EMD马氏距离与SOM神经网络在故障诊断中的应用研究  

Application of EMD Mahalano-bis Distance and SOM Neural Network in Fault Diagnosis

作　　者：姚海妮 王珍[1] 邱立鹏[1] 陈建国[1] 杨铎[1] 

机构地区：[1]大连大学机械工程学院,辽宁大连116622

出　　处：《噪声与振动控制》2016年第1期138-141,162,共5页Noise and Vibration Control

基　　金：国家自然科学基金(51405153);辽宁省教育厅一般项目(L2012446)

摘　　要：为实现对微弱动态响应的准确辨识及故障状态的早期诊断,提出EMD马氏距离与SOM神经网络的故障诊断方法,该方法首先对原始振动信号进行粒子滤波,提高信噪比,然后对其进行EMD分解,并对分解后的各模式分量进行分析,获得相关特征值组成特征向量,并求原始信号特征向量,为了选取能代表信号特征的模式分量,求各模式分量与原信号特征向量的马氏距离,将最优模式分量输入训练好的SOM神经网络,对故障分类,以轴承诊断为应用实例结果表明该方法切实有效。To achieve accuracy identification of weak dynamic response and early fault diagnosis, a fault diagnosis method based on EMD Mahalano-Bis distance and SOM neural network was proposed. First of all, to improve the signal-to-noise ratio, particle filtering was conducted to the original signals of vibration. Then, the signals were decomposed by the EMD. Each intrinsic mode function was analyzed to obtain the eigenvectors which include their corresponding eigenvalues. And the eigenvectors of the original signals were found. In order to select the intrinsic mode function which can represent the signal characteristics, the Mahalano-Bis distance between the intrinsic mode functions and the eigenvectors of the original signals was calculated. The best intrinsic mode function was chosen and input to the well-trained self-organizing feature map (SOM) neural network. Finally, the faults were classified. The application examples of bearing fault diagnosis show the effectiveness of this method.

关 键 词：振动与波 粒子滤波 EMD 马氏距离 SOM神经网络 故障诊断 

分 类 号：TH133.3[机械工程—机械制造及自动化] TH165.3