基于卷积神经网络与图卷积网络的水力机械故障诊断  

Hydraulic-mechanical Fault Diagnosis Based on Convolutional Neural Networkand Graph Convolutional Network

作　　者：吴学春 夏臣智 肖湘曲 李超顺[3] 李英玉 莫兆祥 吴韬为 WU Xue-chun;XIA Chen-zhi;XIAO Xiang-qu;LI Chao-shun;LI Ying-yu;MO Zhao-xiang;WU Tao-wei(Jiangsu Water Source Co.,Ltd.,East Route of South-to-North Water Diversion Project,Nanjing 210019,Jiangsu Provine,China;South-to-North Water Diversion(Jiangsu)Digital Intelligence Technology Co.,Ltd.,Nanjing 210019,Jiangsu Provine,China;Huazhong University of Science and Technology,Wuhan 430074,Hubei Province,China;Jiangsu Pumping Station Engineering Technology Research Centre,Nanjing 210019,Jiangsu Provine,China)

机构地区：[1]南水北调东线江苏水源有限责任公司,江苏南京210019 [2]南水北调(江苏)数智科技有限公司,江苏南京210019 [3]华中科技大学,湖北武汉430074 [4]江苏省泵站工程技术研究中心,江苏南京210019

出　　处：《中国农村水利水电》2025年第2期143-147,共5页China Rural Water and Hydropower

基　　金：江苏省水利科技重点项目(2022001)。

摘　　要：水力机械设备在当前国民生产中扮演着重要角色,其安全稳定运行至关重要。针对单一深度特征难以有效反映机组故障信息的难题,提出了基于卷积神经网络与图卷积网络特征融合的水力机械设备故障诊断模型。首先利用卷积神经网络获取水力机械设备监测信号卷积深度特征,同时利用快速傅里叶变换获取监测信号频谱值,构建监测信号图数据,建立图卷积网络提取样本关联特征。然后利用注意力机制对不同类型特征进行加权求和实现多模态特征融合。最后利用全连接层实现设备的故障诊断。通过水电机组、水泵主机组故障实测数据以及轴承故障数据进行验证,结果表明所提模型能有效实现水力机械设备故障诊断。Hydraulic machinery and equipment play an important role in the current national production,and its safe and stable operation is very important.In order to solve the problem that it is difficult for a single depth feature to effectively reflect the fault information of the unit,a fault diagnosis model for hydraulic machinery and equipment based on the fusion of convolutional neural network and graph convolutional network features is proposed.Firstly,the convolutional neural network is used to obtain the convolution depth characteristics of the monitoring signal of hydraulic mechanical equipment,and the fast Fourier transform is used to obtain the spectral value of the monitoring signal,and the monitoring signal graph data is constructed,and the graph convolutional network is constructed to extract the sample association features.Then,the attention mechanism is used to perform weighted summation of different types of features to achieve multimodal feature fusion.Finally,the full connectivity layer is used to realize the fault diagnosis of the equipment.The model is validated by the measured data of the hydroelectric unit and main pump unit faults as well as the bearing fault data.The results show that the proposed model can effectively realize the fault diagnosis of hydromechanical equipment.

关 键 词：水力机械 卷积神经网络 图卷积网络 故障诊断 

分 类 号：TV312[水利工程—水工结构工程]