引用本文:尹逊龙,牟宗磊,王友清.基于DVMD降噪的旋转机械故障诊断[J].控制理论与应用,2022,39(7):1324~1334.[点击复制]
YIN Xun-long,MU Zong-lei,WANG You-qing.Fault diagnosis of rotating machinery based on DVMD denoising[J].Control Theory and Technology,2022,39(7):1324~1334.[点击复制]
基于DVMD降噪的旋转机械故障诊断
Fault diagnosis of rotating machinery based on DVMD denoising
摘要点击 1229  全文点击 358  投稿时间：2021-06-15  修订日期：2022-06-21
查看全文  查看/发表评论  下载PDF阅读器
DOI编号  10.7641/CTA.2022.10509
  2022,39(7):1324-1334
中文关键词  深度变分模态分解  麻雀搜索算法  降噪  深度学习  特征提取  故障诊断
英文关键词  depth variational mode decomposition(DVMD)  sparrow search algorithm(SSA)  denoising  deep learning  feature extraction  fault diagnosis
基金项目  青岛创业创新领军人才计划项目(19-3-2-4-zhc), 山东省自然科学基金项目(ZR2021MF027), 青岛市博士后应用研究项目资助.
作者单位E-mail
尹逊龙 山东科技大学电气与自动化工程学院 2591819014@qq.com 
牟宗磊 山东科技大学电气与自动化工程学院  
王友清* 山东科技大学电气与自动化工程学院 wang.youqing@ieee.org 
中文摘要
      针对振动信号噪声难以剔除而造成故障诊断精度低的问题, 提出了一种基于深度变分模态分解(DVMD)的 旋转机械故障诊断方法. 首先, 利用麻雀算法(SSA)对变分模态分解(VMD)算法的参数进行优化. 然后, 通过SSA– VMD对信号进行自适应深度分解得到模态分量, 将每层深度的分量与原始信号作皮尔逊相关系数分析, 再对分量 进行奇异值分解(SVD)或者直接剔除, 将处理后分量重构后, 实现振动信号的深度降噪. 最后, 提取降噪信号的一维 多尺度排列熵特征和二维时频特征, 将特征依次放入轻量级梯度提升机(LightGBM)中进行训练, 实现故障诊断. 设 计方法在风力涡轮传动系统的齿轮箱故障信号上进行验证, 不仅能够剔除信号的大量噪声, 并且提高了故障诊断精 度, 具有良好的工程应用前景.
英文摘要
      Aiming at the problem that the noise of vibration signal is difficult to eliminate, which results in low accuracy of fault diagnosis, a fault diagnosis method of rotating machinery based on deep variational mode decomposition (DVMD) is proposed. First, the sparrow search algorithm (SSA) is used to optimize the parameters of variational mode decomposition (VMD). Then, SSA–VMD is used to perform the adaptive depth decomposition of the original signal, which obtains components of the signal. The components of each depth are analyzed by Pearson correlation coefficient with the original signal. Then the components are processed by singular value decomposition (SVD) or directly removed. The processed components are reconstructed, which achieves the denoising of vibration signals. Finally, the one-dimensional multiscale entropy features and the two-dimensional time-frequency features of the signal are extracted. Those features are put into the light gradient boosting machine (LightGBM) for training, which achieves fault diagnosis. This method is verified on gearbox fault signals of wind turbine drivetrain, which can not only eliminate a large number of noise in the signal, but also improve the accuracy of fault diagnosis, and has a good engineering application prospect.