The KIKERP project aims to explore the potential of AI for identifying and assessing used electrical appliances in the circular economy. With the help of an AI assistance system, employees can identify and classify unknown old appliances more easily and more quickly so that they can be properly refurbished, reused and recycled.
Using mobile devices, employees will be guided through a dialog to capture visual data, evaluate assessments, further narrow down the scope of the search with text and other metadata if possible and validate suggested decisions.
Robotic process automation (RPA) and software bots will be used to integrate AI-based technologies into cross-domain process landscapes in order to unlock efficiency and cost-cutting potential in future-proof digital ERP ecosystems. With the help of a cloud-based management platform, a process landscape for the remanufacturing and reuse of large household appliances is to be designed and implemented as an application demonstrator. This allows the fully automated identification of used appliances (product model), diagnostic operations, quality classification and quality assurance. It facilitates movement of goods at different stages in the logistics system, price evaluation and offer preparation as well as their synchronization with e-commerce platforms for selling the analyzed products.
Within the KIKERP project, Fraunhofer IPK is largely responsible for developing and integrating two key AI innovations.
For the first of these innovations, various multimodal AI modules are combined into a single architecture that can efficiently serve multiple applications. It will then be possible to feed this AI with different data in a dialog-based application until a well-founded classification and diagnosis of the inspected (used) electrical appliance can be provided to the user. For the first time, simulations are used to develop the AI as well as the data acquisition and processing in order to design, test, optimize and synthetically pre-train the architecture as efficiently as possible.
The second innovation comprises an automated data management system and continuous improvement of the AI based on process data feedback. The data management system will systematically sort the – sometimes redundant – masses of data obtained in the process using a new AI-based evaluation methodology and select the appropriate data for future training processes. This enables efficient training and data storage on the one hand and the largest possible database for AI training that is free of statistical bias on the other. Drawing on the data management system in combination with performance indicators, the aim is to automatically derive when and with which data the AI needs to be retrained in order to achieve the most significant improvement.
When integrated into the cloud-based process landscape envisioned by KIKERP, these two innovations together should facilitate and accelerate the identification and assessment of (used) electrical appliances. The dialog-based assistance system will make it possible to quickly onboard even untrained employees. Additionally, the process is documented automatically and digitally.