NEURAL VIRTUAL SENSOR FOR MONITORING MEAD FERMENTATION

Abstract

Mead is a fermented alcoholic beverage based on water, honey, and yeast, normally strains of Saccharomyces cerevisiae. This beverage has a great historical and cultural importance, and despite this, its production still occurs in an empirical and artisanal way, which can result in several problems during the fermentation process. Bearing this in mind, it was developed a virtual neural sensor capable of predicting, during fermentation, the concentration of cells (X) and substrate (S) from simpler and faster-to-measure variables, namely: pH, °Brix and optical density. Based on experimental data from a mead fermentation carried out at 25 °C and 31º Brix, two strategies were implemented to obtain the sensor. In strategy I, the X and S variables were individually predicted by two ANNs, and in strategy II, a single ANN was responsible for the simultaneous prediction of the two variables. The best ANN was obtained from strategy II, containing 3 neurons in the input layer, 12 in the hidden layer and 2 in the output layer, trained with the Levenberg-Marquardt optimization algorithm and with activation function hyperbolic tangent in the hidden layer and linear in the output layer. For this ANN, Pearson’s correlation coefficient (ρ) was equal to 1 and the mean absolute percentage error was 5,46x10^-4 % e 8,24x10^-4 % for X and S, respectively. The sensor can be used to monitor and optimize mead production, to obtain high yield and productivity.