Control Systems 2

Lecture 10: Robust stability and performance for MIMO systems

Roy Smith

8:15, Wednesday 4th May, 2022

2022-5-3 10.1

Robustness analysis

Nominal stability (NS) Is the closed-loop system stable when the plant is known exactly?

Robust stability (RS) Is the closed-loop system stable when there is uncertainty in our knowledge of
the plant?

Nominal performance (NP) Does the closed-loop system meet the performance specifications when
the plant is known exactly?

Robust performance (RP) Does the closed-loop system meet the performance specifications when
there is uncertainty in our knowledge of the plant?

2022-5-3 10.2
Robustness analysis example: rotating satellite

y1 r1
+
G(s) − Kr
+ −
y2 − r2

 
1 s − α2 α(s + 1)
G(s) = 2
s + α2 −α(s + 1) s − α2

2022-5-3 10.3

Spinning satellite: Bode plots

Magnitude

10

1 10
1 log ω (rad/sec)

G12 (jω)
0.1 G21 (jω) G11 (jω)
G22 (jω)

180
G11 (jω) G22 (jω)
90
G12 (jω)
1 10
0 log ω (rad/sec)

−90 G21 (jω)

−180

−270 Phase (deg.)

2022-5-3 10.4
Singular values: spinning satellite

 
1 s − α2 α(s + 1)
G(s) = 2 , α = 10.
s + α2 −α(s + 1) s − α2

Magnitude
100

10

σ (G)
log ω
1 (rad/sec)
0.1 1 10
σ (G)

0.1

2022-5-3 10.5

An example: rotating satellite

y1 r1
+
G(s) − Kr
+ −
y2 − r2

   
1 0 1 1 −α
K= , Kr = (α = 10 in plots)
0 1 1 + α2 α 1

   
y1 −1 r1
= (I + G(s)K) G(s)Kr
y2 r2
  
1 1 0 r1
=
s + 1 0 1 r2

2022-5-3 10.6
Loop-at-a-time analysis

u1 ζ1
y1 r1
+
G(s) − Kr
+ −
y2 − r2

Loop transfer function: Margins:

−1
ζ1 = u1 Gain margin: ∞
s
Phase margin: 90 degrees

2022-5-3 10.7

Robustness analysis

  v z  
y1 ∆ r1
y2 r2
u
G(s) + + Kr
−

Perturbations (actuator uncertainty):

 
u1 = (1 + δ1 )z1 δ1 0
∆=
0 δ2
u2 = (1 + δ2 )z2

2022-5-3 10.8
Robustness analysis

  v z  
y1 ∆ r1
y2 r2
u
G(s) + + Kr
−

Closed-loop characteristic polynomial:

s2 + (2 + δ1 + δ2 )s + [1 + δ1 + δ2 + (α2 + 1)δ1 δ2 ] = 0

If δ2 = 0 the smallest destabilizing δ1 is δ1 = −1.

2022-5-3 10.9

Robustness analysis

  v z  
y1 ∆ r1
y2 r2
u
G(s) + + Kr
−

Closed-loop characteristic polynomial:

s2 + (2 + δ1 + δ2 )s + [1 + δ1 + δ2 + (α2 + 1)δ1 δ2 ] = 0

If δ2 = 0 the smallest destabilizing δ1 is δ1 = −1.

But it can be destabilized by a much smaller choice of δ1 , δ2 :

1
δ1 = √ ≈ 0.1 and δ2 = −δ1 .
α2 +1

2022-5-3 10.10
Robustness analysis

Stability region
δ2
2

Unstable Stable
region region
1
1 + δ1 + δ2 + (α2 + 1)δ1 δ2 = 0

−2 −1 1 2 δ1

1 + δ1 + δ2 + (α2 + 1)δ1 δ2 = 0

−1
1 + δ1 + δ2 = 0

−2

2022-5-3 10.11

Systems with multiple perturbations

G2 (s) w2 ∆1 W1 (s)
w3

w1
+ + + G1 (s)

e1 G3 (s)
+ G4 (s)

G5 (s) +
w4 + ∆2 W2 (s) e2

G7 (s) + + G6 (s)

W4 (s) ∆4 ∆3 W3

w5
K(s)

2022-5-3 10.12
Common perturbation structures

∆ W
Input perturbation:
y u
G(s) +

∆ W
Output perturbation: y u
+ G(s)

∆ W
Additive perturbation:
y u
+ G(s)

2022-5-3 10.13

General structure for robustness analysis

z v

e N (s) w

N (s) is a stable closed-loop interconnection

∆ is a block-diagonal matrix of perturbations.

 
  ∆1 0
N11 (s) N12 (s)  .. 
N (s) = ∆= . 
N21 (s) N22 (s)
0 ∆m

2022-5-3 10.14
General structure for robustness analysis

z v

N11 (s) N12 (s)

e N21 (s) N22 (s) w

Robust case:
 
e = N21 (s)∆ (I − N11 (s)∆)−1 N12 (s) + N22 (s) w
= Fu (N (s), ∆) w

Nominal case: if ∆ = 0 then e = N22 (s) w.

2022-5-3 10.15

Robust stability analysis

∆ is an unknown stable system.

∆
∆ defines the “structure” of ∆.
z v
∆∈∆
N11 (s)

2022-5-3 10.16
Robust stability analysis

∆ is an unknown stable system.

∆
∆ defines the “structure” of ∆.
z v
∆∈∆
N11 (s)

Is the interconnection stable for given ∆?

∆ is destabilizing ⇐⇒ det(I − N11 (jω)∆(jω)) encircles 0.

2022-5-3 10.17

Robust stability analysis

∆ is an unknown stable system.

∆
∆ defines the “structure” of ∆.
z v
∆∈∆
N11 (s)

Is the interconnection stable for given ∆?

∆ is destabilizing ⇐⇒ det(I − N11 (jω)∆(jω)) encircles 0.

What is the smallest destabilising ∆?

min σ (∆) such that det(I − N11 (jω))∆(jω)) = 0
∆∈∆

2022-5-3 10.18
Structured singular value

∆ is a complex matrix.
∆
M is a complex matrix.
z v


Definition:  0 if no ∆ ∈ ∆ solves det(I − M ∆) = 0



otherwise
µ∆ (M ) =  n o−1



 min β ∃∆, k∆k ≤ β, det(I − M ∆) = 0
∆∈∆

= 1/size of smallest destabilizing ∆.

2022-5-3 10.19

Robust stability

z v

N11 (s) N12 (s)

e N21 (s) N22 (s) w

The closed-loop interconnection is stable for all

∆ ∈ ∆, k∆kH∞ ≤ 1,

if and only if

µ∆ (N11 (jω)) < 1 for all ω.

2022-5-3 10.20
Properties of µ
If an invertible matrix D commutes with all ∆ ∈ ∆, then,
µ∆ (M ) = µ∆ (DM D−1 )

2022-5-3 10.21

Properties of µ
If an invertible matrix D commutes with all ∆ ∈ ∆, then,
µ∆ (M ) = µ∆ (DM D−1 )

∆ D−1 ∆ D

z v
⇐⇒

M D M D−1
⇐⇒

∆ ∆

z v
⇐⇒

DM D−1 D M D−1

2022-5-3 10.22
Properties of µ

For all possible structures, ∆,

µ∆ (M ) ≤ σ (M ) “unstructured case”

So, for every invertible matrix commuting with ∆,

µ∆ (M ) = µ∆ (DM D−1 ) ≤ σ DM D−1

An upper bound:

Search over all invertible matrices that commute with all ∆ ∈ ∆.

Notation: D = { D | D∆D−1 ∈ ∆ for all ∆ ∈ ∆ }


µ∆ (M ) ≤ inf σ DM D−1
D∈D

See mussv command in Matlab.

2022-5-3 10.23

Satellite example (revisited)

  v z  
y1 ∆ r1
y2 r2
u
G(s) + + Kr
−

The transfer function for v to z can be seen via,

z = (I + KG(s))−1 Kr r − (I + KG(s))−1 KG(s) v,

so,

N11 (s) = −(I + KG(s))−1 KG(s).

Robust stability (for perturbations up to size 1)

 
δ1 0
RS ⇐⇒ µ∆ (N11 (jω)) < 1 for all ω, ∆=
0 δ2

2022-5-3 10.24
Robust stability: satellite example

10

8
Robust stability: µ∆ (N11 (jω))

0
1 10 log ω
(rad/sec)

Because µ∆ (N11 (jω)) is not less than 1 for all ω,

the closed-loop system is NOT stable for all perturbations (up to size 1).
It is stable for all diagonal perturbations up to size 1/10.
2022-5-3 10.25

Robust performance

z v

e N (s) w

Nominal performance (∆ = 0):

kek2 ≤ 1 for all kwk2 ≤ 1.

σ (N22 (jω)) < 1 for all ω ⇐⇒ kN22 (s)kH∞ < 1

Robust performance:

kek2 ≤ 1 for all kwk2 ≤ 1 and all ∆ ∈ ∆, k∆kH∞ < 1.

2022-5-3 10.26
Robust performance
As an equivalent robust stability problem

Robust performance

z v

e N (s) w

RS : µ∆ (N11 (jω)) < 1, for all ω

RP : kFu (N (s), ∆) kH∞ ≤ 1
for all ∆ ∈ ∆

2022-5-3 10.27

Robust performance
As an equivalent robust stability problem

Robust performance ⇐⇒ Robust stability

∆ 0
0 ∆p
∆

z v
N11 (s) N12 (s)
e N (s) w N21 (s) N22 (s)

µ∆˜ (N (jω)) < 1, for all ω

RS : µ∆ (N11 (jω)) < 1, for all ω  
˜ ∆ 0
RP : kFu (N (s), ∆) kH∞ ≤ 1 ∆=
0 ∆p
for all ∆ ∈ ∆
∆ ∈ ∆, ∆p ∈ C dim(w)×dim(e)

2022-5-3 10.28
Robust performance

Robust performance:

kek2 ≤ 1 for all kwk2 ≤ 1 and all ∆ ∈ ∆, k∆kH∞ < 1.

Define a larger block structure:

 
˜ ∆ 0
∆= , ∆ ∈ ∆, ∆p ∈ C dim(w)×dim(e)
0 ∆p

Robust performance is satisfied if and only if,

µ∆
˜ (N (jω)) < 1 for all ω.

2022-5-3 10.29

Robust performance example: DV distillation column (revisited)

Nominal plant

 
1 −87.8 1.4
G(s) =
(100s + 1)(s + 1) −108.2 −1.4

Perturbation model
∆ Wm

   
y u
y= 1 G(s) + u= 1
y2 u2

   
(s + 0.2) δ11 δ12 δ11 0
Wm (s) = I2×2 , ∆= or ∆=
(0.5s + 1) δ21 δ22 0 δ22
| {z } | {z }
“unstructured” “structured”
perturbation perturbation

2022-5-3 10.30
Example: nominal performance and robust stability

1.0

0.75


σ Wp (jω)So (jω)
 
0.5 δ11 δ12
∆=
δ21 δ22

0.25
σ (Wm (jω)Ti (jω))

0
0.01 0.1 1 log ω
(rad/sec)

σ (Wp (jω)So (jω)) < 1, for all ω ⇐⇒ nominal performance

σ (Wm (jω)Ti (jω)) < 1, for all ω ⇐⇒ robust stability

2022-5-3 10.31

Example: interconnection model

∆ Wm (s)
y r
G(s) + K(s) +
−

z v

e N (s) w

2022-5-3 10.32
Example: interconnection model

e Wp (s)

∆ Wm (s)
y r
G(s) + K(s) +
w
−

z v

e N (s) w

2022-5-3 10.33

Example: interconnection model

e Wp (s)
v z
Wm (s)
y r
G(s) + K(s) +
w
−

∆ N11 = −Wm K(I + GK)−1 G

z v N12 = Wm K(I + GK)−1
N21 = Wp (I + GK)−1 G
e N (s) w
N22 = Wp (I + GK)−1

2022-5-3 10.34
Example: robust performance (with input cross-coupling)

3.0

µ∆
˜ (N (jω))
2.0
 
δ11 δ12
∆=
δ21 δ22
1.0

σ Wp (jω)So (jω)

σ (Wm (jω)Ti (jω))

0
0.01 0.1 1 log ω
(rad/sec)

σ (Wp (jω)So (jω)) < 1, for all ω ⇐⇒ nominal performance

σ (Wm (jω)Ti (jω)) < 1, for all ω ⇐⇒ robust stability
µ∆
˜ (N (jω)) > 1, for some ω =⇒ robust performance violated

2022-5-3 10.35

Example: robust performance for diagonal input perturbations

3.0

2.0
 
δ11 0
∆=
0 δ22
1.0
µ∆
˜ (N (jω))


µ∆ (N11 (jω)) σ Wp (jω)So (jω)

0
0.01 0.1 1 log ω
(rad/sec)

σ (Wp (jω)So (jω)) < 1, for all ω ⇐⇒ nominal performance

µ∆ (N11 (jω)) < 1, for all ω ⇐⇒ robust stability
µ∆
˜ (N (jω)) < 1, for all ω ⇐⇒ robust performance

2022-5-3 10.36
Notes and references

Skogestad & Postlethwaite (2nd Ed.)

Pertubation models: sections 8.1, 8.2 and 8.3
MIMO Robust stability: sections 8.4, 8.5 and 8.6
Structured singular value (µ∆ ): sections 8.7 and 8.8
Robust performance: section 8.10

2022-5-3 10.37