial Samples of pure vegetal oils, such as palm, soybean
and rapeseed, were supplied by Iberol and Biovegetal, two Portuguese biodiesel producing companies, whereas the waste frying
oils were supplied by Space, another Portuguese biodiesel producing company. Laboratory-scale samples were prepared by transesterification of different pure vegetable oils and WFO using the
following procedure: a sample of oil was transferred into a stirred
tank reactor equipped with a reflux condenser and immersed in a
temperature-controlled water bath. For stirring, a single paddle
round impeller (diameter = 6.5 cm) at 350 revolutions min1 was
used. The oil was heated until the desired temperature was
reached (65 °C). At this point, a mixture of methanol and sodium
methoxide was added to the oil and the transesterification reaction
began. The reactor was kept at around 65 °C for 180 min. At the
end of the reaction period, the glycerol-rich phase was separated
from the methyl esters layer in a decantation funnel. The esters
phase (crude biodiesel) was washed with water, with a 0.1 M HCl
solution and again with water to provide a purified biodiesel. The
washed methyl esters were then centrifuged (Sigma 4K10, Oste-
353
rode am Harz, Germany) and dried at 80–90 °C under vacuum
(200 mbar) using a rotary evaporator (RE111; Büchi, Flawil,
Switzerland).
To produce biodiesel samples with a wide range of variation of
the oxidative stability, mixtures of soybean, rapeseed and waste
frying oils with palm oil were used as raw-materials. In fact, palm
oil has a high level of unsatured fatty acids thus leading to the production of biodiesel with high values of the oxidative stability. The
addition of antioxidants (HQ, BHT and BHA) to some of the samples
was also carried out.
2.1.2. Analyzes
The oxidative stability was determined by the Rancimat method
[10] according to EN 14214 using a Metrohm 679 equipment. Air
flow was set at 10 L/h and the temperature of the heating block
was set at 110 °C. The determinations of the induction period were
performed in duplicate and the mean values were used for calibration purposes.
The near-infrared spectra of the biodiesel samples were acquired using an ABB BOMEM MB160 (Zurich, Switzerland) spectrometer equipped with an InGaAs detector and a transflectance
probe from SOLVIAS (Basel, Switzerland). The spectra were recorded twice for each sample at room temperature (25 ± 1 °C), with
the aid of the Galactic Grams software package (Galactic Industries,
Salem, NH, USA), in the wave number range of 12,000–4000 cm1,
with a spectral resolution of 16 cm1. The average of the two measurements was used for models development. Some of the other
biodiesel properties were determined using NIR models already
developed [14,15,19,20].
2.2. Data analysis and calibration development
All calculations were carried out using Matlab version 7
(MathWorks, Natick, MA, USA) and the PLS Toolbox version 4.0
(Eigenvector Research Inc., Manson, WA, USA). Only the region
between 9000 and 4500 cm1 was used for calibration because
the noisy (<4500 cm1) and non-informative (>9000 cm1) ranges
of the spectra were excluded.
As mentioned above, data analyzes were performed using principal components analysis (PCA) for a qualitative analysis of the
spectra, followed by the PLS regression to develop the calibration
models between spectral and analytical data. Different preprocessing methods and spectral regions were evaluated and the choice
was made by analyzing the ones that led to the best performance
parameters (minimum number of principal components or latent
variables, lower root mean square errors of calibration and validation, etc.). In this work, the PCA model was developed using the
second order Savitsky–Golay derivative with a filter width of 15
data points and the third-order polynomial fit followed by the
standard normal variate scaling (SNV) and mean centering
(SV2 + SNV + MC), whereas for the PLS model developed with the
samples without antioxidants, the best performance parameters
were obtained using the first order Savitsky–Golay derivative with
a filter width of nine data points and the third-order polynomial fit
followed by the standard normal variate scaling (SNV) and mean
centering (SV1 + SNV + MC). However, when the samples containing anti-oxidants were introduced, the orthogonal signal correction (OSC) was also used (SV1 + SNV + MC + OSC). This preprocessing allows to reduce the data variance in the spectra (X)
due to light scatter effects and to more general types of interferences that have no correlation with the measured property y (in
this case the oxidative stability) [25].
In the PLS regression, the optimal number of latent variables
(LVs) needed in the calibration model was obtained by cross-validation (CV). The internal validation strategy used was the Venetian
Blinds method [24]. It is a recurring procedure, which sets aside
354
N. Canha et al. / Fuel 97 (2012) 352–357
every nth sample of the calibration set at a time, then builds a calibration model without the excluded sample and, finally, makes a
prediction of the excluded sample using the calibration model
developed.
The detection of outliers was performed based on the leverage
values, Q-residuals, and Studentized y-residuals. As a result, a sample was considered to be an outlier if its leverage value was twice
as large as the average leverage value (given by 2(1 + LV)/N, where
LV is the number of latent variables and N the number of samples),
or if its Q-residual falls above the 95% confidence limits for the considered model, or yet if y-residual of the sample was larger than
twice the residual standard deviation [20].
The performance of the calibration models was analyzed by calculating the root mean squares errors of cross-validation, RMSECV,
and of external validation, RMSEP, and the respective determination coefficients, Q 2y , between the predicted and the measured values [25,26]. The latter coefficient, which quantifies the variance in
y being predicted by the model and should be close to 100%, was
calculated as
Q 2y ¼
1
!
^ÞT ðy y
^Þ
ðy y
100
yT yÞ
ð1Þ
The calibration models were developed using data sets randomly split using the Shuffle function of Matlab. Therefore, after
the random reorder of the matrix rows, the complete data set
was divided into the calibration (the first 2/3 of the matrix’ rows)
and the external validation sets (the last 1/3 of the matrix’ rows).
Finally, iPLS was also applied to identify the regions of the spectra that better capture the variance of the oxidative stability [20].
However, the best results were obtained using the complete
spectra.
3. Results and discussion
3.1. Characterization of biodiesel samples
The set of samples used to calibrate the oxidative stability contains 199 samples of biodiesel produced from soybean, palm, rapeseed, sunflower and waste frying oils, from mixtures of these oils
and also several of these samples after different storage periods
and conditions (Table 1). In fact, as mentioned above, the type of
oil used for biodiesel production, as well as the storage time and
conditions greatly influences the oxidative stability values. Therefore, the calibrations were developed using biodiesel samples produced from different oils and mixtures of oils commonly used as
raw materials for biodiesel production, fresh or stored in different
conditions (light in transparent glass or plastic bottles, light in opaque glass or plastic bottles, darkness, heated in an oven at 50 °C).
This strategy allowed to extend the calibration range to oxidation
stability values or induction periods several times lower and higher than the limit values imposed by the European (6 h) and the USA
(3 h) legislations. Actually, to have a conveniently developed NIR
calibration model, well defined in the range imposed by legislation,
it is convenient to extend the calibration range. In addition, it is
also possible to have samples out of specification and so the calibration range was chosen to cover the entire expected variability
of the oxidative stability of real biodiesel samples.
The objective of this work is to evaluate the ability of NIRS together with multivariable data analysis to determine the oxidative
stability of biodiesel samples with and without antioxidants. For
that reason, most of the results concerning the study of the influence of raw-materials, storage conditions, etc., on the oxidative
stability will be presented elsewhere.
The induction periods of several biodiesel samples produced
from pure oils that were also used to prepare the mixtures for
the NIRS calibration models are presented in Table 2.
As expected, Table 2 shows that the induction period decreases
from palm to soybean/rapeseed biodiesel. Actually, due to the
higher content of unsaturated fatty acids, soybean and rapeseed
biodiesel are more reactive to oxidation than saturated palm biodiesel [5,8]. In what concerns waste frying oils, they usually present low values of the oxidative stability due also to the reactions
that occur during the frying process. In fact, depending on the degree of heating, the oils suffer degradation and polymerization
reactions that are responsible for changes in their chemical and
physical properties [2] that led to low values of the oxidative
stability.
It is worth noting that the lower limit of the calibration interval
(0.43 h) was attained after storage of some of the samples (or mixtures) presented in Table 2 for various time periods in different
conditions and containers. In this way, it was possible to lower
the oxidative stability from 2 h (Table 2) down to 0.43 h. This is
important because it is possible to have industrial or laboratory
biodiesel samples out of the specification imposed by EN 14214,
namely when they are produced from waste frying oils or other
low-cost raw-materials and NIRS calibration should be developed
for the entire range of interest. Furthermore, in order to study
the effect of the presence of antioxidants on the performance
parameters of the NIRS model, the three antioxidants mentioned
above were added to several soybean and rapeseed biodiesel samples in concentrations from 100 to 810 mg/kg. This addition allowed an increase of the induction time of 1 up to 16 h,
depending on the antioxidant and concentration used.
It is important to emphasize that the strategy used in this work
was to develop two calibration models: one with biodiesel samples
without additives and the other with samples with and without
antioxidants. In fact, the addition of antioxidants is not always necessary, and so it is important to have a robust NIR model trained
with biodiesel samples without additives for a large range of oxidative stability values. On the other hand, instead of deriving a model
based only in samples containing antioxidants, it was decided that
it is more useful to have a robust calibration model, well defined
over a wide range of oxidative stability values where it is expected
Table 1
Biodiesel samples used for NIR calibration.
Type of biodiesel/number of
samples
Animal fat (AF)
Sunflower (SF)
Rapeseed (R)
Soybean (SB)
Palm (P)
WFO
Pure
oil
Mixtures
Stored samples
Nr
Nr
Type of oil
Composition range
(%)
Nr
Time
Conditions
1
2
5
93
–
–
25
66
65
41
–
–
5–95, 20–80
5–95, 20–80, 60–70,
80
5–95, 5–95, 10–20
20, 20
–
1
2
119
4
8
–
–
P, SB
P, R, WFO,
WFO + P
SB, R, SB + WFO
SB, SB + P
–
4 weeks
4 weeks
2 days to
32 weeks
2–32 weeks
2–32 weeks
–
Light
Light
Light/darkness/oven at 50 °C/glass or plastic
containers
Light/darkness
Light/darkness
22
35
355
N. Canha et al. / Fuel 97 (2012) 352–357
Table 2
Oxidative stability of biodiesel samples.
Oil
SB 1
SB 2
R1
R2
P1
P2
SF
WFO 1
WFO 3
AF
Induction time (h)
5.8
7.3
5.7
6.2
17.8
18.9
2.1
2.8
2.9
2.9
Raw-materials: SB – soybean oil; R – rapeseed oil; P – palm oil; WFO – waste frying oils: AF – animal fat.
to have most of the biodiesel samples produced at a laboratory or
at industrial scale (even the ones that are far from specification).
On the contrary, a model derived with the samples containing antioxidants separately would have less practical interest because
most of the samples, or at least those produced from virgin oils,
would have oxidative stability values higher than the minimum
value imposed by EN 14214.
3.2. Development of NIRS calibration models for the oxidative stability
3.2.1. NIRS model developed with samples without antioxidants
In what concerns the PCA analysis of the spectra, previous work
showed that biodiesel samples are grouped according to the type
of oil used in its production [20], which influences their iodine value and oxidative stability. However, the results obtained in this
work showed that the PCA analysis did not allow the clear identification of the oxidative stability variation pattern probably because, as mentioned above, the oxidative stability is also
dependent on other factors.
For the PLS model’s development, from the 199 samples, 149
were used for calibration and 50 for validation. The calibration
range goes from 0.43 to 18.95 h. After excluding six outliers following the above mentioned procedure, the PLS model was developed
after applying the second order Savitsky–Golay derivative followed
by the standard normal variate scaling (SNV) and mean centering
(SV1 + SNV + MC) as pre-processing to the complete spectra. The
lowest RMSECV was obtained using six latent variables that allowed capturing 99.57% of the data variance in the spectra and
97.31% of the data variance concerning the oxidative stability.
Table 3 presents the performance parameters of this model.
From Table 3 it is possible to conclude that this model performs
very well with acceptable values of R2 and Q 2y . Actually, it allows
the prediction of the oxidative stability with an error similar to
the maximum error expected for the reference method (±0.5 h).
Furthermore, in spite of the large calibration range, with only six
latent variables it was possible to obtain a prediction error correspondent to only 3.5% of the entire calibration range, which is a
very good result. In fact, for example, Lira et al. [21] developed a
calibration model for the oxidative stability of biodiesel for a narrower calibration range (1.5–8.7 h) and with 16 latent variables
obtained a RMSEP correspondent to around 6.9% of the calibration
range.
Fig. 1 shows the performance of the model for both the calibration and validation sets. From this figure it is possible to observe
that the calibration and validation samples are well distributed
around the diagonal, with a higher number of samples with an oxidative stability lower than 8 h, which is the region where it is expected to have most of the industrial samples. Furthermore, in
spite of its large calibration range, the model allows obtaining
small deviations and there is an excellent agreement between
the values predicted by the model and those determined by the
reference method. Therefore, these results confirm that this model
can be used to predict the oxidative stability of biodiesel without
antioxidants with errors similar to the Rancimat ones.
3.2.2. NIRS model developed with samples containing antioxidants
To develop the calibration model for samples with and without
antioxidants, 48 biodiesel samples were spiked with the above
mentioned antioxidants in concentrations between 100 and
810 mg/kg and these samples were added to the dataset containing
the 199 samples without antioxidants. Thus, the resulting set, containing 247 samples, was randomly divided into the calibration
(185 samples) and validation (62 samples) sets, covering a calibration range between 0.43 and 22.1 h.
As presented in Table 4, when SV1 + SNV + MC was used as
spectra pre-processing, 10 latent variables are necessary to capture
99.88% of the spectra variance and 93.48% of the y variance. This
large number of latent variables indicates that this model is more
difficult to construct than the one that contains only samples without anti-oxidants. Additionally, such a large number of latent variables may indicate an overfitted model, with too many LV, that is
describing not only the systematic variations but also the random
variations or noise. In this case, the model will fail to predict new
samples [27]. Thus, in an attempt of improving these results, the
orthogonal signal correction was used as pre-processing technique
(SV1 + SNV + MC + OSC). In this case, it was possible to capture
with only one latent variable 81.6% of the data variance in the
spectra and 94.14% of the samples data variance y (the oxidative
Table 3
Cross-validation and external validation results for the prediction of the oxidative
stability.
Parameter
Spectral region 9000–4500 cm1
Latent variables
Calibration range (h)
R2calibration
6
0.66–17.75
0.973
R2cross validation
0.967
R2prediction
0.950
RMSEC (h)
RMSECV (h)
RMSEP (h)
Rancimat error (h)
Q 2y cross validation
0.53
0.59
0.60
0.50
96.68
Q 2y prediction
94.90
Fig. 1. Correlation between measured and predicted oxidative stability (h) of
biodiesel in the region 4500–9000 cm1 (: calibration set; j: validation set).
356
N. Canha et al. / Fuel 97 (2012) 352–357
Table 4
Cross-validation and external validation results for the prediction of the oxidative
stability of biodiesel samples with and without antioxidants.
Parameter
Spectral region 9000–4500 cm1
SV1 + SNV + MC
SV1 + SNV + MC + OSC
R2calibration
10
0.43–22.10
0.935
1
0.43–22.10
0.941
R2cross validation
0.899
0.910
R2prediction
0.906
0.908
RMSEC (h)
RMSECV (h)
RMSEP (h)
Rancimat error (h)
Q 2y cross validation
0.97
1.22
1.26
0.5
89.81
0.92
1.15
1.28
0.5
90.98
Q 2y prediction
89.72
89.44
Latent variables
Calibration range (h)
stability). Additionally, in spite of the decrease of the number of latent variables, the performance parameters of the calibration models are similar. It is important to mention that the calibration
model containing samples with antioxidants was developed after
the removal of 12 outliers from the calibration set and one outlier
from the validation set, but only three of the 13 samples were samples containing antioxidants.
The difficulty of developing a NIRS calibration model to determine the oxidative stability in samples with and without antioxidants is illustrated by the higher number of latent variables that
are necessary to correlate the spectra variance with the oxidative
stability of the samples or by the necessity of using an additional
pre-treatment, by the higher number of samples identified as outliers (12 instead of 6) and also by the higher RMSEC, RMSECV and
RMSEP values. This difficulty is also clear from Fig. 2, which shows
a greater dispersion of the data points around the diagonal 1:1,
when compared to Fig. 1.
Fig. 2 also shows that the dispersion of the values around the
diagonal seems to increase for the samples with high oxidative stability values, which may indicate a dependency of the error on the
antioxidant concentration. In this situation, the oxidative stability,
instead of being only affected by the structure of the fatty acid
chains of biodiesel that NIRS is able to identify, is also affected
by the presence of antioxidants. Therefore, two samples with similar NIR spectra may have completely different values of the oxidative stability. It is thus understandable that in this case the
prediction errors are higher than in the case of the model developed with the samples without antioxidants.
From Table 4 it is also possible to conclude that regardless the
fact that the RMSEP value is higher than the maximum error of
the reference method (0.50 h), this model still allows the prediction of the oxidative stability of biodiesel samples with and without antioxidants. In fact, the prediction error of 1.28 h may be
considered acceptable because it corresponds to only 5.9% of the
entire calibration range. It is also important to emphasize that this
error will decrease if a narrower calibration range is used. However, this strategy will reduce the robustness and applicability of
the calibration model.
The results obtained in this work allow to conclude that NIRS
may be used to predict the oxidative stability of samples without
and with antioxidants. In fact, despite a prediction error twice
times the one of the reference method, acceptable predictions are
still possible. Furthermore, the speed of the NIRS analysis is an
important advantage when compared with the Rancimat one,
namely for additives addition control or blending operations for
targeted specifications at an industrial level.
4. Conclusions
The aim of this work was to study the application of NIRS to
predict the oxidative stability of biodiesel. The oxidative stability
of biodiesel depends on the fatty acid composition of raw-materials and on the storage conditions, which may include exposure to
air and/or light, high temperature, as well as the presence of several contaminants and can be improved by the addition of antioxidants. The determination of the oxidative stability by the
Rancimat method is time consuming and therefore it is of great value to have a simple, fast and reliable alternative method to determine that important quality parameter of biodiesel. The results
obtained indicate that NIRS is a suitable accurate alternative to
the Rancimat method to determine the oxidative stability of biodiesel samples without antioxidants. In fact, the prediction error
was only 0.6 h for a calibration range of oxidative stability values
between 0.66 and 17.75 h. The introduction of samples containing
antioxidants led to a prediction error twice times higher than the
reference method error (1.28 h), which may be considered acceptable. Furthermore, despite the higher prediction error, the speed of
the NIRS analysis is an important advantage when compared with
the Rancimat one, namely for antioxidants addition control or
blending operations for targeted specifications at an industrial
level.
Acknowledgements
Thanks are due to Iberol, Biovegetal and Space for supplying
industrial samples of oils and to Professor Gabriela Gil from IST
for the use of the Rancimat equipment. Pedro Felizardo would also
like to thank Fundação para a Ciência e Tecnologia (SFRH/BDE/
15566/2005) and Space for financial support of his PhD.
References
Fig. 2. Correlation between measured and predicted oxidative stability (h) of
biodiesel in samples with and without antioxidants (: calibration set; j:
validation set).
[1] Bakea E, Karavalakis G, Stournas S. Biodiesel emissions profile in modern diesel
vehicles. Part 1: Effect of biodiesel origin on the criteria emissions. Sci Total
Environ 2011;409(9):1670–6.
[2] Felizardo P, Neiva Correia MJ, Raposo I, Mendes JF, Berkemeier R, Bordado JM.
Production of biodiesel from waste frying oils. Waste Manage 2006;26(5):
487–94.
[3] Sharma Y, Singh B, Korstad J. Lates developments on application of
heterogeneous basic catalysts for an efficient and eco friendly synthesis of
biodiesel: a review. Fuel 2010;90(4):1309–24.
[4] European Standard EN 14214, Automotive fuels – fatty acid methyl esters
(FAME) for diesel engines – requirements and test methods. CEN – European
Committee for Standardization, Brussels, Belgium; 2002.
N. Canha et al. / Fuel 97 (2012) 352–357
[5] Karavalakis G, Hilari D, Givalou L, Karonis D, Stournas S. Storage stability and
ageing effect of biodiesel blends treated with different antioxidants. Energy
2011;36:369–74.
[6] Schober S, Mittelbach M. The impact of antioxidants on biodiesel oxidation
stability. Eur J Lipid Sci Technol 2004;106(6):382–9.
[7] Park J, Kim D, Lee J, Park S, Kim Y, Lee J. Blending effects of biodiesels on
oxidation stability and low temperature flow properties. Bioresour Technol
2008;99(5):1196–203.
[8] Knothe G. Some aspects of biodiesel oxidative stability. Fuel Process Technol
2007;88(7):669–77.
[9] Kivevele TT, Mbarawa MM, Bereczky A, Laza T, Madarasz J. Impact of
antioxidant additives on the oxidation stability of biodiesel produced from
Croton Megalocarpus oil. Fuel Process Technol 2011;92:1244–8.
[10] European Standard EN 14112. Produit dérivés des corps gras - Esters
méthyliques d’acides gras (EMAG) – Détermination de la stabilité à
l’oxydation (Essai d’oxydation accélérée). CEN – Comité Européen de
Normalisation, Bruxelles, Belgique; 2002.
[11] Armenta S, Garrigues S, Guardia M. Determination of edible oil parameters by
near infrared spectrometry. Anal Chim Acta 2007;596(2):330–7.
[12] Yang H, Irudayaraj J, Paradkar M. Discriminant analysis of edible oils and fats
by FTIR, FT-NIR and FT-Raman spectroscopy. Food Chem 2005;93(1):25–32.
[13] Baptista P, Felizardo P, Menezes JC, Correia JN. Monitoring the quality of oils
for biodiesel production using multivariate near infrared spectroscopy models.
J Near Infrared Spectrosc 2008;16:445–54.
[14] Felizardo P, Baptista P, Uva MS, Menezes JC, Correia JN. Monitoring biodiesel
fuel quality by near infrared spectroscopy. J Near Infrared Spectrosc 2007;
15(2):97–105.
[15] Felizardo P, Baptista P, Menezes JC, Correia JN. Multivariate near infrared
spectroscopy models for predicting methanol and water content in biodiesel.
Anal Chim Acta 2007;595(1–2):107–13.
View publication stats
357
[16] Knothe G. Rapid monitoring of transesterification and assessing biodiesel fuel
quality by near-infrared spectroscopy using a fiber-optic probe. J Am Oil Chem
Soc 1999;76(7):795–800.
[17] Knothe G. Monitoring a progressing transesterification reaction by fiber-optic
near infrared spectroscopy with correlation to 1H nuclear magnetic resonance
spectroscopy. J Am Oil Chem Soc 2000;77(5):489–93.
[18] Knothe G. Analytical methods used in the production and fuel quality
assessment of biodiesel. Trans ASAE 2001;44:193–200.
[19] Baptista P, Felizardo P, Menezes JC, Correia JN. Multivariate near infrared
spectroscopy models for predicting the methyl esters content in biodiesel.
Anal Chim Acta 2008;607(2):153–9.
[20] Baptista P, Felizardo P, Menezes JC, Correia JN. Multivariate near infrared
spectroscopy models for predicting the iodine value, CFPP, kinematic viscosity
at 40 °C and density at 15 °C of biodiesel. Talanta 2008;77(1):144–51.
[21] Lira LFB, Albuquerque MS, Pacheco JGA, Fonseca TM, Cavalcanti EHS, Stragevitch
L, et al. Infrared spectroscopy and multivariate calibration to monitor stability
quality parameters of biodiesel. Microchem J 2011;96:126–31.
[22] Dorado MP, Pinzi S, Haro A, Font R, Garcia-Olmo J. Visible and NIR
Spectroscopy to assess biodiesel quality: determination of alcohol and
glycerol traces. Fuel 2011;90:2321–5.
[23] Otto M. Chemometrics: statistics and computer application in analytical
chemistry. New York: Wiley-VCH; 1999.
[24] Wise BM, Gallagher NB, Bro R, Shaver JM, Winding W, Koch RS. PLS toolbox 4.0
– reference manual for use with Matlab. Eigenvector – Research Incorporated.
[25] Næs T, Isaksson T, Fearn T, Davies T. A user-friendly guide to multivariate
calibration and classification. Chichester, UK: NIR Publications; 2002.
[26] Martens H, Næs T. Multivariate calibration. Chichester, UK: John Wiley & Sons;
1991.
[27] Esbensen KH. Multivariate data analysis – in practice. 5th ed. Camo Process AS;
2002.