Advanced Real-Time Control and Optimisation of Mixed-Culture Beer Fermentation

Following the first isolation of an individual yeast strain in 1883, monoculture fermentation emerged as the primary method for fermenting nearly all beer produced today. However, evolving consumer preferences have spurred the craft beer movement, resulting in a considerable increase in beer diversity. Unfortunately, traditional process control methods used to ferment consistent monoculture beers are inadequate to address the complexities of producing craft and mixed-culture fermented beers. This thesis addresses this challenge by first replicating well-established mechanistic models for monoculture beer fermentation in order to establish a baseline for acceptable performance. Building upon this foundation, a data-driven understanding of both monoculture and mixed-culture beer fermentation processes is subsequently developed. Finally, an LSTM-based plant model is utilised within a model predictive control (MPC) framework for fermentation temperature scheduling. The work in this thesis draws significantly upon data. Therefore, significant efforts have been made to produce a dataset comprising monoculture and mixed-culture beer fermentation processes in pilot and small-scale fermentors. Furthermore, a monoculture fermentation dataset provided by Sennos, composed of real-world commercial-scale monitoring within craft breweries, has been used to support this research. Visualisation leveraging data processing methods has been used to develop an understanding of real-world beer fermentation processes. The replication and parameter tuning of a first-principles mechanistic model provides a standard against which to assess the novel approaches proposed in this thesis. Advanced empirical models trained using the datasets have demonstrated robust performance across a vast fermentation configuration space and accurate forecasting capabilities. Overall, this thesis provides robust models for monoculture beer fermentation process prediction and inferential measurement with a particular focus on serving the needs of the craft beer sector. Furthermore, the development of a mixed-culture beer fermentation dataset and associated predictive modelling efforts provides a proof of concept for further study.