34384

Monitoring the production of multiple alcoholic beverages at industrial scale by using FT-NIR spectroscopy

Leonardo Ramirez-Lopez 1 ; Giuseppe Ferrari 1 ; Giovanni Campolongo 1 ; Margarete de Koning 1 ; G Benetti 2 ; C. Furiosi 2 ; D. Cagni 2 ; I Truzzi 2
Favorite this paper

INTRODUCTION
The strict regulations for the production of alcoholic beverages (and the quality specifications defined by the producers) demand efficient analytical methods for quality control in terms of reliability and costs. Furthermore, in order to identify and cope with different factors that might hamper the quality and consistency of the final product, it is necessary to carry out a detailed physico-chemical monitoring of the main processes involved in the production. On the other hand, to be compliant with the strict regulations on the declared alcoholic content of beverages, analytical methods with negligible uncertainty are essential. Despite traditional techniques of beverage analysis are highly accurate, they are time-consuming and expensive and therefore not suitable for continuous quality control, especially in large scale production. Fourier-Transform Near infrared spectroscopy (FT-NIR) offers an efficient solution to overcome the drawbacks associated to the analysis of different components for detailed process monitoring. In this respect, the main objective of this work was to assess the usefulness of FT-NIR for monitoring the production of alcoholic beverages at an industrial scale.

EXPERIMENTAL
A total of 487 samples comprising about 20 different alcoholic beverages produced by ILLVA Saronno were selected for this research. For all the samples the alcohol content was determined by the specific gravity (SG) method. The SG was in turn determined by using the oscillating U-tube method. All the samples were scanned by using a BÜCHI N-500 FT-NIR spectrometer (BUCHI Labortechnik, Flawil, Switzerland) in order to obtain their NIR spectra (10,000 - 4,000 cm−1). The sample set was randomly split into calibration (75%) and validation (25%) subsets. For each property, a quantitative NIR model was fitted by using the partial least square regression algorithm on the calibration subset. To assess the predictive performance of these models they were applied to the validation subset and the root mean square error (RMSE) and the coefficient of determination (R2) were computed.

RESULTS AND DISCUSSION
The FT-NIR model of alcohol content exhibited a remarkable prediction performance (RMSE = 0.05%; R2 = 0.99). From a practical perspective, this high accuracy and precision allows the users to use the NIR predictions to infer the sort of alcoholic beverage that is being analyzed (from a known group of beverages). This is due to the fact that: i. alcohol content diverges considerably among most of the beverages and ii. the alcohol content intra-variability for the beverages is low. Taking into account that at industrial scale the same production plant infrastructure is often used to produce multiple sorts of alcoholic beverages, it is necessary to monitor the purity of the product being produced at a specific time. This is done to ensure that the bottled product does not contain residues (<0.05 %) of the previous one, but just the one currently under production. Overall, in this study we show that NIR can be used for monitoring both the quality and purity of a given product in a detailed fashion (~20 measurements per day per production line) in large scale production of alcoholic beverages.