基于MED-EEMD的滚动轴承微弱故障特征提取

针对滚动轴承在强噪声环境下故障信号微弱、故障特征难以提取等问题,提出了基于最小熵反褶积(minimum entropy deconvolution,MED)和总体平均经验模态分解(ensemble empirical mode decomposition,EEMD)两者相结合的方法来提取滚动轴承微弱故障特征。通过对仿真信号和风电齿轮箱的振动信号分析,结果表明:为了弥补在强背景噪声下EEMD对微弱信号特征提取的局限性,该文选取MED作为EEMD的前置滤波器,最后对敏感的本征模态函数进行循环自相关函数解调分析,得出了风电齿轮箱的故障来自于高速轴的微小弯曲和高速轴输出端#10轴承外圈点蚀。同时与EEMD进行对比分析,表明了这种方法对微弱故障特征提取有较好的适用性。该文为多故障共存并处于强背景噪声下的微弱特征提取提供了参考。

Weak fault diagnosis for rolling element bearing based on MED-EEMD

Abstract: Under the complex environment, rotating machinery such as wind turbine gearboxes has multi-gearing and multi-bearing. The dynamic responses of these components are complex and interfering with each other. It is usually difficult to diagnose their potential faults. Especially when multiple faults and strong background noise coexist, vibration signals excited by several faults are combined with each other nonlinearly and non-stationary, which makes the observed vibration signals rather complex and difficult to identify each fault by using traditional methods. Especially the rolling bearing's fault feature under strong background noise is very weak and usually overwhelmed by noise. In this paper, minimum entropy deconvolution (MED) and ensemble empirical mode decomposition(EEMD) were combined for rolling bearing's weak fault diagnosis. EEMD is a self-adaptive analysis method and can decompose a complicated signal into a series of intrinsic mode functions (IMFs) according to the signal's local characteristics. EEMD is more accurate and effective for diagnosing the faults of rotating machinery than MED, so it has been used in the fault feature extraction of rolling bearing. However, if the frequency components in the signal are too complex and the background noise is very strong, they will affect the decomposition result, therefore it is very important to improve the signal-to-noise ratio of the original signal. MED searches for an optimum set of filter coefficients that can recover the output signal with the maximum value of kurtosis, which is an indicator that reflects the peak of a signal, so MED technique aims to extract the fault impulses while minimizing the noise and therefore resulting in clear detection results even under high noise, which can remedy the shortage of EEMD. In this paper, MED and EEMD were combined for the simulation signal and the wind turbine gearbox. Firstly, MED was used as the pre-filter to refine the vibration signal, and the strong background noise of rolling bearing was decreased by the MED method. Secondly, the given signal above was processed by the EEMD, and the amplitude of the added white noise could be determined through a trial-and-error method, which could be implemented based on minimizing the problems of mode aliasing. At last the sensitive intrinsic mode functions (IMFs) were analyzed by cyclic autocorrelation function (CAF), which could be applied to separate out the modulators effectively, especially for the weak modulators due to fault effects that could not be detected by other conventional technologies. We analyzed the wind turbine gearbox by combining CAF with EEMD, when the amplitude of the added white noise was 0.4 and the number of the ensemble was 100, the components 28 and 302 Hz corresponded to twice rotational frequency of high speed shaft and fault frequency of outer ring of #10 bearing. The result showed that high-speed shaft of wind turbine gearbox was slightly bent and there was slight pit on the face of outer rings of #10 bearings. The analyzed results demonstrate that the proposed method is an effective approach in identifying weak fault feature under strong background noise of rotating machinery.