您好,欢迎访问三七文档
当前位置:首页 > 机械/制造/汽车 > 综合/其它 > 弧焊机器人焊缝跟踪神经网络控制器
弧焊机器人焊缝跟踪神经网络控制器高向东黄石生毛利彰山本元司摘要:介绍一种能提高弧焊机器人焊缝跟踪精度的神经网络控制器.通过神经网络的补偿作用,弥补了由于无法知道机器人精确模型所造成的控制上的误差.不同于机器人控制中传统的神经网络控制器,本文提出并应用了基于笛卡尔空间轨迹控制的机器人焊缝跟踪神经网络,大大简化了控制算法.计算机模拟及实验结果表明,该控制器非常适用于机器人的实际焊接,对于现有的机器人,无须改变其控制器内部结构,即可应用该技术.与常用的机器人关节力矩控制法相比,有效地提高了跟踪精度并具有较强的鲁棒性.关键词:弧焊机器人;焊缝跟踪;神经网络ApplicationofNeuralNetworkControllerintheSeamTrackingofArc-WeldingRobot*GaoXiangdongandHuangShisheng(DepartmentofMechatronicEngineering,SouthChinaUniversityofTechnology.Guangzhou,510641,P.R.China)AkiraMohriandMotojiYamamoto(DepartmentofIntelligentMachineryandSystems,FacultyofEngineering,KyushuUniversity,Japan)Abstract:Aneuralnetwork(NN)controllerwhichimprovestheaccuracyofseamtrackingofanarc-weldingrobotispresentedinthispaper.TheimprovementoftrackingaccuracycanbeachievedbyapplyingtheNNcontrollerforcompensatingformodeluncertaintiesofrobotmanipulator.UnlikethetraditionalNNcompensationofmodeluncertaintieswhichwascarriedthroughbymodifyingthejoint/forceoftherobot,theproposedNNcompensationisusedtomodifythereferenceCartesianseamtrajectory,whichiseasilyappliedinpractice.TherequiredinternalsignallevelofproposedNNfortheseammodificationismuchsmaller.Simulationsandexperimentshavebeenperformedonanactualarc-weldingrobotmanipulatortotesttheeffectivenessofNNcontrolscheme.IthasbeenfoundthatNNcangeneratebettertrackingperformancethanthetraditionalcomputedtorque(CT)controlmethodwhichisbasedonthemanipulatordynamicsonly.OnegoalofthispaperistostimulatefurtherdiscussionofapplicationofNNinthearc-weldingrobotcontrol.Keywords:arc-weldingrobot;seamtracking;neuralnetwork1IntroductionRobotarc-weldingrepresentsawidspreadtechniqueinthefieldofmetalmanufacture.About40%ofallindustrialrobotsarebeingusedforweldingtasks[1].However,theincreasingtrendtowardsmechanizationcannotbeeasilyextendedtowardsarc-welding,becausetheweldingprocessissubjecttoagreatnumberofinfluencingfactorssuchasheatradiation,arcflares,spatters,fumes,nottosaythehighlynonlinearandcoupledcharacteristicsofarc-weldingrobot.Thoughtherearemoreandmoreadvancedweldingrobotsaswellascontrolmethods,themajorproblemappearstobetheseamtrackingandthecontrolofweldingparameters.Theintelligentseamtrackingabilityisstronglyrequiredbecausethepositionalerrorscausedbythermaldistortionofworkpiecesduringweldinganddimensionalvariationsoffixturespreventtheformationofgoodwelds.Whenanindustrialrobotisusedinarc-weldingapplications,therobotguidestheweldingtorchalongtheweldingseam.Itisrequiredthattheweldsareinthecorrectposition,onlysmalldeviationsbeingacceptable.Thiscannormallybeachievedwiththehelpofnotonlyagoodvisionsensorsystembutalsoanaccurateroboticcontrol.Duringrecentyearsmanypowerfulrobotcontrolconceptsandalgorithmshavebeenproposed.However,mostroboticcontrolmethodsarebasedonanalyticalrobotmodels[2,3].Basedonthesemathematicalmodelsservocontrollerscanbedesignedwhichareassociatedwithdifferentoptimizationcriteria.Itiswellknownthatthepracticalefficiencyofthesemodel-basedcontrolconceptsdependsstronglyontheaccuracyinwhichthemodelrepresentstheactualstaticanddynamicsystembehavior.Duetothestrongkinematicnonlinearities,dynamiccouplingofadjacentrobotjointsandtheunavoidableelasticitiesofthemechanicalstructure,almostallmethodsofmodelidentificationarebasedonthesimplifiedassumptionoflinearsystemresponse.Insuchcasescontrolconceptsbasedonsuchmodelscanonlyprovideinsufficientperformanceandpoorrobustness.Intelligentcontrolhasastrongimpactonrobotics,inthatflexibilityandheuristiccharacteristicsofnewcontroltechniques,suchasfuzzylogicandneuralnetworks(NNs),haveessentiallychangedthemethodologyoftraditionalcontrollerdesign.Therecentresurgenceofresearchandapplicationsofartificialneuralnetworkstoadiverserangeofdisciplinesmakesitpossibletoseekoutsolutionsforroboticproblems.ANNisahighlyparalleldynamicalsystemwiththetopologyofadirectedgraphthatcancarryoutinformationprocessingbymeansofitsstableresponse.Whatmakesaviabletoolisthefactthatautonomoussystem,suchasrobotmanipulators,requireahighdegreeofflexibilitytodealwithsignificantvariationsintheenvironment.Thesevariationsareoftenunpredictableanddifficulttoformulatewithtraditionalmathematicalapproaches.TheNNsaretypicallyusedtomodelthehighlynonlinearstructuredandunstructureduncertaintiesofrobotdynamics,andtheNNmodelisusedtogenerateacompensatingtorque[4].Whiletheanalyticalformulationofeachcontrolschemeisamajortask,itsapplicationtoanactualrobotisanotherimportanttask.AlthoughmanypapersonNNcontrolofrobotmanipulatorshavebeenpublished,experimantalworkonanactualarc-weldingrobotreportedisrare.Inthispaper,wepresentaNNcompensatorforarc-weldingroboticseamtracking.Unlikethewidespreadcompensationinjointspacecontrolproblems[5],westudiedtheCartesianspacecontrolproblemswhicharemorecomplexbutappropriatetoanactualarc-weldingprocess.ItisdemonstratedviasimulationsandexperimentsthattheaccuratetrackingimprovementisobtainedbyapplyingtheNNcompensationatthereferenceseamtrajectorylevelinsteadofthatatthejointtorque/forcelevel.Thisapproachisveryattractiveinpracticeweldingbecauseitcanbeincorporatedintheseamtrajectoryplannerofanyexistingarc-weldingrob
本文标题:弧焊机器人焊缝跟踪神经网络控制器
链接地址:https://www.777doc.com/doc-4353247 .html