4.6 MIMO case position control
4.6.2 Unbalance force rejection control
Inthemagneticallybornerotor,whichspinsaboutaxedaxis,therearealways
somesinusoidaldisturbanceforcesactingontherotor. These forcesarecaused
bytheunbalance,whichcanbedescribedasadiscrepancybetweenthexedaxis
ofrotation(usually,theassumedaxisofgeometryoftherotor)andtheprincipal
inertiaaxis. Apartfromanunbalance,allotherdiscrepanciesandimperfections,
which make the system non-axisymmetrical, may cause additional harmonic
disturbances. The resultingvibrational bearing forces are proportional to the
squareoftherotationalspeed.
The unbalance compensation in the AMB applications is referredto asan
andthe possibilityof changingthestiness anddampingprovide better
capa-bilities to deal with the unbalance than in the traditional bearings. Control
theory, rotordynamics, and the literature on AMBs provide dierent UFRC
methods. Basically, acompensationmechanismissynchronizedwith the
rota-tional speed. It injects compensating harmonicsignals, of proper amplitudes
andphases,tothecontrolsystem. Ingeneral,itispossibleeither tocancelthe
position vibrations(i.e. cancel the eect of the unbalance forces onthe rotor
position)ortocancelthemagneticforcevibrations(i.e. canceltheeectofthe
unbalanceforcesonthecontrolcurrents).
Asanexample,anadaptivefeedforwardcompensationanddiscreteFourier
transform (DFT) are applied by Bleuler et al. (1994). Grochmal and Lynch
(2006)andtheysugestthereduced-orderdisturbanceobserverwiththeobserver
gainscomputedanalytically(based onthedesiredlocationofeigenvalues)and
scheduled accordingto the rotationalspeed. The interestingapproach is
pre-sentedbyLum etal.(1996),wheretheobserver-basedimbalance compensator
performsanon-lineidenticationofthephysicalcharacteristicsoftheimbalance
andusestheresultto tunethecompensator. This lastcompensatortechnique
workesundervaryingrotorspeed. Manyother compensationmethodsarealso
available. Ashort reviewofdierentcompensationmethods andreferenceson
themaregivenbyBleuleretal.(1994).
Inthiswork,theobserver-basedUFRCisconsideredintwovariants. Inthe
rstone,thepositionvibrationsarecanceledand thesystemcantoleratehigh
sinusoidalunbalanceforceswhileitmaintainsaccuraterotorposition.Itresults
in low measured vibration amplitudes. In the second variant, the magnetic
force vibrations are canceled and the rotor (considered as rigid) spins about
its principal inertia axis withouttouching thesafety bearings (assuming that
theclearances are largeenoughto accommodatethe unbalance). This results
inminimal controlcurrents. Both casesaredesignedasaperiodicdisturbance
estimatorthat iscombinedwiththeconstantdisturbanceestimator.
Inthecaseof positionvibrationcancellation,twodierentdesignmethods
areemployedandcomparedwithrespecttotheirrobuststability,forthe
equiva-lentperformance. Forthebettercomparison,boththedesignmethodsareused
in such a waythat theyresult in the sameclosed-loopeigenvalues. The rst
designgeneralizes the LQ designof the full stateestimator with theconstant
disturbanceestimate,asin (4.11), totheonethatincludes asinusoidal
distur-bance. Thecomputationoftheproportionalgainmatrix
L
andthedisturbance estimatorgainmatrixL dist
isbasedontheaugmentedplantmodelx ˙
Similarlytothedesignoftheconstantdisturbancemodel,alsoherethe
distur-Thenegativedamping
ζ = − 0.05
ofasinusoidaldisturbancemodelisassumed.This perturbation is introduced in order to move the roots of the augmented
plantmodelfrom theimaginaryaxis. Inaddition,
ζ
aectstheresultingspeedoftheintegration,forthesinusoidaldisturbanceestimate. Now,theaugmented
modelusedin theimplementationappliesthedisturbancestatematrixsuchas
A 0 dist =
The second design is based on the pole placement. It applies the closed-loop
roots, which result from theKalmanlter designcomputed forthe state
esti-mateswithoutthedisturbances,plusthearbitrarilyaddedintegralpoles.
Now, the state and disturbanceestimate feedback matrix is
[K, K dist ] = [K, C dist ]
.Theunbalanceresponsesofthesimulatedcontrol systemwith andwithout
the UFRC are presented in Fig.4.24 and 4.25. The applied step disturbance
forceis
f = 0.2 · i c,max k i ≈ 402 N
andtheapplied coupleunbalance1isequalto500 g · mm
thatcorrespondstothedisturbanceforceamplitudeequalto137 N
,ateachradialbearing,for
Ω = 5000 rpm
.Theresultingclosed-looppolesarethesameforbothaforementioneddesign
methods, when theplant model without the residualmodes is applied.
How-ever, the closed-loop pole patterns for the designed controllers and the plant
that included theresidual modes(Fig. 4.25)aredierent for both controllers;
namelyforthepoleplacementdesignofintegrators,theresidualmodesareless
damped(aboutsixtimes)thanfortheoptimaldesign. Thisdierencecanalso
be noticed when comparing the singularvalue plots of the output sensitivity
transfer functions for both controllers (4.26). The singular values plot of the
outputsensitivitytransferfunction correspondstothedisturbanceattenuation
performanceinthemeasuredsignal,anditisequivalenttothesensitivityofthe
SISOcase,whichindicates therobuststability.
In thecaseof magneticforce cancellation,thecomputation of
L
andL dist
and the feedback are the same as for the disturbance force cancellation, but
the augmented plantmodels used in the design and implementation have the
output equation such as
y = [C, 0, I, 0] x
andy ¯ = [C, 0, I, 0] ¯ x
, respectively.Thiseliminatestheoscillatingcomponentfrom theerrorsignal.
The experimental evaluations ofUFRCfor thepositionvibration
cancella-tion and forthe magnetic force cancellation are presented in Fig. 4.27and in
Fig.4.28, respectively. Infact,the rotorwasnotrotating; however,theeect
ofcoupleunbalancewasgeneratedthroughthecurrents.
4.6.3 Comparison of LQ and PID controllers
Aslongasarigidrotor(forslowrotation)isconsidered,itispossibleto apply
the decentralized control (four SISOcontrollers) to support the rotorradially
in x and y directions. The low-order PID controllersmay be easily designed
manually in asimilar fashion asthe one for the axial suspension (see section
1
coupleunbalanceassumingnostaticunbalance,theremainingforcesineachplane(in
oppositesidesofthecenterofrotormass)actinoppositedirectionsandareequalinmagnitude
0 0.1 0.2 0.3 0.4 0.5 0.6
Figure 4.24: System withoutUFRCand thesystem with UFRC that cancels
theunbalanceforcesarepresentedunderthecoupleunbalance. The(A)stands
forthemeasuredposition
x A
andcontrolcurrenti c,x,A
;the(B)isthereferenceposition andthescaleddisturbanceforce
f dist /k i
.0 0.1 0.2 0.3 0.4 0.5 0.6
controlled flexible mode roots residual mode roots
integrator roots
Figure4.25: Intheleft-handillustrations,thesystemresponseswithUFRCfor
cancelingthemagneticforcevibrations(thereforeminimizingcontrolcurrents),
arepresented. Themeasuredposition
x A
andcontrolcurrenti c,x,A
areindicatedby(A). Thereference position
x ref,A
andthe scaleddisturbance forcef dist /k i
are indicated by (B). In the right-hand illustration, the closed-loop roots of
thesystemusedin thecontroldesign(B)andthecontrolsystemthatincludes
non-controlledexiblemodes(A)in theplant,areshown.
10 200 1000 3000 10000 10 −2
10 −1 10 0 10 1
Sensitivity
frequency [rad/s]
gain [dB]
Figure4.26: Singularvaluesplotoftheoutputsensitivityfunction forthe
sys-temswithUFRC,arepresented. Theblueplotcorrespondstothesystemdesign
basedontheoptimalgainselectionandtheredplotcorrespondstothesystem
designbasedonthepole placement.
0 0.02 0.04 0.06 0.08 0.1
−2 0 2
control current [A]
without UFRC
0 0.02 0.04 0.06 0.08 0.1
−50 0 50
time [s]
position [ µ m]
i cx,B i cy,B
x B y B
0 0.02 0.04 0.06 0.08 0.1
−2 0 2
control current [A]
with UFRC
0 0.02 0.04 0.06 0.08 0.1
−50 0 50
time [s]
position [ µ m]
i cx,B i cy,B
x B y B
Figure4.27: MeasuredresponsesofthecontrolsystemwithoutandwithUFRC
for theunbalance force cancellation,when the coupleunbalance is
500 g · mm
for
Ω = 5000 rpm
.0 0.02 0.04 0.06 0.08 0.1
Figure4.28: MeasuredresponsesofthecontrolsystemwithoutandwithUFRC
forthemagneticforcecancellation,whenthecoupleunbalanceis
250 g · mm
forΩ = 5000 rpm
.4.4). However,fortherotormodelthatincludestheexiblemodes,theobtained
PIDcontrollersareunstable,and theyhavetobeimproved.
Let us consider a leadcompensatorin Eq. (4.5). We include the bending
modes in the design. Now, we include the rst bending mode into the SISO
plantmodel. Weassumethat it doesnotaecttheend-B oftherotor(as the
nodeof thebending modeis veryclose to theactuatornode). Next, webuild
theapproximatedplantmodelfortheend-Aas
G(s) = k i
modalmassoftherstbendingmode,modaldamping,modalstiness,andthe
rst modal inuence factor at the end-A (Lantto, 1999), respectively. Inthis
case,theinuencefactormayberoughlyapproximatedbythedistancebetween
theactuator node and thenode of therst bending mode. Other parameters
areobtainedfrom theFEM model. Now,wecanuse theapproximatedmodel
inthecontrollerdesign.
Wemodifytheoldcontrolbyadjustingthefrequencyofthemaximumphase
inordertoobtainthesucientstabilitymargins,foreachloopatatime. After
someiterationwiththecontrollerparameters(decreasingtheproportionalgains,
slightlyincreasing thecurrentcontrol bandwidth,andselecting the
ω ld = 0.7 · 2π · 260 rad/s
), whentaking thevalues accordingtoTable4.3, asthe startingpoint, weobtain thestable decentralizedcontroller. In further iterations, the
resultingmultiple SISO,overallcontrolleris testedfor stabilitywith theplant
modelthat includestherstthreebending modes.
Thestepresponsesoftheclosed-loopsystem(withaleadcompensator),and
thecomparisonoftheunbalanceresponsesofthedecentralizedPIDcontrolwith
the centralized LQ position control,for dierent rotational speeds, are shown
inFig.4.29.
Forthe PIDcontrol oftheradialsuspension, thestepresponses arenotas
goodasfortheLQcontrol,eventhoughthePID-basedcontrolassumesa40%
0 0.1 0.2 0.3 0.4 0.5 0.6
1000 2000 3000 4000 5000 6000
0.1
Figure 4.29: Onthe left, the stepposition and disturbanceresponses for PID
radialsuspension. Ontheright,thecomparisonof thecentralizedLQposition
control with high stiness (A), low stiness (B), and the decentralized PID
control(C), fordierentrotationalspeeds,insimulations,ispresented.
responses,thebehaviorforlowerfrequenciesisalsonottoogoodbecauseofthe
lowstiness(requiredforstability).
FortheLQMIMO control,which isableto providehigher gains(and
bet-ter damping) without loosing the stability, the gyroscopic eect is taken into
account. Inconsequence,theoptimalcontrolsolutionvarywith therotational
speed. TheLQcontrolcanbetuned,andremainstable,tohavegoodunbalance
rejectionbothatloworhighfrequencies. Thelowstiness controldesignmay
be obtained by, for example, increasing the available maximum displacement
in the weightingmatrix
Q ¯
. It occurs that the eect of gyroscopic coupling isrelativelyweak,whencomparingthecentralizedLQcontrollerwithUFRCand
thedecentralizedonewithUFRC,theunbalanceresponsesarethesameforthe
under-criticaloperation.
Moreover,forbothcontrolsystems(PID-and LQ-based),thevariationsof
rotationalspeedresultinrelativelyinsignicantchangesoftheclosed-looppoles
andzeros(Fig4.30). InFig4.30,onlythepolesandzeros,whichhavepositive
orzeroimaginaryparts,andwhichareclosetotheimaginaryaxis,areshown.
Figure 4.31presentsthe start-up of the rotorwhen a centralized LQ
con-trollerwith theactivelycontrolled rstexible mode andthe constant
distur-banceestimatorwasemployed. Sincetheestimatorhasnoknowledgeaboutthe
initial states, it isstarted together with themeasurementsbefore theintegral
action isenabled to work asalter withzero control currents. It is sucient
toenableonlytheestimationoftherigidbodymodesduringthisinitialphase.
InFig.4.32,thestart-upoftherotorwhenPIDcontrollerisapplied,isshown.
The measured responses of the LQ controller and PID controller to the
stepdisturbanceforce introducedthrough theAMB at theend-Bof therotor
(
f dist = k i i c,y,B
,wherei c,y,B = 2A
)arepresentedinFig.4.33and 4.34,respec-tively. Figure 4.35 shows the control currentsof the LQ controller when the
stepdisturbanceisapplied.
Tosumup,in thestudied rotortherelativepolarinertia isverysmalland
thereforethegyroscopiceecthassmallinuence ontheresponsesofthe
mul-tiple SISOandMIMO control. The dierencesin theunbalance responses for
−250 −200 −150 −100 −50 0
Figure4.30: Intheleft-handillustration,thepole-zeromap oftheclosed-loop
systemwiththePIDcontroller,andintheright-handillustration,thepole-zero
mapoftheclosed-loopsystemwiththeLQcontrollerarepresented. Thepoles
areplottedas x'sandthezerosareplottedaso's.
stiness solution). It isnot possibleto obtain optimalsolutionfor allthe
fre-quencyranges,andhencethecontroldesignis acompromisebetweenlowand
highfrequencies.