Determinação de parâmetros de qualidade de gasolinas automotivas utilizando espectroscopia de emissão por chama e métodos quimiométricos

• Ano de defesa: 2013
• Autor(a) principal: Jaqueline Martins de Paulo
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Minas Gerais; UFMG
• Palavras-chave em Português: Mínimos quadrados parciais; Ferramentas quimiométricas; Flame Emission Spectroscopy (FES); Partial Least Squares (PLS); Espectroscopia de Emissão em Chama; Gasolinas adulteradas; Análise discriminante com método de mínimos quadrados parciais (PLS-DA); Química analítica; Espectroscopia de chama; Gasolina; Adulterações; Quimiometria
• Link de acesso: http://hdl.handle.net/1843/SFSA-9BWQS6

In this work a system was developed combining Flame Emission Spectroscopy with chemometric tools such as Partial Least Squares (PLS) and Partial Least Squares DiscriminantAnalysis (PLS-DA) with the objective of improving the quality monitoring of fuels commercialized in Brazil through the employment of simple, fast and inexpensive methodologies. The applications of these methodologies were organized as follows: Chapter 5presents a PLS-DA model that classifies different types of gasoline commercialized in Brazil: common gasoline (CG), gasoline with additives (GA) and premium gasoline (PG). In this case a PLS-DA model was built using three latent variables (LV) with accumulated variance of99.98% in X and 51.05% in Y. The model yielded good sensitivity and specificity values for the calibration set and 100% accuracy in the prediction. Chapter 6 presents the classificationof Brazilian gasoline samples adulterated with kerosene, turpentine, thinner, rubber solvent and ethanol and a PLS-DA model was constructed using five latent variables (LV) with cumulative variance of 100% in X and 84.6% Y. Chapter 7 presents the obtained spectral datausing FES associated with PLS to predict octane parameters, Motor Octane Number (MON) and Research Octane Number (RON). Low values of Root Mean Square Error of Calibration(RMSEC) and Root Mean Square Error of Prediction (RMSEP) were obtained: 0.14 and 0.56 respectively. The proposed methodologies are very promising especially because they are simple, fast and do not require pre-treatment of samples. The system can be enhanced for screening in routine analysis presenting low cost and good accuracy. In this context, the methodology provides significant possibilities for laboratories to perform quality control with greater efficiency in the inspection.