Adulteration Detection in Olive Oil Using a PLS Model Augmented by Synthetic NIR Spectra
Olive oil is a high value product with huge global consumption which has always been an easy target for adulteration by lesser quality cheaper oils. There have been several successful attempts to use NIR spectroscopy for detecting such adulterants (Casale et al., 2014). However, these studies have generally suffered from lack of representative samples. In this work we have tried to tackle this issue by augmenting the calibration dataset by synthetic spectra generated form real oil spectra.
40 different samples of extra virgin olive oils from different countries and suppliers were collected for this study. Rapeseed, corn, groundnut, sunflower and vegetable oils were used as adulterants to spike olive oil samples at concentration levels of 0.5%, 1%, 2%, 5% and 10%. In order to get the widest possible range of mixtures, this was done using a random design which led to 265 samples. These were scanned in triplicate in transmission mode on a Bruker MPA FTNIR using 8 mm glass vials at a constant temperature of 50C (wavelength range: 1100-2500 nm).
Synthetic mixture spectra were then generated by linear combination of the real oil mixture spectra resulting in a total of ca. 3300 spectra covering the above adulterant range uniformly. Opus software (Bruker) and WinISI (FOSS Analytical) were used for data analysis.
Cluster analysis by PCA showed that providing the level of adulteration in the olive oil is greater than 5%, it is possible to determine (1) if adulteration had taken place and (2) to identify the adulterant with good degree of certainty.
PLS was then used for quantitative evaluation and detection limit determination of individual adulterants.
A global model including all adulterants was made which resulted in RMSEP=2.91 and RMSEP=1.23 (before and after augmentation respectively). This translates to an accuracy of about ±3% at 95% confidence level. Finally a successful attempt was made to build a model solely based on synthetics spectra followed by validating it against real spectra.