计算机应用 ›› 2016, Vol. 36 ›› Issue (10): 2933-2939.DOI: 10.11772/j.issn.1001-9081.2016.10.2933

• 行业与领域应用 • 上一篇    下一篇

融合极点对称模态分解与时频分析的单通道振动信号盲分离方法

叶卫东1,2, 杨涛1,2   

  1. 1. 西南科技大学 信息工程学院, 四川 绵阳 621010;
    2. 特殊环境机器人技术四川省重点实验室(西南科技大学), 四川 绵阳 621010
  • 收稿日期:2016-03-30 修回日期:2016-06-26 发布日期:2016-10-10
  • 通讯作者: 杨涛,E-mail:yangtao@swust.edu.cn
  • 作者简介:叶卫东(1989—),男,湖北荆门人,硕士研究生,主要研究方向:振动信号处理、机械故障诊断;杨涛(1972—),男,四川绵阳人,教授,博士,主要研究方向:机电系统、声学阵列信号处理。
  • 基金资助:
    国家自然科学基金资助项目(F011102);特殊环境机器人技术四川省重点实验室开放基金资助项目(13zxtk06);四川中烟工业有限责任公司公司科技项目(川渝烟工技研[2015]62号)。

Single-channel vibration signal blind source separation by combining extreme-point symmetric mode decomposition with time-frequency analysis

YE Weidong1,2, YANG Tao1,2   

  1. 1. School of Information Engineering, Southwest University of Science and Technology, Mianyang Sichuan 621010, China;
    2. Key Laboratory of Robot Technology Used for Special Environment of Sichuan Province (Southwest University of Science and Technology), Mianyang Sichuan 621010, China
  • Received:2016-03-30 Revised:2016-06-26 Published:2016-10-10
  • Supported by:
    BackgroundThis work is partially supported by the National Natural Science Foundation of China (F011102), the Open Foundation of Key Laboratory of Robot Technology Used for Special Environment of Sichuan Province (13zxtk06), the Research Project of Sichuan Tobacco Industry Company Limited (Chuanyu Tobacco Industry Technology Research[2015] 62).

摘要: 针对单通道振动信号盲源分离的观察信号少于源信号,且传统的盲源分离方法往往忽视信号非平稳性的问题,提出一种基于极点对称模态分解和时频分析的盲分离算法(ESMD-TFA-BSS)。首先,采用极点对称模态分解方法将观察信号分解成不同的模态,采用贝叶斯信息准则(BIC)估计源信号个数并利用相关系数法选取最优观察信号,由原观察信号与最优观察信号组成新的观察信号;其次,根据新的观察信号计算白化矩阵并将其白化,利用平滑伪Wigner-Ville分布将白化后的信号拓展到时频域,采用矩阵联合对角化方法计算酉矩阵;最后,根据白化矩阵和酉矩阵估计源信号。在盲源分离仿真实验中,ESMD-TFA-BSS的估计源信号与仿真信号的相关系数分别为0.9771、0.9784、0.9660,基于经验模态分解和时频分析的盲分离算法(EMD-TFA-BSS)的相关系数分别为0.8697、0.9706、0.8548,ESMD-TFA-BSS比EMD-TFA-BSS的相关系数分别提高了12.35%、0.80%、13.00%。实验结果表明,ESMD-TFA-BSS在实际工程中能够有效地提高源信号分离精度。

关键词: 振动信号, 盲源分离, 极点对称模态分解, 时频分析, 故障诊断

Abstract: As the number of the observed signals for single-channel vibration signal-separation is less than that of the source signals, and in traditional methods, the Blind Source Separation (BSS) of vibration signals commonly ignores the non-stationarity, a BSS algorithm based on Extreme-point Symmetric Mode Decomposition (ESMD) and Time-Frequency Analysis (ESMD-TFA-BSS) was proposed. Firstly, the single observed signal was decomposed into different modes by ESMD method, and the number of source signals was estimated by Bayesian Information Criterion (BIC) and the optimal observed signals were selected by using correlation coefficient method. The original observed signals and the optimal observed signals were used to construct the new observed signals. Secondly, the whitening matrix and the whitened signals were obtained based on the new observed signals, and the whitened signals were extended to the time-frequency domain by using smoothed pseudo Wigner-Ville distribution, then the unitary matrix was calculated by utilizing matrix joint diagonalization. Finally, the source signals were estimated by the whitening matrix and the unitary matrix. In the BSS simulation experiments, the similarity coefficients between the estimated signals with ESMD-TFA-BSS and the source signals were 0.9771, 0.9784 and 0.9660, and the similarity coefficients with the BSS algorithm based on Empirical Mode Decomposition and Time-Frequency Analysis (EMD-TFA-BSS) were 0.8697, 0.9706 and 0.8548. Compared with EMD-TFA-BSS, the similarity coefficients with ESMD-TFA-BSS was increased by 12.53%, 0.08% and 13.00%. Experimental results indicate that ESMD-TFA-BSS can effectively improve separation accuracy of source signals in practical application.

Key words: vibration signal, Blind Source Separation (BSS), Extreme-point Symmetric Mode Decomposition (ESMD), Time-Frequency Analysis (TFA), fault diagnosis

中图分类号: