基于多尺度卷积神经网络的动车组电缆终端故障诊断研究  

Research on Fault Diagnosis of EMU Cable Terminal Based on Multi Scale Convolution Neural Network

作　　者：焦京海 曹伟东 JIAO Jinghai;CAO Weidong(CRRC Qingdao SIFANG Co.,Ltd.,Qingdao 266000,China;College of Electrical Engineering Southwest Jiaotong University,Chengdu 611756,China)

机构地区：[1]中车青岛四方机车车辆股份有限公司,山东青岛266000 [2]西南交通大学电气工程学院,四川成都611756

出　　处：《电工技术》2021年第13期29-32,共4页Electric Engineering

摘　　要：为对动车组电缆典型故障进行高效、准确评估,提出了一种基于多尺度卷积神经网络的诊断方法。首先人工制作含四种典型缺陷的动车组电缆试样,对其进行局部放电测试,并将采集的局部放电信号去噪;然后将去噪后的局部放电信号导入多尺度卷积神经网络进行深度特征学习;最后通过softmax分类器对电缆故障进行评估。结果表明,针对四种典型的电缆故障,该方法能保持较高识别率,且在识别率和耗费时间方面均优于其他传统故障诊断方法,具有较好的工程应用前景。In order to evaluate the typical faults of EMU cable efficiently and accurately,a diagnosis method based on multi-scale convolution neural network is proposed.Firstly,the samples of EMU cable with four typical defects are made by hand,the partial discharge test is carried out,and the collected partial discharge signal is de-noising.Secondly,the de-noising partial discharge signal is imported into multi-scale convolution neural network for deep feature learning.Finally,the cable fault is evaluated by softmax classifier.The results show that:for four typical cable faults,the proposed method can maintain a high recognition rate,and is superior to other traditional fault diagnosis methods in recognition rate and time-consuming.The proposed method has a good engineering application prospect.

关 键 词：动车组电缆 卷积神经网络 局部放电 故障诊断 

分 类 号：TM855[电气工程—高电压与绝缘技术]