时频联合分析及其在非平稳振动信号处理中的应用

发布时间：2018-07-11 10:30

本文选题：非平稳信号 + 时频联合分析　；参考：《哈尔滨工业大学》2016年博士论文

【摘要】：机械结构的振动信号中包含着结构本身的参数信息,也包含着复杂的故障信息。通过对振动信号进行深入分析处理可以有效地获得结构的模态参数、激励方式和故障类型等信息。传统的傅里叶分析方法已经成功地将振动信号从时域转换为频域,但只能处理平稳的信号。对于频率特征或统计特性随时间变化的随机振动信号,傅里叶分析则变得不再适合。自然界和工程应用中,非线性和非平稳信号十分常见,经典的线性和平稳性假设在很多情况下并不能准确地描述这些特性,特别是时变系统和含有故障的系统,其时变参数和故障特征往往包含在非平稳的信号特性中,因此,对非平稳信号频率随时间变化的特征进行分析变得尤为重要。近年来,许多学者针对非平稳信号处理方法展开了研究,其中最有效的分析手段是时频联合分析,它可以同时表述信号的时域和频域特征。时频分析手段包括经验模式分解法、小波变换、Wigner-Ville分布等非参数化方法以及自适应调频小波分解等参数化方法,这些方法都能表示信号的频率随时间变化的特性。本文基于上述各种时频分析方法,对时频联合分析理论进行系统地分析,并比较了各种方法的优势和局限性。在此基础上,提出了一些改进方法,分别利用仿真和实际信号进行了有效性和精度验证。主要研究内容分为以下几个方面:首先,论文对目前较热门的几种时频分析方法进行了综述,并通过实例比较了各自的适用范围和优缺点。其次,针对经验模式分解存在的模态混叠问题,引入解析模态分解,并针对解析模态分解抗噪性能较弱的问题,利用离散小波变换对信号进行降噪处理。仿真试验表明,该方法能在一定噪声条件下准确地辨识出线性系统的刚度、阻尼等模态参数,并且能根据其时频特性判断系统是否存在非线性和确定非线性类型。再次,建立了一套完整的基于连续小波变换的时变系统模态参数辨识算法。该算法利用Morlet小波尺度图提取小波脊和骨架曲线,再利用小波脊辨识系统的时变刚度,利用小波骨架包络辨识时变阻尼,并通过单自由度和二自由度系统进行了仿真验证。仿真系统包含了实际工程中普遍存在的参数突变和渐变情况,结果表明,本文的辨识方法能有效地识别系统时变参数。第四,研究了自适应线性调频小波变换在处理非线性调频信号时的局限性,并针对性地提出了一种改进算法。该方法通过约束最优线性调频小波的时域长度来减小分段逼近带来的辨识误差。仿真结果表明,改进算法对提升瞬时频率辨识精度效果明显,特别是针对被分析信号频率变化较快的情况。第五,针对Wigner-Ville分布的交叉项问题展开了研究。提出了一种既保留Wigner-Ville分布高时频分辨率,同时又有效抑制交叉项的方法:利用小波尺度图构造时频滤波器,对Wigner-Ville分布进行时频滤波。仿真和实际信号分析结果表明,该方法对线性调频和非线性调频信号都具有很强的分析能力,且计算量较小,十分适合工程应用。最后,利用三个工程实例,进一步论证了论文提出的几种时频分析方法及其改进方案的有效性。工程实例包括列车轮对多边形磨损后的振动信号、列车-桥梁系统振动信号和蝙蝠回声定位信号等。
[Abstract]:The vibration signal of the mechanical structure contains the parameter information of the structure itself, and also contains complex fault information. Through the deep analysis and processing of the vibration signal, the modal parameters, the mode of excitation and the type of the fault can be effectively obtained. The traditional Fourier analysis method has successfully transferred the vibration signal from the time domain. The Fourier analysis is no longer suitable for the random vibration signals of frequency characteristics or statistical characteristics with time. In nature and engineering applications, nonlinear and non-stationary signals are very common, and the classical linear peace stability hypothesis can not be accurately described in many cases. Some characteristics, especially the time-varying system and the system containing the fault, are often included in the nonstationary signal characteristics. Therefore, it is very important to analyze the characteristics of the nonstationary signal frequency with time. In recent years, many scholars have studied the nonstationary signal processing methods, among which the most important The effective analysis method is the time frequency joint analysis, which can simultaneously express the time and frequency characteristics of the signal. The time-frequency analysis means include the empirical mode decomposition method, the wavelet transform, the Wigner-Ville distribution and other parametric methods, such as the adaptive frequency modulation wavelet decomposition and so on. These methods can all express the frequency of the signal with time. Based on the various time-frequency analysis methods mentioned above, this paper systematically analyzes the theory of time frequency joint analysis, and compares the advantages and limitations of various methods. On this basis, some improvement methods are proposed, and the effectiveness and precision of the simulation and actual signals are used respectively. The main research contents are divided into the following aspects: First of all, this paper summarizes several popular time frequency analysis methods, and compares their respective applications and advantages and disadvantages through examples. Secondly, the analytical modal decomposition is introduced to the modal decomposition of empirical mode decomposition, and the discrete wavelet transform is used to solve the problem of weak anti noise performance of analytical modal decomposition. The simulation test shows that the method can identify the stiffness, damping and other modal parameters of the linear system accurately under certain noise conditions, and can determine whether the system has nonlinear and nonlinear types according to its time-frequency characteristics. Thirdly, a set of complete time variable system based on continuous wavelet transform is established. The algorithm uses the Morlet wavelet scale map to extract the wavelet ridge and the skeleton curve, and then uses the wavelet ridge identification system to identify the time-varying stiffness, uses the wavelet skeleton envelope to identify the time-varying damping, and through the single degree of freedom and two degree of freedom system, the simulation system is carried out. The simulation system contains the general existence of the actual engineering. The results show that the identification method in this paper can effectively identify the time-varying parameters of the system. Fourth, the limitations of adaptive LFM wavelet transform in the processing of Nonlinear FM signals are studied, and an improved algorithm is proposed. The method can restrain the time domain length of the optimal linear frequency modulation wavelet. To reduce the identification error caused by the piecewise approximation, the simulation results show that the improved algorithm is effective in improving the accuracy of the instantaneous frequency identification, especially in the case of the fast change in the frequency of the analyzed signal. Fifth, a study is carried out for the cross term problem of the Wigner-Ville distribution. A high time frequency division of the Wigner-Ville distribution is proposed. The method of discrimination and effective suppression of cross terms: using the wavelet scale to construct time frequency filter to filter the Wigner-Ville distribution. The simulation and actual signal analysis show that the method has strong analytical energy for both linear frequency modulation and Nonlinear FM signals, and the calculation is small. Finally, it is very suitable for engineering applications. By using three engineering examples, several time frequency analysis methods and the effectiveness of the improved scheme are further demonstrated. The engineering examples include the vibration signals of the train wheel pair after polygon wear, the vibration signal of the train bridge system and the echo location signal of the bat.
【学位授予单位】：哈尔滨工业大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：TH113.1;TN911.7

【相似文献】