Determinação do teor de biodiesel metílico de palmiste (BMPALM) e pinhão manso (BMPM) usando espectroscopia no infravermelho médio e calibração multivariada por resolução multivariada de curvas com mínimos quadrados alternados (MCR-ALS) e regressão por vetor de suporte (SVR)

Abstract

The growing demand for biodiesel has increased the risk of marketing fuels that do not meet the standards established by the Brazilian National Agency of Petroleum, Natural Gas and Biofuels (ANP), with the concentration of biodiesel in diesel blends being one of the main parameters of non-compliance. In response to this analytical demand, this study developed methods for quantifying the content of palm kernel methyl biodiesel (BMPALM) and jatropha methyl biodiesel (BMPM) in diesel blends, using mid-infrared (MIR) spectroscopy combined with multivariate calibration techniques: multivariate curve resolution with alternating least squares (MCR-ALS), support vector regression (SVR), and partial least squares (PLS) as the reference method. Samples were prepared in different concentrations (1.00 to 30.00% v/v), submitted to spectral acquisition, and then modeled using chemometric approaches. The models built for each type of biodiesel showed good accuracy, with low root mean square errors (RMSEs < 1%) in all stages (calibration, cross-validation, and prediction), and good linearity, as evidenced by determination coefficients (R²) very close to 1. Statistical tests (F-test) indicated similar variances between most models, with few exceptions in specific comparisons between techniques. The results confirm that the combination of FT-MIR spectroscopy with advanced multivariate calibration methods, especially SVR and MCR-ALS, enables reliable quantification of BMPALM and BMPM in diesel blends. Moreover, these two methods offer advantages over PLS, such as lower sample requirements for model construction and reduced operational costs, representing promising alternatives for quality control in the fuel industry and for regulatory monitoring.