机械故障稀疏特征提取及诊断方法研究

发布时间：2018-05-07 07:57

本文选题：谱峭度 + 可调Q因子小波变换　；参考：《武汉科技大学》2016年博士论文

【摘要】：机械设备是现代化生产的重要工具,对其开展运行状态监测及故障诊断对于保障安全化生产具有重要意义。识别早期故障征兆并提取故障特征,是机械设备状态监测及故障诊断的关键。近年来快速发展的信号稀疏表示理论,为基于振动分析的早期故障特征提取及诊断提供了强有力的工具。本论文以国家自然科学基金项目“低速重载机械早期故障稀疏特征识别的研究”为依托,以信号稀疏表示为主要理论工具,围绕机械设备易发生故障部件——轴承和齿轮的早期故障特征提取及诊断开展了深入研究。主要研究内容如下:1.针对早期故障特征易被噪声覆盖而难以准确提取的问题,提出了基于可调Q因子小波变换的早期故障特征提取方法。该方法先利用可调Q因子小波变换对设备的振动信号在不同的Q因子和尺度下分解,以峭度值最大原则确定最佳的Q因子和尺度带,再利用相邻系数降噪方法处理尺度带内的变换系数,最后通过小波逆变换提取故障特征。实验研究表明,该方法可有效提取设备在中速运转下的早期故障特征,相比于传统小波方法,其提取结果噪声更小,包络谱图上故障特征频率更突出。2.针对利用稀疏表示原理对早期故障特征提取时、准确匹配特征成分的字典难以构造的问题,提出了基于字典学习的早期故障稀疏特征提取方法。该方法以故障信号与正常信号的差值为训练信号,利用改进型K均值奇异值分解字典学习算法构造匹配特征成分的字典;在稀疏分解过程中,通过计算每次迭代后逼近信号的峭度值,找出峭度值最大时对应的逼近信号,自适应确定特征成分与噪声成分的稀疏分解分界点。对比实验结果表明,该方法可有效提取设备在低速运转下的早期故障特征,相比于参数化的字典,其提取结果具有更高的精度。3.针对故障部件参数未知的单一故障诊断问题,提出了基于组稀疏分类的故障诊断方法。该方法先将已知故障类型的训练样本和未知故障类型的待测样本转换至频域,利用训练样本的频域系数组合成稀疏分解的字典,再将待测样本的频域系数在该字典上进行组稀疏分解,最后根据各组重构误差的最小值所在的类别确定故障类型。通过故障实验测试,验证了该方法在理论特征频率的未知情况下,可准确诊断出滚动轴承和齿轮的单一故障类型。4.针对故障部件参数未知的复合故障诊断问题,提出了基于小波包系数稀疏分类的故障诊断方法。该方法先对已知各单一故障类型的训练样本进行小波包变换,凭借距离评价参数筛选出具有类别差异的频带,并利用这些频带内的小波包系数构造稀疏分解的字典组,再将待测复合故障类型的测试样本小波包频带系数在对应字典上稀疏分解,通过各组稀疏重构误差最小值所在类别逐一判断复合故障类型。轴承和齿轮的复合故障诊断实验结果验证了该方法的有效性。
[Abstract]:Mechanical equipment is an important tool in modern production. It is of great significance to carry out operation state monitoring and fault diagnosis to ensure safe production. Identifying early fault signs and extracting fault features are the key to condition monitoring and fault diagnosis of machinery and equipment. The rapid development of signal sparse representation theory in recent years provides a powerful tool for early fault feature extraction and diagnosis based on vibration analysis. This thesis is based on the project of National Natural Science Foundation of China "Research on early Fault sparse feature recognition of low Speed heavy haul Machinery", and the signal sparse representation is the main theoretical tool. The early fault feature extraction and diagnosis of bearing and gear are studied in detail. The main research contents are as follows: 1. In order to solve the problem that early fault features are easily covered by noise and difficult to extract accurately, an early fault feature extraction method based on Q-adjustable factor wavelet transform is proposed. In this method, the vibration signal of the equipment is decomposed under different Q factors and scales by using the adjustable Q factor wavelet transform, and the best Q factor and scale band are determined by the principle of maximum kurtosis. The transform coefficients in the scale band are processed by the adjacent coefficient denoising method, and the fault features are extracted by the inverse wavelet transform. The experimental results show that this method can effectively extract the early fault features of the equipment under medium speed operation. Compared with the traditional wavelet method, the result of the method is less noise and the frequency of fault feature on the envelope spectrum is more prominent. 2. In order to solve the problem that it is difficult to construct a dictionary that can accurately match the feature components in early fault feature extraction using sparse representation principle, a dictionary learning based method for early fault sparse feature extraction is proposed. Using the difference between the fault signal and the normal signal as the training signal, the improved K-means singular value decomposition dictionary learning algorithm is used to construct a dictionary that matches the characteristic components. By calculating the kurtosis value of the approximate signal after each iteration, the approximate signal corresponding to the maximum kurtosis value is found, and the sparse decomposition boundary point of the characteristic component and the noise component is determined adaptively. The experimental results show that this method can effectively extract the early fault features of the equipment at low speed. Compared with the parameterized dictionary, the method has a higher precision of .3. To solve the problem of single fault diagnosis with unknown parameters of fault components, a fault diagnosis method based on group sparse classification is proposed. In this method, the training samples of the known fault type and the unknown fault type samples are converted to the frequency domain, and the frequency-domain coefficients of the training samples are combined into a sparse decomposed dictionary. Then the frequency domain coefficients of the samples to be tested are decomposed in the dictionary and the fault type is determined according to the category of the minimum reconstruction error of each group. Through the fault test, it is proved that the method can accurately diagnose the single fault type of rolling bearing and gear in the case of unknown theoretical characteristic frequency. A fault diagnosis method based on sparse classification of wavelet packet coefficients is proposed for composite fault diagnosis with unknown parameters of fault components. In this method, the training samples of each known single fault type are first transformed by wavelet packet transform, and the frequency bands with different categories are screened by the distance evaluation parameters, and the sparse decomposition dictionary group is constructed by using the wavelet packet coefficients in these bands. Then the wavelet packet frequency band coefficients of the test samples are decomposed in the corresponding dictionaries, and the types of composite faults are judged by the categories in which the minimum error of each group of sparse reconstruction is located. The experimental results of bearing and gear fault diagnosis show that the method is effective.
【学位授予单位】：武汉科技大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：TH17

【相似文献】