Image processing with deep-learning and transfer learning for cutting tool degradation monitoring

Abstract

Monitoring the degradation of cutting tools is of utmost importance in the manufacturing world. Tools with substantial wear fail to produce high-quality parts in terms of geometry, residual stress, and surface finish. Furthermore, replacing tools in a non-optimal manner can lead to increased production costs and downtime. Therefore, monitoring the condition of the tool is essential to avoid these additional costs and ensure good production quality. This article explores various classification models, specifically VGG19, EfficientNetV2, and Vision Transformers. These models classify the state of tools using their images. Using transfer learning, a comparison of the best-performing artificial intelligence-based image analysis models is conducted to identify those most suitable for monitoring cutting tools. A comparative analysis of their generalizability, performance and explainability is realized. The model with the best performance is VGG19 with an accuracy of 94%, followed by EfficientNetV2 and ViT with an accuracy of 87%. A full comparison of these results is carried out.