【作者】 孙自强；

【导师】 陈长征；

【作者基本信息】 沈阳工业大学 ， 机械设计及理论， 2013， 博士

【摘要】 风能作为一种清洁能源受到普遍重视，我国风电制造和风场建设都持续占据世界第一大国的地位，但同时风电设备面临着提高运行稳定性和降低故障发生率的问题，这已经成为制约风电行业进一步发展的瓶颈。我国风电建设正从重视规模和速度化向重视质量和效益化的稳定发展模式转变，因而风电设备早期故障诊断和状态监测技术是一项非常重要的研究内容。风电机组工作环境十分恶劣，叶轮转速随风速变化而波动，将交变载荷传递给整个风电传动链上，各个部件在刚性连接下受到复杂的交变冲击，会出现各种故障。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，因此对传动系统进行状态监测是至关重要的。目前，一些商业系统已经能够对风电设备运行数据进行实时获取，具有预警和报警功能，但缺乏对传动系统故障分析的能力。此外风电设备强噪声、不稳定和非线性的特征也使得一些传统故障诊断技术对于风电设备故障诊断率不高，特别是对一些早期故障。基于此本论文选择风力机传动系统为重点研究对象，利用混沌分形理论展开振动数据分析，对相关故障诊断和状态预测进行研究。论文的主要工作如下：论文分析了风电设备振动信号特征，在信噪比相对较低的情况下，基于混沌振子的早期故障振动方法充分利用了Duffing混沌振子对微弱周期信号敏感和对噪声免疫的特点，提高了故障特征信号的分辨率，并经过实验验证了该方法具有可行性。同时结合取样积分技术进一步提高了Duffing混沌振子对信号的提取门限，并利用最大Lyapunov指数定量判断故障是否产生。为了克服风电转速波动造成时域信号不稳定的现象，对时域信号进行了角域重采样，提出对角域信号进行Duffing振子特征信号提取的方法。对风电设备不同状态下的振动信号，提出用振动信号的分维数进行故障判断区分。分别采用了盒维数、关联维数和多重分形对风电设备进行了分维数提取和分析，并且采用双尺度和小波包变换进行了相关改进。通过对待测信号和已知状态之间进行维数距离的比较，发现分维数可以作为风电运行状态以及不同故障识别的一种有效手段。最后对于大型风力机的运行状态提出了基于最大Lyapunov指数局部混沌预测法。作为识别系统混沌行为的工具，用重构相空间和非线性函数逼近法来产生动态模型。这种预测手段依据的是信号自身演化规律属性，在很大程度上减少主观错误，所以对于工作状态的预报精度和可靠性上都有极大提升。更多还原

【Abstract】 Wind power, as a kind of green energy, draws wide attention, and China has retainedits crown as the world’s largest wind manufacturing and wind farm building nation inrecent years. At the same time, wind power industy faces the problems of how to improveoperational stability and decrease fault ratio, which is the the bottleneck restricting furtherdevelopment to a certain extent. Wind power industy development shifts from scale andspeed to quality and efficiency in China, so it is an important content in the research ofincipient fault diagnosis and condition monitoring for wind power equipments. Every rigidconnecting part of wind power equipments is prone to various faults due to abominableworking environment, the big alternant loads produced by changing vane rotating based onwind speed. Some research work shows that the vibration fault in transmission system hashigher proportion compared to other wind turbine parts, so the condition monitoring to thetransmission system is critical. Currently lots of commercial systems can obtain the windturbine operating data in real time and provide the early warning and alarming function.However, these systems are lack of vibration monitoring and relative analysis. Traditionalfault diagnosis technology is also inefficient to the wind turbine equipments because ofstrong noise, instability and nonlinearity, especially to incipient fault diagnosis. Thereforethis paper selects the transmission system as the keystone research subject, and utilze chaosand fractal theory to analyze the vibration signals fault diagnosis and running stateprognosis. The main research work and conclusion are as follows:Vibration signal characteristics of wind turbines are analyzed in the paper. And at arelatively lower signal-to-noise ratio (SNR) the accuracy of fault signal feature extractionis greatly improved, which is effective proved by experimental results, because Duffingoscillator is high sensitive to weak periodical signals and immune to noise. Sampling integraltechnology is applied to improve the signal tracking ability, and maximum Lyapunov exponentis adopted as the quantitative judgment of fault signals. Angular resampling algorithm for applying in conditions monitoring of speed variability, as occurs in wind turbines, can changethe signals from time domain to angle domain in order to remove the speed fluctuations. Then,Duffing oscillator is utilized to extact the angle domain characteristic signals.Fractal dimensions of vibration signals are calculated to distinguish wind turbine differentworking state. Box dimension, correlation dimension and multifractal dimension of windturbine vibration signals are calculated separately. The method of monitor conditioning basedon fractal dimension is modified combined with bi-scale characteristic of vibration signal andthe wavelet packet transformation. Fractal dimension can be used as an efficient measure ofwind turbine different working state or fault by comparing results of dimension distance of themeasured signals to different known state.Based on maximum Lyapunov exponent，the local chaotic time series prediction model ispresented to forecast the wind turbine working state. As the tool of recognizing system chaoticbehavior, reconstructing phase space and nonlinear function approximation technique presentsdynamic model. The prediction model is directly from objective laws of time series, whichavoids the defects of human subjectivity and improve the precision and reliability of forecastresults.更多还原