基于包络分析的齿轮故障监测及基于人工神经网络的齿轮故障分类

发布时间：2018-01-10 23:10

本文关键词：基于包络分析的齿轮故障监测及基于人工神经网络的齿轮故障分类　出处：《重庆大学》2011年硕士论文　论文类型：学位论文

【摘要】：齿轮是机械传动系统中应用最为广泛的部件,现代科技的发展使机械设备越来越精密化,给这些设备的状态监测带来困难。齿轮箱是机械设备中的重要一员,其工作状态的监测必须具有实时性和有效性,这样机械设备中任何故障都能被检测到,并能在早期得到及时的维修。包络分析在轴承和齿轮的故障诊断中已经得到广泛的应用,传统方法要求监测信号在带通滤波前需进行时域同步平均处理,以提高包络分析的效果。然而,在实际应用中,齿轮振动信号的时域同步平局处理很难进行,需要其他的方法来辅助齿轮故障诊断中包络分析。针对上述问题,本文提出了一种新的解调方法,它对齿轮振动信号进行以共振频率为中心频率(一般远离具有较大幅值的齿轮啮合频率谐波分量)的带通滤波,再进行包络分析,实现两种类型,不同损伤程度的齿轮局部故障(轮齿裂纹和剥落)的诊断和分类。齿轮的局部故障将导致一对啮合的齿轮,每旋转一圈的过程中会产生低幅脉冲信号,它会和齿轮的结构共振信号产生调制现象。但是这个低能量的信号往往会被淹没在齿轮箱其他信号源产生的高能量信号中。本文提出了一个方法,它通过从监测信号中提取故障激励信号从而获得齿轮的故障信息。本文提出的包络分析方法步骤如下: ①通过观察齿轮箱监测信号频谱中的共振成分,寻找合适的解调频带,以利于提取由齿轮局部故障产生的脉冲信号,用于故障诊断。 ②在结构共振频率附近(一般远离具有较大幅值的齿轮啮合频率谐波分量)选择带通滤波器的中心频率,通过观察原信号的频谱图,选择带通滤波器的带宽,使带通频段能覆盖整个共振频率区间,它能有效的去除齿轮啮合频率分量的影响。 ③通过对带通滤波后的信号进行基于Hilbert变换的解调,解调后的包络信号只包含与齿轮故障频率相关的分量。再对包络信号进行FFT变换获得其频谱图,从而可以提取并观察齿轮箱中齿轮的故障信息。 ④通过观察包络信号频谱中不同频率分量的特征(例如齿轮的啮合频率及其边频),可提取每个试验齿轮包络信号频谱中的可相互区别的基本频率特征,然后将这些特征作为神经网络分类器的输入,可用于不同损伤程度,不同类型的齿轮故障的辨别和分类。本文针对齿轮轮齿裂纹和剥落故障进行研究,研究结果表明,通过上述方法能获取较好的诊断结果,证明了上述方法在齿轮故障诊断中的有效性。
[Abstract]:Gear is the most widely used part in mechanical transmission system. With the development of modern science and technology, mechanical equipment becomes more and more precise, which brings difficulties to the condition monitoring of these equipment. Gear box is an important member of mechanical equipment. The monitoring of its working condition must be real-time and effective so that any malfunction in mechanical equipment can be detected and can be repaired in time at an early stage. Envelope analysis has been widely used in fault diagnosis of bearings and gears. Traditional methods require monitoring signals to be processed simultaneously in time domain before band-pass filtering in order to improve the effectiveness of envelope analysis. In practical application, it is very difficult to process the gear vibration signal in time domain synchronization and equalization, and other methods are needed to assist the envelope analysis in gear fault diagnosis. In view of the above problems, a new demodulation method is proposed in this paper. It uses the resonance frequency as the central frequency (usually far from the harmonic component of the gear meshing frequency with large amplitude) and carries on the envelope analysis to realize two types of gear vibration signal. Diagnosis and classification of local faults (crack and spalling) of gears with different degree of damage. Local faults of gears will lead to a pair of meshing gears, which will produce low-amplitude pulse signals during each rotation. It will produce modulation phenomenon with the structural resonance signal of the gear, but the low energy signal is often submerged in the high energy signal generated by other signal sources in the gear box. A method is proposed in this paper. The fault information of gear is obtained by extracting the fault excitation signal from the monitoring signal. The envelope analysis method proposed in this paper is as follows: 1 by observing the resonance components in the frequency spectrum of the gearbox monitoring signal, the suitable demodulation frequency band can be found in order to extract the pulse signal generated by the local fault of the gear for fault diagnosis. (2) the center frequency of the band-pass filter is selected near the structural resonance frequency (usually far from the harmonic component of gear meshing frequency with large amplitude), and the bandwidth of the band-pass filter is selected by observing the spectrum diagram of the original signal. The bandpass band can cover the whole resonance frequency range, which can effectively remove the influence of gear meshing frequency component. 3Demodulation based on Hilbert transform for the band-pass filtered signal. The demodulated envelope signal contains only the components related to the gear fault frequency, and then the envelope signal is transformed by FFT to obtain the spectrum diagram, so that the fault information of the gear in the gear box can be extracted and observed. (4) by observing the characteristics of different frequency components in the envelope signal spectrum (such as gear meshing frequency and its edge frequency), the basic frequency characteristics of each test gear envelope signal spectrum can be extracted. Then, these features are used as input of neural network classifier, which can be used to distinguish and classify different types of gear faults with different damage degree. In this paper, the crack and spalling fault of gear teeth are studied. The results show that the above method can obtain better diagnosis results, and proves the effectiveness of the above methods in gear fault diagnosis.
【学位授予单位】：重庆大学
【学位级别】：硕士
【学位授予年份】：2011
【分类号】：TH165.3

【共引文献】