基于深度学习的动车轨外闸片磨耗测量算法研究
Research on the Wear Measurement Algorithm of the Outer Brake Plate of the Rail Based on Deep Learning
刘军 1李国刚 1杨涛 2刘厚军 3杨已葱3
作者信息
- 1. 中国铁路青藏集团有限公司,西宁 810007
- 2. 中国铁路沈阳局集团有限公司梅河口机务段,通化 135000
- 3. 苏州华兴致远电子科技有限公司,苏州 215000
- 折叠
摘要
动车的行车安全是铁路运输行业的生命基线,转向架区域的闸片厚度与列车的安全行驶息息相关.轨外闸片的自动化测量能够大大提高列车的自动化检修水平和效率.基于此,提出一种基于深度学习的动车轨外闸片磨耗测量算法,采用深度学习中的YOLOv5 目标检测算法,以交叉注意力网络(Criss-Cross Attention Network,CCNet)的分割模型为载体,实现对动车轨外闸片剩余厚度的自动分析测量.与人工测量相比,所提出的算法的平均绝对误差达到1.123 mm,能够实现动车的闸片自动测量,减轻动车所人工检修的压力,保障列车安全行驶.
Abstract
The running safety of bullet trains is the life baseline of railway transportation industry.The thickness of brake plate in bogie area is closely related to the safe running of trains.The automatic measurement of the outer brake plate can greatly improve the level and efficiency of the automatic maintenance of the train.Based on this,a deep learning-based wear measurement algorithm for the outer brake plate of the rail is proposed.The YOLOv5 object detection algorithm in deep learning is adopted,and the segmentation model of Criss-Cross Attention Network(CCNet)is used as the carrier to realize the automatic analysis and measurement of the remaining thickness of the outer brake plate of the rail.Compared with manual measurement,the average absolute error of the proposed algorithm reaches 1.123 mm,which can realize the automatic measurement of the brake plate of the train,reduce the pressure of manual maintenance of the train,and ensure the safe running of the train.
关键词
动车/轨外闸片/目标检测/语义分割/厚度测量Key words
bullet train/outer brake plate of rail/target detection/semantic segmentation/thickness measurement引用本文复制引用
出版年
2024
NETL
NSTL
万方数据