Leila Aalizadeh
Fuel oil is one of the petroleum fractions and it can be classified into different grades according to its
composition and properties. Discrimination of the types of fuel oil is important in quality control,
environmental monitoring and industry. In this work, some chemometric methods were examined for
the classification of 264 fuel oil samples from three types of oil (furnace oil, gas oil and mazut oil) with
the aid of attenuated total reflectance Fourier transform infrared (ATR-FTIR) spectroscopy. For this
purpose, principal component analysis (PCA), extended canonical variates analysis (ECVA) and interval
extended canonical variates analysis (iECVA) were used. PCA was not able to discriminate fuel oils but
ECVA showed a good ability for classification purposes. On the other hand, iECVA was effective in
indicating the fingerprint sub-regions in the FTIR spectra and decreasing the model’s complexity. The
proposed models were validated by different statistical methods and excellent specificity and selectivity
were obtained. ATR-FTIR spectroscopy coupled with chemometrics helps to present a fast, powerful and
reliable approach for the discrimination of the grade of fuel oils which is important in quality control and environmental monitoring.
A discrimination approach based on the combination of FTIR
spectroscopy and chemometrics was developed to differentiate
three kinds of fuel oils (i.e. furnace oil, gas oil and mazut oil).
The proposed approach for the discrimination of these fuel oils,
which have similar physicochemical properties and FTIR
spectra, could be a simple and efficient method to replace timeconsuming and complex procedures which are currently
common for distinguishing oil classes and quality. Because of
the high collinearity in the spectral data, PCA was not able to
differentiate FO, MO and GO but the potential of ECVA for this
goal was conrmed using internal and external validation
procedures. On the other hand, a variable reduction was performed using a sub-region selection in iECVA. Building classi-
cation models based on iECVA not only indicated nger-printregions in the FTIR spectra but also decreased the complexity of
the ECVA modeling space. It was revealed that spectra in the
range of 2992.98–2827.13 cm1 are the best regions for the
discrimination of FO, MO and GO.