OPLS and PLS multivariate regression of1H NMR spectra of diesel/mafurra biodiesel blends for estimation of biodiesel content

Abstract

Two methodologies were developed to monitor the biodiesel content of mafurra in mixtures with diesel using hydrogen nuclear magnetic resonance (1H NMR) Spectroscopy combined with the multivariate regression by orthogonal projections to latent structure (OPLS) and partial least squares (PLS). The efficiency of these methodologies was analyzed based on the figures of merit and the fit of the models through the correlation of the measured and predicted values of the calibration and prediction sets. The results of the figures of merit in the OPLS model were better than in the PLS model. A high correlation between the measured and predicted values was evident in the OPLS model, with a correlation coefficient (R2) greater than 0.99, demonstrating a better fit of the OPLS model in relation to the PLS model which presented a correlation coefficient (R2) less than 0.98. The OPLS model is more robust and has good predictive capacity than the PLS model because it obtained a higher Q2 value. The excellent results of the application of 1H NMR spectroscopy combined with multivariate regression by OPLS suggest that this analytical methodology is ideal, feasible, efficient and suitable for use by inspection agencies to control the quality of this fuel.