YOLO-BR: A bearing defect detection model with multi-scale defect perception and attention fusion mechanism
Measurement Science and Technology · 21.08.2026 · 37 (33) · 335406
DOI: 10.1088/1361-6501/ae92a7 ↗Kurz gefasst
Fachartikel in der Zeitschrift Measurement Science and Technology, online erschienen am 19.08.2026. Verfasst von Jingzhe Zhang und 8 weiteren Autor:innen. Die Arbeit nennt 52 Literaturangaben. Gefördert von: Science and Technology Department of Jilin Province.
Abstract (Originalsprache)
Bearings serve as critical transmission components in automotive systems, enduring sustained heavy loads and alternating stresses under harsh working conditions; surface defects readily induce fatigue failure and compromise driving safety, making Bearing Defect Inspection a crucial task. However, their tiny scales, diverse morphologies and severe metallic surface reflection pose considerable challenges to accurate detection, especially for capturing Multi-scale Features and suppressing background interference. To address these issues, this paper presents YOLO-BR, a detection model integrated with multi-scale defect perception and Attention Mechanism fusion. Based on the YOLOv13 framework, three core improvements are proposed: first, the backbone is upgraded with the improved DSC3k2- channel prior convolutional attention (CPCA) module embedded with CPCA channel attention to strengthen discriminative feature representation; second, the defect-aware enhanced convolution module is devised for precise Multi-scale Features extraction via differentiated multi-branch structures; finally, the cross-dimensional branch attention mechanism is established to suppress background clutter through cross-dimensional interaction and contextual fusion, leveraging the Attention Mechanism to focus on salient defect regions. Experimental results demonstrate that YOLO-BR maintains efficient inference while increasing mean average precision (mAP) by 2.2% and raising mAP@0.5%–95.4%. Evaluations on the PKU-Market-PCB dataset further verify its favorable cross-scenario generalization, revealing promising application potential in automated Bearing Defect Inspection systems.
Förderung
- Science and Technology Department of Jilin Province
Zitieren
APA
Zhang, J., Kong, D., Sun, R., Li, B., Li, C., Zhu, X., Xing, Y., Zhao, D., & Zhang, J. (2026). YOLO-BR: A bearing defect detection model with multi-scale defect perception and attention fusion mechanism. Measurement Science and Technology, 37(33), 335406. https://doi.org/10.1088/1361-6501/ae92a7
Vancouver
Zhang J, Kong D, Sun R, Li B, Li C, Zhu X, et al. YOLO-BR: A bearing defect detection model with multi-scale defect perception and attention fusion mechanism. Measurement Science and Technology. 2026;37(33):335406. doi:10.1088/1361-6501/ae92a7
BibTeX
@article{Zhang2026,
author = {Zhang, Jingzhe and Kong, Dexin and Sun, Rui and Li, Bo and Li, Changhe and Zhu, Xu and Xing, Yi and Zhao, Dejin and Zhang, Jianhai},
title = {YOLO-BR: A bearing defect detection model with multi-scale defect perception and attention fusion mechanism},
journal = {Measurement Science and Technology},
year = {2026},
volume = {37},
number = {33},
pages = {335406},
doi = {10.1088/1361-6501/ae92a7}
}
EUVIMED