【作者】 潘明清；

【导师】 周晓军；

【作者基本信息】 浙江大学 ， 机械制造及其自动化， 2005， 博士

【摘要】 机械故障诊断本质上是一个模式分类问题。支持向量机由于解决分类问题有着优良的表现，得到日益广泛的应用。本论文针对三种现实分类问题：两类分类、单值分类和多类分类问题，基于支持向量机分类理论，结合工程实际项目对机械模式分类方法进行了研究。 论文首先介绍了机械故障诊断的历史、意义及研究现状，分析了现有故障诊断理论方法的优点及不足之处。接着，就机械故障诊断的应用背景——汽车检测领域作了叙述和总结，描述了汽车总成关键部件驱动桥试验台的开发现状。讨论了支持向量机(SVM)理论研究方法及其国内外的应用研究现状，对于故障模式分类问题，SVM也有相应理论和应用研究。论文针对解决三种机械故障模式分类的问题，提出了一个基于SVM理论的的故障诊断架构。最后，给出了本论文的主要研究内容。 第二章介绍了支持向量机的基本理论和算法。回顾了支持向量机的理论基础——统计学习理论和机器学习，讨论了统计学习理论的结构和需要进一步研究的领域，在此基础上引出支持向量机。分析了支持向量机分类理论的原理和算法，简略的介绍了基于支持向量机的两类、单值和多类分类算法，为后续章节解决实际问题作一个理论铺垫。 第三章对驱动桥故障诊断的两类分类问题进行了研究。由于故障诊断包含特征提取和诊断决策过程，根据驱动桥疲劳试验表现出的故障非线性特征，提出用核主元分析的方法首先对试验过程中采集到的正常和故障样本进行核主元特征提取，作为支持向量机的输入向量。结合核主元和支持向量机共同具有的核函数特点，提出层次向量机故障诊断方法，采用遗传算法优化层次向量机的核函数参数及和折衷参数，对驱动桥疲劳试验的正常和裂纹故障作了识别诊断，得到较满意结果。 第四章针对汽车驱动桥异响的诊断缺乏足够故障样本的问题，提出基于支持向量数据描述算法进行诊断。分别利用时域和频域指标作为支持向量数据描述分类器的输入特征向量。时域方法采用多个时域统计指标，然后用主元进行特征提取。频域方法根据正常和异响信号的能量不均衡，采用谱熵提取频域指标。两种特征提取的方法结合支持向量数据描述分类器对汽车驱动桥的异响诊断都取得更多还原

【Abstract】 Machine fault diagnosis is a problem of pattern classification in nature. Because of the excellent performance on classification, support vector machine (SVM) is more and more widely used to solve classified problems in practical world. Based on SVM theory, this dissertation develops a research on machine fault pattern classification for engineering project applications.The thesis first introduces the history, the significance and the current research status of machine fault diagnosis. After analyzing the virtue and shortcoming of current fault diagnosis theories, the paper summarizes the application background of machine fault diagnosis. In the field of auto detection, driving axle is a key assembly component. The development of driving axle test system is described. The SVM theory and its research status in nation and oversea is analyzed. The availability of SVM application on machine fault diagnosis is discussed. Through the above analysis, a structure of fault diagnosis based on SVM theory is founded. In the end of this chapter, the main research content and method path of this thesis is proposed.Chapter two introduces the theory and arithmetic of SVM. After reviewing the basic theory-statistical learning theory (SLT) and machine learning (ML), the thesis discusses the SLT’s structure and the question which SLT is faced. Based on analyzing the principle and arithmetic of SVM, three classification methods are introduced to present theory matting for latter chapters.Chapter three does a research on two class classifier problem of fault diagnosis for driving axle. Feature extraction and diagnosis decision-making are two parts of fault diagnosis. Because of the non-linearity charactering of axle faulty in fatigue experimentation, the paper firstly proposes kernel primary component analysis (KPCA) method to extract the features from normal and fault samples in the course of fatigue test. Then the kernel primary components are used as the input of SVM classifier. For kernel function is used both in KPCA and SVM, the paper proposes the concept of kernel SVM and uses genetic arithmetico (GA) to optimize the compromise parameters of kernel function and SVM. Under the disassembly situation更多还原