Determinação simultânea dos parâmetros de qualidade, pontos de fulgor, percentagem de biodiesel/diesel, empregando reflexão no infravermelho médio e regressão multivariada.

Abstract

The search for new sources of renewable resources has started various discussions worldwide, especially related to petroleum that represents 35% of all energy resources in the world. Brazil has invested in the production and use of Biodiesel, being commercialized since the beginning of 2008 in mixture with diesel in the proportion of 2% (B2), having this proportion increased, starting in January of 2010, to 5% (B5). This work approaches the use of Fourier transform infrared spectroscopy (FTIR), using transreflectance and total attenuated reflectance (ATR) techniques, being that the second presented better results and it was with it that multivariate regression models were built. Therefore, the 45 samples that utilized metropolitan and countryside diesel (that differed in the quantity of sulfur) were prepared to form blends with different concentrations of biodiesel, which were of 8 to 30% (v/v). For the built of both models, this work proposes a study and the use of the partial least squares (PLS) together with variable selection techniques iPLS and biPLS, for a built of models with greater prediction capacity, which uses spectrum ranges with greater correlation with the studied variable. This work does not only look to quantify the biodiesel in the blends biodiesel/diesel, but also the effect of this quantity on the property quantity of sulfur, specific mass and flash point. For the quantification of sulfur, the more robust model, the algorithm siPLS with a RMSEP of 16.9 mg/kg and combined the spectrum ranges of 858 - 960 cm-1 and 2732 – 2834 cm-1 was used. For the prediction of biodiesel content the iPLS model was chosen, since it reached results superior to siPLS, with a RMSEP of 0.154% (v/v) and used a spectrum range of 1068 – 1484 cm-1 that corresponds to the asymmetric stretching C-C(O) of the esters of biodiesel. In the determination of specific mass the model that most adjusted was the iPLS, for using a fewer number of latent variables, with a RMSEP of 0.865 kg/m3 when range of 1274 – 1376 cm-1 was used. Finally, for the flash point, the iPLS and siPLS models reached the same performance as the RMSEP measures as well as in number of latent variables, being chosen the iPLS with RMSEP of 1.38 oC, when the spectral range 1274 – 1376 cm-1 was used. All the models that utilized the full spectrum showed prediction capacity inferior to those 9 submitted to iPLS or siPLS variable selection procedures, being a variable selection for deterministic methods, a promising strategy for the determination of physical, chemical, and physical-chemical properties in biodiesel/diesel blends.