In machining production, it has not yet been possible to systematically record tool wear during ongoing milling processes. However, since faulty tools lead to quality losses, increasing rejects and high costs for rework, the Fraunhofer Institute for Production Technology IPT has taken on this task: Together with partners, the researchers from Aachen, Germany developed a system of cameras and image processing using artificial intelligence, which is already in use the machine tool can record and evaluate tool wear.
To this day, the wear condition of cutting tools is laboriously checked outside the machine using measuring microscopes, pocket magnifying glasses and tool setting devices. All of these procedures require manual intervention; Microscopes are also expensive, pocket magnifiers do not allow measurement of wear metrics, and tool setting devices capture the cutting edge contour but cannot identify the type of wear. In each of these cases, measurements can only be taken after manufacturing is complete, when it is too late to intervene in the process.
In the ‘CAMWear 2.0’ project, a research team from Fraunhofer IPT, together with project partners, developed a system that precisely records and evaluates the wear condition of the cutting tools almost in real time during the milling process.
Measuring System Overcomes Weak Points in Tool Wear Detection
To do this, the researchers integrated a microscope into the milling machine, which automatically takes images of the milling tool during processing, between the individual processing steps. Inspired by medical technology processes, they developed techniques for image segmentation, on the basis of which industry-typical evaluation parameters of the tool condition can be derived. In order to protect the sensitive microscope in the harsh environment of the machine tool, the researchers constructed a robust housing with a sealing air function that keeps drops of cooling lubricant away from the camera.
AI-Generated Wear Model
The captured images serve as training data for the AI-supported image processing program that the partners developed over the course of the project. The program is able to classify tool types, show worn areas and calculate wear metrics.
In order to reduce the upstream manual effort required for training artificial intelligence, the researchers used a new approach: They use generative algorithms and neural networks to create synthetic image data in order to artificially enlarge the database. In addition, the real images are changed and reproduced using simple augmentation techniques, such as mirroring or rotating.
Successful Practical Test Confirms Performance
The camera system and the image processing program passed the first practical test under real conditions in the final project phase: The automation of image capture and the outstanding quality of the photos taken exceeded the project team’s expectations. The camera housing proved to be robust enough to reliably protect the microscope unit. The AI in the image processing software identified the visually detectable forms of wear extremely reliably and precisely.
The application is now being further optimized for industrial use: Another goal is to further refine the AI models in order to identify and analyze signs of wear even more precisely. In close collaboration with specialized hardware suppliers, it is now important to transfer the new AI application into industrial practice as quickly as possible.