AI Without Data Exchange

Würzburg, 10.08.2026 (PresseBox) - In the future, artificial intelligence (AI) will be able to evaluate the quality of ultrasonic welded joints directly during production without sensitive company data leaving the facility. This is demonstrated by the FLUSH research project, funded by the BMFTR. The project partners—Katulu GmbH, SUCO Robert Scheuffele GmbH & Co. KG, and the SKZ Plastics Center—developed a federated learning system for this purpose, which has already been successfully deployed under real production conditions.

Taking Quality Assurance in Ultrasonic Welding to a New Level

The FLUSH project (Federated Learning in the Ultrasonic Welding Process for Quality Assurance) developed a novel solution for the automated evaluation of ultrasonic welding processes. Until now, the quality of welded joints has usually only been determined through time-consuming and, in some cases, destructive testing methods.

At the same time, modern ultrasonic welding systems generate large amounts of process and sensor data that have been utilized only to a limited extent. The goal of the project was therefore to use artificial intelligence to make this data usable for automatic in-line evaluation of each individual weld.

Powerful AI with Full Data Security
At the heart of the approach is the use of federated learning. In this process, AI models are trained decentrally using data from individual companies, and only model parameters are exchanged—not the underlying production data. In this way, process-relevant know-how remains entirely with the respective companies, while at the same time a shared, more powerful model is created.

This principle—machines learning from one another without disclosing data—was successfully applied to the ultrasonic welding process for the first time in this project.

High prediction accuracy despite decentralized learning
The technical implementation was carried out using an edge-based infrastructure that captures process and time-series data directly at the welding systems and makes it available for model training. In particular, the use of high-resolution time-series data proved to be crucial for model quality. The locally trained AI model can predict weld seam strength with an accuracy of over 90%.

The federated model achieves an accuracy that is only about 5% lower. This demonstrates that the approach enables high prediction quality even without sharing sensitive production data and can continue to improve in performance as more data becomes available.

From Research to Production
Another milestone of the project was the development and integration of a demonstrator that transfers AI-based quality assessment into the production environment in real time. The solution was implemented at SUCO and displays the predicted strength of the welds directly in an intuitive user interface. The demonstrator operates entirely locally on an edge device, requires no internet connection, and can be seamlessly integrated into existing manufacturing processes.

“It is particularly encouraging that the predictive accuracy of the federated-learning model lags only slightly behind that of a locally trained model. This allowed us to demonstrate that data sovereignty and high-performance AI models do not have to be mutually exclusive,” says Mingo Kübert, Digitalisation Scientist at SKZ.

Pooled Expertise
The project was implemented through close collaboration between industry and research. Katulu developed the federated learning system, the AI models, and the demonstrator. As an application partner, SUCO provided real-world production conditions and manufacturing data. SKZ was responsible for scientific and technical support and data analysis. In addition, other industry partners contributed their requirements and practical experience to the development.

Data-Sovereign AI for Small and Medium-Sized Enterprises
With the results achieved, FLUSH makes an important contribution to the digitalisation of plastics processing. For small and medium-sized enterprises (SMEs) in particular, the approach opens up new possibilities for implementing AI-supported quality assurance cost-effectively without having to disclose sensitive data. At the same time, the project demonstrates that cooperative models of data use represent a sustainable path for industrial applications.

Further data and optimizations are required for broad industrial deployment. However, the solution developed in the project lays an important foundation for future applications. Looking ahead, additional sensor data—for example, from thermographic inspections—could further improve prediction quality and pave the way for 100-percent quality monitoring.

Funding
The FLUSH project, “Federated Learning in the Ultrasonic Welding Process for Quality Assurance,” was funded by the Federal Ministry of Research, Technology, and Space (BMFTR) from October 2023 to September 2025 as part of the AI4SME program under grant number 16IS23058B.

More information on Digitalisation at SKZ