Author/Authors :
An-ShikYang، نويسنده , , Tien-ChuanKuo، نويسنده , , Pou-HongLing، نويسنده ,
Abstract :
Thephasetransportphenomenonofthehigh-pressuretwo-phaseturbulentbubblyflowinvolvescomplicatedinterfacialinteractionsofthemass,momentum,andenergytransferprocessesbetweenphases,revealingthatanenormouseffortisrequiredincharacterizingtheliquid–gasflowbehavior.Nonetheless,theinstantaneousinformationofbubblyflowpropertiesisoftendesiredformanyindustrialapplications.Thisinvestigationaimstodemonstratethesuccessfuluseofneuralnetworksinthereal-timedeterminationoftwo-phaseflowpropertiesatelevatedpressures.Threeback-propagationneuralnetworks,trainedwiththesimulationresultsofacomprehensivetheoreticalmodel,areestablishedtopredictthetransportcharacteristics(specificallythedistributionsofvoid-fractionandaxialliquid–gasvelocities)ofupwardturbulentbubblypipeflowsatpressurescovering3.5–7.0MPa.Comparisonsofthepredictionswiththetesttargetvectorsindicatethattheaveragedroot-mean-squared(RMS)errorforeachoneofthreeback-propagationneuralnetworksiswithin4.59%.Inaddition,thisstudyappraisestheeffectsofdifferentnetworkparameters,includingthenumberofhiddennodes,thetypeoftransferfunction,thenumberoftrainingpairs,thelearningrate-increasingratio,thelearningrate-decreasingratio,andthemomentumvalue,onthetrainingqualityofneuralnetworks.
