EUVIMEDEUVIMED entdecken
Crossref

Bolt surface defect detection based on improved YOLOv11n

Wuxiaotian Zheng, Lingyuxiu Zhong, Mengyu Wu, Yujie Jin, Yunzheng Ran, Honghua Liao

Engineering Research Express · 01.06.2026 · 8 (11) · 115406

DOI: 10.1088/2631-8695/ae76ea ↗

Kurz gefasst

Fachartikel in der Zeitschrift Engineering Research Express, online erschienen am 11.06.2026. Verfasst von Wuxiaotian Zheng und 5 weiteren Autor:innen. Die Arbeit nennt 20 Literaturangaben. Gefördert von: Data Center of Management Science, National Natural Science Foundation of China - Peking University.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

To address existing issues in hydropower station maintenance, such as strong environmental noise interference during bolt surface defect identification, low accuracy in feature extraction of surface defects, and insufficient computational power on edge devices, this paper proposes an identification algorithm based on an improved YOLOv11n (RepViT-Slim-neck-UIoU-YOLOv11n) to enhance the accuracy and robustness of bolt surface defect detection. First, RepViT(Revisiting Mobile CNN From ViT Perspective) replaces the traditional backbone network, leveraging a multi-branch training–single-branch inference structure to strengthen feature extraction capability and noise suppression. Second, the Slim-neck neck structure is introduced, innovatively integrating modules such as GSConv(Grouped Spatial Convolution), GS bottleneck(Ghost Shuffle bottleneck), and VoV-GSCSP(Vortex of Vectorized Ghost Shuffle Cross Stage Partial) to reduce computational parameters without compromising feature extraction ability, thereby enabling deployment on edge devices with lower computational power. Finally, the UIoU(Unified-IoU) loss function is employed, improving bounding box regression accuracy through dynamic weight allocation. Experiments were conducted using a self-built dataset containing 370 original images, divided in a 7:2:1 ratio, with separate augmentation applied only to the training set. Ablation studies show that the improved model achieves an mAP of 93.6%, representing a 22.5% improvement over the baseline YOLOv11n, while maintaining an inference speed of 68.4 FPS(Frames Per Second). Comparative experiments demonstrate that YOLOv11n-RSL outperforms other YOLO variants and mainstream detection models such as Faster R-CNN(Faster Region-based Convolutional Neural Network) in bolt detection tasks, particularly excelling in the detection of corrosion and coating defects. This algorithm exhibits significant advantages in both accuracy and efficiency, making it suitable for deployment on edge devices. This study provides an innovative solution for image signal–based bolt surface defect detection. As an edge-device-friendly algorithm, it offers a high-precision and real-time visual inspection solution for the intelligent inspection of bolt health status in large-scale outdoor equipment such as hydropower stations, wind turbines, and bridge steel structures, demonstrating direct engineering application value.

Förderung

  • Data Center of Management Science, National Natural Science Foundation of China - Peking University

Zitieren

APA

Zheng, W., Zhong, L., Wu, M., Jin, Y., Ran, Y., & Liao, H. (2026). Bolt surface defect detection based on improved YOLOv11n. Engineering Research Express, 8(11), 115406. https://doi.org/10.1088/2631-8695/ae76ea

Vancouver

Zheng W, Zhong L, Wu M, Jin Y, Ran Y, Liao H. Bolt surface defect detection based on improved YOLOv11n. Engineering Research Express. 2026;8(11):115406. doi:10.1088/2631-8695/ae76ea

BibTeX
@article{Zheng2026,
  author = {Zheng, Wuxiaotian and Zhong, Lingyuxiu and Wu, Mengyu and Jin, Yujie and Ran, Yunzheng and Liao, Honghua},
  title = {Bolt surface defect detection based on improved YOLOv11n},
  journal = {Engineering Research Express},
  year = {2026},
  volume = {8},
  number = {11},
  pages = {115406},
  doi = {10.1088/2631-8695/ae76ea}
}
Kostenlos starten →