The smart pollutant magnet

Medicines, chemicals, heavy metals – things that are useful in our everyday lives can pose a threat to the environment. A particular collection point for these substances is wastewater – from households as well as from industry. “Wastewater is like a mirror of our society,” says Prof. Dr. Matthias Kohl, head of the “Data Science for Life Sciences” research group at Furtwangen University. “You can see what we consume in it – but unfortunately, many of these substances remain in the environment much longer than in our bodies.”
In recent years, microplastic particles in particular have come into focus. They can accumulate pollutants; a process experts refer to as sorption. “You can think of it as similar to a pollutant magnet,” explains Kohl. “Depending on the type of plastic and pollutant, the molecules ‘stick’ to varying degrees.”
PolymerActive GmbH in Offenburg is exploiting precisely this principle. The company – founded by former students of Furtwangen University – upcycles microplastic waste and transforms it into so-called adsorbents – particles that can absorb pollutants from water. At HFU, Prof. Kohl's team is investigating how data science and artificial intelligence (AI) can be used to optimise these particles in a targeted manner.

HyplastDB: What AI learns from hundreds of experiments

In theory, it sounds simple – different plastics have different sorption properties. This means that for every type of contaminant, an ideal combination of plastics could be found to purify wastewater as efficiently as possible. In practice, however, this is a mammoth project. “You can't test every conceivable combination of plastic, pollutant, and environmental condition in the laboratory,” says Kohl. “That would take a very long time and be very expensive.”
In the Invest BW project “HyplastDB,” the team is therefore relying on AI models that predict sorption properties. To this end, data from several hundred experiments was compiled from scientific literature. What sounds simple turns out to be detective work; inconsistent representations, missing test parameters, different units.
“We were faced with a data landscape that resembled a patchwork quilt rather than a neat table,” Kohl describes. With the help of statistical methods for so-called imputation – i.e., filling in missing values – the researchers were able to close gaps and reveal correlations. Perhaps the most surprising finding, “A relatively small set of well-chosen descriptors is sufficient to make state-of-the-art predictions despite all the heterogeneity,” says Kohl. In other words, if you know the right characteristics, AI can very accurately estimate how strongly a pollutant will stick to the “plastic pollutant magnet” – without any individual tests in the laboratory.

reABSorb: The surface as a fine-tuning screw

While HyplastDB is about learning sorption properties from existing data, the ongoing “reABSorb” SME innovation project at the university goes one step further. Led by Prof. Dr. Magnus Schmidt, it pursues an ambitious goal – to adapt the surface of the particles produced by PolymerActive so that certain pollutants adhere particularly well and others as little as possible.
“Ideally, we want to build a pollutant magnet that only likes certain ‘adhesive partners’,” Schmidt explains. “A particle that binds drug residues very efficiently, for example, but leaves other substances largely untouched.”
The challenge lies in the production process itself. Speeds, temperatures, dwell times in individual steps – all of these factors influence the final appearance of the particle surface and how well it functions as a “pollutant trap.” This is where Prof. Kohl's team comes in again; it is developing AI models to monitor and control the production process.
“We are trying to learn from process data and subsequent sorption properties,” explains Kohl. “In other words – which setting at the beginning leads to which quality at the end. Our goal is to optimise production based on data.” In the long term, the aim is to create particles with precisely defined sorption properties – tailor-made tools for use in sewage treatment plants or industrial cleaning processes.

Artificial intelligence with practical benefits

The two projects at Furtwangen University show that artificial intelligence can be an important tool in environmental protection. “Ultimately, it's about using data to better solve real environmental problems,” Schmidt summarises. “If we succeed in using AI-controlled plastic adsorbers specifically for pollutants, we will be contributing to the protection of our waters – and thus to the health of us all.”