机械外文文献及翻译.docx

机械外文文献及翻译.docx

《机械外文文献及翻译.docx》由会员分享，可在线阅读，更多相关《机械外文文献及翻译.docx（12页珍藏版）》请在冰点文库上搜索。

机械外文文献及翻译

与机械相关的外文及翻译

MultidisciplinaryDesignOptimization

ofModularIndustrialRobotsby

UtilizingHighLevelCADTemplates

1、Introduction

Inthedesignofcomplexandtightlyintegratedengineeringproducts,itisessentialtobeabletohandleinteractionsbetweendifferentsubsystemsofmultidisciplinarynature[1].Toachieveanoptimaldesign,aproductmustbetreatedasacompletesysteminsteadofdevelopingsubsystemsindependently[2].MDOhasbeenestablishedasaconvincingconcurrentdesignoptimizationtechniqueindevelopmentofsuchcomplexproducts[3,4].

Furthermore,ithasbeenpointedoutthat,regardlessofdiscipline,basicallyallanalysesrequireinformationthathastobeextractedfromageometrymodel[5].Hence,accordingtoBow-cutt[1],inordertoenableintegrateddesignanalysisandoptimizationitisofvitalimportancetobeabletointegrateanautomatedparametricgeometrygenerationsystemintothedesignframework.Theautomatedgeometrygenerationisakeyenablerforso-calledgeometry-in-the-loop[6]multidisciplinarydesignframeworks,wheretheCADgeometriescanserveasframeworkintegratorsforotherengineeringtools.

Toeliminatenoncreativework,methodsforcreationandautomaticgenerationofHLCthavebeensuggestedbyTarkian[7].TheprincipleofhighHLCtsissimilartohighlevelprimitives（HLP）suggestedbyLaRoccaandvanTooren[8],withtheexceptionthatHLCtsarecreatedandutilizedinaCADenvironment.Otherwise,thebasicsofbothHLPandHLCtcan,assuggestedbyLaRocca,becomparedtoparametricLEGOVRblockscontainingasetofdesignandanalysisparameters.Theseareproducedandstoredinlibraries,givingengineersoracomputeragentthepossibilitytofirsttopologicallyselectthetemplatesandthenmodifythemorphology,meaningthe

shape,ofeachtemplateparametrically.

2、MultidisciplinaryDesignFramework

MDOisa“systematicapproachtodesignspaceexploration”[17],theimplementationofwhichallowsthedesignertomaptheinterdisciplinaryrelationsthatexistinasystem.Inthiswork,theMDOframeworkconsistsofageometrymodel,afiniteelement（FE）model,adynamicmodelandabasiccostmodel.Thegeometrymodelprovidestheanalysistoolswithgeometricinput.Thedynamicmodelrequiresmasspropertiessuchasmass,centerofgravity,andinertia.TheFEmodelneedsthemeshedgeometryoftherobotaswellastheforceandtorqueinteractionsbasedonresultsofdynamicsimulations.

Highfidelitymodelsrequireanextensiveevaluationtimewhichhasbetakenintoaccount.ThisshortcomingisaddressedbyapplyingsurrogatemodelsfortheFEandtheCADmodels.Themodelsarebrieflypresentedbelow.

2.1HighLevelCADTemplate—GeometryModel

Traditionally,parametricCADismainlyfocusedonmorphologicalmodificationsofthegeometry.However,thereisalimittomorphologicalparameterizationasfollows:

•Thegeometriescannotberadicallymodified.

•Increasedgeometriccomplexitygreatlyincreasesparameterizationcomplexity.

ThegeometrymodeloftherobotisgeneratedwithpresavedHLCts,createdinCATIAV5.Thesearetopologicallyinstantiatedwithuniqueinternaldesignvariables.Topologicalparameterizationallowsdeletion,modification,andadditionofgeometricelementswhichleadstoamuchgreaterdesignspacecaptured.ThreetypesofHLCtsareusedtodefinetheindustrialrobottopologically;DatumHLCtwhichincludeswireframereferencesrequiredforplacementfortheActuatorHLCTsandStructureHLCts,asseenFig.2.

Fig.2Anindustrialrobot（left）andamodularindustrialrobot（right）

ThenamesofthereferencesthatmustbeprovidedforeachHLCtinstantiationarestoredintheknowledgebase（seeAppen-dixA.4）,whichissearchedthroughbytheinferenceengine.InAppendixA,pseudocodeexamplesdescribeshowthereferencesareretrievedandhowtheyarestoredintheknowledgebase.

Theprocessstartsbytheuserdefiningthenumberofdegreesoffreedom（DOF）oftherobot（seeFig.3）andisrepeateduntilthenumberofaxis（i）isequaltotheuserdefinedDOF.InordertoinstantiatethefirstStructureHLCt,twoDatumandtwoactuatorinstancesareneeded.ReferencesfromthetwoDatuminstanceshelporientingthestructureinspace,whilethegeometriesoftheactuatorinstances,atbothendsofthelink,areusedtoconstructtheactuatorattachments,asseeninFigs.2and3.Fortheremaininglinks,onlyonenewinstanceofbothdatumandactuatorHLCtsarerequired,sincethedatumandactuatorinstancesfromadjacentlinksarealreadyavailable.AppendixA.2showsapseudocodeexampleofaninstantiationfunction.ThefirstinstantiateddatumHLCtisdefinedwithreferencetotheabsolutecoordinatesystem.TheremainingdatumHLCtinstancesareplacedinasequentialorder,wherethecoordinatesystemofpreviousinstancesisusedasreferencefordefiningthepositioninspaceaccordingtouserinputs（seealsoAppendixA.3）.Furthermore,thetypeofeachactuatorandstructureinstanceisuserdefined.

Fig.3ThehighlevelCADtemplateinstantiationprocess

SinceitispossibletocreatenewHLCtsintheutilizedCADtool,theusersarenotforcedtomerelychoosefromthetemplatesavailable.NewHLCtscanbecreated,placedinthedatabaseandparametricallyinsertedintothemodels.

2.2DynamicModel

Theobjectiveofperformingdynamicsimulationofarobotistoevaluatesystemperformance,suchaspredictingaccelerationandtimeperformance,butitalsoyieldsloadsoneachactuatedaxis,neededforactuatorlifetimecalculationsandsubsequentstressanalysisbasedonFEcalculations.ThedynamicmodelintheoutlinedframeworkisdevelopedinModelicausingDymola,anditconstitutesaseven-axisrobotarmbasedontheModelicaStandardlibrary[18].

Thedynamicmodelreceivesinputfromthegeometrymodel,aswellasprovidingoutputtotheFEmodel,whichisfurtherdescribedinSec.2.3.However,tobetterunderstandthecouplingsbetweenthemodels,theNewton–Eulerformulationwillbebrieflydiscussed.Inthisformulation,thelinkvelocitiesandaccelerationareiterativelycomputed,forwardrecursively

Whenthekinematicpropertiesarecomputed,theforceandtorqueinteractionsbetweenthelinksarecomputedbackwardrecursivelyfromthelasttothefirstlink

2.3FESurrogateModel

Tocomputethestructuralstrengthoftherobot,FEmodelsforeachrobotlinkiscreatedutilizingCATIAV5,seeFig.4.ForeachHLCt,meshandboundaryconditionsaremanuallypreprocessedinordertoallowforsubsequentautomationforFE-modelcreation.ThetimespentonpreprocessingeachFE-modelisthusextensive.Nonetheless,theobtainedparametricFE-modelpaveswayforautomatedevaluationofawidespanofconcepts.Eachrobotlinkisevaluatedseparatelywiththeloadconditionsextractedfromthedynamicmodel.Theforce（fi-11andfi）andtorque（ţi-1andti）areappliedonthesurfaceswheretheactuatorsareattached.

2.4GeometricSurrogateModels.

Surrogatemodelsarenumericallyefficientmodelstodeterminetherelationbetweeninputsandoutputsofamodel[19].Theinputvariablesfortheproposedapplicationarethemorphologicalvariablesthicknessandlinkheightaswellasatopologicalvariableactuatortype.Theoutputsofthesurrogatemodelsaremassm,InertiaI,andcenterofgravityri,ci.

Toidentifythemostsuitabletypeofsurrogatemodelfortheoutlinedproblem,arangeofsurrogatemodelstypesarecreatedandevaluatedusing50samples.Theprecisionofeachsurrogatemodeliscomparedwiththevaluesoftheoriginalmodelwith20newsamples.Thecomparisonismadeusingtherelativeaverageabsoluteerror（RAAE）andrelativemaximumabsoluteerror（RMAE）asspecifiedbyShanetal.[20],aswellasthenormalizedrootmeansquareerror（NRMSE）,calculatedasseeninEq.（3）.Allprecisionmetricsaredesiredtobeaslowaspossible,sincelowvaluesmeanthatthesurrogatemodelisaccurate

TheresultingprecisionmetricscanbeseeninAppendixBandthegeneralconclusionisthatanisotropickriging[21],neuralnetworks[22],andradialbasisfunctions[23]arethemostpromisingsurrogatemodels.Toinvestigatetheimpactofincreasingnumberofsamples,additionalsurrogatemodelsofthosethreearefittedusing100samples,andtheresultscompiledinAppendixB.TheresultingNRMSEsfor50and100samplesforanistotropickriging,neuralnetworks,andradialbasisfunctionscanbeseeninFig.5.Thefiguresinsidetheparenthesesindicatethenumberofsamplesusedtofitthesurrogatemodels.

Fig.5GraphoftheNRMSEsfordifferentsurrogatemodels,

fittedusing50and100samples

AccordingtoFig.5,anisotropickrigingoutperformstheothersurrogatemodelsandthedoublingofthenumberofsamplesusedforfittingthesurrogatemodelincreasestheprecisiondramatically.

2.5FESurrogateModels

ForgeneratingFEsurrogatemodels,theanisotropickrigingwasalsoproventobethemostaccuratecomparedtothemethodsevaluatedinSec.2.4.Here,onesurrogatemodeliscreatedforeachlink.Inputsarethickness,actuators,force（fi-11andfi）andtorque（ţi-1andti）.Theoutputforeachsurrogatemodelismaximumstress（MS）.Ameanerrorofapproximately9%isreachedwhenrunning1400samplesforeachlink.Thereasonforthevastnumberofsamples,comparedtogeometrysurrogatemodels,hastodowithamuchlargerdesignspace.

利用高水平CAD模板进行

模块化工业机器人的多学科设计优化

1介绍

指出,除了规则,基本上所有的分析都需要信息,而这些信息需要从一个几何模型中提取。

因此,根据Bowcutt[1]中,为了使综合设计分析和优化，最重要的是能够将在设计的复杂和紧密集成的工程产品的过程中,必须要有能力处理不同的子系统的多学科性质之间的相互作用。

达到一个最优的设计,一个产品必须被视为一个完整的系统,而不是正在开发子的独立系统。

此外,已经一个自动化的参数化几何生成系统融入到设计框架中。

自动化的几何生成对于所谓几何循环多学科设计框架是一个关键驱动因素，在这个框架中CAD几何图形可以作为框架连接者来连接其他工程工具。

消除没有创新的工作，Tarkian已经提出了创造和生成HLCt的方法。

高HLCt