Aplicação de técnicas espectroscópicas de reflexão no infravermelho no controle de qualidade de farinha de trigo.

Abstract

In this study are combined the potentialities of chemometrics techniques and the facilities in the acquisition of spectra presented by the techniques Diffuse Reflectance Infrared Fourier Transform Spectroscopy (DRIFTS), Near Infrared Diffuse Spectroscopy (NIRRS) and Horizontal Attenuated Total Reflection Spectroscopy (HATR) bearing in mind the simultaneous prediction of ash and protein levels, as well as the falling number in wheat flour samples. One hundred wheat flour samples, which were supplied by Filler S. A. - industry located on BR 471 km 58, Santa Cruz do Sul, were used during the research. The samples were analyzed in duplicate by infrared reflection spectroscopy techniques. For the determination of ash level reference, the gravimetric method 08-03 from the American Association of Cereal Chemists (1984) was used, for the determination of protein level, the 2055 method from the Association of Official Analytical Chemists (1984) was used and for the determination of falling number, the 56-81B method from the American Association of Cereal Chemists (1982) was used. To build the models the Partial Least-squares regression method (PLS) was used with raw spectra in log (1/R) form and its first and second order derivates for select spectral range. The data were the mean-centered data and/or variance escaling, as pre-processing tools, for a 55-sample set . To validate the models, Standard Error Validation (SEV) was used, and the prevision values were estimated for the other 45 samples not used in the modeling .Models were also developed, in which two methods for variable selection were applied. One of them was based in the spectra subtraction of sample which included extreme values in the properties to be determined; the other method applied the information of regression coefficients, spectra and loading of a PLS referential model simultaneously. The results reflect the importance to choose the data treatment, as well as the kind of pre-processing applied. Best results were obtained when there was simultaneous application of mean-centered data and variance scaling. The modeled data sets (spectra) also presented the best performance for the models in which data with first or second order derivates were applied. In relation to the spectroscopy methods applied, the ones which use mid- infrared spectra (DRIFTS and HATR) presented a better performance than the ones which apply near infrared spectra (NIRRS), when their performance is considered in relation to the validation set (SEV) . These results let us conclude that good models aiming the simultaneous prediction of ash and protein levels and falling number can be obtained, in spite of the fact that there is not any knowledge about bands (wave numbers) directly correlated to the ash level parameter, and that reflection techniques are adequate to allow a fast obtention of spectra and not to create wastes which are harmful to the environment.