Fermentation monitoring by Bayesian states estimators. Application to red wines elaboration

Abstract

Winemakers must understand all chemical aspects involved and make the right decisions to obtain a high quality product. In a winemaking process, the tracking and control of certain variables are keys to achieve a proper fermentation. This paper presents state estimators design based on Gaussian processes, for on-line alcoholic fermentation monitoring in red wines. For this study, 18 fermentations of three different varietals, Cabernet Sauvignon, Malvec and Tannat, were analyzed to train and validate the estimators. Samples were taken from Merced del Estero, a San Juan industrial winery. Then, cell concentration was determined by neubauer chamber count, while ethanol and total sugars concentrations by infrared absorption spectroscopy. Results show a suitable prediction of cell and ethanol content when only substrate measurement is available. Furthermore, the proposed estimator is compared with a competitive approach (neural network) to highlight the suitability of Bayesian theory for this type of application. This paper provides a reliable monitoring tool, with low computational and economic cost to facilitate the work of winemakers.