RAMAN SPECTROSCOPY ASSISTED BY MACHINE LEARNING FOR THE STUDY OF BIODIESEL STABILITY PARAMETERS
Being a biofuel analogous to diesel, biodiesel appears as a bio-renewable energy source. However, its use demands methodologies that aim to maintain quality and reduce damage to the engine and the environment. In this context, spectroscopic techniques have become a suitable alternative that allows non-destructive analyses and requires minimal sample volume. Among them, Raman spectroscopy, which measures the inelastic scattering of radiation, has been shown as a promising technique for monitoring biodiesel quality. However, due to the high amount of information stored in Raman spectra and the complexity of the samples, it is challenging to extract quantitative information directly, making it necessary to use multivariate processing methods to model and extract the relevant information for exploratory or quantitative purposes. From this perspective, this work uses data mining and machine learning methods to classify soy biodiesel samples and determine oxidative stability under current legislation established by the Brazilian National Petroleum Agency (ANP). The combination of Raman spectroscopy and exploratory analysis allowed the tracking of structural modifications in biodiesel molecules through s changes observed mainly at the CH2, CH3, and -C=C modes. Multivariate classifiers as PLS-DA, SVM, and SIMCA were applied to classify biodiesel samples according to their water content (limit value = 200 mg kg-1) through their O-H, CH2, and CH3 vibrational modes. Besides, oxidative stability, defined as the resistance of the biofuel to oxidative processes caused by oxygen and temperatures nearer ambient, and measured through induction time, was assessed by building PLS calibration models. The structural changes caused by the oxidation process showed differences in CH2 and CH3 vibrational modes mainly due to C=C bond cleavage. Also, a soft-modeling procedure using a multivariate curve resolution method (MCR-ALS) was applied to study the kinetics of this process. For the Exam, an article recently submitted for publication is presented.