基于张量分解的地铁车轮健康指数构建

Establishing health indicators of urban rail transit train wheelsets based on tensor breakdown

地铁在运行过程中,车轮踏面、轮缘与轨道持续磨损会导致车轮性能退化。对车轮退化过程中的不同状态进行识别和评估是车辆故障预测与健康管理的关键环节。近年来,基于振动信号和机器学习的地铁车轮退化状态智能识别技术在降低人工依赖、优化镟修策略方面作用显著。但由于在实际运行过程中,地铁轮对通常受载荷、路况等因素影响,振动信号易受强噪声干扰,导致退化趋势不显著,难以直接识别其状态变化。为解决这一问题,文章提出一种基于张量重构的健康指标构建方法。该方法首先基于张量分解可有效挖掘信号高维本质信息的优点,利用张量Tucker分解获得原始信号的核心张量;其次,采用张量重构和Savitzky-Golay滤波方法对信号进行降噪处理;最后,在此基础上,利用深度自动编码器网络提取深度退化特征,建立车轮退化过程的健康指标。依托北京地铁某线路实际运行的车轮振动信号数据所进行的实验结果表明,采用该方法构建的健康指标具有良好的趋势性和单调性,能够准确描述轮对整个退化过程,所得到的异常预警位置与轮径值磨损变化和实际镟修记录准确对应。该方法的提出为地铁轮对健康管理提供一种智能化的解决方案,具有良好的实用价值。

Throughout the operation of metro vehicles,the ongoing wear of the wheel tread and rim against the track results in the degradation of wheel performance.Recognizing and evaluating various states of wheel degradation is a crucial aspect of predicting vehicle faults and managing the health of vehicles.In recent years,intelligent identification of wheel degradation states in metro systems,based on vibration signals and machine learning,has played a significant role in reducing manual dependence and optimizing reprofiling strategies.However,in the actual operation process,metro wheelsets are typically influenced by factors such as load and rail conditions.The vibration signal is susceptible to substantial noise interference,leading to an inconspicuous degradation trend,which makes it difficult to directly identify the changes in state.To address this challenge,this article suggests a method for constructing a health indicator based on tensor reconstruction.This method involves firstly obtaining the core tensor of the original signal using tensor tucker breakdown,which is based on the advantage that tensor breakdown can effectively excavate the highdimensional essential information of the signal.Secondly,the tensor reconstruction and Savitzky-Golay fi ltering methods are used to reduce the noise of the signal.Finally,based on this,a deep autoencoder network is used to extract the deep degradation features,and to establish the health indicators of the wheel degradation process.The experimental results using wheel vibration signal data from a Beijing metro line demonstrate that the health index constructed through this method exhibits a favorable trend and monotonicity.It accurately describes the whole wheelset degradation process,and the abnormal warning position obtained corresponds accurately to the change in wheel diameter wear and the actual reprofi ling records.This proposed method offers an intelligent solution for the health management of metro wheelsets,proving to be of signifi cant practical value.