Documente Academic
Documente Profesional
Documente Cultură
yesterday is
absolutely general .
= t " ST .
I
To design E 're the separation principle that the
,
we
bring assuming
the controller from the observer
design of is
uncoupled .
⇐ ?
① POLE PLACEMENT
Mpp
Tm
Re
By placing the poles properly I could get
⇒ ,
wk I
-
×
NOT COMPLETELY TRUE .
There also
zeros
are
.
I want not d
Maybe
.
In general we want to
position
our poles in the available
region .
closed loop
?
positions
-
The higher the distance of poles from the open -loop position .
the
control !
HIGHER the
effort
↳ I want the poles .o
not Oo far
open
-
E-EE
'
5-
"
a Of pi
s 't des pi pi
t t
OL tan →
-
,
.
. .
.
. . .
, ,
" "
Sn
tanto
't ai s t
pi pi -
t
a s a , .
. .
, ,
. . .
pin
i
DESIRED POLES
pi
=
tan
-
-
o
- -
-
. . .
. .
.
.
. . .
Fg
):
ft
"
Having G as
E-
→
I compute the char .
pdignomial
with as parameters and then I
g .
.
.
.gr impose
aifg . .
Ri ⇒ System guys
of" A equations in n' unknowns
gi
PROBLEMS :
it
21 Usually is a
NON-LINEARSYSTt.rs
EXAMPLE i S ISO
n -
-
thou
III
-
-
age
-
Here the well
!
u
problem is posed
- - -
INPUT SYSTEM .
Let's imagine
to here
! Asf
ku .
48
!
Let's multiply bug :
" is
!
. ⇒ as
I
:÷÷÷÷÷÷÷÷÷
ai
Ic has the same structure of ¥ but with aitgi instead of ai
So . . .
f Bass
.
GUNA FORMULA )
fait gil ALL
di
gi
t
⇒ di T DETERMINED
gi
= =
ai
-
This formula is valid just for I having this particular structure but ,
we can introduce a
transformation b
proper
.
COMANION or cont I
I =
A e t
Hu
fmsiongdar
STATE TRANSFORM .
TON →
I =
IX ,
on¥
N .
B .
I CAN MOVE MY POLES Wherever I WANT MY SISTER IS
CONTROLLABLE
.
E- IEIi I t
Iko u
E -
Has the same poles but it's in the th! form ⇒
⇒
I can now use One BASS -
Curt FORMULA
We design Io
matrix ⇒ e
IE IIe
-
a
- -
= -
E-II
"
E- Ibn -
Abn . . .
A' bn I
EE
'
f see Claudio ⇒
I -
"
E. 15 is .
.
.
A' hit .
-
IE
An - r An -
2 .
.
-
As I
I =
ane any . . .
I 0
0
: : : :
a I 0
,
. . .
0 0
I
. . .
In general ,
Ai
i
at fg , ,
g ,
. . .
gm )
→ NO UNIQUE solution I More gains
them poles )
NOTE :
place can 't be used with coincident poles .
Observer
the
The dynamic of observer can be reconstructed by
the
dynamics
of the error .
ee ± I E ft E Is
-
- -
.
work for
even I .
get a
proper
behaviour of the error .
problem
-
row .
,
.
At
pots
Es
CEE ⇒ PERFECT DUALITY I
place f AI EJ )
-
,
.
We'd like to
put observer poles .
.
.
where
?
desired poles
Of course
,
in the left plane .
To have a
proper
behaviour .
One observer poles should be
faster them the CL of the
poles system .
B x X -
a POLES
D x A -
OBS POLES
s x
D X
Therefore ,
it has to be fast to
properly
reconstruct the state .
,
good sensors )
,
the noise
is multiplied by L !
the observer
I
the MEASURED
We could have
systems with uncontrollable part → the state is not
obtainable leioimatabk from the
output y .
ie .
":!! :
how
observable
can
?
I cheek if a system is
OBSERVABILITY MATRIX
⑤ = similar to
E but involves
Egad # =
fnjnlxn
µ!q¥¥
OBSERVABLE ⇐ rank (E) = N
.
'
Ey AI
Why ? →
Became
of 3 limitation of the POLE PLACEMENT METHOD .
:
We look at the not to what reality f control
3) just poles .
happens in
Control
good perform
"
J =
I VI
d
⇒
IFE
⇐
t
IT #ve
b
) dt I -
-
performance variable
control
performance Kernel matrix
W=w=
It is a
quadratic form .
The
integral there became I want that Oo be optimal
is
quantity
could to from
to
from now on .
In theory ,
I want minimize J
give
He
.
:L
:O
fi www.adf :'t
I -
-
.
. -
wat .
stymmetric
the
:
they weigh same
quantity .
The matrices are :
D
VI. iz
.
-
VIE ,
Huo -
-
VII I sxnrttrkl
2) V=Vzze0
Eh ⑦0
Since we want Oo
find mint ,
it will not depend from
the factor Therefore what to
multiplying a .
,
is
important is
relatively select Wi ,
to Wu to Wvu .
,
to Wwi .
3 des .
performance :
① Different for
weights -2
tfpftttuuuldt
③ Different to
for
-
weights E ⇒
e
-
in
to
small →
available f ?) For 'll the
p
more in
power convenience we see
( solutions )
for identical actuators # we ¥
Usually ,
.
performance I '
I
Ez
PERFORMANCE EQUATION :
I =
R=
€
to I M
loft
.
Replacing J
I EI khz Eze t et VIwel dt
-
weighting .
We VV±z → We use
performance 's
equation → we derive
I We
assign
.
I =
€ z
I
t
D= Eu I
In this care
.
✓ introduced A
Weights the cross product
between I and I
GENERAL form : J -
-
If ! T¥ ! I
¥11
.
dt
We can
choose to model the system such that Izu is not present or
Of course ,
Ohne 's
*
. I :fl¥ at
Problem :
left : J =
'nJ
my
f stabilizing problem
)
Other Than I ,
i want I to be Of
→
TRACKING PROBLEM AGAIN .
SOLUTION → NO PROOF
I =
to
'
It can be shown that there's a closed form solution
E-
-
Ee I assumption ) →
Introducing this assumption
I can find
✓ r
E- E 'RE
I ↳ How the control
The higher
enters into the
the weight
The lower the system
g
.
system
)
is
fourlimited
E
the solution of the
is
equation
:
IIE
EET
t
E -
EEE
BEE -
-
fsimilm
but
different from
lqyaqm.is )
.
very
P is
=
affected by :
•
As : our model
E :
weights of the state
It is :
Ee 0
f usually I 01
•
>
solved
is
automatically by a computer .
28/05
from a
definition in terms
performance ,
we :
"
J -
I/o Iet Q x t ut R u ) It →
problem : ut : J -
.mJ
my
controller :
'
BE p
'
G R
-
P =
SOLUTION Of RICHT 'S EQUATION .
I MATLAB linear
in : Lar
, quadratic regulator ) .
P G
bmigu
→
unique
P is
positive definite and
symmetric .
Moreover
,
we have
solution at
corresponding an
infinite
gain margin
and
very pod
ROBUSTNESS
phase
margin
→ .
This true
is
only if we can feed directly
the state ,
↳ those
otherwise
,
properties are lost !
If we here a
problem
with an S matrix .
the solution
more complex but we recombine ourselves To the
is can
general
care .
I I fish
) at
-
-
{ → Must BE NON .
shown Af usual )
us I + on
We get
a-
iii. Eoff. HEHE . Ill :L at =
TRT
stir )
'
fats
-
I I de
-
I
-
I
⇒ T= -
Ri ST ⇒
⇒ a
zloty lasts :p off
-f¥5tsArt I
at
a-
It'd Tell at
is Ax t Bou → I -
-
A It But I t Buf -0
→ I . -
'T
ABoris I -1
But
A
on .
.
-
Ee
E F F solution the
R ipso where is the
of Rinaldi 's
-
equation in the
fl system .
'
ri =
Art Buu
t Bd d
In particular ,
let's consider a condition Where d .
disturbed padded .
will he a random
response process .
J= Elztvvzzz t ut Khuu ) -
-
Ef # Q x t ut Ru )
I
"
u . -ox G -
-
R But P P -
-
Riccati A a But
, ,
the described
The
dynamic of observer is as .
a- - GI e- =
I A Kyle
-
Let's see if ,
with a
proper
framework .
we car reduce
the effect of the noise .
is Ax t Bout Bdd
NOISE -
RANDOM PROBLEM → STOCHASTIC FRAMEWORK
To One observer I need to here also the
output
deign .
egn
.
Let's consider a
problem without the D matrix :
Eye
①
f-
due to could
t →
noise .
II. n
.
he in general Cn .
n )
with filters )
shaping .
ix. Ae But
Ly
t
It
I A
feel Av
is i Ax t Bout Bdd Lcyx Ln =
- .
-
=
-
-
=
fA - A -
Ky Aer
I Bu -
B) i t
Bdd -
Ln
= 0
=
of
d
d
e' =
fA -
Kyle t Bid -
Ln =
IA Leg )
-
e t I Bd -
L ] I )
n
1
In order to have an optimal Can be
designed I
could
condition find that You
,
we can I by such one
also here
minimizing something managed of error
an
is on us .
J -
-
E Iet Wee e ] →
If I minimize J ,
I minimize the
rec .
error .
Typically falways ) ,
Wee - I became we want to minimize all
the error .
no different relative weights
!
So
,
J -
-
fete
E ]
SECOND REHAK
The fete ) ?
physical meaning of E ⇒
(Trace f scalar ) -
scaler ) ⇒ J=
=THE ])
fete funeral -
-
Eftrfeiel)
prop :
Trl t.rs/--Tr/B.AI ⇒
Ji
EITrf.ee/J--EfTr/eteD-.-TrloeelMlNlMtwoEfete
) means
minimizing the VAN Arik of the reconstruction
error
'd
3 remark .
Providing is an stake →
asyrp
e
.
( A Ley )
-
red tree IA -
Igf tIBD -
L
) ( ? ) I 711=0
( LYAPUNOV EQW since we fore white noises )
| ? ) .
-
! -
-
Own )
ftp.auvinoeeenoenrl-fwdod
.
I
I Ext noise is
Intensity unvbkd
not
MINCES of white typically
to sensor noise
Moises
The represents the constraint of the
Lyapunov equations
minimization .
J-trlreettplfA-kylreetreefA-kyftBDVVddBITLWn.LI/
Lagrange multiplier
{¥
"
= " ° " " ° " ↳ " " "
^
" " " " " ^
" " "
"
"
fconstm.int/LYapfoee
d=0
L )
,
Is (
'
Proof ) L
0 No
oeicjvvnn
-
" IDUAUSMI
his pl ginny
equation in Tee
Are : tree
'
Att Bd Wdd BF
-
f
If Wan L
Wydad
be
Qp
is
small can
large
→
)
.
.
We met finished yet
'
re .
NOTE :
LQO Design state
a
feedback control u=
-
Ci
where Cr -0 LQR I -
FA -
Ky ) It Buu t
Ly
L → KALMW MITER
not perfect .
What happens if ,
instead of A
,
we hare another matrix ?
the be in
Mathematically , uncertainty can
expressed ,
some cases
,
as
DA f ROBUST Desron
,
another chapter ) .
Our
dynamics is
aproned by :
is IATSA ) xx Buu
Gfx -
e ) =
-
Gx t Ge →
FEEDBACK X AND
AN ERROR
Our G is
designed on A .
e' = I -
in =
I A TSAI t Bou
-
A feel -
Bu -
Lcyx ⇒
A A-
;D
I
⇒ et FA
Lcgtbalx ( Bo "
- -
D= Bu
Ky
So . . .
e. =
fA ①
le
-
Lcylet
↳ RESIDUAL TEN
" :D
( At DA Bob )
:
Bob
ii.
i t "
felt
e
-
=
⇒
:L
raise a
.
. .
STRICTLY SPEAKING .
THE DESIGN CAN NOT BE PERFUMED
SEPARATELY ANYMORE .
x. = Ax t Buu MODEL
Let's
suppose
we think that some
dynamics f ie .
flexibility ) are
neglect during
One
design of the controller .
neglected .
if i
Ar .
t.pt Bev
W
Of course ,
the neglected dynamics is influenced by U .
As we can see
.
the Two
equations are
not coupled .
*it : ethno
NB .
The problem can be Written in this decoupled form only if
The here ban derived in Net Givin
dynamics a
proper
way
→ .
i. e .
Cantilever beam
:/
!
#
-
x
-
µ÷
.
to time g.)
we He A-
.
.
.
a . .
i
.
-
Binford
0 ,
rigid body .
they
are
coupled
! →
off -
diag .
terms → We can use another modal
representation
So our representation is valid always that here applied
providing we
the
proper
transformation .
= x.Axt Buu
its Afxr .
t Br .
U
restricted considered
Since we our analysis To the
dynamics only ,
GI
kg
y
.
-
Cgt t Cf XF v. .
-
-
-
Gx .
Ge
← of
EFFECT neglected dynamics .
in But
Ari Ly
-
t
reconstruction will
-
re error be :
.
.
i.
e =
X -
X
e
-
x
-
- -
:
=
A et FA
-
A -
Lcylx t fBuB) -
u - LCE He
¥1 I f AS
If ADDITION A . QUANTITY FORCING TEM )
=
fA
-
Kyle -
Lcrxr
DVNANCS Of SYSTEM I =
fA -
Bu G) x t Bu be
Dynamics of ERROR { e- i
ft .
Kyle
.
Lc . xr
Bro ×
t Br Ge
In material form
:* .lt
"
⇐in :÷÷÷÷÷i÷:L
Since the matrix is not block -
diagonal
→
FULL COUPLING
-
influenced by poles .
§
¥
&*I
is
A
$
X = C. L . POLES
f SEP .
PRINCIPLE)
obs poles
.
As ftp.PNNCIMEI
a
a
× .
→
ACTUAL LOCATION Of POLES
moved )
they
There is a
&Pik0V f CONTROL SPILLOVER -
Br G )
amplify
the
.
spillover effect .
It can
be dangerous .
The residual mode could move towards the
WE HAVE TO CHECK .
spillover effect
c) Replace the model with a new model with additional
dynamics .
LQR →
Gives G in the state feedback controller .
J=
If !IEWzzz t ut Wvu U) dt
energy given
Wzz -
-
prescribed
-
.
sure
To Oke
prescribe stability margin .
.
.
( QUAD
4
Ji I heart i
Rut
.ee?atdt--ffot./eatxIQeedtteatuIEo.e-tldt
We
h
penali.ie the
cost fan with an etatfern
=
ftrunsfuml =
I ItQ It ITEIIdt →
WHICH SYSTEM is
?
ASSOCIATED TO I and on
I Bok
At#
't
't
.
Ije t neat x ⇒ -
-
a ⇒ the design
will provide a
poles
at the left of This new Iii axis
µ
.
L Im
WE IMPOSED A
PRESCRIBED STABILITY f -
-
one .
A-
Ades
ACTUAL : I = A x t Buu B
DESIRED :
I =
X → i.e .
different dumping on a CL BEAM
A-
ftp.ls.w ) ,
→ desired
1%
Lgu , .
)
IMPLICIT
z = A x t Bub -
Ades x =
fA -
AdasI x t
Bu U MODEL
↳ Minniti
FOLLOWING
Uncertainties
Jeff ! Qx t ut Rut
uncertain of
f parameter my system
the
of the system
Changing a changes dynamics .
J ax What
fo
'
to t →
-
- solution ?
k
T penalize One
rensibivitg f
of my solution Wrt a We should transform .
qftafoyfyswseomfngbny-ottfoesp.FM
i Axt Buu to
apply
=
=
foil
fat x +
Agfa +
qdgu.am
↳ Why
not
Buff ?
ffa -
-
iftar
A -
( SENSHI Vnv
Fatten )
. QR