Agent-based AI for the intelligent networking and evaluation of brownfield factories

How productive is a factory? Where are the weak points and areas for improvement? And how can existing machinery and IT systems be networked in such a way that artificial intelligence (AI) can answer these questions autonomously? In the ReFactory-AI research project, Reutlingen University and the Fraunhofer Institute for Manufacturing Engineering and Automation (IPA) are developing a prototype at Werk150 for an agent-based AI system that connects machines and IT systems and automatically records and evaluates their status.

Project description

Many production environments in Baden-Württemberg have evolved over decades. Machinery and software systems from different generations form a heterogeneous system landscape. Small and medium-sized enterprises, in particular, lack the means to harness this data for modern AI applications.

This is precisely where ReFactory-AI comes in. The project is developing a compact on-premises prototype that connects existing production assets and analyses them using specialised AI agents. All data remains on-premises within the company, and sensitive production data is not transferred to external cloud services.

The basis for this is a digital twin of the factory. Machines, plant and IT systems are integrated via interoperable connectors based on the Asset Administration Shell (AAS). Even older machines can be integrated through retrofitting and the addition of sensors. AI agents are designed to support the connection of the various systems and reduce the manual integration effort required.

How healthy is production?Specialised AI agents analyse the state of production based on the data. The project is investigating the concept of ‘Operational Health’ for this purpose. It describes how efficient and resilient a production process is across various dimensions. The initial focus is on productivity. AI agents access analysis and evaluation functions, collate the necessary information and help to assess the current state of production, as well as to identify anomalies and areas for improvement. The results are consolidated and displayed in an Operational Health Dashboard. This is intended to provide users with a quick overview of the state of their production and to support them in analysing root causes and devising appropriate measures. In the long term, the approach can be expanded to include further dimensions.

Prototype at Werk150

A prototype is being developed and tested under realistic conditions at Werk150 at Reutlingen University. The teaching and research factory is equipped with suitable machinery, industrial IT systems, autonomous intralogistics and a modern digital infrastructure, thus offering ideal conditions for investigating typical industrial scenarios.

As part of a transfer partnership, Dr Juliane Gottmann is supporting the project with her extensive experience in the fields of production science and artificial intelligence.