- Email: [email protected]
Multivariate monitoring of soybean oil ethanolysis by FTIR
Talanta 63 (2004) 1021–1025
Multivariate monitoring of soybean oil ethanolysis by FTIR Giuliano F. Zagonel, Patricio Peralta-Zamora, Luiz P. Ramos∗ Department of Chemistry, Research Center in Applied Chemistry (CEPESQ), Federal University of Paraná, P.O. Box 19081, 81531-990 Curitiba, PR, Brazil Received 22 July 2003; received in revised form 5 December 2003; accepted 14 January 2004 Available online 17 March 2004
Abstract In this work, an analytical procedure was developed to monitor the ethanolysis of degummed soybean oil (DSO) using Fourier-transformed mid-infrared spectroscopy (FTIR) and methods of multivariate analysis such as principal component analysis (PCA) and partial least squares regression (PLS). The triglycerides (reagents) and ethyl esters (products) involved in ethanolysis were shown to have similar FTIR spectra. However, when the FTIR spectra derived from seven standard mixtures of triolein and ethyl oleate were treated by PCA at the region that represents the C=O stretching vibration of ester groups (1700–1800 cm−1 ), only two principal components (PC) were shown to capture 99.95% of the total spectral variance (92.37% for the former and 7.58% for the latter PC). This observation supported the development of a multivariate calibration model that was based on the PLS regression of the FTIR data. The prevision capability of this model was measured against 40 reaction aliquots whose ester content was previously determined by size exclusion chromatography. Only small discrepancies were observed when the two experimental data sets were treated by linear regression (R2 = 0.9837) and these deviations were attributed to the occurrence of non-modeled transient species in the reaction mixture (reaction intermediates), particularly at short reaction times. Therefore, the FTIR/PLS model was shown to be a fast and accurate method to predict reaction yields and to follow the in situ kinetics of soybean oil ethanolysis. © 2004 Elsevier B.V. All rights reserved. Keywords: Degummed soybean oil; Transesterification; Ethyl esters; FTIR; Multivariate calibration; PCA; PLS
1. Introduction In recent years, the concept of producing biodiesel from renewable lipid sources regained international attention. This tendency led the Brazilian Government to establish a national program (Probiodiesel) whose mission is to evaluate the technical, economic and environmental competitiveness of biodiesel in relation to the commercially available diesel oil. As a result, several research projects have been initiated nationwide to investigate and/or optimize biodiesel production from soybean oil and ethanol derived from sugarcane (ethyl esters), the only two raw materials that are considered available to support the initial activities of the national program. By definition, biodiesel (mono-alkyl esters of long chain fatty acids) is a diesel oil substitute that can be produced from renewable sources such as vegetable oils, animal fats and recycled cooking oils [1,2]. This alternative fuel is ∗
Corresponding author. Tel.: +55-41-3613175; fax: +55-41-3613186. E-mail address: [email protected] (L.P. Ramos).
0039-9140/$ – see front matter © 2004 Elsevier B.V. All rights reserved. doi:10.1016/j.talanta.2004.01.008
widely recognized as ecologically friendly, biodegradable, not hazardous for handling (flash point above 110 ◦ C) and of superior lubricity, providing a good performance and low emission profile upon combustion in compression ignition engines. Other advantages linked to biodiesel are that it has little (0.001%) or no sulfur content and it can be used in blends or as a neat fuel, without further investment in the current fleet [2,3]. Ethanolysis is a reversible reaction that is typically performed under alkaline conditions: vegetable oils or animal fats are reacted at room temperature with anhydrous ethanol in the presence of sodium or potassium hydroxide [3,4]. To ensure good fuel properties, the reaction must be carried out under optimal conditions because even a low contamination with unreacted triglycerides, glycerin or reaction intermediates may cause damage to the engine and elicit emissions of soot and harmful compounds such as acrolein [5]. Therefore, biodiesel fuel quality must be ensured by solid analytical methods such as capillary gas chromatography (CGC) [6,7], high performance liquid chromatography (HPLC) [8], size exclusion chromatography (SEC) [4] or nuclear
1022
G.F. Zagonel et al. / Talanta 63 (2004) 1021–1025
magnetic resonance (NMR) [9]. However, despite their high accuracy, these methods are time-consuming and can not be easily used for the in situ evaluation of fast reversible reactions such as transesterification. Fourier-transformed infrared (FTIR) spectroscopy has been reported as a fast and accurate method to monitor the methanolysis of vegetable oils [9–11]. However, in complex reaction mixtures like these, classical univariated calibration models are impracticable because serious spectral interferences (or overlapped signals) may result in non-linear correlations between the measured signal and the property of interest. To circumvent this problem, the application of multivariate analysis, together with the development of new methods for the rapid acquisition of multisignal spectra, allowed for the exploitation of the entirety of the analytically relevant information found in FTIR [12,13]. During the last decade, many authors have used multivariate regression to develop calibration models based on FTIR. Ferraz et al. [14] used partial least squares regression (PLS) to correlate the FTIR data with changes in wood composition during chemical or physical processing. Che Man and Setiowaty [15] described a simple FTIR/PLS calibration model to determine free fatty acids in palmolein. Lendl and Schindler [16] developed a FTIR flow-through sensor based on PLS for the simultaneous determination of several organic acids in aqueous solution. Garc´ıa-Menc´ıa et al. [17] developed a predictive model based on FTIR and PLS to evaluate the octane number of naphthas in a petrochemical refinery. Infrared spectroscopy has also been extensively investigated as a suitable method for the on-line monitoring of chemicals in complex mixtures. Fiber-optic probes have been developed to operate in the mid- [10,18] and near-infrared [11,19,20] regions and multivariate calibration methods have been applied to follow reaction kinetics as well as to establish good parameters for quality control in industrial processes. Knothe [19,20] used near-infrared spectroscopy and PLS to develop a model that was able to predict reaction yields during the methanolysis of soybean oil and this same method was proved useful to determine biodiesel fuel quality, regardless of their relative inadequacy to detect trace amounts of biodiesel contaminants. Other authors have also used the same principle to predict the transesterification end-point of methanolysis [11]. Although a variety of applications have been developed with this method, the routine use of near-infrared spectroscopy in industrial facilities remains limited by the cost of the hardware that is required for remote sensing. Under this viewpoint, analytical methods based on mid-infrared spectroscopy could reduce as much as 50% of the capital cost associated with this application. However, mid-infrared spectroscopy cannot be easily applied to analyze complex mixtures because peak overlapping usually compromises the spectral resolution. Multivariate calibration methods have also been used to simplify the complex matrix of chemical data that is gen-
erated by mid-infrared spectroscopy. Acha et al. [21,22] reported the use of several PLS calibration models for the on-line monitoring of a dechlorination process using a novel ATR mid-infrared sensor. Sadeghi-Jorabchi et al. [10] applied FTIR spectroscopy combined with an ATR mid-infrared fiber-optic probe to estimate, by univariate calibration, the occurrence of biodiesel contamination in lubricating oils. Küpper et al. [18] used silver halide fiber-optic probes to evaluate adulteration of extra virgin olive oils by ATR mid-infrared spectroscopy and PLS regression. More recently, Heise et al. [23] developed novel infrared optical probes for process monitoring based on inert infrared-transparent silver halide fibers. Progress in the quality of these fibers, including their extrusion with different cross-sections, enabled the construction of flexible fiber-optic probes of different geometries that are particularly suitable for remote sensing. In this work, an analytical method was developed to monitor the transesterification kinetics of degummed soybean oil (DSO) using FTIR and a multivariate approach. Our working hypothesis was that FTIR could be used as a simple and rapid analytical tool to determine the reaction yield of processes such as the ethanolysis of vegetable oils.
2. Experimental 2.1. Material The batch of degummed soybean oil used in this study was donated by IMCOPA (Indústria e Comércio de Óleos Vegetais Ltda., Araucária, PR, Brazil), whereas a commercial stock of anhydrous ethanol was kindly obtained from ALCOPAR (Associação dos Produtores de Álcool do Estado do Paraná, Maringá, PR, Brazil). Other reagents, as well as analytical standards of monoolein (1-mono[cis-9-octadecenoyl]-rac-glycerol), diolein (mixed isomers containing 85% of 1,3-di[cis-9-octadecenoyl]glycerol and 15% of 1,2-di[cis9-octadecenoyl]-rac-glycerol), triolein (1,2,3-tri[cis-9-octadecenoyl]glycerol) and ethyl oleate (cis-9-octadecenoic acid ethyl ester), were all of analytical grade (Sigma, St. Louis, MO). 2.2. Synthesis Ethyl esters of degummed soybean oil were obtained according to Zagonel [2]. Batches of DSO, containing 0.34% (w/w) of free fatty acids and 0.09% (w/w) of moisture, were placed in round-bottomed flasks and heated up to 40 ◦ C before ethanol was added to the mixture. A pre-heated ethanol solution, containing 1% (w/w) of potassium hydroxide in relation the oil mass, was then added to the vessel and the reaction was carried out for 1 h. Aliquots of 2 ml were withdrawn from the reaction mixtures at varied reaction times (e.g., 1, 10, 20, 40, and 60 min) and treated with 4 ml of an aqueous solution of ammonium chloride (5%, w/v). The mixture was
G.F. Zagonel et al. / Talanta 63 (2004) 1021–1025
centrifuged for 10 min at 11500 g to ensure good phase separation. The upper organic phase was withdrawn, washed with water, dried against anhydrous sodium sulfate (Na2 SO4 ) and kept in the dark for further analysis by SEC and FTIR. 2.3. Methods SEC of reaction aliquots was carried out using a Shimadzu LC10AD workstation provided with a SIL10A autosampler and detection by differential refractometry (RID10A) [2,4]. Each sample was solubilized in tetrahydrofuran (THF), filtered through a Teflon membrane with a pore size of 0.45 m and analyzed using a series of one pre-column and two Tosoh TSK-GEL columns (TSK 2000HXL and TSK 1000HXL). THF was used as the eluting solvent at a flow rate of 1 ml min−1 . Three analyses were carried out for each replicate and calibration was based on injection of external standards including ethyl oleate and mono-, di-, and triolein. A typical relative standard deviation of 1.6% was obtained throughout the calibration range used in this study. Standards (3–30 mg ml−1 ) and reaction aliquots were also analyzed by Fourier-transformed infrared spectroscopy using a Reagen KBr cell for liquid samples [2]. The FTIR spectra were collected using a Bomem FTIR MB-100 spectrophotometer in the 500–4000 cm−1 region, with a resolution of 4 cm−1 , 32 scans and triangular apodization.
1023
3. Results and discussion In the ethanolysis of vegetable oils, the FTIR spectra of reagents and products are very similar because of the high chemical similarity that exists among triglycerides and ethyl esters (Fig. 1). However, small differences can be clearly observed in two regions: one broader region covering wave numbers from 900 to 1500 cm−1 and another region, around 1730 cm−1 , that includes the stretching vibrations of carbonyl groups (Fig. 2). Changes in the latter region were readily attributed to the extent of glycerol substitution in fatty acids by ethoxy radicals (ethanolysis), whereas the former region was more complex and displayed a series of overlapped signals that were likely to interfere with the development of the model. Based on the evidence that soybean oil ethanolysis would impart changes in the FTIR spectra of the reaction mixture, a multivariate method was developed to allow the in situ evaluation of reaction yields. Initially, exploratory studies were carried out by PCA using seven standard mixtures containing triolein and ethyl oleate (Table 1). The entire spectrum was divided in nine regions, each of them covering an
2.4. Multivariate analysis The entire FTIR spectral range of seven standard mixtures, containing different proportions of triolein (representing triglycerides found in vegetable oils) and ethyl oleate (representing ethyl esters produced as a result of transesterification) (Table 1), was treated by principal component analysis (PCA). The region with the greatest spectral differentiation was identified and subsequently used to develop a multivariate calibration model based on partial least squares regression. Data analysis was carried out in a MATLAB 4.0 environment, whereas both PCA and PLS procedures were performed using the PLS-toolbox 1.5 [24]. The FTIR spectra were mean-centered to compensate for any baseline dislocation and the spectral data were processed only at the region where the best correlation could be obtained.
Fig. 1. FTIR spectra of degummed soybean oil and its corresponding ethyl ester from 800 to 3200 cm−1 .
Table 1 Standard mixtures used in the multivariate analysis Sample
Triolein (%)
´ Ester (%)
1 2 3 4 5 6 7
0.00 9.95 29.88 49.85 69.88 89.95 100.00
100.00 90.05 70.12 50.15 30.12 10.05 0.00
Fig. 2. FTIR spectra of degummed soybean oil and its corresponding ethyl ester from 1600 to 1900 cm−1 .
1024
G.F. Zagonel et al. / Talanta 63 (2004) 1021–1025 Table 2 Predictive capability of the multivariate calibration model Samplea
1 2 3 4 5 6 7 8 9
Real
Predicted
Triolein (%)
Ester (%)
Triolein (%)
Ester (%)
0.00 9.95 29.88 49.85 69.88 89.95 100.00 19.91 8.52
100.00 90.05 70.12 50.15 30.12 10.05 0.00 80.09 91.48
0.60 8.91 30.07 50.19 70.12 89.68 99.92 19.27 8.15
99.40 91.09 69.93 49.81 29.88 10.32 0.08 80.73 91.85
a Calibration was performed with samples 1–7, whereas samples 8 and 9 were used to predict the accuracy of the method (provision samples).
Fig. 3. Scores (A) and loadings (B) of the PCA.
interval of 200 cm−1 , and these data sets were individually treated by PCA. A graph of scores was obtained for each of the data intervals and these graphs were used to identify the region in which the transesterification of soybean oil could be better represented. The best result was obtained from the spectral region ranging from 1700 to 1800 cm−1 . By running a PCA in this region, only two principal components were shown to capture 99.95% of the variance, being 92.37% for PC1 and 7.58% for PC2. That is, two “new vectors” were enough to represent the multidimensional space denoted by the entire spectral data set, which consisted of a linear combination of 54 values of frequency. The graph of scores obtained for the 1700–1800 cm−1 region is shown in Fig. 3A. This graph demonstrated that the stretching vibration of carbonyl groups in esters was the key parameter for the differentiation among the seven calibration standards, which contained different amounts of triolein and ethyl oleate. Fig. 3B shows the graph of loadings where the weight of each original variable (or frequency) is represented in relation to the establishment of the PCs. These results confirmed the previous observation that the dislocation of the carbonyl signal, between the variables 12 and 32, was responsible for the spectral differentiation that occurred as a result of a gradual increase in triolein concentration. That is, for the seven standard mixtures described in Table 1, there was a gradual migration of the C=O stretching vibration from a maximum at 1746 cm−1 (νC=O in triglycerides) to a maxi-
mum at 1735 cm−1 (νC=O in ethyl esters), passing through a minimum at an average wave number of 1740 cm−1 (Fig. 2). Therefore, the individual scores of PC1 and PC2 indicated that progressive changes in triolein concentration (see sequence 1–7 in Table 1) could be measured by changes the C=O stretching vibration in the 1700–1800 cm−1 region. As mentioned above, the triolein-to-ester ratio was adequately reproduced by the spectral data, and only two latent variables (LV) (or two PCs) were responsible for capturing 99.95% of the variance. Based on these evidences, a multivariate calibration model was developed by partial least squares regression using the region previously established by PCA. This model was built with the entire calibration set described in Table 1 plus two independent samples used as the provision objects (Table 2). Only two latent variables were shown to predict 99.89% of the total variance of the x-block (99.03% for the first and 0.86% for the second LV) and 99.98% of the total variance of the y-block (99.96% for the first and 0.03% for the second LV). Therefore, the model was proved useful to predict changes in triolein-to-ester ratios based on the spectral variance of the 1700–1800 cm−1 region. The FTIR/PLS calibration model, developed from the seven synthetic mixtures described in Table 1, was then used to predict the ester yield of a new experimental data set related to the kinectic study of degummed soybean oil (DSO) ethanolysis. The reactions were carried out under the conditions used by Zagonel and colleagues [2,4] for the transesterification of DSO and the results obtained through the model were compared to reaction yields originally determined by size exclusion chromatography [4]. By SEC, the ester concentration is evaluated from a signal absolutely free of interference, while through FTIR, this parameter is obtained from an analytical response of much greater complexity. A total of 40 reaction aliquots, collected at different time intervals from eight independent experiments, were applied to validate the model (Fig. 4). In general, there was a good correlation between both data sets (R2 = 0.9837), determined independently by SEC and FTIR/PLS. However, small discrepancies were observed along the linear regres-
G.F. Zagonel et al. / Talanta 63 (2004) 1021–1025
1025
versible reactions such as the transesterification of vegetable oils.
Acknowledgements The authors are grateful to Dr. Martin Mittelbach for providing us analytical standards of monoolein, diolein, triolein and ethyl oleate, as well as to the financial support of CNPq, CAPES, Funpar and Fundação Araucária.
References Fig. 4. Correlation between reaction yields obtained by SEC and through the FTIR/PLS model.
sion, particularly at low conversion yields. This apparent uncertainty of the model was attributed to the interference caused by unmodeled reaction intermediates (unreacted mono- and diglycerides), which are particularly important at short reaction times. There was a considerable simplification of the matrix towards the end of the reaction. Therefore, deviations at short reaction times were expected due to the occurrence of a greater amount of unmodeled reaction intermediates. However, the FTIR/PLS model was built to predict more advanced reaction yields of DSO ethanolysis, which is what really matters to indicate the time in which the reaction should be stopped. Therefore, Fig. 4 indicates that the model was quite reliable for predicting reaction yields at advanced reaction times in which conversions greater than 30% were obtained.
4. Conclusions The FTIR/PLS calibration developed in this study was proven perfectly suitable as an analytical method to predict reaction yields during the ethanolysis of degummed soybean oil. Even though small discrepancies were observed between the chromatographic and the FTIR/PLS data sets, particularly at very short reaction times, the inherent advantages of the infrared spectroscopy, such as simplicity, high analytical velocity, low-cost, and facility for the implementation of on-line monitoring systems, suggest this method as a powerful analytical procedure for the evaluation of fast re-
[1] U. Schuchardt, R. Sercheli and, R.M. Vargas, J. Braz. Chem. Soc. 9 (1998) 199. [2] G.F. Zagonel, M.Sc. Dissertation, Federal University of Paraná, Curitiba, Brazil, 2000. [3] G.F. Zagonel, L.P. Ramos, Rev. Quim. Ind. 717 (2001) 17. [4] G.F. Zagonel, P.G. Peralta-Zamora, L.P. Ramos, Sci. Eng. J. 1 (2003) 35. [5] D.Y.Z. Chang, J.H. Van Gerpen, I. Lee, L.A. Johnson, E.G. Hammond, S.J. Marley, J. Am. Oil Chem. Soc. 73 (1996) 1549. [6] C. Plank, E. Lorbeer, J. Chromatogr. A 697 (1995) 461. [7] M. Mittelbach, G. Roth, A. Bergmann, Chromatographia 42 (1996) 431. [8] B. Trathnigg, M. Mittelbach, J. Liq. Chromatogr. 13 (1990) 95. [9] G. Gelbard, O. Brès, R.M. Vargas, F. Vielfaure, U.F. Schuchardt, J. Am. Oil Chem. Soc. 72 (1995) 1239. [10] H. Sadeghi-Jorabchi, V.M.E. Wood, F. Jeffery, A. Bruster-Davies, N. Loh, D. Coombs, Spectrosc. Eur. 6 (1994) 16. [11] W.D. Mockel, M.P. Thomas, Proc. SPIE - Int. Soc. Opt. Eng. 1681 (1992) 220. [12] C.E. Miller, Chemom. Intell. Lab. Syst. 30 (1996) 11. [13] H. Martens, T. Naes, Multivariate Calibration, John Wiley, Chichester, 1993. [14] A. Ferraz, J. Baeza, J. Rodriguez, J. Freer, Bioresour. Technol. 74 (2000) 201. [15] Y.B. Che Man, G. Setiowaty, Food Chem. 66 (1999) 109. [16] B. Lendl, R. Schindler, Vib. Spectrosc. 19 (1999) 1. [17] M.V. Garc´ıa-Menc´ıa, J.M. Andrade, P. López-Mah´ıa, D. Prada, Fuel 79 (2000) 1823. [18] L. Küpper, H.M. Heise, P. Lampen, A.N. Davies, P. McIntyre, Appl. Spectrosc. 55 (2001) 563. [19] K. Knothe, J. Am. Oil Chem. Soc. 76 (1999) 795. [20] K. Knothe, J. Am. Oil Chem. Soc. 77 (2000) 489. [21] V. Acha, M. Meurens, H. Naveau, S.N. Agathos, Biotechnol. Bioeng. 68 (2000) 473. [22] V. Acha, M. Meurens, H. Naveau, S.N. Agathos, Water Sci. Technol. 40 (1999) 41. [23] H.M. Heise, L. Küpper, L.N. Butvina, Anal. Bioanal. Chem. 375 (2003) 1116. [24] B.M. Wise, N.B. Gallagher, PLS-Toolbox Version 1.5, Eigenvector Technologies, 1995.
Multivariate monitoring of soybean oil ethanolysis by FTIR Giuliano F. Zagonel, Patricio Peralta-Zamora, Luiz P. Ramos∗ Department of Chemistry, Research Center in Applied Chemistry (CEPESQ), Federal University of Paraná, P.O. Box 19081, 81531-990 Curitiba, PR, Brazil Received 22 July 2003; received in revised form 5 December 2003; accepted 14 January 2004 Available online 17 March 2004
Abstract In this work, an analytical procedure was developed to monitor the ethanolysis of degummed soybean oil (DSO) using Fourier-transformed mid-infrared spectroscopy (FTIR) and methods of multivariate analysis such as principal component analysis (PCA) and partial least squares regression (PLS). The triglycerides (reagents) and ethyl esters (products) involved in ethanolysis were shown to have similar FTIR spectra. However, when the FTIR spectra derived from seven standard mixtures of triolein and ethyl oleate were treated by PCA at the region that represents the C=O stretching vibration of ester groups (1700–1800 cm−1 ), only two principal components (PC) were shown to capture 99.95% of the total spectral variance (92.37% for the former and 7.58% for the latter PC). This observation supported the development of a multivariate calibration model that was based on the PLS regression of the FTIR data. The prevision capability of this model was measured against 40 reaction aliquots whose ester content was previously determined by size exclusion chromatography. Only small discrepancies were observed when the two experimental data sets were treated by linear regression (R2 = 0.9837) and these deviations were attributed to the occurrence of non-modeled transient species in the reaction mixture (reaction intermediates), particularly at short reaction times. Therefore, the FTIR/PLS model was shown to be a fast and accurate method to predict reaction yields and to follow the in situ kinetics of soybean oil ethanolysis. © 2004 Elsevier B.V. All rights reserved. Keywords: Degummed soybean oil; Transesterification; Ethyl esters; FTIR; Multivariate calibration; PCA; PLS
1. Introduction In recent years, the concept of producing biodiesel from renewable lipid sources regained international attention. This tendency led the Brazilian Government to establish a national program (Probiodiesel) whose mission is to evaluate the technical, economic and environmental competitiveness of biodiesel in relation to the commercially available diesel oil. As a result, several research projects have been initiated nationwide to investigate and/or optimize biodiesel production from soybean oil and ethanol derived from sugarcane (ethyl esters), the only two raw materials that are considered available to support the initial activities of the national program. By definition, biodiesel (mono-alkyl esters of long chain fatty acids) is a diesel oil substitute that can be produced from renewable sources such as vegetable oils, animal fats and recycled cooking oils [1,2]. This alternative fuel is ∗
Corresponding author. Tel.: +55-41-3613175; fax: +55-41-3613186. E-mail address: [email protected] (L.P. Ramos).
0039-9140/$ – see front matter © 2004 Elsevier B.V. All rights reserved. doi:10.1016/j.talanta.2004.01.008
widely recognized as ecologically friendly, biodegradable, not hazardous for handling (flash point above 110 ◦ C) and of superior lubricity, providing a good performance and low emission profile upon combustion in compression ignition engines. Other advantages linked to biodiesel are that it has little (0.001%) or no sulfur content and it can be used in blends or as a neat fuel, without further investment in the current fleet [2,3]. Ethanolysis is a reversible reaction that is typically performed under alkaline conditions: vegetable oils or animal fats are reacted at room temperature with anhydrous ethanol in the presence of sodium or potassium hydroxide [3,4]. To ensure good fuel properties, the reaction must be carried out under optimal conditions because even a low contamination with unreacted triglycerides, glycerin or reaction intermediates may cause damage to the engine and elicit emissions of soot and harmful compounds such as acrolein [5]. Therefore, biodiesel fuel quality must be ensured by solid analytical methods such as capillary gas chromatography (CGC) [6,7], high performance liquid chromatography (HPLC) [8], size exclusion chromatography (SEC) [4] or nuclear
1022
G.F. Zagonel et al. / Talanta 63 (2004) 1021–1025
magnetic resonance (NMR) [9]. However, despite their high accuracy, these methods are time-consuming and can not be easily used for the in situ evaluation of fast reversible reactions such as transesterification. Fourier-transformed infrared (FTIR) spectroscopy has been reported as a fast and accurate method to monitor the methanolysis of vegetable oils [9–11]. However, in complex reaction mixtures like these, classical univariated calibration models are impracticable because serious spectral interferences (or overlapped signals) may result in non-linear correlations between the measured signal and the property of interest. To circumvent this problem, the application of multivariate analysis, together with the development of new methods for the rapid acquisition of multisignal spectra, allowed for the exploitation of the entirety of the analytically relevant information found in FTIR [12,13]. During the last decade, many authors have used multivariate regression to develop calibration models based on FTIR. Ferraz et al. [14] used partial least squares regression (PLS) to correlate the FTIR data with changes in wood composition during chemical or physical processing. Che Man and Setiowaty [15] described a simple FTIR/PLS calibration model to determine free fatty acids in palmolein. Lendl and Schindler [16] developed a FTIR flow-through sensor based on PLS for the simultaneous determination of several organic acids in aqueous solution. Garc´ıa-Menc´ıa et al. [17] developed a predictive model based on FTIR and PLS to evaluate the octane number of naphthas in a petrochemical refinery. Infrared spectroscopy has also been extensively investigated as a suitable method for the on-line monitoring of chemicals in complex mixtures. Fiber-optic probes have been developed to operate in the mid- [10,18] and near-infrared [11,19,20] regions and multivariate calibration methods have been applied to follow reaction kinetics as well as to establish good parameters for quality control in industrial processes. Knothe [19,20] used near-infrared spectroscopy and PLS to develop a model that was able to predict reaction yields during the methanolysis of soybean oil and this same method was proved useful to determine biodiesel fuel quality, regardless of their relative inadequacy to detect trace amounts of biodiesel contaminants. Other authors have also used the same principle to predict the transesterification end-point of methanolysis [11]. Although a variety of applications have been developed with this method, the routine use of near-infrared spectroscopy in industrial facilities remains limited by the cost of the hardware that is required for remote sensing. Under this viewpoint, analytical methods based on mid-infrared spectroscopy could reduce as much as 50% of the capital cost associated with this application. However, mid-infrared spectroscopy cannot be easily applied to analyze complex mixtures because peak overlapping usually compromises the spectral resolution. Multivariate calibration methods have also been used to simplify the complex matrix of chemical data that is gen-
erated by mid-infrared spectroscopy. Acha et al. [21,22] reported the use of several PLS calibration models for the on-line monitoring of a dechlorination process using a novel ATR mid-infrared sensor. Sadeghi-Jorabchi et al. [10] applied FTIR spectroscopy combined with an ATR mid-infrared fiber-optic probe to estimate, by univariate calibration, the occurrence of biodiesel contamination in lubricating oils. Küpper et al. [18] used silver halide fiber-optic probes to evaluate adulteration of extra virgin olive oils by ATR mid-infrared spectroscopy and PLS regression. More recently, Heise et al. [23] developed novel infrared optical probes for process monitoring based on inert infrared-transparent silver halide fibers. Progress in the quality of these fibers, including their extrusion with different cross-sections, enabled the construction of flexible fiber-optic probes of different geometries that are particularly suitable for remote sensing. In this work, an analytical method was developed to monitor the transesterification kinetics of degummed soybean oil (DSO) using FTIR and a multivariate approach. Our working hypothesis was that FTIR could be used as a simple and rapid analytical tool to determine the reaction yield of processes such as the ethanolysis of vegetable oils.
2. Experimental 2.1. Material The batch of degummed soybean oil used in this study was donated by IMCOPA (Indústria e Comércio de Óleos Vegetais Ltda., Araucária, PR, Brazil), whereas a commercial stock of anhydrous ethanol was kindly obtained from ALCOPAR (Associação dos Produtores de Álcool do Estado do Paraná, Maringá, PR, Brazil). Other reagents, as well as analytical standards of monoolein (1-mono[cis-9-octadecenoyl]-rac-glycerol), diolein (mixed isomers containing 85% of 1,3-di[cis-9-octadecenoyl]glycerol and 15% of 1,2-di[cis9-octadecenoyl]-rac-glycerol), triolein (1,2,3-tri[cis-9-octadecenoyl]glycerol) and ethyl oleate (cis-9-octadecenoic acid ethyl ester), were all of analytical grade (Sigma, St. Louis, MO). 2.2. Synthesis Ethyl esters of degummed soybean oil were obtained according to Zagonel [2]. Batches of DSO, containing 0.34% (w/w) of free fatty acids and 0.09% (w/w) of moisture, were placed in round-bottomed flasks and heated up to 40 ◦ C before ethanol was added to the mixture. A pre-heated ethanol solution, containing 1% (w/w) of potassium hydroxide in relation the oil mass, was then added to the vessel and the reaction was carried out for 1 h. Aliquots of 2 ml were withdrawn from the reaction mixtures at varied reaction times (e.g., 1, 10, 20, 40, and 60 min) and treated with 4 ml of an aqueous solution of ammonium chloride (5%, w/v). The mixture was
G.F. Zagonel et al. / Talanta 63 (2004) 1021–1025
centrifuged for 10 min at 11500 g to ensure good phase separation. The upper organic phase was withdrawn, washed with water, dried against anhydrous sodium sulfate (Na2 SO4 ) and kept in the dark for further analysis by SEC and FTIR. 2.3. Methods SEC of reaction aliquots was carried out using a Shimadzu LC10AD workstation provided with a SIL10A autosampler and detection by differential refractometry (RID10A) [2,4]. Each sample was solubilized in tetrahydrofuran (THF), filtered through a Teflon membrane with a pore size of 0.45 m and analyzed using a series of one pre-column and two Tosoh TSK-GEL columns (TSK 2000HXL and TSK 1000HXL). THF was used as the eluting solvent at a flow rate of 1 ml min−1 . Three analyses were carried out for each replicate and calibration was based on injection of external standards including ethyl oleate and mono-, di-, and triolein. A typical relative standard deviation of 1.6% was obtained throughout the calibration range used in this study. Standards (3–30 mg ml−1 ) and reaction aliquots were also analyzed by Fourier-transformed infrared spectroscopy using a Reagen KBr cell for liquid samples [2]. The FTIR spectra were collected using a Bomem FTIR MB-100 spectrophotometer in the 500–4000 cm−1 region, with a resolution of 4 cm−1 , 32 scans and triangular apodization.
1023
3. Results and discussion In the ethanolysis of vegetable oils, the FTIR spectra of reagents and products are very similar because of the high chemical similarity that exists among triglycerides and ethyl esters (Fig. 1). However, small differences can be clearly observed in two regions: one broader region covering wave numbers from 900 to 1500 cm−1 and another region, around 1730 cm−1 , that includes the stretching vibrations of carbonyl groups (Fig. 2). Changes in the latter region were readily attributed to the extent of glycerol substitution in fatty acids by ethoxy radicals (ethanolysis), whereas the former region was more complex and displayed a series of overlapped signals that were likely to interfere with the development of the model. Based on the evidence that soybean oil ethanolysis would impart changes in the FTIR spectra of the reaction mixture, a multivariate method was developed to allow the in situ evaluation of reaction yields. Initially, exploratory studies were carried out by PCA using seven standard mixtures containing triolein and ethyl oleate (Table 1). The entire spectrum was divided in nine regions, each of them covering an
2.4. Multivariate analysis The entire FTIR spectral range of seven standard mixtures, containing different proportions of triolein (representing triglycerides found in vegetable oils) and ethyl oleate (representing ethyl esters produced as a result of transesterification) (Table 1), was treated by principal component analysis (PCA). The region with the greatest spectral differentiation was identified and subsequently used to develop a multivariate calibration model based on partial least squares regression. Data analysis was carried out in a MATLAB 4.0 environment, whereas both PCA and PLS procedures were performed using the PLS-toolbox 1.5 [24]. The FTIR spectra were mean-centered to compensate for any baseline dislocation and the spectral data were processed only at the region where the best correlation could be obtained.
Fig. 1. FTIR spectra of degummed soybean oil and its corresponding ethyl ester from 800 to 3200 cm−1 .
Table 1 Standard mixtures used in the multivariate analysis Sample
Triolein (%)
´ Ester (%)
1 2 3 4 5 6 7
0.00 9.95 29.88 49.85 69.88 89.95 100.00
100.00 90.05 70.12 50.15 30.12 10.05 0.00
Fig. 2. FTIR spectra of degummed soybean oil and its corresponding ethyl ester from 1600 to 1900 cm−1 .
1024
G.F. Zagonel et al. / Talanta 63 (2004) 1021–1025 Table 2 Predictive capability of the multivariate calibration model Samplea
1 2 3 4 5 6 7 8 9
Real
Predicted
Triolein (%)
Ester (%)
Triolein (%)
Ester (%)
0.00 9.95 29.88 49.85 69.88 89.95 100.00 19.91 8.52
100.00 90.05 70.12 50.15 30.12 10.05 0.00 80.09 91.48
0.60 8.91 30.07 50.19 70.12 89.68 99.92 19.27 8.15
99.40 91.09 69.93 49.81 29.88 10.32 0.08 80.73 91.85
a Calibration was performed with samples 1–7, whereas samples 8 and 9 were used to predict the accuracy of the method (provision samples).
Fig. 3. Scores (A) and loadings (B) of the PCA.
interval of 200 cm−1 , and these data sets were individually treated by PCA. A graph of scores was obtained for each of the data intervals and these graphs were used to identify the region in which the transesterification of soybean oil could be better represented. The best result was obtained from the spectral region ranging from 1700 to 1800 cm−1 . By running a PCA in this region, only two principal components were shown to capture 99.95% of the variance, being 92.37% for PC1 and 7.58% for PC2. That is, two “new vectors” were enough to represent the multidimensional space denoted by the entire spectral data set, which consisted of a linear combination of 54 values of frequency. The graph of scores obtained for the 1700–1800 cm−1 region is shown in Fig. 3A. This graph demonstrated that the stretching vibration of carbonyl groups in esters was the key parameter for the differentiation among the seven calibration standards, which contained different amounts of triolein and ethyl oleate. Fig. 3B shows the graph of loadings where the weight of each original variable (or frequency) is represented in relation to the establishment of the PCs. These results confirmed the previous observation that the dislocation of the carbonyl signal, between the variables 12 and 32, was responsible for the spectral differentiation that occurred as a result of a gradual increase in triolein concentration. That is, for the seven standard mixtures described in Table 1, there was a gradual migration of the C=O stretching vibration from a maximum at 1746 cm−1 (νC=O in triglycerides) to a maxi-
mum at 1735 cm−1 (νC=O in ethyl esters), passing through a minimum at an average wave number of 1740 cm−1 (Fig. 2). Therefore, the individual scores of PC1 and PC2 indicated that progressive changes in triolein concentration (see sequence 1–7 in Table 1) could be measured by changes the C=O stretching vibration in the 1700–1800 cm−1 region. As mentioned above, the triolein-to-ester ratio was adequately reproduced by the spectral data, and only two latent variables (LV) (or two PCs) were responsible for capturing 99.95% of the variance. Based on these evidences, a multivariate calibration model was developed by partial least squares regression using the region previously established by PCA. This model was built with the entire calibration set described in Table 1 plus two independent samples used as the provision objects (Table 2). Only two latent variables were shown to predict 99.89% of the total variance of the x-block (99.03% for the first and 0.86% for the second LV) and 99.98% of the total variance of the y-block (99.96% for the first and 0.03% for the second LV). Therefore, the model was proved useful to predict changes in triolein-to-ester ratios based on the spectral variance of the 1700–1800 cm−1 region. The FTIR/PLS calibration model, developed from the seven synthetic mixtures described in Table 1, was then used to predict the ester yield of a new experimental data set related to the kinectic study of degummed soybean oil (DSO) ethanolysis. The reactions were carried out under the conditions used by Zagonel and colleagues [2,4] for the transesterification of DSO and the results obtained through the model were compared to reaction yields originally determined by size exclusion chromatography [4]. By SEC, the ester concentration is evaluated from a signal absolutely free of interference, while through FTIR, this parameter is obtained from an analytical response of much greater complexity. A total of 40 reaction aliquots, collected at different time intervals from eight independent experiments, were applied to validate the model (Fig. 4). In general, there was a good correlation between both data sets (R2 = 0.9837), determined independently by SEC and FTIR/PLS. However, small discrepancies were observed along the linear regres-
G.F. Zagonel et al. / Talanta 63 (2004) 1021–1025
1025
versible reactions such as the transesterification of vegetable oils.
Acknowledgements The authors are grateful to Dr. Martin Mittelbach for providing us analytical standards of monoolein, diolein, triolein and ethyl oleate, as well as to the financial support of CNPq, CAPES, Funpar and Fundação Araucária.
References Fig. 4. Correlation between reaction yields obtained by SEC and through the FTIR/PLS model.
sion, particularly at low conversion yields. This apparent uncertainty of the model was attributed to the interference caused by unmodeled reaction intermediates (unreacted mono- and diglycerides), which are particularly important at short reaction times. There was a considerable simplification of the matrix towards the end of the reaction. Therefore, deviations at short reaction times were expected due to the occurrence of a greater amount of unmodeled reaction intermediates. However, the FTIR/PLS model was built to predict more advanced reaction yields of DSO ethanolysis, which is what really matters to indicate the time in which the reaction should be stopped. Therefore, Fig. 4 indicates that the model was quite reliable for predicting reaction yields at advanced reaction times in which conversions greater than 30% were obtained.
4. Conclusions The FTIR/PLS calibration developed in this study was proven perfectly suitable as an analytical method to predict reaction yields during the ethanolysis of degummed soybean oil. Even though small discrepancies were observed between the chromatographic and the FTIR/PLS data sets, particularly at very short reaction times, the inherent advantages of the infrared spectroscopy, such as simplicity, high analytical velocity, low-cost, and facility for the implementation of on-line monitoring systems, suggest this method as a powerful analytical procedure for the evaluation of fast re-
[1] U. Schuchardt, R. Sercheli and, R.M. Vargas, J. Braz. Chem. Soc. 9 (1998) 199. [2] G.F. Zagonel, M.Sc. Dissertation, Federal University of Paraná, Curitiba, Brazil, 2000. [3] G.F. Zagonel, L.P. Ramos, Rev. Quim. Ind. 717 (2001) 17. [4] G.F. Zagonel, P.G. Peralta-Zamora, L.P. Ramos, Sci. Eng. J. 1 (2003) 35. [5] D.Y.Z. Chang, J.H. Van Gerpen, I. Lee, L.A. Johnson, E.G. Hammond, S.J. Marley, J. Am. Oil Chem. Soc. 73 (1996) 1549. [6] C. Plank, E. Lorbeer, J. Chromatogr. A 697 (1995) 461. [7] M. Mittelbach, G. Roth, A. Bergmann, Chromatographia 42 (1996) 431. [8] B. Trathnigg, M. Mittelbach, J. Liq. Chromatogr. 13 (1990) 95. [9] G. Gelbard, O. Brès, R.M. Vargas, F. Vielfaure, U.F. Schuchardt, J. Am. Oil Chem. Soc. 72 (1995) 1239. [10] H. Sadeghi-Jorabchi, V.M.E. Wood, F. Jeffery, A. Bruster-Davies, N. Loh, D. Coombs, Spectrosc. Eur. 6 (1994) 16. [11] W.D. Mockel, M.P. Thomas, Proc. SPIE - Int. Soc. Opt. Eng. 1681 (1992) 220. [12] C.E. Miller, Chemom. Intell. Lab. Syst. 30 (1996) 11. [13] H. Martens, T. Naes, Multivariate Calibration, John Wiley, Chichester, 1993. [14] A. Ferraz, J. Baeza, J. Rodriguez, J. Freer, Bioresour. Technol. 74 (2000) 201. [15] Y.B. Che Man, G. Setiowaty, Food Chem. 66 (1999) 109. [16] B. Lendl, R. Schindler, Vib. Spectrosc. 19 (1999) 1. [17] M.V. Garc´ıa-Menc´ıa, J.M. Andrade, P. López-Mah´ıa, D. Prada, Fuel 79 (2000) 1823. [18] L. Küpper, H.M. Heise, P. Lampen, A.N. Davies, P. McIntyre, Appl. Spectrosc. 55 (2001) 563. [19] K. Knothe, J. Am. Oil Chem. Soc. 76 (1999) 795. [20] K. Knothe, J. Am. Oil Chem. Soc. 77 (2000) 489. [21] V. Acha, M. Meurens, H. Naveau, S.N. Agathos, Biotechnol. Bioeng. 68 (2000) 473. [22] V. Acha, M. Meurens, H. Naveau, S.N. Agathos, Water Sci. Technol. 40 (1999) 41. [23] H.M. Heise, L. Küpper, L.N. Butvina, Anal. Bioanal. Chem. 375 (2003) 1116. [24] B.M. Wise, N.B. Gallagher, PLS-Toolbox Version 1.5, Eigenvector Technologies, 1995.