SDE网络在高速列车转向架未知故障诊断中的应用  

Application of SDE-net in Out-of-distribution Fault Diagnosis of High-speed Train Bogie

作　　者：张一鸣 秦娜[1] 吴培栋 杜家豪 吴比 ZHANG Yi-ming;QIN Na;WU Pei-dong;DU Jia-hao;WU Bi(School of Electrical Engineering,Southwest Jiaotong University,Chengdu 611730,China;Beijing Quanlu Communication Signal Research and Design Institute Group Co.,Ltd.,Beijing 100000,China)

机构地区：[1]西南交通大学电气工程学院,四川成都611730 [2]北京全路通信信号研究设计院集团有限公司,北京100000

出　　处：《控制工程》2022年第2期300-306,共7页Control Engineering of China

基　　金：四川省科技计划资助项目(2020YFQ0057,2021JDJQ0012)。

摘　　要：转向架作为高速列车车体与轨道的连接部位,承载着保证列车在轨道上安全运行的重任。然而,在列车长期服役过程中,轨道不平顺以及轮轨磨耗等原因会造成转向架部件故障,严重影响列车的安全运行。实际运行过程中故障的发生具有随机性,无法将转向架故障诊断简单归类于已知组别分类问题。针对深度学习无法辨别未知故障的缺陷,在卷积神经网络(CNN)中引入随机微分方程(SDE)对转向架已知以及未知故障进行判别。实验证明,随机微分方程网络(SDE-net)不仅能高效分辨出已知故障,还能有效判别出未知故障,且准确率都超过93%。与此同时,通过与一维CNN网络比较体现该方法的优越性。As the connecting system of the high-speed train body and the track,the bogie plays an important part in ensuring the safe operation of the trains on the track.However,the track irregularity and wheel-rail wear will cause the bogie to malfunction during the long-term service of the trains,which will seriously affect the safe operation of the trains.The occurrence of faults during actual operation is random.Therefore,it is inappropriate to classify fault diagnosis of bogie into the problem of known category classification simply.Aiming at the defect that deep learning can not distinguish unknown faults,stochastic differential equation(SDE)is added into the convolutional neural network(CNN)to distinguish known and unknown bogie faults.Experiments show that the stochastic differential equation network(SDE-net)can not only distinguish known faults accurately,but also distinguish unknown faults effectively,and the accuracy rates are over 93%.At the same time,the superiority of the proposed method is demonstrated by comparison with the one-dimensional CNN network.

关 键 词：转向架 故障诊断 SDE网络 未知故障 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]