




全文预览已结束
版权说明:本文档由用户提供并上传,收益归属内容提供方,若内容存在侵权,请进行举报或认领
文档简介
StudyonNeuralNetworksControlAlgorithmsfor AutomotiveAdaptiveSuspensionSystems L J Fu J G Cao SchoolofAutomobileEngineering Chongqing InstituteofTechnology XingshengRoadNo 04Yangjiaping Chongqing China400050 E mail flj Abstract Thesemi activesuspension whichconsistsof passivespringandactiveshockabsorberinthelightof differentroadconditionsandautomobilerunningconditions is themostpopularautomotivesuspensionbecauseactive suspensioniscomplicatedinstructureandpassivesuspension cannotmeetthedemandsofvariousroadconditionsand automobilerunningconditions Inthispaper aneurofuzzy adaptivecontrolcontrollerviamodelingofrecurrentneural networksofautomotivesuspensionispresented Themodeling ofneuralnetworkshasidentifiedautomotivesuspension dynamicparametersandprovidedlearningsignalsto neurofuzzyadaptivecontrolcontroller Inordertoverify controlresults amini busfittedwithmagnetorheologicalfluid shockabsorberandneurofuzzycontrolsystembasedonDSP microprocessorhasbeenexperimentedwithvariousvelocity androadsurfaces Thecontrolresultshavebeencompared withthoseofopenlooppassivesuspensionsystem These resultsshowthatneuralcontrolalgorithmexhibitsgood performancetoreductionofmini busvibration I INTRODUCTION Themainfunctionsofautomotivesuspensionsystemare toprovidesupporttheweightofautomobile toprovide stabilityanddirectioncontrolduringhandlingmaneuvers andtoprovideeffectiveisolationfromroaddisturbances Thesedifferenttasksleadtoconflictingdesignrequirements Thesemi activesuspension whichconsistsofpassive springandactiveshockabsorberwithcontrollabledamping forceinthelightofdifferentroadconditionsandautomobile runningconditions isthemostpopularautomotive suspensionbecausetheactivesuspensioniscomplicatedin structureandconventionalpassivesuspensioncannotmeet thedemandsofdifferentroadconditionsandautomobile runningconditions Simi activesuspensionwithvariable magnetorheological MR fluidshockabsorbershassome advantagesinreducingautomobilevibrationatrelativelow castandpower Sofar thereareanumberofcontrol methodsthathavebeendevelopedforsemi active suspension startwithskyhookmethoddescribedby Karnoopp etal l Thismethodattemptstomaketheshock absorberexertaforcethatisproportionaltotheabsolute velocitybetweensprungmasses Someinvestigationsuse C R Liao B Chen SchoolofAutomobileEngineering Chongqing InstituteofTechnology XingshengRoadNo 04Yangjiaping Chongqing China400050 E mail chenbao linearsuspensionmodel whichislinearizedaroundthe operationalpoints andcontrolalgorithmarederivedusing linearmodels suchasLQGandrobustcontrol 2 3 These controlmethodscannotmakeafullexploitationof semi activesuspensionresourcesbecauseofautomotive suspensionisinherentnon linearperformance Inorderto improveperformanceofnonlinearsuspensionsystem some intelligentcontroltechniques suchasfuzzylogiccontrol neuralnetworkscontrolandneurofuzzycontrol havebeen recentlyappliedtononlinearsuspensioncontrolby researchers 4 5 Inthispaper aneurofuzzyadaptivecontrolcontrolleris appliedtocontrolsuspensionvibrationviamodelingof recurrentneuralnetworksofautomotivesuspensionand continuouslyvariableMRshockabsorbers Thecontroller structuresdesignandneurofuzzycontrolalgorithmsare presentedinsection2 Arecurrentneuralnetworks dynamicsmodelingofsuspensionareshownrespectivelyin section3 Thecontrolsystemexperimentationsaregivenin section4andsomeconclusionsarefinallydrawninsection 5 HI NEUROFUZZYADAPTIVECONTROLALGORITHMSFOR AUTOMOTIVESUSPENSIONS Theneurofuzzycontrolsystempresentedinthispaper showninFigure1 iscomposedofaneurofuzzynetwork andarecurrentneuralnetworkmodelingofmini bus suspension Theneurofuzzynetworkisdefinedasadaptive controller whichhasfunctionoflearningandcontrol The functionofrecurrentneuralnetworkistoidentifymini bus suspensionmodelparameters y t andyd t aresystem actualoutputandsystemdesireoutputrespectivelyinFigure 1 xl t issystemerrorofsystemactualoutputbetween systemdesireoutput x2 t issystemerrorrateofsystem actualoutputbetweensystemdesireoutput xi t and x2 t aredefinedasfellows xI t e t y t Yd t 1 X2 t e t e t 1 e t 2 0 7803 9422 4 05 20 00C2005IEEE 1795 Fig 1 structureofneuralnetworkscontrolsystemforsuspension networkscontrolsystem Theglobalsetsoflinguistic variablesaredefinedrespectivelyasfellows E E 1 AtJ uU U U Theneurofuzzycontrollerhas fourlayersne urons inwhichthefirstandthesecondlayers correspondtothefuizzyrulesif part thethirdlayer correspondstotheinferenceandtheforthlayercorresponds tothefuzzyrulesthen part Thesetsxl x2and uare respectivelydivinedintosevenfuzzysubsetsofwhichfuzzy setsX1 X2Uarecomposedasfallowsrules X1 NB NM NS ZE PS PM PB X2 NB NM NS ZE PS PM PB U NB NM NS ZE PS PM PB Inthispaper theGaussianmembershipfunctionareused inelementsoffuzzysets X1 X2andtheelementsof fuzzysetUisdefinedasfollowingmembershipfunction ci u J0 otherwise 0 3 I 3 k 1 2 3 49 j 13 23 3 7 4949 Layer4 4 3 wkand0 4 I 4 0 3 k 1k 1 Wherexl t x2 t aretheinputsofneuralnetworks wkisweightofneuralnetwork 0 4 iStheoutputof neuralnetworksinwhich0 4 U ai b j arethecentral valuesofGaussianmembershipfunction Learning algorithmsoftheneuralnetworkscontrollerisbasedon gradientdescentbymeansoferrorsignalback propagation method Theerrorback propagationalgorithm saccomplish synapticweightadjustmentthroughminimizationofcost function 5 m ALGORITHMFORRECURRENTNEURALNETWORKS SUSPENSIONDYNAMICALMODELING Arecurrentneuralnetworkdesignedtoapproximateto theactualoutputofsuspension y t isthree layerneural networkwithonelocalfeedbackloopinthehiddenlayer whosearchitecturesareshowninFigure3 Thepropertythat isofprimarysignificanceforrecurrentneuralnetworkisthe abilityofthenetworktolearnfromitsenvironmentandto improveitsperformancesbymeansofprocessof adjustmentsappliedtoitsweights Therecurrentnetwork withinputsignal II t u t andI2 t y t 1 has outputy t bylocalfeedbackloopneuroninthehidden layerwhoseoutputsumis Sj t correspondingtothe neuronjth 3 Fig 2 schematicofneuralnetworkscontrollerforadaptivesuspension WhereU Eu Theinput outputispresentedasfollows accordingtoFigure2 Layer1 I 1 x t andO xi t i 1 2 Layer2 I 2 t ai 2 b 2 and O epx i 1 2j 1 2 3 7 Layer3 I 13 t u X2Q I and Fig 3 schematicofneuralnetworksmodelingofsuspensionsystem 4 Sy w i t WJD Xj t 1 i1 i t wjXj t l q yj t 1w xi t j l 5 6 1796 wherew I w areweightoftherecurrentneural network Xj t is outputofneuronwithlocalfeedback loopneuroninthehiddenlayer p qareinputneuron numberandfeedbackneuronnumberrespectively The activationfunctionforbothinputneuronsandoutput neuronsislinearfunction whiletheactivationforneurons inthehiddenlayerissigmoidfunction heobjectivefunctionE t canbedefmedinthetermsof theerrorsignale t as E t y t y t 2 1e2 t 7 22 Thatis E t istheinstantaneousvalueoftheerror energy Thestep by stepadjustmentstothesynapticweights ofneuronarecontinueduntilthesystemreachsteadystate i e thesynapticweightsareessentiallystabilized DifferentiatingE t withrespecttoweightvectorw yields aE t 8 e t 0Y 8 Fromexpression 1 2 and 3 differentiatingA t 0DI withrespecttotheweightvectorw1 w w Yrespectively yields aS t x t As t woax1Q W aXI t aWj J aWj From 4 5 and 6 analyzing valueofsynapticweightisdeterminedby w t 1 w t q e t 89 t 12 whereqtheleaning rateparameter Adetailed convergenceanalysisoftherecurrenttrainingalgorithmis rathercomplicatedtoacquiretheleaning rateparameter value Accordingtoexpression 13 theweightvectorw forrecurrentneuralnetworkcanbeadjusted Weestablisha theLyapunovfunctionasfollowsV t 1 2 e2 t whosechangevalueAV t canbedeterminedaftersome titerations inthesensethat 13 Wehavenoticedthattheerrorsignale t aftersomet iterationscanbeexpressedasfollowsfromexpression 13 and 14 ae t ao t ae t ae t Aw qe t 77e t the aw O waw O w Lyapunovfunctionincrementcandeterminedaftersomet iterationsasfollows 14 Mtt q t v2 e t V t where t 2 2jt1 6 t 2 A 10 lp q 2 5l0 t ll 2 ql 2 77 O 220w 9 7maxa t 29 if q f 2 thenAV t O w ax1 t D and aWj x1 t uxi yieldsrespectivelyrecurrentformulas ax1 t a f S t FX x tt1 1 ax1 O WjD axi t aNi afS t w a t i 4 LaN i ax1 o 11 avn 0 Havingcomputedthesynapticadjustment theupdated namelytherecurrenttrainingalgorithmisconvergent IV ROADTESTANDRESULTSANALYSES Tomakeademonstrationthevalidityofneuralcontrol algorithmproposedinthepaper anexperimentalmini bus suspensionwithMRfluidshockabsorberhasbeen manufacturedinChina Themini busadaptivesuspension systemconsistsofaDSPmicroprocessor 8acceleration sensors 4MRfluidshockabsorbers and 1controllable electriccurrentpowerwithinputvoltage12V TheDSP microprocessorreceivessuspensionvibrationsignalinput fromaccelerometersmountedrespectivelysprungmassand un sprungmass Inaccordancewithvibrationsignaland controlschemeinthispaper theDSPmicroprocessor adjustsdampingofadaptivesuspensionbyapplication controlsignaltothecontrollableelectriccurrentpower connectedtoelectromagneticcoilinMRfluidshock absorbers Magneticfieldproducedbytheelectromagnetic coilinMRfluidshockabsorberscandvarydampingforce inbothcompressionandreboundbyadjustmentofflow 1797 II V t 12 t 1 e2 t 2 behaviorsofMRfluidsindampingchannels Raodtestonmini busadaptivesuspensionbasedneural networkscontrolpresentedinthispaperarecarriedoutinD classroadsurfacesrespectivelyinrunningvelocity 30 40 50km h Duringroadtest experimentalmini busruns eachtestconditionataconstantspeed Thetestexperiments ofadaptivesuspensionwithneuralnetworksandpassive suspensionsystemwerecarriedoutrepeatedlyundersame roadsurfaceandrunningvelocity TestresultslistedinTable 1haveshownthattheadaptivesuspensionwithneural networkscanreducevibrationpowerspectraldensitiesof bothsprungmassandun sprungmass Figure4isthemin bussuspensionvibrationpower spectraldensitiesofbothsprungmassandun sprungmass withpassiveandadaptivesuspensionsystembyDclass roadsurface Itisclearthatneuralnetworkscontrol improvesperformancesofmini bussuspensionwithmainly improvementsoccurringaboutsprungmassresonancepeak Thepowerspectraldensitiesindicatethattheadaptive suspensionsystemwithneuralnetworkscontrolcanreduce mini busvibrationgreatlycomparedwithpassive suspension Ifexcellentfizzycontrolrulesandrational modelingofshockabsorberandsuspensioncanbeobtained theadaptivesuspensionsystemwithneuralnetworkscontrol willimprovefartherridecomfortandroadholdingand handlingstabilityofautomobileinthefuture TABLEI min bussuspensionroadtestresults sprungmassandun sprungmassaccelerationr m s Values Dclassroad Speed30 1km h 40 1m h 50 kmlh PassiveControlreducePassiveControlreducePassiveControlreduce mass10 37560 325213 40 41400 344916 70 46940 396615 5 mass pg1 60111426610 91 89751 660312 52 34682 065212 0 mass IC 4a 1 t 0 ri 0110 1 lo1 Fry 0Qgco1okaId e la r10f1 Frcqv O Fig 4 min bussuspensionvibrationpowerspectraldensitiesofsprungmass left andun sprungmass right withcontrolandpassive runningspeed40km h V CONCLUSIONS Inthispaper anewrecurrentneuralnetworks oriented suspensionmodelandneurofuzzycontrolschemesforthe mini bussuspensionsystemwereinvestigated Uponthe requirementofusing8accelerationsensors aDSP controllerwithgainschedulingwasdeveloped ConsideringthecomplexityoftheMRfluidshock absorber theactuatordynamicshasbeenincorporated duringthehardware in the loopsimulations Itwas demonstratedthattheadaptivecontrolsystemcould 1798 achieveacompetitivecontrolperformancebyadoptingthe neurofuzzycontrolschemesandrecurrentneural networks orientedsuspension Becausethecontrollaw design thegainschedulingstrategy andthe hardware in the loopsimulationmethoddevelopedinthis paperarerestricted
温馨提示
- 1. 本站所有资源如无特殊说明,都需要本地电脑安装OFFICE2007和PDF阅读器。图纸软件为CAD,CAXA,PROE,UG,SolidWorks等.压缩文件请下载最新的WinRAR软件解压。
- 2. 本站的文档不包含任何第三方提供的附件图纸等,如果需要附件,请联系上传者。文件的所有权益归上传用户所有。
- 3. 本站RAR压缩包中若带图纸,网页内容里面会有图纸预览,若没有图纸预览就没有图纸。
- 4. 未经权益所有人同意不得将文件中的内容挪作商业或盈利用途。
- 5. 人人文库网仅提供信息存储空间,仅对用户上传内容的表现方式做保护处理,对用户上传分享的文档内容本身不做任何修改或编辑,并不能对任何下载内容负责。
- 6. 下载文件中如有侵权或不适当内容,请与我们联系,我们立即纠正。
- 7. 本站不保证下载资源的准确性、安全性和完整性, 同时也不承担用户因使用这些下载资源对自己和他人造成任何形式的伤害或损失。
最新文档
- 企业培训评估与反馈模板
- 员工培训资源清单及模板库
- 汽车维修与保养合作合同
- 2025广西钦州市北部湾大学公开招聘高层次人才53人模拟试卷及参考答案详解1套
- 借贷活动合规承诺书7篇
- 历史保护建筑修复质量承诺书3篇
- 山西省忻州市2024-2025学年高三上学期10月月考地理试题(解析版)
- 辽宁省凌源市2024-2025学年高一下学期期末考试地理试题(解析版)
- 使命彻底完成承诺书5篇
- 2025广西职业技术学院博士人才专项招聘64人模拟试卷及完整答案详解
- 页人音版三年级音乐上册音乐教案(2025-2026学年)
- 员工应急救护知识培训课件
- 2025昆明中北交通旅游(集团)有限责任公司驾驶员招聘(60人)考试参考题库及答案解析
- 2026中国航空工业集团金航数码校园招聘备考考试题库附答案解析
- 健康教育培训师资队伍建设方案
- 二类医疗器械零售经营备案质量管理制度
- 2025年医技三基考试试题及答案
- 既有建筑幕墙安全培训课件
- 2025年全国事业单位联考C类《职业能力倾向测验》试题及答案
- 英语A级常用词汇
- 气管切开非机械通气患者气道护理团体标准课件
评论
0/150
提交评论