风力发电机组发电机和齿轮箱故障诊断方法研究

发布时间：2018-01-26 11:06

本文关键词： 风电机组发电机齿轮箱故障诊断频谱分析法主成分遗传神经网络　出处：《华北电力大学》2014年硕士论文　论文类型：学位论文

【摘要】：风电建设快速推进的同时,也带来了一系列的挑战,突出的表现就是风电机组的质量问题频发,严重影响正常的生产发电工作,造成巨大的经济损失。因此,加强风电机组故障诊断,对降低风电场维护费用,提高风电场运行经济效益具有重要意义。风电机组主要包括齿轮箱、发电机、叶片、液压系统和偏航系统等部件,齿轮箱和发电机是关键部件,也是故障发生率最高的部件,其运行的稳定性会影响到整机性能。故本文以风电机组发电机和齿轮箱为研究对象,对其故障诊断方法进行了研究。首先,考虑到发电机轴承故障振动响应较弱,本文提出了倒频谱域相干分析的发电机组轴承故障特征提取方法,该方法利用相干分析减弱测量信号中噪声的干扰,突出故障信息,然后对相干函数做倒频域计算,提取边带特征。针对齿轮箱齿轮故障振动信号频谱结构的特点,提出了基于小波包与倒频谱分析的风电机组齿轮箱齿轮裂纹诊断方法,两种方法使风电机组发电机轴承故障和齿轮箱齿轮故障诊断通过简单易行的频谱分析实现。然后,本文提出了同类信息融合的方法,该方法选取振动信号的峭度、峰值作为时域特征值,利用小波包算法提取频带能量和二范数作为时频域特征值。考虑到特征值之间的相关性,利用主成分分析法确定主成分,从而减少神经网络的输入变量。利用遗传算法对BP神经网络权值和偏置进行优化,建立遗传神经网络的故障诊断模型。仿真测试表明了算法的有效性。最后,提出了异类信息融合的方法,该方法针对风电机组齿轮箱单一故障信号的局限性和故障特征存在较强非线性关系的特点,以采集的振动信号、温度信号和润滑油信号为原始信源,分别提取它们的峭度、小波包频带能量,齿轮箱轴承温度、齿轮箱油温,润滑油粘度作为特征值。考虑到特征值之间的相关性,利用主成分分析法对原始特征值的组合进行降维融合,得到信息互补的特征量。将融合特征通过遗传算法优化的神经网络进行模式识别。仿真测试表明了该方法比同类信息特征融合法具有更高的诊断精度。
[Abstract]:At the same time, wind power construction has brought a series of challenges, the outstanding performance is the frequent occurrence of wind turbine quality problems, seriously affect the normal production and power generation work, resulting in huge economic losses. Strengthening the fault diagnosis of wind turbine is of great significance to reduce the maintenance cost of wind farm and improve the economic benefit of wind farm operation. The wind turbine mainly includes gearbox, generator and blade. Components such as hydraulic system and yaw system, gearbox and generator are the key components, and also the components with the highest fault rate. The stability of its operation will affect the performance of the whole machine. Therefore, the fault diagnosis method of the generator and gearbox of wind turbine is studied in this paper. Firstly, considering the weak vibration response of generator bearing fault, this paper proposes a method of feature extraction of generator bearing fault based on coherent analysis in cepstrum domain. In this method, the interference of noise in the measured signal is reduced by coherence analysis, and the fault information is highlighted, then the coherent function is calculated in inverted frequency domain. According to the characteristic of frequency spectrum of gear fault vibration signal of gear box, a method of gear crack diagnosis based on wavelet packet and cepstrum analysis is proposed. The two methods make the fault diagnosis of generator bearing and gearbox gear by simple and easy spectrum analysis. Then, a similar information fusion method is proposed, in which the kurtosis of the vibration signal is selected and the peak value is taken as the time domain eigenvalue. Wavelet packet algorithm is used to extract frequency band energy and two-norm as time-frequency domain eigenvalues. Considering the correlation between eigenvalues, principal component analysis is used to determine principal components. In order to reduce the input variables of neural network, the weight and bias of BP neural network are optimized by genetic algorithm, and the fault diagnosis model of genetic neural network is established. The simulation results show that the algorithm is effective. Finally, a method of heterogeneous information fusion is proposed, which aims at the limitation of the single fault signal of the gearbox of wind turbine and the strong nonlinear relationship between the fault characteristics, so as to collect the vibration signal. The temperature signal and the lubricating oil signal are the original information sources, and their kurtosis, wavelet packet frequency band energy, gearbox bearing temperature, gear box oil temperature are extracted respectively. The viscosity of lubricating oil is regarded as the eigenvalue. Considering the correlation between the eigenvalues, the combination of the original eigenvalues is reduced by the principal component analysis (PCA). The information complementary feature quantity is obtained. The fusion feature is recognized by the neural network optimized by genetic algorithm. The simulation results show that this method has higher diagnostic accuracy than the similar information feature fusion method.
【学位授予单位】：华北电力大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TM315

【参考文献】