经验模态分解中的模态混叠问题

Mode Mixing in Empirical Mode Decomposition

针对经验模态分解(empirical model decomposition,简称EMD)存在的模态混叠问题,总结了引起模态混叠异常事件的类型,讨论了模态混叠的产生原因,提出了采用加入高频谐波后再进行EMD分解消除模态混叠的方法。根据信号分析频率范围和特征选择高频简谐波的频率和幅值,并使高频谐波作为第1阶IMF分解出来,可以有效消除模态混叠现象,异常事件通常可以包含在第1阶IMF中,必要时可以将加入的高频信号直接减掉,不影响对EMD结果的判断。与总体平均经验模态分解法(ensemble empirical model decomposition,简称EEMD)对比的仿真计算表明,两种方法都可以有效消除模态混叠现象,但高频谐波加入法具有运算速度快、误差小、分解结果物理意义明确和不需后处理的优点,对含复杂异常事件的实际故障信号分析验证了该方法在工程应用中的有效性和可行性。

Mode mixing is an inevitable problem in empirical mode decomposition(EMD),the ingredients which cause mode mixing,such as intermittency,pulse and noise,are defined as abnormal events in the paper.A novel method using high frequency harmonic before EMD is proposed.The amplitude and frequency of the added harmonic are determined according to frequency range and characteristic of the original signal,and the added harmonic will be decomposed as the first IMF,then mode mixing can be alleviated.Commonly the first IMF containes the added harmonic and the abnormal events,the harmonic can be subtracted to get the true first mode if necessary.Compared with EEMD,simulation results show that both the two methods can avoid the mode mixing correctly,but the proposed method allows small errors,fast operation,tangible physical meaning and no post-process.An actual fault signal with complex abnormal events is also analyzed,and the effectiveness and feasibility in engineering are verified.