Characterization and Classification of Crude Oils Using a Combination of Spectroscopy and Chemometrics

Research has been carried out to the utility of chemometric models to predict long residue (LR) and short residue (SR) properties of a crude oil directly from its absorption or magnetic resonance spectrum. Such a combined spectroscopic-chemometric approach might offer a fast alternative for the elaborate crude oil assays that are currently used in petrochemical industries. Six different spectroscopic techniques have been explored: infrared (IR), near IR (NIR), Raman, UV-Vis, 1H-NMR and 13C-NMR. Principal Component Analysis (PCA) and Partial Least Squares (PLS) regression were selected as chemometric modeling techniques. Seven different LR properties have been studied, i.e., yield-long-on-crude (YLC), density (DLR), viscosity (VLR), pour point (PP), asphaltenes (Asph), carbon residue (CR) and sulfur content (S). Four SR properties as function of the atmospheric equivalent flash temperature (AFT) were considered, i.e., penetration depth (P), ring and ball (R&B), density (DSR) and viscosity (VSR). IR and NIR spectroscopy proved to be the most useful techniques for LR and SR prediction. Despite the high information content of the spectra, NMR performed less good, while UV-Vis and Raman spectroscopy turned out to be not useful. To improve the LR and SR prediction models, the combination of IR and NMR spectra as input for modeling was explored, but enhancement was not achieved. Furthermore, the influence of temperature treatments as a tool for improvement was investigated. The crude oils were exposed to 65C to reduce the contribution of volatile constituents in the IR spectra. Modeling of these spectra did not result in better LR or SR prediction models. Besides, the influence of the temperature on the IR spectra of the exposed crude oils measured at 20, 40 and 60C was studied. The PLS models based on these variable temperature data for LR predictions appeared to need fewer latent variables (LV’s) and a slight improvement in the SR predictions were obtained. In the next step, the applicability of the IR models to predict the LR properties of mathematically created IR spectra of blends was investigated. Physically prepared blends of two crude oils in various weight ratios could be mimicked by co-adding the IR spectra in the same ratio. The predictions of the LR properties of the artificial and the real blends were found to be largely the same. The robustness of the LR prediction models was tested by measuring the IR spectra on different instrumental set-ups. The models performed best in case the validation spectra were recorded on the same set-up as the calibration spectra. If another set-up was used, the accuracy decreased by a factor of 2. Finally, as an additional goal, PLS-modeling of IR spectra as a tool for sulfur speciation of crude oils was investigated. This application turned out to be limited compared to 2D gas chromatography analysis, which is widely used for this purpose. However, the models to predict the total sulfur concentration and the concentration of dibenzothiophenes and 3 different benzothiophene classes perform reasonably well. Based on the results of the described study, a computer program to predict LR and SR properties of crude oils and blends has been developed. This program is currently being tested on-site and the underlying methodology has been patented