基于CEEMDAN和双重峭度准则的电动机轴承故障特征频率估计方法

A Fault Characteristic Frequency Estimation Method of Motor Bearing Based on CEEMDAN and Double Kurtosis Criterion

振动传感器采集的轴承故障信号极易被强噪声污染,导致故障特征频率估计精度恶化。针对该问题,提出一种基于自适应噪声完备集合经验模态分解(CEEMDAN)和双峭度准则的轴承故障特征频率高精度估计方法。使用CEEMDAN完成振动信号分解之后,从众多的备选模态中挑选出合适成分重构故障特征信号极具挑战。对信号分解获得的模态分量进行迷向圆变换(标准白化处理)后,噪声对应的模态分量的分布更接近于正态分布。借助该信息,引入双重峭度准则,第一重峭度判定是在原始模态分量中筛选出疑似的故障信号,第二重峭度判定是借助迷向圆变换剔除掉噪声成分,然后再使用这两重判定的交集模态完成对轴承故障特征信号的重构。在此基础上,采用复包络法和FFT变换获取信号的包络谱,然后使用三线谱校正法更为精准地估计轴承故障特征频率。仿真实验和实测外圈、内圈故障数据表明,与一些现有方法相比,所提方法具有信号筛选方法简便、估计精度高等优势。

Bearing fault signal collected by vibration sensor is easily polluted by strong noise,which degrades the accuracy of the fault characteristic frequency estimation.For this problem,a high accurate method of the bearing fault frequency estimation based on complete ensemble empirical mode decomposition with adaptive noise(CEEMDAN)and double kurtosis criterion is proposed.It is very challenging to select the appropriate components from alternative modes to reconstruct the fault characteristic signal after the signal decomposition.The distribution of the modal components corresponding to noise is closer to the normal distribution after the isotropic circle transformation(standard whitening treatment).Consequently,two kurtosis criterions are introduced.The first kurtosis criterion is to screen out suspected fault signals from the original modal components,and the second kurtosis criterion is to remove noise components by the isotropic circle transformation.Hence,the intersection modes of the double kurtosis criterion are used to reconstruct the bearing fault characteristic signal.Furthermore,the complex envelope method and FFT are used to obtain the envelope spectrum of the signal.In addition,the three-line spectrum correction method is employed to achieve more accurate fault characteristic frequency estimation.Compared with some existing methods,simulation experiments and measured bearing fault data of outer and inner rings show that the proposed method can obtain high accuracy estimates by a simple signal screening method.