PÊ, P. R.; http://lattes.cnpq.br/0402706548030700; PÊ, Patrícia Rodrigues.
Resumo:
Vegetable oils are one of the most caloric food sources. Furthermore, the similarity of their
fatty acid chains with petrodiesel is one of the advantages of biodiesel production. In the
world, its use is growing as a source of added value of dual competition between food and
energy use. But the tools used in the process of quality control of vegetable oils have technical limitations, such as the destruction of the sample, low frequency processing and generation of large volumes of waste. In this context, it was aimed to develop models exploratory of the multivariate calibration using non-destructive and rapid measures of low-field IH NMR and PCA, HCA, SIMCA, MLR, PCR and PLS. The signs of relaxation in NMR of T2 were obtained from a total of 65 samples of seven kinds of vegetable oil (cotton, n = 15; soybean, n = 15, olive oil, n = 15; rice, n = 5; sunflower , n = 5; maize, n = 5; canola, n = 5). The measurements were performed in a total of three replicates authentic using 150 mL of sample. The instrument used was a spectrometer 7005 with Oxford MQA electromagnet of 0.47 T of 5 MHz from the signal obtained, to use the techniques of PCA, HCA, SIMCA, MLR, PCR and PLS. In PCA, the graph shows the training of scores of different classes with good separation for cotton, soybean and olive. On four PCs you get 98.4% of variance explained. To validate the results of a PCA was performed HCA. The graph shows the dendogram obtained with an anomalous two samples of soybean and one olive. Moreover, there is a greater similarity between the classes of soybean and olive than for cotton. The observed behavior is explained by the distribution of fatty acids in triglycerides of molecules of each class. With this preliminary information SIMCA models were developed. For this, the samples were selected randomly to be the sets of training, validation and prediction. AH samples were correctly classified at 95% probability. The PCR of multivariate calibration model to predict the viscosity was more robust for MLR and PLS. The relative errors of prediction of the viscosity compared to the reference were less than 6,3%. Considering the observations, the *H low field NMR and multivariate analysis allow the classification of vegetable oils and their prediction of viscosity of a direct, non-destructive, non-invasive, without generating waste and rapid (30s).