AutomaticControl
SystemLinearization
11Oct16
Aimsforthischapter
Forclassiccontroltheorywearegoingtodiscussisbasedon
[Link],itisnotsuitablefornonlinear
systems.
Incontrolengineering,anormaloperationofthesystemmay
bearoundanequilibriumpoint,andthesignalsmaybe
consideredsmallsignalsaroundtheequilibrium.
Ifthesystemoperatesaroundanequilibriumpointandifthe
signalsinvolvedaresmallsignals,thenitispossibleto
approximatethenonlinearsystembyalinearsystem.
Suchalinearsystemcanbeapproximatedtothenonlinear
systemaroundtheoperationpoint.
SystemLinearization
Why linearization process is needed?
The advantages of using linear system model are
1). The principle of superposition can be applied.
2). Operational methods of analysis can be used.
3). The magnitude scale factor is preserved in a linear
system. (The property of homogeneity)
Linearization
1). Smallsignal linearization (in this lecture)
2). Linearization by feedback (in nonlinear control theory)
3). Transformation method (in nonlinear control theory)
Linearization can help designers to use linear system theory
to analyze the nonlinear system at certain interesting operation points.
FunctionLinearization
Toobtainalinearmathematicalmodelforanonlinearsystem,we
assumethatthevariablesdeviateonlyslightlyfromsomeoperating
condition.
Notethatthelinearized modelsonlydescribelocalbehavior.
Considerasingleinputsingleoutput(SISO) nonlinearfunction(ora
nonlinearsystem)whoseinputisx(t)andoutputisy(t).The
relationshipbetweeny(t)andx(t)isgivenby
Ifthenormaloperatingpointcorrespondsto
TheequationcanbeexpandedintoaTaylorseries aboutthispoint.
H.O.T
EquationLinearization
BasedontheTaylorseries(onthepoint)
H.O.T
variation
Ifthevariationissmall,thehigherordertermscanbeneglected.
whereand
Finallywegotthelinearized function
output
deviation
y Kx
input
deviation
FunctionLinearization
Geometricmeaningoflinearization
y f x
y y0 K ( x x0 ), K : slpoe
y0 f x0
x0
nonlinear
curve
FunctionLinearization
Nowconsideranonlinearfunctionwithtwoinputs
Alinearapproximationtothisnonlinearfunction(orsystem)canbe
obtainedbyTaylorseriesaboutthenormaloperatingpoint
variation
variation
H.O.T
Again,ifthevariationissmall,thehigherordertermscanbe
[Link]
Finally,onehas
y K1x1 K 2 x2
FunctionLinearization
Firstly,considerasimplenonlinearequation
Letusobtainalinearized equationforthenonlinearequationneara
[Link]
ApplyingTaylorseries gives
Thefinallinearized equationis
FunctionLinearization
Thefinallinearized equation
Recalltheexactnonlinearequation
Considerainterestingpair(x=5,y=10)aroundtheoperationpoint,
itgivesriseto
theapproximated solution:
theexact solution:
Itcanbeobservedthatthereexistsestimationerror(5049=1)due
tolinearization.
Consideranotherinterestingpair(x=5.9,y=10.9),whichisvery
closetotheoperatingpoint.
Theestimatevalueis64.3,whiletheexactvalueis64.31.
FunctionLinearization
Considerthesinglevariablenonlinearfunctiondescribedby
y x 0 .4 x 3
(a).Findthelinearizedmodel.
(b).Findthecorrespondingequationw.r.t x 1, x 2
y f x
df x
x x
dx x x
x 0.4 x 3 1 1.2 x 2 x x
f x
For x 1
For x 2
df x
dx x x
y x 0.4 x 3 1 1.2 x 2 x x 1.4 2.2 x 1 2.2 x 0.8
y x 0.4 x 3 1 1.2 x 2 x x 5.2 5.8 x 2 5.8 x 6.4
FunctionLinearization
Thegeometricillustrationisshownasfollows:
y x 0 .4 x 3
y 2.2 x 0.8
y 5.8 x 6.4
Differentoperationpointgivesrisetodifferentlinearized model!!
FunctionLinearization
Forthemultivariablenonlinearequation
,
ApplyTyalor seriesgives
f x, y
H.O.T
where
Thelinearizedmodeis
FunctionLinearization
Usually,theoperationpointsisselectedbythemiddlepointofthe
interestingregion.
Fortheexactmodel:
Whentheoperationpointischosenby. Itleadstothe
followinglinearizedmodel.
[Link](x=10,y=4),theestimated z=345
whiletheexact z=356 error=356345=11
[Link](x=9.2,y=3.1),theestimated z=256.2
whiletheexact z=256.38 error=256.38 256.2=0.18
Theexamplesclearshowthattheclosertheoperationpointwe
select,thesmallertheestimatederrorwewillhave.
SystemLinearization
Inordertolinearizegeneralnonlinearsystems,wewilluseagainthe
TaylorSeriesexpansionoffunctions.
Considerasinglestatevariabledynamicequation
x f x
Supposethat x isapointsuchthat f x 0 .
Inthiscase,thepoint x iscalledanequilibriumpointofthesystem
since
x f x 0
Inotherwords,apointx willbecalledanequilibriumpoint(Epoint
inthefollowingcontent)if
x 0
when x x
the system doesnt change
anymore at this point
SystemLinearization
RecallthattheTaylorSeriesexpansionoff(x)aroundtheEpointis
givenby
1 d 2 f x
1 d 3 f x
df x
2
3
x x
x
H .O.T
x
x
f x f x
2
3
2! dx x x
3! dx x x
dx x x
For xsufficientlyclosetoEpoint,thesehigherordertermswillbe
[Link]
f x f x
df x
x x
dx x x
Moreover,bytheequivalences x x x x x x x
(alsoknownasdeltastate)
f x 0
Thelinearizedmodelis: x
df x
df x
x x
x
dx x x
dx x x
SystemLinearization
NotethatthislinearmodelisvalidonlyneartheEpoint(hownear
dependsonhownonlinear.
Basedontheequivalentrepresentationoftheperturbed(delta)
state,thelinearizedsystemisalsocalledaperturbedsystem
dynamics.
df x
x
dx
x Ax
x x
Defineanewvariablez =x,thelinearizedsystemcanbefurther
representedby
z Az
whereAdenotesasalinearizedsystemcoefficientattheEpoint.
Thenonlinearsystemdynamicsnowturnsintoalinearone.
SystemLinearization
Considerasinglevariablenonlinearsystem
x x x 2
x0
TrytolinearizethenonlinearsystemaroundtheEpoint
andfindthelinearperturbedsystemdynamics.
x
df x, u
x
dx x x 0
1 2 x x
x
Itcanalsoberepresentedby
x x
xt x0 et
unstable
SystemLinearization
Actually,therearetwoEpoints
f x x x 2 0 x1 x 0
WehavethefollowingEpoints
x0
x 1
Whenx=1,thecorrespondinglinearizedmodelis
x
df x, u
x
dx x x 1
1 2 x x
x
xt x0 e t
stable
SystemLinearization
Nowweextendtheconcepttosinglestateandsingleinputdynamic
[Link]
f x, u f x , u
f x, u
f x, u
x x
u u
x
x
x
x
x u u
u u u
1 2 f x, u
2 f x, u
2 f x, u
2
2
x x 2
x x u u
u u
2
2
2! x
xu x x
u
x x
x x
H .O.T
x, u
AlsonotethattheEpointcanbedeterminedsuchthat
x f x , u 0
SystemLinearization
Ignoringthehigherorderterms,onehasthefollowinglinearized
model
f x, u
f x, u
x f x , u
x x
u u
x x
x x
x
u u
Usingthefactthat
Thefinalformthelinearizedsystemwillbe
x
u u
x x , f x , u 0
x x x , u u u
f x, u
f x, u
x
u A x B u
x ux ux
u ux ux
where&aresystemstate/input
coefficients,respectively
B
A
SystemLinearization
Considerasingleinputsingleoutputsystem
x x x 2 sin x u
x 0, u 0
TrytolinearizethenonlinearsystemaroundtheEpoint
andfindthelinearperturbedsystemdynamics.
x
f x, u
f x, u
x
u
x x x
u x x
u u
u u
1 2 x cos x x u
2x u
Itcanalsoberepresentedby
x 2 x u
TF :
1
X s
U s s 2
SystemLinearization
Theextensiontofunctionsofmultiplestates(ndimension)and
inputs(mdimension) isverysimilartotheaboveprocedure.
Considerahighordernonlineardynamicswithnstate&minput
x1 f1 x1 , x2 ,, xn , u1 , u2 ,, um
x2 f 2 x1 , x2 ,, xn , u1 , u2 ,, um
xn f n x1 , x2 ,, xn , u1 , u2 ,, um
Supposethattheequilibriumpointsaregivenby
fi x1 , x2 ,, xn , u1 , u2 ,, um 0, i 1,2,, n
NotethattheEpointshouldmakeallofthefunctionsfi equalto
zero.
SystemLinearization
Inherittheprecioussystemlinearizationconcept,thelinearization
ofeachsubsystem fiabouttheequilibriumpointcanbedirectly
derived
n f
m f
j
x j x j j
u j u j
x j
Jacobian Matrix
j 1 x j
where
x1
x
x 2 n1
xn
f j
j 1 x j
f1
x
1
f 2
A x1
f n
x1
f1
x2
f 2
x2
f n
x2
j 1 u j
x j x j
x j
x j x j
f j
j 1 u j
u j u j
u j
u j u j
f1
f1
u
xn
1
f 2
f 2
nn
B u1
xn
f n
f n
xn x x
u1
x Ax Bu
f1
u2
f 2
u2
f n
u2
f1
un
f 2
nm
un
f n
um x x
SystemLinearization
Considerasingleinputsingleoutputsystem
x1 f1 x1 , x2 , x3 , u x1 x22 sin x1 u
x2 f 2 x1 , x2 , x3 , u 2 x1 2 x2 x3
x3 f3 x1 , x2 , x3 , u 3 x3 x1 x22 sin x2 u
FindthelinearperturbedsystemdynamicsaroundtheEpoint
x1 x2 u 0
x1 2x1 u
x2 2x1
x3 3x3 x2 u
Itcanalsoberepresentedby
x1 2 x1 u
x2 2 x1
x3 3 x3 x2 u
x1 2 0 0 x1 1
x 2 0 0 x 0u
2
2
x3 0 1 3 x3 1
SystemLinearization
ConsidertheLorenzsystem(3rd ordernonlineardynamics)withtwo
controlinputs
x 10 x 10 y
y 28 x y xz u1
8
z z xy u2
3
x 0, y 0, z 0
Byletting,theEpointsare
0,0,0, 6 2 ,6 2 ,27 , 6 2 ,6
Basedonthegivensystem,onehas
f1 x, y, z 10 x 10 y
f 2 x, y, z 28 x y xz u1
2 ,27
8
f3 x, y, z z xy u2
3
Trytofindthelinearized modelsw.r.t thoseEpoints
SystemLinearization
Applyingthelinearizationprocedureyields
f1 x, y, z 10 x 10 y
f 2 x, y, z 28 x y xz u1
8
f3 x, y, z z xy u2
3
TheJacobian matricesare
f1
x
f
A 2
x
f3
f1
y
f 2
y
f 3
y
f1
f
f
10, 1 10, 1 0
z
x
y
f 2
f
f
28 z , 2 1, 2 x
x
y
z
f3
f
f
8
y , 3 x, 3
x
y
z
3
f1
f1
u
z
0
10 10
1
f
f 2
B 2
28 z 1
x
u1
z
y
x 8 / 3 x x
f
f3
3
z x x
u1
f1
f
0, 1 0
u1
u2
f
f 2
1, 2 0
u2
u1
f3
f
0, 3 1
u1
u2
f1
u1
0 0
f 2
1 0
u2
0 1 x x
f3
u2 x x
SystemLinearization
BasedonthecalculatedEpoints
Epoint:
0,0,0
differentEpointsleadto
differentlinearized systems
0
0
10 10
10 10
0
A 28 z 1
x
28 1
y
x 8 / 3 x 0,0,0 0 0 8 / 3
Epoint: 6 2 ,6 2 ,27
0
10 10
A 28 z 1
x
y
x 8 / 3
x 6 2 , 6
Epoint: 6 2 ,6 2 ,27
0
10 10
A 28 z 1
x
y
x 8 / 3 x 6
10
0
10
1
1 6 2
6 2 0 6 2 8 / 3
2 , 27
2 , 6
10
0
10
1 6 2
6 2 6 2 8 / 3
2 , 27
0 0
B 1 0
0 1 x x
SystemLinearization
Innonlinearcontrolsystemlinearization,systemstabilityanalysis
w.r.t zerostate(i.e.,theorigin)isconsideredveryoften.
Herearesomeusefultechniquesforlinearizing nonlinearsystemsat
theorigin.
Forthesenonlinearity,itcanbedirectlyreplacedbythe
correspondingequivalence.
sin x x
cos x 1
x sin x 0
x cos x x
x2 0
x1x2 0
x1x22 0
TheequivalencescanbeeasilyprovenbytakingtheTaylorseries
expansion.
SystemLinearization
Letsgobacktotherealdynamicsystem
Rod
Dynamics
(xG,yG)
Cart
Dynamics
H&Vremoval
Therearethreenonlinearfunctionsinvolvedinthedynamics
Frequently seen
in nonlinear systems
f1 sin
f 2 cos
f3 Vl sin Hl cos
SystemLinearization
Forthependulumsystem
WhereMg istheweight,J isthemomentofinertiaofthependulum
aroundthepointofrotation,T isanappliedtorqueandListhe
lengthofthependulum
Thesystemnonlineardynamicsis
Basedonthepreviousresult,onecandirectlyobtainthelinearized
modelw.r.t zeroangle.
d 2 MgL
2 2
d t
sin , cos 1
SystemLinearization
Forthe2nd ordernonlineardynamicswithoutinput
dx
d 2x
2
cos x 0
dt
d 2t 2
Itcanberepresentedby
x1 f1 x2
x2 f 2 2 x2 cos x1
Thelinearized modelw.r.t Epointis
f
f
x1 f1 x1 , x2 1 x1 x1 1 x2 x2 1 x2 x2 x2
x2
x1
f
x2 f 2 x1 , x2 2 x1 x1 2 x2 x2
x2
x1
2 x2 cos x1 sin x1 x1 x1 2 x2 x2 2 x2 cos x1 sin x1 x1 2x2
x1
x2
SystemLinearization
TofindtheEpoint,let
x2 0
x1 f1 0
x1
x2 f 2 0
n
, n 1,3,5,
2
SubstitutingtheEpointtothelinearized system
x1 x2
x2 2 x2 cos x1 sin x1 x1 2x2
gives
x1 x2
x2 x1 2x2
homogeneous
x1, x2 / 4,0
Ifonehasinterestatanotheroperationpoint,ex..
SinceitisnotanEpoint,itleadsto
x1 x2
x2
2
2
x1 2x2
2
2
Nonhomogeneous
SystemLinearization
Consideranonlinearelectricalnetwork.
Findthetransferfunction(TF)
Sincetheelectricalnetworkisnonlinear,wearenotabletogetthe
[Link],systemlinearizationmustbeapplied.
0.1v
Sincethecurrentpassingthroughthenonlinearresistoris ir 2e r
1
Theresultingdropvoltageis vr 10 ln i
2
BasedontheKirchhoffsvoltagelaw,wehave
L
di
1
10 ln i 20 vt
dt
2
di
10 1 20 1
f i, v ln i vt
dt
L 2
L L
Considerthesteadystate,i.e.,v=0,theEpointis
f i, v
10 1 20 1
ln i vt 0 i 2e 2
L 2
L L
SystemLinearization
Forthenonlinearsystem
di
10 1 20 1
f i, v ln i vt
L 2
L L
dt
Thelinearized modelis
f i, v
di
f i, v
v v
i i
f i , v
dt
i iviv
v iviv
20 1 10
1
10 1
v v
ln i v i i
2
L
L
L
i
L
L
f i ,v 0
f i ,v
i
f i ,v
v
10
1
10
1
i v 2 i v
iL
L
L
2e L
TheTFfromV toVL canbeobtainedfrom
I s
1
V s Ls 10 / 2e 2
Ls
LsI s VL s
V s
V s Ls 10 / 2e 2
Summary
Linearized systemisalwaysalocalapproximationaboutthe
operatingpoint.
Forthesamenonlinearsystem,astheoperatingpointchanges,the
linearized modelchangesaswell.
Theusualobjectiveofcontroldesignusingthelinearized dynamics
isdeviationminimization([Link])
Controldesignbasedonlinearizeddynamicsalwaysreliesonthe
strategyofgainscheduling([Link])
Gain changes with different operation conditions~
Homework
Forthenonlinear2nd ordersystem
x1 x22 x1 cos x2
x2 x2 x1 1x1 x1 sin x2
Pleasefindthelinearizedmodelattheorigin
(Notethatdetailderivationmustbeprovided.)
Homework
Findthelinearmodelofthefollowingdoubletanksystematthe
operationpoint:h1=h1s,h2=h2s,F=Fs
F,F1,F2:flowrates
R1,R2:valveresistances
A1,A2:tankbottomareas
h1,h2:waterlevels
h1
h2
R1
F1
F1 R1 h1 h2
R2
F2
F2 R2 h2
F F1 F R1
h
f
h
,
h
,
F
h1 h2
1
1
1
2
A1 A1 A1 A1
state
equations
h2 f 2 h1 , h2 , F F1 F2 R1 h1 h2 R2 h2
A2 A2 A2
A2
output equation
y h2
