uncertain z + w Outline Well-posedness and topological - - PowerPoint PPT Presentation

uncertain z w outline well posedness and topological
SMART_READER_LITE
LIVE PREVIEW

uncertain z + w Outline Well-posedness and topological - - PowerPoint PPT Presentation

Quadratic separation for feedback connection of an uncertain matrix and an implicit linear transformation Dimitri PEAUCELLE & Didier HENRION & Denis ARZELIER all with LAAS-CNRS - Toulouse, FRANCE also with Czech Technical


slide-1
SLIDE 1

Quadratic separation for feedback connection of an uncertain matrix and an implicit linear transformation

Dimitri PEAUCELLE & Didier HENRION‡ & Denis ARZELIER all with LAAS-CNRS - Toulouse, FRANCE

‡ also with Czech Technical University in Prague

slide-2
SLIDE 2

Problem statement

Well-posedness of interconnected system

w + + z z w

A, E possibly non square ∇ ∈ ∇ ∇ uncertain

Outline

① Well-posedness and topological separation ② Robust stability of descriptor systems ③ Main result ④ Some results for robust analysis ⑤ Numerical example

1 GT MOSAR 9-10 juin 2005

slide-3
SLIDE 3

Topological separation

General framework

G (z, w)=0

z w

z z w w F (w, z)=0

Well-Posedness: Bounded ( ¯

w, ¯ z) ⇒ unique bounded (w, z) ∃θ topological separator: F(¯ z) = {(w, z) : F¯

z(w, z) = 0} ⊂ {(w, z) : θ(w, z) > −φ1(||¯

z||)} GI( ¯ w) = {(w, z) : G ¯

w(z, w) = 0} ⊂ {(w, z) : θ(w, z) ≤ φ2(|| ¯

w||)}

Related results :

➞ Stability (θ Lyapunov certificate), Passivity (θ storage function), IQC ... ➞ Robust analysis of Linear uncertain systems (Iwasaki)

2 GT MOSAR 9-10 juin 2005

slide-4
SLIDE 4

Well-posedness in the considered case

w + + z z w

Well-posedness: Null(E − A∇) is empty ∀∇ ∈ ∇

∇ ✪ Includes classical µ theory framework: I − A∇ non-singular for all (structured) norm-bounded ∇ ✪ Results of the paper extend to block diagonal uncertainties: ∇ = diag(δR

1 Ir1, . . . , δR NRIrNR, δC 1 Ic1, . . . , δC NCIcNC, ∆C 1, . . . , ∆C Nf) 3 GT MOSAR 9-10 juin 2005

slide-5
SLIDE 5

Well-posedness and descriptor systems

Stability of E ˙

x = Ax ⇔ Es − A full rank for all s ∈ C+ ⇔ W.P

. of     

w = ∇z Ez = Aw

for all ∇ = s−1In ∈ C+

+ + x x

4 GT MOSAR 9-10 juin 2005

slide-6
SLIDE 6

Well-posedness and pole location

Let a region defined as a half plane or a disk

D = { s ∈ C : d1 + d2s + d∗

2s∗ + d3ss∗ ≤ 0 }

D-Stability of E ˙ x = Ax ⇔ Es − A full rank for all s ∈ D ⇔ W.P

. of     

w = ∇z Ez = Aw

for all ∇ = ∇

where ∇

∇ = { s−1In : d1s−1s−∗ + d2s−∗ + d∗

2s−1 + d3 ≥ 0 }. 5 GT MOSAR 9-10 juin 2005

slide-7
SLIDE 7

Well-posedness and polynomial descriptor systems

Stability of Adx(d) + Ad−1x(d−1) + · · · + A1 ˙

x + A0x = 0 ⇔ W.P

. of     

w = ∇z Ez = Aw

for all ∇ = s−1Idn ∈ C+ where E =         

Ad O · · · O O −I O

. . . ...

O O −I

        

A = −

        

Ad−1 · · · A1 A0 I O O

... . . .

O I O

        

6 GT MOSAR 9-10 juin 2005

slide-8
SLIDE 8

Robust stability of descriptor systems

Robust stability ∀∆ ∈ ∆

∆ of E(∆) ˙ x = A(∆)x

with

E(∆) = EA + (B∆ − EB)(ED − D∆)−1EC A(∆) = A + (B∆ − EB)(ED − D∆)−1C ⇔ W.P

. of     

w = ∇z Ez = Aw

for all ∇ =    s−1In

O O ∆

  

: s−1 ∈ C+ ∆ ∈ ∆ ∆

where E =    EA

EB EC ED

  

A =

   A

B C D

  

7 GT MOSAR 9-10 juin 2005

slide-9
SLIDE 9

Main result: Quadratic separation

W.P . of     

w = ∇z Ez = Aw

for all ∇ = ∇

∇ ⇔ ∃Θ :

              

  • I

E◦∗∇∗

  • Θ

  

I ∇E◦

   ≤ O ,

∀∇ ∈ ∇ ∇

  • EE◦

−A

⊥∗

Θ

  • EE◦

−A

> O .

where the columns of E◦ form an orthogonal basis of E∗ and the columns of

  • EE◦

−A

⊥ span the null-space of

  • EE◦

−A

  • .

8 GT MOSAR 9-10 juin 2005

slide-10
SLIDE 10

Application to descriptor systems

Stability of E ˙

x = Ax ⇔

W.P . of     

w = ∇z Ez = Aw

for all ∇ = s−1In ∈ C+. Choice of separator:

  • I

E◦∗s−∗

  • Θ

 

O −E◦∗P −PE◦ O

     

I E◦s−1

   = −2Re(s−1)E◦∗PE◦ LMI result ❶:

E◦∗PE◦ > O

  • EE◦

−A

⊥∗   

O E◦∗P PE◦ O

  

  • EE◦

−A

< O

Equivalent to ❷:

E∗X∗ = XE ≥ O , A∗X∗ + XA < O

9 GT MOSAR 9-10 juin 2005

slide-11
SLIDE 11

Example of descriptor system

Scalar ’switch’ system:    1    ˙

x =

   1

α

   x If α = 0

  • x(t) = 0 the system is stable

  • EE◦

−A

= [ ] →

LMI p > 0

❷ X =

  • x1

x2

LMI x1 ≥ 0 , 2x1 + 2x2α < 0 If α = 0

  • ˙

x = x the system is unstable ❶

  • EE◦

−A

=

   1

1

  

LMI p > 0 , 2p < 0

❷ X =

  • x1

x2

LMI x1 ≥ 0 , 2x1 < 0

10 GT MOSAR 9-10 juin 2005

slide-12
SLIDE 12

Application to uncertain descriptor systems

Robust Stability of E(∆) ˙

x = A(∆)x

, ∆ = δI : |δ| ≤ ¯

δ ⇔

W.P . of     

w = ∇z Ez = Aw

for all ∇ =    s−1In

O O δIm

  

: s−1 ∈ C+ |δ| ≤ ¯ δ

. Choice of separator:

Θ =

     

E◦∗

2 Θ1E◦ 2

−E◦∗

1 P

E◦∗

2 Θ2

−PE◦

1

O O Θ2∗E◦

2

O Θ3

     

: ∇E◦ =

   s−1E◦

1

∆E◦

2

   constrained by the LMIs               

E◦∗

1 PE◦ 1 ≥ O ,

E◦∗

2 Θ3E◦ 2 ≥ O

E◦∗

2 (Θ1 + ¯

δΘ2 + ¯ δΘ2∗ + ¯ δ2Θ3)E◦

2 ≤ O

E◦∗

2 (Θ1 − ¯

δΘ2 − ¯ δΘ2∗ + ¯ δ2Θ3)E◦

2 ≤ O 11 GT MOSAR 9-10 juin 2005

slide-13
SLIDE 13

Reducing conservatism ◆ Original system

   EA

EB EC ED

      ˙

x z∆

   =    A

B C D

      x

w∆

   , ∇ =    s−1In

O O δIm

  

❖ Augmented system

            

O I O O O EA EB O O EC ED O EA O O EB EC O O ED

                     

¨ x ˙ x z∆ ˙ z∆

        

=

            

I O O O O A B O O C D O A O O B C O O D

                     

˙ x x w∆ ˙ w∆

        

, ∇ =

   s−1I2n

O O δI2m

  

12 GT MOSAR 9-10 juin 2005

slide-14
SLIDE 14

Reducing conservatism ◆ Quadratic separation on Original system ➞ Lyapunov stability with V (η, ∆) = η∗Pη ❖ Quadratic separation on Augmented system ➞ Lyapunov stability with V (η, ∆) = η∗P(∆)η

where in the case E = I (not descriptor)

P(∆) =

   A(∆)

I

  

P

   A(∆)

I

  

13 GT MOSAR 9-10 juin 2005

slide-15
SLIDE 15

Robust stability analysis example

E =

              

8 4 8 10 9 10 3 5 3 1 3

              

, A =

              

3 −16 9 2 −8 −16 −19 −17 −20 1 4 1 2 −10 −6 3 7 −6

              

◆ Quadratic separation on Original system ➞ LMIs feasible up to ¯ δ = 0.22 - infeasible for ¯ δ = 0.23 ❖ Quadratic separation on Augmented system ➞ LMIs feasible up to ¯ δ = 0.45 - infeasible for ¯ δ = 0.46

14 GT MOSAR 9-10 juin 2005

slide-16
SLIDE 16

Conclusions ➞ Quadratic separation for feedback connection of an uncertain matrix and an

implicit linear transformation is valuable for extensions of known results to de- scriptor systems.

➞ Applying existing methods to artificially augmented systems gives new less

conservative results (more about this in Seville)

15 GT MOSAR 9-10 juin 2005