Brazilian automotive gasoline has been the target of constant adulterations in an attempt to raise profit margins as a result of the market flexibility and the ensuing increased competition. Gasoline adulteration with solvents is common because solvents present lower taxation in comparison to gasoline. In the face of this, the importance of developing low-cost analytical techniques such as spectroscopy to certify the quality and authenticity of fuels used in Brazil is conspicuous. This work used IR spectroscopy and multivariate techniques (principal components analysis-linear discriminant analysis, PCA-LDA) to determine gasoline adulteration by solvent and to identify the solvent added. The results show that FTIR associated with PCA-LDA is a powerful technique in the quality control of automotive gasoline. The method sensitivity was 8% v/v with 96% efficiency in the classification of adulterated and unadulterated gasoline and 93% efficiency in the identification of the type of solvent added.

No abstract

Chemometric data analysis was applied to chromatographic data as a modeling tool to identify the presence of solvents in gasoline obtained at gas stations in the Minas Gerais state. As a training set, 75 samples were formulated by mixing pure gasolines with varying concentrations of four solvents and analyzed by gas chromatography-mass spectrometry. Selected chromatographic peak areas were used in chemometric analysis. Sample distribution patterns were investigated with principal component analysis (PCA). Score graphics revealed a clear sample agglomeration according to the solvents added. Classification models were created with linear discriminant analysis (LDA). Because gasoline presents a very complex profile and the chromatographic data contains too many variables, two approaches were tested to reduce the dimensionality of the data before LDA. Fisher weights were used as an exclusion criterion of lesser variables, and the original variables were substituted for a few principal components obtained from the covariance matrix. To test the quality of the models, a test set with a total of 31 new samples was prepared using certified gasolines mixed with the same solvents used in the training set. Both models indicated the presence of solvent in gasoline effectively, failing only for samples whose solvent concentrations were low. The PCA plus LDA model was more efficient in signaling solvent-free samples, which reduced the number of false positive cases.

Chemometric data analysis tools were applied to chromatographic data to identify the presence of solvents in gasoline samples from gas stations in Minas Gerais state, Brazil. A training set of 75 samples was formulated by mixing pure gasoline with various concentrations of four complex solvents. The samples were analyzed by GC-MS, and the selected peaks were used in chemometric studies. Hierarchical cluster analysis, HCA, was used to search for sample distribution patterns according to the solvent added. K-nearest neighbor (KNN) was used to create a classification scheme to differentiate pure and mixed samples and to indicate the type of solvent present. HCA revealed a clear clustering tendency of samples containing the same solvent. However, only after the exclusion of lesser variables (peaks) by means of Fisher weights was it possible to separate samples with low solvent concentrations. After optimization of the KNN algorithm, it was possible to classify 88% of the samples of the training set correctly. To check the quality of the model, another group of samples was prepared with certified gasoline and the same solvents. The algorithm classified the great majority of the samples correctly once again.