AZIBO Shape Descriptor for Monitoring Tool Wear in Milling

Abstract

The aim of this paper is to estimate efficiently insert wear in metal machining and to improve tool replacement operations. Image processing and classification are used to automatize the decision making about the adequate time for tool replacement. Specifically, the shape descriptor aZIBO (absolute Zernike moments with Invariant Boundary Orientation) has been used to characterize insert wear and ensure its optimal usage.A dataset composed of 577 regions with different levels of wear has been created. Two different classification processes have been carried out: the first one using three different classes (Low, Medium and High wear -L, M and H, respectively-) and the second one with just two classes: Low (L) and High (H). Classification was carried out using on the one hand kNN with five different distances and five values of k and, on the second hand, a Support Vector Machine (SVM).aZIBO performance has been compared with classical shape descriptors such as Hu and Flusser moments. It outperforms them, obtaining success rates up to 91.33% for the L-H classification and 90.12% for the L-M-H classification.