基于LMD方法的转子系统故障诊断研究

发布时间：2018-05-13 09:04

本文选题：故障诊断 + 转子系统　；参考：《湖南大学》2011年硕士论文

【摘要】：转子系统的故障诊断过程包括诊断信息的获取、故障特征信息提取和状态识别三部分。其中,故障特征提取是诊断的关键。本文将时频分析的新方法——局部均值分解法(Local Mean Decomposition,简称LMD)应用于转子系统的故障诊断中。该方法的特点是可以获得一系列瞬时频率具有物理意义的PF(Product function,简称PF)分量。本文对LMD方法在转子系统故障诊断中的应用进行了研究。主要研究工作如下: 1.针对转子系统故障振动信号的非平稳特性,提出了一种基于LMD和神经网络相结合的故障诊断方法。该方法首先对信号进行LMD分解,将其分解为若干个PF分量之和,再选取包含主要故障信息的PF分量做进一步分析,从这些分量中提取时域统计量和能量等特征参数作为神经网络的输入参数来识别转子系统的故障类别。结果表明,基于LMD与神经网络的故障诊断方法能够准确、有效地对转子系统的工作状态和故障类型进行分类。 2.提出了基于LMD和AR模型相结合的转子系统故障诊断方法。该方法先用LMD方法将转子振动信号分解成若干个瞬时频率具有物理意义的PF分量之和,然后对每一个PF分量建立AR模型,提取模型参数和残差方差作为故障特征向量,并以此作为神经网络分类器的输入来识别转子的工作状态和故障类型。与内禀模态函数分解法(Empirical Mode Decomposition,简称EMD方法)的对比研究表明,这两种方法均能有效地应用于转子系统的故障诊断。但LMD方法在信号分解方面体现了更大的优势。 3.针对LMD分解法的频率混淆问题,提出了基于改进的LMD和奇异值分解法相结合的转子系统故障诊断方法。该方法先用小波包分解法将转子振动信号分解成若干个小波包分量,进一步对各小波包分量进行LMD分解,得到一系列PF分量,形成初始特征向量矩阵。然后对初始特征向量矩阵进行奇异值分解得到矩阵的奇异值,将其作为特征向量输入神经网络来识别转子系统的工作状态和故障类型。实验结果表明,该方法能有效的用于转子系统故障诊断。 4.提出了基于改进的LMD和时频熵相结合的转子系统故障诊断方法,转子系统振动信号进行小波-LMD分解后,能量分布在具有不同时间尺度的PF分量上,转子系统的状态不同,能量在不同的PF分量上的分布是不一致的,表现为时频分布上的不同,基于改进LMD方法的时频熵就是上述时频分布的定量描述,通过实验数据分析可知,基于改进的LMD的时频熵对转子故障类别十分敏感,可用于转子系统故障诊断。
[Abstract]:The fault diagnosis process of rotor system includes three parts: obtaining diagnosis information, extracting fault feature information and identifying state. Fault feature extraction is the key to diagnosis. In this paper, the local mean decomposition method, a new time-frequency analysis method, is applied to the fault diagnosis of rotor system. The characteristic of this method is that a series of PF(Product function components with physical meaning can be obtained. In this paper, the application of LMD method in rotor system fault diagnosis is studied. The main work of the study is as follows: 1. A fault diagnosis method based on LMD and neural network is proposed for the non-stationary characteristics of rotor system fault vibration signal. The method firstly decomposes the signal into the sum of several PF components by LMD decomposition, and then selects the PF component which contains the main fault information for further analysis. The characteristic parameters such as time-domain statistics and energy are extracted from these components as input parameters of the neural network to identify the fault types of the rotor system. The results show that the fault diagnosis method based on LMD and neural network can classify the working state and fault types of rotor system accurately and effectively. 2. A fault diagnosis method for rotor system based on LMD and AR model is proposed. Firstly, the rotor vibration signal is decomposed into the sum of PF components with physical meaning by LMD method, then AR model is established for each PF component, and the model parameters and residual variance are extracted as fault eigenvector. It is used as input of neural network classifier to identify rotor working state and fault type. Compared with the intrinsic mode function decomposition (EMD) method, it is shown that the two methods can be effectively applied to the fault diagnosis of rotor systems. But the LMD method has more advantages in signal decomposition. 3. In order to solve the frequency confusion problem of LMD decomposition method, a fault diagnosis method for rotor system based on improved LMD and singular value decomposition (SVD) is proposed. In this method, the rotor vibration signal is decomposed into several wavelet packet components by wavelet packet decomposition method, and a series of PF components are obtained by LMD decomposition of each wavelet packet component, forming the initial eigenvector matrix. Then the singular value of the initial eigenvector matrix is obtained by singular value decomposition. The singular value of the matrix is input into the neural network as the eigenvector to identify the working state and fault type of the rotor system. Experimental results show that this method can be used effectively in rotor system fault diagnosis. 4. A fault diagnosis method for rotor system based on improved LMD and time-frequency entropy is proposed. After the vibration signal of rotor system is decomposed by wavelet LMD, the energy distribution on PF component with different time scale is different. The distribution of energy in different PF components is different and the time-frequency distribution is different. The time-frequency entropy based on the improved LMD method is the quantitative description of the time-frequency distribution. The time-frequency entropy based on improved LMD is very sensitive to rotor fault types and can be used in rotor system fault diagnosis.
【学位授予单位】：湖南大学
【学位级别】：硕士
【学位授予年份】：2011
【分类号】：TH165.3

【引证文献】