Predicting Fermentation Rates in Ale, Lager and Whisky

Abstract

Recently there has been an increased interest in characterising the rates of alcoholic fermentations. Sigmoidal models have been used to predict changes such as the rate of density decline. In this study, three published sigmoidal models were assessed and fit to industrial fermentation data. The first is the four-parameter logistic model described in the ASBC Yeast-14 method. The second model is a nested form of the four-parameter logistic function, adding an extra parameter, creating the 5-parameter logistic equation., where an additional parameter was added to allow for asymmetry. The final model is a three-parameter logistic equation which describes the change in the Apparent Degree of Fermentation with time. The three models were compared by fitting them to industrial data from Australian and Canadian lagers, American and Scottish ales and Scotch Whisky fermentations. The model fits were then compared to one another with a technique developed by Akaike and a nested F-test. The Akaike information criterion compares the models and accounts for both the goodness of fit, and the number of parameters in the model. Finally, consideration was given to the establishment of prediction bands (that enclose the area that one can be 99% sure contains the true datapoints). Calculation of these bands was “challenging” but successful as the industrial fermentation data was heteroscedastic.

Keywords

• beer
• whisky
• logistic modelling
• fermentation
• heteroscedastic error prediction