Classificação e determinação dos atributos de qualidade das fibras de algodão utilizando um espectrômetro Raman portátil e quimiometria

Abstract

Cotton down is an important natural fiber with many applications. Its market value is associated with quality, expressed through variables and considerations of aspects such as production location, genetic modifications and maturity. These measurements are performed by a High-Volume Instrument (HVI). This analysis technology demands high costs, lengthy analysis time, laborious sample preparation, among other conditions. With Raman spectroscopy such measurements can be performed in a simple, fast and economically viable way. Therefore, the objective is to develop an analytical methodology through Raman spectroscopy and multivariate analysis to determine quality attributes and classification of cotton fibers of different classes. For this, spectra of 213 cotton fiber samples were recorded in the region of 400 to 2300 cm-1 with a portable Raman spectrometer. With the HVI, reference measurements were made for the attributes length, short fiber index, micronaire index, strength and reflectance. The exploratory analysis was performed through the PCA's with the data of attributes and Raman spectra. Likewise, for the classification of samples, PLS-DA models were built with the data of measured quality attributes and Raman spectral. The determination of the attributes took place through the PLS models. From the spectral profile of the samples, a working range in the region of 700 to 1700 cm-1 was selected. In the PLS-DA Raman models, an RMSE of 17% was obtained for the samples of different maturities in the test stage. With the data of quality attributes, 90.51% accuracy was obtained for the samples from locality B in the test stage. In PLS, the model for the short fiber index parameter presented an RMSEC of 0.590 and RMSEP of 0.525, lower than the coefficient of variation 8,55. Thus, the method has potential for classifying samples through Raman spectroscopy, considering that it is a technique of simple application and low cost for analysis, in addition to the speed that the method can present to classify the fibers according to the productive profile. Also, it is possible to classify fibers from different locations from the models built with the HVI data. As for the quantification models, some experimental adjustments are needed to improve their predictive capacity.