Aplicação de espectrômetro de infravermelho próximo ultracompacto e quimiometria para a análise de gordura hidrogenada de soja
The industrial process of soybean oil hydrogenation, although already well established and widespread, usually has its quality control performed through time-consuming methodologies that expend several reagents, demand sample preparation and generate diverse chemical residues. From this, the need ar...
| Autor principal: | Pereira, Juliana Mendes Garcia |
|---|---|
| Formato: | Dissertação |
| Idioma: | Português |
| Publicado em: |
Universidade Tecnológica Federal do Paraná
2018
|
| Assuntos: | |
| Acesso em linha: |
http://repositorio.utfpr.edu.br/jspui/handle/1/3140 |
| Tags: |
Adicionar Tag
Sem tags, seja o primeiro a adicionar uma tag!
|
| Resumo: |
The industrial process of soybean oil hydrogenation, although already well established and widespread, usually has its quality control performed through time-consuming methodologies that expend several reagents, demand sample preparation and generate diverse chemical residues. From this, the need arises for developing faster low cost instrumentation methodologies, which do not generate expressive amounts of chemical residues. The aim of this work was to evaluate the application of near infrared spectroscopy (NIRS) in tandem with the partial least squares regression (PLSR) or support vector regression (SVR) in the control of soybean oil hydrogenation process. Models were designed to predict the amount of saturated fatty acids (SFA), unsaturated fatty acids (UFA), monounsaturated fatty acids (MUFA), trans fatty acids (TFA), polyunsaturated fatty acids (PUFA), and the iodine value (IV). The values predicted by the PLSR and SVR models were compared to the experimental values obtained by gas chromatography with flame ionization detector (GC-FID). As NIRS spectra present a large number of variables, a methodology for feature selection was also assessed. Good multivariate models were obtained for IV, MUFA, PUFA, and TFA for PLSR, and good models for IV and UFA when using SVR. The feature selection using the correlation informative vector was also efficient, maintaining the performance of the models and reducing by up to 78% the amount of variables used for the PLSR and 85% for the SVR. The values obtained for root mean square error of cross validation (RMSECV), root mean square error for calibration set (RMSEC), root mean square error for prediction set (RMSEP) and correlation coefficient (r2) remained very close for both PLSR and SVR. The residual prediction deviation (RPD) was adequate for the quality control of the hydrogenation process in both models, for both IV and PUFA in the PLSR and for the PUFA in the SVR. Regarding relative standard deviation (RSD), all values are above 5% for PLSR models, although models with values between 10 and 20% still show good predictive capacity. For SVR, RSD presents acceptable values, i.e., less than 5% for both IR and UFA. Compared with the mid-infrared spectroscopy (FTIR-ATR) for the same samples, better results were obtained regarding RMSECV, RMSEC and RMSEP for mid-infrared spectra. However, the r2 values were very similar, which guarantees the use of the models obtained using NIRS spectra. It is worth noting that was used an ultra-compact NIRS (900-1600 nm) with the advantages of being reliable for larger sample volumes, has a low cost and effortless-assembly. Since the NIRS equipment has no moving parts, it can be used in any environment, including in locu. Thus, through the results obtained, it was demonstrated that the NIRS methodology both in tandem with PLSR or SVR could be used to monitor the industrial hydrogenation process of soybean oil. |
|---|