形态滤波与平移不变量小波增强EEMD的故障诊断方法
A Fault Diagnosis Method using Morphological Filtering-translation Invariant Wavelet and Improved Ensemble Empirical Mode Decomposition

针对集成经验模式分解（Ensemble empirical mode decomposition，EEMD）在轴承故障特征提取中的问题，提出一种混合故障诊断方法。首先，将"形态滤波-平移不变量小波"作为EEMD的前置滤波器，实现对原始信号中窄带脉冲和随机噪声干扰的有效消除；其次，针对本征模式分量（Intrinsic mode functions，IMFs）中真实分量的选取问题，提出一种轴承振动信号EEMD分解的筛选规则，即计算各IMFs和原信号的自相关函数并作归一化处理，然后计算各IMFs自相关函数和原信号自相关函数的相关系数，以最大相关系数的一半作为阈值剔除虚假的IMFs，与此同时保留第1和第2阶IMFs，从而实现对EEMD的改进。仿真和实验轴承故障诊断研究表明了该方法的有效性，方法的优点在于：将"形态滤波-平移不变量小波"作为集成经验模式分解的前置滤波器，可有效去除故障轴承振动信号中的窄带脉冲和随机噪声干扰；本文的筛选规则可有效选取去噪信号EEMD分解后的IMFs中真实分量，从而可靠地获取故障特征频率。本研究为轴承故障诊断提供了一种新的手段。
In order to solve the problem in the usage of the ensemble empirical mode decomposition (EEMD) to extract fault feature in rolling element bearing, a hybrid fault diagnosis method is proposed. First, the morphological filter combining with translation invariant wavelet are served as the pre-filter of the EEMD to effectively eliminate the narrowband pulse and random noises from original signals. Second, aiming to the reasonable selection of the true components from intrinsic mode functions (IMFs), a screening rule is proposed to improve EEMD by selecting IMFs from the vibration signals of rolling element bearings. The autocorrelation functions of both the original signal and each IMF are calculated and normalized. All correlation coefficients are obtained by calculating the autocorrelation functions of the original signal with the autocorrelation functions of each IMF. To eliminate the false IMFs, a hard threshold is defined using the half of the maximum correlation coefficient. Therefore, inspect whether the first two IMFs are reserved, if not, reserving the first two IMFs simultaneously. Simulation and experimental investigations for the fault detection of rolling bearing show that the present hybrid method is effective. The advantages of the present method are:the usage of the pre-filter of the hybrid morphological filter and translation invariant wavelet is efficient to eliminate the interference of the narrowband pulse and the random noises from the rolling element bearings with faults; the real IMFs will be effectively selected form the de-noised signal using EEMD decomposition, and finally the feature frequency will be reliably obtained. This study provides a new way to detect faults of rolling element bearings