Fourier-transform-infrared-spectroscopy-based approach to predict engine fuel properties of biodiesel

Robust models are useful to predict the properties of biodiesel to facilitate the careful choice of feedstock for producing biodiesel for automotive engine applications. In the present work, a Fourier transform infrared spectroscopy (FTIR)-based approach is used to predict the kinematic viscosity, density, cetane number, and the higher calorific value of biodiesels. Unlike the standard partial least-squares (PLS) regression performed over a complete infrared spectrum, a novel approach involving few independent variables based on the functional groups present in biodiesel and correlating them with the properties of biodiesel is explored in the present study. To mimic a wide range and the type of methyl esters present in biodiesels, five biodiesels of significantly different compositions, namely, camelina, coconut, karanja, linseed, and palm, are chosen and are mixed in different volumetric proportions to obtain 70 biodiesel blends. The peak absorbance ratios of 70 biodiesel blends obtained from FTIR are correlated with the measured kinematic viscosity, density, cetane number, and the higher calorific value. The property prediction models are developed using multilinear regression and an artificial neural network whose performance is compared with that of standard full spectrum PLS regression. The results obtained show that the proposed approach is simple, reliable, and direct and provides better prediction with mean absolute percentage errors of 4.62%, 1.04%, 2.75%, and 6.85%, respectively, for the kinematic viscosity, density, higher calorific value, and cetane number of biodiesels.