Aplicação de métodos de seleção de variáveis para o controle de qualidade de biodiesel de pinhão manso e de moringa em misturas com diesel usando espectrometria no infravermelho médio

• Ano de defesa: 2019
• Autor(a) principal: Sitoe, Baltazar Vasco
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Uberlândia; Brasil; Programa de Pós-graduação em Química
• Palavras-chave em Português: Quimiometria; Calibração multivariada; Classificação supervisionada; iPLS e iPLS-DA; Teor do biodiesel; Química; Chemometric; Multivariate calibration; Supervised classification; iPLS and iPLS-DA; Biodiesel content; Chemistry; CNPQ::CIENCIAS EXATAS E DA TERRA
• Link de acesso: https://repositorio.ufu.br/handle/123456789/27757; http://doi.org/10.14393/ufu.te.2019.2520

In Brazil, the addition of 10% (v/v) of biodiesel to diesel was mandatory since March 1st 2018 until August 31st, 2019. From the analysis of biodiesel/diesel blends made by the Fuel Quality Monitoring Program, it was found that the biodiesel content corresponds to 55.90% of the total nonconformities. Thus, the present study aimed to apply methods to quantify contents of jatropha and moringa methyl biodiesel and to classify their samples in blends with diesel, using Partial Least Squares Multivariate (PLS) calibration methods and supervised classification by Partial Least Squares Discriminant Analysis (PLS-DA) through interval variable selection methods and the mid-infrared spectrometry technique. Before to model building, pure biodiesel, B100; Then samples of the biodiesel/diesel blends were prepared. The models built from variable selection methods that presented statistically lower Root Mean Square Error of Prediction (RMSEP) value and better classification, compared to the full spectrum models (global models), were validated according to established merit parameters for multivariate analysis. The results of the quantification models for each type of biodiesel show an excellent correlation between the mensured and predicted values (R2 > 0.99) in the concentration range used (0.50 – 30% (v/v)), RMSEP values between 0.11 and 0.25% (v/v) and de models have no systematic errors. The efficiency of the classification models, also constructed by variable selection, was evaluated based on the figures of merit: sensitivity, specificity, accuracy and Matthews correlation coefficient (MCC). The threshold values were determined based on the Bayes theorem in order to minimize false positives and false negatives. The relationship between specificity and sensitivity was also graphically assessed using Receiver Operating Characteristics (ROC) curves. These parameters presented values equal to 1.0 for all models, which represents a correct classification of the samples of the training and test sets. Thus, the developed methods may be a viable alternative to the analysis for biodiesel content determination and classification of jatropha and moringa methyl biodiesel samples in diesel blends.