Researchers at TU Berlin’s Institute of Machine Tools and Factory Management (IWF) have developed a “hybrid intelligence” system combining knowledge from experienced workers and the advantages of artificial intelligence. The system combines the best of both worlds: Humans often intuitively understand correlations, such as those between quality characteristics of a product and various process parameters in manufacturing (including electric performance, temperature, pressure, etc.) based on prior experience. In contrast, AI is able to combine a great number of sensor data and analyze them in real time. As such, hybrid intelligence can not only provide insight into certain stages of the manufacturing process which would otherwise not be possible. The AI cognition-supporting assistance system (KI-kognitionsunterstützendes Assistenzsystem, KIKA) developed by the IWF also consistently learns from the large datasets as well as the direct feedback from workers. This could particularly benefit small and medium-sized companies which often lack the necessary in-house AI expertise
Every production machine is unique in a certain way due to the type of its tools, the component and material properties and their respective states in the production process. If the material and component properties or the tool wear change, the entire machine must be recalibrated step by step via the control system. Over years of trial and error, experienced workers in manufacturing acquire the ability to adjust the machine to changing boundary conditions in such a way that, in the end, components are achieved with an accuracy of sometimes one thousandth of a millimeter. In the process, they develop expert knowledge with which they can characterize processes and components and, if necessary, detect the smallest deviations. “However, many of these experts will be retiring in the next five to eight years. This means we have a very narrow window to transfer this experience to AI assistance tools in companies,” says Dr-Ing. Soner Emec, who is leading the project at IWF. “The challenge is teaching machines how to make good decisions just like humans can, even when the data basis is poor.”
Using AI to understand the internal workings of manufacturing
Despite these valuable human insights, a production team is also often faced with perplexities. For example, variations in material properties, wear and tear, or other undetected disruptions in the respective manufacturing phases can lead to poorer product quality. Here, AI support can provide insights into the complex interrelationships of the manufacturing process that would otherwise not be possible. The research team has already achieved initial results for the so-called additive buildup welding of metallic, large-volume components. In this process, an arc burns between a melting wire electrode and the workpiece. Similar to 3D printing, complex parts can be built up layer by layer from metal.
Integrating human practical knowledge
"The challenge here is that the components are sometimes manufactured over several days, and are often one-offs, for example in the construction of prototypes. During production, all process parameters are iteratively readjusted by workers to the material and component properties. This is not only labor-intensive, but often also consumes a lot of material," says Dr. Soner Emec. Through an AI-supported inline video sequence analysis of the welding process, slag inclusions, porosity or bonding defects can be identified during production and communicated to the workers in order to take countermeasures. The method used for automated pattern recognition also evaluates the current and voltage signals of the welding device, whose frequency is in the range of kilohertz. “We were already able to patent this process prior to the start of the project,” explains Emec1.
Workers’ practical knowledge is transferred to AI using “human in the loop” methods. For example, the AI shows images of the welding process and identifies spatters, certain surface textures, and tool positions on them. The workers at the machines then use a touchscreen to mark those features that are actually relevant to the quality of the welding process in this case. The AI learns from this experiential knowledge and derives better compensation strategies with each iteration.
Particularly beneficial for SMEs
By moving the computing operations in the KIKA-IPK project to the cloud servers operated by project partner PSI Software AG, companies do not need to have high-performance computers on site to train the AI models and use the AI assistance services. In the future, "AI service centers" funded by the German Federal Ministry of Education and Research (BMBF) will also be able to offer such cloud services free of charge. The KIKA-IPK module developed at IWF is also particularly adaptable and thus especially benefits small and medium-sized enterprises (SMEs). Until now, small quantities and frequent changes in production processes have prevented SMEs from fully automating. In addition, they rarely have their own AI expertise.
Tests under real industrial conditions
The developed system is currently being tested under real industrial conditions in two additive manufacturing scenarios to prove its resource efficiency potential: – Printing of large-volume metallic components by arc buildup welding (Wire Arc Additive Manufacturing, WAAM) at the project partner GEFERTEC GmbH – Printing of personalized medicines by so-called drop-on-demand (DoD) processes at the project partner DiHeSys GmbH.
Financed by the Federal Ministry of Research
The project is funded by the BMBF initiative Learning Production Technology – Use of Artificial Intelligence (AI) in Production (ProLern) as part of the German government's high-tech strategy. In order to transfer the results to other manufacturing processes such as turning, milling and drilling (so-called separating manufacturing), a project was launched in November 2022 as part of the initiative "Demonstration and Transfer Network AI in Production" (ProKI-Netz) in the BMBF funding program "Future of Value Creation - Research on Production, Services and Work". In this context, ProKI-Netz Berlin with more than 15 demonstrators and accompanying transfer measures is being created at the IWF.
Cooperation with BIFOLD and HPI
To facilitate operation and achieve further scaling, the team is collaborating with the Berlin Institute for the Foundations of Learning and Data (BIFOLD). Together, the researchers are integrating the AI methods and required frameworks into the data stream management environment NebulaStream, which is being developed by BIFOLD. A cooperation also exists with the newly founded AI Service Center Berlin Brandenburg at the Hasso Plattner Institute (HPI) in Potsdam to make the developed AI applications accessible to a large number of companies and organizations.
A detailed article about the hybrid intelligence project has also been published in the Fraunhofer Institute for Production Systems and Design Technology’s magazine “Futur.”
The German Federal Ministry for Economic Affairs funded the research as part of the Central Innovation Program for SMEs. In Sissach, Switzerland, this new technology has already been successfully used in the construction of a kindergarten.
1TU Berlin patent for automated pattern recognition: “A Method for the Scalable Real-Time State Recognition of Processes and/or Sub-Processes During Production with Electrically Driven Production Plants” – EP 3 705 964