Adaptive Control

Part 5: Direct and Indirect Adaptive Control

1
I.D. Landau: « A Course on Adaptive Control » - 4
 Adaptive Control – A Basic Scheme

SUPERVISION
Adaptation loop

Performance Plant
Controller
specifications Model
Design
Estimation

Adjustable Plant
+ Controller
-

- Indirect adaptive control

- Direct adaptive control (the controller is directly estimated)

2
I.D. Landau: « A Course on Adaptive Control » - 4
 Outline

• Digital control systems

•Tracking and regulation with independent objectives
(known parameters)
• Adaptive tracking and regulation with independent objectives
(direct adaptive control)
• Pole placement (known parameters)
• Adaptive pole placement (indirect adaptive control)

3
I.D. Landau: « A Course on Adaptive Control » - 4
 Digital Control System

The control law is implemented on a digital computer

e(k) u(k) u(t) PLANT y(t) y(k)

r(k)
DAC
+ DIGITAL
COMPUTER + Actuator Process Sensor ADC
- ZOH

CLOCK

ADC: analog to digital converter

DAC: digital to analog converter
ZOH: zero order hold

4
I.D. Landau: « A Course on Adaptive Control » - 4
 Digital Control System

CLOCK

e(k) u(k) u(t) y(t) y(k)

r(k)
DAC
+
COMPUTER + PLANT ADC
- ZOH

DISCRETIZED PLANT

- Sampling time depends on the

system bandwidth
- Efficient use of computer resources

5
I.D. Landau: « A Course on Adaptive Control » - 4
 The R-S-T Digital Controller

CLOCK

r(t) u(t)
D/A y(t)
Computer + PLANT A/D
(controller) ZOH

Discretized Plant

r(t) Bm u(t) y(t)

T + 1 −d
q B
Am S A
-
Plant
R Model
Controller
−1
q y (t )=y (t−1)
6
I.D. Landau: « A Course on Adaptive Control » - 4
 Discrete time model – General form
nA nB
(*) y(t )=− ∑ ai y(t−i)+∑ bi u(t−d−i )
i=1 i=1
d –delay (integer multiple of the sampling period)
nA
−n A +1
−i
1+ ∑ ai q =A( q )=1+q −1 −1 −1
A( q ) ; A( q−1 )=a1 +a2 q−1 +. .. +an q
A

n B i=1
−nB +1
∑ bi q −i −1
=B(q )=q B (q ) −1 −1
; B(q−1 )=b 1 +b 2 q−1 +.. . +bn q
B
i=1
−1 −d −1
(*) A( q ) y(t )=q B(q )u(t )
−1 −1
(*) A( q ) y (t+d )=B( q )u(t ) (Predictive form)
−d −1
−1 q B(q )
(*) y(t )=H (q )u (t ) ; −1
H (q )= - pulse transfer operator
A (q−1 )
−z −1
−1 −1 q B( z )
q →z −1
H ( z )= - transfer function
A ( z−1 )
7
I.D. Landau: « A Course on Adaptive Control » - 4
 First order systems with delay
Fractional delay
−sτ
Continuous time model Ge τ=d . T s +L ; 0 <L<T s
H ( s)=
1+sT
Discrete time z −d (b1 z −1 +b 2 z −2 ) z −d−1 (b 1 +b2 z−1 )
H ( z−1 )= =
model 1+a 1 z
−1
1 +a1 z
−1

L−T T
s
T s s L
−
− T T T
a 1=−e T b1 =G( 1−e ) b 2 =Ge ( e −1)
b2
Remark: For L>0 . 5 T s ⇒ b 2 >b1 ⇒ unstable zero (|− |>1)
+j
b1
b2
<1 z
b1
o o x x o - zero
-a 1 1 x - pole
b2
b1 >1 -j
8
I.D. Landau: « A Course on Adaptive Control » - 4
 Tracking and regulation with independent objectives

It is a particular case of pole placement

(the closed loop poles contain the plant zeros))
It is a method which simplifies the plant zeros
Allows exact achievement of imposed performances

Allows to design a RST controller for:

• stable or unstable systems
• without restrictions upon the degrees of the polynomials A et B
• without restriction upon the integer delay d of the plant model
• discrete-time plant models with stable zeros!!!

Remarks:
•Does not tolerate fractional delay > 0.5 TS (unstable zero)
•High sampling frequency generates unstable discrete time zeros !
9
I.D. Landau: « A Course on Adaptive Control » - 4
 Tracking and regulation with independent objectives

The model zeros should be stable and enough damped

Zero Admissible Zone
1
0.3 0.2

0.8

0.6 f /f = 0.4 0.1

0 s

0.4

0.2
 = 0.1  = 0.2
Imag Axis

-0.2

-0.4

-0.6 0.4 0.1

-0.8

0.3 0.2
-1
-1 -0.5 0 0.5 1
Real Axis

Admissibility domain for the zeros of the discrete time model

10
I.D. Landau: « A Course on Adaptive Control » - 4
 Tracking and regulation with independent objectives
y∗(t+d+ 1)
r(t ) Bm 1 u(t ) −d
q B y(t )
+
T
Am -
S A

R
−(d+1 )
q
P
−(d+ 1)
q
q−( d+1 ) Bm
Am
−1 −1 −1
P(q )=P D (q )P F (q )
−1
Bm (q )
Reference signal (tracking): y (t+d+ 1)= −1
r (t )
A m (q )
Controller: −1 −1
S (q )u(t )=T ( q ) y (t+d+ 1)−R (q ) y (t )
−1

11
I.D. Landau: «(qA) -1Course on Adaptive Control » - 4
 Regulation (computation of R(q-1) and S(q-1))

T.F. of the closed loop without T:

−d+ 1 −1 −d+ 1 −d+1 −1
−1 q B (q ) q q B( q )
H CL (q )= = =
A (q ) S(q )+q B (q ) R( q ) P(q ) B (q−1 ) P( q−1 )
−1 −1 −d+1 −1 −1 −1

The following equation has to be solved :

A( q−1 )S (q−1 )+q−d+1 B( q−1 ) R(q−1 )=B (q−1 ) P(q−1 ) (*)
S should be in the form: S (q−1 )=s 0 +s 1 q−1 +. .. +s n q =B(q−1 ) S' (q−1 )
−n S
S

After simplification by B*,(*) becomes:

−1 −1 −d+1 −1 −1
A( q )S' (q )+q R (q )=P(q ) (**)

Unique solution if: nP = deg P(q-1) = nA+d ; deg S'(q-1) = d ; deg R(q-1) = nA-1
−n A −1 −1 −1 −d
−1 −1
R(q )=r 0 +r 1 q +.. . r n −1 q
S' (q )=1 +s' 1 q +.. . s' d q
A

12
I.D. Landau: « A Course on Adaptive Control » - 4
 Regulation ( computation of R(q-1) and S(q-1))

(**) is written as: Mx = p x = M-1p

nA + d + 1
1 0
0
a1 1
.
a2 a1 0
.
: : 1
.
ad ad-1 ... a1 1
. nA + d + 1
ad+1 ad a1
1 .
ad+2 ad+1 a2
0 .
.
.
.
0
0 0 ... 0 an A 0 0 1

d+1 nA
T ' ' T
x =[ 1,s 1 ,. ..,s d ,r 0 ,r 1 ,...,r n−1 ] p =[ 1 ,p1 ,p 2 ,. . .,p n ,pn
A A +1 , .. . ,p n A +d ]

Use of iReg or predisol.sci(.m) for solving (**)

Insertion of pre specified parts in R and S is possible
13
I.D. Landau: « A Course on Adaptive Control » - 4
 Tracking (computation of T(q-1) )
Closed loop T.F.: r y
q−( d+1 ) B m (q−1 ) Bm (q−1 )T ( q−1 )q−( d+ 1)
H CL (q−1 )= −1
=
A m (q ) A m (q−1 )P (q−1 )
Desired T.F.

It results : T(q-1) = P(q-1)

Controller equation:
−1 −1 −1
S (q )u(t )=P(q ) y ( t+d+1)−R (q ) y (t )
−1 −1
P(q ) y (t+d+1 )−R (q ) y(t )
u(t )=
S (q−1 )
1 −1 −1 −1
u(t )=
b1
[ P( q ) y (t+d+ 1)−S (q )u (t−1 )−R (q ) y (t ) ] (s0 = b1)
−1
Bm (q )
Reference signal (tracking) : y (t+d+ 1)= −1
r (t )
A m (q )
14
I.D. Landau: « A Course on Adaptive Control » - 4
 Tracking and regulation with independent objectives
A time domain interpretation
y* (t+d+1) u(t) y(t)
r(t) Bm
+
1 B∗¿
Am P S q−(d+1) ¿
A
-

−( d+1 ) R
q −1
y (t )= −1
P(q ) y∗(t+d+1 )
P (q ) q−(d+1 )
P

P(q−1 ) y(t )=q−(d+1 ) P (q−1 ) y∗(t+d+ 1) q−(d+1) - + P

−1 −( d+ 1) −1
P(q ) y (t )−q P(q ) y∗(t+d+1 )=0 0
ε (t )≡0
(in case of correct tuning)

Reformulation of the “design problem”:

Find a controller which generate u(t) such that:
0
ε (t+d+ 1)=P [ y (t+d+ 1)− y∗(t+d+ 1) ] =0

15
I.D. Landau: « A Course on Adaptive Control » - 4
 Tracking and regulation with independent objectives
Synthesis in the time domain – an example
*

y* (t+1) u(t) y(t)

For d=0 (S’=1) r(t) Bm
+
1 −1 B∗¿
−1 −1 P S q
A
¿
S (q )=B∗(q ) Am -

A( q−1 )+q−1 R (q−1 )=P(q−1 ) R

R(q−1 )=P∗(q−1 )− A∗( q−1 )
q−1
P(q−1 )=1+q−1 P∗(q−1 ) P
A( q−1 )=1+q−1 A∗(q−1 )
+ P
−1
q -
Example:
y (t+ 1)=−a1 y (t )+b 1 u(t )+b2 u(t−1 ); P(q−1 )=1 +p1 q−1 0
ε (t )≡0
ε° (t+ 1)=P [ y (t+1 )− y∗(t+ 1) ] =y (t+ 1)+p1 y (t )−Py∗(t+ 1)=
[−a1 y ( t )+b1 u(t )+b2 u(t−1 )+p 1 y (t )−Py∗(t+1 )]=0 Solve for u(t)
Py∗(t+ 1)−b 2 u(t−1 )−r 0 y (t )
u(t )= ; r 0 =p 1−a1
b1

P(q−1 ) y∗(t+ 1)=b1 u(t )+b 2 u (t−1)+r 0 y (t )=θT φ(t )

Controller satisfies:
θT =[ b 1 ,b 2 ,r 0 ] φ T (t )= [ u(t ),u (t−1),y ( t ) ]
16
I.D. Landau: « A Course on Adaptive Control » - 4
 Adaptive tracking and regulation with independent objectives
Three techniques:

•Model reference adaptive control (direct)

•Plant model estimation + computation of the controller (indirect)
•Re-parametrized plant model estimation (direct)

17
I.D. Landau: « A Course on Adaptive Control » - 4
 Model Reference Adaptive Control
−1
Objective: P( q ) [ y ( t+ 1)− y∗(t+ 1) ] =0
Py∗( t+1 )− b^ 2 (t )u(t−1 )−r^ 0 ( t ) y (t )
Adjustable controller: u(t )=
b^ (t ) 1
P(q−1 ) y∗(t+1)= θ^ T (t ) φ(t )
θ^ T (t )=[ b^ 1 (t ), b^ 2 (t ),^r 0 (t ) ] ; φT (t )=[ u (t ),u(t−1 ),y(t ) ]
But for the correct values of controller parameters one has:
P(q−1 ) y(t+1)=P(q−1 ) y∗(t+1 )=θT φ(t )
0 T
And therefore one has: ^
ε (t+1 )=[ θ− θ(t ) ] φ(t )
T
^
Define the a posteriori adaptation error: ε (t+1 )= [ θ−θ(t+1 ) ] φ(t )
Use P.A.A.
However one should show in addition that ‖φ(t)‖ is bounded (i.e. plant
input and output are bounded)
18
I.D. Landau: « A Course on Adaptive Control » - 4
Model Reference Adaptive Control

Derivation and
I.D. Landau: « A Course on Adaptive Control » - 4
PAA design [board] 19
Plant model estimation + computation of the controller (indirect)
Step 1 : Plant model estimation
Plant model (unknown): y (t+ 1)=−a1 y (t )+b 1 u(t )+b2 u(t−1 )=θ TP φ (t )

^y 0 (t+ 1)=− a^ 1 (t ) y ( t )+ b^ 1 (t )u(t )+ b^ 2 (t )u(t−1 )=θ^ TP (t )φ(t )

Adjustable predictor: T T
θ^ (t )= [ b^ (t ), b^ (t ),−a^ (t ) ] ; φ ( t )=[ u (t ),u(t −1),y (t ) ]
P 1 2 1

T
a priori prediction error: ε 0 (t+ 1 )=y (t+ 1)− ^y 0 (t+ 1)=[ θ P −θ^ P (t ) ] φ(t )
T
a posteriori prediction error: ε (t+1 )=y (t+ 1 )− y^ (t+1 )= [ θ P− θ^ P (t+1 ) ] φ (t )

Use PAA
Step 2 : Computation of the controller
Compute at each instant t : r^ 0 (t )=p1 −a^ 1 (t )
From θ^ P (t )
Adjustable controller:
−1
P(q ) y∗(t+1)= θ (t ) φ(t ) ^T
φ T (t )= [ u(t ),u(t−1),y ( t )] ; θ^ T ( t )=[ b^ 1 (t ), b^ 2 (t ), r^ 0 (t ) ]

In the general case d > 0 one will have to solve equation (**)
20
I.D. Landau: « A Course on Adaptive Control » - 4
 Re-parametrized plant model estimation (direct)
y(t+1)=−a1 y (t )+b1 u (t )+b2 u(t−1 )± p1 y (t )
Plant model (unknown):
− p1 y(t )+( p1 −a 1 ) y (t )+b 1 u(t )+b2 u(t−1)=− p1 y (t )+θT φ (t )

{
r0
^y 0 (t+1)=− p1 y(t )+ ^r 0 (t ) y(t )+b^ 1 (t )u( t )+ b^ 2 (t )u(t −1)
Re-parametrized
adjustable predictor: − p y(t )+ θ^ T (t )φ (t )
1
θ^ T (t )=[ b^ 1 (t ), b^ 2 (t ), r^ 0(t ) ] ; φ T (t )= [ u(t ),u (t−1),y( t )]
0 0 T
a priori prediction error: ε (t+ 1 )=y (t+1)− ^y (t+ 1)=[ θ−θ^ (t ) ] φ (t )
T
a posteriori prediction error: ^
ε (t+1 )=y (t+1 )− y^ (t+1 )= [ θ−θ(t+1 )] φ(t )

Use PAA
One estimates directly the parameters of the controller

One has:
−1
P( q ) [ y (t+ 1)− y∗(t+ 1) ] =0
21
I.D. Landau: « A Course on Adaptive Control » - 4
 Adaptive tracking and regulation with independent objectives

• Easy generalization for the case d > 0

• Elegant and simple solution for adaptation (direct)
• Unfortunately restricted use in practice because it requires that
the plant zeros remain always stable and well damped

22
I.D. Landau: « A Course on Adaptive Control » - 4
 Direct Adaptive Control – Simulations results
Tracking

Parameters change + adaptation

−1 −1 −1
P(q )=1 P(q )=(1−0.4q )
The choice of the poles for the closed loop (regulation)
has a great influence upon adaptation transient behavior!
23
I.D. Landau: « A Course on Adaptive Control » - 4
 Direct Adaptive Control – Simulations results
Regulation

Constant plant parameters

Parameters change + adaptation

−1 −1 −1
P(q )=1 P(q )=(1−0.4q )
The choice of the poles for the closed loop (regulation)
has a great influence upon adaption transient behavior!
24
I.D. Landau: « A Course on Adaptive Control » - 4
 Indirect Adaptive Control
of non-necessarily zeros- stable plants

25
I.D. Landau: « A Course on Adaptive Control » - 4
 Pole placement

The pole placement allows to design a R-S-T controller for

• stable or unstable systems
• without restriction upon the degrees of A and B polynomials
• without restrictions upon the plant model zeros (stable or unstable)
• but A and B polynomials should not have common factors
(controllable/observable model for design)

It is a method that does not simplify the plant model zeros

26
I.D. Landau: « A Course on Adaptive Control » - 4
 Structure
p(t)

-d + y(t)
r(t) +
-1 1
------------ q B +
T (q)) -1 ---------
S (q ) A
-
PLANT
-1
R(q )

-d -1
q B(q )
--------------------
-1
P (q )

−d −1
−1 q B(q )
Plant: H (q )=
A (q−1 )
−n A −n B
A( q−1 )=1+a 1 q−1 +.. . +an q B(q−1 )=b 1 q−1 +b 2 q−2 +.. . +bn q =q−1 B∗(q−1 )
A B

27
I.D. Landau: « A Course on Adaptive Control » - 4
 Pole placement

Closed loop T.F. (r y) (reference tracking)

−d −1 −1 −d −1 −1
q T ( q )B (q ) q T (q ) B(q )
H CL (q−1 )= =
A (q−1 ) S(q−1 )+q−d B(q−1 )R (q−1 ) P( q−1 )
−1 −1 −1 −d −1 −1 −1 −2
P(q )=A (q )S (q )+q B(q )R (q )=1 +p1 q +p 2 q + .. ..
Defines the (desired) closed loop poles

Closed loop T.F. (p y) (disturbance rejection)

−1 −1 −1 −1
−1 A( q )S (q ) A (q )S (q )
S yp (q )= −1 −1 −d −1 −1
=
A (q )S (q )+q B (q ) R(q ) P( q−1 )

Output sensitivity function

28
I.D. Landau: « A Course on Adaptive Control » - 4
 Choice of desired closed loop poles (polynomial P)

−1 −1 −1
P(q )=P D (q )P F (q )
Dominant poles Auxiliary poles
Choice of PD(q-1) (dominant poles)
Specification discretization
−1
in continuous time 2 order ()
nd P D (q )
(tM, M) Te
0 . 25≤ω 0 T e≤1 . 5
0 . 7≤ζ ≤1
Auxiliary poles
• Auxiliary poles are introduced for robustness purposes
• They usually are selected to be faster than the dominant poles

29
I.D. Landau: « A Course on Adaptive Control » - 4
 Regulation (computation of R(q-1) and S(q-1))

−1 −1 −d −1 −1 −1
(Bezout) A( q )S (q )+q B (q ) R( q )=P(q ) (*)
? ?
−1
n A =deg A ( q ) n B =deg B(q )
−1 A and B do not have
common factors
unique minimal solution for :
−1
n P =deg P(q )≤n A +nB +d−1

n S =deg S (q−1 )=n B +d−1 −1

n R=deg R (q )=n A−1

−1 −1 −n S
S (q )=1 +s 1 q +.. . s n q =1 +q−1 S∗(q−1 )
S

−1 −1 −n R
R(q )=r 0 +r 1 q + .. . r n q
R

30
I.D. Landau: « A Course on Adaptive Control » - 4
 Computation of R(q-1) and S(q-1)

Equation (*) is written as: Mx = p x = M-1p

T T
x =[ 1,s 1 ,...,s n ,r 0 ,...,r n ]
S R
p =[ 1,p1 , ...,pi ,...,p n , 0, ...,0 ]
P

nB + d nA

1 0 ... 0 0 ... ... 0

a1 1 . b' 1
a2 0 b' 2 b' 1
1 . b' 2 nA + nB + d
a1 . .
an A a2 b' n B .
0 . 0 . . .
0 ... 0 an A 0 0 0 b' n B

nA + n B + d
b'i = 0 pour i = 0, 1 ...d ; b'i = bi-d pour i>d
Use of iReg or bezoutd.sci(.m) for solving (*)
31
I.D. Landau: « A Course on Adaptive Control » - 4
 Structure of R(q-1) and S(q-1)

R and S may include pre-specified fixed parts (ex: integrator)

−1 −1 −1 −1 −1 −1
R(q )=R' (q ) H R (q ) S (q )=S' (q ) H S (q )
HR, HS, - pre-specified polynomials
−n S'
−1 −1
R' (q )=r' 0 +r' 1 q + .. .r' n q
−n R'
S' (q−1 )=1 +s' 1 q−1 +.. . s' n q
S'
R'

•The pre specified filters HR and HS will allow to impose certain properties of the
closed loop.
•They can influence performance and/or robustness

−1 −1 −1 −d −1 −1 −1 −1
A( q ) H S (q )S' (q )+q B( q ) H R (q ) R'(q )=P (q )
? ?
32
I.D. Landau: « A Course on Adaptive Control » - 4
 Fixed parts (HR , HS). Examples

Zero steady state error (Syp should be null at certain frequencies)

−1 −1 ' −1
A (q ) H S (q )S (q )
S yp (q−1 )= −1
P( q )
−1 −1
Step disturbance (0 at ω=0 : q = 1): H S ( q )=1−q
−1 −2
Sinusoidal disturbance : H S =1+αq +q ; α=−2cos ωT s
Signal blocking (Sup should be null at certain frequencies)
−1 −1 ' −1
A (q ) H R (q )R (q )
S up (q−1 )=− −1
P (q )
−1 −2
Sinusoidal signal: H R=1 +βq +q ; β=−2cos ωT s
−1 n
Blocking at 0.5fS: H R =(1+q ) ; n=1,2

33
I.D. Landau: « A Course on Adaptive Control » - 4
 Tracking (computation of T(q-1) )
Ideal case
−1
-1 B m (q )
r (t) q Bm y* (t) y* −1
H m (q )= −1
------------ A m (q )
Am
r
Tracking reference desired t Bm (q−1 )=b m0 +bm1 q−1 +. ..
model (Hm) trajectory for y (t) −1 −1 −2
Am ( q )=1+a m1 q +am2 q +. ..
Specification discretization
−1
in continuous time 2 order ()
nd H m (q )
(tM, M) Ts
0 . 25≤ω 0 T s ≤1 .5
0 . 7≤ζ ≤1
The ideal case can not be obtained (delay, plant zeros)
Objective : to approach y*(t)
−( d+ 1) −1
q Bm (q )
y (t )= −1
r (t )
A m (q )
34
I.D. Landau: « A Course on Adaptive Control » - 4
 Tracking (computation of T(q-1) )
−1
Bm ( q )
Build: y (t+d+ 1)= −1
r (t )
A m (q )

Choice of T(q-1) :
• Imposing unit static gain between y* and y
• Compensation of regulation dynamics P(q-1)
−(d +1) −1 −1
1 q T (q )B∗(q )
T(q-1) = GP(q-1) G= H CL (q−1 )=
B(1) P(q−1 )
−( d+1 ) −1
q B m (q ) B∗(q−1 )
F.T. r y: H BF (q−1 )= ⋅
−1
A m (q ) B∗(1 )

1
Particular case : P = Am −1
T (q )=G=
B (1)

35
I.D. Landau: « A Course on Adaptive Control » - 4
Pole placement. Tracking and regulation
y* (t+d+1) u(t) y(t)
r(t) Bm + 1 -d
T q B
Am S A
-

-(d+1) -1
q B*(q )
P(q -1 )
q -(d+1) B*(q -1)
B(1)
-1
q
-(d+1) B (q ) B*(q -1 )
m
-1
A m(q ) B(1)

S (q−1 )u(t )+R(q−1 ) y (t )=T (q−1 ) y∗(t+d+1 )

36
I.D. Landau: « A Course on Adaptive Control » - 4
 Pole placement. Control law
−1 −1
T ( q ) y∗(t+d+ 1)−R(q ) y(t )
u(t )=
S (q−1 )
−1 −1 −1 −1
S(q )u(t )+R(q ) y(t )=GP(q ) y∗(t+d+1)=T (q ) y∗(t+d+ 1)
−1 −1 −1
S (q )=1+q S∗(q )
u(t )=P(q−1 )Gy∗(t+d+1)−S∗(q−1 )u(t−1)−R (q−1 ) y(t )
−1
Bm (q )
y∗(t+d+ 1)= −1
r (t )
Am( q )
−1 −1 −1
Am ( q )=1+q A m∗( q )

y∗(t+d+ 1)=− A m∗(q−1 ) y∗(t+d )+Bm (q−1 )r(t )

−1 −1 −1 −1 −2
Bm (q )=b m0 +bm1 q +. .. Am ( q )=1+a m1 q +am2 q +. ..

37
I.D. Landau: « A Course on Adaptive Control » - 4
 Indirect adaptive control
At each sampling instant:
Step I : Estimation of the plant model ( A^ , B^ )
ARX identification (Recursive Least Squares)

Step II: Computation of the controller

Solving Bezout equation (for S’ and R’)
−d
A^ H S S'+q B^ H R R'=P
Compute:
−1 −1 −1
−1 −1
R(q )=R' (q ) H R (q )
−1
S (q )=S' (q ) H S (q )
T =G
^ P

Remark:
It is time consuming for large dimension of the plant model
38
I.D. Landau: « A Course on Adaptive Control » - 4
 Supervision
Estimation:
• Check if input is enough “persistently exciting”
(if not, do not take in account the estimations)
• Check if A^ and B^ are numerically “sound” (condition number)
(no close poles/zeros)
• If necessary, add external excitation (testing signal)

Control:
• Check if desired dominant closed loop poles are compatible
with estimated plant poles
• Check robustness margins
Additional problem:
• How to deal with neglected dynamics ?
(filtering of the data, robustification of PAA)
39
I.D. Landau: « A Course on Adaptive Control » - 4
 Homework
Homework

40
I.D. Landau: « A Course on Adaptive Control » - 4