Fuel 160 (2015) 57–63

Contents lists available at ScienceDirect

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

Quantification of biodiesel and adulteration with vegetable oils

in diesel/biodiesel blends using portable near-infrared spectrometer
Eduardo Maia Paiva a, Jarbas José Rodrigues Rohwedder b, Celio Pasquini b, Maria Fernanda Pimentel c,
Claudete Fernandes Pereira d,⇑
a
Universidade Federal Rural de Pernambuco, Departamento de Química, CEP 52.171-030 Recife, PE, Brazil
b
Instituto de Química, Universidade Estadual de Campinas, SP 13.083-970, Brazil
c
Departamento de Engenharia Química, Universidade Federal de Pernambuco, Recife, PE 50.740-521, Brazil
d
Departamento de Química Fundamental, Universidade Federal de Pernambuco, Recife, PE 50740-560, Brazil

h i g h l i g h t s

 Portable NIR spectrometer was used to quantify biodiesel and vegetable oil in diesel.
 Performance of the portable NIR spectrometer was compared with the benchtop one.
 PLS and MLR regression models with variable selection were developed.
 Portable NIR spectrometer is feasible for monitoring quality of diesel blends.

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

Article history: The performance of a portable near infrared spectrophotometer to determine biodiesel content in die-
Received 14 May 2015 sel/biodiesel blends, as well as to quantify adulteration by vegetable oil, was evaluated. The NIR transmit-
Received in revised form 7 July 2015 tance/absorbance spectra of 118 samples in the range 950–1650 nm were acquired using an
Accepted 20 July 2015
ultra-compact spectrophotometer (MicroNIR), and also by using a Fourier transform benchtop instru-
Available online 29 July 2015
ment (FT-NIR) from 800 to 2500 nm. In a preliminary step, partial least squares (PLS) models aimed at
the determination of biodiesel quality parameters were developed by using full spectral information.
Keywords:
Thereafter, the variables showing significant regression coefficients were selected and the PLS models
Diesel/biodiesel blends
Diesel/biodiesel adulteration
were re-processed. Successive Projections Algorithm (SPA) was used to screen for the most orthogonal
Portable near-infrared (NIR) variables to be used in a model constructed by using Multiple Linear Regression (MLR). For determining
spectrophotometer the contents of both biodiesel and vegetable oil in diesel/biodiesel blends, using MicroNIR and FT-NIR, the
Multivariate calibration results indicated MLR-SPA as the best modelling strategy. RMSEP values of 0.22% v/v (MicroNIR) and
Variable selection 0.13% v/v (FT-NIR) for biodiesel and 0.34% v/v (MicroNIR) and 0.22% (FT-NIR) for vegetable oil contents
were obtained. LOD values of 0.31% (MicroNIR) and 0.22% (FT-NIR) for biodiesel and 0.40% (MicroNIR)
and 0.34% (FT-NIR) for vegetable oil determination were achieved. The results revealed comparable per-
formances between the portable and the benchtop instruments, demonstrating the feasibility of monitor-
ing in situ the quality characteristics of diesel/biodiesel blends according to biodiesel content and for
detection of adulteration with vegetable oils by using the portable spectrophotometer.
Ó 2015 Elsevier Ltd. All rights reserved.

1. Introduction this cheaper raw material, or as a result of incomplete conversion

The prevalent fuel commercialized in Brazil is a blend of diesel
with (7.0 ± 0.5)% (v/v) of biodiesel. These blends are highly subject
to adulteration with vegetable oil, either as an illegal addition of
⇑ Corresponding author. Tel.: +55 81 2126 7417; fax: +55 81 2126 8442.
E-mail address: clauimp@gmail.com (C.F. Pereira).
of the oil into ester during processing [1–3]. Therefore, it is impor-
tant to ensure that diesel blends are in accordance with quality
standards and regulations.
Standard methods, such as EN 14078 and ASTM D7371-07,
which use middle infrared (MIR) with univariate and multivariate
methods, respectively, are described for determination of biodiesel
in diesel fuel [4,5]. However, these methods do not consider the
presence of vegetable oil as a contaminant.

http://dx.doi.org/10.1016/j.fuel.2015.07.067
0016-2361/Ó 2015 Elsevier Ltd. All rights reserved.
58 E.M. Paiva et al. / Fuel 160 (2015) 57–63
The use of Near Infrared (NIR) spectroscopy is very attractive for
the purpose of quality control of blends since it does not require
sample treatment and enables fast analysis [6–9]. The literature
reports extensively on methods using Near Infrared (NIR) spec-
troscopy with multivariate calibration for the determination of
biodiesel and vegetable oil contents in diesel, suggesting the appli-
cability of this technique.
Pimentel et al. [10] have described the use of NIR spectroscopy
to quantify biodiesel content (0.0–5.0%, v/v) in diesel/biodiesel
blends employing Partial Least Square (PLS) regression models.
Their results demonstrate the possibility of quantifying biodiesel
content using a root mean squared error of external validation
(RMSEP) of 0.18% v/v even in the presence of raw vegetable oil.
Oliveira et al. [11] developed PLS and Neural Networking (ANN)
models using NIR spectra to quantify different methyl esters in bio-
diesel/diesel blends (0.0–100% w/w). Two groups of samples,
Group I (one type of methyl ester + diesel) and Group II (three
types of methyl ester + diesel), were employed. The best RMSEP,
0.06% w/w (Group I) and 0.32%, w/w (Group II), were obtained
using PLS models.
Fernandes et al. [12] have evaluated the use of NIR spec-
troscopy, PLS models and Multiple Linear Regression (MLR), with
and without variable selection by the Jack-Knife algorithm
(PLS-JK), Successive Projection Algorithm (MLR-SPA) and
Stepwise (MLR-SW) to determine the biodiesel content in die-
sel/biodiesel blends. The best RMSEP was 0.64% w/w using
MLR-SPA. Overall, the use of variable selection improved the
results compared with PLS models constructed by using full
spectrum.
Calibration models based in PLS and support vector machine
regression (SVMR) methods without and with variable selection
by synergy interval (siPLS and siSVR) were developed by
Filgueiras et al. [13], to determine animal fat biodiesel content in
soybean biodiesel blends (RMSEP = 0.18% w/w) and 20% w/w bio-
diesel (animal fat plus soybean)/diesel blends (RMSEP = 0.10%
w/w) employing NIR spectroscopy. Vasconcelos et al. [1] have
developed several calibration models (PLS, PLS-JK and MLR-SPA)
using the NIR overtone region (833–1666 nm) to quantify biodiesel
and vegetable oil contents in diesel/biodiesel blends (0.0–10.0%
v/v). The use of sample cells with optical path lengths of 10 mm,
20 mm and 50 mm indicated that the best RMSEP were 0.08%,
0.09% and 0.12% v/v for biodiesel content, and 0.37%, 0.45% and
0.57% v/v for vegetable oil content, respectively.
Oliveira et al. [2], demonstrated that the ASTM 4052, ASTM D
445, ASTM D 4737, ASTM D 93, and ASTM D 86 methods are not
suitable for identifying diesel samples adulterated with vegetable
oil. The authors demonstrated the use of NIR and Raman spec-
troscopy and Principal Components Regression (PCR), PLS, and
ANN based calibration models to determine vegetable oil (0–5%
w/w) in diesel/biodiesel blends. For PLS model employing NIR
spectra the RMSEP obtained was 0.045% w/w.
The determination of the vegetable oil content in diesel/biodie-
sel blends (0.0–30% w/w of vegetable oil) was also described by
Gaydou et al. [3]. The authors investigated the use of NIR and
MIR spectroscopies separately and combined. For the separate
regions, the best RMSEP (0.36% w/w) was obtained using the NIR
region. For the combined regions, the lowest RMSEP (0.27% w/w)
was achieved using Hierarchical-PLS (H-PLS). Recently, Alves and
Poppi [14] developed a NIR spectroscopy quantitative models,
based on Support Vector Machine (SVM) and PLS, for determina-
tion of biodiesel content (0.0–21% v/v) in diesel/biodiesel/hydro-
carbon renewable diesel blends. The PLS model shows a RMSEP
value of 0.19% v/v for determination of biodiesel content.
Common to all these works is the fact that they employed
benchtop spectrophotometers often based on Fourier transform
(FT), which undoubtedly offer high spectral resolution,
signal-to-noise ratio and better wavelength reproducibility [15–
17]. However, FT-NIR spectrophotometers are expensive and less
suitable for field analysis, which is important for fuel quality con-
trol. In contrast, portable NIR instruments are very attractive for
field analysis, despite their reduced spectrophotometric character-
istics. Recently, an extremely miniaturized NIR spectrometer,
called the MicroNIR (JDSU Corporation, Milpitas, USA), employing
a thin film linear-variable filter (LVF) as the radiation analyzing
element, became commercially available [18,19]. This instrument
uses two small internal tungsten light bulbs as the radiation
source, and a 128-element detector (InGaAs) and it is powered
and controlled through an USB port of a computer running propri-
etary software. The instrument is able to obtain spectra in the
range 950–1650 nm and has been primarily designed to be used
for spectral reflectance measurements.
Several recently published works can be found in the literature
employing the MicroNIR spectrophotometer. Alcalà et al. [20] used
the MicroNIR in qualitative discrimination between authentic and
illegal pharmaceuticals and quantitative determination of acetyl-
salicylic acid, ascorbic acid and caffeine. O’Brien et al. [21] used
reflectance spectra measured with MicroNIR to differentiate very
similar fish species with quite different prices, according to their
quality. Singh et al. [22,23] employed the MicroNIR and PAT
(Process Analytical Technology) tools to monitor a tablet manufac-
turing process. Rohwedder et al. [24] have developed a miniatur-
ized gas analyzer based on the use of a substrate-integrated
hollow waveguide (iHWG) coupled to a MicroNIR. Lutz et al. [25]
developed a transflectance measurement cell to be used with
MicroNIR to quantify the ethanol content in gasoline.
Up to now, no work has reported on using MicroNIR to access
quality parameters of diesel blends, although it is possible to find
some commercially available portable FT-MIR spectrophotometers
(middle infrared) designed for quantifying biodiesel in diesel/bio-
diesel blends. The present work describes simple instrumental
modifications required to employ the MicroNIR for in-field trans-
mittance/absorbance measurements, and the development of ana-
lytical methods based on multivariate regression and variable
selection to determine biodiesel and vegetable oil content in diesel.
2. Materials and methods
2.1. Samples
A mixture design with a central point and including internal
points was employed to prepare blends using different diesel fuels,
oilseeds and soybean ester. Maximum and minimum levels of both
biodiesel and oil were 0.0% and 15.0% (v/v). The ratio of ester/oil
concentration was varied from 0.25 to 4.0. A set of 118 samples
was obtained. The biodiesel, oil, and diesel samples used to prepare
the blends in the laboratory were all within the quality specifica-
tions and were acquired in the market, from different manufac-
tures, to match real samples.
2.2. Near infrared spectra acquisition
A portable NIR spectrometer, MicroNIR (JDSU Corporation,
Milpitas, USA), which operates in the spectral region 900–
1650 nm, was used to acquire the NIR spectra. This instrument,
powered (5 V) and controlled via USB port of a computer, originally
employs two tungsten light bulbs as radiation source, a Linear
Variable Filter (LVF) as dispersing element and an uncooled
128-element detector (InGaAs). The MicroNIR resolution is
12.5 nm at 1000 nm. Spectra of the samples were obtained by aver-
aging 50 scans with an integration time of 50 ms resulting in a total
measurement time of 2.5 s per sample.
 E.M. Paiva et al. / Fuel 160 (2015) 57–63 59

Two different adaptations were made to the original MicroNIR:

(1) one, containing a plane mirror placed after the sample cell com-
partment, was constructed, enabling transflectance measurements
to be made by using the internal radiation sources of the instru-
ment, and (2) the other used external radiation sources to allow
for transmittance/absorbance measurement employing (a) an
external air cooled tungsten source (25 W, 12 V) and (b) tungsten
mini-lamp (1 W, 5 V), powered via the USB port of the computer.
The performance of the MicroNIR was compared to a benchtop
FT-NIR spectrophotometer (model MB 160D, ABB Bomem). In this
case, each spectrum was acquired as an average of 100 scans with a
spectral resolution of 8 cm 1 in the spectral region of 800–
2500 nm. Total time required for one spectra acquisition is equal
to 60 s.
For both instruments, the background (reference, 100% trans-
mittance) spectra were obtained using a clean empty sample cell
with a 20 mm optical path length. Room temperature was con-
trolled at 23 ± 1 °C throughout the spectral acquisition process.

2.3. Chemometric procedure and software Fig. 1. Two different configurations developed to acquire spectra of liquids using
the MicroNIR spectrophotometer: (a) transflectance mode; (b) transmittance mode.
(1) MicroNIR, (2) compartment of cell (20 mm optical path length), (3) plane mirror,
All spectral data were pre-processed using different strategies (4) tungsten mini-lamp (1 W) powered via USB port of a laptop.
such as baseline correction (offset), moving average and first
derivative with a Savitzky–Golay filter second-order polynomial
using different window dimensions. In addition, the spectral data
the radiation from the MicroNIR internal bulbs. The spectra
were also mean-centered before the modeling procedures.
obtained when employing this configuration show quite low val-
The SPXY (sample set partitioning based on joint x–y distances)
ues of absorbance, less than half, when compared with the spectra
algorithm [26,27] was applied to divide the sample set into calibra-
obtained using the benchtop spectrophotometer (FT-NIR) in the
tion (70%) and validation (30%) subsets. Score, residual and lever-
transmittance mode for the same sample cell with 20 mm optical
age plots were used to detect and eliminate outliers.
path. This arrangement made it difficult to control and avoid the
Multivariate calibration models were built using Partial Least
spurious radiation that reached the detectors, arising from radia-
Squares (PLS) with the full spectral data (PLS full spectrum) and
tion reflected by the parts of the cell, support, and even from the
after variable selection made by: (1) Successive Projections
sapphire window of the MicroNIR.
Algorithm (SPA) for the MLR models (MLR-SPA) [28], and (2)
The transmittance setup (Fig. 1b) was initially evaluated using
Jack-Knife algorithm [29], which selects variables to be used in
an external air cooled tungsten source (25 W), with the radiation
the PLS regression according to the stability and significance of
sources of the MicroNIR turned off. The MicroNIR worked, in this
the regression coefficients (PLS-JK). Full cross-validation (CV) was
case, only as wavelength selector and detection system. The
carried out to optimize the models for the best number of PLS fac-
absorbance intensities obtained with this configuration were more
tors or variables, in case of MLR.
similar (about 60%) to those obtained by the FT-NIR spectropho-
The root mean squared error of prediction (RMSEP) obtained by
tometer. The radiation source, however, requires a large and heavy
external validation subsets was employed to evaluate the perfor-
high voltage power supply (110 V) counteracting the portability of
mance of the models. The statistical significance of differences
the instrument. Therefore, this same setup was tested using a
between RMSEP values were assessed using an F-test at a confi-
tungsten mini-lamp (1 W, 5 V) as the radiation source (Fig. 1b).
dence level of 95%. The F-values were calculated as the ratio of
This radiation source can be driven by the 5 V supply present in
the squares of the largest and smallest RMSEP values. This ratio
any USB port of a laptop, for example, contributing to the instru-
was compared with the critical F(n, n, 0.95) value, where n is the
ment portability. The spectra of diesel/biodiesel obtained using
number of samples in the prediction sets. A t test [13,30], at a con-
this simple setup also showed lower values of absorbance when
fidence level of 95%, was carried out to evaluate the significance of
compared with the FT-NIR instrument, (about 40% lower), as can
the bias of the estimates obtained by external validation. Limits of
be noticed in Fig. 2. The cause is probably associated with the pres-
detection values (LOD) were estimated according to reference [25].
ence of some spurious reflected radiation still present, and to the
The predicted values of biodiesel and oil for ten samples containing
lower spectral resolution of the MicroNIR. Nevertheless, it was pos-
zero biodiesel and zero oil were employed to estimate the average
sible to quantify biodiesel and vegetable oil in diesel/biodiesel
standard deviations (SD). The LOD was calculated as 3.3 SD.
blends employing these spectra as demonstrated below.
The data pre-treatment and PLS regression were carried out
Fig. 2 depicts the average spectrum of diesel/biodiesel blends
using UnscramblerÒ X.2 (CAMO S.A.) software. The SPXY and
(0–15% v/v) registered using the portable (MicroNIR) and a bench-
MLR-SPA algorithms were performed with MatlabÒ R2010a 199
top (FT-NIR) spectrophotometers in the spectral region of 900–
7.10.0.499.
1700 nm. In this spectral range, the bands are generally assigned
to first overtone of combinations (1350–1400 nm) and second
3. Results and discussion overtone (1120–1250 nm) of CAH stretching. In addition, the
fourth overtone of the C@O is expected to appear around
Fig. 1 depicts the two different setups evaluated in the present 1160 nm [31]. The low intensity band close to 900 nm, is assigned
work to acquire absorbance spectra of liquid samples using the to the third overtone of CAH stretching, and can be clearly
MicroNIR spectrophotometer. The first one (Fig. 1a) operates in observed in the FT-NIR spectrum. Although this band can also be
transflectance mode (the nominal optical path length is doubled). observed in the MicroNIR spectra, it is close to the limit of the
A plane mirror was placed after the cuvette compartment to reflect working spectral region of the instrument, where the signals show
60 E.M. Paiva et al. / Fuel 160 (2015) 57–63

Fig. 2. Average spectrum of diesel/biodiesel blends in NIR region registered with:

(a) MicroNIR using the transmittance mode (tungsten mini-lamp: 1 W), (b) FT-NIR
spectrophotometers. Optical path length: 20 mm.

poor signal-to-noise ratio. As can be verified in Fig. 2, the spectral

resolution of MicroNIR is worse than for the FT-NIR along the full
spectral region. It is worth mentioning that the FT-NIR operates
from 800 to 2500 nm; but above 1700 nm the spectra acquired
using the 20 mm optical path length showed saturated absorption
signals (results not shown). On the other hand, the MicroNIR spec-
trophotometer used in the present work was designed to operate
in the spectral region 900–1650 nm. Therefore, the comparison
between the performances of these instruments was made in this
spectral region.
PLS regression models were developed using the spectral region
900–1650 nm, and the region including only the bands assigned to
first overtone of combinations and second overtone of CAH
stretching bands (982–1571 nm) to evaluate the more suitable
spectral region for the determination of biodiesel and vegetable
oil in diesel/biodiesel blends. Overall results from PLS models using
the spectral region 900–1650 nm (results not shown) were worse
than that employing the region 982–1571 nm, indicating this last
wavelength range as the best choice for both instruments.
Considering this spectral region, the number of variables for the
MicroNIR and FT-NIR spectra is 96 and 990, respectively.
The results of the regression models developed to determine
the biodiesel content employing the spectra sets acquired with
MicroNIR and FT-NIR spectrophotometers are summarized in
Table 1. For the PLS and PLS-JK models, the RMSEP values obtained
Fig. 3. Wavelengths selected by SPA and employed by MLR models using spectral
data obtained by the MicroNIR and FT-NIR spectrophotometers to determine (a)
biodiesel (b) vegetable oil contents in diesel/biodiesel blends.
with MicroNIR are statistically similar to those obtained with
FT-NIR at a confidence level of 95%, according to F-test. However,
the PLS-JK model, built using the MicroNIR, presents a significant
bias, according to the t test (at a confidence level of 95%).
The best model based on MLR-SPA, developed using the spectra
acquired with MicroNIR, showed a RMSEP of 0.22% v/v for the
determination of biodiesel in diesel blends. The value is statisti-
cally different, according to the F-test, from that obtained with
FT-NIR (RMSEP = 0.13% v/v) using the same modelling strategy.
The results obtained by the MLR-SPA model for MicroNIR and
FT-NIR are statistically different (at 95% confidence level) from
those obtained with PLS and PLS-JK models. To quantify the biodie-
sel content, the number of spectral variables employed by
MLR-SPA models using MicroNIR and FT-NIR was fifteen and

Table 1
Overall result for regression models developed for the NIR determination of biodiesel content in diesel/biodiesel blends using the two instruments.

Spectrophotometer Model Calibration (N = 83) External validation (N = 35) LOD (% v/v)

Fa RMSECV (% v/v) R2 Bias RMSEP (% v/v) R2 Bias
MicroNIR PLSb 5 0.39 0.990 0.0163 0.32 0.988 0.0864 1.36
PLS-JKb 5 0.34 0.992 0.0067 0.26 0.991 0.1080 1.04
MLR-SPAc 15 0.17 0.999 0.0023 0.22 0.998 0.0492 0.31
FT-NIR PLSd 5 0.24 0.996 0.0009 0.26 0.994 0.0146 0.67
PLS-JKd 4 0.34 0.992 0.0064 0.27 0.993 0.0096 0.98
MLR-SPAe 12 0.14 0.999 0.0004 0.13 0.999 0.0349 0.22
a
Number of PLS factors or selected variables by SPA.
b
With 1st derivative (SG 2nd order polynomial and 3 window points).
c
With baseline + moving average with 3 window points.
d
With 1st derivative (SG 2nd order polynomial and 5 window points).
e
With baseline + moving average (5 window points).
 E.M. Paiva et al. / Fuel 160 (2015) 57–63 61

Fig. 4. Predicted versus reference plots obtained to determine biodiesel content in diesel/biodiesel blends using: (a) MLR-SPA with MicroNIR, (b) MLR-SPA with FT-NIR, (c)
PLS-JK with FT-NIR. (s) calibration (cross validation results) (+) external validation sets.

Table 2
General results for regression models developed for the NIR determination of vegetable content in diesel/biodiesel blends using two different instruments.

Spectrophotometer Model Calibration (N = 83) External validation (N = 35) LOD (%v/v)

Fa RMSECV (% v/v) R2 Bias RMSEP (% v/v) R2 Bias
MicroNIR PLSb 6 0.49 0.982 0.0070 0.46 0.973 0.0195 1.60
PLS-JKb 6 0.41 0.988 0.0061 0.41 0.978 0.0228 1.41
MLR-SPAc 18 0.21 0.998 0.0024 0.34 0.992 0.0458 0.40
FT-NIR PLSd 5 0.34 0.992 0.0008 0.30 0.992 0.0610 0.98
PLS-JKd 4 0.35 0.991 0.0106 0.34 0.990 0.0984 1.15
MLR-SPAe 11 0.26 0.998 0.0001 0.22 0.998 0.0491 0.34
a
Number of PLS factors or selected variables by SPA.
b
With 1st derivative (SG 2nd order polynomial and 3 window points).
c
With baseline + moving average with 3 window points.
d
With 1st derivative (SG 2nd order polynomial and 5 window points).
e
With baseline + moving average (5 window points).

twelve, respectively. As can also be seen in Table 1, all developed
models explained more than 98% of the total variance of
the data. Undoubtedly, the MLR-SPA models are the better
choice to quantify biodiesel in diesel/biodiesel blends for both
spectrophotometers.
Fig. 3a shows the spectral variables selected by SPA for the MLR
models built using MicroNIR and FT-NIR spectrophotometers to
quantify biodiesel in diesel/biodiesel blends. In general, these vari-
ables are close; however, as the spectra from FT-NIR have better
resolution than those from MicroNIR, the FT-NIR results are better
to quantify biodiesel content, as demonstrated by the overall
results of the regression models. Among the selected variables,
those assigned to third (around 1350 nm) and fourth (around
1160 nm) overtones of C@O bond, which are characteristic of bio-
diesel, were included [31,1].
Although the RMSEP value (0.22% v/v) obtained with MLR-SPA
using the MicroNIR spectrophotometer to quantify the biodiesel
content is poorer than that obtained with FT-NIR, it is comparable
with those values described in the literature by Pimentel et al.,
2006 (RMSEP = 0.18% v/v) [10], and Alves and Poppi, 2013
(RMSEP = 0.19% v/v) [14], employing expensive benchtop
spectrophotometers.
Plots of the predicted versus reference values for biodiesel con-
tent obtained from the MLR-SPA and/or PLS-JK models using both
spectrophotometers are shown in Fig. 4. As can be seen, all the
points are randomly distributed about the bisectrix line through
the entire range of y-values.
As stated before, raw oil can be present in the diesel/biodiesel
blends due to illegal addition or due to incomplete conversion to
ester during the process of biodiesel production. Considering these
62 E.M. Paiva et al. / Fuel 160 (2015) 57–63
Fig. 5. Predicted versus reference plots obtained to determine vegetable oil content in diesel/biodiesel blends using: (a) MLR-SPA with MicroNIR, (b) MLR-SPA with FT-NIR,
(c) PLS-JK with MicroNIR, (d) PLS-JK with FT-NIR. (s) calibration (cross validation) (+) external validation sets.
facts, we also developed models to quantify vegetable oil in die-
sel/biodiesel blends. The results obtained with both spectropho-
tometers are presented in Table 2. All regression models using
the spectra acquired with the MicroNIR show similar predictive
performances (RMSEP from 0.34% to 0.46% v/v). The MLR-SPA
model achieved the lowest RMSEP value of 0.34% v/v, when only
eighteen spectral variables were used (Fig. 3b).
When the regression models developed using the spectra
acquired in the MicroNIR for the quantification of oil content are
compared with those obtained with the FT-NIR, the FT-NIR shows
better predictive performances (RMSEP values between 0.22 and
0.34% v/v). An exception is the PLS-JK model (RMSEP = 0.41% v/v)
obtained using MicroNIR, which achieved a result (according to
F-test) similar to the PLS-JK model (RMSEP = 0.34% v/v) using the
FT-NIR spectrophotometer. Again, the MLR-SPA attained the low-
est RMSEP value (0.22% v/v), which is statistically different from
those values obtained with the PLS and PLS-JK models. Only eleven
spectral variables were selected by SPA to quantify vegetable oil in
diesel/biodiesel blends using the FT-NIR spectrophotometer
(Fig. 3b).
The results obtained using the MicroNIR spectrophotometer to
quantify the vegetable oil content are comparable to those
obtained by Vasconcelos et al. [1] (RMSEP = 0.45% v/v), who
employed the same spectral region, vegetable oil content from 0
to 10% v/v, MLR-SPA model, and spectra measured with a benchtop
spectrophotometer. Other works [2,3] also have described the
determination of vegetable oil in diesel/biodiesel blends using
NIR benchtop spectrophotometers but reporting the results as per-
centage of weight (% w/w), which makes it difficult to compare
with our results because we did not know the specific masses of
the diesel/biodiesel blends.
Fig. 5 shows plots of the predicted versus reference values for
vegetable oil content determination in blends using PLS-JK and
MLR-SPA models with both spectrophotometers. No evidence of
systematic error is observed according to the t test at a confidence
level of 95% and the points are randomly distributed close to the
bisectrix line and along the whole range of y-values.
The better performance of models when the FT-NIR instrument
was used for the quantification of oil content in diesel/biodiesel
blends can be explained by its higher spectral resolution and high
signal-to-noise ratio compared with the MicroNIR. The oil spec-
trum is very similar to the ester resulting from biodiesel conver-
sion process. On the other hand, the results attained by using the
MicroNIR are not so different from those of the FT-NIR, being suit-
able to quantify vegetable oil in diesel/biodiesel blends.
Furthermore, the MicroNIR is a less expensive portable spec-
trophotometer, which allows in field measurements and fast
screening of samples.
Finally, Limits of detection (LOD) for both biodiesel (Table 1)
and vegetable oil (Table 2) were also calculated. As can be
observed, the lowest LOD values were obtained using the
MLR-SPA models. For biodiesel quantification, the LOD values
obtained were 0.31% v/v and 0.22% v/v using MicroNIR and
FT-NIR, respectively. On the other hand, for vegetable oil determi-
nation, LOD values of 0.40% v/v (MicroNIR) and 0.34% v/v (FT-NIR)
were obtained. These results pointed to the comparable perfor-
mance of both instruments in the detection of biodiesel and veg-
etable oil in diesel blends.
 E.M. Paiva et al. / Fuel 160 (2015) 57–63
4. Conclusion
This work evaluates the performance of a commercial portable
spectrophotometer (MicroNIR) to quantify the biodiesel and veg-
etable oil content in diesel/biodiesel blends. A customized compact
configuration employed to acquire transmittance spectra from liq-
uids using the MicroNIR was developed. The resulting spectropho-
tometer employs an external tungsten mini-lamp as radiation
source. Because the bulb is powered, as is the MicroNIR, by an
USB port of a laptop, for example, the setup remains portable.
With this compact system, the quantifications of biodiesel and veg-
etable oil in diesel/biodiesel blends were accomplished using PLS,
PLS-JK and MLR-SPA models.
For the determination of biodiesel content, we demonstrated
that the adapted MicroNIR provides similar performance to a
FT-NIR benchtop spectrophotomer when using PLS or PLS-JK mod-
els. On the other hand, for quantification of vegetable oil, the
MicroNIR results were similar for FT-NIR and only the with
PLS-JK model. For both, the determination of biodiesel and veg-
etable oil content in diesel/biodiesel blends, using MicroNIR and
FT-NIR, the results point to MLR-SPA as the best modelling strat-
egy. Although the MicroNIR shows an overall predictive ability
worse than the FT-NIR benchtop spectrophotometer, even when
using the best modeling strategy (MLR-SPA), the setup is more
appropriate for monitoring the quality of diesel blends in-field, at
points of sale, since it is compact, robust, fast, and less expensive.
Acknowledgments
The authors are grateful to FINEP, FACEPE/NUQAAPE
(APQ-0346-1.06/14), CNPq/FAPESP/INCTAA (Proc. No.
573894/2008-6, FAPESP 2008/57808-1) and LAC/UFPE for financial
support. Research fellowships and scholarship granted by CNPq
and CAPES are also gratefully acknowledged. The English version
was revised by Sidney Pratt, Canadian, BA, MAT (The Johns
Hopkins University), RSAdip (TEFL) (Cambridge University).
References
[1] Vasconcelos FVC, Souza Jr PFB, Pimentel MF, Pontes MJC, Pereira CF. Using
near-infrared overtone regions to determine biodiesel content and
adulteration of diesel/biodiesel blends with vegetable oils. Anal Chim Acta
2012;716:101–7.
[2] Oliveira FCC, Brandão CRR, Ramalho HF, Costa LAF, Suarez PAZ, Rubim JC.
Adulteration of diesel/biodiesel blends by vegetable oil as determined by
Fourier transform (FT) near infrared spectrometry and FT-Raman
spectroscopy. Anal Chim Acta 2007;587:194–9.
[3] Gaydou V, Kister J, Dupuy N. Evaluation of multiblock NIR/MIR PLS predictive
models to detect adulteration of diesel/biodiesel blends by vegetal oil.
Chemometr Intell Lab Syst 2011;106:190–7.
[4] EN 14078. Liquid petroleum products – determination of fatty methyl ester
(FAME) content in middle distillates – infrared spectrometry method.
European Standard; 2009.
[5] ASTM D7371. Standard test method for determination of biodiesel (fatty acid
methyl esters) content in diesel fuel oil using mid infrared spectroscopy (FTIR-
ATR-PLS Method). West Conshohocken (PA): ASTM International; 2014.
[6] Brito LR, Silva MPF, Rohweeder JJR, Pasquini C, Honorato FA, Pimentel MF.
Determination of detergent and dispensant additives in gasoline by ring-oven
and near infrared hypespectral imaging. Anal Chim Acta 2015;863:9–19.
[7] Pontes MJC, Rocha AMJ, Pimentel MF, Pereira CF. Determining the quality of
insulating oils using near infrared spectroscopy and wavelength selection.
Microchem J 2011;98:254–9.
63
[8] Lima SM, Silva BFA, Pontes DV, Pereira CF, Stragevitch L, Pimentel MF. In-line
monitoring of the transesterification reactions for biodiesel production using
NIR spectroscopy. Fuel 2014;115:46–53.
[9] Alves JCL, Henriques CB, Poppi RJ. Determination of diesel quality parameters
using support vector regression and near infrared spectroscopy for an in-line
blending optimizer system. Fuel 2012;97:710–7.
[10] Pimentel MF, Ribeiro GMGS, Cruz RS, Stragevitch L, Pacheco Filho JGA, Teixeira
LSG. Determination of biodiesel content when blended with mineral diesel
fuel using infrared spectroscopy and multivariate calibration. Microchem J
2006;82:201–6.
[11] Oliveira JS, Montalvão R, Daher L, Suarez P, Rubim JC. Determination of methyl
ester contents in biodiesel blends by FTIR-ATR and FTNIR spectroscopies.
Talanta 2006;69:1278–84.
[12] Fernandes DDS, Gomes AA, Costa GB, Silva GWB, Véras G. Determination of
biodiesel content in biodiesel/diesel blends using NIR and visible spectroscopy
with variable selection. Talanta 2011;87:30–4.
[13] Filgueiras PR, Alves JCL, Poppi RJ. Quantification of animal fat biodiesel in
soybean biodiesel and B20 diesel blends using near infrared spectroscopy and
synergy interval support vector regression. Talanta 2014;119:582–9.
[14] Alves JCL, Poppi RJ. Simultaneous determination of hydrocarbon renewable
diesel, biodiesel and petroleum diesel contents in diesel fuel blends using near
infrared (NIR) spectroscopy and chemometrics. Analyst 2013;138:6477–87.
[15] Pasquini C. Near infrared spectroscopy: fundamentals, practical aspects and
analytical applications. J Braz Chem Soc 2003;14(2):198–219.
[16] Chase B. Fourier transform near-infrared Raman spectroscopy. In: Chalmers
JM, Griffiths PR, editors. Handbook of vibrational spectroscopy 1: theory and
instrumentation. Chichester: J. Wiley & Sons; 2002. p. 522–33.
[17] Gonzaga FB, Pasquini C. A low cost short wave near infrared
spectrophotometer: application for determination of quality parameters of
diesel fuel. Anal Chim Acta 2010;670:92–7.
[18] Friedrich DM, Hulse CA, von Gunten MK, Williamson EP, Pederson CG, O’Brien
NA. Miniature near-infrared spectrometer for point-of-use chemical analysis.
In: Soskind YG, Olson C, editors. Photonic instrumentation engineering. San
Francisco: Proc SPIE; 2014. p. 1–11.
[19] O’Brien NA, Hulse CA, Friedrich DM, Van Milligen FJ, von Gunter MK, Pfeifer F,
et al. Miniature near-infrared (NIR) spectrometer engine for handheld
applications. In: Druy MA, Crocombe RA, editors. Next-generation
spectroscopic technologies V. Baltimore: Proc SPIE; 2012. p. 1–8.
[20] Alcalà M, Blanco M, Moyano D, Broad NW, O’Brien NA, Friedrich DM, et al.
Qualitative and quantitative pharmaceutical analysis with a novel hand-held
miniature near infrared spectrometer. J Near Infrared Spectrosc 2013;21:
445–57.
[21] O’Brien NA, Hulse CA, Pfeifer F, Siesler HW. Near infrared spectroscopic
authentication of seafood. J Near Infrared Spectrosc 2013;21:299–305.
[22] Singh R, Sahay A, Karry KM, Muzzio F, Ierapetritou M, Ramachandran R.
Implementation of an advanced hybrid MPC–PID control system using PAT
tools into a direct compaction continuous pharmaceutical tablet
manufacturing pilot plant. Int J Pharm 2014;473:38–54.
[23] Singh R, Sahay A, Muzzio F, Ierapetritou M, Ramachandran R. A systematic
framework for onsite design and implementation of a control system in a
continuous tablet manufacturing process. Comput Chem Eng
2014;66:186–200.
[24] Rohwedder JJR, Pasquini C, Fortes PR, Raimundo IM, Wilk A, Mizaikoft B.
IHWG-mNIR: a miniaturised near-infrared gas sensor based on substrate-
integrated hollow waveguides coupled to a micro-NIR spectrophotometer.
Analyst 2014;139:3572–6.
[25] Lutz OMD, Bonn GK, Rode BM, Huck CW. Reproducible quantification of
ethanol in gasoline via a customized mobile near-infrared spectrometer. Anal
Chim Acta 2014;826:61–8.
[26] Galvão RKH, Araújo MCU, José GE, Pontes MJC, Silva EC, Saldanha TCB. A
method for calibration and validation subset partitioning. Talanta 2005;67:
736–40.
[27] Silva VH, Rebouças MV, Salles AR, Pimentel MF, Pontes MJC, Pasquini C.
Determination of naphtha composition by near infrared spectroscopy and
multivariate regression to control steam cracker processes. Fuel Process
Technol 2015;131:230–7.
[28] Araújo MCU, Saldanha TCB, Galvão RKH, Yoneyama T, Chame HC, Visani V. The
successive projections algorithm for variable selection in spectroscopic
multicomponent analysis. Chemometr Intell Lab Syst 2001;57:65–73.
[29] Karaman I, Qannari EM, Martens H, Hedemann MS, Knudsen KEB, Kohler A.
Comparison of Sparse and Jack-knife partial least squares regression methods
for variable selection. Chemometr Intell Lab Syst 2013;122:65–77.
[30] ASTM E 1655-05. Standard practices for infrared multivariate quantitative
analysis. West Conshohocken (PA): ASTM International; 2012.
[31] Workman J, Weyer L. Practical guide to interpretative near-infrared
spectroscopy. 1st ed. New York: CRC Press; 2008.