基于多特征量提取和PSO优化神经网络的轴承故障诊断

发布时间：2018-02-26 07:11

本文关键词： 滚动轴承故障诊断局部均值分解(LMD) 提升小波粒子群优化BP网络　出处：《燕山大学》2016年硕士论文　论文类型：学位论文

【摘要】：滚动轴承作为机械设备中使用非常广泛的基础部件,其运行状况的良好与否与整个系统的功能实现和安全生产有着直接的关系。工业技术如今发展飞速,人们意识到滚动轴承的稳定高效运行非常重要。论文以滚动轴承振动信号为基础,研究了滚动轴承故障诊断方法并做仿真及实验验证。首先,介绍了滚动轴承的构造、分类、振动原理以及常见的轴承故障诊断分析方法。针对以往轴承故障诊断技术的局限性,提出基于多特征量提取及粒子群优化神经网络的滚动轴承故障诊断方法。其次,引入了局部均值分解算法(Local Mean Decomposition,LMD)对振动信号分解。LMD分析方法能够有效地处理复杂信号,显示信号的局部特征。但是,由于实际信号常常会包含有大量的噪声,噪声会对局部均值分解算法的精度和效果产生严重影响。提升小波结构简单,构造方法灵活,因此本文引入提升小波消噪方法,对信号进行提升小波消噪后再进行LMD分解。通过一系列的仿真实验验证了该方法的有效性。然后,提取故障信号的时域特征指标、样本熵以及LMD分解后PF分量的能量,共同组成特征向量。利用时域特征参数可以对轴承故障进行分类,样本熵可以用来反映信号的复杂程度,LMD分解后各个PF分量的能量可以体现信号较深层次的信息。时域特征指标、样本熵、能量特征指标分别从系统的不同特性出发来描述系统的内部信息。这些特征量共同输入神经网络分类模型,可以使信息互补,弥补单一特征量的信息缺失。在对轴承故障的状态识别时,针对BP神经网络的收敛性问题,引入粒子群对BP网络优化。粒子群优化算法(Particle Swarm Optimizer,PSO)是一种新的基于群体智能的优化算法,有极强的全局搜索能力。PSO算法参数中,针对其重要的参数惯性权重,本文采用随机惯性权重法结合压缩因子的混合粒子群算法。对于BP神经网络的权值和阈值,利用混合PSO算法进行优化,有助于BP网络快速收敛到全局最优点。最后,将美国凯斯西储大学滚动轴承故障数据作为研究对象,分别对滚动轴承不同类型和不同损伤程度两个方面进行实验分析,结果表明,基于多特征量提取和粒子群优化神经网络的轴承故障诊断方法能够取得良好的效果。
[Abstract]:As a widely used basic component in mechanical equipment, rolling bearing has a direct relationship with the function realization and safe production of the whole system. People realize that the stable and efficient operation of rolling bearing is very important. Based on the vibration signal of rolling bearing, the fault diagnosis method of rolling bearing is studied, and the simulation and experimental verification are done. Firstly, the structure and classification of rolling bearing are introduced. Vibration principle and common bearing fault diagnosis and analysis method. In view of the limitation of the previous bearing fault diagnosis technology, a rolling bearing fault diagnosis method based on multi-feature extraction and particle swarm optimization neural network is proposed. Secondly, The local mean decomposition algorithm (LMD) is introduced to decompose the vibration signal. LMD analysis method can deal with the complex signal effectively and display the local characteristics of the signal. However, the actual signal usually contains a lot of noise. Noise will have a serious impact on the accuracy and effect of the local mean decomposition algorithm. Lifting wavelet is simple in structure and flexible in construction, so the lifting wavelet denoising method is introduced in this paper. The signal is de-noised by lifting wavelet and then decomposed by LMD. The validity of the method is verified by a series of simulation experiments. Then, the time-domain characteristic index of fault signal, sample entropy and the energy of PF component after LMD decomposition are extracted. Using time domain characteristic parameters to classify bearing faults, The sample entropy can be used to reflect the complexity of the signal and the energy of each PF component after LMD decomposition can reflect the deeper information of the signal. The energy characteristic indexes describe the internal information of the system according to the different characteristics of the system. These features are input into the neural network classification model together, which can make the information complement each other. In order to solve the problem of the convergence of BP neural network, when identifying the state of bearing fault, Particle swarm optimization algorithm (PSO) is a new optimization algorithm based on swarm intelligence. In this paper, a hybrid particle swarm optimization algorithm based on random inertial weight method and compression factor is used. For the weight and threshold of BP neural network, the hybrid PSO algorithm is used to optimize the BP neural network, which helps the BP network converge quickly to the global optimum. Taking the rolling bearing fault data of case Western Reserve University as the research object, the experimental analysis of different types and different damage degree of rolling bearing is carried out. The results show that, The bearing fault diagnosis method based on multi-feature extraction and particle swarm optimization neural network can achieve good results.
【学位授予单位】：燕山大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TH133.3

【相似文献】