基于多维声纹特征与XGBoost模型的城轨钢轨波磨故障识别方法  

Rail corrugation fault recognition of urban rail transit based on multidimensional acoustic features and XGBoost model

作　　者：刘力 袁英强 王鑫 吕彦朋 戴泽宇 LIU Li;YUAN Yingqiang;WANG Xin;LYU Yanpeng;DAI Zeyu(Urban Rail Transit Center of China Academy of Railway Sciences Group Co.,Ltd.,Beijing 100081,China;Beijing Engineering Consultation Co.,Ltd.of CARS,Beijing 100081,China;Beijing Ensonictech Co.,Ltd.,Beijing 100080,China)

机构地区：[1]中国铁道科学研究院集团有限公司城市轨道交通中心,北京100081 [2]铁科院(北京)工程咨询有限公司,北京100081 [3]北京谛声科技有限责任公司,北京100080

出　　处：《现代城市轨道交通》2025年第3期111-116,共6页Modern Urban Transit

基　　金：中国铁道科学研究院集团有限公司基金资助项目(2023YJ093)。

摘　　要：为提高城市轨道交通钢轨波磨故障的检测精度和效率,文章提出一种基于多维声纹特征与XGBoost模型的识别方法。研究过程中,在北京地铁某线路部署声学采集设备,收集波磨和正常轨道的声学数据,构建钢轨波磨故障识别数据集。通过对时域与频域特征的提取与分析,建立多维声纹特征集,并采用特征选择与模型参数优化策略对模型进行训练和测试。实验结果表明,相较于传统单一声压级特征,多维声纹特征在不同机器学习算法上的性能均得到显著提升,验证了多维特征在故障识别中的有效性。同时,采用XGBoost算法的钢轨波磨故障识别模型平均准确率达到95.6%,显著优于其他经典机器学习算法。此外,该模型的漏检率和误检率均维持在较低水平,进一步证实其在实际应用中的可靠性与有效性。To improve the detection accuracy and efficiency of rail corrugation faults in urban rail transit,this article proposes a recognition method based on multidimensional acoustic features and the XGBoost algorithm.During the process of the research,acoustic acquisition devices were deployed on a line of Beijing metro to collect acoustic data from both corrugated and normal rails,so as to construct a rail corrugation fault recognition dataset.Through the extraction and analysis of time-domain and frequency-domain features,a multidimensional acoustic feature set was established,and feature selection and model parameter optimization strategies were employed to train and test the model.Experimental results demonstrate that,compared to traditional single sound pressure level features,multidimensional acoustic features significantly improved the performance across various machine learning algorithms,validating the effectiveness of multidimensional features in fault recognition.Furthermore,the rail corrugation fault recognition model utilizing the XGBoost algorithm achieved an average accuracy of 95.6%,significantly outperforming other classical machine learning algorithms.Additionally,the model maintained low levels of rate of missed detection and false detection,further confirming its reliability and effectiveness in practical application.

关 键 词：城市轨道交通 钢轨波磨 声学检测 XGBoost 

分 类 号：U221.3[交通运输工程—道路与铁道工程]