基于LMD信号重构和支持向量机的柱塞泵故障诊断分析  

Fault Diagnosis Analysis of Plunger Pump Based on LMD Signal Reconstruction and Support Vector Machine

作　　者：洪晓艺[1] 翟东媛 乔庆鹏 HONG Xiao-yi;ZHAI Dong-yuan;QIAO Qing-peng(Department of Electronic Information,Xinxiang Vocational and Technical College,Xinxiang,Henan 453006;Department of Electronic and Information Engineering,Hunan University,Changsha,Hunan 410082;College of Artificial Intelligence,Henan University of Finance and Economics,Zhengzhou,Henan 450046)

机构地区：[1]新乡职业技术学院,电子信息系,河南新乡453006 [2]湖南大学,电子信息工程系,湖南长沙410082 [3]河南财政金融学院,人工智能学院,河南郑州450046

出　　处：《液压与气动》2021年第6期91-96,共6页Chinese Hydraulics & Pneumatics

基　　金：湖南省自然科学基金青年基金(2019JJ50091);河南省高等学校重点科研项目(18A150013)。

摘　　要：为了提高基于机器学习的柱塞泵故障诊断效率,在柱塞泵故障5种状态振动信号基础上,提出基于局部均值分解(Local Mean Decomposition,LMD)信号重构和支持向量机(Support Vector Machine,SVM)的柱塞泵故障诊断方法。对消噪信号进行LMD分解,将重构信号与原始信号的样本熵进行对比。通过相关系数法处理分解后的PF分量和原始振动信号,以低相关性的分量作为噪声信号,同时重构高相关性的分量。结果表明:每种状态重构信号和原始信号之间的相关系数都达到0.9以上,说明重构信号内已经含有原始信号主要信息。各状态重构信号样本熵形成了比原始信号样本熵更优的分布状态,说明LMD重构信号可以减弱噪声对故障特征提取造成的影响。200组样本中识别准确率高达99%,表明以SVM多类分类器可以获得较高的故障识别诊断准确率。相对于原始信号,LMD重构信号达到了更高的训练准确度与测试准确性,表现出很好的计算精度。In order to improve the efficiency of plunger pump fault diagnosis based on machine learning,based on the vibration signals of five states of plunger pump fault,a fault diagnosis method based on LMD signal reconstruction and support vector machine is proposed.The denoising signal was decomposed by LMD,and the reconstructed signal was compared with the original signal's sample entropy.The decomposed PF component and the original vibration signal are processed by the correlation coefficient method.The low-correlation component is taken as the noise signal,and the high-correlation component is reconstructed at the same time.The results show that the correlation coefficient between the reconstructed signal and the original signal in each state reaches more than 0.9,indicating that the reconstructed signal already contains the main information of the original signal.The reconstructed signal sample entropy of each state forms a better distribution state than the original signal sample entropy,which indicates that the reconstructed signal of LMD can reduce the influence of noise on fault feature extraction.In 200 groups of samples,the recognition accuracy is up to 99%,which indicates that the SVM multi-class classifier can achieve high fault recognition and diagnosis accuracy.Compared with the original signal,the reconstructed LMD signal achieves higher training accuracy and test accuracy,showing good calculation accuracy.

关 键 词：故障诊断 柱塞泵 极限学习 支持向量机 仿真分析 

分 类 号：TH137[机械工程—机械制造及自动化]