基于经验模态分解和相似性挖掘的水电机组振动故障诊断研究

Based on existing research findings, the intelligent diagnosis problems of vibration fault for hydropower units were investigated in-depth by integrated various methods such as theoretical analysis, numerical simulation and experimental investigation. Through collecting the vibration signal data of the water power station with the fault diagnosis system and simulating the partial vibration fault in a model hydraulic turbine test rig, the fault samples of vibration fault were obtained, then these vibration signal data were analyzed and processed to set up the fault sets of hydropower units vibration fault. By reference from latest research achievements of modern signal processing and artificial intelligence fields, and by using lifting algorithm-based 2nd generation wavelet transform, wavelet packet decomposition, rough set reduction and support vector machines, and so on, the vibration signal de-noising, fault features extracting and fault diagnosis methods of hydropower units vibration fault were carried out to propose a intelligent on-line diagnosis method for the hydropower units vibration fault and to establish a diagnosis model of hydraulic turbine vibration fault.

在总结现有研究成果的基础上，采取理论分析和数值仿真相结合的总体技术路线，理论分析与实际情况相互参照，仿真计算与真机观测密切结合，深入研究水电机组振动故障的智能诊断问题。采集原型水轮机振动故障信号数据，对振动信号进行分析处理，获取水轮机振动的故障样本，建立水电机组振动故障的故障集。借鉴现代信号处理与智能化技术领域的最新研究成果，将经验模态分解方法与模糊数学、海量数据挖掘技术相结合，对水电机组振动故障诊断进行应用研究，建立水电机组振动故障的诊断模型。

本项目以水电机组振动故障为研究对象，借鉴现代信号处理领域的最新研究成果，对水电机组振动信号特征提取和状态识别方法进行了系统的研究。针对Hilbert-Huang变换中的虚假分量问题，提出能量波动法，构建波动阈值来跟踪、识别、重构虚假分量，解决了Hilbert-Huang变换理论存在的不足，完善了水电机组振动信号分析处理和故障特征的提取。系统研究了分形理论，提取水电机组振动信号的多重分形特征参数，探索在不同状态和不同工况下特征参数的分布规律和区分程度，为水电机组振动故障的特征提取提供了新的思路。将分形和支持向量机做了有机结合，先提取振动信号的多重分形特征参数，再应用改进支持向量机算法进行识别，充分发挥支持向量机良好的泛化推广能力和较强的小样本学习能力，以较高的准确度完成了多类轴心轨迹分类识别。在提取到水电机组转子不同运行状态下的多重分形特征参数后，应用模糊聚类的方法对信号进行模式识别，通过特征优选提高了分类的精度。项目组研发了水电机组状态监测与故障诊断系统和水电机组自动化盘车测量系统，获得了良好的应用效果。