基于EMD的起重机齿轮箱故障特征提取研究

发布时间：2018-11-04 15:28

【摘要】：近年来,起重机械在国民经济中的影响越来越大,其安全问题一直受到人们的广泛关注。一旦事故发生,对事故性质的认定往往存在一定的难度。其中最易发生故障的零部件之一就是起重机齿轮箱,本文主要对起重机齿轮箱故障特征提取方法进行了研究。齿轮箱被普遍应用于机械设备传动中,作为连接和传递动力的部件,齿轮的磨损、裂纹和断齿故障都会导致机器运行不正常,因此在线准确监测和诊断齿轮箱的故障是十分必要的。由于齿轮箱振动信号的非线性、非平稳特性,当齿轮出现故障时,通常都有较强的背景噪声存在,这样会影响齿轮箱故障诊断的准确性。本文首先利用在传统软、硬阈值方法基础上改进的小波阈值方法,对采集的齿轮箱振动信号进行降噪预处理,利用EMD方法对信号分解,将分解后得到的信号进行谱分析,结合齿轮故障振动信号的调制频率及其边频带分布特点,实现齿轮故障诊断分析,最后引入BP神经网络故障识别方法,能够准确的识别齿轮故障状态。本文主要的研究内容和结果包括:(1)研究齿轮故障表现出的常见损伤形式及其产生缘由,从而能够准确地判断此故障检测参数的有效性;基于齿轮发生故障时出现啮合频率调制和边频带分布现象,得到齿轮典型故障和相应振动信号的特征频率之间的关系。(2)为抑制齿轮故障信号中噪声的干扰,突出故障特征频率,本文利用一种改进的小波阈值降噪方法,通过对加噪信号仿真实验验证与传统的软、硬阈值降噪方法进行了对比,证明了此降噪方法的有效性。(3)将改进小波分析阈值法和EMD方法相结合分析振动信号。综合对比齿轮箱在不同故障状态下的振动信号的时域波形、幅值谱、Hilbert谱和边际谱,获得齿轮箱在不同故障状态下的故障特征频率及其附近调制边频带特征,成功完成齿轮箱故障诊断分析。(4)引入BP神经网络,利用EMD提取的相应特征向量,作为神经网络的训练样本和测试样本。通过对BP神经网络的学习和识别,能够分类出相应的工作状态,对齿轮箱的相应故障做出判定。实验表明此方法适合于齿轮箱故障识别。
[Abstract]:In recent years, lifting machinery has become more and more important in the national economy, and its safety has been paid more and more attention. Once an accident occurs, it is often difficult to identify the nature of the accident. One of the most prone parts is crane gearbox. In this paper, the fault feature extraction method of crane gearbox is studied. Gearbox is widely used in mechanical transmission. As a part of connecting and transmitting power, gear wear, crack and broken tooth failure will lead to abnormal operation of the machine. Therefore, it is necessary to accurately monitor and diagnose the gearbox faults on line. Because of the nonlinear and non-stationary characteristics of the gear box vibration signal, there is usually a strong background noise when the gear is in trouble, which will affect the accuracy of the gear box fault diagnosis. In this paper, the wavelet threshold method, which is based on the traditional soft and hard threshold method, is firstly used to pre-process the vibration signal of the gearbox. The signal is decomposed by EMD method, and the decomposed signal is analyzed by spectrum analysis. According to the modulation frequency of gear fault vibration signal and the characteristics of frequency band distribution, the gear fault diagnosis and analysis is realized. Finally, the fault identification method of BP neural network is introduced, which can accurately identify the gear fault state. The main contents and results of this paper are as follows: (1) the common damage forms and their causes of gear failure are studied, so that the validity of the fault detection parameters can be accurately judged; Based on the phenomenon of meshing frequency modulation and side band distribution when gear fault occurs, the relationship between the characteristic frequency of gear typical fault and corresponding vibration signal is obtained. (2) in order to suppress the interference of noise in gear fault signal, Highlighting the characteristic frequency of fault, this paper uses an improved wavelet threshold denoising method, and compares it with the traditional soft and hard threshold denoising method through the simulation of the noise-added signal. It is proved that this method is effective. (3) the improved wavelet analysis threshold method and the EMD method are combined to analyze the vibration signal. The time domain waveform, amplitude spectrum, Hilbert spectrum and marginal spectrum of vibration signal of gearbox in different fault state are compared synthetically, and the fault characteristic frequency and the modulation side band characteristic of gearbox in different fault state are obtained. The analysis of gearbox fault diagnosis is completed successfully. (4) the BP neural network is introduced and the corresponding eigenvector extracted by EMD is used as the training sample and test sample of the neural network. By learning and recognizing the BP neural network, the corresponding working states can be classified and the corresponding faults of the gearbox can be judged. Experiments show that this method is suitable for gearbox fault identification.
【学位授予单位】：上海应用技术大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TH21

【参考文献】