基于振动响应的高速动车车轮踏面磨耗状态识别  

Identification of Wheel Tread Wearing Condition of High-speed EMU Based on Vibration Response

作　　者：吴锐东 沈龙江 姚远[1] 孟凡愚 WU Ruidong;SHEN Longjiang;YAO Yuan;MENG Fanyu(State Key Laboratory of Rail Transit Vehicle System,Southwest Jiaotong University,Chengdu 610031,China;State Key Laboratory for Integration of High-power AC Drive Electric Locomotive,CRRC Zhuzhou Locomotive Co.,Ltd.,Zhuzhou 412001,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 [2]中车株洲电力机车有限公司大功率交流传动电力机车系统集成国家重点实验室,湖南株洲412001

出　　处：《铁道车辆》2024年第4期26-32,95,共8页Rolling Stock

基　　金：国家自然科学基金(U2268211);四川省自然科学基金(2022NSFSC0034,2022NSFSC1901);牵引动力国家重点实验室自主研究课题(2022TPL-T02)。

摘　　要：针对高速动车车轮踏面磨耗状态识别问题,融合车辆振动响应及其幅值滑动平均值特征,提出一种基于双向门控循环单元(Bi-directional Gated Recurrent Unit,Bi-GRU)网络的车轮踏面磨耗状态识别方法。首先,以国内某型高速动车为对象,采用SIMPACK进行车辆动力学仿真,定义五类不同等效锥度磨耗状态的车轮踏面,考虑悬挂结构参数、轮轨接触状态等随机因素,获取车体与构架横向振动加速度;其次,结合横向加速度及其幅值滑动平均值,使用特征融合技术建立不同磨耗踏面的多维特征融合数据集;最后,采用Bi-GRU网络建立踏面磨耗状态识别模型,训练模型结构参数并验证其识别效果。结果表明:该模型能够有效识别高速动车车轮踏面磨耗状态,识别准确率优于传统最近邻、决策树和支持向量机模型,并具备一定抗噪能力。对于线路实测数据集,迁移模型准确率超过98%,证明模型具有良好的泛化性能。To identify the wheel tread wearing condition of high-speed EMU,a wheel tread wearing condition identification method based on the Bi-GRU(Bi-directional Gated Recurrent Unit) network is proposed by combining the vibration response of the vehicle and the feature of its average amplitude sliding value.Firstly,a certain model of Chinese high-speed EMU is taken for object,the SIMPACK is used for dynamics simulation of the vehicle,five types of wheel tread with different equivalent taper wearing conditions are defined,the parameters of the suspension structure,the wheel-rail contact condition and other random factors are taken into account to obtain the lateral vibration acceleration between the carbody and the frame.Secondly,by combining the lateral acceleration and its average amplitude sliding value,the feature fusion technology is used to establish the multi-dimensional feature fusion data set for different wearing treads.Finally,the Bi-GRU network is used to build the tread-wearing condition identification model,the structural parameters of the model are trained and its identification effect is verified.It turns out that this model can effectively identify the wheel tread wearing condition of high-speed EMU and its accuracy of identification is better than that of conventional nearest neighbors,decision trees,and support vector machine models.It also has anti-noise properties.For the actual measured data set of the track,the accuracy of the transport model is higher than 98%,proving the good generalization performance of this model.

关 键 词：车轮踏面磨耗 等效锥度 特征融合 状态识别 双向门控循环单元 

分 类 号：U213.5[交通运输工程—道路与铁道工程]