基于灰色ELM的滚动轴承故障预测

发布时间：2018-11-18 07:42

【摘要】：随着旋转机械设备趋于大型化、高效化、集成化,这对旋转机械安全可靠运行的要求也相应提高。滚动轴承是现代旋转机械设备中最具关键作用的组成部件之一,它的工作状态正常与否直接关系着机械设备乃至整个系统的运行状态,但由于工作条件的恶劣造成其寿命参差性很大和容易损伤的缺点,因此需要对滚动轴承实施有效的故障预测。本课题“基于灰色ELM的滚动轴承故障预测”对采用灰色理论方法和ELM神经网络进行滚动轴承故障预测中所涉及的理论、方法和关键技术进行了深入的研究,主要研究工作如下:通过系统分析滚动轴承故障预测主要方法及国内外在该领域的研究现状,指出快速、准确的轴承故障预测技术已成为研究重点。在三大类故障预测技术中,重点分析介绍了了本文采用的基于数据的故障预测技术。针对滚动轴承振动信号具有非平稳性和非线性的特点以及各种经典方法的不足之处,运用灰色理论来预测轴承故障发展。深入研究了灰色模型的建模机理和适用范围。针对传统灰色GM(1,1)模型的不足,探讨了一系列改进灰色模型。其中,灰色多变量预测模型MGM(1,n)模型从系统的角度对多个诊断指标进行统一描述,本文用滚动轴承实验予以验证。由于传统的神经网络梯度学习算法存在着训练时间长、过度拟合训练样本和易陷入局部最优等问题。本课题引入了 ELM神经网络,ELM(极限学习机)具有学习时间短、算法简单容易实现、良好的泛化性能和能避免陷入局部最优等优点,已经成功应用于函数拟合和预测等应用领域。介绍了极限学习机的相关基础理论,对前馈神经网络学习算法和极限学习机算法进行了研究。在实际环境中的滚动轴承的振动信号在的噪声背景下不易提取,兼有非线性和非平稳的特点。本文对滚动轴承的振动信号进行经验模态分解,经过理论和仿真分析,发现在非线性信号处理方面EEMD分解较EMD分解有更好的抗混叠效果;自相关函数降噪性能优越,在故障提取特征方面具有很大的优势。灰色模型能够预测发展序列,ELM神经网络具有高度的非线性映射特性,本文提出一种新的组合权系数的计算方法,将二者有机结合,建立灰色ELM组合预测模型,能够描述轴承兼具确定性和波动性的复杂趋势。将组合模型用于轴承故障预测,充分利用信息以提高精度。对用EMD分解轴承振动数据后得到一系列IMF分量,将包含故障频率的IMF的均方根值作为轴承的故障特征向量和预测模型输入参数。得到的实验结果表明,其较单一模型有更高的预测精度。
[Abstract]:With the large scale, high efficiency and integration of rotating machinery, the requirement of safe and reliable operation of rotating machinery is raised accordingly. Rolling bearing is one of the most important components in modern rotating machinery. Its normal working condition is directly related to the running state of machinery and even the whole system. However, due to the poor working conditions, it is necessary to carry out effective fault prediction for rolling bearings due to the disadvantages of great variation of life and easy damage. In this paper, the theory, method and key technology of rolling bearing fault prediction based on grey ELM and ELM neural network are deeply studied. The main research work is as follows: through the systematic analysis of the main methods of rolling bearing fault prediction and the current research situation in this field at home and abroad, it is pointed out that the rapid and accurate bearing fault prediction technology has become the focus of research. In the three kinds of fault prediction technology, the data based fault prediction technology is analyzed and introduced in this paper. In view of the non-stationary and nonlinear characteristics of rolling bearing vibration signal and the shortcomings of various classical methods, the grey theory is used to predict the development of bearing fault. The modeling mechanism and application scope of grey model are studied in depth. Aiming at the deficiency of traditional grey GM (1K1) model, a series of improved grey models are discussed. Among them, the grey multivariable prediction model MGM (1n) model is used to describe several diagnostic indexes uniformly from the point of view of the system, which is verified by rolling bearing experiments in this paper. The traditional neural network gradient learning algorithm has many problems, such as long training time, over-fitting of training samples and easy to fall into local optimum. This paper introduces the ELM neural network, ELM (extreme learning machine), which has the advantages of short learning time, simple algorithm, good generalization performance and the ability to avoid falling into local optimum. It has been successfully applied in the fields of function fitting and prediction. This paper introduces the basic theory of extreme learning machine and studies the learning algorithm of feedforward neural network and ultimate learning machine. The vibration signal of rolling bearing in real environment is difficult to be extracted under the background of noise and has the characteristics of nonlinearity and nonstationarity. In this paper, the vibration signals of rolling bearings are decomposed by empirical mode decomposition. Through theoretical and simulation analysis, it is found that EEMD decomposition has better anti-aliasing effect than EMD decomposition in nonlinear signal processing. Autocorrelation function has excellent performance in noise reduction and has great advantages in fault feature extraction. Grey model can predict development sequence, ELM neural network has high nonlinear mapping characteristics. This paper presents a new calculation method of combined weight coefficient, combining the two methods organically, establishes the grey ELM combination prediction model. Can describe the bearing both certainty and volatility of the complex trend. The combined model is applied to the bearing fault prediction, and the information is fully utilized to improve the accuracy. A series of IMF components are obtained by using EMD to decompose the bearing vibration data. The root mean square (RMS) value of IMF which contains the fault frequency is taken as the bearing fault eigenvector and the input parameter of the prediction model. The experimental results show that the prediction accuracy of the model is higher than that of the single model.
【学位授予单位】：北京交通大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TH133.33

【参考文献】