局部均值分解及其在机械故障诊断中的应用研究

发布时间：2018-07-28 08:53

【摘要】：工程中故障振动信号一般都为非线性非稳态的信号,其振动频率会随着时间不断的发生变化,传统的傅里叶变换(Fourier Transform, FT)只能得到振动信号的频率分布范围,不能辨别信号频率发生突变的时间点。短时傅里叶变换及后来发展起来的Gabor变换还有同期发展起来的Wignar-Ville分布(WVD)都是利用窗函数对原始信号进行截取,且其所加窗函数是固定不变的,在分析多分量信号时,WVD在时频谱上会出现交叉项。小波变换的窗函数宽度可以随其在时间轴上平移而发生变化,但小波变换本质上是对原始信号的机械截取,仍缺乏对信号分解的自适应性。1998年N.E. Huang提出一种新的时频分析方法希尔伯特-黄变换,2005年J. S.Smith提出另一种新的时频分析方法方法——局部均值分解(Local Mean Decomposition, LMD),都是对以傅里叶变换为基础的线性稳态谱分析的一个重大突破。LMD是一种自适应的时频分析方法,可以将振动信号分解为一组频率从高到低自动排列的乘积函数(Product Function, PF), PF分量是一个包络值为1的调频函数和一个包络函数的乘积。LMD方法对信号的分解方式很好地解决了以往时频分析方法对信号的机械划分的问题,从而使分解结果更准确。文章对LMD方法原理及方法中存在的问题进行了详细的介绍,并着重针对LMD方法的端点效应进行了改进。将LMD方法与Teager能量算子和1.5维谱相结合,使滚动轴承的故障特征频率更明显；LMD方法与增强包络谱结合,并提出新的PF分量筛选准则,使轴承故障特征频率更容易辨别。随后将LMD方法与支持向量机(Support Vector Machine, SVM)相结合,同时提出新的特征值提取方法提取故障特征,用SVM对这些故障特征进行分类,利用该方法对故障轴承的故障程度和转子碰摩位置进行了识别,都取得了良好的效果,验证了所提方法的可行性。
[Abstract]:In engineering, the fault vibration signal is usually nonlinear and unsteady, and its vibration frequency will change with time. The traditional Fourier transform (Fourier Transform, FT) can only get the frequency distribution range of the vibration signal. The time point at which the frequency of the signal is mutated cannot be distinguished. The short time Fourier transform (STFT), the later developed Gabor transform and the Wignar-Ville distribution (WVD) developed in the same period are all used to intercept the original signal by window function, and the added window function is fixed and invariant. When analyzing the multicomponent signal, the WVD will appear cross term in the time spectrum. The width of window function of wavelet transform can be changed with its translation on the time axis, but wavelet transform is essentially mechanical interception of the original signal. In 1998, N. E. Huang proposed a new time-frequency analysis method, Hilbert-Huang transform. In 2005, J. S.Smith proposed another new time-frequency analysis method-local mean decomposition (Local Mean Decomposition, LMD),. An important breakthrough in linear steady-state spectrum analysis based on Fourier transform. LMD is an adaptive time-frequency analysis method. The vibration signal can be decomposed into a set of product functions with frequency from high to low, which is the product of a frequency modulation function with an envelope value of 1 and an envelope function. LMD method is a good way to decompose the signal. The problem of mechanical division of signals in previous time-frequency analysis methods is solved. Thus, the decomposition result is more accurate. In this paper, the principle and problems of LMD method are introduced in detail, and the endpoint effect of LMD method is improved. By combining the LMD method with the Teager energy operator and 1.5-D spectrum, the fault characteristic frequency of rolling bearing is more obvious than that of the enhanced envelope spectrum. A new PF component screening criterion is proposed to distinguish the fault feature frequency of the bearing more easily. Then the LMD method is combined with the support vector machine (Support Vector Machine, SVM), and a new feature extraction method is proposed to extract the fault features. The fault features are classified by SVM. The method is used to identify the fault degree of the bearing and the rubbing position of the rotor. Good results are obtained and the feasibility of the proposed method is verified.
【学位授予单位】：华北电力大学(北京)
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TH17

【相似文献】