基于延拓与可变余弦窗的经验模态分解改进算法研究

针对经验模态分解(Empirical Mode Decomposition，EMD)中存在的边界效应及边界发散现象随着筛选层次的增加而增加的问题，提出一种利用延拓与可变余弦窗相结合的改进新方法。首先对信号进行延拓处理，增加一定长度的数据，实现延拓数据与原始信号交界处的光滑过度。其次，根据信号边界的发散程度，在逐层提取各阶本征模函数（Intrinsic Model Function，IMF）之前，在信号两端加上宽度可变的余弦窗函数，使得每一个IMF分量边界发散问题最小化，保证信号有效数据的正确分解，实现EMD边界处理算法的改进。仿真和实例信号分析表明，该方法能较好地抑制EMD边界效应，有效地提取故障信号中的特征信息。

Improved algorithm against end effect of EMD based on extension and variable cosine window

For the end effect of empirical mode decomposition (EMD) in pace with the increase of the filtering level, a new improved method using a combination of extension with variable cosine window is proposed. Firstly, the signal is extended with a certain length of data. Then the extended signal is processed at both ends with variable cosine window before extracting the every Intrinsic Model Function（IMF）, which makes the end effect of each IMF minimized and ensures the correct decomposition and the improvement of the EMD algorithm. Simulation analysis and crack fault diagnosis example show that the improved method can inhibit end effect effectively.