基于GRU模型的发电机转子绕组匝间短路故障诊断  

Vibration characteristic analysis and diagnosis of generator interturn short circuit fault

作　　者：马高海 刘大炜 杨宏宇 史宗辉 MA Gaohai;LIU Dawei;YANG Hongyu;SHI Zonghui(China Energy Investment Zhejiang Yuyao Fuel Gas Power Generation Co.,Ltd.,Ningbo 315400,China;Shanghai Proinvent Information tech ltd,Shanghai 110006,China;School of Mechanical Engineering,Shenyang University of Technology,Shenyang 110870,China)

机构地区：[1]国能浙江余姚燃气发电有限责任公司,浙江宁波315400 [2]上海博英信息科技有限公司,上海110006 [3]沈阳工业大学机械工程学院,辽宁沈阳110870

出　　处：《电力科技与环保》2022年第6期467-474,共8页Electric Power Technology and Environmental Protection

基　　金：国家自然科学基金(51675350)。

摘　　要：针对发电机转子绕组匝间短路早期故障不易检测的问题,本文提出了一种基于门控循环单元(Gated Recurrent Unit,GRU)的深度学习网络模型来进行发电机转子匝间短路故障诊断方法。发电机在发生匝间短路故障时会形成不平衡的电磁脉冲力,使发电机产生振动;采用振动特性方法分析绕组匝间短路故障时气隙磁势、励磁电流、气隙磁密等电气参数的变化,计算出使发电机产生振动的电磁脉冲力大小。并以某250 MW燃气发电机组为例进行了匝间短路故障检测试验,试验数据由发电机正常状态、匝间短路状态、其他故障状态的振动数据组成,共计900组,数据处理采用深度学习中的GRU神经网络模型,提取振动信号中的故障特征与时间序列特征,进而实现发电机转子绕组匝间短路故障的诊断。结果表明,绕组匝间短路故障时会产生f、2f、3f、4f、5f倍频的振动,其中2f倍频振动相对较大,符合理论分析结果;利用GRU神经网络模型对发电机转子绕组匝间短路故障进行诊断,准确率达到95.14%,高于AlexNet、VGG16等其他深度学习方法。通过试验可知,GRU模型可有效对发电机转子绕组匝间短路故障进行诊断。该研究可以在故障初期通过振动信号发现发电机匝间短路故障,为绕组匝间短路故障诊断提供了新思路。Aiming at the problem that the early fault of generator rotor winding inter-turn short circuit is difficult to detect,this paper proposes a fault diagnosis method of generator rotor inter-turn short circuit based on Gated Recurrent Unit(GRU)deep learning network model.Firstly,it is pointed out from the mechanism that the winding inter-turn short circuit fault will cause unbalanced electromagnetic force(electromagnetic force per unit area is pulse force),which will cause generator vibration.Secondly,the variation of air gap magnetic potential,excitation current,air gap magnetic density and other electrical parameters during winding inter-turn short circuit fault are analyzed by using the vibration characteristic method,and the pulsating magnetic pull that makes the generator vibrate is calculated.Finally,taking a 250 MW gas turbine generator unit as an example,the GRU neural network model in deep learning is used for data processing.Fault features and time series features in actual vibration signals are extracted.Combining the two features,GRU neural network is used to diagnose rotor winding inter-turn short circuit fault.The results show that the frequency doubling vibration of f,2f,3f,4f,5f will occur when the winding turns short circuit fault occurs,and the frequency doubling vibration of 2f is relatively large,which conforms to the theoretical analysis results.The accuracy rate of GRU network model generator rotor winding inter-turn short circuit fault diagnosis reaches 95.14%,which is superior to AlexNet,VGG16 and other in-depth learning methods.The test results show that GRU model can effectively diagnose inter-turn short circuit fault of generator rotor winding.It provides a new idea for fault diagnosis of inter-turn short circuit of rotor winding,so as to find the fault of generator interturn short circuit through vibration signal at the initial stage of fault.

关 键 词：发电机 匝间短路 故障诊断 振动分析 

分 类 号：TM301.3[电气工程—电机] TH17[机械工程—机械制造及自动化] TK229.8[动力工程及工程热物理—动力机械及工程]