CNN-Based Classification of Optically Critical Cutting Tools with Complex Geometry: New Insights for CNN-Based...

Area of Science:

• Manufacturing Engineering
• Materials Science
• Computer Vision

Background:

• Sustainability initiatives drive the need for recycling and repairing high-demand items like CNC machining tools.
• Repair and regrinding of cutting tools are crucial due to their essential role in manufacturing.
• Automated classification of diverse cutting tools (drills, end mills, taps) is challenging due to optical properties and coatings.

Purpose of the Study:

• To investigate the influence of varying lighting conditions on the automated classification of cutting tools for regrinding.
• To address the growing need for automation in tool repair and maintenance, mitigating skilled worker shortages.
• To compare the effectiveness of different training strategies for tool classification under diverse lighting.

Main Methods:

• Development of two unique tool-specific datasets, each comprising 36 distinct cutting tools.
• Recording tool images under two distinct lighting conditions: direct diffuse ring lighting and normal daylight.
• Utilizing Grad-CAM heatmap analysis to interpret classification features and model attention.

Main Results:

• Lighting conditions significantly affect the accuracy of automated cutting tool classification for regrinding.
• Different training strategies yield varying performance levels depending on the lighting environment.
• Grad-CAM analysis identified key visual features crucial for distinguishing tool types under different lighting.

Conclusions:

• Optimizing lighting is critical for robust and accurate automated classification of cutting tools in repair and regrinding workflows.
• The findings support the development of more reliable automated systems to address manufacturing labor shortages.
• Further research into lighting-invariant feature extraction can enhance classification performance across diverse industrial settings.