基于PCA和TCN-Attention的重载铁路钢轨剥离伤损退化趋势预测  

Prediction of Stripping Damage and Degradation Trend of Heavy-haul Railway Rails Based on PCA and TCN-Attention

作　　者：王忠美 吴海波 刘建华[1] 何静[1] 聂芃轩 WANG Zhong-mei;WU Hai-bo;LIU Jian-hua;HE Jing;NIE Peng-xuan(College of Railway Transportation,Hunan University of Technology,Zhuzhou 412007,China)

机构地区：[1]湖南工业大学轨道交通学院,株洲412007

出　　处：《科学技术与工程》2024年第28期12333-12341,共9页Science Technology and Engineering

基　　金：国家重点研发计划(2021YFF05011);湖南省教育厅青年项目(22B0586);湖南省教育厅项目(2022JGYB185)。

摘　　要：重载铁路在高强度的运输过程中极易导致钢轨产生剥离、磨耗等伤损影响行车安全,为了保证铁路的安全运行,对钢轨的伤损状态监测和预测是非常重要的。然而,目前钢轨伤损检测方法主要以人工道路巡检为主,检测结果存在主观性强、伤损程度量化难、伤损退化趋势预测难等问题。针对现有问题,提出一种基于主成分分析(principal component analysis,PCA)和TCN-Attention(temporal convolutional networks with attention)的重载铁路钢轨剥离伤损退化趋势预测新方法。首先,从钢轨剥离伤损振动信号中提取时域、频域特征,并采用PCA对提取到的高维特征进行降维;其次,利用时序样本间特征的差异性,构建出钢轨剥离伤损退化指标描述退化趋势性,解决伤损状态度量难的问题;利用TCN网络模型结合Attention机制对有效特征的关注提升模型的预测精度;最后,利用某铁路机务段采集的钢轨从正常到出现损伤直至失效的全生命周期振动数据,对所提方法的有效性进行验证,实验结果表明:所提出的方法能准确地预测钢轨剥离伤损的退化趋势。During high-intensity transportation on heavy-haul railways,rail stripping damage,abrasion,and other forms of damage that compromise driving safety can occur easily.Monitoring and predicting the condition of the rails are crucial steps to ensure the safe operation of the railway.However,current methods for detecting rail damage primarily rely on manual road inspections.These methods often suffer from issues such as strong subjectivity,challenges in quantifying the extent of damage,and difficulty in predicting the degradation trend of the damage.In response to the aforementioned issues,a novel approach was introduced to predict the stripping damage degradation trend of heavy-haul railway rails based on principal component analysis(PCA)and temporal convolutional networks with attention(TCN-Attention).Firstly,time domain and frequency domain features were extracted from the vibration signal associated with rail stripping damage.Subsequently,PCA was applied to reduce the dimensionality of the extracted high-dimensional features.Secondly,leveraging the distinctions in characteristics among time series samples,a degradation index for rail stripping damage was formulated to depict the degradation trend and address the challenge of measuring the damage state.The TCN network model,coupled with the Attention mechanism,was employed to focus on significant features and enhance the prediction accuracy of the model.Finally,the effectiveness of the proposed method was validated using vibration data collected throughout the entire life cycle of the rail,from normal to damaged to failure stages,obtained from a railway locomotive depot.Experimental results demonstrate that the proposed method accurately predicts rail stripping damage and loss of degradation trend.

关 键 词：剥离伤损 退化趋势 时间卷积网络 注意力机制 

分 类 号：U239.4[交通运输工程—道路与铁道工程]