基于MUDW和峭度的齿轮故障信号预处理方法

摘要

针对齿轮故障信号特征提取困难、易淹没在噪声干扰中等问题,采用形态非抽样小波(Morphological UnDecimated Wavelet, MUDW)分解和峭度对振动信号进行预处理,以强化齿轮故障信号特征和提高特征信息比重.首先,采用MUDW对信号进行分解,利用网格搜索法优选其初始参数;然后,采用峭度作为评价指标来表征各分解层近似信号对故障特征的贡献量,在此基础上进行加权融合运算,以提高有用的近似信号比重;最后,利用仿真信号和实测的齿轮故障振动信号验证了该方法的有效性和实用性.%Aiming at the problems that the gear fault signal feature extraction is difficult, and it is easily lost in the noise interference, Morphological Un-Decimated Wavelet (MUDW) decomposition and kurtosis are adopt to pretreat the vibration signal, so as to strengthen the gear fault signal characteristics and raise the proportion of feature information. Firstly, MUDW is adopted to decompose the signal, and the optimal parameters are selected using the grid search method. Secondly, kurtosis is used as evaluating index to characterize the approximate signal s contribution in decomposition layers to the fault characteristics, on that basis the weighted fusion operation is carried out to raise the proportion of useful approximation signals. Finally, the effectiveness and practicability of this method is verified by the simulation signal and the measured gear failure vibration signal.