Researchers at the Karlsruhe Institute of Technology (KIT) have developed a system for the fully automatic monitoring of ball screws in machine tools. A camera integrated directly into the nut of the ball screw is used. Based on the image data generated, an artificial intelligence (AI) continuously monitors wear and thus reduces machine downtime.

The maintenance and timely replacement of defective components in machine tools is an important part of the production process in mechanical engineering. In the case of ball screw drives, such as those used in lathes for precision guidance in the manufacture of cylindrical components, wear has so far been detected manually. "Maintenance is therefore associated with assembly work. The machine then comes to a standstill," says Professor Jürgen Fleischer from the Institute of Production Engineering (wbk) at KIT. "Our approach, on the other hand, is based on the integration of an intelligent camera system directly into the ball screw drive. This allows a user to continuously monitor the condition of the spindle. If there is a need for action, he is informed automatically."

Components of the camera system © KIT

The new system consists of a camera with lighting attached to the nut of the ball screw drive, which is combined with artificial intelligence to evaluate the image data. As the nut moves on the screw, it takes individual images of each section of the screw. This allows the entire surface of the screw to be analyzed.

Demonstration of the system developed at KIT for fully automatic wear monitoring of ball screws using artificial intelligence. © KIT

Artificial intelligence for mechanical engineering
The combination of image data from ongoing operations with machine learning methods enables users of the system to directly assess the condition of the spindle surface. "We have trained our algorithm with thousands of images so that it can now confidently distinguish between spindles with and without defects," says Tobias Schlagenhauf from wbk, who was involved in the development of the system. "By further evaluating the image data, the wear can also be precisely quantified and interpreted. This allows us to distinguish whether discoloration is simply dirt or harmful pitting."

When training the AI, all conceivable forms of visually visible degeneration were taken into account and the functionality of the algorithm was validated with new image data that had never been seen by the model before. The algorithm is suitable for all applications in which image-based defects on the surface of a spindle are to be identified and can also be transferred to other applications. as