Abstract (EN):
The quality control of a commercial gasoline includes the assessment of several physical-chemical properties. The main objective of the present work is to investigate the simultaneous use of NIR and H-1 NMR data to accurately estimate 13 of those characteristics [i.e., density, vapour pressure, percent evaporated at 70 degrees C, percent evaporated at 100 degrees C, percent evaporated at 150 degrees C, final boiling point, research octane number, motor octane number, aromatic content, olefinic content, benzene content, oxygen content and methyl t-butyl ether content]. The data driven models implemented in this work were single PLS using a concatenated block including two data types (i.e., NIR and H-1 NMR), and serial PLS using the different data blocks separately. This work also aims at comparing the predictive capability of these regression models with the models calibrated with one type of spectroscopic data only. The optimum complexity of each model (i.e., number of latent variables) has been determined according to the repeated cross-validation strategy, consisting of 100 repetitions of a cross-validation loop. Serial PLS models have been proven effective in property estimation and, for most of the cases, outperform the models developed using a single spectroscopic technique. This proves the complementarity of both spectroscopic techniques, which are based on distinct principles. The developed models yielded accurate predictions of 11 of the 13 gasoline properties within the reproducibility of ISO standards. The implementation of these techniques in industry enables a significant reduction in the total time needed for gasoline quality control and the simultaneous use of two different spectroscopic techniques allows for more reliable estimations of the properties in study.
Idioma:
Inglês
Tipo (Avaliação Docente):
Científica
Nº de páginas:
10