Trond. Brekke

Prediction of physical properties of hydrocarbon mixtures by partial-least-squares calibration of carbon-13 nuclear magnetic resonance data

By using the partial-least-squares (PLS) method, bulk properties of 12-component synthetic mixtures containing n-alkanes, iso-alkanes, cyclo-alkanes and aromatics are calibrated against intensities and chemical shifts of 13C-NMR spectra. The standard error of prediction (SEP) for the determinations of density, refractive index, mean molecular weight and carbon-type distribution was found to be less than 3.2% of the observed range. The SEP for excess densities is significantly larger, especially for values based on chemical shift data. The chemical shift variation supplies unique chemical information on solute/solvent interactions.

Assignment of 13C nuclear magnetic resonance spectra of complex mixtures by multivariate analysis

Abstract 13 C nuclear magnetic resonance in combination with multivariate data analysis has been applied to the identification of single constituents in hydrocarbon mixtures. Both model mixtures and real petroleum fractions (b.p. 180–220 °C) have been analyzed. Mixtures containing the same constituents in varying relative amounts show variations in spectral intensifies which give correlation patterns that correspond to individual chemical constituents in the mixtures. 85–90% of the chemical shift variance in the spectra is shown to be induced by variations in aromaticity. Aromatic methine carbons, saturated methine/methylene carbons and methyl carbons require separate chemical shift referencing due to the aromaticity-induced solvent shifts.

Chapter 20 Analysis of nuclear magnetic resonance spectra of mixtures using multivariate techniques

Publisher Summary The aim of this chapter is to present data-analytical methods that can cope with the amounts of potential information obtained by modern nuclear magnetic resonance (NMR) spectroscopy. The systems investigated are petroleum fractions and related systems of hydrocarbon mixtures. However, the methodology is the main issue. The NMR mixture spectra are subjected to correlation analyses, projection analyses, principal component analysis, and partial-least-squares regression analysis. These methods provide: (1) correlation of samples, (2) spectral interpretation and thereby identification of constituents, and (3) quantification of constituents. 13 C NMR provides spectra in which the individual constituents are resolved even in highly complex mixtures. It is not possible by simple visual investigation to recognize the information content because of the complexity of the spectra. Multivariate analysis (MVA) provides a tool for this task. This chapter examines how to make 13 C NMR spectra amenable to MVA—that is, how to circumvent the solvent shift problem. Once this is done, a battery of MVA techniques is available. With the high resolution and S/N ratio obtained with modern NMR spectrometers, it is not always necessary to turn to sophisticated methods to resolve overlapping or noisy peaks. Thus, the direct use of the correlation matrix is a viable method for constituent identification. Solvent shifts are complex observables, and requires other methods to resolve the different contributions. Principal component analysis is a powerful method for this purpose. Constituent quantification is achieved either by a direct method such as marker-object projections, or indirectly by Partial-Least-Squares (PLS) calibration and prediction.