基于SFA-余弦相似度的轮轨异常磨耗识别方法  

Wheel-Rail Abnormal Wear Identification Method Based on SFA-Cosine Similarity

作　　者：王佳月 陈建政[1] 刘震锋 赵春云 WANG Jiayue;CHEN Jianzheng;LIU Zhenfeng;ZHAO Chunyun(State Key Laboratory of Rail-Transit Vehicle System,Southwest Jiaotong University,Chengdu 610031,China)

机构地区：[1]西南交通大学轨道交通运载系统全国重点实验室,四川成都610031

出　　处：《机械》2025年第1期52-58,共7页Machinery

摘　　要：传统的轮轨异常磨耗检测方法过于依赖于人工经验和理论研究,检测结果容易受主观因素和理论局限性的影响,提出了一种基于符号傅里叶(SFA)-余弦相似度的轮轨异常磨耗识别方法,选取轴箱振动信号作为监测信号并对加速度信号进行时间序列分割处理,将原始时间序列信号分割成若干子序列,以充分捕捉轮轨异常磨耗的时间尺度信息,进而利用SFA时间序列符号化方法将若干个子序列时域信号转换为符号序列,利用TF-IDF词频统计方法对符号序列进行特征提取,结合改进的余弦相似度分类方法完成轮轨异常磨耗的识别。结果表明,该方法可有效识别混合叠加轮轨异常磨耗下的磨耗类型及磨耗程度,其中对叠加磨耗类型的识别准确率可达到97%以上,对叠加磨耗程度的识别准确率可达到92%以上。The traditional methods for detecting abnormal wheel rail wear rely too much on manual experience and theoretical research,and the detection results are easily affected by subjective factors and theoretical limitations.This paper proposes a wheel rail abnormal wear recognition method based on symbolic Fourier(SFA)-cosine similarity.The vibration signal of the axle box is selected as the monitoring signal and the acceleration signal is segmented into time series.The original time series signal is divided into several subsequences to fully capture the time scale information of abnormal wheel rail wear.Then,the SFA time series symbolization method is used to convert the time-domain signals of several subsequences into symbol sequences,and the TF-IDF word frequency statistical method is used to extract features from the symbol sequences.Combined with improved cosine similarity,the feature extraction of the symbol sequences is carried out.The classification method is used to identify abnormal wheel rail wear.The results show that this method can effectively identify the types and degrees of wear under mixed superimposed wheel rail abnormal wear,with an accuracy rate of over 97% for identifying superimposed wear types and over 92% for identifying superimposed wear degrees.

关 键 词：轮轨异常磨耗识别 时间序列符号化 余弦相似度度量 特征提取 

分 类 号：U216[交通运输工程—道路与铁道工程]