基于EMD-SVD-BIC的机械振动源数估计方法  

Mechanical Vibration Source Number Estimation Based on EMD-SVD-BIC

作　　者：叶红仙[1,2] 杨世锡[2] 杨将新[2] 

机构地区：[1]杭州电子科技大学机械工程学院,杭州310018 [2]浙江大学机械系,杭州310027

出　　处：《振动．测试与诊断》2010年第3期330-334,共5页Journal of Vibration,Measurement & Diagnosis

基　　金：国家自然科学基金资助项目(编号:50675194);国家高技术研究发展计划("八六三"计划)资助项目(编号:2008AA04Z410)

摘　　要：提出了一种复合EMD-SVD-BIC(经验模态分解,empirical mode decomposition,简称EMD;奇异值分解,singular value decomposition,简称SVD;贝叶斯信息准则,bayesian information criterion,简称BIC)的机械振动源信号数量估计方法,解决卷积混合的机械振动源在观测数小于振动源数情况下的源数估计问题。应用EMD方法获得信号的本征模函数,对两观测信号的本征模函数复合矩阵的相关矩阵进行奇异值分解,获得反映源数信息的特征值分布;再采用BIC信息准则,判断源信号的数目。仿真和试验结果表明,该方法可以在观测数小于振动源数的情况下正确获取信号源数,为机械振动故障诊断中的振动源分析及其源信号的正确分离提供了方法保障。To estimate the number of mechanical vibration sources from the convolutive mixtures with fewer sensors than sources,a new method composed of EMD-SVDBIC（Empirical Mode Decomposition,EMD;Singular Value Decomposition,SVD;Bayesian Information Criterion,BIC） is proposed.The characteristic of vibration sources,which is intrinsic mode function（IMF） is obtained through the EMD of the mixtures.The complex matrix of the IMF of different mixtures is used to get the eigenvalue of sources by using SVD.The eigenvalue is employed to determine the number of sources by means of BIC.It is verified that the new method yields a correct estimate of source number in the simulation test and experiment. This work is help to analyze the mechanical vibrations in the fault monitoring system and to separate the useful mechanical vibration signals efficiently from the mixtures.

关 键 词：旋转机械 故障诊断 盲源分离 源数估计 

分 类 号：TN911.7[电子电信—通信与信息系统] TH165.3[电子电信—信息与通信工程]