K3I-Cycling

The overall objective of Innovation Lab 2 is to significantly improve the mechanical recycling of post-consumer plastic packaging waste, both quantitatively and qualitatively, and thus to enable the sustainable use of plastics in both industrial and private everyday life. This goes hand in hand with the aim of increasing the acceptance of products made from recycled materials among manufacturers and consumers. In order to achieve this goal, the following two key points will be addressed in the innovation lab: By creating and providing a new, open and standardizable artificial intelligence (AI) interface for the cross-sector collection of relevant information in terms of a light packaging (LVP) product passport, the digital networking of all stakeholders along the LVP value chain will be enabled for the first time. Building on the collected data and applying AI methods, the optimization of the entire LVP cycle is then addressed. Here, both the AI-based optimization of the processes in the individual sectors (logistics, sorting, processing, regranulation/recycling, etc.) plays a role, as well as the AI-based overarching optimization of the entire value chain. The approach is realized by establishing an Artifical Neural Twin (ANT). This is a fully differentiable digital representation that has the ability to optimally adjust the individual components, and thus the overall system, to a global quality measure, while taking local conditions into account. In order to be able to develop the planned work on an industrial scale, the described points will be developed and optimized using an LVP sorting plant in real operation. Thus, a direct implementation is guaranteed, which is at the same time a strength and special feature of this project.

The goal of Innovation Lab 2 is the definition, development and use of a generally valid AI data interface, which each component within the overall system must fulfill. This makes it possible for the first time to map the value chain as an Artificial Neural Twin (ANT), whose individual layers correspond to the respective sectors (cf. Figure 1). The basis for the optimization of the individual layers, but also of the overall system, is the sector-specific as well as the cross-sector digital recording of the material flow as well as its properties. For this purpose, new approaches are being taken with regard to the sensory acquisition of all kinds of information. By determining a global quality measure, the neural network can be optimized. The AI interface allows further local quality measures specific to the individual sectors to be respected and taken into account for the individual layers. Furthermore, the AI interface allows the deviation from the global quality measure to be propagated back through the individual layers (sectors). Thus, the influence of individual

components along the value chain and identify critical elements. This also makes it possible to evaluate correlations along the value chain and, in particular, to adjust the parameters of the individual sectors (life cycle phases) on the basis of the back-propagated deviation in such a way that the deviation from the global quality measure is minimized. A direct consequence of such a setup is that when a change occurs in the value chain, be it wear or the replacement of a component, the overall system automatically readjusts itself optimally (as long as the new components implement the AI interface). Since optimization of the entire value chain can counteract local optimization of individual sectors, the ANT will also serve as an analysis tool for society and politics. In this way, a political momentum is created for the first time, for example, to identify bottlenecks in the cycle management and to enable an objective assessment of sustainability. Based on this, mechanisms can be developed and established (e.g. similar to CO2 dividends) that compensate for economic disadvantages in individual sectors while optimizing the value chain as a whole.