Scholarlink

To prevent small surface defects in high-strength TC4 bolts from compromising the accuracy and reliability of precision equipment, this study employs the widely used YOLOv5 algorithm for experimental testing on a self-built dataset. The model’s performance analysis reveals that the mean average precision (mAP) consistently improves with training epochs and stabilizes at 76.5% after 400 epochs, demonstrating good convergence. The model achieves an average accuracy of 94.3% and a detection speed of 56 FPS, ensuring efficient and reliable defect identification. Furthermore, robustness tests under complex lighting conditions confirm the model’s ability to accurately detect defects in both high and low illumination environments, as well as under uneven lighting, without misclassification. These results validate the feasibility and reliability of computer vision-based non-destructive testing for high-strength bolt defect detection.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Detection of Small Surface Defects in High-Strength Bolts Based on Computer Vision

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Liangliang Wei,
  • Yanfeng Xia,
  • Long He,
  • Yanhong Ding,
  • Feilong Xu,
  • Xu Ma

摘要

To prevent small surface defects in high-strength TC4 bolts from compromising the accuracy and reliability of precision equipment, this study employs the widely used YOLOv5 algorithm for experimental testing on a self-built dataset. The model’s performance analysis reveals that the mean average precision (mAP) consistently improves with training epochs and stabilizes at 76.5% after 400 epochs, demonstrating good convergence. The model achieves an average accuracy of 94.3% and a detection speed of 56 FPS, ensuring efficient and reliable defect identification. Furthermore, robustness tests under complex lighting conditions confirm the model’s ability to accurately detect defects in both high and low illumination environments, as well as under uneven lighting, without misclassification. These results validate the feasibility and reliability of computer vision-based non-destructive testing for high-strength bolt defect detection.