【作者】 付杰；

【导师】 曾洪涛；

【作者基本信息】 武汉大学 ， 动力机械及工程， 2018， 硕士

【摘要】 我国对电能的需求正在与日俱增。水力发电作为我国第二大发电方式,其清洁、成本低廉的特点符合我国制定的可持续发展政策。水电机组是水电站的核心部分,保证其安全稳定运行对电站的效益甚至整个电力系统的稳定性具有重大意义。然而当今对水电机组的检修形式存在一定的弊端,新型的状态检修形式逐渐受到人们的重视。状态检修需要通过测试信号判断水电机组的运行状态,因此开展水电机组故障诊断的研究,是目前研究的热点,而其中振动故障判断是研究的重点。由于水电机组的振动信号在采集过程中存在较多的背景噪声,必须对振动信号实施降噪处理。经验模态分解能够有效降低噪声干扰。故障的出现会使得水电机组振动信号中包含许多杂乱的频率成分,经过经验模态分解后故障特征将隐含在各阶分量中,采用一些无量纲特征参数可以将故障特征提取出来。水电机组是一个相当复杂的系统,故障与其特征之间的存在非线性的对应关系,神经网络经过训练后能够模拟这种非线性的映射关系,可以有效地判断水电机组的运行状态,而其中BP神经网络在拥有旋转结构的设备故障诊断领域中技术较为成熟,被广泛使用。本文主要的研究内容包括:(1)阐述水轮机工作原理和结构。分析水电机组经常出现的故障类型及其发生的原因。(2)系统阐述了经验模态分解的原理和特性,指出该方法用于降噪时存在的缺陷,提出模拟小波阈值降噪的经验模态阈值降噪方法,并通过实验检测了该方法的效果。(3)分析了经验模态分解中存在的端点效应和模态混叠对特征提取的不利影响。然后说明用镜像对称法可以约束端点效应的副作用,用聚合经验模态分解可以改善模态混叠现象,并通过实验检测了这两种方法的效果。(4)通过对特征提取的实际操作,选用了对故障敏感的6个无量纲特征参数作为故障特征用于故障诊断。(5)系统介绍了神经网络理论和BP神经网络的改进算法,利用BP神经网络处理水电机组故障诊断,得到最后的判断结果。更多还原

【Abstract】 Electric energy is the main energy nowadays,and power demand is growing in China.As the second largest power generation mode in China,hydro-power conforms with the sustainable development of our country by its characteristics of clean and low cost.Hydro-power unit is the key equipment of hydro-power station,and ensuring its safe and stable operation is very important to the the efficiency of power station and even the stability of the entire power system.However,there are some shortcomings of the tradition maintenance mode of hydroelectric generating units,so the new way of condition-based maintenance is attracting more and more attention.Condition-based maintenance needs the test signal to judge the running state of the hydroelectric unit.Therefore,it is a hot spot to do some researches on the fault diagnosis of the hydroelectric unit,especially on the vibration fault diagnosis.Due to the large amount of background noise in the acquisition process of vibration signals,we need to reduce the noise of vibration signals.The noise signals usually have high frequency and low energy.The empirical mode decomposition can reduce the noise effectively.The occurrence of the fault makes the vibration signal of the hydroelectric unit contain a lot of complex frequency components.After the empirical mode decomposition,the fault features are hidden in the intrinsic functions.The fault features can be extracted with some non-dimensional symptom parameters.The hydroelectric generating unit is a rather complex system,and its faults and characteristics have a nonlinear mapping.After training,neural network can simulate the nonlinear mapping relationship,so it can judge the running state of hydro-power units effectively.BP neural network is relatively mature and widely applied in the field of rotating machinery fault diagnosis.The main work of this paper are as follows:(1)Introduce the working principle and structure of the hydraulic turbine.Analyze the common fault types and the reasons in hydroelectric generating units.(2)Introduce the principle and characteristics of empirical mode decomposition.Point out the problems of empirical mode noise reduction.Propose an empirical mode threshold denoising method by simulating wavelet threshold denoising.Test the effect of this method with experiment.(3)Analyze the negative effects of endpoint effect and modal aliasing on the feature extraction.Point out the mirror extending method can restrain endpoint effect and ensemble empirical mode decomposition can restrain modal aliasing.Test the effect of two methods with experiment.(4)Select 6 non-dimensional symptom parameters which are sensitive to faults through the practical operation of feature extraction as fault features.(5)Introduce neural network theory and improved training algorithms of BP neural network.Obtain the final judgement of hydro-power unit with this method.更多还原