摘要
高铁的飞速发展使其基础设施的在线智能维保技术需求更加迫切,其中包括基于机器视觉技术实现高铁吊弦结构的缺陷检测。在预处理后使用深度学习目标检测的SSD算法学习大量数据中需要定位的吊弦线夹位置,并在校正后使用SVM学习这些数据的“相对峭度”阈值。通过实验数据评估定位准确性并根据学习到的“相对峭度”阈值判断每个吊弦的受力状态,检出吊弦严重缺陷,实验结果验证了该方法的有效性。
Upon the rapid development of the high-speed train,the speed train dropper defect detection based on machine vision technology is in urgent need by online intelligent maintenance technology.The SSD algorithm for deep learning object detection was used to learn the dropper in need to be located in a large amount of data after preprocessing.SVM was applied to learn‘relative kurtosis’from these data after image correction.The accuracy of positioning was evaluated with experimental data,and the state of each dropper was judged according to the learned threshold.The results prove the method to be effective.
作者
陈云莎
张兵
孙琦
闫磊
CHEN Yunsha;ZHANG Bing;SUN Qi;YAN Lei(State Key Laboratory of Traction Power,Southwest Jiaotong University,Chengdu 610036,China;CRRC Qingdao Sifang Co.,Ltd.,Qingdao 266031,China)
出处
《机械制造与自动化》
2021年第3期167-170,189,共5页
Machine Building & Automation
基金
国家重点研发计划项目(2017YFB1201103)。
关键词
吊弦
机器视觉
峭度表征
缺陷检测
dropper
machine vision
kurtosis representation
defect inspection
