基于EMD和小波包的轴承故障特征提取方法研究.pdf

摘 要
论文首先分析了经典信号处理方法的特点和局限，研究了当今处理非平稳信号的热
点——小波变换和希尔伯特-黄变换，并对这两种变换的基本理论和算法做了深入研究。
其次，本文重点研究了经验模态分解（EMD）和小波变换结合的方法，指出了这种
结合大多是建立在利用小波良好的降噪性能对信号做预处理，但实际信号非常复杂且伴
有噪声影响，因而降噪模型是难以得到的，这使得特征参数的提取非常困难。在对 EMD
的理论研究后发现，其分解过程不仅是自适应地选取基函数而且具有一定的自适应滤波
功能。因此，本文提出了将 EMD 的自适应性和小波包分解结合的方法（文中简称为
EMD_WP）来处理非平稳信号，该方法是在 EMD 对信号分解后用小波包对 IMF 做进一
步分析，通过仿真试验说明，该方法在有噪声干扰的情况下，仍可以提取出较精确的特
征频率。试验中还验证了在噪声强度比较大时，引用二代小波对信号做降噪处理后
EMD_WP 的效果会得到改善。
最后，本文以非平稳的轴承故障信号为研究对象，叙述了轴承故障产生的机理，并
用 EMD_WP 分析了滚动轴承故障数据，结果表明该方法提取的故障特征明显，可以实
现对滚动轴承故障的识别。与文献中已有方法相比，EMD_WP 可以提高电机振信号非
平稳特征参数的提取精度，能更好地解决实际问题，具有一定的实用价值。


关键词：希尔伯特-黄变换；经验模态分解；小波变换；小波包变换；特征提取；
包络谱；内圈故障
Abstract
First, after analyzing the characteristics and limitations of classic signal processing, the
paper introduces today's research focus——Hilbert-Huang transform(HHT) and Wavelet
transform(WT), which are the main method of non-stationary signals processing, and deeply
studied the basic theory and algorithms.
Second, this paper is focused on combining empirical mode decomposition and wavelet
and point out that most of this combination is used a good reduction noise of wavelet to do
pre-processing. Since the actual motor vibration signals are very complex and mix some noise,
it is necessary to find a noise reduction method. However, it is difficult to get unified noise
reduction model, and so it is very difficult to extract fault feature. After studying the theoreti-
cal of EMD found that the decomposition process is not only self-adaptive but also adaptive
filtering. Therefore, the paper combines adaptive filtering of EMD and wavelet packet decom-
position(the paper is known as EMD_WP) to deal with non-stationary signal. EMD_WP is
further analyzing the IMFs decomposed by wavelet packet. Simulation results show that in a
case of noise, the method can st