33
Separation Principle

Separation Principle

Embed Size (px)

DESCRIPTION

Separation Principle. Controllers & Observers. Plant Controller Observer. Picture. Composite state model. Similarity Transformation. Controllability of Composite System. Composite system cannot be controllable. (rows of zeros) Look at transfer function of composite system - PowerPoint PPT Presentation

Citation preview

Page 1: Separation Principle

Separation Principle

Page 2: Separation Principle

Controllers & Observers

( ) ( ) ( )

ˆ

( )

ˆ( ) ( ) ( )

(

ˆ( ) ( ) ( ) ( )

)

) (

u t Kx t

x t A L

x t Ax t B

C x t Bu t

u t y t Cx

y t

t

L

r t

Plant

Controller

Observer

Picture

Page 3: Separation Principle

Composite state model

( ) ( ) ( ) ( ) ( )

ˆ ˆ( ) ( ) ( ) (

ˆ( ) ( )

) ( )

ˆ( )ˆ

(

( )(

)

0(

))

)(

x t A LC x t Bu t Ly t

L

A B x t B

C A LC BK x tx t

x t Ax t Bu t y t Cx t

x

u t Kx t r t

Kr t

t

Page 4: Separation Principle

Similarity Transformation

1

( )

( ) 0

ˆ ˆ( )

(ˆ( )ˆ( )

0 0

0

()

0

) Kr t

A B x t B

C B

x t I x xP

e t I I x x

IP P

I I

I A B I

I I C I I

A BK I

A LC A

L A LC BK x tx t

L A LC B

C

K

L I

K

I

x t

0

( ) ( )

( ) 0 ( ) 0

A BK BK

A LC

x t A BK BK x t Br

e t A LC e t

Page 5: Separation Principle

Controllability of Composite System

2 2 1

( ) ( )

( ) 0 ( ) 0

0 0 0 0

nc c c

Comp

x t A BK BK x t Br

e t A LC e t

B A B A A

Composite system cannot be controllable. (rows of zeros)•Look at transfer function of composite system

•Poles and zeros

Page 6: Separation Principle

Stability Discussion

• K stabilizes “control system”

• L stabilizes “observer”

• Values of K and L are largely independent of each other

• Observer must be “more stable” than controller

• Separation Principle

Page 7: Separation Principle

Satellite Example

Page 8: Separation Principle

More Stability

Page 9: Separation Principle

Energy Storage & State

component Energy/Power

Resistor I2R=V2/R

Capacitor, v V2/(2C)

Inductor, i LI2/2

Friction element Bv2/2

Mass, v Mv2/2

Spring, dx K(dx)2/2

Page 10: Separation Principle

Free Response and Energy

• Free response indicates how energy stored in the system “leaves” the system

( ) (0)Atx t e x• If x(t) goes to zero, then the energy initially

stored in the system (kinetic, potential, inductors, capacitors) is dissipated in friction elements, resistors, or radiates to the ambient world

Page 11: Separation Principle

Forced Response and Energy

• The forced response shows how energy from the input function (and bias sources) enters the states

Page 12: Separation Principle

Stability and Energy Dissipation

• The conceptual (not literal) connection between energy stored in the states and whether or not it dissipates to zero in the free response, leads to a powerful test of stability associated with the name of Lyapunov.

• See Chapter 8: pps. 193, 198, 200, 209

Page 13: Separation Principle

Definitions

• The origin is stable if, for any given value of ε > 0, there is a value of δ(ε) > 0 such that if ||x(0)|| < δ then ||x(t)|| < ε for all t > 0 (see Fig. 8.1 pg. 195).

• The origin is asymptotically stable if it is stable, and if there exists δ’ > 0 such that whenever ||x(0)|| < δ’ then limt||x(t)|| = 0.

• Diagonal systems

Page 14: Separation Principle

Energy Functions & Lyapunov stability

• Lyapunov functions energy functions• Consider a mechanical system with no

energy sources, consider any function built up from kinetic and potential energy of its components. This function does not increase as time passes.

• Consider an electrical system with no energy sources, consider any function built from the energy stored in the capacitors and inductors. This function does not increase as time passes.

Page 15: Separation Principle

2nd Method of Lyapunov

Along every solution x(t) contained in Ω

1. If v(x(t)) is decreasing except at the origin, then the origin is asymptotically stable, and all initial conditions inside Ω produce trajectories that approach the origin;

2. If v(x(t)) is increasing except at the origin, then the origin is unstable.

Page 16: Separation Principle

Energy Functions for LTI systems• v(x) = xTPx, P symmetric positive-definite.

( ) ( )

( )

( )

T T

T T

T T T

T T

T

T

x t Ax t

v x x Px x Px

Ax Px x PAx

x A Px x PAx

x A P PA x

x Qx

A P PA Q

• Choose real, symmetric, positive definite Q.

• Find, if possible, real, symmetric, positive definite P

• If successful, conclude stability.

• Approach generalizes to non-linear systems

Page 17: Separation Principle

Positive/Negative (semi-) definite matrices, x != 0

• xTQx > 0 positive definite• xTQx < 0 negative definite• xTQx >= 0 positive semi-definite• xTQx <= negative semi-definite

• Symmetric QT=Q• Real every element of Q is real• Q is positive definite if, for some orthogonal S, STQS is

– Diagonal,– Has non-zero, positive elements on the diagonal

• Diagonal matrix examples

Page 18: Separation Principle

Positive Definiteness

• A real symmetric matrix M is positive definite if and only if any one of the following conditions holds:

– All eigenvalues of M are positive

– There exists a nonsingular matrix N such that M = NTN

Page 19: Separation Principle

Connection between Q and P• Q is any positive definite matrix and let

( ) ( ) ( ), ( ) (0)T Atv t x t Qx t x t e x • Total energy dissipated is

0

0

0 0

( ( )) ( ) ( ) (0) (0)

(0) (0) (0) (0)

T

Tt

A t At

t

t t tT T A t At

t t t

T T

v t dt x t Qx t dt x e Qe x

e Qe

dt

x x x xdt P

• Now consider ATP+PA

Page 20: Separation Principle

ATP+PA=?

0

0 0

0 0

0

lim

0

T T

T T

T T

T

T

T T

t tT

t tA t At A t At

t t

A t At A t At A t A

A t At A t A

t t

t

t

t

t

t t

A t At

P P e Qe dt e Qe dt

e Qe e Qe

A A A A

dA e Qe

e Qe e Q

A dt dtdt

e Q e

Q Q

e

Page 21: Separation Principle

Solving Lyapunov Eqns

• ATP+PA=-Q– Given Q, A, find P– “vectorize” to get Mp=q– Solve for p– Reconstruct P from p

– Solutions exist “if and only if” λi+ λj ≠0 for eigenvalues of A, including when i = j.

Page 22: Separation Principle

Satellite Example

1,1 1,2 1,1 1,2 1,1 1,2

1,2 2,2 1,2 2,2 1,2 2,2

1,2 1

2

,1

2

,2

,2 1,2

1,2

2,2

0 1

0 0

0 1 0 0

0 0 1 0

2

0

T

A

AP PA Q

p p p p q q

p p p p q q

p q

p q

q

p q

Q cannot be totally arbitrary since q2,2 must be 0.Can Q and P be positive definite?

Page 23: Separation Principle

Satellite Example

1,1 1,2 1,1 1,1

1,2 1,1 1,2

2 21,1 1,1 1,22 2 2

1,1 1,2 1,1 1,2 , /

21,1 1,2 1,1 , /

2,

0 1

0 0

/ 2

0 / 2

4det( ) ( )

2

det( ) ( )( ) / 4

det( )

T

Q

P

A

A AP PA Q

q q p qQ P

q q q

q q qsI Q s s q q s q s q

sI P s p s q q

sI A s

, , , ,0 0A A A A

Q cannot be positive definite since it has a negative EV.

Page 24: Separation Principle

With Feedback

11 12 11 12 11 12

12 22 12 22 12 22

12 11

11 12 2

22 11 1

2 12 12 22 22

2 12

0 1

2 2

0 1 0 2

2 2 1 2

2

2 2 4 4

0 2 0

2

2 2

2 1

0 4 4

fb

Tfb fb

A A BK

A P PA Q

p p p p q q

p p p p q q

p q

p p p

p p p q

q p p q

11 111

12 12

22 22

11 11 12 22

12 11

22 11 22

6 4 11

4 0 084 0 2

6 4 /8

/ 2

2 / 4

p q

p q MP Q M

p q

p q q q

p q

p q q

Any Q produces a P.Does pos-def Q produce pos-def P?

Page 25: Separation Principle

With Feedback

11 12 11 12 11 12

12 22 12 22 12 22

11

12

22

211 22 12

0 1

2 2

0 1 0 2 1 0

2 2 1 2 0 2

1

1/ 2

1

det( ) ( )( ) ( 1)( 2)

det(

Tfb fb fbA A BK A P PA Q

p p p p q q

p p p p q q

p

p

p

sI Q s q s q q s s

s

2 2 2

11 22 11 22 12) ( ) 2 3/ 4

( 1.5)( .5)

I P s p p s p p p s s

s s

Any Q produces a P.Does pos-def Q produce pos-def P?

Page 26: Separation Principle

With Feedback and Observer

Page 27: Separation Principle

Uncancelled pole-zero pairs

• CCF– Observable?

• OCF– Controllable?

• Subspaces– CO– Not C, O– Not O, C– Not O, Not C

• Controllable if input affects all state components directly or indirectly

• Observable if all state components affects output directly or indirectly

Page 28: Separation Principle

Example

23 2

0

1( )

3 3 10 1 0 0

0 0 1 0 1 1 0

1 3 3 1

0

( )

( )

0 1 0

1 2 1 1 0

1

2

??

1

10 1

0

0

0

c

c

o

sH s

s s s

x x u y x

rank

rank

z z u y z

z Px

s

P

s

Page 29: Separation Principle

Kalman Canonical form (1)

0 0

00

c c cc c cc c

c c c

o c o oo

o oo c o o

x A A x Bu y C C x

x A x

x A x Bu y C x

x A A x B

Uncontrollable states can affect controllable states.Controllable states cannot affect uncontrollable states.

Observable states can affect unobservable states.Unobservable states cannot affect observable states.

Picture.

Compute Transfer functions.

Page 30: Separation Principle

Kalman Canonical Form (2)

?0

? ?

0 0 0 0

0 ? 0

0 0

0

?

0co co co co

co co co co

co co co

co co co

co

coco co

co

co

x A x B

x A x Bdu

x A xdt

x A x

x

xy C C

x

x

Compute Transfer Function.

Depends only on Observable and controllable portion.

Hence lower degree polynomials.

Page 31: Separation Principle

Design via Similarity Transformations -- Picture

Page 32: Separation Principle

Signal Energy

• Example 7, pg. 202

• Define signal energy

• Compute energy in free response due to x(0)

• P is solution to Lyapunov Eqn.

Page 33: Separation Principle

Optimal Control• See Example 8, Pg 203

• Introduce SVFB, need to choose K

• Define quadratic cost functional in terms of – Positive definite R– Positive semidefinite Q

• Substitute y=Cx, u=-Kx+u’

• Substitute free response, with initial state x(0)

• Compare integrand with signal energy integrand, solve resulting Lyapunov eq.