AI and automation in the industrial bioeconomy

Three practical perspectives from the TransBIB Learning Lounge 

AI and automation in the industrial bioeconomy are no longer just a pipe dream – they are already transforming the way processes are planned, scaled and optimised. In the TransBIB Learning Lounge entitled ‘Automated, digital, future-proof’, three experts demonstrated how this works in practice: from the fields of research, simulation and industrial fermentation. 

What do AI and automation mean for the bioeconomy today? 

The industrial bioeconomy faces a twofold challenge: on the one hand, it is expected to replace fossil-based raw materials and establish sustainable value chains; on the other hand, it must match the efficiency of traditional industrial processes. This is precisely where digital technologies come into play. 

Data-driven decisions, automated workflows and machine learning systems enable more efficient use of resources, shorter development cycles and safer processes – across research, technology and industrial applications. 

TransBIB – the knowledge transfer network funded by the BMWE with 5.5 million euros – brings together 18 project partners, over 45 centres of expertise and more than 110 knowledge experts across Germany. The Learning Lounge is the open digital platform for low-threshold knowledge exchange from the regions, for the regions. 

How does an AI agent learn to design chemical process plants? 

Professor Dominik Grimm from TUM’s Straubing Campus provided a particularly fascinating insight into research. His question: Can machine learning design chemical processing plants – without any prior knowledge of chemistry or process engineering? 

The answer is yes – and the way to achieve this is through reinforcement learning. A flow chart is, at its core, a sequential planning problem: an AI agent must decide, step by step, which piece of equipment to use next in order to produce a target product of the desired purity. 

Prof. Grimm’s group extended AlphaZero-based algorithms so that they can be applied to any deterministic planning problem. During training, the agent competes against earlier versions of itself and continuously improves. 

The result: in more than 99 per cent of all cases studied – over 200 instances and four different chemical systems – the agent generated high-quality separation processes. In doing so, it independently discovered concepts such as azeotropic and extraction distillation: an agent on a par with a Master’s student in chemical process engineering. 

Digital simulation as a strategic tool for scaling 

Stefan Merkle, founder of Merkle CAE Solutions, presented the perspective of digital process simulation. His argument is clear: Scaling is a digital problem today. 

Many biotechnological processes work perfectly in the laboratory – but fail when scaled up. The reason usually lies in the physics: the Reynolds number changes with scale. What was still homogeneously mixed in the small-scale reactor suddenly forms dead volumes in the large-scale plant. Simulation makes these effects visible before they are cast in steel. 

Practical examples from the presentation: mixing in biogas plants, hydrogen explosions in electrolysers with pressure wave calculations in the millisecond range, and oxygen distribution in fermenters. In every case, simulation reduces uncertainty and speeds up development cycles. 

Time savings through simulation: realistically 20–30 per cent; in some projects, a factor of 20. The next step is the digital twin – a parallel simulation that runs in real time alongside the actual plant and serves as a basis for operational and management decisions. 

AI in Fermentation-Based Production – Opportunities and Limitations in Practice 

Dr Peter Welters, founder and CEO of Phytowelt, shared the perspective of an SME that is trialling AI in real-world fermentation processes – with all the successes and limitations that entails. 

Phytowelt has been developing bio-based production processes for flavourings and fragrances since 1998. Its most prominent product is R-alpha-ionone (raspberry flavouring), produced by fermentation from sugar – it is enzyme-pure and therefore has a clear advantage over petrochemical alternatives. 

Phytowelt made its first foray into AI in 2020: data from 20 fermentation runs were analysed. Result: a clear correlation between CO₂ production and alpha-ionone yield. This was followed in 2022 by an Interreg project, which identified an optimal feeding rate range. 

The current ZIM project ‘ENANTIONONE’, carried out in collaboration with the Technical University of Kaiserslautern, combines strain optimisation, process development and online sensor technology – with the aim of optimising alpha-ionone production in a data-driven manner and preparing it for industrial scale-up. 

What do all three perspectives have in common? 

Research, simulation and fermentation practice speak different languages – but they share the same fundamental conviction: those who think digitally today will have a head start tomorrow. The path to an industrial bioeconomy lies in data. Those who digitise processes, utilise sensor technology and systematically build up knowledge create the conditions – not only for AI, but for competitive bio-based production as a whole. 

TransBIB: Your network for the bioeconomic transition 

TransBIB is more than just a research network. In the one-stop shop , you’ll find bioeconomy databases, a certification system, an AI chatbot to act as a sparring partner, and a pool of expertise. An interactive map shows scaling infrastructures across Germany. 

Particularly useful: the matching platform. Here you can post technical queries, ideas or requirements as listings – or search through existing offers using keywords. Do you have a specific digitalisation project or are you looking for partners for an AI project? Get in touch – TransBIB brings the right people together.