基于自适应多小波与流形学习的水电机组振动故障特征提取方法研究

Hydro-turbine generating unit is a complicated nonlinear system which is affected by coupling of hydraulic, mechanical, and electromagnetic factors. Its fault diagnosis problem has always been the research hotspot and difficulty in hydropower industry. Vibration fault feature extraction is the key procedure in the process of fault diagnosis. Since traditional vibration fault feature extraction methods lack adaptability, the extraction result exhibits poor sensitivity and the fault recognition rate is low. Aiming at this problem, this project will combine adaptive multiwavelets and manifold learning method to investigated adaptive fault feature extraction method for fault diagnosis of hydro-turbine generating unit. Firstly, multiwavelets will be given adaptability by research on multiwavelets construction method with variables to realize optimal matching between multiwavelets basis functions and vibration signal of hydro-turbine generating unit and reveal fault features of the unit more effectively. Then, sensitivity evaluation index is investigated for vibration fault feature selection to eliminate irrelevant features from the obtained high-dimensional vibration fault feature set so as to realize initial dimensionality reduction. After that, manifold learning method will be explored to excavate the essential rules and inner links among the feature parameters in the vibration fault feature subset after initial dimensionality reduction to remove the redundant information and extract low-dimensional high sensitive vibration features. Through these series of research, the hydro-turbine generating unit vibration fault recognition rate is expected to be improved and the realization of condition based maintenance is expected to be promoted.

水电机组是受水、机、电多因素耦合影响的复杂非线性系统，其故障诊断问题一直是水电行业研究的热点和难点。振动故障特征提取是水电机组故障诊断的关键环节，鉴于传统特征提取方法缺乏自适应性，特征提取结果敏感性差，导致诊断结果准确率低的问题，本项目拟将自适应多小波与流形学习方法相结合，研究水电机组振动故障特征自适应提取方法。通过带可变参数多小波构造方法研究，赋予多小波自适应特性，实现多小波基函数与水电机组振动信号的最佳匹配，更有效地揭示振动信号故障特征；引入敏感性评估指标进行特征选择，去除不相关故障特征参数，实现水电机组高维振动故障特征集初步约减；研究基于流形学习的水电机组非线性振动故障特征融合方法，充分挖掘初步降维后非线性振动故障特征子集中特征参数的本质规律和相互联系，消减冗余特征参数，提取低维强敏感振动故障特征，从而提高水电机组振动故障识别率，为状态检修目标的实现提供理论基础和方法体系。

水电机组是受水、机、电多因素耦合影响的复杂非线性系统，其故障诊断问题一直是水电行业研究的热点和难点。振动故障特征提取是水电机组故障诊断的关键环节，鉴于传统特征提取方法缺乏自适应性，特征提取结果敏感性差，导致诊断结果准确率低的问题，本项目将自适应多小波与流形学习方法相结合，研究水电机组振动故障特征自适应提取方法。.主要研究成果包括：.1、结合鲇鱼山水电站，研发了水电机组状态监测诊断系统软件，用于振动信号采集及专家系统诊断。.2、通过水电机组非线性动力学模型的构建，完成了动力学分析和特征参数敏感性分析，揭示了水电机组振动的产生原因和内部规律，为水电机组振动故障特征提取和诊断提供理论指导。.3、水电机组振动信号常受到大量噪声干扰，影响故障诊断结果的准确性。为提高振动信号降噪效果，提出了一种自适应冗余二代小波降噪方法，与二代小波降噪方法相比，利用自适应冗余二代小波降噪法对振动信号进行降噪，获得了较高的故障识别率。.4、结合水电机组的特点以及小波分析与局部线性嵌入（Local Linear Embedded, LLE）流形学习算法的优点，提出了基于LLE的水电机组故障特征提取方法，实现了水电机组振动故障特征的最优提取。利用转子实验台与水电机组振动数据进行实验，结果表明，所提出的方法能够有效区分不同的机组状态。.5、提出了基于自适应多小波与局部切空间排列（Local Tangent Space Alignment，LTSA）流形学习算法的水电机组振动故障特征提取方法。通过水电机组原始振动故障特征集的自适应构建、特征选择和特征融合三方面的努力，获取低维强敏感振动故障特征参数，利用转子实验台与水电机组振动数据进行实验，结果表明，所提出的方法能够提高机组振动故障诊断的准确性。.通过以上研究，构建了水电机组振动故障特征自适应提取方法体系，提高了振动故障诊断结果准确性，为水电机组状态检修目标的实现提供了理论支撑。