• Journal of the Korea Society of Computer and Information
• /
• v.29 no.10
• /
• pp.149-156
• /
• 2024

This paper implemented a drill bit foreign matter inspection system based on the YOLO V3 algorithm and evaluated its performance. The study trained the YOLO V3 model using 600 training data to distinguish between the normal and foreign matter states of the drill bit. The implemented inspection system accurately analyzed the state of the drill bit and effectively detected defects through automatic inspection. The performance evaluation was performed on drill bits used more than 2,000 times, and achieved a recognition rate of 98% for determining whether resharpening was possible. The goal of foreign matter removal in the cleaning process was evaluated as 99.6%, and the automatic inspection system could inspect more than 500 drill bits per hour, which was about 4.3 times faster than the existing manual inspection method and recorded a high accuracy of 99%. These results show that the automated inspection system can dramatically improve inspection speed and accuracy, and can contribute to quality improvement and cost reduction in manufacturing sites. In future studies, it is necessary to develop more efficient and reliable inspection technology through system optimization and performance improvement.

• Journal of the Institute of Convergence Signal Processing
• /
• v.16 no.3
• /
• pp.90-95
• /
• 2015

This study presents a method to test quite small-sized and light-weighted micro-drill bits which are used to make holes in printed circuit boards(PCB). After getting images of micro-drill bits through the high resolution microscope, we developed image processing algorithms to detect fiducial points, and then measured diverse factors of the drill-bit based on these points. We also developed the window-based inspection system to automatically discriminate normal and abnormal status. For the relative comparison of its performance, the system was compared with an existing inspection system using test images. Experimental results showed that the proposed system slightly improved performance, and also classified correctly some misjudged errors which were occurred in the existing system.