基于协方差矩阵流形的风电机组齿轮箱故障诊断方法研究

发布时间：2018-02-04 16:59

本文关键词： 风电机组故障诊断黎曼流形协方差矩阵流形熵　出处：《哈尔滨工业大学》2014年博士论文　论文类型：学位论文

【摘要】：由于能源危机和环境问题日益严重,风能作为绿色无污染能源已受到世界各国的关注和重视。随着装机规模的不断扩大,及时发现风电机组故障并进行维修变得越来越重要。齿轮箱是风电机组中故障率较高的部件,振动信号分析是齿轮箱故障诊断中最常用的技术。风电机组齿轮箱振动信号成分复杂,具有较强的背景噪声和明显的非平稳性。研究可以适应振动信号非平稳性、非高斯特征的新方法是提高其故障诊断技术水平的关键。因此,开展对风电机组齿轮箱振动信号的特征提取、故障检测与故障分类等方法研究,对于保证风电机组正常安全运行具有重大的实际意义。本文在对风电机组齿轮箱进行振动测试的基础上,针对传统振动信号分析方法往往仅对单通道信号分别提取,无法提取多通道振动信号各通道间关联和结构信息的问题,提出了一种基于协方差矩阵流形分析的齿轮箱故障诊断方法。主要研究内容如下:首先,综述了风电机组齿轮箱故障诊断的基本理论和方法,阐述了风电机组的基本组成,分析了国内外研究现状,介绍了常用的振动信号特征提取方法,在分析多通道振动信号的基础上,给出了基于多通道振动信号的协方差矩阵流形表示方法。其次,针对风电机组齿轮箱多通道振动信号的关联和结构特征提取问题,提出了一种基于协方差矩阵流形表示的椭球可视化故障诊断方法。该方法将多通道振动时间序列转化为一个协方差矩阵序列进行处理,可以有效表达各通道间的关联结构信息。对协方差矩阵通过奇异值分解转化为椭球和四元数表示,可以实现齿轮箱振动状态的可视化,然后进行各向异性分析,从而有利于理解振动信号特点并进行故障诊断。再次,在多通道振动信号协方差矩阵流形表示的基础上,针对齿轮箱振动信号非平稳性、非高斯的特点,提出了一种基于协方差矩阵流形黎曼距离的故障检测与定位方法。该方法用协方差矩阵流形作为描述子,以黎曼距离作为相似性度量,结合统计过程控制图来实现对齿轮箱故障的检测与定位。通过实验测试与分析,该方法不仅可以有效检测多通道振动信号间的相关性,而且在故障检测准确率和算法复杂度上均具有一定的性能优势。最后,针对风电机组齿轮箱故障分类问题,基于振动信号的协方差矩阵流形表示和传统的多元多尺度熵算法,提出了一种多尺度流形熵故障分类方法。该方法将传统的多元多尺度熵推广到协方差矩阵上的多尺度流形熵,解决了传统方法计算量大、难以量化各通道间相关性的问题,实现了齿轮箱振动信号的复杂性和相关性特征的有效提取。在故障数据上进行验证,结果表明该方法可以有效地降低计算复杂度,同时避免了传统时频分析中高频信号的干扰,具有较高的故障分类识别率。
[Abstract]:As a result of energy crisis and environmental problems, wind energy as a green non-pollution energy has been paid attention to by many countries all over the world. With the expansion of the scale of installation, wind energy has been paid more and more attention. It is becoming more and more important to find wind turbine faults and maintain them in time. The gearbox is the component with high failure rate in wind turbine. Vibration signal analysis is the most commonly used technology in gear box fault diagnosis. It has strong background noise and obvious non-stationarity. The study can adapt to the non-stationary vibration signal. The new method of non-#china_person0# feature is the key to improve the technical level of fault diagnosis. The methods of feature extraction, fault detection and fault classification of wind turbine gearbox vibration signal are carried out. It is of great practical significance to ensure the normal and safe operation of wind turbine units. This paper is based on the vibration test of wind turbine gearbox. To solve the problem that the traditional vibration signal analysis method usually only extracts the single channel signal separately, it can not extract the correlation and structure information among the multi-channel vibration signals. A method of gearbox fault diagnosis based on covariance matrix manifold analysis is proposed. The main research contents are as follows: firstly, the basic theory and method of gearbox fault diagnosis for wind turbine are summarized. This paper expounds the basic composition of wind turbine, analyzes the present research situation at home and abroad, and introduces the commonly used vibration signal feature extraction methods, on the basis of analyzing the multi-channel vibration signal. The covariance matrix manifold representation method based on multi-channel vibration signal is presented. Secondly, the correlation of multi-channel vibration signal and the extraction of structural features of wind turbine gearbox are discussed. An ellipsoidal visual fault diagnosis method based on covariance matrix manifold representation is proposed, which transforms the multi-channel vibration time series into a covariance matrix sequence for processing. The covariance matrix can be transformed into ellipsoid and quaternion representation by singular value decomposition, which can visualize the vibration state of the gearbox and then analyze the anisotropy. This is helpful to understand the characteristics of vibration signal and fault diagnosis. Thirdly, on the basis of multi-channel vibration signal covariance matrix manifold representation, aiming at the non-stationary vibration signal of the gearbox, non-#china_person0# characteristics. A fault detection and location method based on Riemannian distance of covariance matrix manifold is proposed, in which covariance matrix manifold is used as descriptor and Riemannian distance as similarity measure. Combined with the statistical process control chart, the gearbox fault detection and location can be realized. Through the experimental test and analysis, the method can not only effectively detect the correlation between the multi-channel vibration signals. And in fault detection accuracy and algorithm complexity has certain performance advantages. Finally, for wind turbine gearbox fault classification problem. Covariance matrix manifold representation based on vibration signal and traditional multi-scale entropy algorithm. A multi-scale manifold entropy fault classification method is proposed, which extends the traditional multi-scale entropy to the multi-scale manifold entropy on the covariance matrix. It is difficult to quantify the correlation between the channels, which can effectively extract the complexity and correlation characteristics of the gearbox vibration signal, and verify the fault data. The results show that this method can effectively reduce the computational complexity and avoid the interference of high frequency signals in the traditional time-frequency analysis. It has a high fault classification and recognition rate.
【学位授予单位】：哈尔滨工业大学
【学位级别】：博士
【学位授予年份】：2014
【分类号】：TH165.3

【参考文献】