Kazan, Russia - September 11-13, 2007
Conditional Averaging Methodology for Periodi
Data with Time Jitter and Spatial S atter
Berend G. van der Wall,OliverS hneider
DLR, Inst.of FlightSystems, Brauns hweig, Germany
The onditional averaging methodology is shown to be mandatory to proper average
essentiallyperiodi datathatarebiasedbyspatials atterandtimejitteree tswhi hare
always presentin both ightand windtunneltest.This isdemonstratedatvarious data
of ompletely dierent physi al origin as obtained from the Higher Harmoni Control
Aeroa ousti RotorTest(HARTII)performedintheLargeLow-SpeedFa ility(LLF)of
the German/Dut hWind Tunnel (DNW):blade positionand tip vortex ow eld data
(bothobtainedinstantaneouslyand withopti almeasurementte hniques),bladesurfa e
pressureanda ousti data(measured ontinuouslybyabsolutepressuretransdu ersand
mi rophones,respe tively).
1 Introdu tion
Inheli opterrotorexperimentaldata-eitheroriginatingfromthe yingheli opterorfromthewindtunnel
model-periodi andnon-periodi motionsofthetestvehi learepresentduetogustsandtransientmotion
oftheheli opterordueto exibilityofmodelsupportandunsteadinessinthewindtunnelenvironment.
Thesemotionsusuallyareoflow-frequen ynature omparedtotherotorrotationalfrequen y,but they
slightly hangetheaerodynami environmentrevolutionbyrevolution.Consequentlythedatalikerotor
bladeposition,bladepressure,vortexposition,mi rophonepressureexhibitdisturban esinspa eand/or
timethatmustproperlybeaddressedindataanalysis,espe iallywhenthetime(orspa e)averageddata
aretobe omputed.Theseaverageddata(velo ityelds,positionorpressuretimehistories)arethebasis
for odevalidation eortsandthusspe ialalgorithmsmustbeapplied toeliminatethesespatials atter
and/ortimejitteree tsappropriately.
Themostfrequentlyusedmethod ofaveragingis theso- alledsimple(= arithmeti )averaging.
Pro-videdthatanumberofperiodsN
p
(oneperiod=onerotorrevolution)havebeenre ordedforea hsensor
-favorablywitha onstantnumberofsamplesN
s
perperiodandadditionallyea hsampley
i
atthesame
phaselo ation
i
withinea hperiod(whi hisusuallythe aseinrotarywingdatathataretriggeredto
themain rotorazimuth with1/revasthedominatingfrequen y)-, su h analgorithmto omputean
averagesampley
i
(j); i=1;2;:::;N
s
ofatimehistoryreadsinageneralizedform
y i = m v u u t 1 N p Np X j=1 y m i (j) i=1;2;:::;N s (1) with m= 1 harmoni average m!0 geometri average
m=1 simple(arithmeti )average
m=2 quadrati average(oree tivevalue)
and y i;min y i (m= 1)y i (m!0)y i (m=1)y i (m=2)y i;max
Thesimpleaverage(m=1)iswidelyusedinstatisti alanalysisofvariationsintheindividualsamples
andeliminatesnoiseinapropermannerinordertoobtainasmoothtime-averagedhistory.Theharmoni
ofthesamplesis tobeinterpreted,thegeometri averagem!0isoftenused, thatis y i = N p v u u t Np Y j=1 y i (j) i=1;2;:::;N s (2)
Anotherimportantparametertojudgetheindividualsamples onden eintherelationtotheaverage
valueistomakeuseofthestandarddeviation,whi hisdenedas
i = v u u t 1 N p 1 N p X j=1 ( y i (j) y i ) 2 i=1;2;:::;N s (3)
In a Gaussian normal distribution, whi h is expe ted in most ases, about 68% of all samples are
withintherangeofy
i i ,about95%within y i 2 i
, andabout99.8%within y
i 3
i .
In ase the data are ontaminated by spurious elements being either by far too large and/or too
small omparedto themajorityofsamplesanotherpro edure shouldbeappliedrstbyeliminating the
samefra tion ofN
p
at both endsof thes alebefore applying oneof theaveragingmethods mentioned
before. This is alled the trun ated average,but the trun ation threshold(usually 5to 15%) must be
adapted manuallyto theindividualdistribution of dataand thusrequires somedegreeofexperien ein
order to distinguishbetweeen those data beingobviouslyerroneousand those that should beretained.
Oftenavaluebetween2 and3 is usedas athreshold toidentify those samplesas outliers.However,
the mathemati al formulation requires trun ation at both ends of the s ale with the same number of
samples,whi hin some asesdoesnotmat hthe physi softhedata.Some otherhigherorder variants
ofthetrun atedaverageexistthatuseweighingfun tions.
As anexamplefor instantaneous datathat are notavailable in timerather thanin spa ethe result
of ow eld measurementslikePIV an be seen.Herein neighboring data (velo ity ve tors) are
spa e- orrelated.Theprepro essingofparti leimagestovelo ityve torsusuallyleadstosomespuriousve tors
thatmustbeeliminatedbeforeanalysingthevelo ityve toreld.Theiridenti ationisbasedon
statis-ti alanalysisofa ertainsubsetofve torsaroundthatoneinquestion.On eidentiedasspurious,these
ve torsnormallyareofsigni antlylargermagnitudeanddierentorientation(althoughtheorientation
alone annot be used as indi ator, forexample theorientationnaturally hangesto the oppositewhen
passingthe enterofavortex)thanthoseve tors omputed orre tly,butrarely ofsigni antlysmaller
length.Thus, lipping thesamenumberofve torsatbothendsofthelengths alewouldeliminatealot
of orre tsamplesatthelowerendandappearsnotappropriatetotheproblem.
Another example is given with ontinuous data as obtained from strain gauges, pressure sensors
or mi rophones. Within ea h sensor time history the individual samples are time- orrelated to form a
ontinuous urve. Any spikes - like those aused by slip-ring problems or broken ables - an easily
be dete ted either by orrelation to the neighboring samples within an individual time history, or by
orrelationwiththesamplesatthesamephaseoftherestoftheperiodsmeasured.However,in asethis
is arepetetive problem o uring in ea h period relyingon the lattermethod alonewould notwork.A
spa e- orrelationmustalsoexistto ontinuousdataobtainedfrom anothersensorlo ated loseenough,
su hthatthisinformation ouldalsobeusedtoidentifyspuriousdata,but aremustbetakensin ethis
oftenin ludesnon-linearphaseshiftsbetweenthetimehistories.
So far, the averaging methods mentioned are mainly useful when statisti ally distributed, mainly
one-dimensional, data are to be averaged and no other information about the physi s behind has to
be obeyed. Yet, a fundamental problem annot be addressed by those averaging methods, and this is
the spatial s atter of individual events of interest. Applying any of the averagingmethods mentioned
wouldresultin arti iallysmoothingtheindividualeventswhi hmakesananalysisoftheseobsolete.To
illustratetheproblemseeFig.1.
As an be seen in this example, in ea h individual time history an identi al event is seen and the
nal goalis to providean averagedtimehistoryas indi atedby CA = onditionally average.The line
denotedSArepresentstheresultofsimpleaveragingusingEq.1withm=1anditis learlyvisiblethat
the individual hara teristi sare widely lost.In asethis was ameasurement of theindu ed velo ities
ofavortexwhenpassingtheprobeea hindividual analysiswouldleadto orre t oreradiusandswirl
velo ity,buttheaveragedtimehistorywouldgive ompletelywrongresultsforbothparametersdespite
a orre taveragepositionof theevent.Thereasonisthetime jitteroftheeventthat leadstoarti ial
-1,5
-1,0
-0,5
0,0
0,5
1,0
1,5
0
10
20
30
40
50
time, index
y
i
(j)
SA
CA
Figure1:Ee toftimejitteronaveragedtimehistory(SA-simpleaverage;CA- onditional average).
of the individual time history. To ir umvent the time jitter ee t, two possible solutions have been
elaboratedand usedinthepast:
1. in thespa e-timedomain: toidentify theparametersoftheeventin ea h individualmeasurement
andthenapplyoneoftheaveragingmethodstotheparametersidenties.However,theindividual
analysis oftenisbiasedorhinderedbynoiseandtheresultisasetofparameters,but noaveraged
timehistoryorspatialdistribution ofdataisavailable.
2. in the frequen y domain: for ea h individual period of the time history the Fourier spe trum is
omputed, thenall spe traare averaged.This isappropriatefor themagnitudesof thespe trum,
but thetimeinformation(=phase)getslostand againnoaveragedtimehistoryisavailable.
Theee t oftimejitterontheresultsofthespe trumand orre tionstoitin thefrequen ydomain
were elaborated, for example by [5℄. In this arti le an alternative method is presented known as the
onditionally averagingthat takes into a ount spatial s atter and time jitter and as a resultprovides
averagedspatial distributions or timehistories that eliminatenoisebutstill in lude alltheinformation
oftheindividualevent.Intimehistorydata,thegoalistogenerateaveragedtimehistories thathavea
Fourierspe trumsimilartotheaveragedspe trum.
2 The HART II data base
AlltheanalysismethodologiesareappliedtotheHARTIItestdataobtained2001intheDNWbyDLR,
ONERA, NASA Langley, US Army AFDD and DNW. These data en ompass measurements of wind
tunnel data,rotorbalan e,blademotion,bladepressure,a ousti radiationanddetailed tipvortex ow
eld data. The HARTII test is des ribedin [1℄ and details of thetest set-up and of themeasurement
te hniques applied are given in [2℄. Representativeresultsapplying partof the te hniques des ribed in
thispaperareshownin[3℄.Thesereportsandpartofthedataareavailablefortherotor raft ommunity
withintheframeworkoftheinternationalHARTIIworkshopheldsemi-annuallyatboththeAHSForum
andtheEuropean Rotor raftForum[4℄,sponsoredbytheHARTIIteam.
Within HARTII aMa h-s aledanddynami allys aledBo105modelrotorwithN
b
=4re tangular
blades at a pre- oneof
p
=2:5deg, having alinear twist of
tw
= 8deg=R , aradius of R =2m, a
hordof =0:121mandaNACA23012airfoilwithtrailingedgetabwas operatedin6degdes ent ight
ondition that is known to generate strong blade-vortexintera tion (BVI) noise.The blade rotational
speedwas =109rad=s andthe wind speedV
1
=33m=s, resultingin anadvan eratio of=0:151.
Thethrust oeÆ ientwassettoC
T
=0:0044representingalightlyloadedBo105andzerohubmoments.
ThesetupisshowninFig.2.
Oneofthemaingoalswastoidentify thephysi alme hanismsofhigherharmoni ontrol(HHC)on
noiseandvibrationredu tionbymeansofextensive oweldmeasurements.Threedistin tive onditions
weremainlyinvestigated,thesearethebaseline ase(referredtoasBL)withashaftangleof
S
=5:3deg
and zero roll and pit h moments. The appli ation of HHC features twovariants trimmed to the same
onditionastheBL ase:theminimumnoise ase(MN)with3/revpit h ontrolof
3
=0:8degat the
bladerootat aphaseof
3
=300deg,andtheminimumvibration ondition(MV)with thesamepit h
amplitude,butadierentphaseof
3
=180deg.These onditionswereidentiedintherstHARTtest
mentedBo105modelrotorwas usedwheretwobladeswereequippedwithin total51absolutepressure
transdu erspriortothetest.Theleadingedgedierentialpressureat3% hord ouldthusbemeasured
from 40-97%radius at 11 radial se tions,and the hordwise pressuredistribution at 87% radius ould
bere ordedbymeans of 17sensors assket hedin Fig. 2.By hordwiseintegration ofthese sensor
sig-nalsthe lo al lift timehistoryin termsof thenormal for e oeÆ ientC
n M
2
or lo al blade ir ulation
b =C
n
V =2 with V( ) as the time-varying airspeed at the se tion an be evaluated. 80 su essive
rotorrevolutionswerere orded,triggeredtotherotorazimuthatadatarateof2048samples/revwithin
ea h ondition.
(a) ModelrotorintheLLFoftheDNW
0
-2
-1.5
-1
-0.5
0.5
1
1.5
2
x
V
0.45m
0.5m
0
1
-1.5
-1
-0.5
0.5
1.5
y
hub
hub
/R
/R
hub
z
/R = 0.4575
z
mic
+
/R = -0.65
(b)Mi rophonemeasurementpositions
*
*
*
*
*
*
#
#
#
#
*
*
*
*
*
*
*
#
#
#
*
#
*
*
*
*
*
**
*
#
*
r/R
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
0.0
Absolute pressure transducers (51)
Strain gages: flap (3), lead−lag (2), torsion (1)
*
17
reference blade (25 Kulites)
preceeding blade (26)
( ) Distributionofpressuresensorsontheblades
.228
.318
.408
.498
.588
.678 .768
.858 .948
.993
.903
.813
.723
.633
.543
.453
.363
.273
r/R
5 6
7
21
22
23
40
Marker no.
(d)DistributionofSPRmarkersontheblades
Figure2:HARTII equipmentusedformeasurementofa ousti s,bladepressuresandbladede e tion.
A ousti pressuredata were re orded1:1075R belowthe rotorusing anarraywith 13 mi rophones
spa ed equallyat adistan eof 0:225Rthat was traversedforeandaftof therotor entreby arangeof
2Rat 17 downstream lo ationsseparatedby 0:25R , seeFig. 2.Atea htraverselo ation mi rophone
datawerere ordedfor100su essiverotorrevolutionsatadatarateof2048samples/rev,againtriggered
totherotorazimuth.Thepost-pro essingprovidesnoiselevelsateverylo ationbasedonspe tralanalysis
Thebladepositionwasmeasuredopti allyusingStereoPatternRe ognition(SPR)[9℄.Four ameras
weremountedonthe oortakingimagesoftherotorat24spe iedazimuthallo ationsequallydistributed
aroundtherevolution.Allfour bladeswereequippedwith 18markersallalongtheleadingedgeas well
as alongthetrailingedge asgivenin Fig.2.The omputationofthemarker enters bymeansofimage
pro essingandaproper alibrationresultedintheabsolutepositioninspa eofthesemarker enterswith
an a ura y of 0:4mm, whi h is 0.33% hord (or 0.02% radius). Sin e the ameras ould not measure
ontinuouslyratherthanwerelimitedbyamaximumfrequen yofabout10Hz,oneimagewas re orded
every7threvolutionandaminimumof50repeatswastakenateverybladeazimuthpositiontoallowa
statisti alanalysis.
The post-pro essing is des ribed in detail in [6℄ and allows the identi ation of ap and lead-lag
positionofthequarter hordlinealongthespanatea hazimuth.Puttingtogetherallazimuthalpositions
insequen ethetimehistoryofblademotion anbe omputedbymeansofFourieranalysisandsynthesis
for intermediate azimuth lo ations. The elasti deformation is obtained subtra ting the pre- one rigid
bladeposition.Dierentiating theleadingedgeandtrailingedge markerpositions providesinformation
abouttheradialdistribution ofthelo al pit h anglewith ana ura yof0:4deg.Whensubtra ting the
build-inpre-twistandthe ommandedbladerootpit h ontroltheelasti torsionis obtained.
Floweld measurementwereperformedusingStereoParti leImageVelo imetry(SPIVod3C-PIV)
[10℄. A large observation area of 0:45m0:37m was taken by DNW ameras for the global velo ity
distribution and globalwakestru ture analysis and simultaneouslyasmall observationareaof0:15m
0:13mwas gatheredbyDLR amerasforanalysis ofthebladetipvortexstru tures. Thepre-pro essing
of the amera images results in velo ity ve tor elds, while the post-pro essing is devoted to analyse
theseve tormapsrstforthevortexspatialposition,thenforthevortexpropertiesinorder toidentify
parameters like the ore radius r
, the maximum swirl velo ity at the ore radius V
and the swirl
velo ityproleV
s
(r).Otherparameterslikevortex ir ulationmaybe omputedbasedonthese results.
Themethodsofpost-pro essingaredex ribedin[7℄.
(a)PIVlo ations,MV (b)Rawdata PIVimage,BL,Pos.21,lower
amera
Figure3:PIVmeasurementpositionsandexemplaryrawdataimages.
As in SPR measurements, the ameras were triggered to the rotor azimuth and re orded an
im-age every7th revolution, with 100 repeats. A traversingsystem allowed to overthe entire rotor disk
outside jyj=R = 0:4 su h that on either side of the rotor disk the tip vorti es were tra ed from their
reation at the trailing edge downstream until the rear end of the disk in lateral planes lo ated at
y=R=0:4;+0:55;0:7;0:85;0:97.Thepositionsmeasuredareindi atedinFig.3attheexampleof
theMV ase,where dualvortexsystemsexistoverpartof therevolutiondueto downloadattheblade
tip itself.The rightgureexemplarilyshowsarawdata imageof theBL aseat position 21where the
fullsize representsthelargeobservation areaand theinnerimagerepresentsthesmallobservationarea
withaboutthreetimesthespatialresolution.
A se ondary post-pro essing puts together the downstream measurements of the tip vortex whi h
resultsin the vortextraje tory (andlo ations where thevortex passedthepath ofthe blades),as well
as theagingof thevortexby meansofthe developmentofits identiedparametersliker
or V
.Some
By means of stereo pattern re ognition te hnique (SPR) the spatial position of markers atta hed to
ea h of thefour blades and to the bottom of the fuselage (Fig. 4) was determined opti ally. The SPR
te hniqueis basedona3-dimensionalre onstru tion ofvisible markerlo ationsbyusingstereo amera
images.Thea ura yofmarkerpositionre ognitiondependsontheresolutionandangularset-upofthe
ameras and on the marker shape and size.For the onditions of the HART II test measurement the
theoreti alresolutionis0:4mminx ,y andz dire tion.Amoredetaileddes riptionofthemethodis
presentedin[6℄and[9℄.The amerasweretriggeredtotherotorazimuthand50,sometimes100repeats
perposition were re orded.Thedata ontainthelowfrequen ymotionof themodelas spatial s atter.
Thismotionis in reasingwithblade radialpositionsin etheairloads generatingtheblademotionare
alsosubje ttovarya ordingtothemodelmotion.
Figure4:SPRimageofthedownsreamright ameraat90deg
Togetsmoothdatawithredu ederrorsandeliminatedvibrationsitisne essarytodetermineaverages
ofthe oordinates.Toaveragethemarker oordinatesasimplemeanvalueforea h orre tlyre ognized
marker of all repeats was omputed. In Fig. 5(a)and Fig.5(b) an exampleis shown for twodierent
markersatthe90degazimuth position ofblade1ofthebaseline ase.
−0.1
0
0.1
−0.2
−0.1
0
0.1
0.2
100y/R
100z/R
σ
2
σ
σ
y
= 0.0463
σ
z
= 0.0202
(a)Bodymarker
−0.1
0
0.1
−0.2
−0.1
0
0.1
0.2
100y/R
100z/R
σ
2
σ
σ
y
= 0.0446
σ
z
= 0.1020
(b)Bladetipmarker(trailing
edge)at99%R
0.2
0.4
0.6
0.8
1
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0.08
0.09
0.1
0.11
r/R
100
σ
x
/R, 100
σ
z
/R
σ
x
LE
σ
y
LE
σ
z
LE
σ
x
TE
σ
y
TE
σ
z
TE
( ) Standarddeviationinx andz dire tionfor
leadingandtrailingedgemarkers
Figure5:S atterofbodyandblademarkersat90degBL
Atthelo ationofthebodymarker(leftgure)thes atteringisbyabout4mminy dire tion(lateral)
horizontaldire tionno hange ins atteringisvisible(the bladeisstiradially).
In Fig.5( ) the standarddeviations in alldire tions are shown for both theleading and trailing
edge markersdepending onthe bladeradius. It learly an beseenthat there isno dependen e of the
y s atterontheradiusduetolargeradialstinessoftheblade,itisnearly onstantandonlyin uen ed
by the low elasti ity of the wind tunnel sting. The x and z s atter in rease with the blade radial
lo ationbe auseofthe hangingairloads,thebladeelasti ityin ap,andmodelmotion.
TheanalysisofSPRresultsrequiressomepost-pro essingsin ethedata ontainonlypositionsofthe
markers alongleading and trailingedge in spa e, i.e. in thewind tunnel oordinate system.The goals
are ap,lead-lagandtorsiondispla ementsofthequarter hordlinein theshaft oordinatesystemwith
origin in the enter of the rotor hub. To obtain these results, the position of the hub enter must be
knownandseveralpro edureshavetobeapplied(Fig. 6).
Rawdatainthewindtunnelsystem
+
Averagingofrawdata
+
Motionanddrift omp ensation
+
Rotationbytheshaftangle
+
Rotationbytherollangle
+
Rotorhub enterx ; y p ositionidentifi ation
+
Rotorhub enterz p ositionidentifi ation
+
Shiftof o ordinatesintotherotorhub enter
+
Analysisofelasti blademotion
+
Fourieranalysisoftheblademotion
(a)SPRanalysis owdiagram
0
90
180
270
360
−2
−1.5
−1
−0.5
0
0.5
1
ψ
/ deg
100z
el
/ R
precone alone
BL
MN
MV
(b)Flapmotion0
90
180
270
360
1
1.5
2
2.5
ψ
/ deg
100y
el
/ R
BL
MN
MV
( )Lagmotion0
90
180
270
360
−4
−3
−2
−1
0
1
2
3
ψ
/ deg
θ
el
/ deg
BL
MN
MV
(d)TorsionmotionFigure6:Post-pro essingpro edure andresultsofSPR:Elasti tip motiondepending onrotorazimuth
forHHC-sweep,blade1(measurementa ura y:100z=R=0:02in apandlag;=0:5deg;
airfoilthi kness:100z=R=0:726)
During the measurements dynami markerposition displa ements due to low frequen y motionsof
thewind tunnel sting,vibrationsoftherotormodel andof thestingsupportwerepresent.Alsoadrift
in stingyawled to alateraldrift ofthe entire model. To orre t this drift(up to 10mm within the23
azimuth lo ations)thedriftofthebody markersisused. Thedieren e betweenthe meanvalueat one
azimuth tothemeanvaluefromallazimuths denethebody drift.Basedonthisdriftallblademarker
oordinatesweremodied.Theresultshaveshownthatthebodymarkerpositionsdierinmeasurements
ofthe advan ing and retreatingside.That is why thedrift ompensationwas madeseparatelyfor ea h
side.
After drift ompensationthedatain DNW oordinatesystemhavetobetransformedintotherotor
hub oordinatesystemtoobtaintheparametersofblademotion.Thisisdonebyrotatingthe oordinates
bytherotorshaftangleandtherotorrollangleinto a oordinatesystemparallel totherotorshaftand
transformation(shifting)ofallmarker oordinatesintotherotorhub enter oordinatesystem.Thenthe
elasti blademotion an beextra tedandasmotheddistributionofthethree omponents ap, leadlag
shaftanglethatwerenon-zero.Therollanglewasmeasuredto0.145deginsteadofzero.Duetosupport
elasti ity therotor shaftangle hashad an osetof about0.15deg that was alsotaken into a ountin
thefurtheranalysis.
Unfortunatelytherotorhub enter ouldnotdire tlybemeasured,sin etheSPR ameraswerelo ated
belowthemodel.Asuitablemethod forhub enter identi ationwas foundin the ir ularregressionof
singleblademarkers( omputebestt ir lesofthepositionsfromonerevolutiontogetthex y enter
point). There is one ir le enter point obtainedfor ea h blade marker whi h resulted in a s attering
of about 1:0mmin x dire tion and 1:1mmin y dire tion in all ongurations.When thehub enter
positions al ulatedbyusingdierentbladesare ompared,the maximumdeviationisabout0:8mmin
x-dire tionandonly0:2mmin y dire tion whi hisin theorderoftheSPRmeasurementa ura y.
Afterthehub enter oordinatesinx andy dire tionarefound,thepositionin z dire tionofthe
rotorhub enter isindemand.Atrstthe oordinatesystemmustbeshiftedintothehub enterfound
so far. This means atransformation of all blade and body markersinto a oordinate systemwith the
originin therotor hubx , y enter point, while theoriginin z is yet anywhere ontheshaft axis.To
identifytherotorhubz oordinate,apolynomialoffourthorderwithanadditionally onstraintisused.
It is assumed that the gradientdz=dr at theposition where theblade is xed is equalto the pre one
angleof2:5deg asaboundary ondition.Finallythe oordinats anbeshiftedintotherotorhub enter
oordinatesystemandtheblademotionparameters anbe omputed.
The elasti blade ap de e tion z
el
(positiveup) is omputedby the distan e between the quarter
hordlineandastraightlinedened bythepre- oneangle.Thereforeapproximatelythedistan eofthe
quarter hordline z position to thepre- one line atdened radialpositions is used. Theelasti blade
lead-lagde e tiony
el
isgivenbythedistan ebetweentheradialpositionofthequarter hordlineand
astraightlinedenedbythe urrentazimuthpositionoftheblade(lagpositive).Thepureelasti pit h
deformation
el
(positivenoseup) anbe al ulatedbythedistan eofthez oordinateofthefrontand
rearblademarkeraftersubtra tingtheasso iatedpit h ontrolangle, thepre-twistangle andthepit h
osetinz dire tionduetothedierentdistan eofthefrontandrearblademarkerstothequarter hord
line.
Fig.6(b),Fig.6( )andFig.6(d)showthe omparisonoftheelasti ap,lead-lagandtorsionmotion
of the rotorblade tip for the referen e bladedepending on azimuth forthe ongurations with higher
harmoni ontrol andthebase line aseat thebladeposition r=R=99%.When3=rev HHC isapplied
( asesMNandMV),a3=rev appingdominatesthegure(Fig.6(b))asexpe ted.Wendlo al
ampli-tudesofupto0:6%RotheBLposition atthebladetipintheminimumnoise ase.Theresultsofthe
bladelagmotionshownearlyidenti alvalueswith1=revamplitudesofabout0:5%Rindependentofthe
higherharmoni ontrol (Fig.6( )).With 3=rev HHC astrong3=rev torsionis theresponsewhi h was
expe tedduetothenaturalfrequen yintorsionat3:6=revofthisrotor.Thelo alamplitudesintorsion
areupto1:5degotheBLvaluesintheminimumnoise aseandupto2:5degintheminimumvibration
ase.
4 Tip vortex position and ow eld analysis
ThePIVmeasurementsweretriggeredtotherotorreferen ebladeazimuthsu htoobtainimagesofthe
tipvortexfromthesamereferen ebladeandthesameobservationareainspa eandintimeforaveraging
purposesandstatisti alanalysis.
As an example,the tip vortex of theMV ase reatedat y = 0:7Ronthe advan ing side at about
=135degis reatedbyalo al downloadwithaprettysharpgradientoftheblade ir ulationtowards
thetip. This vortex isni ely shaped rightfrom the beginning and basedon theanalysis of 100repeat
measurements these have a verti al s atter of 0:068%R and a horizontal s atter of 0:058%R in the
observation areawhen thevortexisjust reated,Fig.7(a).Theverti als atterrepresentsthevariation
in thebladetipposition asgivenby meansofSPRresultsshownin Fig.5,while thehorizontals atter
representstheradialposition onthebladewhere thevortexis generated, whi h isbiased alsobybody
motioninthisdire tion.Morethanonerotorrevolutionlaterthes atterhassigni antlygrowninboth
dire tionsduetotheso alledvortexwanderee t[3,7℄anddueto loseen ounterswithpassingblades
(BVI),Fig.7(b).It analsobeseenthatthedistributionofvortexpositionsisratherhomogenouswithin
therangeof insteadofbeing lustered withhigherdensityaroundthe meanvalueandthus doesnot
represent aGaussian normaldistribution. The growthtrend is progressiveas an be seen in both the
V V
a,b:s atterof vortex entrepositionatdierentvortexage,MV,y=R=0:7
0,0
0,5
1,0
1,5
2,0
0
90
180
270
360
450
540
Ψ
V
/deg
1
0
0
σ
x
/R
+0.7, BL
-0.7, BL
+0.7, MN
-0.7, MN
+0.7, MV
-0.7, MV
y/R =
BVI at
( )Standarddeviationofvortexhorizontalposition
0,0
0,5
1,0
1,5
2,0
0
90
180
270
360
450
540
Ψ
V
/deg
100
σ
z
/R
+0.7, BL
-0.7, BL
+0.7, MN
-0.7, MN
+0.7, MV
-0.7, MV
y/R =
BVI at
(d)Standarddeviationofvortexverti alposition
Figure7:Tipvortexpositions atter(observationarea oordinates).
wellasthegrowthofs atterisfoundtobesimilarforallthree asesandindependentofapplyinga tive
ontrolor not.
Twomethods of averaginghavenow been applied: the simple averaging (SA), and the onditional
averaging(CA)method.TheSAissuspe tedtoleadtoin reasinglyerroneousresultsintermsofvortex
properties like ore radius, swirl velo ityand vorti itywhen the vortex position s atter is growing.To
ir umventthese problems, a onditional averagingmust be applied as des ribedin [7℄. TheSA needs
nofurtherexplanationssin eitappliesthearithmeti averagingtothe100datasetsoneveryindividual
ve torof the eld. TheCA method rst identies all individual vortex entre oordinates, then shifts
allindividual vorti eswiththeir entreto theaverage entre position(= alignmentofthe entres)and
nallyaveragesallindividual measurementsforea hve tor.Duetotheshifts,theve torsat theborder
of the observation area are not overed by ea h data set su h that the resulting eld is based on a
non-homogenous number of samples. However, this does not ae t the vortex analysis as long as the
vortexis lose to the image entre, whi h is the ase in virtuallyall measurements.The ow hartof
post-pro essingtheindividualve tormapsisillustratedinFig.8.Mostoptions anbeswit hedonoro
using ontrolparameters,andanautomati pro essingis possibleinmostofthedataavailable.Manual
ontrol however, is ne essaryin tri ky ases, for example, when a blade has passed very lose to the
vortexandade isionmustbemadewhi hofthevorti estosele tforaveraging.
Theee tofaveragingmethodologyontheresultingvelo ityeldisshownnextattheexamplefora
vortexof theMN aseontheadvan ing side.Dueto a tive ontrol thisvortexis generatedbyahigher
blade loadingat thepoint of vortex reation, ompared to theBL ase.In Fig.9(a)and Fig.9(b) the
rstindividual measurementsat avortex ageof
V
=5:5deg and335:5deg,respe tively,are shownfor
omparision with the averagingmethods. In-plane velo ity omponents are shown as ve tors and the
ross- owvelo ityis olor- oded. Theyare noisydue to owturbulen e andmeasurementun ertainty.
ω
y
λ
2
ω
y
λ
2
ω
y
λ
2
ω
y
λ
2
Read control parameter
Select Dpt, image, averaging method
Average (simple, conditional)
Compute Gauss’ bell function
Identification of vortex center
Remove mean flow components
Compute rotation angles
Rotate into vortex axis
Iteration
Compute Gauss’ bell function in sheared grid
Dialog / automatic
Statistical analysis, plot of scalars and profiles
Stop
Plot of scalars f(x), f(z), f(x,z), f(r)
Store scalars and profiles
Subtract identified vortex from field
New image: yes / no
Data smoothing (Fourier synthesis)
Spurious vector elimination
Image clipping
Identification of sense of rotation
Read Image
Compute flow derivatives
Compute flow derivatives in sheared grid
Convolution with u, v, w, , Q,
Convolution with u, v, w, , Q,
Compute operators , Q,
Compute operators , Q,
Non−dimensionalization of
Ω
coordinates (R) and velocities ( R)
Figure8:Post-pro essing owdiagramfor3C-PIVdata.
in-plane velo ity omponents and the ross- ow velo ity, see Fig. 9( ) and Fig. 9(e). This means the
positions atterissmallenough omparedtothedimensionsofthetipvortexandtheshearlayerbehind
therotorblade.The oreradius,indi atedbythe ir le,isessentiallythesameinbothaveragingmethods,
andthemeasurementnoiseisalsosu essfullysuppressedbybothmethods.Fortheoldvortex,however,
the dieren es be ome visible in all parameters.The axial velo ity in the vortex enter is mu h more
pronoun edandthe oreradiusissigni antlysmallerintheresultoftheCAmethod,Fig.9(f), ompared
totheSAmethod inFig.9(d). Inthis asethepositions atter issigni antlyex eedingthedimensions
ofthe owstru turetobeobservedandthusCAmustbeapplied.
Thisresultisfurtherdemonstratedatthevelo ityprolesinahorizontal utthroughthevortex entre
for the samedata sets. For theyoungvortex in Fig. 10(a) and the old vortex in Fig. 10(b) the result
ofthe SA andtheCA method are ompared withea hother andwith therst 5ofthe100individual
measurements.Duetothesmallspatials attertheyoungvortexallowstheappli ationoftheSAmethod,
but thisis obsoletefortheold vortexwithlarges atter wherethe SAmethod leadstoarti ially large
ore radiiand smaller swirl velo ities than any of the individuals. In ontrast, the CA method retains
boththe oreradiiandtheswirlvelo ity.Thesespatialdistributionsoftheswirlvelo ityareananalogon
to the prin iple shown in Fig.1, and sin e the vortex entre is s attered in bothdire tions this ee t
isvisible in bothvelo ityprolesu(z)andw(x). Thus, atwo-dimensional orre tionhasto beapplied,
whi hisrepresentedbytheCAmethod.
Itmustbere ognizedthatthisspatials atterintwodimensionswillbere e tedtwo-foldintheblade
aerodynami loadingand onsequentlyin thea ousti time histories. First,theverti al vortexposition
s atter will be visible in as atter of BVI-related pressurepeaks (and theasso iated C
n M
2
) and thus
as atter ofBVI intensity, whi h an behandled using standardaveragingpro edures.In ontrast, the
horizontalvortexposition s atterwilltranslateinto azimuthalvariationsofwherethisBVI happens. In
this ase,astandardaveragingpro edurearti iallysmoothes theBVI signatureandagainaderivative
oftheCAmethodmustbeapplied.
5 Blade pressure and se tion loading time histories
DuringtheHARTIItest,bladepressuredatawerere ordedbymeansofabsolutepressuresensorswith
adatarateof2048=rev,triggeredto therotorazimuth.80 ontinuous revolutionsweretakensu hthat
low frequen yos illationsof themodel aredire tly visible in these data. Thevortexwanderaddressed
in the last se tion leadsto atwofold ee t:First,the verti alposition s atter translatesinto as atter
( ) simpleaverage(SA) (d)simpleaverage(SA)
(e) onditionalaverage(CA) (f) onditionalaverage(CA)
a, ,e:Pos.17a,
V
=5:5deg b,d, f:Pos.22,
V
=425:5deg
Ve tors:in-planevelo itiesuandw; olor- oded: ross- owvelo ityv.Meanvaluesaresubtra ted,
observationarea oordinates.The oreradiusisindi atedbythe ir le, every5thve torshown.
Figure 9:Ee tofaveragingmethodonthevelo ityeld, MN,y=R=0:7.
revolutionbyrevolution.Se ond,thehorizontalpositions atterofthevortextranslatesintoanazimuth
(or time)jitterfor bladepressuretimehistories, withoutmagnitudeee t. Allthese ee ts ombine to
produ e both a magnitude s atter and a time jitter of these BVI-events in ea h revolution. A simple
averagingwould leadto arti ialsmoothingof su h BVI-events(as isthe aseforPIV analysis),while
the onditional averagingmethod applied to thetimehistories isableto ompletely eliminatethetime
jitterbefore averagingonlythemagnitudes atter.
The omputationsweremade usingthe for e oeÆ ientnormal tothe hord enterline, C
n M
2
Bla ksolid:CA,dashed:SA;red:5individual;meanvaluesaresubtra ted,observationarea
oordinates.
Figure10:Ee tofaveragingmethodontheswirlvelo ity,MN,y=R=0:7.
thethree operational onditionsBL, MNand MV. Toget aphysi ally orre t averagedblade pressure
time historyofall 80 onse utivelymeasuredrotorrevolutionswith respe t to BVIphenomena (whi h
arein thefrequen yrangeofabout20 200=rev),itis notenoughto simpleaveragetheC
n M
2
values
at ea h sample(forrotorloadingpurposes,whi h are overingthefrequen yrangeof about0 6=rev,
simpleaveragingisbyfarsuÆ ient).Further,amoredetailed viewontherawdataisneededto getthe
orre ttime historyby onditional averaging.For al ulationof thenormalfor e oeÆ ientC
n M
2
the
pressuretime histories at aradius of r=R = 0:87of 11 Kulite sensors on theupperside and 6 on the
lowersideofthereferen ebladewereavailable.The hord-wisedistributionisshowninFig.11.
Corre tionofdefe tsensorsignals
+ C n M 2 omputation +
Bandpassfilterfrom20to250/rev
+
Simpleaverage(SA)of80revolutionsasreferen e
+
DefinitionofBVIeventsasreferen ep ositions
+
ConvolutionofindividualwithSAtimehistory)QC
+
ShiftvaluedefinedatmaximumQC
+
Corre tionofindividualtimehistories
+
Conditionalaverage(CA)of80 orre tedtimehistories
(a)Pressureanalysis ow hart
.32
.86
.66
.44
.24
.18
.06
.12
.03 .09 .15
.55
.80
(b)Airfoilse tionFigure11:Chordwisebladepressuresensordistribution asusedin HARTII,r=R=0:87
In[3℄aliasingproblemsaredes ribedfordistin tfrequen iesofspe i multiplesofthesamplingrate.
Isolatedspikeswerefoundat128=rev,256=rev,512=rev,and768=revwithabout20dBmagnitudemore
than thesignalshould exhibit. An explanation of this problem ould beveryhighfrequen y signalsin
somedataa quisition ablesthatleadtodisturbedpressuresignals.Toremovethedisturbedfrequen ies
aharmoni analysisofthea ordingpressuresignalismadeforea hfrequen ywhi hhastobe orre ted
to get the sine and osine oeÆ ients. Thereafter, by using a harmoni synthesis, all parts of these
frequen iesare addedtogetherandnally subtra tedfromtheoriginaltimehistory.
Totallindividualtimehistories togetherfor orre tedamplitudeandphasewidthoftheaveraged
timehistorya oupleofpro eduresareneeded(Fig.11(a)).Themainproblemsarethedierentlo ations
ofthe entreofaBVI-event(=itstimes atter)andthusthes atterinlo ationoftheasso iatedminimum
andmaximumpeaks(beginningandendingofaBVI-event)whi his ausedby u tuationsinrotational
speed and varying vortex lo ations due to vortex wander when passing the blade. The typi al BVI
entretheswirlvelo ityisdire tedupwardswhi hprodu esanin reasingC
n M
2
.Attheretreatingside,
theBVIsignatureis theotherwayround sin ethebladeapproa hesthevortexfrom theoppositeside.
A steep de reasing ank is presentwhile passing the vortex entre. The azimuth lo ations where BVI
takespla e are strongly dependent on the operating onditionand an best be visualizedby the high
frequen y ontentofthebladeleadingedgepressurealongradiusandazimuth.Forthe asesinvestigated
here,theyaremainlybetween =10degand =90degattheadvan ingsideandbetween =270deg
and =350degattheretreatingside.
TondtheBVI-eventswhere a orre tionof thetimephaseisneededaband passlteringbetween
20=rev and 250=rev is donefor allindividual C
n M
2
timehistories in order to eliminate thelarge
low-frequen y ontent and thus to leave overonly the interesting frequen y range of BVI-events.Therein,
the signature of one BVI-event time historyis hara terizedby a de reasing ank followed by asteep
in reasing ank and again a de reasing ank at the advan ing side and the other way round at the
retreating side due to the physi sof vortexintera tions as des ribed above.The simpleaveragedtime
historyalready provides the orre t lo ation of BVI-events,but neither the orre tmagnitude nor the
orre tazimuthalextensiontotherightandleftoftheeventitself.Itisusedasreferen efora onvolution
withtheindividualtimehistories.Thevortex entreofaBVIlo ationisdenedusingthesimpleaveraged
data whereC
n M
2
=0. For ea h BVI-eventthere isone onvolutionfun tion (=CF),whi his dened
bytheband-passlteredvaluesof thesimpleaveragedtime historybetweenthea ordingstartingand
ending point ofa BVI-event.For the onvolutionitself, theregion ofinterest of ea h individual C
n M
2
timehistoryismultipliedwiththeCF togetaquality riteria(=QC).Thestartingpointisshiftedfrom
15samplesto+15samples(2:6degofazimuth),whi h oversthemaximumshiftofeventsobserved.
TheresultingQC (Eq.4) is ameasurefor the oin iden e betweenboth theindividual andthe simple
averagedtimehistories.
QC(j)= +15 X j= 15 BVIend X i=BVIstart CF(i)C n M 2 (i+j) ! (4)
This pro edure,applied to all BVI-events,leads to individual valuesof shift for theadvan ing and
retreating side a ording to the number of BVI-events. In general, the shift values obtained dier by
about 5.5samples, whi h orrespondsto about =1degof rotorazimuth. It was found, that there is
nodependen ybetween thevaluesof shiftofthe advan ing andtheretreating side. For one individual
rotor revolutionthe shiftsof advan ing and retreating side are ompletely dierent and no systemati
behaviouris visiblein all three ight ases.On theother hand theshiftis nearly onstant within ea h
side, thus, the phaseshift orre tion hasto be applied independently on theadvan ing and retreating
side. When all shifts of ea h BVI-event are known, the phase orre tion an be done. The orre tion
is made by the assumption that the referen e lo ation is at an integer sample numberat theposition
where thesteepin reasing ank(advan ing side)or steep de reasing ank (retreatingside)hasavalue
nearC n M 2 =0.Todispla etheC n M 2
valuesin timealinearstret hingor ompression(depending on
positiveornegativeshift) betweentworeferen epointsisused. Sin etherequiredshiftsfoundhavereal
inde esan interpolationis ne essarytoget nally thenew C
n M
2
valuesdependingon anintegertime
index(whi hisneededforaveraging).
All individualtimehistoriesare orre ted(shiftedandinterpolated)that wayandanewmeanvalue
anbe omputed.InFig.12all80C
n M
2
timehistoriesareplottedfortheretreatingsideoftheBL ase
before(Fig.12(a))andafter(Fig.12(b))thetimephase orre tion.After orre tion,thelo ationofthe
steepde reasing ankisnearlyidenti alforalltimehistoriesandthemaximumandminimumpeaksare
atthesamelo ationsaswell.Finally,theaveragetimehistory anbe omputed,whi hisnow alledthe
onditionallyaveragedtimehistory.
For omparisonbetweenthesimpleaveragedand onditionalaveragedtimehistories,thepeak-to-peak
amplitudes and peak-to-peak time dieren es for the sele ted BVI-events were investigated. Changes
ould be found with respe t to the peak-to-peak amplitudes between the onditional average and the
simpleaverage,whi h ledalwaysto in reasesofupto C
n M
2
=+3:610 3
or +8:4%aroundthe
BVI-events.However,thesevaluesarenotsomu hexpressive,sin ethemagnitudeoftheamplitudesarevery
dierent. The peak-to-peak azimuthal distan es between the C
n M
2
extreme valuesof the onditional
averagedC
n M
2
values omparedtothepeak-to-peakazimuthwidthofthesimpleaveragedvaluesshow
de reasesofupto 1:8samples( 0:32deg).Theper entagedieren esarebetween+0:3%upto 7:8%
(seealsoFig.13(a)).
Inany ight ase,the onditional averagehaslargerC
n M
2
amplitudesat theBVI-events,while the
290
295
300
305
0.07
0.08
0.09
0.1
0.11
ψ
/deg
C
n
M
2
(a)before orre tion
290
295
300
305
0.07
0.08
0.09
0.1
0.11
ψ
/deg
C
n
M
2
(b)after orre tion Figure12:80 individual C n M 2time histories before (a) and after (b) time jitter orre tion (BL, data
band passlteredfrom20-250/rev)
tothesimpleaveragedtimehistories.Althoughthedieren esbetweensimpleand onditionallyaveraged
datamayappearasnotimportant,thephysi sofrotorBVInoisearebasedontimederivativeofC
n M
2
,
whi hsigni antlyexaggeratesthedieren es(Fig.13(b)).
290
295
300
305
0.075
0.08
0.085
0.09
0.095
0.1
0.105
0.11
Ψ
/deg
C
n
M
2
C
n
M
2
(SA)
C
n
M
2
(CA)
(a)Normalfor e oeÆ ientCnM 2
290
295
300
305
−1.2
−1
−0.8
−0.6
−0.4
−0.2
0
0.2
0.4
0.6
Ψ
/deg
dC
n
M
2
/d
Ψ
/ rad
dC
n
M
2
/d
Ψ
(SA)
dC
n
M
2
/d
Ψ
(CA)
(b)TimederivativeofCnM 2Figure13:Comparisonofnormalfor e oeÆ ientC
n M
2
anditstimederivativeforsimpleand onditional
average(BL,databandpasslteredfrom 20-250/rev),r=R=0:87
6 Mi rophone time histories
Sin e blade pressuretime histories (Se t. 5) are the sour e ofnoise generation allee ts of timejitter
andmagnitudes atterarealsofoundin mi rophonepressuretimehistorydatainthesamemanner.For
investigationsoftheee t of onditionalaveragingonmi rophonesignalstheunltered rawdata ofthe
baseline ase(BL)of mi rophone11attheadvan ing sideandmi rophone4at theretreatingsideare
used. For ea h mi rophonelo ationdata were storedfor100 onse utiverevolutionswith anin rement
11.Theselo ationswere hosenbe ausetheyare very losetothemaximumpeaksofthenoise ontour
(Fig.14),one attheadvan ing andoneattheretreatingside.
(a)Simpleaverage (b)Spe traaverage
Figure14:Noisedire tivity ontourbaseline ase(BVI-SPL)
Thespe trumofthesimpleaveragedmi rophonetimehistoriesthenleadstosigni antlylowernoise
levelsthantheindividualspe tra,sin ethepeaksofpressurearesmoothedunrealisti allylarge.To over
theproblems asso iatedwith thistime jitter,in thepasttheaverageof allindividual spe trawasused
asbeingrepresentativefortheaveragednoisespe trum.Ea hofthemi rophonepressuretimehistories
show fourtypi aleventswithinone rotorrevolution.These eventsare ausedby theBVI-eventsof the
rotorbladesandsin ethemodelrotorisfour-bladedtherearefourmain eventsatea hmi rophoneper
revolution.AsmentionedinSe t.5thesimpleaveragedtimehistoryalreadyprovidesthe orre tlo ation
ofBVI-events,butnotthe orre tmagnitudenorthe orre tazimuthalwidthoftheeventitself.Again,a
onvolutionismadeandthebest onvolutionfun tiontobe omparedtoasingletimehistoryisassumed
tobethesimpleaveragedtimehistoryofallthe100revolutions.Itisalsousedtosetthereferen epoints
respe tivelythereferen eazimuthlo ations.Forthemi rophonedatanolteringisne essary,sin ethere
areonlypressureos illationsaroundthestati pressure.Herethereferen epointswere hosentobeon
thesteepin reasing ankatthezero rossingbetweentheminimumandthefollowingpositivemaximum
valueofaBVI-event.
Shiftto overmainBVIevents
+
Simpleaverage(SA)of100revolutionsasreferen e
+
DefinitionofBVIeventsasreferen eazimuthlo ations
+
ConvolutionofindividualwithSAtimehistories)quality riterionQC
+
ShiftvaluedefinedatminimumQC
+
Corre tionofindividualtimehistories
+
Conditionalaverage(CA)of100 orre tedtimehistories
+
Cal ulationofp owersp e tra
+
Cal ulationofBVISPLandBWISPL
history at the main BVI-events again a onvolution is made in a range of 32 samples around the
a ording referen e point (RP) to havebest oin iden e. By means of the least error squares method
aquality riteriaQC(j)is omputed where CF is the onvolutionfun tion extra ted from the simple
averagedtime history. The onvolutionismade in arangeof16samples, whi h oversthe maximum
shiftof eventsobserved(Eq. 5). Atthe shiftvaluewhere the individual timehistorybest ts with the
CFthe QC at this sample is minimal. To ndthe minimumQC, abest t polynomialof 2 nd
order is
omputedbymeansofregressionanalysisusingvevaluesaroundtheminimum.
QC(j)= +16 X j= 16 RP+32 X i=RP 32 (CF(i) p(i+j)) 2 ! (5)
Sin e four main eventsare presentwithin one rotor revolution,this pro edure has to be applied to
allofthem.Finallywegetfour individualshiftvaluesforea hmi rophonetimehistory.All BVI-events
havenearlythesameshiftvalueswithinone revolution.As inbladepressuredata theshiftvaluesdier
byabout5:5samplesinmaximum,whi h orrespondstoabout =1degofrotorazimuth.Comparing
this magnitudeto theshift resultsfoundin the C
n M
2
analysis in Se t. 5asimilar behaviour isfound.
The u tuationsinrotationalspeedandvaryingvortexlo ations,whi hleadtothese shifts,arepresent
at thebladesas lo ationofnoisesour eas wellas atthemi rophonepositions. Finallytheadjustment
of the individual time histories is done a ording to the pro edure (by linearstret hing/ ompression)
mentionedinSe t.5andanewaverage anbe omputed-now alledthe onditionalaverage.InFig.15
thepost-pro essing ow hartformi rophonedataisshownandFig.16showsthe omparisonoftheraw
dataandthe orre tedmi rophonepressuretimehistories.
294
296
298
300
302
304
−40
−30
−20
−10
0
10
20
30
40
50
ψ
/ deg
p / Pa
(a)before orre tion
294
296
298
300
302
304
−40
−30
−20
−10
0
10
20
30
40
50
ψ
/ deg
p / Pa
(b)after orre tionFigure16:100 individualpressure time histories before (a) and after (b)time jitter orre tion
(Mi ro-phone4)
The omparison of simple and onditional averages leads to in reases in peak-to-peak amplitudes
between +4:4% and +6:5% for both mi rophones while the peak-to-peak azimuth dieren es always
de reasebyabout 1%to 10%.Thisleadstoremarkable hanges ofthegradientsdp=d atthemain
BVI-events.Thedieren eofthegradientsofmi rophone4(+15:4%to+34:2%)isabouttwi easlarge
asthedieren esfoundinthemi rophone11results(+8%to +13:4%).
In heli optera ousti s thesoundpressurelevelor powerspe trumof mi rophonedatais important
for noise estimations. There are twomain frequen y bands of interest. With respe t to the BVI-noise
therelevantfrequen yrangeisbetweenthe6 th
and40 th
bladepassagefrequen y(bpf)whi h is24=rev
to 160=rev forafour-bladedrotor,fortheBWI-noise(Blade WakeIntera tion)therangebetween40 th
and100 th
bpf(160=revto400=rev forafour-bladedrotor).Inthesetwofrequen yrangesthea ording
sound pressurelevels(SPL in de ibel) an be omputed as logarithm of the square root of the sumof
pressureamplitudes,dividedbyareferen epressurep
ref
BVISPL=20log 0 u u t 160 X i=24 p(i) 1 p ref 1 A (p ref =210 5 Pa) (6)
Atrstthepowerspe tra(SPL)are al ulatedbymeansofaFFTforboththesimpleand
ondition-allyaveragedtimehistories(Fig.17(a)).Additionallythespe traaverageisplotted,whi histheaverage
oftheindividualspe trafromea hofthe100timehistories.The onditionalaveragespe trum, ompared
tothesimpleaveragespe trum,showshigheramplitudes inthelowerfrequen yrangesasexpe tedand
thusis losertotheaveragespe trum.Tohaveabetterrelationbetweenthethreespe traonlythepeaks
atmultiplesofthebladepassagefrequen iesaresele tedandplotted intheBVI-SPLfrequen yrangeas
theupperenvelopeinFig.17(a). It learly an beseenthat the onditionalaverageisvery loseto the
spe traaverageandthus more apabletogeta urateBVI-SPL al ulations.
Finally theBVI-SPL results an be ompared(Fig. 17(b)). Thevaluess atter byabout2to 2:5dB
forbothmi rophonedata. Whilefor mi rophone4theBVI-SPLof thesimpleaverageddatais 0:38dB
lowerthan thespe traaverageBVI-SPL (111:66dB),the BVI-SPLof the newly omputed onditional
average is only 0:1dB lower.The same tenden y ould be found in the mi rophone 11 results , where
thedieren eto thespe traaverageBVI-SPL (113:56dB)nowisredu edfrom 0:64dB (SA)to 0:18dB
(CA). Even the blade wake intera tion (BWI) noise spe trum is mu h better omputed basedon the
onditionalaveragedpressuretimehistories, omparedtosimpleaveraging.
40
60
80
100
120
140
160
65
70
75
80
85
90
95
100
105
n/rev
SPL [dB]
Spectra average
Spectrum SA t.h.
Spectrum CA t.h.
(a)Upperenvelopeofthepowerspe traofmi rophone4,
onlybladeharmoni sinBVI-SPLrange
0
20
40
60
80
100
110.5
111
111.5
112
112.5
Revolution
BVISPL [dB]
Individual t.h.
Simple average t.h.
Cond. average t.h.
Spectra average
(b)BVI-SPLofmi rophone4Figure17:Resultsof onditionalaveragingwrtsoundpressurelevel(Mi rophone4)
7 Con lusions
The physi s of time jitter and its ee ts on simple averaged time history data are demonstrated and
lariedin thispaper.Themethod of onditionalaveragingprovidesagoodmeansto eliminatespatial
s atter ee ts requiredfor the analysis of ow eld ve tor maps, and also to eliminate thetime jitter
of BVI events in the individual time histories. This is espe ially valid for highly sensitive data like
mi rophonepressuretime histories. Consequently, onditionally averagingis intended tobemandatory
forthegenerationofreliableaveragedtimehistoriesofblade(andmi rophone)pressureor owelds(in
ordertoretainthehighfrequen yee tslikeBVIor smallstru turesliketipvorti esin theirindividual
[1℄ B.G.vanderWall,C.L.Burley,Y.H.Yu,K.Pengel,P.Beaumier,TheHARTIITest-Measurement
ofHeli opter Rotor Wakes,Aerospa e S ien eandTe hnology,Vol.8,No.4,pp.273-284,2004
[2℄ B.G.van derWall, 2nd HHC Aeroa ousti Rotor Test (HARTII) Part I: Test Do umentation-,
DLR-IB111-2003/31,2003
[3℄ B.G.vanderWall,C.L.Burley,2ndHHCAeroa ousti RotorTest(HARTII) PartII:Representative
Results-,DLR-IB111-2005/03,2005
[4℄ HARTIIinternationalworkshop website,ftp://HART-IIftp.dlr.de
[5℄ T.F.Brooks,Ee tofSignalJitterontheSpe trumofRotorImpulsiveNoise,Verti a,Vol.12,No.3,
pp.257-265,1988
[6℄ O. S hneider, Analysis of SPR measurementsfrom HART II, Aerospa e S ien e and Te hnology,
Vol.9,No.5,pp.409-420,2005
[7℄ B.G. van der Wall, H. Ri hard, Analysis methodology for 3C-PIV data of rotary wing vorti es,
ExperimentsinFluids,Vol.40,No.5,pp.789-812,2006
[8℄ W.R.Splettstoer,R.Kube,U.Seelhorst,W.Wagner,A.Boutier,F.Mi heli,E.Mer ker,K.Pengel,
KeyResultsfromaHigherHarmoni ControlAeroa ousti RotorTest(HART)intheGerman-Dut h
Wind Tunnel,21 st
EuropeanRotor raftForum, Saint-Petersburg,Russia,1995
[9℄ Pengel, K., Muller, R., van der Wall, B.G., Stereo Pattern Re ognition - the Te hnique for
Reli-able Rotor Blade Deformation and Twist Measurement, AHS InternationalMeeting on Advan ed
Rotor raftTe hnologyandLifeSavingA tivities, Utsunomiya,To higi,Japan,2002
[10℄ Rael,M.,Ri hard,H., S hneider, G.,Klinge,F.,Ehrenfried,F., Pengel, K.,Feenstra, G.,
Re ord-ing andEvaluation Methods of PIV Investigation on a Heli opter Rotor Model, 11 th
International