K3I-Cycling: Using AI to achieve better-sorted plastic waste and more reliably assessed recycled material qualities
13.10.2026
From hidden lithium batteries to digital representations of entire recycling chains: K3I-Cycling demonstrates the wide range of applications for artificial intelligence in plastics recycling. This is not just about improving sorting, but also about how recycled materials can be assessed more reliably and utilised in a more targeted manner.
At a Glance
- K3I-Cycling develops AI-powered solutions for sorting, recycling, and recycled plastic quality assessment.
- An AI tool helps analyze material data and guide decisions about further processing.
- The Artificial Neural Twin (ANT) digitally models the entire recycling chain, showing how individual process steps interact and affect the quality of the resulting recycled plastic.
- DangerSort uses X-ray technology and AI to detect hidden lithium batteries and other foreign objects in the material stream.
Twenty years ago, sorting often ended where materials could not be clearly distinguished. Today, packaging waste can be recorded much more accurately using cameras, sensors and other measurement methods. Systems recognise material types, foreign substances and quality characteristics, and feed this information directly back into the ongoing process.
A glance inside a modern sorting plant shows just how far this development has come: films, cups, packaging scraps and composite materials pass by on a conveyor belt. Amongst them lies a lithium battery, almost completely obscured by other material. An X-ray system detects it nonetheless; an AI evaluates the image, and shortly afterwards the object is ejected by a jet of air.
The fact that such systems are possible today is the result of a fundamental shift. Stricter requirements for recycled materials, regulatory guidelines and advances in sensor technology, data processing and machine control have significantly transformed plastics recycling in recent years.
This is precisely where the K3I-Cycling project comes in. It combines materials analysis, sensor technology and artificial intelligence with a central question: How can a recycled material be extracted from a heterogeneous waste stream, the quality of which can be assessed as reliably as possible and used specifically for a particular application?
When material data informs decision-making
Composition and condition vary greatly, particularly in the case of post-consumer plastics: ageing, contaminants or product residues can influence further processing. “Post-consumer waste is always a complex mixture,” says Dr Simon Gölden from the Fraunhofer Institute for Structural Durability and System Reliability LBF.
K3I-Cycling combines materials analysis with machine learning. An AI-supported tool collates measurement data and can derive specific recommendations for action from it – for example, whether a recycled material can be reprocessed straight away, requires an antioxidant, or is no longer suitable for practical use.
The highest possible quality is not automatically the most important factor here. What matters is whether the material is suitable for the specific application. Or, as Dr Gölden puts it: “What one user considers to be inadequate property profiles may be sufficient for another.”
ANT: Not just one machine, but the entire recycling chain
The quality of a recycled material is not determined only at the very end. A great deal happens beforehand: the material is collected, sorted, washed and processed. Each of these steps can influence the final outcome.
This is precisely where the’s Artificial Neural Twin (ANT) comes into play. Put simply, it digitally maps the entire recycling chain and brings together data from the individual stages. This makes it easier to understand what happens when something is changed at any one point.
For example: a customer may not require the highest possible purity for their application. In that case, the ANT can show whether costs can be saved elsewhere – for instance, in the washing process or in sorting accuracy.
Data from collection or logistics can also be incorporated, where available. This means that it is not just a single machine that is considered, but the interplay of the entire recycling chain.
The idea behind this is simple: Not every step needs to perform at its absolute best on its own. What matters is that, in the end, the material quality produced is exactly what is actually required.
DangerSort: When AI detects hidden batteries
Whilst the ANT takes the entire recycling chain into account, DangerSort focuses on a very specific aspect: hazardous foreign objects in the material stream.
This brings us back to the lithium battery mentioned at the start. The system combines a high-throughput X-ray system with AI-supported image analysis. This enables it to detect even batteries that are partially or completely concealed amongst other waste or firmly embedded in products. They are then ejected from the material stream using a jet of air.
For sorting facilities, this is more than just a matter of ensuring clean material flows. If a battery ends up in an industrial shredder and is damaged there, it can cause a fire. “The main priority in sorting is to remove batteries,” explains Dr Gölden.
DangerSort can also detect other problematic foreign objects, such as silicone cartridges. However, the focus is on lithium batteries – and thus a very tangible contribution to safety in sorting plants.
What this means in practice
Ultimately, it comes down to practical decisions: Which material can still be processed effectively? What kind of treatment is required? And for which applications is its quality sufficient?
This is precisely where the approaches from K3I-Cycling are intended to help:
- Recyclers can assess material flows more accurately and manage processing steps in a more targeted manner.
- Plastics processors and packaging manufacturers can better assess whether a recycled material meets the requirements of their application.
- Sorting facilities benefit from solutions such as DangerSort, which can detect problematic foreign objects at an early stage and remove them.
For this to work, however, suppliers and customers must have a shared understanding of what constitutes a specific grade of recycled material. This is where DIN SPEC 91446 comes in. It sets out criteria by which recycled plastics can be described and compared with one another. According to Dr Simon Gölden, findings from the analysis and evaluation of recycled materials from K3I-Cycling have been incorporated into this specification.
With ANT and DangerSort, the project also demonstrates how data and digital tools can be utilised directly in sorting and recycling.
This brings us full circle: plastic waste remains diverse. AI cannot make these differences disappear – but it can help us to understand them better and use that understanding to make more appropriate decisions for the next stage of processing.
Did you know?
Even the odour of a recycled material can be analysed using measurable markers. Within the project, methods have been developed that detect specific chemical compounds and can derive information about odour characteristics from them – a factor that may be relevant for the subsequent use of recycled materials.
