基于低秩和稀疏分解的滚动轴承故障特征提取方法对比研究  

Contrastive study on fault feature extraction methods for rolling bearing based on low rank and sparse decomposition

作　　者：王冉 黄裕春 张军武 余亮 WANG Ran;HUANG Yuchun;ZHANG Junwu;YU Liang(School of Logistics Engineering,Shanghai Maritime University,Shanghai 201306,China;State Key Laboratory of Mechanical System and Vibration,Shanghai Jiao Tong University,Shanghai 200240,China)

机构地区：[1]上海海事大学物流工程学院,上海201306 [2]上海交通大学机械系统与振动国家重点实验室,上海200240

出　　处：《振动与冲击》2023年第21期182-191,共10页Journal of Vibration and Shock

基　　金：国家自然科学基金(51505277,12074254);上海市自然科学基金(21ZR1434100);机械系统与振动国家重点实验室自主课题(MSVZD202201);机械系统与振动国家重点实验室开放基金课题(MSV202305)。

摘　　要：滚动轴承是机械设备中广泛使用的关键部件,其故障特征的准确提取对设备稳定运行至关重要。轴承的初始故障很微弱,容易被背景噪声掩盖,这使故障特征的提取较为困难,需要对轴承故障特征与噪声的特性进行准确刻画。针对上述问题,为了深入探究轴承故障特征及噪声在时频域中的低秩与稀疏特性及其内在关联,对轴承故障特征提取低秩稀疏分解框架下的两种代表性方法开展对比研究,以便充分利用故障特征与噪声成分的性质,为噪声干扰下的轴承故障提取方法选择提供一定的依据。利用周期性瞬态冲击信号在时频域中的稀疏与低秩特性建立矩阵分解模型,对比了Go分解(go-decomposition,Go-Dec)和非负矩阵分解(non-negative matrix factorization,NMF)两种具有代表性的分解方法,并将其应用于时频域中滚动轴承的故障特征提取。首先,基于短时傅里叶变换(short time Fourier transform,STFT)生成振动信号的时频矩阵,并揭示了轴承故障脉冲在时频域中具有的稀疏性和低秩性。利用Go-Dec和NMF两种矩阵分解方法,分解出表征故障特征的矩阵。最后,对分解的故障矩阵采用逆短时傅里叶变换重构瞬态脉冲信号,并对该信号取包络谱从而确定滚动轴承的故障类型和频率信息。仿真分析和试验对比了两种故障特征分解方法,结果表明Go-Dec可以更好地去除噪声干扰,有效提取出表征滚动轴承故障特征的稀疏分量。Rolling bearings are widely used key components in mechanical equipment,and accurate extraction of their fault features is crucial for stable operation of equipment.Initial fault of bearing is very weak and easily covered by background noise to make it be difficult to extract its fault features.It is necessary to accurately characterize characteristics of bearing fault features and noise.Here,aiming at the above problems mentioned,low rank and sparse characteristics of bearing fault features and noise in time-frequency domain as well as their inherent correlations were studied,contrastive study on 2 representative methods for bearing fault feature extraction was conducted under the framework of low rank and sparse decomposition to fully utilize properties of fault features and noise components,and provide a certain basis for selecting bearing fault extraction methods under noise interference.Matrix factorization model was established using sparse and low rank characteristics of periodic transient impact signals in time-frequency domain.2 representative decomposition methods,go-decomposition(Go-Dec) and non-negative matrix factorization(NMF) were compared and applied in fault feature extraction of rolling bearing in time-frequency domain.Firstly,time-frequency matrix of vibration signals was generated based on short time Fourier transform(STFT),and sparsity and low rank of bearing fault pulses in time-frequency domain were revealed.Then,Go-Dec and NMF were used to decompose the matrix characterizing fault features.Finally,inverse STFT was performed for the decomposed matrices to reconstruct transient pulse signal,and the envelope spectrum of transient pulse signal was used to determine fault type and frequency information of rolling bearing.The two fault feature decomposition methods were compared using simulation analysis and experiments.The results showed that Go-Dec can better remove noise interference,and effectively extract sparse components characterizing fault features of rolling bearing.

关 键 词：滚动轴承 故障特征提取 短时傅里叶变换(STFT) Go分解(Go-Dec) 非负矩阵分解(NMF) 

分 类 号：TH212[机械工程—机械制造及自动化] TP29[自动化与计算机技术—检测技术与自动化装置]