Comparison of non-invasive NIR and Raman spectrometries for determination of alcohol content of spirits

Abstract

The suitability of non-invasive NIR and Raman spectrometries for determination of % ethanol content has been investigated. Samples of whisky, vodka and sugary alcoholic drinks were analysed in 200 mL (flat) and 700 mL (round) glass bottles. The NIR spectrometer used double transmission measurements and was limited mainly to analysis of the signal produced at about 10,000 cm −1 by water and ethanol in the samples. The Raman measurements, produced using a 785 nm laser, were based on a sharp signal from ethanol at 880 cm −1. A multivariate calibration model, based on a combined PCA–PLS algorithm, was required for analysis by NIR spectrometry, whereas a univariate calibration model was suitable for Raman spectrometry. Both techniques were limited to measurements in clear glass bottles as strong absorption/fluorescence occurred with coloured glass bottles. Bottle-to-bottle variations contributed the largest uncertainty to the measurements obtained for a 20% (v/v) ethanol solution in flat bottles: 2.3% R.S.D. for NIR spectrometry and 2.2% R.S.D. for Raman spectrometry. For 700 mL round bottles, which have stricter manufacturing tolerances on glass thickness, the bottle-to-bottle variability for Raman spectrometry was 1.4% R.S.D. When spirit samples with ethanol concentrations in the range 19.9–61.7% (v/v) were analysed, the precision (average R.S.D.) was 0.4 and 0.5% for NIR (flat bottles) and Raman (round bottles) spectrometries, respectively, and the average accuracy was 2.1 and 2.9%, respectively. When a calibration model constructed from NIR data acquired on 1 day was applied to data sets collected over a 15-day period, the average error was 3.9%.