Carbon Nuclear Magnetic Resonance Spectroscopic Profiles coupled to Partial Least-Squares Multivariate Regression for Prediction of Several Physicochemical Parameters of Brazilian Commercial Gasoline

• Published in: Energy & fuels 2012-09, Vol.26 (9), p.5711-5718
• Main Authors: Flumignan, Danilo Luiz, Sequinel, Rodrigo, Hatanaka, Rafael Rodrigues, Boralle, Nivaldo, de Oliveira, José Eduardo
• Format: Article
• Language: English
• Subjects: Automotive components; Brazil; Ethyl alcohol; Fingerprinting; Gasoline; Mathematical analysis; Mathematical models; Nuclear magnetic resonance; Regression
• ISSN: 0887-0624; 1520-5029
• DOI: 10.1021/ef300722c

Brazilian commercial gasoline follows a rigid quality control, regulated by Brazilian Government Petroleum, Natural Gas, and Biofuels Agency, ANP, following international analytical protocols, such as ASTM and ABNT, covered by Regulation ANP No. 309. Each property is a complicated function of the gasoline chemical composition, which would be represented by diverse types of mathematical correlations. However, these correlations are not adjusted to Brazilian gasoline, whose chemical composition is modified by anhydrous ethanol addition. The purpose of this work is to find correlations, using partial least-squares (PLS) regressions, between 13C NMR Brazilian gasoline fingerprintings and several physicochemical parameters, such as relative density, distillation curve (temperatures related to 10, 50, and 90% of distilled volume, final boiling point and residue), octane numbers (motor and research octane number and antiknock index), hydrocarbon compositions (olefins, aromatics, and saturated) and anhydrous ethanol and benzene. 150 representative gasoline samples, collected randomly from different gas stations, were analyzed following international analytical protocols. All 13C NMR spectroscopic fingerprintings, reported in parts per million (ppm), FIDs (free induction decays) were zero filled and Fourier transformed. A data matrix, composed of 13C NMR chemical shifts and physicochemical parameters, was constructed and used in PLS regression. 13C NMR fingerprinting of 100 gasoline samples were employed in the training set, and 50 samples formed the prediction set. In 13C NMR-PLS models, root-mean square error of calibration (RMSEC) and prediction (RMSEP) were the mains parameters considered to select the “best model”, which shown results roughly similar in magnitude to the repeatability and reproducibility of ASTM and NBR officials analytical protocols. 13C NMR-PLS multivariate regression, as an alternative analytical methodology, offers an appealing procedure for commercial automotive gasoline quality control.