[PPT] - A B 1 + 2 = A + B ( I D ) 1 C = 1 PowerPoint Presentation

SLIDE 1

Affine and LFT modeling of systems with uncertainty: some benefits of descriptor forms

Dimitri PEAUCELLE LAAS-CNRS - Universit´ e de Toulouse - FRANCE

µControl’12

Besanc ¸on May 31-June 1, 2012

SLIDE 2

Introduction ■ Outline ❶ Motivations for uncertain descriptor modeling ❷ Affine polytopic models ❸ LFT models - frequency dependent models ❹ Augmented descriptor models and conservatism reduction

1 µControl’12, May 31, Besanc

¸on

SLIDE 3

❶ Motivations for uncertain descriptor modeling

■ Models issued from physics are naturally in descriptor form E ˙ x = Ax + Bu

Example: mechanical system

M ¨ θ + C ˙ θ + Kθ = u M: inertia ; C: friction ; K: stiffness ; u external forces   M O O I     ¨ θ ˙ θ   =   −C −K I O     ˙ θ θ   +   I O   u

Example: robotic systems with Lagrange formulations [MG89]

E may not be invertible

Example: networks of systems with algebraic constraints describing links

2 µControl’12, May 31, Besanc

¸on

SLIDE 4

❶ Motivations for uncertain descriptor modeling

■ Descriptor models can be converted to usual models

Example: mechanical system

¨ θ + M −1C ˙ θ + M −1Kθ = M −1u

Assumes that M −1 is well-conditioned and known
If some parameters are unknown: M(∆), C(∆), K(∆)

¨ θ + M −1(∆)C(∆) ˙ θ + M −1(∆)K(∆)θ = M −1(∆)u

Increased complexity of the model

■ Descriptor models are preferable for describing systems with uncertainty

3 µControl’12, May 31, Besanc

¸on

SLIDE 5

❶ Motivations for uncertain descriptor modeling

■ Example: DC motor I ˙ ω = bu regulated in speed u = −ω

Parameters are assumed uncertain I = 1 + δ1, b = 1 + δ2

(1 + δ1) ˙ ω = −(1 + δ2)ω ⇒ ˙ ω = −1 + δ2 1 + δ1 ω

Model is rational w.r.t. uncertainties: exists an LFT representation

z∆ w

∆

Σ

−1 + δ2 1 + δ1 = A + B∆∆(I − D∆∆)−1C∆ =   A B∆ C∆ D∆   ⋆ ∆ ▲ Can be build with Robust Control toolbox of Matlab or LFRT [Mag05]

4 µControl’12, May 31, Besanc

¸on

SLIDE 6

❶ Motivations for uncertain descriptor modeling

Building the LFT for ˙

ω = −1+δ2

1+δ1ω

▲ 1st step: descriptor form with no denominators ˙ ω + δ1 ˙ ω = −ω − δ2ω ▲ 2nd step: all multiplications correspond to a feedback ˙ ω + w1 = −ω − w2 : w1 = δ1z1 z1 = ˙ ω , w1 = δ2z2 z2 = ω ▲ 3rd step: descriptor LFT: ∆ =

δ1

δ2

, z∆ =
z1

z2

, w∆ =
w1

w2

,

 

1 −1 1 1

 

˙

ω z∆

=

 

−1 −1 −1 1

 

ω

w∆

, w∆ = ∆z∆

▲ last step: invert the left-hand side matrix

5 µControl’12, May 31, Besanc

¸on

SLIDE 7

❶ Motivations for uncertain descriptor modeling

LFT for ˙

ω = −1+δ2

1+δ1ω − 1+δ2

1+δ1 =

2 6 6 4 1 −1 1 1 3 7 7 5

−1 2

6 6 4 −1 −1 −1 1 3 7 7 5 ⋆ 2 4 δ1 δ2 3 5 = 2 6 6 4 −1 −1 −1 −1 −1 −1 1 3 7 7 5 ⋆ 2 4 δ1 δ2 3 5 = −1 + h −1 −1 i 2 4 δ1 δ2 3 5 @I − 2 4 −1 −1 3 5 2 4 δ1 δ2 3 5 1 A

−1 2

4 −1 1 3 5 = −1 − h δ1 δ2 i 2 4 1 + δ1 δ2 1 3 5

−1 2

4 −1 1 3 5 = −1 − h δ1 δ2 i 2 4 − 1+δ2

1+δ1

1 3 5 = −1 + δ1−δ2

1+δ1 = − 1+δ2 1+δ1

■ The example shows the interest of descriptor models,

even if only for technical manipulations

6 µControl’12, May 31, Besanc

¸on

SLIDE 8

❶ Motivations for uncertain descriptor modeling

■ All fractional models have affine descriptor representations [MAS03] ▲ Proof: All fractional models can be converted to an LFT   ˙ x z∆   =   A B∆ C∆ D∆     x w∆   , w∆ = ∆z∆ ▲ the LFT gives the affine descriptor form:   I −B∆∆ O I − D∆∆     ˙ x z∆   =   A C∆   x

Can give representations of smaller dimensions

▲ Example  

1 δ1 δ2 1 + δ1 δ2 1

   

˙ ω z1 z2

  =  

−1 −1 1

  ω ⇔ (1 + δ1) ˙ ω = −(1 + δ2)ω

7 µControl’12, May 31, Besanc

¸on

SLIDE 9

❶ Motivations for uncertain descriptor modeling

■ General descriptor models Exx ˙ x + Exππ = Ax + Bu y + Exy ˙ x + Eyππ = Cx + Du

x: state ; u: inputs
π: linearly constrained signals
Exx and A may not be square

▲ Can be converted to ˆ E ˙ ˆ x = ˆ Aˆ x+ ˆ Bu with ˆ E and ˆ A square and ˙ ˆ x =  

˙ x π λ

  ▲ Not recommend: increased size of the model

8 µControl’12, May 31, Besanc

¸on

SLIDE 10

Outline ❶ Motivations for uncertain descriptor modeling ❷ Affine polytopic models ❸ LFT models - frequency dependent models ❹ Augmented descriptor models and conservatism reduction

9 µControl’12, May 31, Besanc

¸on

SLIDE 11

❷ Affine polytopic models

■ All fractional models have affine descriptor representations Exx(∆) ˙ x + Exπ(∆)π = A(∆)x + B(∆)u y + Exy(∆) ˙ x + Eyπ(∆)π = C(∆)x + D(∆)u

Models also used for polynomial non-linear models [CTF02]

■ General affine descriptor data   −Exx(∆) −Exπ(∆) A(∆) B(∆) −Eyx(∆) −Eyπ(∆) C(∆) D(∆)   = M(∆)

Different classes of affine models

▲ intervals: M M(∆) M

(element-wise mij ≤ mij(∆) ≤ mij)

▲ parallelotopes: M(∆) = M |0| + ¯

p p=1 δpM |p|

: |δp| ≤ 1 ▲ polytopes: M(∆) ∈ co

M [v=1...¯

v]

intervals ⊂ parallelotopes ⊂ polytopes

10 µControl’12, May 31, Besanc

¸on

SLIDE 12

❷ Affine polytopic models

■ Example: two mass spring system

M1 M2 x1 u k f x2eq w1 w2

M1¨ x1 + f( ˙ x1 − ˙ x2) + k(x1 − x2) = u + w1 M2¨ x2 + f( ˙ x2 − ˙ x1) + k(x2 − x1) = w2

General affine data model

  M1 f −f k −k M2 −f f −k k  

=M(∆)

      

¨ x1 ¨ x2 ˙ x1 ˙ x2 x1 x2

       =

u + w1

w2

11

µControl’12, May 31, Besanc

¸on

SLIDE 13

❷ Affine polytopic models

■ Example: two mass spring system - continued

4 uncertain parameters Mi ≤ Mi ≤ M i,

f ≤ f ≤ f, k ≤ k ≤ k M(∆) =

M1

f −f k −k M2 −f f −k k

Interval model generates conservatism (elements assumed independent)

  M1 f −f k −k M2 −f f −k k  



 M1 f −f k −k M2 −f f −k k  



 M1 f −f k −k M2 −f f −k k  

12 µControl’12, May 31, Besanc

¸on

SLIDE 14

❷ Affine polytopic models

■ Example: two mass spring system - continued M(∆) =

M1

f −f k −k M2 −f f −k k

Parallelotopic model

▲ Nominal model: M |0| (center of the intervals)

M(∆) = M|0| + δM1

1 2(M1 − M1)

  1  

. . .

+δk 1 2(k − k)   0 1 −1 −1 1  

M|k|

▲ M |k| ”axis” of variations along δk.

13 µControl’12, May 31, Besanc

¸on

SLIDE 15

❷ Affine polytopic models

■ Example: two mass spring system - continued M(∆) =

M1

f −f k −k M2 −f f −k k

Polytopic model: 24 = 16 vertices (δi = ±1 in parallelotopic model)

M [1] =

M1

f −f k −k M2 −f f −k k

M [2] =
M1

f −f k −k M2 −f f −k k

.

. .

M [16] =

M1

f −f k −k M2 −f f −k k

14

µControl’12, May 31, Besanc

¸on

SLIDE 16

❷ Affine polytopic models

■ Some techniques to deal with affine uncertain models

General type of analysis criteria

ηTΘ(∆)η < 0 , ∀η = 0 : M(∆)η = 0 ▲ Example: negative definite derivative of Lyapunov function V = xTP(∆)x ˙ V = ηT

O

P(∆) P(∆) O

η < 0, ∀η =
˙

x x

= 0 :
−I

A(∆)

η = 0
Usual LMI type result

(M ⊥(∆) matrix generating the null-space of M(∆))

M ⊥T(∆)Θ(∆)M ⊥(∆) < O ▲ Problem: M ⊥(∆) is a rational function of ∆ ▲ Solutions: SOS [HG05, Sch06], Poly´

a [CGTV09, OdOP08]

15 µControl’12, May 31, Besanc

¸on

SLIDE 17

❷ Affine polytopic models

■ Some techniques to deal with affine uncertain models

General type of analysis criteria

ηTΘ(∆)η < 0 , ∀η = 0 : M(∆)η = 0

Usual LMI type result

(M ⊥(∆) matrix generating the null-space of M(∆))

M ⊥T(∆)Θ(∆)M ⊥(∆) < O

Slack variables results

(variables issued from the ”creation” Finsler lemma)

∃F : Θ(∆) < FM(∆) + M T(∆)F T ▲ Conservative, but sufficient to test at the vertices ▲ Θ(∆) affine (parameter-dependent Lyapunov function) ▲ [OBG99, PABB00, EH04, PDSV09]

16 µControl’12, May 31, Besanc

¸on

SLIDE 18

Outline ❶ Motivations for uncertain descriptor modeling ❷ Affine polytopic models ❸ LFT models - frequency dependent models ❹ Augmented descriptor models and conservatism reduction

17 µControl’12, May 31, Besanc

¸on

SLIDE 19

❸ LFT models

■ Affine polytopic models well adapted for scalar parametric uncertainty ■ LFT suitable for other types of uncertainties:

Frequency dependent uncertainties
Linear / non-linear operators
Time-varying parameters
...

z∆ w

∆

Σ

■ Linear combination of repeated uncertainties [Sch06] ∆ =

Jj(Irj ⊗ ∆j)Kj ,

[J1 · · · J¯

] orthonormal

▲ Example: bloc-diagonal uncertainties ∆ =  

∆1 δ2 δ2

 

18 µControl’12, May 31, Besanc

¸on

SLIDE 20

❸ LFT models

■ Dissipative structured parametric uncertainties

An uncertain matrix ∆ is said to be {Ψ1, Ψ2, Ψ3}-dissipative if
I

∆∗



 Ψ1 Ψ2 Ψ∗

2

Ψ3     I ∆   ≤ O .

An uncertain matrix ∆ is said to be {Φ1, Φ2, Φ3}-structured if
I

∆∗



 Φ1 Φ2 Φ∗

2

Φ3     I ∆   = O .

19 µControl’12, May 31, Besanc

¸on

SLIDE 21

❸ LFT models

Examples of {Ψ1, Ψ2, Ψ3}-dissipative uncertainties

▲ Norm-bounded uncertainties are {−γ2I, O, I}-dissipative

I

∆∗ −γ2I O O I I ∆

= −γ2I + ∆∗∆ ≤ O .

w∆ = ∆z∆ ⇒ w∆ ≤ γz∆ ▲ Positive real uncertainties are {O, −I, O}-dissipative

I

∆∗ O −I −I O I ∆

= −(∆∗ + ∆) ≤ O .

w∆ = ∆z∆ ⇒ < w∆|z∆ > ≥ 0

20 µControl’12, May 31, Besanc

¸on

SLIDE 22

❸ LFT models

Examples of {Ψ1, Ψ2, Ψ3}-dissipative uncertainties - continued

▲ Uncertainties in an ”ellipsoid” are {∆∗

0Z∆0 − R, −∆∗ 0Z, Z}-dissipative

I

∆∗ ∆∗

0Z∆0 − R

−∆∗

0Z

−Z∆0 Z I ∆

= (∆ − ∆0)∗Z(∆ − ∆0) − R ≤ O .

(discs of the complex plane for scalar ∆ = δ)

▲ Uncertainties in a ”half space” {−∆∗

0N + N∆0, N ∗, O}-dissipative

I

∆∗ −∆∗

0N + N∆0

N∗ N O I ∆

= (∆ − ∆0)∗N + N ∗(∆ − ∆0) ≤ O .

(half of the complex plane limited by a line for scalar ∆ = δ)

{Φ1, Φ2, Φ3}-structured uncertainties: border of ”ellipsoids” or of ”half spaces”

21 µControl’12, May 31, Besanc

¸on

SLIDE 23

❸ LFT models

{Ψ1, Ψ2, Ψ3}-dissipative {Φ1, Φ2, Φ3}-structured, examples

[IH05]

▲ δ = (jω)−1 with bounded frequencies ω ∈ [ ω ω ]: {2, −j(ω + ω), 2ωω}-dissipative {0, 1, 0}-structured ▲ δ = (jω)−1 with high frequencies ω ≥ ω ≥ 0: {0, −jω, 2ω2}-dissipative {0, 1, 0}-structured ▲ δ = e−jω with bounded frequencies ω ∈ [ ω ω ]: {2 cos ω−ω

2 , −ej ω+ω

2 , 0}-dissipative

{−1, 0, 1}-structured

−ω
ω

ω −j / −j / −ω

▲ δ ∈ R bounded δ ∈ [ δ δ ]: {2δδ, −(δ + δ), 2}-dissipative {0, j, 0}-structured

22 µControl’12, May 31, Besanc

¸on

SLIDE 24

❸ LFT models

■ System analysis in frequency intervals: s−1 = (jω)−1 , ω ∈ [ ω ω ]

Allows loop-shaping type specifications
Have interpretations in time domain [IHF05]

23 µControl’12, May 31, Besanc

¸on

SLIDE 25

❸ LFT models

■ Performances?

Stability: no poles in C+ ⇔ well-posedness w.r.t. s−1 ∈ C+
L2 perf: z2 ≤ γw2 ⇔ well-posedness w.r.t. w = ∇z : ∇2 ≤ 1

γ

▲ Can be tested for given frequencies or for frequency intervals

Dissipativity:

∞

z(t)

w(t)

∗

Ψ1 Ψ2 Ψ∗

2

Ψ3 z(t) w(t)

dt ≥ 0

⇔ well-posedness w.r.t. w = ∇z : ∇ {Ψ1, Ψ2, Ψ3}-dissipative ■ Performances ⇔ well-posedness w.r.t. uncertain operators

24 µControl’12, May 31, Besanc

¸on

SLIDE 26

❸ LFT models

■ Well-posedness [Saf80]

z w w z

For all bounded external signals ( ¯

w, ¯ z)

the internal signals (w, Ez) are unique and bounded

▲ Example: z = ˙ x, w = ∇ ˙ x = x (i.e. ∇ integrator).

Well posedness ⇔

∀ initial conditions ¯ w = x(0) and bounded disturbances ¯ z x is unique, no impulsive modes, no divergence (stability)

25 µControl’12, May 31, Besanc

¸on

SLIDE 27

❸ LFT models

■ Well-posedness [PAHG07]

z w w z

Integral quadratic separation (IQS) result

▲ Assume E = E1E2 with E1 full column rank and let M =

−E1

A

▲ The loop is well-posed iff there exists Θ such that

ηTΘη < 0 , ∀η = 0 : Mη = 0

(1) and for all “uncertainties”

∞   E2z(t) [∇z](t)  

∗

Θ   E2z(t) [∇z](t)   dt ≥ 0

(2)

26 µControl’12, May 31, Besanc

¸on

SLIDE 28

❸ LFT models

IQS result

▲ (1): LMI-type constraint (M ⊥TΘM ⊥ < O) ▲ (2): Integral quadratic constraint (IQC)

IQS applies easily to uncertainties defined, or included in IQCs

▲ Example: ∇ = Ir ⊗ ∆, ∆ is {Ψ1, Ψ2, Ψ3}-dissipative {Φ1, Φ2, Φ3}-structured Θ =   P ⊗ Ψ1 P ⊗ Ψ2 P ⊗ Ψ∗

2

P ⊗ Ψ3   +   R ⊗ Ψ1 R ⊗ Ψ2 R ⊗ Ψ∗

2

R ⊗ Ψ3   P > O R = R∗

(generalized version of D/G-scaling)

27 µControl’12, May 31, Besanc

¸on

SLIDE 29

❸ LFT models

■ IQC also defined in the frequency domain (Parseval) ∞   E2z(jω) [∇z](jω)  

∗

Θ   E2z(jω) [∇z](jω)   dω ≥ 0

(3)

IQC formulas [MR97, JM99] apply for building IQS results
∇ can contain

▲ Saturations [FLR06, PTGSB12] ▲ Delays [JS01, GP07, PAHG07] ▲ ... to be continued ■ Parametrizing Θ solutions to the IQC is in general conservative [MSF97, IH05] ▲ Lossless formulas iff ∆ =

¯ 

j=1

Jj(Irj⊗∆j)Kj    ¯  = 2 , r1 > 1 , r2 = 1 ¯  = 3 , r1 = r2 = r3 = 1

28 µControl’12, May 31, Besanc

¸on

SLIDE 30

Outline ❶ Motivations for uncertain descriptor modeling ❷ Affine polytopic models ❸ LFT models - frequency dependent models ❹ Augmented descriptor models and conservatism reduction

29 µControl’12, May 31, Besanc

¸on

SLIDE 31

❹ Augmented descriptor models and conservatism reduction

■ Except in special cases finding Θ solution to the IQCs is conservative

Existing parameterizations of Θ apply both for constant and time-varying ∆
But ∆

∆∞ ⊂ ∆ ∆d ⊂ ∆ ∆0 where ▲ ∆ ∆0 set of constant uncertainties such that Σ(∆) is stable ▲ ∆ ∆d set of uncertainties with ˙ ∆ ≤ d such that Σ(∆) is stable ▲ ∆ ∆∞ set of TV uncertainties with switches such that Σ(∆) is stable

Reducing conservatism: can be achieved by introducing information about ˙

∆

30 µControl’12, May 31, Besanc

¸on

SLIDE 32

❹ Augmented descriptor models and conservatism reduction

■ Descriptor augmentation method - example

Assume system in LFT form
˙

x = Ax + Bw∆ z∆ = Cx + Dw∆

⋆

x = s−1 ˙

x w∆ = ∆z∆

Add information about derivative of uncertainty:

˙ w∆ = ∆ ˙ z∆

=w∆1

+ ˙ ∆z∆

=w∆2
Define derivative of exogenous signals:

˙ z∆ = C ˙ x + D ˙ w∆ = C ˙ x + Dw∆1 + Dw∆2 , w∆ = s−1 ˙ w∆

31 µControl’12, May 31, Besanc

¸on

SLIDE 33

❹ Augmented descriptor models and conservatism reduction

■ Descriptor augmentation method - example - continued

˙

x = Ax + Bw∆ z∆ = Cx + Dw∆

⋆

x = s−1 ˙

x w∆ = ∆z∆

All equations gathered separating linear constrains and ”uncertainties”

                

˙ x = Ax + Bw∆ z∆ = Cx + Dw∆ ˙ z∆ − C ˙ x = Dw∆1 + Dw∆2 ˙ w∆ = w∆1 + w∆2 0 = w∆ − w∆

⋆                 

x = s−1 ˙ x w∆ = s−1 ˙ w∆ w∆ = ∆z∆ w∆1 = ∆ ˙ z∆ w∆2 = ˙ ∆z∆

32 µControl’12, May 31, Besanc

¸on

SLIDE 34

❹ Augmented descriptor models and conservatism reduction

■ Descriptor augmentation method - example - continued

Descriptor LFT form of system augmented with information on derivatives

z =

˙

x ˙ w∆ z∆ ˙ z∆ T , w =

x

w∆ w∆ w∆1 w∆2 T        

I O O O O O I O −C O O I O I O O O O O O

        z =        

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

        w ⋆ w =     

s−1In+m O O O ∆ O O O ∆ O ˙ ∆ O

     z

33 µControl’12, May 31, Besanc

¸on

SLIDE 35

❹ Augmented descriptor models and conservatism reduction

■ IQS results applicable to the original & the augmented descriptor model

LMI formulas for augmented model are less conservative
IQS formulas implicitly contain Lyapunov certificates

▲ Lyapunov certificate in case of LMI for original model V = xTPx ▲ Parameter-dependent Lyapunov certificate in case of augmented system V =   x w∆  

T

P   x w∆   , w∆ = ∆(I − D∆)−1Cx

The system augmentation technique is also proved efficient for systems with

delays [PAHG07], with saturations [PTGSB12]...

34 µControl’12, May 31, Besanc

¸on

SLIDE 36

Conclusions ■ Some advantages of descriptor models

Natural models from physics
More flexibility for system description
Affine descriptor models are alternatives to LFT models
Well-posedness of descriptor LFT loops

= general framework for robust analysis problems (includes frequency dependent specifications/uncertainties)

System augmentation with derivatives = less conservative results

(made possible thanks to descriptor modeling)

35 µControl’12, May 31, Besanc

¸on

SLIDE 37

REFERENCES

R´ ef´ erences

REFERENCES

References

[CGTV09]

G. Chesi, A. Garulli, A. Tesi, and A. Vicino, Homogeneous polynomial forms for robustness analysis of

uncertain systems, Lecture Notes in Control and Information Sciences, Springer-Verlag, 2009. [CTF02]

D. Coutinho, A. Trofino, and M. Fu, Guaranteed cost control of uncertain nonlinear systems via polynomial

Lyapunov functions, IEEE Trans. on Automat. Control 47 (2002), no. 9, 1575–1580. [EH04]

Y. Ebihara and T. Hagiwara, New dilated LMI characterizations for continuous-time multiobjective controller

synthesis, Automatica 40 (2004), no. 11, 2003–2009. [FLR06]

H. Fang, Z. Lin, and M. Rotea, On IQC approach to the analysis and design of linear systems subject to

actuator saturation, IEEE Conf. Decision and Control (San Diego), 2006. [GP07]

F. Gouaisbaut and D. Peaucelle, Robust stability of time-delay systems with interval delays, IEEE Conference
n Decision and Control (New Orleans), December 2007.

[HG05]

D. Henrion and A. Garulli (eds.), Positive polynomials in control, Lecture Notes in Control and Information

Sciences 312, Springer Verlag, Berlin, 2005. [IH05]

T. Iwasaki and S. Hara, Generalized KYP lemma: Unified frequency domain inequalities with design appli-

cations, IEEE Trans. on Automat. Control 50 (2005), no. 1, 41–59. [IHF05]

T. Iwasaki, S. Hara, and A.L. Fradkov, Time domain interpretations of frequency domain inequalities on

(semi)finite ranges., Systems & Control Letters 54 (2005), no. 7, 681–691.

36 µControl’12, May 31, Besanc

¸on

SLIDE 38

REFERENCES

R´ ef´ erences

REFERENCES

[JM99]

U. J¨
nsson and A. Megretski, IQC characterizations of signal classes, European Control Conference (Karl-

sruhe), September 1999. [JS01]

M. Jun and M.G. Safonov, Multiplier IQC’s for uncertain time-delays, American Control Conference (Arling-

ton), June 2001. [Mag05] J.F. Magni, User manual

f

the linear fractional representation toolbox version 2.0, Tech. report, ONERA

Systems

Control and Flight Dynamics Depertment, October 2005, URL: www.cert.fr/dcsd/idco/perso/Biannic% verb+/toolboxes.html+. [MAS03]

I. Masubuchi, T. Akiyama, and M. Saeki, Synthesis of output-feedback gain-scheduling controllers based
n descriptor LPV system representation, IEEE Conference on Decision and Control, December 2003,
pp. 6115–6120.

[MG89] J.K. Mills and A.A. Goldenberg, Force and position control of manipulators diring constrained motion tasks, IEEE Trans. Robot. Autom. 5 (1989), no. 1, 30–46. [MR97]

A. Megreski and A. Rantzer, System analysis via integral quadratic constraints, IEEE Trans. on Automat.

Control 42 (1997), no. 6, 819–830. [MSF97]

G. Meinsma, Y. Shrivastava, and M. Fu, A dual formulation of mixed µ and on the losslessness of (D, G)-

scaling, IEEE Trans. on Automat. Control 42 (1997), no. 7, 1032–1036. [OBG99] M.C. de Oliveira, J. Bernussou, and J.C. Geromel, A new discrete-time stability condition, Systems & Control Letters 37 (1999), no. 4, 261–265. [OdOP08] R.C.L.F. Oliveira, M.C. de Oliveira, and P .L.D. Peres, Convergent LMI relaxations for robust analysis of un-

37 µControl’12, May 31, Besanc

¸on

SLIDE 39

REFERENCES

R´ ef´ erences

REFERENCES

certain linear systems using lifted polynomial parameter-dependent Lyapunov functions, Systems & Control Letters 57 (2008), 680–689. [PABB00]

D. Peaucelle, D. Arzelier, O. Bachelier, and J. Bernussou, A new robust D-stability condition for real convex

polytopic uncertainty, Systems & Control Letters 40 (2000), no. 1, 21–30. [PAHG07]

D. Peaucelle, D. Arzelier, D. Henrion, and F. Gouaisbaut, Quadratic separation for feedback connec-

tion of an uncertain matrix and an implicit linear transformation, Automatica 43 (2007), 795–804, doi: 10.1016/j.automatica.2006.11.005. [PDSV09] Goele Pipeleers, Bram Demeulenaere, Jan Swevers, and Lieven Vandenberghe, Extended LMI character- izations for stability and performance of linear systems, Systems & Control Letters 58 (2009), no. 7, 510 – 518. [PTGSB12] D. Peaucelle, S. Tarbouriech, M. Ganet-Schoeller, and S. Bennani, Evaluating regions of attraction of LTI systems with saturation in IQS framework, IFAC Symposium on Robust Control Design (Aalborg), June 2012. [Saf80] M.G. Safonov, Stability and robustness of multivariable feedback systems, Signal Processing, Optimization, and Control, MIT Press, 1980. [Sch06] C.W. Scherer, LMI relaxations in robust control, European J. of Control 12 (2006), 3–29.