基于LMD和HMM的转子故障诊断方法

发布时间：2018-06-21 01:46

本文选题：故障诊断 + 转子系统　；参考：《兰州理工大学》2012年硕士论文

【摘要】：在故障诊断过程中，故障信号特征量提取和故障模式识别是整个过程的关键步骤。基于此，本文将一种新的时频分析方法局部均值分解（Local MeanDecomposition，简称LMD）和另一种基于统计学的模式识别技术隐Markov模型（hidden Markov model，简称HMM）应用于转子系统的故障诊断中。本文模拟了转子试验台常见故障，为提高特征提取的信息量，采用基于LMD的方法对故障信号进行特征提取，同时，使用了统计理论十分严谨的HMM方法进行故障模式识别。围绕试验台提取的多种故障信号数据，用以上理论以及算法，本文做出的主要工作和结论如下：（1）在转子试验台上分别模拟出四种典型转子故障振动信号，对故障信号的振动机理和过程进行了深入研究，并且进行了小波滤波消噪和频谱分析。实验证明，小波消噪后得到的信号比较平滑，频谱图的各频率成分十分明显，很适合转子系统的数据处理，可以为特征提取提供原始数据。（2）提出了一种基于LMD和近似熵的故障特征提取新方法，LMD的局域化特征方法可以自适应分解信号为多个有用的乘积函数（Product function，简称PF）和一个余量之和，再结合改进的快速近似熵计算方法计算出每个PF的近似熵作为故障特征值，综合所有故障特征值组合为特征向量集。实验结果表明，基于LMD和近似熵的故障特征向量集可以准确反映故障信号特征，并且适于输入至HMM分类器进行故障分类。（3）以构造最优分类器为目标，针对HMM找到全局最优点的概率相对较低的问题，提出了一种经粒子群算法（Particle Swarm Optimization，简称PSO）优化的HMM的模式分类器。该方法可以优化HMM训练模型，使之准确寻找全局最优概率。将齿轮箱故障数据提取的幅值作为特征向量输入到该分类器中进行数据分类，实验表明，收敛误差较小，并可以成功识别齿轮箱故障。（4）提出基于LMD近似熵和PSO优化的HMM诊断方法，对转子试验台上采集的故障信号进行故障特征提取和模式识别，该方法能自适应分解原始故障信号，，提取出故障特征后输入至分类器进行故障识别便可实现多故障识别与诊断。通过实验数据分析，并与经典HMM的对比实验，说明了该方法的有效性。
[Abstract]:In the process of fault diagnosis, fault signal feature extraction and fault pattern recognition are the key steps in the whole process. Based on this, a new time-frequency analysis method, Local mean decomposition (LMDM), and another statistically based pattern recognition technique, Hidden Markov Model (HMMMMM), are applied to fault diagnosis of rotor systems. In this paper, the common faults of the rotor test-bed are simulated. In order to improve the information of feature extraction, the LMD-based method is used to extract the fault signal. At the same time, the hmm method, which is very rigorous in statistical theory, is used for fault pattern recognition. According to the above theory and algorithm, the main work and conclusions of this paper are as follows: 1) four typical rotor fault vibration signals are simulated on the test bed. The vibration mechanism and process of fault signal are deeply studied, and wavelet filtering and spectrum analysis are carried out. Experimental results show that the signal obtained by wavelet de-noising is smooth, and the frequency components of the spectrum are very obvious, which is suitable for the data processing of rotor system. A new fault feature extraction method based on LMD and approximate entropy is proposed. The sum of function and a surplus, Then the approximate entropy of each PF is calculated as the fault eigenvalue and all the fault eigenvalues are combined into eigenvector sets by using the improved fast approximate entropy calculation method. The experimental results show that the fault eigenvector set based on LMD and approximate entropy can accurately reflect the fault signal features, and is suitable for fault classification with hmm classifier. Aiming at the problem of relatively low probability of finding global best for hmm, a pattern classifier for hmm optimized by particle swarm optimization (PSO) is proposed. This method can optimize hmm training model to find the global optimal probability accurately. The amplitude extracted from the gearbox fault data is input into the classifier as the eigenvector to classify the data. The experimental results show that the convergence error is small. The fault diagnosis method based on LMD approximation entropy and PSO optimization is proposed to extract fault features and recognize the fault signals collected on the rotor test-bed. This method can decompose the original fault signal adaptively, extract the fault feature and input it to the classifier for fault identification, and then realize the multi-fault identification and diagnosis. The effectiveness of the proposed method is demonstrated by analyzing the experimental data and comparing it with the classical hmm.
【学位授予单位】：兰州理工大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：TH165.3

【引证文献】