搜索
26520115JournalofAerospacePowerVol.26No.5May2011:10008055(2011)05117807佘云峰,黄金泉,鲁峰(,210016):,.,,,,.,.:;;;;;:V2337:A:20100426;:20100704:(1986-),,,,.GaussianweightedsumclusteringmethodofadaptivehybriddiagnosticmodelforturboshaftengineSHEYunfeng,HUANGJinquan,LUFeng(CollegeofEnergyandPowerEngineering,NanjingUniversityofAeronauticsandAstronautics,Nanjing210016,China)Abstract:AGaussianweightedsumclusteringmethodofonboardadaptivehybridmodelwasproposedforhealthparameterestimationofturboshaftengineinthefullflightenvelope.TheonboardadaptivehybridmodelwascomposedofKalmanfilterandneuralnetwork.Gaussianweightedsummodelwasusedtoclusterahugeamountofflightdatainrealtimeandupdatetheneuralnetworkweighting.Theworkingrangeofadaptivehybridmodelwasincreasedautomaticallywiththismethod.Thesimulationresultsshowthevalidityoftheproposedmethod.AdaptivehybridmodelbasedonGaussianweightedsumclusteringincreasestheaccuracyforturboshaftengineperformancetrackinginthefullflightenvelope.Keywords:turboshaftengine;Gaussianweightedsummodel;onboardadaptivehybridmodel;performanceparameterestimation;Kalmanfilter;neuralnetwork,,,.,.,,,.5:,,,..1972Urban.30,,[1][2][3][4][5][67].,F119F135.,[811].,,,[12].,,,.,,.1.1.,.1).uuc,ysycr=yc-ys,uc,r,,r=f(uc).2).,.1Fig.1Structureofhybridmodelx=AL00x+B0u+y=[CM]x+Du+(1)=[cWcgWgp]T,,.x=[ngnp]T,y=[ngnpT3p3T45]T,,.A,B,C,D,L,M,.,.,np.,5(2).HMa,20.LevenbergMarquardt.1179262Fig.2Structureofneuralnetwork,,0.11000m12,,.(5),12(60),12.,,.2,.,,,.,,.:,,,.,.,,.,,.,,,.2.1nX={x1,x2,,xn}xi=[uTi,rTi]T,i=1,2,nui=[ui(1),ui(2),ui(3)]T=[H,Ma,]Tri=[ri(1),ri(2),,ri(5)]Tri15.KKP(x)=Kk=1wkgk(x,x-k,k)(2)wkk,Kk=1wk=1,x-k=[u-k,r-k]T,k=[(u)k,(r)k]Tk.gkkgk(xi;x-k,k)=1(2)d|k|exp-12(xi-x-k)T-1k(xi-x-k)(3)(3),.,k,u-k,i(j)=i-1iu-k,i-1(j)+1iuk,i(j)(u)k,i(j)=i-2i-1[(u)k,i-1(j)]2+[u-k,i-1(j)]2-ii-1[u-k,i(j)]2+1i-1[uk,i(j)]2j=1,2,3(4)r-k,i(j)=i-1ir-k,i-1(j)+1irk,i(j)(r)k,i(j)=i-2i-1[(r)k,i-1(j)]2+[r-k,i-1(j)]2-ii-1[r-k,i(j)]2+1i-1[rk,i(j)]2j=1,2,,5(5)11805:u-k,i(j)(u)k,i(j)kij.r-k,i(j)(r)k,i(j).,imin,i=imin,(k),(,uk,0,k,0),ui(iimin),ui,|ui(j)-uk,0(j)|a(j)k,0(j)j=1,2,3(6)a(j)j.(6),,(4)(5);(6),,(u-k,(u)k)(r-k)ik,,.2.21),,,,.2),,.,,R=argmini=1,2,,Nc(Hi-H)2+(Mai-Ma)2(7)Nc,Nc=12.3),,.4),.|ui(j)-u-l(j)|a(j)l(j)l=1,2,,nj=1,2,3(8)n.(8),,,.,2),3),4).5),,(6),,,.mn|u-m(j)-u-n(j)|min(u)m(j),(u)n(j)j=1,2,3(9)mn,u-mn=u-mim+u-ninim+in,imn=im+inmn=1imn-1[(im-1)2m+imu-2m+(in-1)2n+inu-2n]-imnimn-1u-2mn(10),,,.6).3,.3.13~5.6.6,,,,.,imin=200;a=[3,3,3],194,71181263Fig.3Variationcurveofaltitudewithtimeinthefirstflight4Fig.4VariationcurveofMachnumberwithtimeinthefirstflight5Fig.5Variationcurveoftorqueanglewithtimeinthefirstflight3,157,72,7157.,,17,17(6()Fig.6Kalmanfilter(KF)sestimationwithoutneuralnetwork7Fig.7Clustersinthefirstflight,).,8.,,(29534,59),.8()Fig.8KFsestimationwithneuralnetwork11825:3.25%,.9~10.,10:0~170s;170~335s;335~425s;425~500s;500~650s,10.,,,,,5%.,,,,.,121,15,1115.9Fig.9Variationcurveoftorqueanglewithtimeinthesecondflight10Fig.10FlightdatapointsofaltitudeandMachnumberinthesecondflight11Fig.11Newaddedclustersinthesecondflight,1(72),.,,12.125%Fig.12KFsestimationwith5%degradedincompressorefficiency4,,,,.,..,118326.,,.:[1]GanguliR.Applicationoffuzzylogicforfaultisolationofjetengines[R].ASME2001GT13,2001.[2],.[J].,2008,23(1):163168.ZHAOShirong,HUANGXianghua.Faultdiagnosisforaeroenginegaspathcomponentsbasedonneuralnetworkmultisensordatafusion[J].JournalofAerospacePower,2008,23(1):163168.(inChinese)[3]JensenFV.Bayesiannetworksanddecisiongraphs[M].NewYork:SpringerVerlag,2001.[4],.[J].,2009,24(4):880885.LUFeng,HUANGJinquan.AGALSSVRcompensationforonboardselftuningmodelofaeroenginemodel[J].JournalofAerospacePower,2009,24(4):880885.(inChinese)[5],.[J].,2009,24(7):16491653.LUFeng,HUANGJinquan.Researchonperformancefaultfusiondiagnosisofaeroenginecomponent[J].JournalofAerospacePower,2009,24(7):16491653.(inChinese)[6]VolponiAJ,BrothertonT.Abootstrapdatamethodologyisforsequentialhybridenginemodelbuilding[R].BigSky,MT:IEEEAerospaceConferenceProceedings,2005.[7]VolponiAJ.Enhancedselftuningonboardrealtimemodel(eSTORM)foraircraftengineperformancehealthtracking[R].NASA/CR2008215272,2008.[8]LuppoldR,BrothertonT,VolponiA.Adaptiveonwinggasturbineengineperformanceestimation[R].BigSky,MT:IEEEAerospaceConferenceProceeding,2007.[9]FrithP,KarvounisG.ModelbaseddecisionsupporttoolsforT700enginehealthmonitoring[R].Melbourne,Australia:DSTOInternationalConferenceonHealthandUsageMonitoring,2001.[10]LittJS,SimonDL.Towardarealtimemeasurementbasedsystemforestimationofhelicopterenginedegradationduetocompressorerosion[R].NASA/TM2007214843,2007.[11]VolponiAJ,DePoldH,GanguliR.TheuseofKalmanfilterandneuralnetworkmethodologiesingasturbineperformancediagnostics:acomparativestudy[J].JournalofEngineeringforGasTurbineandPower,2003,125(10):917924.[12],.[J].,2008,23(3):581584.HUANGWeibin,HUANGJinquan.Onboardselftuningmodelforaeroenginefaultdiagnostics[J].JournalofAerospacePower,2008,23(3):581584.(inChinese)1184