【作者】 刘文艺；

【导师】 汤宝平；

【作者基本信息】 重庆大学 ， 机械电子工程， 2010， 博士

【摘要】 随着风力发电机组(以下简称风电机组、风机)的单机容量越来越大,装机量也逐年增加,相关的第三产业即风电机组运行维护、监测、故障诊断等将成为行业新的增长点。而风电机组的工作环境恶劣,风速有很高的不稳定性,在交变负载的作用下,机组的传动系统等部件最容易损坏,而风电机组又安装在偏远地区且距地面甚高,维修不便,风电机组的状态监测和故障诊断在这种情况下具有重要的意义。研究表明风电机组传动系统的振动故障在常见故障中占有较高的比重,对传动系统进行状态监测是至关重要的。目前大型商业化的风电机组自带有监测控制和数据获取(SCADA)系统,为提高风电场运行的稳定性和可靠性提供了强有力的技术平台和支撑,但是SCADA系统缺乏对传动系统的振动监测及相关分析。而目前SKF公司针对风电机组振动故障监测的系统虽然具有预警及报警功能,却缺乏精确故障诊断等功能。因而论文选择传动系统为重点监测对象,开展针对传动系统的振动监测与故障诊断研究。论文的主要工作如下:①论文分析了传动系统叶轮、齿轮箱、发电机等机构及其常见振动故障特点。风电机组传动系统振动信号受到背景噪声的干扰,且齿轮、轴承等旋转部件故障信号多呈现周期非平稳特性。同时,由于风速的不平稳性,交变载荷作用到传动系统,又使得振动信号呈现高斯噪声混杂及非线性特性。根据对传动系统关键部件的故障及其时频域特点分析,确定了风电机组传动系统的监测点的位置,据此可以设置相关的振动传感器采集关键部件的振动信号,对传动系统进行振动监测。②在风电机组振动信号预处理方面,提出了一种交叉验证优化Morlet小波参数的消噪方法。风电机组工作环境恶劣,振动监测获取的数据多包含强烈的背景噪声,传统的滤波消噪很难将噪声和有用成分在时频域区分开,小波消噪在此种情况下有较优的分析效果。但小波消噪方法存在小波基选取和分解层数确定、阈值方法确定等问题,论文分析了传统硬阈值方法和软阈值方法的局限并提出了一种自适应阈值小波消噪方法,进一步提出了交叉验证优化Morlet小波参数的消噪方法。利用交叉验证法对改进Morlet小波的参数进行优化选择并确定了最佳尺度。通过齿轮振动信号实例并对比了传统的小波消噪方法,证明了该方法具有较好的消噪效果。③针对风电机组振动信号的周期非平稳性特点,提出了一种基于自项窗抑制魏格纳分布(WVD,Wigner-Ville distribution)交叉项的故障诊断方法。由于风电机组旋转部件振动信号的周期非平稳性特点,单纯的时域或频域分析方法很难取得理想的效果,时频分析是分析此类信号的有效方法,其中WVD的时频分辨率和能量聚集性具有无可比拟的优势,可用于旋转机械的特征提取及故障诊断。但由于WVD具有交叉项的干扰,需要寻找合适的方法对其交叉项进行抑制。论文在研究WVD自项和交叉项相互关系的基础上提出了一种阈值自适应STFT(ASTFT,Adaptive Short-Time Fourier Transform)故障诊断方法,进而针对STFT自身分辨率较低的缺陷,提出了一种基于自项窗WVD的故障诊断方法。设计了自适应自项窗函数,并用其替代自项对WVD进行加窗处理,可以有效地抑制交叉项,还能够使自项能量很紧密地集中在各分量瞬时频率的附近。通过仿真分析和振动信号分析验证了该方法具有较好的特征提取及故障诊断效果。④针对风电机组振动信号非高斯非线性特点,提出了一种模糊高阶谱故障诊断方法。该方法既可以消除振动信号中混叠的高斯噪声,同时可以很好地分析非线性特性振动信号,实现正确的故障诊断。论文首先利用双谱估计方法分析了不同类别下的滚动轴承的振动信号,研究表明了双谱分布区域信息与故障类别间存在映射关系,且这种映射关系不受工作转频影响。对双谱估计特征值进行阈值化处理,并在此基础上构造由核图、域图构造的目标模板,通过测试样本到目标模板之间的距离来进行不同类别的判别。最后通过滚动轴承故障诊断实例进行测试表明测试样本的分类都完全正确,验证了该故障分类方法的有效性。⑤初步研究了风电机组振动监测及故障诊断系统及其软件的设计与实现。采用面向对象的编程技术进行该监测系统的软件开发,并基于本文提出的方法初步开发了一套用于风电机组振动监测及故障诊断的分析系统,该系统包括辅助功能模块、信号处理模块、特征提取模块和故障诊断模块,可以完成对风电机组振动信号的预处理、特征提取、故障诊断等分析功能,为故障特征的提取及故障的诊断提供了有效的帮助。文章最后对本文的工作进行了总结和对相关的研究技术进行了展望。更多还原

【Abstract】 As the increase of the wind turbine unit capacity and the new installed capacity every year, the relative tertiary industry such as maintenance, monitoring and fault diagnosis will be a new growth point in the wind industry. As the wind turbine work environment is very poor and the wind has high instability, the alternant force makes the transmission system the much easier damaged components in wind turbine. The wind turbine is installed in remote areas and is high from the ground, which makes it difficult to maintenance. Therefore the condition monitoring and fault diagnosis of the wind turbine in this case has significant meaning. Some research work shows that the vibration fault in transmission system has higher proportion compared to other wind turbine parts, and the condition monitoring to the transmission system is important. Currently most large commercial wind turbine has their own kinds of Supervisor Control And Data Acquisition (SCADA) systems, which supplied strong technology flat roof and sustentation for the improvement of the wind farm’s stability and reliability. However, at the same time, the quality of the SCADA systems can’t satisfy the needs of the vibration monitoring and fault diagnosis. It has relatively simple analysis functions and is lack of time-frequency analysis method which has better effect in dealing with no-stationary signals. The SCADA system is also lack of vibration monitoring and relative analysis. Though the WindCon system aimed at the wind turbine vibration fault monitoring has the early warning and alarming functions, it at the same time is lack of precision fault diagnosis function. Therefore this paper select the transmission system as the keystone monitoring objects and carry out the research work on the vibration monitoring and fault diagnosis. The main research work and conclusion are as follows:①The components such as wheels, gear box and generator in the transmission system and its vibration fault characteristic are analyzed in detail, which helps to ensure the measurement point positions in monitoring. The wind turbine vibration signal is disturbed by the background noise and the fault components such as bearing and gear in the rotation part shows cycle non-stationary characteristic. At the same time the alternate load by the non-stationary wind forced on the transmission system, makes the vibration signal showing Gauss noise immingled and non-linearity characteristic. Through the analysis on the key components in the gear-box, the mesh frequency and fault frequency are calculated and the monitoring points are confirmed. Then the vibration sensors can be settled and the vibration signal can be collected.②In the pretreatment research of the wind turbine vibration signal, a new method based on cross validation method optimized Morlet wavelet is put forward and discussed in detail. In the wind turbine structures the signals under considerations are known to be non-stationary, for which the signal parameters are time-varying. But for early fault signals, the fault feature signal is not strong enough to be caught, which can be drowned in the strong noise signals. In this case the traditional filter methods can’t separate the noise and useful components. The wavelet de-nosing method has better analysis affect but at the same time has some difficult in the selection of wavelet base and decomposition level. Aimed on the characteristic that the wind turbine work condition is rush and full of strong noise pollution, an adaptive wavelet de-noising method was proposed according to the inverse characteristics of useful signal and noise in different wavelet scales and the limitation of the traditional threshold methods. Then a new de-noising method based on parameter optimized Morlet wavelet is put forward. The simulation and experiment results reveal that both these two methods can considerably improves the capability of feature extraction and incipient fault diagnosis under strong noise background.③Aimed at the cycle non-stationary characteristic of the wind turbine vibration signal, a fault diagnosis method based on auto term window repressed Wigner-Ville distribution (WVD) is discussed in detail. As the wind turbine vibration signal has cycle non-stationary characteristic, the simple time domain methods and frequency domain methods can’t obtain perfect effect. The time-frequency methods have good effect in dealing with no-stationary signals, in which the WVD theoretically has an infinite resolution in time-frequency domain, is chosen to extract feature of the wind turbine vibration signal. But the WVD has the fault in cross term interface, which need to be suppressed by appropriate methods in the feature extraction analysis. Based on the relationship between the auto terms and the cross terms of WVD, a new threshold adaptive short-time Fourier transform (ASTFT) method is put forward. Then the auto term window suppressed WVD feature extraction method is discussed in detail. The auto term window is designed based on the smoothed pseudo Wigner-Ville distribution (SPWVD) and takes the place of the auto term in window analysis. All these three methods can not only remove the cross terms efficiently, but also reserve most advantage of WVD at the same time. The simulation and experiment results show that the proposed methods are validity tools for TFR of multi-component non-stationary signals in feature extraction.④Aimed at the non-gaussian and non-linearity characteristic of the wind turbine vibration signal, a fuzzy high-order spectrum fault diagnosis method is presented. This method can not only de-noises the Gauss noise in the vibration signal, but also has good effect in analyzing the no-linearity characteristic vibration signal and realize the correct fault diagnosis. At first the research using bi-spectrum analysis on the rolling bearing fault vibration signal in different fault styles show that, the bi-spectrum analysis results has relationship with the fault styles and this relationship has no effects by the rotate speed. On the base of the bi-spectrum analysis threshold result, the target template combined of kernel map and region map is constructed. Then by testing the distance between the test sample and the target template, the different fault can be distinguished on the value of the distance. Theoretical analysis and rolling bearing fault diagnosis show that the new method has good validity in fault diagnosis and the classification of all the test samples are correct.⑤In the pilot study on the wind turbine vibration monitoring and fault diagnosis system, the system and the design and realization of hardware and software are discussed in detail. Investigating the system structure of the parameter-sharing module software, the uniform frame work of the system module and the apparatus interface are designed. The wind turbine vibration monitoring and fault diagnosis system is in principium exploited based on the methods in this paper, which supplies strong help to the fault character extraction and fault diagnosis. The project applications and wind turbine vibration analysis proved the software be practical and availability. At the end of the thesis, the summarization of the article and expectation of the relative technology development are presented.更多还原