基于小波包分解多尺度排列熵及2阶特征选择的转辙机故障诊断方法  

Fault Diagnosis Method of Switch Machine Based on Wavelet Packet Decomposition,Multi-Scale Permutation Entropy and Two-Stage Feature Selection

作　　者：孙永奎[1,2] 曹源 李鹏[1] 李旭 SUN Yongkui;CAO Yuan;LI Peng;LI Xu(School of Electronic and Information Engineering,Beijing Jiaotong University,Beijing 100044,China;National Engineering Research Center of Rail Transportation Operation and Control System,Beijing Jiaotong University,Beijing 100044,China;China Railway Society,Beijing 100844,China)

机构地区：[1]北京交通大学电子信息工程学院,北京100044 [2]北京交通大学轨道交通运行控制系统国家工程研究中心,北京100044 [3]中国铁道学会,北京100844

出　　处：《中国铁道科学》2023年第3期178-188,共11页China Railway Science

基　　金：国家重点研发计划项目(2021YFF0501102);国家自然科学基金资助项目(U1934219,52202392,52022010,52102472);中央高校基本科研业务费专项资金资助项目(2021RC276)。

摘　　要：针对转辙机高精度故障诊断的需求,结合声音信号非接触、易采集等优势,提出一种基于声音信号的非接触式故障诊断方法。首先,基于小波包分解与多尺度排列熵,实现对声音样本的特征提取;其次,提出基于ReliefF和二进制粒子群优化算法的2阶特征选择方法,得到最佳特征集合,实现对声音样本的特征选择;最后,基于支持向量机算法对最佳特征集进行训练和测试,完成对转辙机的故障诊断。依托10种常见工况下共计800组声音样本开展实验,结果表明:该方法在反位—定位和定位—反位转换过程中得到的特征点数分别为13和39个,故障诊断准确率分别为99.67%和100%;相比于单一特征选择方法,采用的2阶特征选择方法能够大大降低特征维度,提高故障诊断准确率;相比于k近邻和线性判别分析这2种分类器,支持向量机分类器在转辙机故障诊断中更具优势。Aiming at the demand for high-precision fault diagnosis of switch machines,and combining the advantages of non-contact and easy acquisition of sound signals,a non-contact fault diagnosis method based on sound signals was proposed.Firstly,feature extraction of sound samples was carried out based on wavelet packet decomposition and multi-scale permutation entropy.Secondly,a two-stage feature selection method based on ReliefF and binary particle swarm optimization algorithm was proposed to obtain the optimal feature set and realize the feature selection of sound samples.Finally,the optimal feature set was trained and tested based on the support vector machine to realize the fault diagnosis of the switch machine.The experiment was carried out based on a total of 800 groups of sound samples under 10 common working conditions.The results show that the numbers of feature points obtained by the proposed method in the switching processes of reversenormal and normal-reverse are 13 and 39,respectively,and the accuracy rates of fault diagnosis are 99.67%and 100%,respectively.Compared with the single feature selection method,the two-stage feature selection method adopted in this paper can greatly reduce the feature dimension and improve the accuracy rate of fault diagnosis.Compared with the other 2 classifiers of k-nearest neighbor and linear discriminant analysis,the support vector machine classifier has more advantages in fault diagnosis of switch machines.

关 键 词：转辙机 故障诊断 小波包分解 多尺度排列熵 2阶特征选择 支持向量机 

分 类 号：U284.92[交通运输工程—交通信息工程及控制]