基于鲁棒预白化的机械噪声信号盲分离.pdfVIP

第 24卷 第 12期 机 械 科 学 与 技 术 Vol. 24 No. 12 2005年 12月 M ECHANICAL SCIENCE AND TECHNOLOGY December 2005 : 10038 28( 2005) 李加文, 李从心 ( 上海交通大学 国家 具 CAD工程中心, 上海 200030) 李加文 : 机械噪声故障诊断的难度在于实际检测的噪声是多个设备或零部件噪声 号的混合, 噪比低, 基于二阶 统计量盲源分离算法的故障噪声诊断技术, 利用二阶协方差矩阵的联合对角化, 从测量噪声中分离出感兴趣故障噪 声进而提取特征, 但该算法抗干扰噪声性能差。本文利用多个协方差矩阵平滑滤波后的矩阵进行白化, 进一步提高 了抗干扰噪声能力, 在样本数据较少时仍能实现较好的盲源分离效果, 仿真实验证实了该算法的有效性。 : 盲源分离; 故障诊断; 号采集; 特征提取 : TB53 : A R obust Prewhiteningbased Blind Separation ofM achine N oise L i Jiaw en, Li Congxin ( NationalD ie Mou ld CAD Engineering Research Center, Shanghai Jiaotong Un iversity, Shanghai 200030) Abstract: M ach ine noise fau lt diagnosis d ifficu lties lie in the fact that the detected interested signal is m ixed by many other noiseswh ich lead to low SNR. By using secondorder covariancematrices joint diag onalization, blind source separation d iagnosis technology based on secondorder statistics can extract fea tures of interesting failure noise from detected signals in h igh SNR, but has a bad performance in low SNR. In this paper, the average eigenstructure of d ifferent tmi e delayscovariancematrices is investiga ted to compute singular value decomposition whenw hiten ing signals, w hich can mi prove the antijamm ing capab ility even in case of small sample data. Smi u lation tests show that the approach is feasible for blind source separation of distorted rotating machine sources. Key w ords: blind source separation; fau lt diagnosis; f