【作者】 郭雄伟；

【导师】 伍星；

【作者基本信息】 昆明理工大学 ， 机械电子工程， 2010， 硕士

【摘要】 滚动轴承是各种旋转机械中应用最为广泛的一种通用机械部件,也是确保机械设备功能和性能的关键部件。由于滚动轴承的结构特点及工作条件恶劣,极易造成损坏,对滚动轴承进行故障诊断与状态监测意义重大。故障轴承振动信号呈现出较强的非高斯、非线性、非平稳特性,使用传统的基于线性平稳假设的信号处理方法处理效果不理想。高阶统计量(Higher-Order Statistic, HOS)是最近十几年发展较快的一种现代信号处理方法,由于其出色的噪声消除性能及相位保留能力,是处理非线性、非高斯、非平稳、非最小相位、非因果信号的有效手段,被广泛应用于雷达、声纳、生物医学、地球物理、盲信号处理、盲系统辨识、机械故障诊断等众多领域。本文对双谱、三谱、双相干谱、双谱切片、Hilbert双谱、高阶时频谱的定义、性质、估计方法及物理意义进行了研究；讨论了信号幅值和相位重构的意义及常用方法,使用基于双谱的最小二乘法对轴承振动信号进行相位和幅值重构分析；讨论了滚动轴承故障机理及振动模型,对滚动轴承振动信号在高频共振处带通滤波,对滤波后的信号进行双谱分析并与传统的双谱分析结果进行了对比,结果表明该方法在消除噪声、突出故障特征方面具有优势；高阶时频分布兼有时频分布及高阶统计量的优点,针对滚动轴承振动信号的高阶时频及其切片分析表明该方法的有效性。本文还研究了常用的机械故障分类方法和特征提取方法,使用双谱峰值频率对信息、振动信号AR (Autoregressive, AR)模型参数等构造用于轴承模式识别的特征向量；研究了常用的优化算法及故障分类算法,分别使用遗传算法和微粒群优化算法对支持向量机的主要参数进行优化,使用优化的支持向量机对测试数据进行分类,取得较高的分类准确率。最后,基于MATLAB的GUIDE对本文中所使用和改进的算法开发了MATLAB工具包。更多还原

【Abstract】 Rolling bearings are one of the most common elements in rotating machinery; rolling bearings are also critical parts to insure function and performance of rotating machinery. Because of the structural characteristics and adverse working conditions, rolling bearings are vulnerable to damage. Thus, fault diagnosis and condition monitoring of rolling bearings is important. The vibration signals caused by faulty bearings are typical non-Gaussian, non-linear, and non-stationary; the conventional methods based on linear time-invariant are not effectiveness.Higher-Order statistic is a kind of new signal-processing method which with rapid development in recent decades. Because of its capability of eliminating Gaussian noise and retaining the signal phase information, HOS is a useful tool for non-linear, non-Gaussian, non-stationary, non-minimum phase, non-causal signals, and it is now applied in many fields, such as radar, sonar, physical geography, biomedicine, blind system identification, blind signal separation, fault diagnosis.In this dissertation, the definition, characteristics, algorithm and physical meaning of mainly used HOS is summarized. Signal amplitude and phase reconstruction from its bispectrum is proposed, different reconstruction methods are discussed. Mechanism of rolling bearing vibration and models of different defect bearings are given. Vibration signals of bearings are pre-processed and the high frequency resonance signals are used for bispectrum analysis, experiment results have shown the effectiveness of these methods. Higher-Order time frequency distributions have both the advantages of HOS and time frequency distribution; Higher-Order time frequency distributions analyses of bearing vibration signals are shown feasible. Defect classification and feature extraction methods are also researched in this dissertation, peak information of bispectrum, AR model parameters are used for feature vectors of support vector machine. PSO and genetic algorithm are use for optimizing key parameters of SVM, the optimized SVM is used for bearing condition classification, and the results have shown that the diagnosis precision has been improved. Finally, a toolbox is developed for the proposed HOS analysis algorithms based on the MATLAB GUIDE.更多还原