Application of multivariable calibration techniques to determine physical-chemical properties and quality of gasoline products

Gasoline is one of the main fuels used in transportation and is produced in petroleum refineries via distillation of crude oil, followed by some other processes. For every batch of gasoline prepared at petroleum refineries, all the parameters given in TS EN 228 specifications are analyzed according to reference method published by ISO 17025 accredited laboratories to determine if the product meets the limits. Multivariate calibration techniques, a subject in chemometrics, take an important role here. Multivariate calibration method is a technique in which spectroscopic measurement is combined with the reference test methods results to obtain a model for future prediction. In this thesis work, 46 gasoline samples, obtained from the Quality Control Laboratory of Tupras Izmit Refinery, were studied to establish multivariate calibration models to see if they can be used instead of reference test methods and the prediction error is determined to compare. Octane number (RON and MON), distillation, hydrocarbon types (aromatics, olefins and benzene) and density parameters were the parameters studied, and all gasoline samples were analyzed by using reference test methods for these parameters. Same samples were analyzed by NIR spectroscopy in wavelength range of 800 – 2500 nm. After baseline correction, mean centering and removal of outlier, PCR and PLS multivariate calibration methods used to obtain calibration models. One sample is removed from data set as outlier. 30 samples were used as a calibration data set and 15 samples were used as a validation data set, which are not used in calibration data set. Predicted results, residuals, root mean squared error of prediction (RMSEP) were calculated, and some useful plots were plotted for both calibration and validation data sets. According to the results, all the residuals calculated for MON and Aromatics by using PCR and for RON, MON and Aromatics by using PLS are smaller than reproducibility of reference test method. The PLS model results indicated that more than 90% of residuals are even smaller than half of the reproducibility (R/2) for RON, MON, Aromatics and E150. It can be concluded that the PLS model can be safely used as a replacement for the reference test methods for RON, MON, Aromatics and E150 with the condition of remaining in studied range and with the condition of using samples of similar molecular structure. The residuals and RMSEP values were found higher for olefins and density parameters, a detailed study with different parameters needed in order to get better results.