基于EEMD-LSTM的汽轮机转子碰磨故障诊断模型及其工程应用  

EEMD⁃LSTM⁃based Turbine Rotor Rub⁃impact Fault Diagnosis Model and Its Engineering Application

作　　者：陈尚年 李录平[1] 张世海 王颖[1] CHEN Shang-nian;LI Lu-ping;ZHANG Shi-hai;WANG Ying(School of Energy and Power Engineering,Changsha University of Science and Technology,Changsha,China,Post Code:410014;Guizhou Chuangxing Electric Power Research Institute Co.,Ltd.,Guiyang,China,Post Code:550002)

机构地区：[1]长沙理工大学能源与动力工程学院,湖南长沙410014 [2]贵州创星电力科学研究院有限责任公司,贵州贵阳550002

出　　处：《热能动力工程》2023年第8期159-168,共10页Journal of Engineering for Thermal Energy and Power

基　　金：国家重点研发计划(2017YFB0903600);南方电网公司重点科技项目(GZKJXM20172214)。

摘　　要：汽轮机转子与静子间的碰磨严重影响着机组的安全运行。为了解决汽轮机转子发生在早期和中期的碰磨故障难以通过基于振动信号检测诊断方法进行有效识别的问题,本文提出一种基于EEMD-LSTM的汽轮机转子碰磨故障诊断方法。首先,该方法通过声发射技术监测汽轮机转子的碰磨故障信号;然后,利用EEMD信号分解方法处理获取的声发射信号,并提取能量特征参数和相关的时域特征参数,从而获得碰磨故障特征数据集;最后,利用划分的数据集对LSTM神经网络进行训练与测试,从而获得碰磨故障诊断模型。工程应用结果表明,本文提出的方法能够有效识别机组在不同转速时期的早期碰磨故障,且故障诊断的准确率较高。The rub⁃impact between rotor and stator components seriously affects the safe operation of the turbo⁃generator unit.In order to solve the problem that it is difficult to effectively identify the steam turbine rub⁃impact faults in the early and intermediate period by vibration signal detection and diagnosis method,a rub⁃impact fault diagnosis method of steam turbine rotor based on EEMD⁃LSTM was proposed in this paper.Firstly,the rubbing fault signal of turbine rotor was monitored by acoustic emission(AE)technology;then,the EEMD signal decomposition method was used to process the obtained AE signal,from which the energy feature parameters and related time domain feature parameters were extracted to obtain the characteristic data set of rub⁃impact fault;finally,the divided data set was used to train and test the LSTM neural network to obtain the diagnosis model of rub⁃impact fault.The engineering application results show that the method proposed in this paper can effectively identify the initial rub⁃impact fault of unit at different rotating speeds with high fault diagnosis accuracy.

关 键 词：汽轮发电机组 碰磨故障 集合经验模态分解 长短时记忆网络 故障诊断 

分 类 号：TK268[动力工程及工程热物理—动力机械及工程]