基于谐波小波包和神经网络的旋转机械故障诊断系统研究

发布时间：2019-03-18 18:03

【摘要】：目前，振动检测是大型旋转机械故障诊断的主要手段，一般旋转机械发生故障时会产生复杂的动态非平稳振动信号，故如何准确地提取出此类信号的特征量是故障诊断的首要条件。谐波小波理论基于其频域严格的“盒形”特性，非常适合于非平稳信号的特征提取，但是由于旋转机械振动信号本身的复杂性，该方法还没有在旋转机械的故障诊断领域得到广泛的应用。本文在其他学者研究的基础上，提出了谐波小波包对旋转机械振动信号能量特征自动提取的方法，避免了转速和采样频率不同对信号特征提取的影响。本文首先对旋转机械故障诊断意义及其发展状况做了简单的介绍，选择了几种比较典型的传统信号处理方法，对它们的优缺点进行了对比研究；然后对谐波小波理论做了系统的研究，通过仿真信号详细分析了它在微弱信号、局部突变信号和近频信号的特征提取中的优势；研究了谐波小波包对不同转速和不同采样频率下的信号的能量特征提取方法；其次介绍了Elman神经网络的基本结构及其算法，并且与BP网络进行了对比，突出其在学习稳定性、收敛速度和故障识别率方面的优势；最后提出了谐波小波包和Elman神经网络相结合的思想，，设计了基于这种思想的旋转机械故障智能诊断系统的基本结构。采用LabVIEW和MATLAB相结合的方法，完整地设计了基于谐波小波包和Elman神经网络的旋转机械故障智能诊断系统；通过在转子试验台上模拟转子的四种典型故障，采集振动信号，输入诊断系统，结果显示该系统诊断性能良好。
[Abstract]:At present, vibration detection is the main means of fault diagnosis for large-scale rotating machinery. In general, complex dynamic and non-stationary vibration signals can be generated when rotating machinery fails. Therefore, how to accurately extract the characteristics of such signals is the first condition of fault diagnosis. Harmonic wavelet theory is very suitable for feature extraction of non-stationary signals based on its strict "box-shaped" characteristics in frequency domain, but because of the complexity of the vibration signal of rotating machinery, This method has not been widely used in fault diagnosis of rotating machinery. In this paper, based on the research of other scholars, a method of automatically extracting the energy feature of rotating machinery vibration signal by harmonic wavelet packet is proposed, which avoids the influence of different rotational speed and sampling frequency on the signal feature extraction. In this paper, the significance and development of rotating machinery fault diagnosis are introduced briefly, and several typical traditional signal processing methods are selected, and their advantages and disadvantages are compared. Then, the harmonic wavelet theory is studied systematically, and its advantages in the feature extraction of weak signal, local mutation signal and near-frequency signal are analyzed in detail through the simulation signal. The energy feature extraction method of harmonic wavelet packet at different rotation speed and different sampling frequency is studied. Secondly, the basic structure and algorithm of Elman neural network are introduced, and compared with BP neural network, its advantages in learning stability, convergence speed and fault recognition rate are highlighted. Finally, the idea of combining harmonic wavelet packet with Elman neural network is put forward, and the basic structure of intelligent fault diagnosis system for rotating machinery is designed based on this idea. The intelligent fault diagnosis system of rotating machinery based on harmonic wavelet packet and Elman neural network is designed based on the method of LabVIEW and MATLAB. By simulating four typical faults of rotor on the rotor test-bed, the vibration signal is collected and the diagnosis system is inputted. The results show that the diagnosis performance of the system is good.
【学位授予单位】：燕山大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：TH165.3

【参考文献】