Integrating object detection and image segmentation for detecting the tool wear area on stitched image

Area of Science:

• Manufacturing Engineering
• Materials Science
• Computer Vision

Background:

• Flank wear is a prevalent issue in end milling, complicating tool condition monitoring.
• Manual detection of flank wear is time-consuming and labor-intensive.
• Automated detection is crucial for optimizing machining processes and tool lifespan.

Purpose of the Study:

• To develop a comprehensive and automated method for detecting flank wear areas on spiral end milling cutters.
• To improve the efficiency and accuracy of tool wear assessment.
• To provide a system for predicting tool replacement needs.

Main Methods:

• Utilized template matching and deep learning techniques to process curved surface images.
• Expanded curved surface images into panorama images for enhanced wear detection.
• Employed the You Only Look Once v4 (YOLOv4) model for automatic cutting tip detection.
• Applied segmentation models (U-Net, Segnet, Autoencoder) to extract flank wear regions.
• Evaluated segmentation models using the Dice coefficient.

Main Results:

• The U-Net model achieved the highest Dice coefficient score of 0.93 for flank wear segmentation.
• The proposed method effectively extracts tool wear regions from spiral cutting tools.
• The system can predict wear trends to inform tool change decisions.
• Panorama image expansion facilitated wear detection without specific tool positioning.

Conclusions:

• The developed method offers an effective solution for automated flank wear detection in spiral end milling.
• Accurate wear area extraction enables proactive tool management and reduces downtime.
• The system provides valuable insights into tool condition, preventing severe wear and optimizing tool replacement schedules.