Fraunhofer Institute for Production Technology IPTResearch project "FL.IN.NRW"
Quality requirements in the machining industry are high, especially in heavily regulated sectors such as aviation, the automotive industry and medical technology. In these sectors, manufacturing companies process high-strength materials. Errors in the cutting process can have serious consequences, ranging from product failures to safety issues. Strict quality controls are therefore essential, but they are also time-consuming and expensive.
Artificial intelligence (AI) offers great potential for improving production quality and control: automated monitoring and analysis of production processes can significantly reduce inspection times and the cost of quality assurance, while improving the accuracy of quality assessment.
Cutting as the first application study
In the "FL.IN.NRW" research project, we are working with our partners to develop a learning platform for decentralized training of predictive AI models. As a first use case, our project team is investigating the complex process of cutting: The large number of tool and process parameters in cutting is a challenge for quality control, which can usually only be overcome by time-consuming manual inspection of the components.
By training the models with process data directly from the production machine, the AI can detect quality problems during cutting: Deviations in the desired component profile due to tool wear are detected by fluctuations in the spindle load and clamping pressure. The AI model immediately detects this tool behavior as a dimensional deviation outside specified tolerances. This means that time-consuming quality control only has to be carried out when necessary and can be significantly reduced, making quality assurance and manufacturing more efficient.
Data protection risks and cost disadvantages of using centralized cloud services
So far, companies have been relying on centralized cloud services to develop their AI-powered quality control to avoid expensive initial investments in on-premises digital infrastructure. However, the large amount of production data stored in the cloud is outside the company's control and is therefore exposed to greater data protection and security risks. In addition, the ongoing, service-dependent fees of cloud services can become a cost disadvantage for companies in the long term.
Decentralized AI training for more cost efficiency and data sovereignty
The decentralized approach to machine learning enables small and medium-sized companies to use the advantages of AI for their quality control while ensuring data protection and data security of their sensitive production data:
The data remains securely on local servers while it can be used for decentralized, collaborative training of even more powerful AI models. Across multiple company locations, the AI model is trained in a network of local devices and company servers without the manufacturing data leaving the local databases. Only the model parameters are sent to a central server, where they are aggregated and merged into a global model, ensuring that data sovereignty remains with the companies.
During the project, our experts will investigate how model performance can vary depending on the configuration and which aggregation algorithm is best suited to the specific requirements of the use cases.
Diverse production applications for federated learning
In a next step, the project team will generalize the research approach to enable the future use of federated learning in the production of other components and in entirely different production applications, such as optical metrology and additive manufacturing. This opens up promising opportunities to increase process efficiency and component quality in these application areas.
Project partners
- Fraunhofer Institute for Production Technology IPT, Aachen (coordination)
- gemineers GmbH, Aachen
- Innoclamp GmbH, Aachen
- Kaitos GmbH, Münster
- dataMatters GmbH, Cologne
Funding
The "FL.IN.NRW" project is funded by the European Union and the state of North Rhine-Westphalia as part of the EFRE/JTF program NRW 2021-2027.
- Project Management Jülich
- Funding code: EFRE-20800207
