Avaliação de espectrômetros Raman e modelos de classificação multivariada para monitoramento da estabilidade oxidativa do biodiesel

Abstract

Biodiesel oxidation reactions result in the formation of undesirable products that can cause clogging and solids deposition in diesel engines. Biodiesel quality control is regulated in Brazil by the National Agency of Petroleum, Natural Gas and Biofuels (ANP), according to Resolution No. 920/2023, which establishes that the oxidative stability of biodiesel must be equal to or greater than 13 hours, being determined by the Rancimat method with approximately 12 hours of duration. This work proposes the use of Raman spectroscopy associated with chemometric tools for rapid classification of B100 biodiesel samples in relation to their oxidative stability. In the first part of the work, two Raman spectrometers were evaluated: a portable Raman spectrometer with SORS technology and excitation laser at 830 nm and a compact benchtop spectrometer with laser at 1064 nm. The multivariate classification models (SIMCA, PLS-DA and random forest) constructed with the obtained data were evaluated in terms of sensitivity, specificity, efficiency and Matthews coefficient. The results demonstrated that both spectrometers are suitable for use in the analysis of the oxidative stability of biodiesel, with the classification model using random forest presenting better performance. In the second part of the work, two class balancing methods were compared, ADASYN and SMOTE. The ADASYN method presented better performance in balancing the minority class, resulting in a model with greater efficiency. With the proposal of a hierarchical model, the samples classified as compliant were classified again, reaching an efficiency of 94.7% in the classification of compliant samples with oxidative stability greater than and less than 17 hours. The use of Raman spectroscopy associated with multivariate classification methods proved to be suitable for the analysis of biodiesel samples, reducing the analysis time compared to Rancimat and enabling its application in different stages of the process.