基于形态分量分析的变工况齿轮箱故障诊断研究

发布时间：2018-06-17 00:46

本文选题：形态分量分析 + 字典学习　；参考：《石家庄铁道大学》2017年硕士论文

【摘要】：在机械传动系统中,齿轮箱是故障多发的部件。在变工况条件下,传统故障诊断方法难以对齿轮箱故障进行准确检测和有效识别。通过搭建实验台采集变工况条件下的齿轮箱振动信号来研究其信号特点,验证了齿轮箱在变工况运行时信号表现为转速和负载的变化,且振动信号是非线性的多分量信号等特点。信号的非线性多分量特点使故障特征难以提取;转速的不断变化会在频谱中出现频率“涂抹”现象,难以确定特征频率;负载变化导致正常齿轮与故障齿轮信号的特征相似,故障难以识别。针对以上问题本论文开展了以下研究工作:首先,在基于固定字典的形态分量分析(Morphological component analysis,MCA)方法基础上,提出了基于字典学习的形态分量分析方法用于滚动轴承故障特征提取。分别以轴承内圈、外圈故障信号为训练字典的样本,应用K-SVD(K-singular value decomposition)字典学习算法对样本进行字典学习,找到最优的字典空间;将学习到的字典替代MCA中的固定字典,根据信号所包含各成分的形态差异性,利用MCA对滚动轴承故障信号中的内、外圈故障特征和噪声成分进行分离;根据包络后特征分量的频谱图分析和诊断轴承的故障。其次,分别使用基于数字微分的计算转速曲线方法和线性调频小波路径追踪(Chirplet Path Pursuit,CPP)的瞬时频率估计方法获取振动信号中的转速信息;对转速曲线积分得到时间转角函数,根据阶次跟踪理论确定角度重采样阶次和重采样角度,通过对时间转角函数插值计算等角度采样的时间序列;使用计算出的时间序列对原信号进行样条曲线插值,将非平稳信号转化为角域的平稳信号,解决变转速引起的频率“涂抹”问题。最后,提出基于形态分量分析的变工况齿轮箱故障诊断方法。先使用阶次跟踪方法将非平稳信号转化为角域平稳信号,然后使用形态分量分析方法从角域平稳信号中分离出冲击、简谐分量与噪声成分,提取齿轮箱非线性、多分量信号中的故障特征,再对冲击分量做角域平均突出故障特征,最后进行瞬时功率谱(Instaneous power spectrum,IPS)分析识别齿轮是否存在故障。本文形态分量分析方法在选取字典时使用了字典学习的方法,从仿真和实验的结果可知,学习到的字典比固定字典更能匹配复杂信号的结构特征;针对齿轮箱变转速、变负载和振动信号是非线性、多分量信号的特点,将形态分量分析方法,阶次跟踪方法和瞬时功率谱有机的结合起来,实现了变工况情况下的齿轮箱故障诊断。
[Abstract]:In mechanical transmission system, gearbox is a fault-prone component. Under the condition of variable working conditions, the traditional fault diagnosis method is difficult to detect and identify the gearbox fault accurately. The characteristics of the vibration signals of the gearbox are studied by setting up an experimental bench to collect the vibration signals of the gearbox under the condition of variable working conditions. It is verified that the signals of the gearbox under the variable working conditions show the variation of the speed and the load. And the vibration signal is nonlinear multi-component signal and so on. The nonlinear multi-component characteristic of the signal makes it difficult to extract the fault feature, the frequency "smear" appears in the frequency spectrum with the constant change of rotational speed, the characteristic frequency is difficult to be determined, and the characteristics of the normal gear signal and the fault gear signal are similar to each other due to the load change. The fault is difficult to identify. In order to solve the above problems, the following research work has been carried out in this paper: firstly, based on morphological component analysis (MCA) method based on fixed dictionary, a dictionary learning-based morphological component analysis method is proposed for fault feature extraction of rolling bearing. Taking the fault signals of bearing inner ring and outer ring as the samples of training dictionary, the dictionary learning algorithm of K-SVD / K-singular value decompositionis used to study the dictionary to find the best dictionary space, to replace the fixed dictionary in MCA. According to the morphological difference of each component contained in the signal, MCA is used to separate the inner and outer ring fault features from the noise component of the rolling bearing fault signal, and the fault of the bearing is analyzed and diagnosed according to the spectrum diagram of the envelope characteristic component. Secondly, the rotational speed information in vibration signal is obtained by using the method of calculating rotational speed curve based on digital differential and the instantaneous frequency estimation method of chirplet path pursuit CPP based on linear frequency modulation wavelet path tracing, and the time rotation angle function is obtained by integrating the rotational speed curve. According to the order tracking theory, the order of angle resampling and the angle of resampling are determined, the time series are calculated by interpolation of time rotation function, and the original signal is interpolated by spline curve with the calculated time series. The non-stationary signal is transformed into the stationary signal in the angle domain to solve the problem of frequency smear caused by variable rotational speed. Finally, a fault diagnosis method based on morphological component analysis is proposed. The order tracking method is used to transform the non-stationary signal into the stationary signal in the angular domain, and then the impact, harmonic and noise components are separated from the stationary signal in the angular domain by using the morphological component analysis method, and the nonlinearity of the gear box is extracted. The fault features in multi-component signals are analyzed to identify whether the gears have faults or not. Then the angular mean outburst feature of the impulse components is made. Finally, the instantaneous power spectrum is used to identify the faults of gears. In this paper, the dictionary learning method is used in the selection of dictionaries. The results of simulation and experiments show that the dictionaries can match the structural characteristics of complex signals better than the fixed dictionaries. Variable-load and vibration signals are nonlinear and multi-component signals. By combining morphological component analysis method, order tracking method and instantaneous power spectrum, the gearbox fault diagnosis is realized.
【学位授予单位】：石家庄铁道大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TH132.41

【相似文献】