Process optimisation of beer fermentation through dynamic simulation

Beer production involves three essential, lengthy process stages: brewing, fermentation and conditioning. Process times have been established over many years' experience, and they vary due to different company philosophies. A commonly held belief by the purist brewer is that shortening fermenting or conditioning time compromises beer quality, as today's lagers can still take 4­6 weeks to produce. Shortening process times, however, could significantly impact plant capacity and consequently production efficiency. The highly complex chemical reaction system comprises over 600 species (Vanderhaegen
et al., 2006), rendering rigorous dynamic modelling extremely uncertain. A reduced-­order dynamic fermentation model must therefore consider only the key reaction pathways. This paper presents a computational implementation of two differential­-algebraic equation (DAE) models. The Gee and Ramirez (1988) model (five ODEs) considers three sugars, ethanol and a biomass form; the de Andrés-­Toro (1998) model (seven ODEs) considers a single sugar, ethanol, three biomass forms and two undesirable by­products (ethyl acetate, diacetyl molecules). Temperature profile inputs (Carrillo­-Ureta et al., 1999, 2001; Xiao et al., 2003; de Andrés­Toro, 2003) have been successfully used to validate both codes. Process optimisation uses the validated dynamic models to compute the minimum fermentation time and optimal temperature manipulation profile, while adhering to prescribed composition quality constraints. In contrast to some prior studies, we emphasise the use of real industrial plant data for model validation: our goal is to predict with confidence the effects of varying fermentation time (and other key parameters) on product quality, in order to optimise processing times, and potentially plant capacity.