Abstract
Wine consumers are increasingly health- and environmentally conscious. At the same time, white wine and rosé drinkers favour freshness and varietal aromas, which requires low-temperature regimes that extend fermentation time and increase energy demand. Additionally, global warming accelerates grape ripening which increases alcohol level in wine. To reduce cost and alcohol levels while maintaining quality, predictive tools that forecast how fermentation conditions impact fermentation time, and primary and secondary metabolite concentrations, can provide practical benefits to wineries by expediting oenological decisions-making and in turn reducing energy demand. Additionally, literature highlights static models in smart manufacturing suffer from performance degradation with data drift. In light of this, we successfully developed and evaluated pipelines for the automated design and training of three ML methods - support vector regression, random forest and artificial neural networks - to predict fermentation outcomes from ten initial features. The dataset employed captures the fermentation of nine commercially available Saccharomyces cerevisiae strains, each performed in triplicate, across two synthetic media and four temperature regimes, sampled at nine time points. The targets we focused on included time to complete fermentation, Ethanol, Acetate, given its undesired spoilage effects and ten secondary metabolite concentrations, given their synergistic impact on aroma. We were able to recommend effective pipelines for each task and identified our feature set contained adequate predictive signal for ethanol, acetate and fermentation duration predictions, but inadequate for the multi-output regression of secondary metabolite concentrations, which exhibited target heterogeneity. Ultimately, embedding these pipelines into optimisation frameworks, offers an actionable route to tailoring wine characteristics to evolving consumer preference, while reducing energy demand.