Determinação de parâmetros físico-químicos do óleo diesel a partir de curvas de destilação utilizando técnicas quimiométricas

Abstract

Diesel is the most consumed fuel in Brazil (49 biL in 2010), which is related to extensive road network in this country. For the consumption of this fuel, it is necessary to verify whether it is appropriate to operating in the diesel engine. This is carried out by analyzing physicochemical parameters, such as distillation, specific gravity, and viscosity, according to Resolution No. 33 of ANP. These physicochemical parameters are time consuming, have high cost of implementation and maintenance of equipment, and consume high purity solvents. Thus, alternative methods for predicting parameters related to the fluidity, flammability and content biodiesel in diesel were proposed in this paper, using multivariate calibration PLS and distillation curves. Low prediction errors were obtained in all predictions, when compared to other analytical techniques, such as infrared spectroscopy, which proves the efficiency of the calibration models constructed from distillation curves. In addition, high correlation between reference and predicted values were obtained in all predictions, indicating the good fit of the models built. This method has low cost, reduces analysis time, is easy to implement and can replace the currently used standard methodology. In addition to evaluate whether the samples of diesel oil are suitable for consumption, it is necessary to determine their origin and type, since this prediction may be an efficient mechanism in enforcement actions and combat cases of tax evasion that are related to simulation of the commercialization of diesel to the states where the tax is lower. Thus, the chemometric techniques PCA and LDA jointly with distillation curves were used to classify and predict origin and type of diesel samples from five refineries and two types. With the PCA it was possible to classify the samples into six groups, according to the origin and type, while by using LDA the origin and type were predicted, with 95.3% accuracy.