Uso de espectrometria no infravermelho médio, calibração multivariada e seleção de variáveis por intervalos na quantificação de biodieseis em misturas com diesel

• Ano de defesa: 2016
• Autor(a) principal: Gontijo, Lucas Caixeta
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Uberlândia; BR; Programa de Pós-graduação Multi-Institucional em Quimica (UFG - UFMS - UFU); Ciências Exatas e da Terra; UFU
• Palavras-chave em Português: Biodiesel; Infravermelho; Calibração multivariada; PLS; iPLS; Seleção de variáveis; Espectroscopia de infravermelho; Calibração; Infrared; Multivariate calibration; Variable selection; CNPQ::CIENCIAS EXATAS E DA TERRA::QUIMICA
• Link de acesso: https://repositorio.ufu.br/handle/123456789/17536; https://doi.org/10.14393/ufu.te.2016.47

In Brazil, since November 2014, it is required by law an addition of 7% (v/v) of biodiesel to diesel, and this mixture is called B7. According to the report issued by National Brazilian Agency of Petroleum, Natural Gas and Biofuels in 2015, 33.6% of commercial diesel samples examined presented biodiesel content outside quality specifications. Therefore, this work aims to develop and validate methodologies to quantify methyl and ethyl biodiesel from soybeans, used frying oil, cottonseed and jatropha into binary mixtures with diesel, using Mid Infrared Spectroscopy (680 to 3100 cm-1) and multivariate calibration by Partial Least Squares (PLS). Furthermore, we present a study of application of variable selection methods for intervals as Interval Partial Least Squares (iPLS), Backward Interval Partial Least Squares (biPLS) and Synergy Interval Partial Least Squares (siPLS). Thus, the models built from spectral intervals, which provided, statistically, lower Root Mean Square Error of Prediction (RMSEP) and fewer variables compared to the model of full spectrum, were validated to determine the respective type of biodiesel mixed with diesel. The results of the global PLS models, to all kinds of biodiesels, demonstrated excellent correlation between actual and predicted values of the prediction set (R > 0.99), in the range of concentration of 1.00 to 30.00% (v/v), values of RSMEP between 0.0473 and 0.1020% (v/v) and did not show systematic errors according to ASTM E1655 norm. The proposed global PLS models met the requirements of accuracy demanded by ABNT NBR15568 (RMSEP < 0.1%) besides using only one calibration model against two of the respective norm. In relation to Variable Selection methods by intervals, were efficient in the elimination of non-informative variables significantly improving the prediction of validation samples if compared to multivariate calibration by PLS models, with the exception of the models containing biodiesels from used frying oil and jatropha methyl biodiesel. Therefore, the developed methodologies are presented as a viable alternative to the usual analysis for determining the biodiesel content in mixtures as diesel.