Whereas human knowledge has so far played an important role mainly as a source of data when artificial intelligence is being trained, the Fraunhofer IPT, together with eleven consortium partners, is going one step further in the research project "GeMeKI - Generalization of human-centered AI applications for production optimization": The research team is developing the components for a user-friendly expert system for the three selected manufacturing processes of joining, cutting and forming, which focuses on people rather than software applications. The aim is to improve human-machine interaction in both directions and in this way also to increase the acceptance of digital assistance systems in everyday operations. The result of the research project is an overall system comprising AI, sensor technology and production technology that - in contrast to the isolated solutions frequently used to date - can be seamlessly integrated into the value chain and adapted to any other fields of application.
For the users, there are always several interfaces for interaction in this system: Instead of the usual unidirectional process, in which the human evaluates the AI's found solutions but does not receive any feedback himself, a dynamic, bidirectional process is to take place here. This means that users can intervene more actively than before in the application of the solutions found, correct them independently, and use sensors to generate new, improved raw data that enrich the AI models with further information.
In the project, the Fraunhofer IPT is focusing on cold forming processes, for example in the automotive or aerospace industries. These require good specialist knowledge on the part of the machine operators who, as experts of many years' standing, often recognize the wear of complex progressive and transfer tools intuitively or on the basis of the noises that change over time when the tools are used. In the "GeMeKI" project, tool manufacturer Franz Pauli GmbH & Co KG, Meastream GmbH and the Fraunhofer IPT are investigating how predictive tool maintenance and compliance with quality standards can be improved through the use of AI and augmented reality (AR). Using the AR system, workers can mark defective component locations. This defect data flows back into the AI system and enriches the AI's learning base. The system is supported by a laser marking unit that is integrated into the mold. Based on a unique laser coding, the manufactured workpieces can be identified by the AI system. In this way, the system can derive correlations between workpiece quality, process control and tool condition from the complete database of the manufacturing process. The information helps workers with documentation and process diagnostics to reduce scrap and unplanned maintenance. In this way, between five and ten percent of manufacturing costs could be saved in contract manufacturing in the future.
The three example applications serve the research partners to set standards for the development and introduction of AI-based expert systems in production. In all three cases, therefore, the transferability of the solutions developed to other application fields is also taken into account. The project consortium's goal is to reduce AI learning time by up to twenty percent through more human-centric development of AI tools and the integration of augmented reality. Its introduction in companies, which today takes several weeks to several months, is expected to be shortened by up to 25 percent through human-centric, digital introduction and support concepts.
The project partners Aixbrain GmbH, MT Analytics GmbH and Youse GmbH also provide support in developing suitable services and implementing the AI systems in the participating companies. Thus, at the end of the project, in addition to new basic knowledge about the control loops of human-machine interaction and the use of artificial intelligence, an overall package of three best practices is available whose concepts can be easily adapted to further use cases and manufacturing scenarios.
The research project "GeMeKI - Generalization of human-centered AI applications for production optimization" is funded by the German Federal Ministry of Education and Research (BMBF) as part of the research program "Innovations for tomorrow's production, services and work" under the funding code 02P20A111.
Project coordinator: