改进经验模态分解在动平衡信号提取中的应用

在变频结构干扰和强噪声背景下，传统方法从原始振动信号中提取动平衡信号的精度不高。本文采用经验模态分解可以根据实时振动信号的局部特征时间尺度，将其自适应分解为有限多个由高频到低频排列的、正交的本征模态函数；同时利用自回归预测模型延拓信号端点，以消除分解过程的边界效应对低频动平衡信号的影响；最后，根据功率谱密度可以快速、有效地判断出代表基频信号的本征模态函数。实验结果证明，该方法可以高精度提取动平衡信号，在相同测量条件下，能够获得较高的一次不平衡量降低率和较好的重复性能。

Extracting dynamic balancing signal based on improved empirical mode decomposition

In strong background noise and variable-frequency disturbance, dynamic balancing signal can not be effectively extracted from the original vibration signal with high accuracy using conventional methods. The empirical mode decomposition method is employed to extract dynamic balancing signal from such background condition. According to the local characteristic time scale of real-time signal, the original vibration signal can be effectively and adaptively decomposed into a finite number of orthogonal intrinsic mode functions that are arranged in the order from high frequency to low frequency. The auto-regression prediction model is introduced to extend both endpoints of the signal, which eliminates the influence of end effects on low frequency dynamic balancing signal. Finally, the power spectral density is adopted to identify the fundamental signal from all intrinsic mode function components easily. The experimental results verify the effectiveness of the presented method.