基于改进物理信息神经网络的轴流泵流场重构方法研究

打开文本图片集

关键词：改进物理信息神经网络；三维卷积神经网络；流场重构；轴流泵；有限体积法中图分类号：TH312；035 文献标志码：A doi：10.3969/j.issn.1000-1379.2026.03.

Research on Axial Flow Pump Flow Field Reconstruction Method Based on Improved Physics-Informed Neural Network

LIU Kang1.²，LIU Xingning³，SUN Yong4，LIU Liang4， JIA He1，²， ZENG Tao1，²， ZHANG Yaofei 1，2 （1.StateKeyLaboratoyofHydraulicEnginengIteligentConstructioandOpeationanjinUnversityTanj3O4ia； 2.School ofCivilEnginering，Tianjin University，Tianjin 300354，China；3.PowerChina Kunming Engineering Co.，Ltd.， Kunming 650051，China；4.China Water Huaihe Planing，Design and Research Co.，Ltd.，Hefei 2306O1，China） Abstract：Thflowfieldinformationofaxialfowpumpsservesathebasisforoperationalstabilityanalysisndstructuraloptiizationdesign.Duetotelatiosofauretoitisalgttaiopletefoeldfoatiouringeratiofor animprovedPhsics-InforedNeuralNetork（PINN）modelwasproposedforreconstructingtheflowfieldundersparsedataconditions. Firstlytheflowfieldproblemasdsibedbyaalyingphysicalconstraints，oundaryonditions，ndflowfeldonstraints.hn，aD ConvolutionalNeuralNetwork（3DCN）wasintroducedtosolvethfowfeldproblem.Lastly，theFinite VolumeMethod（FVM）wasused for numerical simulation toobtainsteady-statevelocityand pressuredistribution information.After meshing preprocessing， 1% of the flow felddatawassampledformodeltraiing.Tovalidatethproposedmethod，asmplifiedaiaflowpumppipelineasusedasatestcase.The resultsndicaetatterecostructedflofieldusingteiprovedPodelcloselyatcestheFsimulatedfofeld，ipsue beinglargelyconsistentandvelocitytrendsbeingsimilar，exhibitingonlyminordeviatonsintheflowfieldregionsnearthipelerand guidevanes.Thisdemonstratesthattheproposedmethodcanaccuratelypredictthethree-dimensionalflowfieldundersparsedatand complex boundary conditions.

Keywords：improvedPhsics-InforedNeuraletwork；3DConvolutionaleuralNetwork；flowfieldreconstruction；aialflowpump Finite Volume Method

0 引言

轴流泵广泛应用于水利工程、石油化工等行业，其高效运行对系统安全稳定至关重要。（剩余9515字）