"visIT" science journal: Insight into our current research

In "visIT", our science journal or theme brochure, we inform you in a concentrated form about current focal topics of our research. On this webpage, you can find all issues in digital form (mainly in German language). If you wish, we will be happy to add you to the distribution list for the digital or print edition - you will then receive visIT two to three times a year, free of charge. Simply write an e-mail to: publikationen@iosb.fraunhofer.de

The four-page newsletter InfOSB, which informs about recent developments in all the institute's business units (in German), accompanies visIT. visIT subscribers receive it automatically.

In March 2022, after two years of construction and with support and funding from the federal government and the state of Baden-Württemberg, we launched the Karlsruhe Research Factory. With our chosen motto at the time—“Research Factory for AI-Integrated Production”—we are—more so than we realized back then—in step with the current trends in German industry. Meanwhile, projects are underway at the Karlsruhe Research Factory with various partners: startups, medium-sized machine builders, BMWE projects such as Catena-X and Factory-X, and a DFG research group on “immature production processes.” 

Its counterpart in Lemgo, SmartFactoryOWL, has developed since its inauguration in 2016 into a widely recognized real-world laboratory for smart automation, energy optimization, and, more recently, cognitive robotics. The core idea behind both facilities is to make Industry 4.0 technologies literally tangible: This makes them understandable, their benefits for developers and users transparent, and it also becomes clear what potential and development opportunities still lie within them. The same applies to the use of machine learning and artificial intelligence: Our research factories help identify what AI can do, what possibilities it offers, but also where human creativity and expertise remain irreplaceable.

Research factories serve as showrooms, test beds, and pilot environments for new materials, new production processes, new machinery, and new IT solutions—specifically, how they interact: It is only through interdisciplinary collaboration and at the interfaces between fields of expertise that innovations emerge with the potential to win over industrial users and ultimately gain widespread adoption in practice. In Karlsruhe, the Fraunhofer ICT, the Institute for Manufacturing Engineering (wbk) at KIT, and the Fraunhofer IOSB work closely together; in Lemgo, TH OWL and Fraunhofer IOSB-INA collaborate.

In this issue, marking the fourth anniversary of the Karlsruhe Research Factory and the tenth anniversary of SmartFactoryOWL, we present several examples that have been realized in our research factories; you are cordially invited to share your own questions with our contacts and then solve them together with us.

Content

The links will take you directly to the corresponding page of the ePaper (German version):

• KI in der Produktion:vForschung, Anwendung und Transfer in vernetzten Fabriken
  Florian Mayr, Constanze Hasterok, Nissrin Heymann
• KI-Methoden zur schnellen Ertüchtigung unreifer Produktionsprozesse – DFG-Forschungsgruppe FOR 5339
  Jürgen Beyerer, Frank Döhner
• Souveräne Datenräume in der Praxis – Vom Messedemonstrator zur kognitiven Forschungsfabrik
  Boris Schnebel
• Next Level für die Forschung: Realproduktion in der SmartFactoryOWL bringt Forschung und Industrie zusammen
  Jürgen Jasperneite, Nissrin Heymann
• Wenn Batteriezellpoduktion lernfähig wird – Ansätze des wbk für eine effiziente Produktion
  Jan-Niklas Sturm, Jürgen Fleischer
• Autoinspect – Multisensorielle Inspektion großer Werkstücke zur Inline-Qualitätskontrolle
  Ilja Kaufmann, Dirk vom Stein
• KI-unterstützter Assistent für die manuelle Montage – Von automatischen Arbeitsanweisungen bis zur Zukunftsvision der humanoiden Assistenz
  Frederik Diederichs, Matthis Leicht
• SafeInLoc – Software-definierte Sicherheit für dynamische Produktionsumgebungen
  Holger Flatt, Dennis Sprute
• GLUE – Ein Framework für den sicheren Datenaustausch in industriellen Produktionsprozessen
  Paul-Georg Wagner, Pascall Birnstill
• Drohnen im Praxistest: Testfeld in der Karlsruher Forschungsfabrik
  Matthias Kollmann, Wilmuth Müller
• Schwenkprüfstand – Innovation zur Validierung elektrischer Antriebe für Luftfahrtanwendungen eVTOL-Antriebssysteme realitätsnah testen
  Lars-Fredrik Berg, Simon Dillmann

Following the rapid technological evolution in the field of Artificial Intelligence (AI), the title of this brochure – Agentic Industrial AI – seems quite straightforward. How can we apply the concepts and tools of AI agents to industrial applications? Indeed, there is a huge potential if optimization tasks in production processes can be delegated to autonomous software components, or if operators can reliably interact with machines or entire production plants in natural language. However, industrial environments have specific requirements, not only with respect to safety, security and reliability constraints, but also regarding the methods to acquire and manage the data needed to train and operate the AI agents, as well as to systematically engineer industrial AI systems. Hence, this brochure takes a broader perspective and includes methods, tools and applications that demonstrate how to integrate agentic AI into industrial IT environments based on Industry 4.0 standards and industrial dataspaces.

It starts with an essay about the revival of agent technology in the AI era, before presenting, in a guest contribution from EDI AG, an industrial AI framework product that is evolving towards agentic AI. We then consider the data foundations enabling agentic AI, based on open standards for sensor data management, Industry 4.0 digital-twin modelling standards, industrial dataspaces and data ecosystems. We continue with the question of how to apply large language models (LLMs) to industrial engineering before presenting one of the key interfaces for agentic AI – the Model Context Protocol (MCP). The brochure concludes with an explanation of how to use agentic decision support for power system control rooms.

Agentic industrial AI is highly dynamic, multidisciplinary and diverse. It requires the targeted and careful combination of systems engineering, AI and IT skills within and across companies. We would appreciate the opportunity to get in touch and stay in contact with you in order to collaborate on dependable and trustworthy agentic industrial solutions.

Content

The links will take you directly to the corresponding page of the ePaper.

• Revival of the agents: what is new in the AI era?
  Thomas Usländer
• Agentic industrial AI for cross-layer automation: Unlocking data and knowledge silos with the EDI hive IoT & AI Framework
  Mohanad El-Haji, Thomas Freudenmann
• Unlocking AI potential: Open standards as the foundation for AI agents
  Philipp Hertweck
• Data ecosystems as enablers of industrial AI
  Olaf Sauer
• Agentic AI in the FA³ST Ecosystem 
  Marc Leon Haller
• LLM agents for industrial engineering: From code generation to process optimization
  Joschka Kersting
• MCP as an open interface for intelligent production
  Jan Hermes, Benedikt Stratmann
• GridCompanion: Agentic decision support for power system control rooms
  Dennis Rösch