【作者】 周莹；

【导师】 程卫东；

【作者基本信息】 北京交通大学 ， 机械制造及其自动化， 2011， 硕士

【摘要】 滚动轴承是旋转机械中比较常见的而且易损坏的部件,轴承的好坏对机器的工作状况影响很大。它的运行状态直接影响整台机器的性能,一旦发生故障就成为引起机械设备失效的重要原因,据统计旋转机械的故障有30%是由轴承引起的。因此,对滚动轴承的状态监测和故障诊断是机械设备故障诊断的重要研究内容。在设备故障诊断中,构成故障的特征空间是非常复杂的非线性关系,很难建立数学模型。人工神经网络能够映射任意复杂的非线性关系,具有自学习、自组织和自适应等特征,被广泛地应用于机械故障诊断中。本文提出了MIV算法,时/频域分析法和BP神经网络算法对滚动轴承做了故障诊断。主要研究了以下内容：首先,介绍了滚动轴承的结构特征以及产生的故障原因,以及人工神经网络的基本知识,选取BP神经网络做滚动轴承故障诊断的原因。设计了滚动轴承的信号采集系统,采集并处理了正常轴承、内圈故障轴承、外圈故障轴承、滚动体故障轴承的振动数据,作为基于BP神经网络的滚动轴承故障诊断实验仿真的数据样本。其次,结合BP神经网络和MIV算法对常用的时域特征参数和频域特征参数做了特征筛选,根据仿真结果即MIV值的大小,得到8个敏感故障特征参量,即他们更能反映故障特征。之后选用时域、频域分析法,分别分析这8个特征参量的对故障特征的敏感程度,得到方差、均方根值、峰值、裕度因子、总功率谱和以及峭度系数对故障最为敏感,并确定为BP神经网络的输入。再次,设计了对滚动轴承做故障诊断的BP神经网络的结构。确定了BP神经网络的结构参数,分析了BP神经网络的结构参数的设置对网络性能的影响。主要包括：激活函数、学习算法、数据预处理方法、网络初始权值以及期望误差的选取,输入神经元、隐层节点数、网络学习率和训练系数对网络性能的影响情况。最后,利用前文的实测的滚动轴承的数据,以及设计好的BP神经网络故障诊断系统,对滚动轴承做了故障诊断仿真实验。确定了训练样本和预测样本,完成了对轴承的故障诊断和分类,并求取了分类误差率。仿真研究结果表明了诊断系统的有效性,验证了BP神经网络对滚动轴承做故障诊断的可行性,诊断效果良好。更多还原

【Abstract】 Rolling bearing is a very common and easily damaged part in rotating machinery. The quality of the bearing impacts badly the working conditions of machine. It directly affected the operational status of the performance of the whole machine, and its failure becomes a major cause of the failure of the whole machinery, in statistics,30% of machines’ failure is caused by bearings. Therefore, bearing condition monitoring and fault diagnosis of machinery fault diagnosis is an important research. In fault diagnosis, Fault space is very complicated, and difficult to build mathematical models. But the artificial neural network can reflect the complexity of the linear relationship. Artificial neural network has self-learning, self-organizing, adaptive features, is widely applied to machinery fault diagnosis. In this paper,In this paper, the MIV method, time/frequency domain analysis method and BP neural network algorithm was maked the rolling bearing fault diagnosis. The following main elements:First, I introduce the bearing structure and the cause of the malfunction, and the basic knowledge of artificial neural networks. I designed a data acquisition system, And collecting the normal bearing, failure bearing inner ring, outer ring fault bearings, rolling element bearing fault vibration data.Secondly, I selected the fault parameters, as the neural network input. Including the variance, RMS, peak, margin factor, and the total power spectrum and the kurtosis coefficientAgain, I designed a roller bearing fault diagnosis system, analysis of the coefficient on the impact of network performance.Finally, I complete the simulation of fault diagnosis. Strike the classification error rate.更多还原