【作者】 郑晓斌；

【导师】 史铁林；

【作者基本信息】 湖北工业大学 ， 测试计量技术及仪器， 2007， 硕士

【摘要】 随着机械故障智能诊断系统的发展,知识获取(KA)变得越来越重要。知识获取是人工智能(AI)领域最关键的问题之一,也是机械故障智能诊断系统丞待解决的关键问题。目前各种智能故障诊断专家系统的共同缺陷就是系统拥有的知识量太少,获取知识精度不高。本文针对以上问题,以故障特征诊断知识获取为中心,对自组织特征映射(SOFM)神经网络在故障特征知识获取方面的应用做了研究,本文主要研究工作如下:1研究了SOFM网络的算法原理和步骤及其在故障诊断知识自动获取方面的应用。由于所获取的知识隐藏于网络连接权值中,难以被理解,所以对网络训练结果的可视化方法进行研究,应用U-矩阵法和在此基础上改进的可视化方法对诊断结果进行直观的表达。2探讨了小波分析理论,对小波包变换(WPT)用于信号的分解和重构进行了研究,在此基础上探讨了一种利用小波包变换提取信号能量特征的方法,它能够有效地对非线性相关的高维数据进行降维和特征提取。3针对时域特征和SOFM结合的知识获取方法中存在的不足,提出了一种基于小波-SOFM的知识获取模型,能够进一步提高获取知识的精确度。4以滚动轴承故障诊断知识获取为例,介绍了小波-SOFM方法在故障诊断知识自动获取方面的应用,并和时域特征为输入向量的方法做了比照,验证了该方法的有效性。之后,以滚动轴承为研究对象,以MATLAB为开发工具,结合SOM Toolbox 2.0工具箱,开发了一个以小波-SOFM自动知识获取模型为核心简单的滚动轴承故障诊断系统。最后,对全文作了总结回顾,归纳了本文解决的问题,指出了SOFM的在机械故障诊断知识获取方面今后值得关注和深入研究的方向。更多还原

【Abstract】 With the development of the mechanical fault intelligence diagnosis system, Knowledge Acquisition (KA) becomes more and more important. KA is one of the key problems in the domain of Artificial Intelligence (AI) and mechanical fault intelligence diagnosis system. At present, one common defect of intelligence fault diagnosis expert system is that the knowledge which included the system is limited and inaccurate. According these questions, taking the fault feature diagnosis knowledge acquisition as the core, the applications of fault feature knowledge acquisition based on Self-Organizing Feature Map (SOFM) neural network are researched. The main researches and contributions are fourfold:1 How the SOFM network works is researched and the application of SOFM in fault diagnosis knowledge acquisition is discussed. It is very difficult to be understood that the knowledge is hide in the network’s weight. So the visualization of the SOFM training results is investigated, U-matrix method and a new visualizing technique is used.2 The theory of Wavelet Analysis is discussed. The Application of Wavelet Packet Transform (WPT) in signal decomposition and restructure is researched. According the theory, a method of energy feature extraction which can be used to reduce the dimensionality of nonlinear high dimensional data space and extract feature, is proposed.3 The modal of KA based on WPT and SOFM, which can be used to acquire knowledge automatically, is proposed. It is more suitable for fault feature knowledge acquisition from the fault signals than time domain feature combining with SOFM.4 The new model is applied to automatic knowledge acquisition of fault diagnosis of roll bearing. Its simulation results indicate that the method based on WPT and SOFM can acquire the knowledge of roll bearing fault classification accurately. Later, taking roll bearing as object, regarding MATLAB as developing instrument, combining with SOM Toolbox2.0, a roll bearing fault diagnosis system based on WPT and SOFM model is developed.Finally, the whole thesis is summarized, and some future research areas of condition monitoring and fault diagnosis technique based on SOFM are highlighted.更多还原