基于自组织网络的故障诊断算法改进及可视化研究

发布时间：2018-07-20 18:58

【摘要】：随着生产自动化程度的提高，为了保证机械设备高效安全的运行,故障诊断系统已成为机械设备中不可缺少的一部分。结合当前故障诊断和状态监测的发展趋势和要求，本学位论文从机械故障聚类分析和可视化表达角度出发，结合自组织映射神经网络的最新研究成果，改进了自组织映射(Self-Organizing Map. SOM)网络的学习算法，缩短了样本学习时间；研究了基于自组织神经网络的高维数据可视化方法，并将改进后的SOM网络运用于机械设备的故障诊断中。本课题的研究旨在提高SOM神经网络自组织学习的效率，寻求高维数据可视化方法，为机械设备故障诊断和状态监测等提供新方法、新工具；并进一步丰富和发展机械设备状态监测和故障诊断智能化的基础理论和方法。本文提出了一种基于自组织神经网络的故障诊断算法改进及可视化研究方案，其主要研究工作如下：（1）探讨了SOM神经网络在故障诊断方面的国内外应用及研究现状，研究了SOM网络的算法原理及其在故障诊断领域的应用。（2）阐述了自组织神经网络算法，并针对Kohonen竞争学习机制中有些结点始终不能获胜的问题，对匹配神经元策略进行了改进；同时，由于网络训练由任意初始条件开始，网络结构完全由经验决定，针对这一缺陷，本文还提出了一种动态神经元数目改进方案。（3）针对网络学习得到的知识隐藏于连接权值中，对神经网络的可视化理论进行研究，对自组织网络诊断系统进行可视化，并对U-矩阵可视化算法进行了改进，为动态监测被检测对象的运行状态提供了新的方法。（4）以滚动轴承为研究对象，MATLAB7.0为开发工具，对该方案进行了验证。结果表明：该方案生成一种更灵活的拓扑结构以便能更好的匹配复杂数据分析问题；同时缩短了SOM计算时间。最后，对全文作了总结回顾，归纳本文提出的方案能够解决的问题，指出了SOM在机械故障诊断方面值得关注及研究的方向。
[Abstract]:With the improvement of production automation, in order to ensure the efficient and safe operation of mechanical equipment, fault diagnosis system has become an indispensable part of mechanical equipment. According to the development trend and requirements of fault diagnosis and condition monitoring, this dissertation improves the self-organizing map from the point of view of clustering analysis and visual representation of mechanical faults and the latest research results of self-organizing mapping neural networks. The learning algorithm of SOM) network shortens the sample learning time, studies the high dimensional data visualization method based on self-organizing neural network, and applies the improved SOM network to fault diagnosis of mechanical equipment. The purpose of this research is to improve the efficiency of SOM neural network self-organization learning, to seek high-dimensional data visualization methods, and to provide new methods and tools for mechanical equipment fault diagnosis and condition monitoring. And further enrich and develop the basic theory and method of condition monitoring and fault diagnosis of mechanical equipment. In this paper, an improved and visualized algorithm for fault diagnosis based on self-organizing neural network is proposed. The main research work is as follows: (1) the application and research status of SOM neural network in fault diagnosis at home and abroad are discussed. In this paper, the algorithm principle of SOM network and its application in fault diagnosis are studied. (2) Self-organizing neural network algorithm is expounded, and the problem that some nodes in Kohonen competitive learning mechanism can not win is discussed. At the same time, the network training begins with arbitrary initial conditions, and the network structure is completely determined by experience. This paper also proposes a dynamic neuron number improvement scheme. (3) aiming at the knowledge of network learning hidden in the connection weight, the visualization theory of neural network is studied, and the self-organizing network diagnosis system is visualized. The U- matrix visualization algorithm is improved to provide a new method for dynamic monitoring of the running state of the detected object. (4) taking rolling bearing as the research object, MATLAB 7.0 as the development tool, the scheme is verified. The results show that the proposed scheme generates a more flexible topology to better match the complex data analysis problem and shortens the SOM computing time. Finally, the paper summarizes and reviews the full text, sums up the problems that can be solved by the scheme proposed in this paper, and points out that SOM is worthy of attention and research direction in the field of mechanical fault diagnosis.
【学位授予单位】：湖北工业大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：TH165.3

【相似文献】