The control of large-scale industrial bioreactors is one of the most complex tasks in the process industry. In the often huge plants, the smallest differences in temperature, pH value or oxygen content determine the success or failure of a process. An interdisciplinary research team at Graz University of Technology (TU Graz) led by Stefan Radl from the Institute of Process and Particle Technology has now presented the CLARA software toolbox, which can be used to simulate the processes in the reactors 10,000 times faster, in the publication Chemical Engineering Science . At the same time, the calculations are so precise that the simulation results can be used in practice to control large-scale plants, for example in fermentation, the energy utilisation of biomass or polymer production. The solution was developed and financed as part of the TU Graz lead project DigiBioTech .
CLARA (CLustering AlgoRithm Austria) automates the creation of reduced order models (ROMs) using unsupervised machine learning methods. This makes it possible to calculate detailed reactions and mass transport processes in complex flows more efficiently. “Up to now, simulations based on computational fluid dynamics (CFD) have mainly been used for the calculations,” says Stefan Radl, “they are also very accurate and detailed, but also very computationally intensive. Simulations that depict real processes over days or weeks might take months to calculate.”
The core of CLARA is an AI-supported clustering process. Millions of data points from shorter CFD simulations are automatically grouped into a manageable number, between five and 50, of interconnected zones, known as compartments. What previously had to be done manually and with a great deal of expertise is now automated and mathematically precise. Even with complex multiphase flows, in which solid, liquid and gaseous material flows are mixed in the bioreactor, the accuracy is maintained.
The advantage of this approach lies in the drastic reduction in the complexity of the models to be calculated, while the result data continues to be highly informative. This means that important physical effects are retained despite the greatly increased calculation speed. For example, dead zones in the bioreactor with insufficient mixing or areas with high shear forces can still be reliably taken into account. The species exchange between gas and liquid phases is also modelled realistically, which is a decisive factor for biotechnological processes.
A particular focus was on ensuring physical consistency. Simplified models often suffer from inaccuracies in the mass balance. For example, when an organism grows in a bioreactor, at a certain point it needs more gases and other nutrients. Simplified models are not very conclusive here, especially at large scales. For this reason the researchers at TU Graz developed algorithms for optimising ROMs that guarantee constant mass conservation. Initial validations using analytical benchmarks and detailed CFD data confirm their high accuracy.
“The potential applications of CLARA are diverse and range from reactors for CO2 capture to processes in the production of battery cells,” says Stefan Radl. “The technology offers particularly great potential for the development of digital twins that can map and optimise processes in real time. We are now making CLARA available as an open source tool in a public repository . All the tested features are included, and we are gradually introducing new developments.” CLARA is already in use as part of a doctoral thesis at the COMET centre BEST and at the TU Graz spin-off SimVantage.
Publication : Compartment Modelling of Multiphase Reactors using Unsupervised Clustering
Authors : Michael Mitterlindner, Maximilian Graber, Regina Kratzer, Markus Reichhartinger, Stefan Radl
In : Chemical Engineering Science
DOI : https://doi.org/10.1016/j.ces.2026.125455
Chemical Engineering Science
Computational simulation/modeling
Not applicable
Compartment modelling of multiphase reactors using unsupervised clustering
27-Oct-2026