波形曲率延拓在局域均值分解中的应用

局域均值分解（LMD）是一种能够将复杂的调幅调频信号自适应地分解为一系列单分量的调幅调频信号的处理方法，其分解过程存在端点效应，分解结果有一定程度的失真。针对此问题，提出根据波形曲率特征对信号端点进行极值延拓，通过特征波的曲率波动来筛选与边界波形最为相似的数据段，在此基础上将波形匹配曲率估计应用于LMD分解过程中, 并与镜像延拓及自适应波形匹配延拓方法相比较，验证了所提方法的优点。使用仿真信号与实际的齿轮故障数据进行试验与检测，结果表明，所提方法可以有效改善LMD分解过程的端点效应，提高分解精度。

Waveform Curvature Extension in Applications of Local Mean Decomposition

LMD was a kind of signal processing method to self-adaptively decompose the complex amplitude modulation-frequency modulation signals into a series of single-component amplitude modulation-frequency modulation. To some degrees, the decomposition results were of distortions because of the influences of end effects. In order to improve the end effects from LMD, the data segments which were most similar to wave boundary data segments were searched according to the curvature fluctuations of characteristic waves. The waveform curvature was applied in LMD. According to the comparison of mirror extension and self-adaptive waveform match, the advantages of waveform curvature extension were verified. The experiments of simulation signals and actual gear failure were made, the results show that this method can suppress the end effect effectively and improve the decomposition precision.