Fraunhofer Institute for Production Technology IPTResearch project "FL4AI"
Tool wear: a barrier to efficient production and a cost driver
In the cutting industry, tool wear due to milling, turning and drilling is a significant cost factor: Delayed replacement of worn tools leads to machine downtime, quality defects and scrap, thus causing high costs.
To reduce these costs, visual inspection of cutting tools are performed during production. The standard tool inspection procedures are conducted in accordance with ISO 8688, which measures the flank wear width parameter throughout the tool's lifespan. Yet, this method is subjective and highly time-consuming. Conventional tool inspection systems are installed outside the machine and require complex manual operation, resulting in extended production downtime.
As a result, companies tend to replace tools prematurely after a specified cumulative tool path length or usage time in order to avoid costs and quality issues. However, this significantly shortens the lifespan of the tools and wastes considerable resources.
Precise tool evaluation through AI-supported wear detection
To meet these challenges, we are working with our partners on the "FL4AI" research project to develop an automated and objective method for measuring tool wear directly in the machine. Our goal is to determine the optimal time for tool changes in order to avoid both premature and delayed changes.
Images from high-resolution camera systems analyzed with artificial intelligence (AI) enable automated tool condition monitoring. The AI recognizes wear patterns, marks relevant pixels in the image and processes them to create key performance indicators. This way, the progression of damage is recognized and monitored – and tool evaluation is significantly more precise. Such an automated monitoring system increases the efficiency and sustainability of production processes.
Decentralized AI training with federated learning: cost efficiency and data sovereignty
Until now, companies have used centralized cloud services for their AI-based quality control to avoid high initial investments in local infrastructure. However, this puts large amounts of sensitive production data outside their control and exposes it to major data protection risks. In addition, long-term, service-dependent fees can lead to cost disadvantages.
Federated learning offers a decentralized solution that enables small and medium-sized companies to use AI for manufacturing while ensuring data protection. The data remains on local servers and is used for the collaborative training of more powerful AI models. Companies only transfer the model parameters for a few iterations to a central server, where they are aggregated and merged into a global model. This way, companies retain data sovereignty.
Camera system technology and prototype development
In the project, an optimized camera system is developed to capture high-resolution images of cutting tool edges during machine operation. The research team is developing a machine-integrated camera technology and a powerful image processing interface that incorporates both raw and pre-processed images into the AI model. This programming interface enables efficient analysis and automatic identification of wear characteristics and adapts flexibly to different tool types and cutting processes.
The project outcome should be a functional prototype that combines camera technology, interface programming and federated learning and is then thoroughly tested in realistic production environments.
Project partners
- Fraunhofer Institute for Production Technology IPT, Aachen
- Kaitos GmbH, Münster (Coordination)
- Rotoclear GmbH, Heidelberg
- S & D Metall- und Kunststoffbearbeitung GmbH, Ketsch
Funding reference
The "FL4AI" project is funded by the Federal Ministry of Education and Research (BMBF) as part of the "KMU-innovativ: IKT" funding program.
- Funding reference: 01|S24054D
- Project management agency: German Aerospace Center (DLR)