摘要
辅助给水汽动泵(汽辅泵)作为核电站专设安全设施,在保障核电站安全可靠地停堆、减轻事故后果方面发挥着重要作用。因此,针对核电现场汽辅泵运行特点,提出了一种基于深度卷积收缩网络(deep convolutional shrinkage network,DCSN)模型的汽辅泵故障诊断方法。该方法首先针对汽辅泵压力、转速等参数数据,按照一定尺寸将采集的时间序列状态信号矩阵化,构成了多故障类型的故障样本;然后,将软阈值模块嵌入卷积神经网络(convolutional neural network,CNN),构建了DCSN模型用于故障诊断;最后,利用核电厂全范围模拟机中的故障数据集合,对所提出方法进行验证。研究结果表明:与CNN模型相比,所提出的DCSN模型具有更优越的性能。
The steam-driven auxiliary feedwater pump,as a dedicated safety facility in nuclear power plants,plays a critical role in ensuring safe reactor shutdown and mitigating accident consequences.In terms of the operation characteristics of the pumps in nuclear power scenarios,a fault diagnosis method was proposed based on deep convolutional shrinkage network(DCSN)model.Firstly,time-series condition signals of key parameters(e.g.,pressure and rotational speed)were processed into matrices with predefined sizes,constructing multi-category fault samples.Subsequently,a DCSN model was developed by embedding soft thresholding modules into a convolutional neural network(CNN).Finally,the proposed method was validated using a fault dataset generated from a nuclear full-scope simulator.Case study results demonstrate that compared with the conventional CNN mo-del,the DCSN achieves superior diagnostic performance.
作者
刘丙月
赵新文
姜佳行
曾利民
万舒
LIU Bingyue;ZHAO Xinwen;JIANG Jiahang;ZENG Limin;WAN Shu(Naval Univ.of Engineering,Wuhan 430033,China;China Wuhan Second Ship Design and Research Institute,Wuhan 430064,China;Hainan Nuclear Power Co,Ltd.,Changjiang 572800,China;Fujian Fuqing Nuclear Power Co,Ltd.,Fuqing 350300,China)
出处
《海军工程大学学报》
北大核心
2025年第3期94-98,共5页
Journal of Naval University of Engineering
关键词
汽辅泵
故障诊断
深度卷积收缩网络
卷积神经网络
软阈值化
steam-driven auxiliary feedwater pumps
fault diagnosis
deep convolutional shrinkage network
convolutional neural network
soft thresholding
