Despite advanced digitalization through CAD or digital twins in production, today's industrial processes are largely linear: raw materials are processed into products that become waste at the end of their life cycle. While the costs of production have steadily fallen and production volumes have increased, this growth is proving to be unsustainable. One example is the household appliance industry: its market volume of around 700 billion euros today is expected to grow to 1.2 trillion euros by 2032. With this growth, the amount of waste will also explode. One particularly critical example is electronic waste, almost two thirds of which consists of disused kitchen and laundry appliances. There are two key bottlenecks to dealing with this flood of material: the lack of skilled workers for manual sorting and the lack of intelligent, scalable solutions for evaluating old appliances.

The data acquisition pipeline is implemented directly in real remanufacturing and processing lines to assess the condition of products and components. © Indústria Fox

Researchers at the Fraunhofer Institute for Production Systems and Design Technology IPK are meeting this challenge with a new AI-based framework. They are developing an intelligent assistance system that enables better data collection and analysis as well as a well-founded evaluation of end-of-life products. This is an end-to-end system for intelligent data management. The technological basis for this are modern smartphones with their high-quality cameras. They allow data to be captured easily, in high quality and digitally. The assistance system is primarily designed for use by specialist personnel in industry, but can also be adapted for end users at home. This process-centered approach solves the problem at the root. If the end user can already understand the condition of their appliance and make an informed decision, this reduces logistical problems during sorting and relieves the entire downstream value chain. The collected data is structured and made available for continuous AI training and in-depth analyses of product life cycles.

From mobile scanning to precise analysis

An application example shows how this user-centered approach works: Vivek Chavan, researcher at Fraunhofer IPK, demonstrates how the user's perspective can be effectively captured, whether with a mobile camera or, in the future, with AI-supported smart glasses. The data captured can be multimodal and include audio comments, text or other metadata in addition to visual information. These are sent to a specially developed AI pipeline. The reward function of the AI model forces it not only to correctly identify the device, but also to precisely understand its visual state. To achieve this, the researchers in the BMFTR-funded KIKERP project worked closely with industrial partners Indústria Fox and Yes Technologies, which specialize in the remanufacturing of "white goods". The solution was developed using data from thousands of devices and is already being implemented in real use cases and directly in production.

A blueprint for circularity

The technology enables effective identification, sorting, quality assessment and decision-making. Its use transforms recycling from a cost factor into a profitable business area. Companies can maximize the value of their products beyond the first life cycle and tap into new revenue streams. The data-supported evaluation creates transparency and enables a precise calculation of the processing costs compared to the potential resale value. At the same time, the sustainability balance is measurably improved by reducing the_{CO2 footprint} through the extended use of products and the saving of primary raw materials. The key to this effect lies in the technological core of the system: a combination of intelligent data collection, sophisticated data management and consistently process-centered AI development. Because the system is designed to map and optimize real workflows, it is highly customizable. This fundamental approach enables it to be transferred to numerous other application domains, including the automotive industry, the energy sector and industrial mechanical engineering. However, the researchers' vision goes even further: entire product histories can be mapped through integration with digital twins. Coupled with robot-assisted disassembly, the AI decision can be translated directly into a physical action in the future, taking the level of automation in the circular economy to a new level. The project is funded by BMFTR and DLR Project Management Agency.

Source: Vivek Chavan, MSc; Prof. Dr.-Ing. Jörg Krüger, Fraunhofer IPK