基于人工神经网络的半刚性钢框架性能点预测(IJISA-V11-N10-5)

I.J.IntelligentSystemsandApplications,2019,10,42-53PublishedOnlineOctober2019inMECS()DOI:10.5815/ijisa.2019.10.05Copyright©2019MECSI.J.IntelligentSystemsandApplications,2019,10,42-53PredictionofPerformancePointofSemi-RigidSteelFramesUsingArtificialNeuralNetworksZahraBahmaniaaDepartmentofComputerEngineering,BehbahanKhatamAlanbiaUniversityofTechnology,Behbahan,IranE-mail:[e-mail]ilEngineering,UniversityofSistanandBaluchestan,Zahedan,IranE-mail:[e-mail]ofCivilEngineering,YazdUniversity,Yazd,IranE-mail:[e-mail]partmentofCivilEngineering,BehbahanKhatamAlanbiaUniversityofTechnology,Behbahan,IranE-mail:[e-mail]:19April2019;Accepted:07June2019;Published:08October2019Abstract—Oneofthemainstepsintheperformancebasedseismicanalysisanddesignofstructuresisdeterminationofperformancepointwherethenonlinearstaticanalysisapproachisused.Theaimofthispaperistopredicttheperformancepointofsemi-rigidsteelframesusingArtificialNeuralNetworks.Assuch,togeneratedatarequiredfortheprediction,severalsemi-rigidsteelframesweremodeledandtheirperformancepointwasdeterminedthen.Teninputvariablesincludingnumberofbays,numberofstories,bayswidth,momentofinertiaofbeams,crosssectionalareaofcolumns,crosssectionalareaofbraces,rigiditydegreeofconnectionsandsoftstory(existenceornonexistence)wereconsideredintheprediction.Inaddition,theactualresultswereobtainedatthepresenceofdifferentearthquakeintensitylevelsandsoiltypes.BackPropagationwithelevendifferentalgorithmsandRadialBasisFunctionArtificialNeuralNetworkswereusedintheprediction.Thepredictionprocesswascarriedoutintwosteps.Inthefirststep,allsampleswereusedforthepredictionandtheperformancemetricswerecomputed.Inthesecondstep,threeofthebestnetworkswereselected,andtheoptimumnumberofsampleswasfoundconsideringaveryslightreductionintheaccuracyofthenetworksused.Finally,itwasshownthat,despiteusingratherlimitednumberofsamples,thegeneratedArtificialNeuralNetworksaccuratelypredicttheperformancepointofsemi-rigidsteelframes.IndexTerms—Artificialneuralnetworks,prediction,semi-rigidconnection,steelstructures,performancepoint.I.INTRODUCTIONPastdestructiveearthquakes(e.g.1989LomaPrieta;1990Manjil-Rudbar;1994Northridge)havelefttheirsignaturebelowthedocumentsoftheeconomicalandlifelossesreportedthen.Inaccordancewiththeseismicdesigncodes,somedegreeofdamageisexpectedforordinarybuildingssubjectedtodesignbasisearthquakes.Althoughpreventingthenaturalearthquakeoccurrenceisbeyondthehumancontrol,thewaysofmitigatingsuchlosseshavebeentheenforcedinterestofresearchersandengineeringcommunities.Theinappropriateperformanceofstructuresunderthepastdestructiveearthquakescausedthecommunitiestoeliminatetheunexpectedlossesthen.Oneofthemodernmethodologiesgainingsignificantattentionistheperformance-basedseismicdesign(PBSD)[1].Overthelast30years,theconceptualframeworkofPBSDwasdevelopedbythevariousguidelinespublishedbythewell-knownengineeringassociationssuchasStructuralEngineersAssociationofCalifornia[2],AppliedTechnologyCouncil[3],andFederalEmergencyManageAgency[4].Conceptually,itisexpectedtohavedifferentseismicperformancelevelsofstructuresduetovariousstructuralseismicbehaviorsandhazardlevels.Becauseofthis,theabovementionedguidelineshaveintroduceddifferentseismicperformancelevels(orobjectives)correspondingtothedifferenthazardlevelsandthus,eachstructureofinterestshouldbeevaluatedbasedonitsrelevantperformancesandhazardlevels.BasedonthestudiessoPredictionofPerformancePointofSemi-RigidSteelFramesUsingArtificialNeuralNetworks43Copyright©2019MECSI.J.IntelligentSystemsandApplications,2019,10,42-53farcarriedout,itwasfoundthatthesteelframestructuresexperiencesomedegreeofnonlinearityundersevereearthquakesandthus,thesingleparametersofstrengthcouldnotbeacceptedfortheseismicanalysisanddesign.Hence,astheguidelineshaverecommended,theanalysisanddesignprocesscanbecarriedoutusingtheexpectedseismicperformance(e.g.forceand/ordeformation)ofthesubjectstructure.Infact,inordertodesignorretrofitthenewandexistingbuildingsrespectively,aperformanceobjectiveischosenbyowner/designerbasedonthedefinedperformanceandhazardlevelsinthementionedguidelines[4].Performance-basedseismicanalysisanddesignofstructuresrequiresdeterminingtheperformancepoint.Theperformancepointisobtainedfromtheintersectionpointoftargetdisplacementandbaseshearonthepushovercurveofthesubjectstructure.Todeterminetheperformancepoint,nonlinearstaticanalysisisrequiredtobeperformed.Theprocessiscase-sensitiveandthenonlinearanalysisisinherentlyatime-consumingprocesswhichneedshighcomputationalefforts.Analternativewayistopredictthepointusingsoftcomputing-basedpredictivetools.Overthelastthreedecades,themuchcomputationalburdenofcomputationalapproachesandsometimestheirweaknesswereresultedindiscoveringthesoftcomputing-basedapproach,whichhavebeenincreasinglyapplyinginthedifferentaspectsofcomplexengineeringproblems.Softcomputing-basedapproa