3.3 Modeling of actuators
3.3.3 Actuator dynamics
The bearing dynamics are examined using the voltage equation of the coilof
a single electromagnet. The voltage of the coil
u
equals to the voltage dropaccordingtoOhm's Law
Ri
plusthechangeoftheux linkageintime.u = dψ
−5
Figure 3.10: Forcecharacteristicsof twopairedhorseshoeelectromagnetsasa
function ofthecontrol currentand displacementinthex andy planes.
0
Figure3.11: Calculateddynamicinductanceofoneelectromagnet,asafunction
of coil current anddisplacementof therotorfrom the central position toward
theelectromagnet.
−0.35 −0.2 0 0.2 0.4
Figure 3.12: Top two graphsdepict the control currentsas afunction of
dis-placements,measured(dots)andcomputedwith theRNM(solidlines),in the
radialbearings,aredepicted. Inthebottomgure,themeasuredmagneticforce
asafunctionofcontrolcurrentisshown.
−6 −4 −2 0 2 4 6
Figure3.13: Forcecharacteristicsofanaxialhorseshoeelectromagnet,forthree
biascurrents: (A)
i bias = 0.5 [A]
,(B)i bias = 2.5 [A]
,(C)i bias = 5 [A]
.Consideringthedisplacementoftherotorin
x
direction(towardthe electromag-netwiththeinductanceL
),andusingthedynamicinductance,dened earlier, thevoltageequationcan bewrittenasu = L dyn di
Here, we introduce the ux linkage
ψ = Li
and the velocity-induced voltagek u x ˙
(or motion-induced backelectromotiveforce). The inductanceof asingle horseshoe electromagnet (in the operating point, i.e.,x = 0
, and when themagnetizationoftheironisneglected) canbecomputedusing theux density
in theair-gap(2.5)orreluctanceoftheair-gap(3.36)as
L = N S air dB di = N 2
2 < air = µ 0 N 2 S air
2(l 0 − x) .
(3.43)Hence, using this result, comparing with Eq. (2.8), and (2.15), the
velocity-inducedvoltagecoecientmaybeapproximatedas
k u = i dL
dx = µ 0 N 2 iS air
2(l 0 − x) 2 ≈ 1
2 k i .
(3.44)Thesameresultmaybeobtainedfrom(2.5)anddirectlyfromthepartial
deriva-tiveoftheuxlinkage
ψ
withrespecttotheair-gap. Intheoperatingpoint,thek u
equalsthecurrentstiness ofthesingleelectromagnetthat ishalf ofthek i
(the currentstiness oftwopaired, opposite electromagnetsin the dierential
drivingmode). Ifamoreaccuratemodelfor
k u
isrequired,itcouldbeobtainedbybuildingthelook-uptablefromthedynamicinductance.
Werecallthatweassumedacurrent-controlledAMBswiththebiasedcontrol
currents
i ref = max(i bias ± i c , 0)
. In the system, the input to the actuatorsconsistsoffourradialreferencecurrentsandoneaxialone. Mostoften,ineach
actuator,anindependentlycontrolledcurrentloopisbuiltusingaproportional
controlleras
u ref = G P (i ref − i m ) ,
(3.45)where
u ref , G P
,andi m
arethereferencevoltage,theproportionalgain,andthe measuredcoilcurrenti
,respectively. Whenanalyzingthedynamicsoftheinner currentclosed-loop,itistypical(e.g.Schweitzeretal.,2003,Lantto,1999,andLarsonneur, 1990) to simplify the problem by neglecting the velocity-induced
voltage. Thisispossiblebecauseofitsrelativelyinsignicantmagnitude,
com-paredwiththevoltageofthecoil. Additionally,the
k u
introducesthederivativethathasastabilizingeect;itcounteractstheeectofthenegativeposition
sti-ness. Furthermore,itistypicalthattheresistanceofthecoilissmallandcould
beneglected. With allthese assumptions and utilizing (3.42), and (3.45), the
simplestapproximationoftheclosed-loopdynamicsbecomes
G cl (s) = i m
i ref ≈ G P
sL + G P
= 1
sτ cl + 1 ,
(3.46)where
τ cl
istheclosed-looptimeconstant.ThemoredetailedmodelincludesthePWMdelay. ThePWMdelaycanbe
approximatedusing a shiftin time in theLaplace domain and dierentseries
G f f
Figure3.14: Actuatorblockdiagram modelcomprisestheforceeld (withthe
current stiness
k i
and position stinessk x
), LR circuit, switched ampliermodeledasthedelay,andinternalcurrentcontroller,where
G P
andG F
indicatetheproportionalcontrollergainandfeedforwardcontrollergain,respectively.
e sξ ≈ 1 + sξ + (sξ) 2
2! + ... + (sξ) n
n! + ...,
(3.47)where
ξ
is the time shift in seconds. We can use (3.47) for the linear timeinvariantmodels(likestate-variableformandtransferfunctions). Forexample,
a second-order approximation of a delay is
e sξ ≈ 1/[s 2 T md 2 /4 + sT md /2 + 1]
.In Matlab Simulink we can utilize a Transport Delay block, for PWM delay
toprovidemuch shortersimulationtimes comparedwiththeswitchingmodels
(Wurmsdobler,1997). Therst-orderMaclaurinapproximationcorrespondsto
therst-order(0/1)Padéapproximation(Franklinet al.,1998). Forthedelay
ξ = − T md /2
this rst-orderapproximation,asthetransferfunction,yieldsG del (s) ≈ e sξ ≈ exp 0/1 (sξ) = 1
1 − sξ = 1
s T md 2 + 1 ,
(3.48)where
T md
istheaveragemodulationdelaythat equalshalf ofthemodulationperiodowingtotheasymmetricregularsampling.
Finally, the slightly more detailed approximation of the current control
closed-loopdynamicsis
G cl (s) ≈ G P G del (s)
sL + R + G P G del (s) .
(3.49)Thevoltagerelationtogetherwiththeforcerelation(2.14),internalcurrent
controller and PWM delay form the complete actuator model, as presented
in Fig. 3.14, where the feedforward controller gain compensates the resistive
voltagedrop.
TheschematicofloadcapacitylimitationsoftheAMBactuatorispresented
in Fig. 3.15. Themaximum force of thebearing systemis determined by the
maximumcoil current
f max = F (i max )
. Themaximumconstantcoilcurrentisdeterminedbythevaluesoftheresistance
R
andavailableDClinkvoltageu DC
(
i DC,max = u DC /R
). However, usually this limit is high and thus theoretical.Inpractice,the maximumforce forthe continuousoperation islimitedby the
coiltemperaturelimit. Typically,thecoiltemperaturelimitisdescribedasthe
coilcurrent,at which amaximumcurrentdensityin acoilreachesthecertain
limit
J max = 4 − 6 A/mm 2
. Onebut adicultopportunityfor moreaccuratethermal analysis (Pöllänen et al., 2006). Another option could be to use the
fact that the static load capacity of the amplier is limited by the size and
geometryofthebearing(seeEq.2.17). Hence,tolimitthesaturationinuence
on system dynamics and linearity, we may select the limiting current to be
equaltothesaturationcurrent(seeEq.2.22). Apeaktransientcurrent,atlow
frequencies, is determined by the articially selected maximum peak current,
atwhichthecontroller,andcontrolelectronics(limitedmeasuringrangeofthe
currentsensorsandlimitednumberformat)canprovideastablesuspension. For
highfrequencies,thedynamicloadcapacityisdeterminedbyeitherthecurrent
control bandwidth (using Eq. 3.46,
ω cl ≈ G P /L
) orby the power bandwidthω BW
,whicheverismorerestrictive.A powerbandwidth resultsfrom thelimitedDC link voltage, thelow-pass
characteristicsofthecoiltransferfunctionappliedtosinusoidalsignals,andthe
maximumcurrent. It canbeestimatedbyusing theopen-loopplantscaled by
the
u DC
,whichis equaltothevoltagelimitedcurrentamplitudeasi u,max = u DC G pl = u DC
G del
(sL + R) ,
(3.50)wherethepeak-to-peakcurrentof
2 · i u,max
equalstothemaximumcurrenti max
.Thefrequencyatwhichthevoltageentersthesaturation
ω sat
maybedeterminedbythe crossingofthe magnitudeBode plotsof thetransferfunction basedon
thecontrollerdynamics (3.49) andthevoltage-limitedcurrent-amplitude ratio
(3.50),specically
2u DC /i max · 1/(sL +R)
. Thepowerbandwidthapproximationω BW
canberead,directlyfromtheBodemagnitudeplot,approximatelyatthe -3dBcrossingofthevoltage-limitedcurrent-amplituderatio(usedasatransferfunction). Another approximation of the dynamic force limitation that takes
theinductancevariationsinto account,asgivenby Lantto(1999),is basedon
therisetime
t rise
ofthecurrentinthecoiloftheelectromagnetL dyn
fromzeroto
i max
. Utilizingaresponseoftherst-orderapproximationofthesystem,the powerbandwidthiswhere
u DC
is theDC link voltage applied to thecoil. Replacing thedynamicinductance by the nominal one, assuming the
f max
at thei max = i sat
, using(3.43),and(2.22),thepowerbandwidthcanbepresentedin termsofthe
max-imum force amplitude
f max
(2.17) and the maximum amplier performanceP max = i max u DC
,asIfthecurrentcontrolbandwidthisabovethepowerbandwidth(
ω cl > ω BW
),the loadcapacity is limitedby theamplier performance and bythe DC link
voltage. Inconsequence, at thefrequenciesgreater thanthe
ω sat
, thebearingcannotproduceforceswithamplitudes determinedbythecontrollerdynamics.
The
ω sat
isdeterminedat thepointwhere thevoltagelimitcurvecrosseswithlo g f ( i )
Figure3.15: LimitsonAMB:(A)thestaticloadcapacity,(B)thedynamicload
capacityundercontinuousoperation,(C)thepeaktransientloadcapacity
themagnitude plotof thetransferfunction of thecontroller. Therough
rst-orderapproximationoftheactuatordynamicsbecomes
G cl (s) ≈ ω BW
s + ω BW
.
(3.53)Alternatively, forhigher-orderapproximationsoftheclosed-loopdynamics,we
canmodifyEq.(3.49),byutilizingadescribingfunctionmethod(Franklinetal.,
1998) and (Slotine and Li, 1991). The derivation of the describing function
Γ(G P i ref )
fortheactuatorsaturationispresentedinAppendixA.4. Thequasi-linearizedactuator,forsinusoidalinputand output,yields
G cl (s) ≈ Γ(u ref )G P G del (s)
where
u ref
isassumed,here,to betheamplitudeofthesinusoidalinput(3.45).Thecontrolbandwidthcanbeestimatedasabandwidthof (3.54),where
u ref = 0.5 · G P i max
(forsingle electromagnet). Forhigh amplitudes of acoilcurrent, theeect of thedynamic inductancemaybeincludedinto (3.54) by using(inthecomputations),suchavaluefor
L
,thatgivesthesamerisetimeastheL dyn
in Eq. (3.51). However,for high frequencies,the voltage is saturatedand coil
currents are bound in the regionwhere the inductanceis closeto its nominal
value.
TheexemplaryBodeplotsfordierentapproximationsofactuatordynamics
arepresentedin Fig.3.16. Themain benetof using (3.54)isthe inclusionof
boththesaturationandthemodulationdelayin onemodel.
Astothedynamicsoftwooppositehorseshoeelectromagnets,themaximum
loadcapacity at dierent frequenciesdepends on the maximumpossibleforce
slew rate, for
x = 0
,df /dt = k i (i) · di c (i)/dt
. As explained earlier, in theoperatingpoint,thecurrentstinessofasingleelectromagnetequalshalfofthe
corresponding
k i
oftwooppositeelectromagnets,asdoestheforceslewrateand apeak-to-peakvalueoftheforce. Therateofchangeofthereferencecurrent,inthesingleelectromagnet,correspondstotherateofchangeofthebiasedcontrol
current,fortwooppositeelectromagnets,whereeachofthemgeneratestheforce
−60
−40
−20 0
Magnitude (dB)
10 0 10 2 10 4 10 6
−180
−135
−90
−45 0
Phase (deg)
(A) (B) (C) (D)
Bode Diagram
Frequency (rad/sec)
Figure3.16: Bodeplotsoftheapproximatedactuatordynamics,for
ω cl > ω sat
:(A)basedon(3.49),(B)basedon(3.54),(C)basedon(3.53),(D)theDClink
voltagelimitfortheopen-loopEq.(3.50).
1. Ifthecontrolcurrentamplitudeislowerthanorequalto
i bias
,therateofchangeofthecontrolcurrentisthesameastherateofchangeofthe
refer-encecurrents(bothcoilsareactive). Consequently,thevoltagesaturation
appearsapproximatelyat
ω sat
,and thefrequencyresponses ofthesingleelectromagnet are the same as the frequency responses of two opposite
electromagnets(Fig.3.17).
2. If thecontrol currentamplitudeisgreaterthan
i bias
, theratesofchangeof tworesultingreference currents(fromtheapplied
i c
)are greaterthantherateofchangeofthe
i c
. Consequently,thevoltagesaturationappears earlierthanatω sat
(determinedforasingleelectromagnetwithoutreduced premagnetization current), and the frequency responses dier from thepredictedones(thederivedlinearmodelsaremoreuncertain). Theactual
ω sat
changes and depends on the control signalamplitude and bias. Inaddition,theintrinsicsystemnonlinearitiesbecomelargerforhighervalues
ofpositionandcurrentsignals(Fig.3.18).
Toput the second point in another way, for lowand medium frequenciesand
for amplitudes of the control current that are greater than the value of the
biascurrent,therateofchangeofamagneticforceiscontributedmostlybyone
electromagnet(forveryhighfrequencies,thecoilcurrentamplitudesarereduced
below
2 · i bias
andbothelectromagnetsproduce force). Forthese reasons, the forceresponsemaydeviateconsiderablyfromthelinearizedone,attheoperatingpoint,modelprediction.
Figures 3.17and 3.18showcomparisonsof thenormalizedgains (resulting
from dierent force magnitudes) and phaseBodediagrams obtained from the
singleelectromagnet(A),twooppositeelectromagnets(B), anddierent
actu-atorapproximationsof thesystem with
ω cl > ω BW
. TheBodeplots (A)and0 1000 2000 3000 4000 5000 6000 0.4
0.6 0.8 1
magnitude [pu]
0 1000 2000 3000 4000 5000 6000
−80
−60
−40
−20 0
frequency [rad/s]
phase [degrees]
(A) (B) (C) (D) (E)
Figure 3.17: Frequency responses ofthe simulatedmodelsand dierent linear
approximations are presented. The
i max = 2 · i bias = 5 A
. The saturationoccurred at
ω sat = 2800
rad/s, accordingto prediction (3.50). Thesaturationoccurredat
ω sat = 2762
rad/s, accordingtosimulation.(B) were obtained using the Simulink models, which included the force
non-linearities,inductancenonlinearities,modulationdelay,andvoltagesaturation.
Theresponses of theapproximatedmodels comprise(C) approximationbased
on(3.49),(D)approximationbasedon(3.54),and(E)approximationbasedon
(3.53). Forfrequencieslowerthan
ω sat
,thefrequencyresponsesoftheSimulinkmodels aresimilarto theresponse ofEq. (3.49)(the agreementis notsoclose
forthesystemwiththehigh signalamplitudes), andthevoltagesdonot
satu-rate. Forthefrequencieshigherthan
ω sat
, theampliersentersaturation,and thefrequencyresponsesoftheSimulinkmodelsapproachtheresponseof(3.54).In thesimulations thesame current controllers (withthe samegains), with a
considerablyhighbandwidth
ω cl
(controldesignfort r = 500µs
),wereused. Theapplied reference for single electromagnet
i ref = 0.5 · i max sin(ωt) + 0.5 · i max
,andtheappliedcontrolcurrentforoppositeelectromagnets