摘要
针对机械振动信号特征提取中的去噪问题,联合集合经验模式分解(EEMD)和最小均方算法(LMS)发展了一种自适应去噪方法。首先研究了LMS的固定步长固定阶数、变步长(VS)和变阶数(VT)的算法性能,提出在迭代过程中以比较阶数和步长变化时的最小均方误差期望为收敛方向,发展了一种联合变步长变阶数最小均方算法(VSVTLMS)的去噪方法;通过对原信号的EEMD分解,使各模式分量窄带化,进而通过VSVT-LMS对每个IMF分量进行去噪,有效避免LMS算法对宽频信号的不稳定性,同时也避免了EMD分解的不唯一性和去噪中阈值的选择问题。最后通过对仿真和实际车辆振动信号去噪,验证了方法在工程上的可行性。
To reduce noise in mechanical vibration signals, a new adaptive denoising method combining ensemble empirical mode decomposition （EEMD） and least mean square algorithm （LMS） was proposed here. Firstly, the algorithm performance of LMS with a fixed step and a fixed filter order and that with variable steps and variable orders were studied, the expectation of least mean square error due to order or step changing in iterative process was taken as the convergence direction, a kind of variable step and variable order LMS （VSVO-LMS） algorithm was developed. Using EEMD, the original signal was decomposed so that each mode component was narrowband, and then through VSVO-LMS each IMF was denoised, the algorithm instability of LMS for wideband signals algorithm, the EMD threshold selection problem was also avoided effectively. Simulations and vehicle actual signals were analyzed and it was shown that the new method has a good adaptive feature and a better accuracy, and the feasibility of the method is verified.
出处
《振动与冲击》
EI
CSCD
北大核心
2013年第20期61-66,共6页
Journal of Vibration and Shock
基金
国家自然科学基金资助项目(50675099)
中央高校基本科研业务费专项资金资助(CX10B_094Z)
关键词
振动信号
集合经验模式分解
自适应滤波器
变阶数
变步长
vibration signal
ensemble empirical mode decomposition （EEMD）
adaptive filter
variable order
variable step
