Abstract : The fermentation rate, the wort density and the ethanol concentration are estimated on-line, based on the measurement of the CO2 evolution rate. The future density evolution and the fermentation end-time are predicted using dynamic models based on neural networks and on the average fermentation kinetics. The decrease of the prediction error with time is demonstrated. The results are based on 33 industrial fermentations involving two sorts of beer. Implementation problems and practical difficulties are discussed.
