Efficient Cutting Tool Wear Segmentation Based on Segment Anything Model

TL;DR

This work tackles tool wear segmentation under limited data by integrating Segment Anything Model (SAM) with U-Net as an automated prompt generator. The proposed pipeline binarizes U-Net outputs to derive Point-of-Interest prompts and uses three PoI strategies (MS, CoGA, RCoGA) to refine SAM segmentations, with RCoGA emerging as the most effective for varied wear. Across data-scarce regimes, the SAM-enhanced approach matches or surpasses U-Net trained on larger datasets, demonstrating substantial gains at 20% training data and robust performance as data increases. The study highlights practical potential for industrial deployment with limited labeled data while acknowledging SAM's domain-generalization limits and the need for domain-specific retraining in future work.

Abstract

Tool wear conditions impact the surface quality of the workpiece and its final geometric precision. In this research, we propose an efficient tool wear segmentation approach based on Segment Anything Model, which integrates U-Net as an automated prompt generator to streamline the processes of tool wear detection. Our evaluation covered three Point-of-Interest generation methods and further investigated the effects of variations in training dataset sizes and U-Net training intensities on resultant wear segmentation outcomes. The results consistently highlight our approach's advantage over U-Net, emphasizing its ability to achieve accurate wear segmentation even with limited training datasets. This feature underscores its potential applicability in industrial scenarios where datasets may be limited.

Figures (19)

• Figure 1: U-Net on tool wear segmentation task
• Figure 2: overview of SAM's structure
• Figure 3: the concept of the proposed approach
• Figure 4: procedural flow of MS
• Figure 5: procedural flow of CoGA