期刊
MECHANICAL SYSTEMS AND SIGNAL PROCESSING
卷 135, 期 -, 页码 -出版社
ACADEMIC PRESS LTD- ELSEVIER SCIENCE LTD
DOI: 10.1016/j.ymssp.2019.106443
关键词
Empirical mode decomposition; Manifold learning; Machinery fault diagnosis; Noise assistance; Time-frequency signal decomposition
资金
- National Natural Science Foundation of China [51805342, 51875375, 51875376]
- National Program for Support of Top-Notch Young Professionals
- Natural Science Foundation of Jiangsu Province [BK20180842]
- China Postdoctoral Science Foundation [2018M640514]
- Jiangsu Planned Projects for Postdoctoral Research Funds [2018K006B, 2018K004C]
- Joint Fund of Equipment Pre-Research and Ministry of Education [6141A02022141]
- Natural Science Fund for Colleges and Universities in Jiangsu Province [18KJB470022]
- Suzhou Prospective Research Program [SYG201912]
One challenge of the existing noise-assisted methods for solution of mode mixing problem of empirical mode decomposition (EMD) is that, the decomposed modes contain much residual noise derived from both added and self-contained noise. This paper proposes a new noise-assisted method, called EMD manifold (EMDM), for enhanced fault diagnosis of rotating machines. The major contribution is that the new method nonlinearly and adaptively fuses the fault-related modes containing different noise via a manifold learning algorithm, by which true fault-related transients are preserved, while fault-unrelated components including mode-mixing-induced components and the residual noise derived from both the added and self-contained noise are greatly suppressed. First, the sensitive mode is located among the modes obtained by the EMD method according to a newly proposed criterion. Then, a high-dimensional matrix is constructed of the sensitive modes obtained through a small number of EMD trials on the signals plus noise of different amplitudes. Finally, the manifold learning algorithm is performed on the high-dimensional matrix to extract intrinsic manifold of the fault-related transients. The high-dimensional matrix is repeatedly constructed with random noise added to adjust local data distribution of the matrix for adaptive EMDM feature learning. Experimental studies on gearbox and bearing faults are conducted to validate the proposed method and its enhanced performance over traditional noise-assisted EMD methods. (C) 2019 Elsevier Ltd. All rights reserved.
作者
我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。
推荐
暂无数据