机电工程技术2025，Vol.54Issue(6)：144-149,6.DOI:10.3969/j.issn.1009-9492.2025.06.025
基于改进YOLOv5的机车圆弹簧缺陷检测
Defect Detection of Locomotive Circular Springs Based on Improved YOLOv5
摘要
Abstract
The current magnetic particle inspection of locomotive circular springs is performed through manual observation and analysis,commonly leading to issues such as missed detections,false detections,and low automation.An improved YOLOv5-based defect detection algorithm for locomotive circular springs is proposed to better assist workshop personnel in detecting crack defects in locomotive circular springs.First,to ensure sufficient sample support for model training,data augmentation is applied to expand the existing defect samples of locomotive circular springs.Second,to make the trained defect detection model more suitable for edge deployment,MobileNetv3 is used to replace YOLOv5 original backbone,achieving a lightweight model and reducing computational costs during inference.Finally,experimental analysis is conducted on a dataset of locomotive circular spring crack defects.The improved algorithm,YOLOv5-M,significantly reduces model parameters and computational costs without excessively sacrificing detection accuracy.The model's parameter count is reduced to 3.39 M,a decrease of 53.1%compared to the original model,while GFLOPs are reduced from 16.6 to 6.1.Compared to the base algorithm,the improved algorithm offers better deployability,demonstrating the feasibility and applicability of the algorithm enhancements.关键词
圆弹簧缺陷检测/YOLOv5/数据增强/轻量化Key words
circular spring defect detection/YOLOv5/data augmentation/model lightweighting分类
信息技术与安全科学引用本文复制引用
戴永刚,朱亚斌,高国章,马文娟,裴志彪,王栋,高鹏..基于改进YOLOv5的机车圆弹簧缺陷检测[J].机电工程技术,2025,54(6):144-149,6.基金项目
中国铁路兰州局科技发展计划项目(FWTPZDJXJ-170) （FWTPZDJXJ-170）
