Detecção de adulterações em gasolina e previsão de parâmetros físico-químicos a partir de curvas de destilação associadas a ferramentas quimiométricas

Abstract

Despite the action of ANP (National Agency of Petroleum Natural Gas and Biofuels) in the process of monitoring and inspection, there are samples of gasoline being sold out of specification which may be an indication of tampering. The existing specifications were adopted, largely based on a proper functioning of the engine, not to identify tampering. Thus, alternative methods for the identification and quantification of solvents in gasoline have been developed combining the distillation curves with PCA (Principal Component Analysis) and PLS-DA (Discriminant Analysis with Partial Least Squares Method). It was possible to perform the discrimination of adulterated samples and not adulterated from the PQMC (Monitoring Program of the Fuel Quality) and the distinction of samples unadulterated and added with solvent 5-50% (v/v). The use of PLS (Partial Least Squares) allowed the prediction of the level of contaminants with low prediction errors when compared to other methods. Besides identifying tampering are also necessary agility and economy in the quality control of gasoline. Tests are performed from the analysis of physico-chemical parameters of the ANP, according to Decree No 309. In general, the analysis results take a long time, and consume large quantities of samples and reagents producing residues that are prejudicial to health and the environment. In this context, alternative methods for predicting the octane number MON (Motor Octane Number) and RON(Reserch Octane Number) than the vapor pressure have been developed in this work, combining the distillation curves and multivariate PLS calibration. The results showed that the predictions generated low errors compared to the literature using techniques such as chromatography and infrared spectroscopy, showing the quality of models constructed. Adding to this, the graphs of actual and predicted values were constructed and showed correlation coefficients higher than 0.99.