Escolar Documentos
Profissional Documentos
Cultura Documentos
R-CdO
and
C-C
) and 372 cm
-1
(assigned as
CO-O-C
).
28
Raman spectra present well-defined bands and less congestion
than FTIR and NIR spectra, and as a consequence, there is a
smaller possibility of band overlapping, making quantitative
measurements simpler.
51,52
As mentioned earlier, in our previous work,
28
FT-Raman/
PLS methods showed good prediction capabilities when applied
to known samples treated as unknown. A data set of 63 spectra
from 21 standard soybean oil/ethyl ester mixtures (triplicate
measurements) was used, and from the numerous correlations
and regression methods tested for analysis, FT-Raman/PLS
methods at the CH stretching region (3100-2740 cm
-1
, mainly
the peak at 2932 cm
-1
) could be distinguished from the others
due to their predicted capabilities: (i) R
2
) 0.9994 for the model
using 21 samples averaged from 63 measurements as a data set
(model 4, notation used in ref 28) and (ii) R
2
) 0.9993 for the
model using all 63 measurements as a data set (model 5). A
detailed description of both methods was given elsewhere,
28
and
no further discussion will be mentioned here. The two models
were applied to the heterogeneous transesterification reaction,
and similar results were obtained. A plot of model 4 predicted
values versus model 5 predicted values results in an R
2
)
0.9976. However, when compared to the predicted values found
by
1
H NMR methods, model 4 was slightly better then model
5. Although multivariate analysis generally requires a large
number of samples to build reliable calibration models (at least
20 samples for each component), the articles of Knothe
3,27
using
NIR/PLS, and Zagonel et al.
13
using FTIR/PLS, were success-
fully applied to progressive transesterification reactions with
smaller data sets in their models. In our case, the effects of
increased data sets on FT-Raman/PLS models led to increased
predicted capabilities and linearity (see Table 2).
3.3. Correlation Between FT-Raman/PLS and
1
H NMR
Methods. In preceding articles dealing with spectroscopic
measurements of biodiesel content in transesterification reac-
(50) Jie, M. S. F. L. K.; Lam, C. C. Chem. Phys. Lipids 1995, 77, 155-
171.
(51) Baranska, H.; budzinska, A.; Terpinski, J. Laser Raman Spectros-
copy: Analytical Applications; Ellis Horwood Limited: Chichester, U.K.,
1987; pp 151-168.
(52) Skoog, D. A.; Holler, F. J.; Nieman, T. A. Principles of Instrumental
Analysis; Saunders College Publishing: Philadelphia, PA, 1997; pp 429-
443.
Table 1. Biodiesel True Concentration Values and Those Predicted
by Eq 2 with and without Assumption of Six Standard Soybean
Oil/Ethyl Esters Mixtures
true CEE
values (wt %)
predicted CEE
values (wt %)
without assumption
a
predicted CEE
values (wt %)
with assumption
b
0.00 0.00 0.63
20.45 20.64 21.19
40.39 42.46 42.90
59.85 60.62 60.94
79.62 76.80 77.00
100.00 99.40 99.41
a
CEE values calculated from eq 2 without assuming that glyceryl
methylenic hydrogens areas at 4.25-4.35 and 4.10-4.20 ppm are identical.
b
CEE values calculated from eq 2 assuming that glyceryl methylenic
hydrogens areas at 4.25-4.35 and 4.10-4.20 ppm are identical.
Table 2. Predicted CEE Values by Raman/PLS Models with
Increased Data Set for Standard Soybean Oil/Ethyl Ester Mixtures
of Known Composition Treated as Unknown
predicted CEE values (wt %)
true CEE
values (wt %)
a
6 samples
data set size
(R
2
) 0.9990)
11 samples
data set size
(R
2
) 0.9991)
21 samples
data set size
(R
2
) 0.9997)
0.00 3.73 3.71 0.52
a
20.01 25.91 25.72 21.88
a
49.82 51.23 51.34 50.28
a
79.69 80.52 80.42 80.64
a
100.00 99.64 99.18 99.31
a
Values obtained from ref 28.
FT-Raman Spectroscopy Quantification of Biodiesel Energy & Fuels, Vol. 21, No. 5, 2007 2479
tions, the results obtained were correlated with other techniques
to validate their values. Knothe
27
correlated NIR spectroscopy
and two
1
H NMR equations with good agreement. Neto and
co-workers
14
correlated their
1
H NMR analytical curves with
viscosity measurements, obtaining a linear correlation coefficient
of 0.9981. Zagonel and co-workers
13
used GPC to validate their
FTIR/PLS methods, obtaining an R
2
) 0.9837 between both
techniques. Figure 5 displays FT-Raman/PLS (model 4) and
1
H NMR methods applied to a progressive heterogeneous
transesterification reaction. The results obtained from all four
methods (FT-Raman/PLS and three
1
H NMR methods) showed
good correlation (Figure 5) with each other, indicating their
mutual validation. Indeed, the linear fit between FT-Raman
measurements and all NMR methods tested (i.e., analytical
curve, eq 1, and eq 2) showed R
2
values of 0.9974, 0.9847, and
0.9972, respectively (Figure 6). The difference observed between
biodiesel yield values from eq 1 and the other methods after 2
h of reaction is associated with integration difficulties produced
by signal superimposition from glyceridic intermediates (MAGs
and DAGs) and R-acyl CH
2
.
Although both methods presented in this work do not attend
the detection limits required by ASTM to determine low levels
of contaminants in biodiesel fuel as chromatographic methods
do, spectroscopic techniques allows a faster analysis of the
reaction by easily identifying the formation of peaks or bands
related to ethyl ester products. These methods, when coupled
to multivariate analysis, can be applied to fast reversible
reactions, such as transesterification,
3,13,14,26,27,32,46
to evaluate
a new catalyst potentiality to produce biodiesel. In conclusion,
FT-Raman/PLS models are an attractive alternative to rapidly
monitor biodiesel synthesis due to their inherent qualities and
their reduced costs when compared with other techniques,
including NMR. However,
1
H NMR methods, especially when
involving simple equations, are more easily adapted to other
vegetable oils methanolysis or ethanolysis reactions. Both
procedures described here were successfully correlated with each
other and validate by reported methods, evidencing and increas-
ing their potential for biodiesel synthesis and quality monitoring.
Acknowledgment. Acknowledgement is given to CNPq, CAPES/
PQI, FINATEC, Finep-CTPetro, Finep-CTInfra 970/01, UnB-IQ,
UnB-IG, and FAPDF/SCDT/CNPq.
EF060657R
Figure 5. Biodiesel yield from a heterogeneous transesterification
reaction determined by four methods:
1
H NMR analytical curve, eq 1,
eq 2, and FT-Raman/PLS model.
Figure 6. Plot from FT-Raman/PLS model predicted reaction values
vs
1
H NMR analytical curve, eq 1, and eq 2 predicted reaction values
(R
2
) 0.9974, 0.9847, and 0.9972, respectively).
2480 Energy & Fuels, Vol. 21, No. 5, 2007 Ghesti et al.