Determinação do número de cetano de blendas de biodiesel/diesel utilizando espectroscopia no infravermelho médio e regressão multivariada

Abstract

The aim of this work is to determine the cetane number of biodiesel/diesel blends using mid-infrared spectroscopy coupled with multivariate regression. Thirty five samples were used into the calibration set and twenty for the prediction set. The partial least squares algorithms used were: by Interval (iPLS), by Backward Interval (biPLS) and by Synergy Intervals (siPLS). In the best iPLS model was selected the spectral range 1050-1500 cm-ˡ obtained with root mean square error of prediction (RMSEP) of 0.314 and determination coefficient (r²) of 0.952 and root mean square error of cross-validation (RMSECV) of 0.728. For biPLS model the spectral range selected is a combination of several spectrum bands which showed a RMSEP of 0.353, r² of 0.960 and RMSECV of 0.664. The best model is the siPLS split the spectrum into 16 intervals and the combined spectral regions 1485-1277 cm-ˡ and 858-650 cm-ˡ presenting RMSECV = 0.642, RMSEP = 0.352 and r² = 0.962. The proposed methodology was adequate, with prediction errors less than 1%, being cleaner, faster and easy to perform.