A comparative study between multivariate calibration and artificial neural network in quantification of soybean biodiesel

Abstract

Biodiesel is an alternative fuel that can be obtained by the transesterification of vegetable oils. Spectrophotometric methods can be used for the quantification of mixtures of oils, associated with chemometric tools, enabling the analysis of overlapping signals. The aim of this work was to apply the multivariate calibration with PLS (partial least squares) and artificial neural network (ANN), to estimate the concentration of esters in the transesterification of soybean oil using molecular absorption spectrophotometry as analytical technique. The absorbance measurements were performed in a spectrophotometer UV/VIS. Synthetic solutions were prepared with standards of the five major compounds of soybean biodiesel and the real samples were obtained by the reaction of transesterification of soybean oil with two types of catalysts: NaOH and enzymatic method using Lipozyme® IM (Novozymes). According to the results, all components of the reaction medium absorb in the wavelength range of 190-280 nm. The reactions of the basic catalysis reached conversions close to 100%, whereas the enzymatic reactions reached lower conversion values. For both methods, the calibration and validation groups were composed, respectively, by the synthetic and the real samples. Results showed that the concentrations of esters estimated by the ANN model in the real samples are more accurate (R2 of 0.93), showing the great ability of the ANN in estimating the data.