Calibração multivariada associada à ressonância magnética nuclear na caracterização de petróleos

Abstract

The use of Nuclear Magnetic Resonance (NMR) spectroscopy with chemomet-rics tools has become an efficient methodology for estimating physicochemicalproperties of Brazilian oils with their high precision, speed and reliability. In thiswork, the hydrogen NMR (1H NMR) technique associated to the multivariatecalibration PLS, OPLS and variable selection (iPLS, siPLS, and MPA) was usedto estimate the UOP characterization factor K (KUOP), nitrogen content, solubil-ity parameter, sulfur content, maximum and minimum pour point and kinematicviscosity at 20 °C, 30 °C, 40 °C and 50 °C of 138 Brazilian crude oil samples.The models were evaluated by coefficient of determination (R2), root meansquare error of predicted (RMSEP) and the residuals. The sulfur content andthe solubility parameter had their estimates improved by PLS and OPLS, re-spectively, with a RMSEP of 0.09 wt% (sulfur content) and 1.06 (MPa)1/2 (solu-bility parameter). Model population analysis presented better results in the esti-mation of KUOP, nitrogen content, kinematic viscosity at 40 °C and pour pointwith prediction mean error of 0.06, 0.048 wt%, 0.036 and (14.6 °C and 19.2 °C)respectively. Minimum and maximum pour point models had the weakest per-formances; suggesting the complexity of this property. With very promising re-sults, MPA has shown that it is an innovative chemometrics technique.