摘要
随着铁路运输系统的高速发展,车载非接触式钢轨磨耗自动化检测任务日益增加,然而,线路上丰富多样的钢轨轨形轮廓和多样离群点干扰,导致磨耗值在线计算错误频发,且运营列车日趋海量的检测数据对于实时性的要求越来越高。因此,提出一种基于随机森林(RF)的钢轨轮廓在线识别方法,分类模型1利用主分量分析(PCA)提取钢轨轮廓的全局特征,分类模型2采用相对高度二叉树提取钢轨轮廓的局部特征,随后采用RF算法模型对转化后的低维特征向量进行分类识别。与支持向量机(SVM)分类识别算法进行对比,结果表明,所提出的基于RF算法的分类模型1对于普通钢轨轮廓和非普通钢轨轮廓的整体分类识别准确率为98.7%,单帧识别耗时8.57 ms;分类模型2对于鱼尾板、道岔尖轨和其他轮廓的整体分类识别准确率为96.7%,单帧识别耗时11.95 ms,满足运营列车75 km/h的实时在线检测钢轨轮廓需求,具有工程应用价值。
With the rapid development of railway transportation systems,the tasks of automated detection of non-contact steel rail wear have increased dramatically.However,the rich and diverse rail profiles and various outliers on the track lead to frequent errors in online wear calculation,and the massive amount of detection data from operating trains increase requirement for real-time detection.This paper proposes an online rail profile recognition method based on Random Forest algorithm(RF).Classification model one utilizes Principal Component Analysis(PCA)to extract the global features of the rail profile,while classification model two uses a relative height binary tree to extract the local features of the rail profile.Subsequently,the RF algorithm is used to classify and recognize the transformed low-dimensional feature vectors.Compared with the Support Vector Machine(SVM)classification recognition algorithm,the results show that the proposed classification model one based on RF algorithm achieves an overall recognition accuracy of 98.7%for common and non-common rail profiles,with a single-frame recognition time of 8.57 ms.Classification model two achieves an overall recognition accuracy of 96.7%for fishplate,switch point rail,and other profiles,with a single-frame recognition time of 11.95 ms.The proposed recognition method meets the real-time online detection requirements of rail profiles for operating trains at 75 km/h and thus it has certain engineering application value.
作者
刘震锋
陈建政
赵春云
王佳月
LIU Zhenfeng;CHEN Jianzheng;ZHAO Chunyun;WANG Jiayue(State Key Laboratory of Rail Transit Vehicle System,Southwest Jiaotong University,Chengdu 610031,China)
出处
《机械》
2024年第10期52-58,共7页
Machinery
关键词
钢轨轮廓
在线计算
特征提取
随机森林
轮廓识别
rail profile
online calculation
feature extraction
random forest
profile recognition
