AI-Driven Biofilm Detection for Sustainable Fermentation
Biofilm formation during industrial fermentation presents a significant challenge for food production companies like Quorn Foods. Biofilms accumulate on fermentor walls, causing lumps in the final product and often leading to premature termination of the fermentation cycle. This results in productivity loss, higher production costs, and increased energy waste, impeding the sector’s goal of sustainable, low-carbon food production.
An NBIC-funded Flexible Talent Mobility Accounts (FTMA) project between researchers at Teesside University and Quorn Foods developed an AI-driven biofilm detection system using microscopy images collected at different stages of Quorn’s fermentation process.
By leveraging transfer learning on Convolutional Neural Networks and integrating meta-information via multi-modal approaches, the model semantically segmented biofilm regions to predict biofilm percentage. This enabled the optimisation of fermentation cycle termination timing to reduce waste and energy consumption.
The AI model can identify biofilm formation in microscopy images and predict optimal fermentation termination times, directly addressing quality control and reducing process inefficiencies. This is expected to lower energy use and carbon emissions, aligning with “Clean Growth” objectives. Quantitative validation and deployment are ongoing.
By improving fermentation process control, the technology benefits the food production industry through enhanced product quality, reduced waste, and a lower environmental footprint. Indirectly, this supports public goals of sustainable food production and climate change mitigation.
Professor Annalisa Occhipinti from Teesside University said,
“NBIC’s support enabled access to valuable industrial datasets and facilitated close collaboration between Quorn fermentation experts (Mark Taylor) and AI researchers at Teesside University. This led to the development of a successful semantic segmentation pipeline based on a UNet architecture that can detect biofilm presence with high accuracy. The NBIC funding helped overcome data annotation challenges and supported iterative model improvements by providing industry feedback”.
Following the NBIC collaboration, the team successfully secured BBSRC funds to expand this research. This funding, led by Professor Claudio Angione at Teesside University, supports further development and scaling of the AI model, accelerating its translation into industrial applications.