【作者】 周昊；

【导师】 陈长征；

【作者基本信息】 沈阳工业大学 ， 测试计量技术及仪器， 2014， 博士

【摘要】 近年来，随着人类社会对能源需求的急速增长和日益严重的环境问题，煤、石油、天然气等传统能源所暴露的问题越来越突出。风能作为一种新型可再生能源，以其储量巨大，价格低廉，环境污染小的优势越来越受到人们的重视，使得风电装备也随之得到了迅速的发展。近年来随着我国大力发展风力发电事业，风电机组逐步增多，但随之而来的是风电机组事故频发，对风力发电机的状态监测和故障诊断显得尤为重要。在风电机组的各组成部件中，主轴承是最为重要，也是最容易出现故障的部件之一。而主轴承的工作状态是否正常，将直接影响到整个风电机组的正常运转。因而对风力机主轴承的状态监测和故障诊断显得十分有必要。目前针对风力机主轴承的诊断方法很多，其中最常用的是振动诊断法和声发射诊断法。但由于风力机运行环境经常十分恶劣，在运行过程中，反映其故障状态的特征信息经常淹没在噪声干扰信号之中，有效地提取其故障信息，对风力机主轴承的监测和诊断十分必要。国内外很多学者在这方面做了大量工作，如将专家系统、模糊系统、神经网络、小波变换、Hilbert-Huang变换、Wigner分布、支持向量机等方法应用于风力机主轴承的诊断之中，取得了很多有价值的研究成果，但同时也存在一些问题。鉴于此，本文采用盲源分离理论来探索风力机主轴承振动和声发射故障信号的提取方法，并做了如下工作：第一，介绍了风电技术的发展状况，阐述了课题的研究背景、研究的目的和意义，论述了风力机主轴承振动和声发射诊断的国内外研究现状，并指出本文的思路和采用的研究方法。第二，探讨了盲源分离的基本理论和盲源分离算法，主要阐述了FastICA算法和JADE算法的计算过程，并指出这些算法存在的不足之处。针对盲源分离算法存在的不足，探讨了采用粒子群优化算法对盲源分离过程进行的优化，并比较了各分离算法的性能。第三，建立了基于盲源分离的风力机主轴承振动诊断系统。首先探讨了振动信号的提取方法，认为包络分析对振动信号的提取较为有效，然后分别对转子试验台、风力发电机试验台和实际风力发电机主轴承的振动信号进行了分离，以实现振动故障信号的特征提取。第四，建立了基于盲源分离的风力机主轴承声发射诊断系统。首先探讨了声发射信号的提取方法，认为小波分析对声发射信号的提取较为有效，然后对风力发电机主轴承的声发射信号进行了分离，以实现声发射故障信号的特征提取。第五，根据提取的风力机主轴承信号的特点，采用集成小波神经网络对风力机主轴承进行故障诊断。针对振动信号和声发射信号的特点分别设计子神经网络，并采用决策融合神经网络进行诊断信息融合，提高了诊断效率。并对诊断算法进行了软件实现，增强了诊断方法的实用性。第六，总结本文的主要结论并对相关的研究技术进行了展望。更多还原

【Abstract】 In recent years, along with the rapid increasing demand for energy and seriousenvironmental problem, the problem of coal, oil, natural gas and other traditional energybecome more and more serious. As a new renewable energy, wind energy has been paid moreand more attention for its advantages of huge reserves, low price and less environmentpollution. So the wind power equipment also obtained a rapid development. In recent years,with the vigorous development of wind power industry in China, the number of wind turbinesincrease gradually. Because of the frequent wind turbine accidents, it is particularly importantfor the state monitoring and fault diagnosis of wind power generator. Among all thecomponents of wind turbine, main bearing is the most important, also one of the most prone tofailure part. The working station of main bearing will directly affect the operation of the wholewind turbine. Therefore, condition monitoring and fault diagnosis of main bearing in windturbine is very necessary.At present there are quite a lot of diagnosis methods for main bearing in wind turbine. Themost commonly used methods are vibration diagnosis method and acoustic emission diagnosismethod. Due to the operating environment of wind turbine is often very poor, in the operationprocess, characteristic information which reflects the fault state is often submerged in noisejamming signals. Extracting the fault information effectively is of great importance inmonitoring and diagnosis main bearing of wind turbine. Many domestic and foreign scholarshave made great efforts in this respect. They acquire methods such as expert system, fuzzysystem, neural network, wavelet transform, Hilbert-Huang transform, Wigner distribution,support vector machine in the diagnosis of main bearing in wind turbines, which has mademany valuable research results, but there still exist some problems. In view of this, this paperexplores the extraction method of wind turbine main bearing vibration and acoustic emissionsignals based on blind source separation theory, and the following work has been done:Firstly, the development status of wind power technology is introduced and thebackground, purpose and significance of the research are expounded. The research situation ofvibration and acoustic emission diagnosis for main bearing in wind turbine at home andabroad is discussed, and the ideas and methods in this paper are pointed out. Secondly, the basic theory of blind source separation and the blind source separationalgorithm are discussed, mainly on FastICA algorithm and JADE algorithm, and theshortcomings of these algorithms are pointed out. In view of the existing problems of theseseparation algorithms, this paper discusses the optimization of particle swarm optimizationalgorithm for blind source separation process, and compared the performance of the variousseparation algorithm.Thirdly, vibration fault diagnosis system of wind turbine main bearing based on blindsource separation is established. The extraction method of vibration signals is discussed andenvelope analysis is regarded as an effective method in the extraction of vibration signals. Thevibration signals of rotor test-bed, the wind turbine test-bed and real wind turbine mainbearings are decomposed for the feature abstraction of vibration fault signal.Fourthly, acoustic emission fault diagnosis system of wind turbine main bearing based onblind source separation is established. The extraction method of acoustic emission signals isdiscussed and wavelet analysis is regarded as an effective method in the extraction of acousticemission signals. Then the wind turbine main bearing acoustic emission signals aredecomposed for the feature abstraction of acoustic emission fault signal.Fifthly, according to the characteristics signals of wind turbine main bearing, faultdiagnosis method of wind turbine main bearing based on integrated wavelet neural network isestablished. According to the characteristics of vibration signals and acoustic emission signals,sub neural network are designed and the decision fusion neural network is designed fordiagnosis information fusion, which has improved the efficiency of fault diagnosis. Softwareimplementation for the diagnosis algorithm has enhanced the practicability of diagnosticmethods.Sixthly, summarize the whole texts and prospect the relevant technique..更多还原