Development of ball end-mill wear condition judgment system aided by abnormality detection and image classification

Abstract

During the cutting process, it is crucial to replace cutting tools at the right time to meet machining accuracy requirements. However, determining tool life relies on the unsystematic tacit knowledge of skilled engineers. Moreover, the increasing variety of cutting tools needed to accommodate the trend towards small quantities of various products has made it even more challenging to determine tool life. This study aimed to address this issue by developing a system that could accurately determine when to replace ball end-mills without requiring the judgment of skilled engineers. The results of the study showed that anomaly detection methods can be applied to images of worn tools to determine tool life accurately. By converting current value data during cutting into images and using image classification methods, it is possible to classify tools according to their wear conditions, such as new, early wear, end stage of wear, and end of tool life. The study also showed that constructing a multimodal judgment system using both tool appearance image abnormality detection and current value data image classification resulted in a 98% judgment accuracy rate for tools that had reached the end of their service life, with a correct answer rate of over 90%.