【作者】 周小勇；

【导师】 叶银忠；

【作者基本信息】 上海海运学院 ， 电力电子与电力传动， 2001， 硕士

【摘要】 故障诊断技术是正在迅速发展的研究领域，是实际应用需求与多学科理论发展两个方面交替作用的结果。从实际应用方面看，随着现代化技术水平的不断提高，各类工程系统的复杂性大大增加，系统的可靠性与安全性已成为保障经济效益和社会效益的一个关键因素，得到了广泛的重视；从学科理论的发展方面而言，故障诊断具有很强的学科交叉性，现代控制理论、信号处理、模式识别、人工智能等学科领域近20年来的迅速发展，为解决复杂系统的故障诊断问题提供了有力的理论基础。 小波分析(Wavelet Analysis)即小波变换是80年代中期发展起来的一门新兴的数学理论和方法，它被认为是傅立叶分析方法的突破性进展，它具有许多优良的特性。小波变换的基本思想类似于Fourier变换，就是用信号在一族基函数张成的空间上的投影表征该信号。经典的Fourier变换把信号按三角正、余弦基展开，将任意函数表示为具有不同频率的谐波函数的线性迭加，能较好地刻划信号的频率特性，但它在时空域上无任何分辨，不能作局部分析，这在理论和应用上都带来了许多不便。小波分析优于傅立叶之处在于，小波分析在时域和频域同时具有良好的局部化性质，因为小波函数是紧支集，而三角正、余弦的区间是无穷区间，所以小波变换可以对高频成分采用逐渐精细的时域或空间域取代步长，从而可以聚焦到对象的任意细节。因此，小波变换被誉为分析信号的显微镜，傅立叶分析发展史上的一个新的里程碑。小波分析是一个新的数学分支，它是泛函分析、傅立叶分析、数值分析的最完美结晶；在应用领域，特别是在信号处理、图象处理、语音分析、模式识别、量子物理、生物医学工程、计算机视觉、故障诊断及众多非线性科学领域都有广泛的应用。 本文正是基于小波分析的特点，利用小波变换所得的小波系数模值和突变信号的关系来诊断故障；利用小波变换的多尺度分析技术对信号进行分解和重构，对信号进行去噪处理直接观测信号的波形特征来判别系统的故障或对信号的高频部分进上梧每运 颀士学位过文 人达分析庄轿厉哮饼中g地谓行小波系数模值检测来诊断故障；利用小波网络具有对任意信号或函数的良好逼近特性对故障进行分类诊断。本文的主要内容如下： 第一章概要介绍了故障诊断的由来和意义，总结了目前故障诊断的理论和方法，并介绍了小波分析技术目前在故障诊断中的应用情况。 第二章简要介绍了小波分析和傅里叶分析的差异，并分析了MATLAB所提供的小波函数的特点，探讨了小波基选择的标准，最后用实例说明了利用小波变换后的系数模值和信号突变的关系来诊断故障。 第三章利用小波变换的多尺度分解技术来诊断故障，并给出了一个应用实例。 第四章介绍了神经网络在故障诊断中应用情况，着重阐述了小波网络的结构、特点和实现方法。结出了一个小波网络的实现模型。 第五章介绍了小波神经网络应用于船舶主柴油机冷却系统的故障诊断实现方法。 第六章总结了全文所做的工作，并对小波分析在故障诊断中应用发展提出一些看法。更多还原

【Abstract】 The techniques for fault diagnosis are under rapid development. This is due to the interlacing between the demands from the practical applications and the achievements in various fields of the theory and the technology. As the modern technology develops, the engineering control systems become more and more complicated and advanced in practical applications, therefore the reliability and the security of those systems will play a critical role to both the society and the economy. More and more attention has been paid in the reliability and security in the last two decades. From the academic view, fault diagnosis has a closer relationship with many areas on science and technology such as modern control theory, signal processing, pattern recognition and artificial intelligence. The rapid development and the increasing achievements on these areas during the past two decades lay a solid foundation for the fault diagnosis of complicated systems.Wavelet analysis is a novel mathematical theory and method proposed in 1980s. It is regarded as a breakthrough of Fourier analysis because it has many wonderful characteristics. Fourier analysis consists of breaking up a signal into sine waves of various frequencies. Similarly, wavelet analysis is the breaking-up of a signal into shifted and scaled versions of the original wavelet. Lacking of space locality in time domain, Fourier analysis can only make certain of the integral singularity of a function or signal. As a result, it is difficult to detect the spatial position and distribution of broken signal by Fourier analysis. Wavelet analysis has the characteristic of spatial locality, and its wideness in both windows of the time and the frequency can be adjusted, so it can analyze the details of a signal. Therefore Wavelet analysis is called a microscope of signal analysis and a milestone of Fourier analysis. As a new embranchment of mathematics, Wavelet analysis is the most perfect combination of the functional analysis, Fourier analysis and numerical analysis. Wavelet analysis has been widely used in signal processing, image manipulation, voice analysis, pattern recognition, quantaphysics, biomedicine engineering, computer vision, fault diagnosis, some nonlinear fields and so on.In this dissertation, the fault diagnosis techniques based on Wavelet analysis were investigated. The relationship between the wavelet coefficient modules and the broken signal, the multiple-level Wavelet decomposition and reconstruction for de-noising, and the Wavelet neural network were analyzed or studied and were applied to detect the variant types of faults. The dissertation is outlined below.In Chapter 1 we will briefly introduce the state-of-art of fault diagnosis. The major aspects on its theory and techniques will be summarized. The applications of the Wavelet analysis in fault diagnosis will also be introduced.Chapter 2 is devoted to compare Wavelet analysis and Fourier analysis. The features of each element in Wavelet family will be studied, and the criteria in choosing the wavelet function for different purposes will be outlined. An example is given to show how to use the relationship between the wavelet coefficient modules and the broken signal to achieve fault diagnosis.In Chapter 3 the multiple-level Wavelet decomposition and reconstruction for de-noising will be investigated and applied to detect faults. A practical example will be given to show the fault diagnosis procedure.Chapter 4 will be on the application of the Wavelet Neural Network (WNN) in fault diagnosis. The basic knowledge on artificial neural network and its application in fault diagnosis will be introduced. The structure and the algorithm of the WNN will be studied. A realization model for WNN will be constructed.In Chapter 5, the WNN is applied to build up a fault diagnosis system for the cooling system of the marine engine.Chapter 6 will give a summary of this dissertation. Also we will put forward some problems to be studied further and trends of development in this field.更多还原