Comparison of two supervised methods pattern recognition for classification of middle petroleum distillates using Infrared spectroscopy

Abstract

In the oil refining industry, the use of crudes from several origins is frequent. This leads to considerable
variations in the chemical composition of the products obtained during the refining process. The application of the
chemometric techniques for pattern recognition has made possible the development of alternative methods that allow
the classification and quality control of several types of fuels in a quick way, using small quantities of samples, through
spectroscopy and chromatography methods. The objective of this study is the application of the supervised pattern
recognition methods: Soft Independent Modeling of Class Analogy (SIMCA) and Partial Least Squares Discriminant
Analysis (PLS-DA) in the classification of kerosene through infrared spectroscopy. This study is also intended to carry
out a comparative study of the results achieved with the application of both methods. The models developed allowed to
differentiate two kerosene groups with different chemical compositions, which were corroborated through mass spectrometry.
The comparative study carried out through a validation with external samples showed that the SIMCA model
had an adequate sensitivity and selectivity: it didn’t show false negatives and was unable to classify the turbo fuel
samples alien to the modeled system. Nevertheless, the PLS-DA model showed selectivity problems and was unable to
differentiate the turbo fuel samples.