基于MKLSVM的MVB端接故障诊断

Terminating fault diagnosis of MVB based on MKLSVM

列车通信网络的故障诊断一直是列车健康管理中的难点,复杂的工作环境、分布式的系统结构也使得故障难以被发现和定位.在分析多功能车辆总线(Multifunction Vehicle Bus,MVB)网络端接故障和反射形成机理的基础上,提出了一种以MVB网络物理波形参数为样本特征,结合多核学习支持向量机(Multiple Kernel Learning Support Vector Machine,MKLSVM)的网络故障诊断方法,以完成由端接电阻造成的网络故障诊断.搭建平台,进行了数据采集、模型训练、结果测试.分别利用普通支持向量机及MKLSVM对样本集进行了测试,并从不同性能度量角度评估了分类器性能.结果显示:以物理波形参数为样本特征能够表征端接网络故障的故障模式,结合支持向量机(Support Vector Machine,SVM)的模式识别方法能够有效对端接网络故障进行诊断.在查准率、查全率、分类精度、代价函数方面,MKLSVM均优于普通SVM分类器.

Due to the complex working environment and distributed structure,train communication network fault diagnosis has always been a difficult issue in the field of train prognostic and health management. Based on the analysis of MVB terminating fault and waveform reflection, a fault diagnosis method that combined physical wave parameters and MKLSVM is proposed to complete terminating fault diagnosis. A test bed has been constructed for wavefrom sampling, classifier model training and testing. In the expriments, both SVM and MKLSVM classifiers are built to evaluate their performance in different indicators. It proves to be effective in terminating fault diagnosis that extracts physical layer wave parameters to represent the terminating fault characteristics and combines SVM to diagnose treminating fault. MKLSVM is superior to single kernel SVMs on the evaluation indicators inculding precision, recall, accuracy and cost-sensitive function.