基于改进的局部均值分解方法在齿轮故障诊断中的应用研究

发布时间：2018-02-03 20:11

本文关键词： 齿轮故障局部均值分解卡尔曼滤波支持向量机稀疏表示　出处：《重庆三峡学院》2017年硕士论文　论文类型：学位论文

【摘要】：随着齿轮在旋转机械中所占比重的增加,齿轮的研究也显得愈加重要。齿轮作为重要的传动原件之一,在处于高负荷工作状态下,极易出现故障,影响生产的正常进程,严重时会导致整个系统瘫痪,引发安全问题和带来巨大经济损失。齿轮的主要故障形式为:磨损、断齿、点蚀、胶合等,当齿轮出现这些故障时,会引起相应的振动信号幅值和相位变化,产生幅值和相位的调制。对于振动信号的准确解调是齿轮故障研究的一个重点。在振动信号处理上,局部均值分解方法可以将多分量信号分解成多个时域分量和频域分量,将这些分量重组可以得到信号的完整时频分布,非常适合于非线性、非平稳信号的处理。但对于振动信号存在强噪声或信号长度过大时,会影响局部均值分解的计算,甚至会给诊断结果带来误差。因此,本文在运用局部均值分解方法进行故障诊断的基础上,结合卡尔曼滤波、支持向量机和稀疏表示方法的优势,进行振动信号的降噪、故障分类和信号的压缩,以提高齿轮故障诊断的准确性、故障识别率,并缩短诊断时间。全文主要结论概况如下:(1)振动信号中包含了大量的噪声,容易影响故障诊断结果。结合卡尔曼滤波的局部齿轮故障诊断方法,对振动信号进行降噪处理,对比降噪前后的频谱图,可以看出降噪后故障特征数量由一个增加到了三个,且特征频率更为明显,说明该方法能有效减小噪声对齿轮信号诊断的影响。(2)齿轮故障类型的识别对系统进行故障诊断具有重要作用。结合局部均值分解和支持向量机的方法,对齿轮故障进行分类判别。实验研究表明本文方法能对磨损故障和断齿故障进行有效的分类,对比经验模态分解和支持向量机的齿轮故障分类法,故障识别率明显提高,准确度几乎可达到100%。(3)采用基于匹配追踪和局部均值的齿轮故障诊断方法,创建Gabor原子库,对信号进行稀疏表示,结合谱峭度原则选取最佳分量,进行频谱分析,结果表明该方法能缩短信号重构时间,加快运行速度,准确提取出故障的特征频率。(4)采用基于正交匹配追踪和局部均值分解的齿轮故障诊断方法,对比信号重构前后的效果图看出,本文方法提取的故障特征数量明显增加,说明在信号重构的过程中剔除了干扰信息,明显提高诊断的准确性。
[Abstract]:With the increase of the proportion of gear in rotating machinery, the research of gear becomes more and more important. As one of the important parts of transmission, the gear is prone to malfunction under the condition of high load. Affecting the normal process of production, serious will lead to paralysis of the whole system, causing safety problems and huge economic losses. The main failure forms of gears are: wear, tooth breakage, pitting, gluing and so on. When the gear has these faults, it will cause the corresponding vibration signal amplitude and phase change. The modulation of amplitude and phase is produced. The accurate demodulation of vibration signal is an important point in the research of gear fault. The local mean decomposition method can decompose the multi-component signal into multiple time-domain components and frequency-domain components. The complete time-frequency distribution of the signal can be obtained by recombination of these components, which is very suitable for nonlinear. But when the vibration signal has strong noise or the signal length is too large, it will affect the calculation of local mean decomposition and even bring error to the diagnosis result. In this paper, based on the local mean decomposition method for fault diagnosis, combined with the advantages of Kalman filter, support vector machine and sparse representation method, the vibration signal noise reduction, fault classification and signal compression are carried out. In order to improve the accuracy of gear fault diagnosis, fault recognition rate, and shorten the diagnosis time. The main conclusions of the paper are as follows: 1) the vibration signal contains a lot of noise. It is easy to affect the result of fault diagnosis. Combined with the Kalman filter method of local gear fault diagnosis, the vibration signal is de-noised, and the spectrum before and after noise reduction is compared. It can be seen that the number of fault features increases from one to three after noise reduction, and the feature frequency is more obvious. It shows that this method can effectively reduce the influence of noise on gear signal diagnosis. (2) the identification of gear fault type plays an important role in fault diagnosis of the system. The method of local mean decomposition and support vector machine is combined. The experimental results show that this method can effectively classify the wear fault and the broken gear fault, and compare the empirical mode decomposition with the support vector machine. The fault recognition rate is improved obviously, and the accuracy can almost reach 100%. The gear fault diagnosis method based on matching tracing and local mean value is adopted to create Gabor atomic library to represent the signal sparsely. According to the principle of spectral kurtosis, the optimal component is selected and the spectrum analysis is carried out. The results show that this method can shorten the time of signal reconstruction and speed up the operation. The characteristic frequency of fault is extracted accurately. (4) the gear fault diagnosis method based on orthogonal matching tracing and local mean decomposition is adopted, and the results before and after signal reconstruction are compared. The number of fault features extracted by this method is obviously increased, which shows that the interference information is eliminated in the process of signal reconstruction, and the accuracy of diagnosis is improved obviously.
【学位授予单位】：重庆三峡学院
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TH132.41

【参考文献】