Rapid screening of designer fuel frauds by Raman spectroscopy

•A portable Raman spectrometer was used to analyze diesel fuels aiming at the detection of fraudulent mixtures containing a heavier hydrocarbon fraction from lubricant / waste oils.•More than 200 diesel fuels collected from gas stations, fuel deposits and oil refineries have been tested, among which 28 resulted “designer fuels”, as verified by GC and distillation parameters.•Multivariate analysis of the Raman data, in particular Linear Discriminant Analysis (LDA), proved superior when classifying the diesel fuels as either compliant or containing a heavier fraction with zero false negatives and less than 4 % false positives. Designer fuel fraud consists in the smuggling of modified diesel blends as engine lubricant oils and their illegal trade avoiding payment of the excise duty applied to energy products. The fraudulent mixture contains regular diesel fuel plus a heavier hydrocarbon fraction, originating from waste automotive lubricant or cheap, residual base oils. Raman spectroscopy was tested as a rapid in-situ screening method to separate regular diesel fuel samples from those suspected to contain a heavier component, and thus demanding a more extensive characterization. The Raman fingerprint region from the screened sample is matched to purposely created spectral libraries of compliant and non-compliant diesel fuels using the instrumental search algorithm. Overall, 177 compliant fuel samples and 28 non-compliant samples (all designer fuels with a confirmed heavier fraction and/or anomalous distillation parameters) were measured. The designer fuels were all positively identified, with ∼18 % false positives. Subsequently, the Raman data-set was studied by Principal Component Analysis (PCA) and then classified as either compliant or non-compliant using Linear Discriminant Analysis (LDA). PCA using up to three principal components for data visualization shows only an incipient separation but still a partial overlap between compliant and non-compliant samples. LDA, on the opposite, performed superiorly in the binary classification task, with no false negatives and less than 4 % false positives. [Display omitted]