铁道科学与工程学报2025，Vol.22Issue(5)：2346-2354,9.DOI:10.19713/j.cnki.43-1423/u.T20241206
基于瓦片编码网络的钢轨焊缝几何不平顺识别
A tile-coding-based neural network for rail weld irregularity identification
摘要
Abstract
Rail welds,as one of the three vulnerable components of the track,can affect the safety and stability of railway operations.Thus,the rapid and efficient measurement of the geometric irregularity of rail welds is crucial for ensuring the safety of railway traffic and enhancing the work efficiency of railway maintenance personnel.In this regard,this study initially utilized a hand-pushed rail short-wave geometric detection device to collect the short-wave track irregularity.Subsequently,a deep learning approach based on the fuzzy tile coding neural network was proposed,which could not only determine the centerline mileage of the rail welds from the short-wave irregularity signals but also calculate the reliability of the identification at corresponding positions.This enabled the swift separation of the geometric waveform of rail welds from a variety of mixed and complex short-wave track irregularity signals,thereby improving the efficiency of data utility.Finally,field experiments were conducted on one railway line to verify the accuracy and stability of the proposed method,and the results of the weld irregularity were compared with the standard measurement using a MCRuler straightedge by the maintenance department,to enhance the practicality of the engineering application.The findings of this research can be concluded as follows.(1)The identification accuracy of rail welds based on the tile coding network can reach 92.01%,with a recall rate of up to 94.98%.(2)Moreover,the tile coding network can accurately pinpoint the center of the rail welds,with the deviation from the actual weld center kept within 0.03 m.(3)The final results of the 1-meter-chord irregularities of the rail welds is fundamentally consistent with the amplitude results obtained from the MCRuler straightedges,with the maximum amplitude difference not exceeding 0.1 mm.Overall,this study can offer valuable engineering technical reference for enhancing the efficiency of maintenance inspection data utilization and reducing the detection costs associated with rail welds.关键词
钢轨焊缝/瓦片编码网络/短波不平顺/智能检测/焊缝平直度Key words
rail weld/tile-coding-based neural networks/short-wave track irregularity/intelligent inspection/rail weld irregularity分类
交通工程引用本文复制引用
高天赐,史一帆,江乐鹏,王源,刘晓舟,罗钦,王平..基于瓦片编码网络的钢轨焊缝几何不平顺识别[J].铁道科学与工程学报,2025,22(5):2346-2354,9.基金项目
国家自然科学基金资助项目(52008198,52208441) （52008198,52208441）
广东省普通高校创新团队资助项目(2022KCXTD027) （2022KCXTD027）
