摘要
在现有轨道车辆轴承温度预警研究中,因监测数据复杂度不一致导致特征难以选择,同时现有预警方法往往只能在轴承故障发生前的几分钟进行预警,为此,提出一种基于特征选择的轨道车辆轴承温度预警方法。首先采用皮尔逊系数计算特征相关性后分析引入关联轴承,然后依据线性相关性将低线性相关特征数据与关联轴承数据一起输入LightGBM模型,以对特征进行再次选择;其次,利用大量正常状态下的履历数据,基于深度学习模型双向门控循环单元构建轴承温度预测模型;最后利用某轨道车辆实测数据进行预警方法验证。结果表明:对于正常轴承,轴承温度预测模型的温度预测值和实际值的差异小于4℃且稳定;而对于异常轴承,在轴承故障发生前的数小时即可发现两者间存在大于4℃以上的持续显著差异。
In the existing early warning research of rail vehicle bearing temperature,the inconsistency of the monitoring data makes it difficult to select features,and the existing early warning method only provide early warning a few minutes before the bearing failure occurs.Therefore,a bearing temperature warning method for rail vehicles based on feature selection was proposed.Firstly,the Pearson coefficient was used to calculate the feature correlation and then the low linear correlation feature and the related bearing data were input into the LightGBM model according to the linear correlation to select the feature again.Secondly,the bearing temperature prediction model was constructed based on the deep learning model bidirectional gate recurrent unit using a large amount of normal data.Finally,the actual measurement data of a rail vehicle was used to verify the early warning method,and the result shows that:for the normal bearing,the difference between the predicted value and the actual value of the bearing temperature prediction model is less than 4℃and stable;for the abnormal bearing,the difference between the predicted value and the actual value is more than 4℃in several hours before the bearing failure.
作者
蒋雨良
彭飞
曾大懿
邹益胜
张波
JIANG Yuliang;PENG Fei;ZENG Dayi;ZOU Yisheng;ZHANG Bo(School of Mechanical Engineering,Southwest Jiaotong University,Chengdu 610031,China;CRRC Industry Institute Limited Corporation,Beijing 100070,China;Technology and Equipment of Rail Transit Operation and Maintenance Key Laboratory of Sichuan Province,Chengdu 610031,China)
出处
《现代制造工程》
CSCD
北大核心
2021年第4期139-145,89,共8页
Modern Manufacturing Engineering
基金
国家重点研发计划项目(2018YFB1201901-05)
重庆市教委科学技术研究项目(KJZD-K201805801)。
