Modelling beer fermentation variability

In this paper we present the outcomes of three different approaches to characterising beer fermentations, with the particular aim of predicting the likelihood of the target 'present gravity' (PG) being reached within a given time window. The study uses data collated at real brewery sites, and from three different beer qualities. The approaches include: the modelling of the PG curve by a mathematical function; a nearest neighbour (NN) approach; and the generation of centile curves. We show that it is useful to combine these approaches; a software package allowing them to be integrated has been developed, which enables an informed judgement to be made as to whether a given fermentation deviates from normal behaviour.