基于变分模态分解与流形学习的滚动轴承故障特征提取方法  

Fault feature extraction method of rolling bearings based on VMD and manifold learning

作　　者：戚晓利[1] 叶绪丹 蔡江林[1] 郑近德[1] 潘紫微[1] 张兴权[1] QI Xiaoli;YE Xudan;CAI Jianglin;ZHENG Jinde;PAN Ziwei;ZHANG Xingquan(School of Mechanical Engineering,Anhui University of Technology,Ma'anshan243032,China)

机构地区：[1]安徽工业大学机械工程学院,安徽马鞍山243032

出　　处：《振动与冲击》2018年第23期133-140,共8页Journal of Vibration and Shock

基　　金：国家自然科学基金资助项目(51505002);安徽省高校自然科学研究重点项目(KJ2017A053)

摘　　要：提出了一种基于变分模态分解(VMD)和局部切空间排列算法(LTSA)相结合的滚动轴承早期故障诊断方法。首先利用VMD算法分解圆柱滚子轴承不同运行状态下的振动信号,通过求取瞬时频率均值并绘制特征曲线筛选出与原始信号最为相关的几个分量;然后,提取有效模态分量的时域指标和小波包频带分解能量所构成的频域指标,两者结合初步提取高维故障特征后,再应用LTSA对故障特征进行二次提取;最后输入到K-means分类器进行故障类型识别;通过对圆柱滚子轴承故障诊断的对比实验分析,发现:(1)与时频特征+LTSA、EMD+LTSA特征提取方法相比,VMD+LTSA方法在分类效果和识别精度上更具优势;(2) LTSA算法相比较于PCA、LPP、LE、ISOMAP和LLE这5种算法,其降维后的特征故障敏感性最好。研究结果表明所提出的方法在圆柱滚子轴承故障诊断方面具有一定的优越性。A method for early fault diagnosis of rolling bearings based on the variational mode decomposition(VMD)combined with the local tangent space alignment(LTSA)algorithm was proposed.Firstly,the VMD algorithm was used to decompose vibration signals of a rolling bearing under different operational conditions,and several components most correlated to the original signal were screened out by solving the instantaneous frequency mean and plotting the feature curve.Then,the time domain index of the effective modal component and the frequency domain index formed with the wavelet packet frequency band decomposition energy were extracted.After these two indexes were combined to primarily extract higher dimensional fault features,LTSA was used to extract fault features again.Finally,the extracted fault features were inputted into a K-means classifier to recognize fault types.Through the contrastive test analysis of cylindrical roller bearings’fault diagnosis,it was shown that compared with the method of time frequency features extraction+LTSA and the method of EMD+LTSA,the method of VMD+LTSA has more advantages in classification effect and recognition accuracy;the LTSA algorithm has the best sensitivity to fault features after dimension reduction compared to the algorithms of PCA,LPP,LE,ISOMAP and LLE;the proposed method has a certain superiority in the fault diagnosis of cylindrical roller bearings.

关 键 词：变分模态分解 流形学习 局部切空间排列算法 故障诊断 圆柱滚动轴承 

分 类 号：TN911.7[电子电信—通信与信息系统]