CHEMOMETRIC TOOLS OF MULTIVARIATE CALIBRATION FOR MONITORING OF DIESEL OIL QUALITY
Monitoring, Diesel Oil, Chemometrics, Multivariate Calibration and Artificial Neural Networks
Diesel oil is one of the main derivatives of petroleum, fundamental for the Brazilian road sector, in the transportation of passengers and cargo. The adulteration of this fuel, with low-cost products, such as vegetable oils and petrochemical solvents, is worrying, since it brings numerous damages, both financial and environmental. For the rapid identification of these changes in fuels, it is necessary to develop more practical and efficient methods applied to the monitoring of diesel quality and to the detection and quantification of adulterants. This work was developed with the aim of contributing to the repertoire of analytical techniques applied to the monitoring of diesel, using spectroscopic methods associated with chemometric techniques. This research was conducted using two types of approaches, the Multivariate Curve Resolution (ALCR) and Partial Least Squares (PLS), for the identification and quantification of residual vegetable oils added as adulterants in the commercialized diesel oil and the Artificial Neural Networks (ANNs) for determination of specification parameters of diesel oil. In the identification and quantification of the adulterant residual oil, 16 commercial samples of diesel containing 8% (v / v) biodiesel and 10 mg of sulfur / kg (denominated S10B8) were mixed with the residual frying oil at various volumetric concentrations (1 - 60%), and then determined their physicochemical properties specified by the National Agency for Petroleum Natural Gas and Biofuel - ANP (specific mass, kinematic viscosity, flash point and atmospheric distillation), according to American Society standards for Testing and Materials (ASTM). At the same time, infrared medium with Fourier transform, FT-MIR, and near infrared, FT-NIR, for the creation of multivariate calibration models. The two models created (MCR-NIR, MCR-MIR, PLS-NIR and PLS-MIR) were able to accurately predict, not presenting a statistical difference between the estimated concentrations of adulterant and the reference values, being valid for a confidence level of 95%. In addition, the MCR-ALS was able to recover the pure spectral profile related to fuels and adulterants. For the modeling of ANNs, 162 diesel samples of different compositions (50, 500 and 1800 mg kg-1) were used, thus revealing the variety of fuel in the Brazilian market, which were analyzed according to ASTM methodologies recommended by the ANP, with a total of 810 tests. The ANNs were used to predict, not simultaneously, flash point, cetane number and sulfur content (S1800) of diesel blends with 7% (v / v) biodiesel, using distillation curves (ASTM D86), specific mass (ASTM D405), cetane number (ASTM D4737), flash point (ASTM D93) and sulfur content (ASTM D4294) as input data for modeling. The low error values obtained in comparison to other chemometric models described in the literature and high correlation coefficients between the reference and predicted values showed that the ANNs were efficient in determining the flash point, cetane number / cetane number and sulfur (1800 mg kg-1). In addition, the proposed method presents advantages such as low cost and easy implementation, since it uses data of the own routine monitoring to evaluate the quality control of diesel.