【作者】 任达千；

【导师】 吴昭同； 杨世锡； 严拱标；

【作者基本信息】 浙江大学 ， 机械制造及其自动化， 2008， 博士

【摘要】 机械设备故障诊断技术对于保证设备的安全、可靠、高效运行具有重要的理论意义和应用价值。机械设备的信号处理和故障特征提取是机械设备故障诊断技术中的关键问题,是机械学科的研究热点之一。局域均值分解(Local Mean Decomposition,LMD)是一种近年来新出现的时频分析方法,该办法的特点是可以获得信号在任何时刻的时频分布及有物理意义的瞬时频率,为机械设备故障诊断提供一种新的故障特征提取方法。本文结合国家自然科学基金项目,研究了基于LMD的旋转机械故障特征提取方法,并开发了基于嵌入式系统的故障诊断系统。论文的主要研究内容和章节安排如下:第一章论述了旋转机械故障诊断的研究意义,介绍了旋转机械故障诊断方法与技术的研究现状;综述了时频分析方法的发展以及在机械故障诊断中的应用;分析了基于LMD的时频分析方法和故障诊断系统的国内外研究现状,提出了本文的研究思路与研究内容。第二章介绍了基于LMD时频分析方法的一些基本概念,接着介绍LMD的理论基础和算法,然后给出实例说明LMD在旋转机械故障特征提取中的应用,最后比较LMD和经验模态分解(Empirical Mode Decomposition,EMD)的不同之处,指出LMD在旋转机械故障特征提取中的优势。第三章研究了纯调频信号瞬时频率的直接求取法,针对极值点附近畸变情况引进平滑处理以改进该算法,得到较好的效果;提出并证明了纯调频的瞬时频率直接求取法适用性的判别条件:纯调频信号的极值点为±1;分别应用直接法、Hilbert变换和Teager能量法对仿真信号和旋转机械实测振动信号进行分析比较,结果表明:直接法可以获得调频调幅信号的频率和幅值随时间变化情况,能够直观地反映波内调制现象,验证了瞬时频率直接求取法的有效性,为旋转机械故障振动信号时频特征提取提供了一种方法。第四章研究了采样效应对LMD分解的影响,给出纯调频信号的判据的实现方法,在实际信号处理时,必须在±1处设置一增减量δ。提出在信号的相邻极值点距离变化比较大的情况下,应用新的滑动平均跨度的选择方法,使LMD算法易于收敛。分析LMD端点效应的特点,提出了一种LMD端点效应的评价指标和一种基于端点延拓的LMD端点效应抑制方法。第五章建立了旋转机械的故障模拟实验装置,开发基于LMD的旋转机械故障分析原型系统;应用基于LMD的时频分析方法,研究了转子裂纹、油膜涡动和轴不对中三种典型故障振动信号的时频特征,验证了该方法提取故障特征的有效性。应用BP神经网络对故障分类做了实验研究。结论是:裂纹转子的振动信号有明显的波内频率训制现象;油膜涡动振动信号LMD分解以后,可以得到0.5倍频和1倍频两个频率成分,边缘谱的能量主要分布在0.5倍频和1倍频附近:不对中振动信号在时频谱上主要表现为频率调制现象,边缘谱在1倍频附近有一较宽的频带。LMD分解PF特征提取方法结合BP神经网络,与FFT频谱特征方法比较可以有效地对裂纹转子、油膜涡动和转子不平衡三种故障分类。第六章分析旋转机械故障分析系统的结构,给出了面向旋转机械的嵌入式故障分析系统整体方案;设计了一种带有自动放大控制(AGC)的键相信号预处理电路,提高了系统的可靠性;提出了一种跟踪转速的整周期、等相位的信号采集控制方法;研究了基于重采样技术的整周期采样方法,仿真表明重心法检测脉冲周期比过零法有更高的抗噪能力;对所开发的面向旋转机械嵌入式故障分析系统进行了性能测试,说明系统能满足实际数据采集和故障特征分析的要求。上述的技术和系统已经在相关的企业现场投入实际运行,效果良好。第七章总结了论文的研究内容,展望了论文研究工作的发展。更多还原

【Abstract】 In recently decade years, fault diagnosis technology has been deep researched and some important development has been gotten. Time-frequency analysis method based on Local Mean Decomposition (LMD) is a signal process method appeared in recent years. Through this method we can acquire the instantaneous frequency of the signal, which always has physical meaning in any time. The instantaneous frequency may be an important characteristic of rotating machine fault diagnosis. Based on the ’Study on the new method for rotating machine faults diagnosis based on independent component analysis’ (National Nature Science Fund Project, No: 50205025) and ’Study on the new method of noise source recognition of complex system based on ICA-EMD analysis’ (National Nature Science Fund Project, No: 50505016), the basic principle and algorithm of LMD, and its application in the rotating mechanic faults diagnosis is discussed in this paper. The main research content is as follows:In chapter one, we discuss the importance and the main method of rotating mechanic faults diagnosis. A survey of faults diagnosis of machines and a summary of recently development are presented. LMD based time-frequency method and the state-of-the art in home and abroad are introduced.In chapter two, we introduce some basic concept of LMD -based time-frequency analysis method. After that, the theory and the algorithm of LMD are presented. Then examples are given to show the application of LMD in the rotating mechanic faults diagnosis. In the end, the difference between LMD and EMD (Empirical Mode Decomposition) is described to point out the advantage of LMD.In chapter three, we discuss the direct method acquiring instantaneous frequency from a signal. We apply a smooth method to improve the direct method. A limit of direct method is given out. We also discuss the relationship between instantaneous frequency of the purely frequency modulation signal and instantaneous frequency of the PF(Product Function). By simulation signal and practice signal, the validity of the instantaneous frequency which is received by the direct method is testified. The direct method, Hilbert transform and Teager energy method are compared. The result shows that as to the signal instantaneous frequency, the extreme value of signal must be±1. For AM-FM signal, the direct method can gain the variation of frequency and amplitude. The direct method can reflect the intra-wave modulation phenomena.In chapter four, we discuss the sampling effect of LMD and the judge of purely frequency modulation signal. Aiming at the non-convergent problem of LMD algorithm, the new selection method of moving averaging span is provided, which can make LMD convergent in the case the distances between neighboring extreme points have great variety. A kind of evaluation index of LMD end effect is provided. The difference of the EMD end effect and LMD end effect is analyzed and compared, and we provide a new kind of signal extend technology to reduce LMD end effect. The conclusion is that the selection method of moving averaging span will influence the convergence of LMD algorithm. Compared to EMD, the degree of end effect of LMD is very light and the influence range is small.In chapter five, we apply the LMD-based time-frequency analysis method at vibration signal analysis of rotating machine. Firstly, we introduce the experiment device and the LMD-based faults diagnosis prototype system. Then we discuss the time-frequency character of three kinds of faults such as rotor crack, oil film whirling and shaft-misalignment, which testified the feasibility of LMD-based time-frequency analysis method in the faults diagnosis of rotating machine.In chapter six, we discuss the framework of the faults diagnosis system, the overall scheme of embedded faults diagnosis system is provided, and the new keyphase signal preconditioning circuit is developed which has been applied in a power unit. The practice testifies that reliability of data acquiring system can be greatly increased. The new circuit has obtained the Chinese invention patent authorization. Keyphase signal frequency multiplied technology is improved and the resampling-based integer period sample method is improved. The simulation shows that testing pulse period using gravity method will have higher noise-resistant ability than that using zero-cross method which is always used in the past. In this paper, relative detailed test result is given.In chapter seven, we summarized the research content of the whole paper. The prospect is given on the research direction and possible research content in the future.更多还原