基于稀疏分解及图像稀疏表征的滚动轴承微弱故障诊断
发布时间：2018-11-22 07:44
【摘要】：作为国民经济支柱型企业中的关键大型现代化旋转机械设备,对其工作及运行环境苛刻性的要求越来越高,同时对保证其长期安全运行的监测机制要求也愈来愈高。设备能否安全运行不仅牵涉到企业经济利益,而且关系到操作设备员工的生命财产的安全保证与否。能否有效提取出旋转机械的微弱故障特征,进而制定有效的针对治理措施以确保设备的安全高效运行显得尤为重要。作为旋转机械中广泛应用的零部件-滚动轴承,其安全运行与否往往决定着整个设备能否安全运行。对滚动轴承进行有效、及时的故障诊断有着非常重要的安全及经济意义。然而,实际工程应用中滚动轴承的故障特征往往表现得非常微弱,究其原因无异于以下三种情况:采集路径较长以致信号衰减严重;早期微弱故障阶段及其他噪源干扰严重;复合故障状态下。研究上述三种情况下滚动轴承的故障诊断方法有着重要的实际工程应用价值及安全经济意义。滚动轴承发生故障时其振动信号往往呈现出非高斯、非平稳及非线性特性,传统的信号处理方法不能再有效提取出滚动轴承发生故障时的非线性、非高斯特征。稀疏分解方法是一种能有效匹配滚动轴承发生故障时冲击信号特征的处理方法,并在滚动轴承的故障诊断中已经取得了一定应用。基于此,本文在经典稀疏分解方法的基础上提出改进方法对滚动轴承微弱故障诊断进行深入的理论及实验研究;借鉴图像非负矩阵分解处理的思想,将非负矩阵分解方法与稀疏分解的思想相结合,提出基于双谱图像稀疏性非负矩阵分解的滚动轴承复合故障诊断方法。论文的主要内容包括以下几个方面:(1)首先以旋转机械微弱故障特征提取所面临的理论及实际问题为出发点,阐述本学位论文的研究背景及其相关重要意义。总结近年来关于机械设备的相关故障诊断方法、智能诊断方法以及图像稀疏表征等方面的国内外研究现状并分析所总结方法的利弊,确立论文研究内容。(2)详细介绍了稀疏分解的基本思想、基础数学理论、常用的典型求解算法、稀疏性度量及冗余字典的构建等内容;简要介绍基于稀疏分解思想的图像稀疏表征的发展历程,并对图像稀疏表征的多尺度几何分析方法作以详细的介绍。此章节的内容为后续章节具体研究内容奠定坚实的理论支撑。(3)实际工程应用中,某些设备在安装振动传感器时由于受实际条件的限制,造成信号采集路径较长(传感器安装位置所采集到的振动信号离实际故障振源比较远)以致信号衰减严重及受背景噪声影响比较大,直接对此工况下采集到的信号进行故障特征提取很难取得好的效果。最小熵解卷积(Minimum Entropy De-convolution,MED)方法有效减弱了采集路径信号衰减的影响,能有效突出滚动轴承发生故障时的瞬态冲击成份;稀疏分解算法能用最佳的原子去有效的匹配滚动轴承发生故障时的瞬态冲击成份。将二者的优点相结合用于滚动轴承的微弱故障特征提取,提出基于MED-稀疏分解的滚动轴承微弱故障诊断方法,通过仿真和实验验证了所述方法的有效性及优点。并比较了所述方法相对于小波分析方法、总体经验模态分解方法、时频切片小波变换方法及基于谱峭度处理方法的优点。(4)共振稀疏分解方法是一种基于多字典库的稀疏分解方法,可以同时分解出信号中的瞬态冲击成分及其持续震荡成分(工频及其谐频成分)。该方法在EEMD前处理基础上,对分解后峭度指标最大的固有模态函数分量进行共振稀疏分解分析:根据共振稀疏分解中信号品质因子的定义,分别构建高、低品质因子小波基函数字典库、并利用形态学分析方法建立信号稀疏表示的目标函数进而实现对滚动轴承发生早期微弱故障或受其他高品质因子噪源干扰严重时具有低品质因子的瞬态故障成份及其他持续振荡高品质因子噪声成份的成功分离。(5)滚动轴承发生复合故障时,由于不同部位故障信号之间的相互干扰及耦合效应,复合故障信号表现得非常复杂,基于信号处理的滚动轴承复合故障方法往往难以取得好的效果。双谱三维图像信息比单纯频谱蕴含更多故障信息,适用于滚动轴承复合故障特征提取。但是如何有效精炼的提取三维图谱的特征以实现智能诊断是一个亟需解决的问题。基于此,将图像非负矩阵分解与稀疏分解的思想相结合,提出稀疏性非负矩阵分解方法对双谱三维图像进行有效特征提取进而实现滚动轴承复合故障的高效智能诊断。最后并与基于双谱图像非负矩阵分解的特征提取效果作以对比突出了所述方法的优越性。
[Abstract]:As the key large-scale modern rotating machinery in the national economy pillar-type enterprise, the requirement for its working and operating environment is higher and higher, and the monitoring mechanism for ensuring its long-term safe operation is getting higher and higher. Whether the equipment can operate safely involves not only the economic benefits of the enterprise, but also the safety guarantee of the life and property of the employees of the operation equipment. It is very important to effectively extract the weak fault features of the rotating machinery and to develop effective measures to ensure the safe and efficient operation of the equipment. As a part-rolling bearing widely used in the rotating machinery, the safe operation of the rolling bearing often determines whether the whole equipment can operate safely. It has very important safety and economic significance for the effective and timely fault diagnosis of rolling bearing. However, the fault characteristics of rolling bearing in practical engineering application are often very weak, the reason is as follows: the acquisition path is longer so that the signal attenuation is serious; the early weak failure stage and other noise source interference are serious; and in the composite fault condition. The method of fault diagnosis of rolling bearing is of great value and safety and economic significance in three cases. The non-Gaussian, non-stationary and non-linear characteristics of the rolling bearing are often presented in the fault of the rolling bearing, and the traditional signal processing method can not extract the non-linear and non-Gaussian features of the rolling bearing failure. The sparse decomposition method is a kind of treatment method which can effectively match the characteristics of the impact signal when the rolling bearing is in fault, and has obtained some application in the fault diagnosis of the rolling bearing. On the base of the classical sparse decomposition method, this paper makes an in-depth theoretical and experimental study on the weak fault diagnosis of the rolling bearing, and combines the non-negative matrix decomposition method with the idea of sparse decomposition based on the thought of the non-negative matrix decomposition processing of the image. The invention provides a rolling bearing composite fault diagnosis method based on a double-spectrum image sparse non-negative matrix decomposition. The main contents of the thesis include the following aspects: (1) Firstly, the thesis starts with the theory and practical problems of the weak fault feature extraction of the rotating machinery, and expounds the research background and relevant significance of the thesis. The present situation of relevant fault diagnosis method, intelligent diagnosis method and image sparse representation of mechanical equipment in recent years are summarized, and the advantages and disadvantages of the summarized methods are analyzed and the research contents of the paper are established. (2) The basic idea of the sparse decomposition, the basic mathematical theory, the typical solution algorithm, the sparsity measure and the construction of the redundant dictionary are introduced in detail, and the development course of the sparse representation of the image based on the sparse decomposition is briefly introduced. and the multi-scale geometric analysis method for sparse representation of the images is described in detail. The contents of this section provide a solid theoretical support for the specific study of the subsequent sections. (3) In practical engineering applications, certain equipment is limited by the actual conditions when the vibration sensor is installed, The signal acquisition path is long (the vibration signal acquired by the sensor installation position is far from the actual fault source) so that the signal attenuation is serious and the influence of the background noise is relatively large, and the signal that is collected directly under the condition is difficult to obtain good effect. The Minimum Entropy De-convolute (MED) method effectively reduces the influence of the signal attenuation of the acquisition path, and can effectively highlight the transient impact component in the fault of the rolling bearing; the sparse decomposition algorithm can effectively match the transient impact component in the fault of the rolling bearing with the best atom. The advantages of the method are verified by simulation and experiment by combining the advantages of the two with the weak fault feature extraction of the rolling bearing. The method based on the wavelet analysis method, the general empirical mode decomposition method, the time-frequency slice small-wave transformation method and the spectral kurtosis processing method are compared and compared. and (4) the resonance sparse decomposition method is a sparse decomposition method based on a multi-dictionary library, and can simultaneously decompose the transient impact component in the signal and the continuous oscillation component (power frequency and the harmonic frequency component thereof). according to the definition of the signal quality factor in the resonance sparse decomposition, a high and low quality factor wavelet basis function dictionary library is respectively constructed, and a morphological analysis method is used to establish a target function of the signal sparse representation, so as to realize the success of the transient fault component with low quality factor and other continuous oscillation high-quality factor noise components in the early-stage weak fault of the rolling bearing or the interference of other high-quality factor noise sources. It's out of here. (5) The composite fault signal is very complex due to the mutual interference and coupling effect between the fault signals of different parts, and the composite fault method of the rolling bearing based on the signal processing is often difficult to achieve. The double-spectrum three-dimensional image information contains more fault information than the pure frequency spectrum, and is suitable for the feature extraction of the composite fault of the rolling bearing. But how to extract the characteristics of the three-dimensional map effectively to realize the intelligent diagnosis is an urgent problem to be solved. On the basis of this, combining the non-negative matrix decomposition of the image with the idea of sparse decomposition, a sparse non-negative matrix decomposition method is proposed to carry out effective feature extraction on the double-spectrum three-dimensional image, so as to realize the high-efficiency intelligent diagnosis of the composite fault of the rolling bearing. Finally, the advantages of the method are compared with the feature extraction effect based on the non-negative matrix decomposition of the dual-spectrum image.
【学位授予单位】：上海交通大学
【学位级别】：博士
【学位授予年份】：2015
【分类号】：TH133.33
本文编号：2348550
[Abstract]:As the key large-scale modern rotating machinery in the national economy pillar-type enterprise, the requirement for its working and operating environment is higher and higher, and the monitoring mechanism for ensuring its long-term safe operation is getting higher and higher. Whether the equipment can operate safely involves not only the economic benefits of the enterprise, but also the safety guarantee of the life and property of the employees of the operation equipment. It is very important to effectively extract the weak fault features of the rotating machinery and to develop effective measures to ensure the safe and efficient operation of the equipment. As a part-rolling bearing widely used in the rotating machinery, the safe operation of the rolling bearing often determines whether the whole equipment can operate safely. It has very important safety and economic significance for the effective and timely fault diagnosis of rolling bearing. However, the fault characteristics of rolling bearing in practical engineering application are often very weak, the reason is as follows: the acquisition path is longer so that the signal attenuation is serious; the early weak failure stage and other noise source interference are serious; and in the composite fault condition. The method of fault diagnosis of rolling bearing is of great value and safety and economic significance in three cases. The non-Gaussian, non-stationary and non-linear characteristics of the rolling bearing are often presented in the fault of the rolling bearing, and the traditional signal processing method can not extract the non-linear and non-Gaussian features of the rolling bearing failure. The sparse decomposition method is a kind of treatment method which can effectively match the characteristics of the impact signal when the rolling bearing is in fault, and has obtained some application in the fault diagnosis of the rolling bearing. On the base of the classical sparse decomposition method, this paper makes an in-depth theoretical and experimental study on the weak fault diagnosis of the rolling bearing, and combines the non-negative matrix decomposition method with the idea of sparse decomposition based on the thought of the non-negative matrix decomposition processing of the image. The invention provides a rolling bearing composite fault diagnosis method based on a double-spectrum image sparse non-negative matrix decomposition. The main contents of the thesis include the following aspects: (1) Firstly, the thesis starts with the theory and practical problems of the weak fault feature extraction of the rotating machinery, and expounds the research background and relevant significance of the thesis. The present situation of relevant fault diagnosis method, intelligent diagnosis method and image sparse representation of mechanical equipment in recent years are summarized, and the advantages and disadvantages of the summarized methods are analyzed and the research contents of the paper are established. (2) The basic idea of the sparse decomposition, the basic mathematical theory, the typical solution algorithm, the sparsity measure and the construction of the redundant dictionary are introduced in detail, and the development course of the sparse representation of the image based on the sparse decomposition is briefly introduced. and the multi-scale geometric analysis method for sparse representation of the images is described in detail. The contents of this section provide a solid theoretical support for the specific study of the subsequent sections. (3) In practical engineering applications, certain equipment is limited by the actual conditions when the vibration sensor is installed, The signal acquisition path is long (the vibration signal acquired by the sensor installation position is far from the actual fault source) so that the signal attenuation is serious and the influence of the background noise is relatively large, and the signal that is collected directly under the condition is difficult to obtain good effect. The Minimum Entropy De-convolute (MED) method effectively reduces the influence of the signal attenuation of the acquisition path, and can effectively highlight the transient impact component in the fault of the rolling bearing; the sparse decomposition algorithm can effectively match the transient impact component in the fault of the rolling bearing with the best atom. The advantages of the method are verified by simulation and experiment by combining the advantages of the two with the weak fault feature extraction of the rolling bearing. The method based on the wavelet analysis method, the general empirical mode decomposition method, the time-frequency slice small-wave transformation method and the spectral kurtosis processing method are compared and compared. and (4) the resonance sparse decomposition method is a sparse decomposition method based on a multi-dictionary library, and can simultaneously decompose the transient impact component in the signal and the continuous oscillation component (power frequency and the harmonic frequency component thereof). according to the definition of the signal quality factor in the resonance sparse decomposition, a high and low quality factor wavelet basis function dictionary library is respectively constructed, and a morphological analysis method is used to establish a target function of the signal sparse representation, so as to realize the success of the transient fault component with low quality factor and other continuous oscillation high-quality factor noise components in the early-stage weak fault of the rolling bearing or the interference of other high-quality factor noise sources. It's out of here. (5) The composite fault signal is very complex due to the mutual interference and coupling effect between the fault signals of different parts, and the composite fault method of the rolling bearing based on the signal processing is often difficult to achieve. The double-spectrum three-dimensional image information contains more fault information than the pure frequency spectrum, and is suitable for the feature extraction of the composite fault of the rolling bearing. But how to extract the characteristics of the three-dimensional map effectively to realize the intelligent diagnosis is an urgent problem to be solved. On the basis of this, combining the non-negative matrix decomposition of the image with the idea of sparse decomposition, a sparse non-negative matrix decomposition method is proposed to carry out effective feature extraction on the double-spectrum three-dimensional image, so as to realize the high-efficiency intelligent diagnosis of the composite fault of the rolling bearing. Finally, the advantages of the method are compared with the feature extraction effect based on the non-negative matrix decomposition of the dual-spectrum image.
【学位授予单位】：上海交通大学
【学位级别】：博士
【学位授予年份】：2015
【分类号】：TH133.33
【参考文献】
相关期刊论文 前1条
1 ;Compressed sensing SAR imaging based on sparse representation in fractional Fourier domain[J];Science China(Information Sciences);2012年08期
，本文编号：2348550
本文链接：https://www.wllwen.com/jixiegongchenglunwen/2348550.html
