摘要
轨道车辆的运行安全问题在工程装备应用中日益受到重视,车辆长时间高速运行时转向架上易出现螺栓松脱风险,针对这一问题,文章提出基于3D点云和改进Unet网络的检测算法。该算法首先通过训练YOLOv7网络对转向架中螺栓位置进行定位,再利用改进Unet网络训练完成对螺栓标记线的语义分割,提取螺栓标记线,并对标记线对应的点云进行平面拟合,最后计算所有点到拟合平面的内点的数量占比,通过标记线的偏离程度判断螺栓故障。最终通过试验验证了本算法在检测5°以上的松动螺栓故障准确率达到98.9%。
The operation safety of rail vehicles is increasingly valued in the application of engineering equipment,and bolt looseness is likely to occur on bogies when vehicles run at high speed for a long time.In response to this issue,this paperr proposes a detection algorithm based on 3D point cloud and improved Unet network.This algorithm first locates the bolt position on the bogie by training YOLOv7 network,and then completes the semantic segmentation of the bolt marking line through improved Unet network training,extracts the bolt marking line,and carries out plane fitting for the point cloud corresponding to the marking line.Finally,the proportion of all points to the number of interior points in the fitting plane is calculated,and the bolt fault is determined by the deviation of the marking line.In the end,the accuracy of this algorithm in detecting loose bolt deviating above 5 degrees reaches 98.9%through experiments.
作者
周勇
卞耀辉
田腾翔
滑瑾
ZHOU Yong;BIAN Yaohui;TIAN Tengxiang;HUA Jin(CRRC Nanjing Puzhen Rolling Stock Co.,Ltd.,Nanjing,Jiangsu 210031,China)
出处
《轨道交通材料》
2023年第6期56-61,共6页
MATERIALS FOR RAIL TRANSPORTATION SYSTEM
基金
省级、城轨地铁车辆智能制造产线关键技术研究及示范应用(2021CKZ014-5)。
