Sustainable Breweries
Real World Challenge
Breweries and pubs are under growing pressure to cut carbon emissions, reduce energy and water use, and improve sustainability — all while maintaining product quality and profitability. The challenge is multifaceted: Scope 3 emissions from ingredients, packaging, and logistics dominate the footprint (often 57–95% of total), but in-brewery processes like wort boiling, fermentation control, and cold storage are also major contributors.
Smaller producers face additional barriers: limited data collection, variability between sites, and the difficulty of applying advanced analytics without dedicated technical teams. Traditional monitoring methods are slow, resource-intensive, and offer little scope for real-time optimisation. As a result, breweries and pubs often run processes conservatively — over-cooling, over-cleaning, and over-packaging — to avoid quality risks, but at the expense of efficiency, cost, and sustainability.
A Machine Learning pipeline for energy and environmental prediction in FDCSs. The pipeline employs feature extraction and selection methods, utilises real-world datasets, and evaluates models using MAE and SHAP analysis for forecasting electricity consumption, as well as indoor temperature and humidity.
Our AI Solution
At the Food AI Lab, we combine open carbon accounting, IoT sensing, and machine learning to help breweries and pubs make sustainability decisions backed by real data.
Transparent Carbon Footprinting– We developed a free, open-source carbon calculator tailored to UK craft breweries, benchmarking emissions across brewing, packaging, and distribution. Analysis shows total footprints ranging from ~205–1483 gCO₂e/L depending on format, with packaging and Scope 3 supply chains as key targets for action.
Smart Fermentation Monitoring – Using low-cost ultrasonic sensors and ML, we predict alcohol concentration in real time (R² ≈ 0.95, MAE ~0.27–0.28), reducing manual sampling and enabling earlier intervention.
Energy-Efficient Cold Storage – Our tailored ML pipeline forecasts cellar electricity use, temperature, and humidity up to a week ahead. Random Forest models achieved the lowest MAEs (~151 and 385) on real pub datasets, supporting more efficient stock rotation and maintenance.
Collaborative Learning Across Sites – Transfer and federated learning approaches improve accuracy without sharing raw data, making AI viable even when each site has limited training examples.
By integrating real-time monitoring, predictive modelling, and actionable carbon data, we give breweries and pubs the tools to target the biggest impacts first, cut resource use without compromising quality, and communicate sustainability progress with confidence.
Partners and Funding: University of Nottingham, EPSRC