Fraunhofer Institute for Software and Systems Engineering
Fraunhofer Institute for Software and Systems Engineering

FLaME

Sovereign Federated Learning in the Edge-Cloud Continuum

How can companies train AI collaboratively when sensitive production data cannot be consolidated centrally? In the “FLaME” project, Fraunhofer ISST and Fraunhofer IOSB, in collaboration with IVA SCHMETZ GmbH, have developed a prototype for secure federated learning in the edge-cloud continuum. The use case is predictive maintenance for industrial furnaces.

Schiffs- und Kommunikationsnetzkonzept

The Challenge

When it comes to predictive maintenance for industrial furnaces, operational data from various customer sites can help develop better AI models. However, this data is sensitive: It reveals insights into production processes and, for technical, economic, or contractual reasons, often cannot be centrally aggregated. Added to this are differing IT infrastructures and requirements for data sharing. Shared AI models therefore require an approach that enables cross-site learning without depriving companies of control over their data.

Our Solution

At the heart of FLaME is the freedom of choice for plant operators: They decide for themselves which data to share, how it is used, and how much control they wish to retain. To this end, Fraunhofer ISST and Fraunhofer IOSB, in collaboration with IVA SCHMETZ GmbH, have developed a flexible federated learning architecture that takes into account varying requirements regarding data protection, IT resources, and costs. Three training options—Edge-FL at the customer’s site, proxy training at a trusted service provider, and centralized cloud training—enable participation tailored to specific needs. The architecture is complemented by binding approval processes and a governance and operational model with tiered service offerings.

 

The Result

In the application scenario with IVA SCHMETZ GmbH and its predictive maintenance solution, key system processes were successfully demonstrated in a prototype—from site registration through federated training to controlled model deployment. FLaME thus establishes a technical foundation for sovereign, collaborative industrial AI. The next step will involve evaluating model quality and benefits using real plant data.

The Partners

  • Fraunhofer IOSB
  • IVA SCHMETZ GmbH
  • Duration: 03/2026–10/2026