基于小波神经网络的双馈感应风力发电机故障诊断研究

发布时间：2018-05-30 15:34

本文选题：双馈感应风力发电机 + 故障诊断　；参考：《上海电机学院》2017年硕士论文

【摘要】：随着风力发电的强劲复苏,截止2015年末,我国风电装机容量新增32.97GW,装机容量又创历史新高。双馈感应风力发电机作为双馈型风电机组的关键设备,在运转时其工作状况决定整个风电机组的效率。为保证双馈感应风力发电机安全长久稳定运转,对双馈感应风力发电机故障诊断研究极其关键。本论文的主要研究内容如下:(1)研究分析双馈感应风力发电机的故障类型及对应故障产生机理,选用故障率较高的定子故障和轴承故障作为研究对象。设计3.5KW双馈感应风力发电机故障模拟实验平台,该实验平台能够实现定子绕组的匝间短路故障、相间短路故障、单相接地故障和轴承故障。根据定子绕组和轴承的故障特征表现,确定分别以定子绕组电压和轴承振动信号作为数据采集对象。(2)提出一种改进的小波阈值函数和阈值方法,改进后的阈值函数能够对含噪信号实现更好的去噪效果,保留更多有效信号,信噪比更高且均方根误差更小相比于传统小波阈值函数。以定子电压和轴承振动信号作为实例分别进行去噪处理,对比分析去噪前后的信号,验证了改进的小波阈值去噪方法的有效性。(3)对经小波去噪后的定子电压和轴承振动信号分别采用小波包进行频带分解,计算各个频带的能量值并归一化处理组成故障诊断的特征向量,与没有故障的频带能量信号对比分析,验证了小波包提取故障特征的精确性。(4)提出一种改进的BP神经网络隐含层节点算法,改进的隐含层节点数算法能够确定更优的隐含层节点数,使网络的性能更优。以定子绕组和轴承数据为样本训练改进的BP神经网络,并用测试数据对训练好的网络进行检测。实验结果表明,改进的BP神经网络对发电机定子绕组和轴承故障类型识别准确率较高。(5)利用虚拟仪器LabVIEW设计了双馈感应风力发电机的故障诊断系统,包括用户登入、数据采集、数据显示、数据分析和故障诊断模块,然后通过LabVIEW的Web服务器功能,实现故障诊断系统与互联网结合,增强了故障诊断系统的共享性和交互性。
[Abstract]:With the strong recovery of wind power generation, by the end of 2015, the installed capacity of wind power in China has been increased by 32.97 GW, and the installed capacity has reached a new record. As the key equipment of the doubly-fed wind turbine, the efficiency of the wind turbine is determined by the working condition of the doubly-fed induction wind turbine. In order to ensure the safe and stable operation of doubly-fed induction wind turbine, it is very important to study the fault diagnosis of doubly-fed induction wind turbine. The main contents of this thesis are as follows: (1) the fault types and corresponding fault generation mechanism of doubly-fed induction wind turbine are analyzed, and the stator faults and bearing faults with high failure rate are selected as the research objects. The fault simulation experiment platform of 3.5KW doubly-fed induction wind turbine is designed. The platform can realize inter-turn short-circuit fault, inter-phase short-circuit fault, single-phase grounding fault and bearing fault of stator winding. According to the fault characteristics of stator winding and bearing, this paper proposes an improved wavelet threshold function and threshold method, taking stator winding voltage and bearing vibration signal as data acquisition objects, respectively. The improved threshold function can achieve better denoising effect, retain more effective signals, have higher signal-to-noise ratio and lower root mean square error than the traditional wavelet threshold function. Taking stator voltage and bearing vibration signal as examples, the signal before and after de-noising is compared and analyzed. The validity of the improved wavelet threshold denoising method is verified. The wavelet packet is used to decompose the stator voltage and bearing vibration signal after wavelet denoising. The eigenvector of fault diagnosis is formed by calculating the energy value of each frequency band and normalizing processing, and compared with the energy signal of frequency band without fault. This paper verifies the accuracy of wavelet packet extraction of fault features and proposes an improved BP neural network hidden layer node algorithm. The improved hidden layer node number algorithm can determine the number of hidden layer nodes and make the network performance better. The improved BP neural network is trained with stator winding and bearing data as samples, and the trained neural network is detected by test data. The experimental results show that the improved BP neural network has high accuracy in identifying the stator windings and bearing fault types of generators. The fault diagnosis system of doubly-fed induction wind turbines is designed by using virtual instrument LabVIEW, including user login, data acquisition, etc. Data display, data analysis and fault diagnosis module, and then through the Web server function of LabVIEW, the fault diagnosis system is combined with the Internet, which enhances the sharing and interaction of the fault diagnosis system.
【学位授予单位】：上海电机学院
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TM315

【参考文献】