基于谱图和约束NMF的滚动轴承特征频谱提取算法
|
摘要
为了克服传统时频分析中信号特征频谱提取技术中参数敏感问题,设计了一种基于谱图和约束NMF的特征频谱提取算法。该算法首先对振动信号进行归一化预处理和短时Fourier分析,获得代表原始非平稳信号特性的瞬时参数即谱图;然后对谱图进行约束NMF分解;最后由基矩阵获得原始振动信号的特征频谱。理论分析、仿真和工程试验验证了该算法的可行性和有效性。
In order to solve the problem about sensitive parameters in signal feature spectrum extraction of traditional time-frequency analysis,a feature spectrum extraction algorithm is proposed based on spectrogram and Constrained Non-Negative Matrix Factorizations( CNMF). Firstly,the vibration signal is normalized,and then the short time Fourier analysis is carried out,obtaining the instantaneous parameters that represent original non-stationary signal characteristics. Secondly,the spectrogram is decomposed using CNMF. Finally,the feature spectrum of original vibration signal is got by the basic matrix. The feasibility and effectively of the proposed method are verified by theoretical analysis, simulation and engineering experiment.
In order to solve the problem about sensitive parameters in signal feature spectrum extraction of traditional time-frequency analysis,a feature spectrum extraction algorithm is proposed based on spectrogram and Constrained Non-Negative Matrix Factorizations( CNMF). Firstly,the vibration signal is normalized,and then the short time Fourier analysis is carried out,obtaining the instantaneous parameters that represent original non-stationary signal characteristics. Secondly,the spectrogram is decomposed using CNMF. Finally,the feature spectrum of original vibration signal is got by the basic matrix. The feasibility and effectively of the proposed method are verified by theoretical analysis, simulation and engineering experiment.
引文
[1]Gabor D.Theory of Communication[J].Institution of Electrical Engineering,1946,93:429-457.
[2]Ville J.Theorie et Application de la Notion de Signal Analytique[J].Cables et Transmission,1948,2A:61-74.
[3]陈卫刚,戚飞虎.可行方向算法与模拟退火结合的NMF特征提取方法[J].电子学报,2003,31(S1):2 190-2 193.
[4]宦若虹,杨汝良.基于小波域NMF特征提取的SAR图像目标识别方法[J].电子与信息学报,2009,31(3):588-591.
[5]高涛,何明一.改进投影梯度非负矩阵分解的单训练样本特征提取研究[J].电子与信息学报,2010,32(5):1 121-1 125.
[6]李乐,章毓晋.基于线性投影结构的非负矩阵分解[J].自动化学报,2010,36(1):23-39.
[7]李建宏,姜同敏,何玉珠,等.基于NMF的SVM故障诊断方法[J].北京航空航天大学学报,2012,38(12):1-5.
[8]Lee Danlel D,Seung Sebastion H.Learning the Parts of Objects by Non-negative Matrix Factorization[J].Nature,1999,401:788-791.
[9]Donoho D,Stodden V.When does Mon-negative Matrix Factorization Give a Correct Decomposition into Parts[J].Advances in Neural Information Processing Systems,2004,16:1 141-1 148.
[10]李宁,史铁林.基于非负矩阵分解的盲信号源数估计[J].中国机械工程,2007,18(22):2 734-2 737.
[11]Bearing Data Center.Case Western Reserve University Bearing Dada[EB/OL].[2011-12-10].http://csegroups.case.ed./bearingdatacenter/pages/downloaddata-file.
[2]Ville J.Theorie et Application de la Notion de Signal Analytique[J].Cables et Transmission,1948,2A:61-74.
[3]陈卫刚,戚飞虎.可行方向算法与模拟退火结合的NMF特征提取方法[J].电子学报,2003,31(S1):2 190-2 193.
[4]宦若虹,杨汝良.基于小波域NMF特征提取的SAR图像目标识别方法[J].电子与信息学报,2009,31(3):588-591.
[5]高涛,何明一.改进投影梯度非负矩阵分解的单训练样本特征提取研究[J].电子与信息学报,2010,32(5):1 121-1 125.
[6]李乐,章毓晋.基于线性投影结构的非负矩阵分解[J].自动化学报,2010,36(1):23-39.
[7]李建宏,姜同敏,何玉珠,等.基于NMF的SVM故障诊断方法[J].北京航空航天大学学报,2012,38(12):1-5.
[8]Lee Danlel D,Seung Sebastion H.Learning the Parts of Objects by Non-negative Matrix Factorization[J].Nature,1999,401:788-791.
[9]Donoho D,Stodden V.When does Mon-negative Matrix Factorization Give a Correct Decomposition into Parts[J].Advances in Neural Information Processing Systems,2004,16:1 141-1 148.
[10]李宁,史铁林.基于非负矩阵分解的盲信号源数估计[J].中国机械工程,2007,18(22):2 734-2 737.
[11]Bearing Data Center.Case Western Reserve University Bearing Dada[EB/OL].[2011-12-10].http://csegroups.case.ed./bearingdatacenter/pages/downloaddata-file.