Sie sind auf Seite 1von 67

ROBUST CONTROL:

ANALYSIS AND DESIGN

Olivier SENAME
GIPSAlab Control Systems department
INPG-CNRS
INPG CNRS
ENSE3-BP 46
38402 Saint Martin d'Hères Cedex, FRANCE
Olivier.Sename@grenoble-inp.fr

Why Robust H∞ control


„ MIMO systems
„ Performance specifications linked to control design
„ Analysis of robustness properties
„ Design of robust controllers
„ Advanced optimisation tools for control synthesis
„ Extensions: Gain-scheduling, Linear Parameter Varying
systems…

2
Robust control : analysis and design» © Olivier Sename 2010

1
Bibliography
S. Skogestad and I. Postlethwaite, Multivariable Feedback Control: analysis
and design, John Wiley and Sons, 2005.
www.nt.ntnu.no/users/skoge

K. Zhou, Essentials of Robust Control, Prentice Hall, New Jersey, 1998.


www.ece.lsu.edu/kemin

J.C. Doyle, B.A. Francis, and A.R. Tannenbaum, Feedback control theory,
Macmillan Publishing Company, New York, 1992.
www.control.utoronto.ca/~francis

G.C. Goodwin, S.F. Graebe, and M.E. Salgado, Control System Design,
Prentice Hall, New Jersey, 2001.
http://csd.newcastle.edu.au/

G. Duc et S. Font, Commande Hinf et -analyse: des outils pour la robustesse,


Hermès, France, 1999.

D. Alazard, C. Cumer, P. Apkarian, M. Gauvrit, et G. Ferreres, Robustesse et


commande optimale, Cépadues Editions, 1999.
3
Robust control : analysis and design» © Olivier Sename 2010

OUTLINE

‰ Motivation Industrial examples

‰ H∞ norm, stability

‰ Performance analysis/specifications Performances quantifiers


A first robustness criteria

‰ H∞ control design Mixed sensitivity problem

‰ Uncertainties and robustness Representing uncertainties


Robust stability
stability, Robust performance
Robust control design

‰ Performances limitations Bode and Poisson sensitivity integral

4
Robust control : analysis and design» © Olivier Sename 2010

2
INTRODUCTION

Modern industrial plants have sophisticated control systems crucial to their successful
operation:
– robotics
– aerospace
– semiconductor manufacturing industry
– nuclear industry
– Energy production and distribution
– ……...
– automotive industry :
» SI and Diesel engines
» suspension
» braking
» Global chassis control
» intelligent
g highways
g y
» driver supervision
» ……….

5
Robust control : analysis and design» © Olivier Sename 2010

AUTOMOTIVE CONTROL
Some actual important fields of investigation concern:
1. Environmental protection (Limiting of pollutant emissions –
Nox; CO; CO2) Engine control
Automatic driving
Traffic optimisation
Energy consumption optimisation
Electrical and Hybrid vehicles

2. Road safety and monitoring (decrease the number of


accidents)
Braking in dangerous situations
Detection of critical situations
Chassis control
Traffic control
Driver assistance (stop & start, anti-collision)
“by wire” technology
Diagnosis of embedded system
6
Robust control : analysis and design» © Olivier Sename 2010

3
ENGINE CONTROL

ƒ injection control (Common


Rail))
ƒ idle-speed control
ƒ air to fuel ratio control
ƒ cylinder balancing
ƒ Torque control
ƒ throttle control
ƒ EGR + VGT
ƒ driveline control
ƒ Post-treatment
ƒ Energy recovery
ƒ Downsizing

7
Robust control : analysis and design» © Olivier Sename 2010

VEHICLE DYNAMICS CONTROL

Topics
Active Control for safety and comfort
Multi actuators (suspensions, braking, steering)

Methodologies:
- Physical and behavioral modelling
- H∞ control: LPV, fault-tolerant
- On line adaptation of
comfort/handling criteria
- Global Chassis Control ( suspension,
suspension
braking, steering)

Robust control : analysis and design» 8 © Olivier Sename 2010

4
Analysis and robustness of frequency synthesizers

– Modelling and
optimization of
frequency
synthesizer
lloops ffully
ll
integrated on
chip
– Analysis of
semi-global
stability,
robustness,
b t
observation
and robust
control

9
Robust control : analysis and design» © Olivier Sename 2010

CONTROL OF GLASS FIBER BUSHING

Process

Objective :enhance product


quality i.e.,
- avoid variations of fiber
diameter
-less production breaks
accounting for disturbances
(air, input glass temp.)
- Robustness requirements
as bushings are changed
Control scheme each year
- MIMO system

10
Robust control : analysis and design» © Olivier Sename 2010

5
DVD PLAYERS CONTROL

Control problem: minimize


position error between the
D=120mm
laser spot and the real track
Disk
d=15mm position, both in the radial and
in the vertical direction.
Tracking movement
Neither the track position nor
in-out
the true spot position can be
Focus movement
up-down
measured.

Robustness pb: non ideal


System Controller
Measurement unit and D/A conv.
construction of the device and
Photodiodes

pickup A/D
non perfect location of the hole
Pre-processing Digital
Current
unit Converter controller
Amplifier
p
at the center of the disc
6300 and 6301

Actuation unit and D/A conv


Motors PWM
Current
(PDM)
Amplifier unit

pickup TDA...

11
Robust control : analysis and design» © Olivier Sename 2010

OUTLINE

‰ Motivation Industrial examples

‰ H∞ norm, stability

‰ Performance analysis/specifications Performances quantifiers


A first robustness criteria

‰ H∞ control design Mixed sensitivity problem

‰ Uncertainties and robustness Representing uncertainties


Robust stability
stability, Robust performance
Robust control design

‰ Performances limitations Bode and Poisson sensitivity integral

12
Robust control : analysis and design» © Olivier Sename 2010

6
About H∞ norm: MIMO GAIN

For a SISO system, y=Gd, the gain at a given frequency is simply

y (ω ) G ( jω )d (ω )
= = G ( jω )
d (ω ) d (ω )

The gain depends on the frequency, but since the system is linear it is
independent of the input magnitude

For a MIMO system we may select :

y (ω ) G ( jω )d (ω )
2
= 2
= G ( jω )
d (ω ) 2
d (ω ) 2
2

Which is « independent » of the input magnitude. But this is not a correct


definition. Indeed the input direction is of great importance

13
Robust control : analysis and design» © Olivier Sename 2010

MIMO GAIN

⎡5 4⎤
G=⎢ ⎥ How to define and evaluate its gain ??
⎣3 2⎦
Five different inputs

d1 = 1 d2 = 0 d3 = 0.7070 d4 = 0.7070 d5 = 0.6000


0 1 0.7070 -0.7070 -0.8000
Input magnitude : norm2= 1
Norm(d1)=norm(d2) =norm(d3)=norm(d4)=norm(d5)=1

Corresponding outputs

y1 = 5 y2 = 4 y3 = 6.3630 y4 = 0.7070 y5 = -0.2000


3 2 3.5350 0.7070 0.2000

and gains
5.8310 4.4721 7.2790 0.9998 0.2828

14
Robust control : analysis and design» © Olivier Sename 2010

7
MIMO GAIN

MAXIMUM SINGULAR VALUE = 7.34


7
Gd
max 2
= σ (G ) 6
d ≠0 d 2
5

|| y ||2 / || d ||2
4

Gd 3
min 2
= σ (G )
d ≠0 d 2
2

MINIMUM SINGULAR VALUE = 0.27


0
-5 -4 -3 -2 -1 0 1 2 3 4 5
d20 / d10

We see that, depending on the ratio d20/d10, the gain varies between
0.27 and 7.34 .
15
Robust control : analysis and design» © Olivier Sename 2010

MIMO GAIN

Eigenvalues are a poor measure of gain. Let


⎡0 100⎤
G=⎢ ⎥
⎣0 0 ⎦
Eigenvalues are 0 and 0
⎡0 ⎤ ⎡100⎤
But an input vector ⎢ ⎥ leads to an output vector ⎢ ⎥.
⎣1 ⎦ ⎣ 0 ⎦
Clearly the gain is not zero.
Now, the maximal singular value is = 100
It means that any signal can be amplified at most 100 times
This is the good gain notion.

16
Robust control : analysis and design» © Olivier Sename 2010

8
MIMO GAIN

In the case of a transfer matrix G(s) : (m inputs, p outputs)


u vector of inputs, y vector of outputs

y (ω )
σ (G ( jω )) ≤ 2
≤ σ (G ( jω ))
u(ω ) 2

Example of
A two-mass/spring/damper system

2 inputs: F1 and F2
2 outputs: x1 and x2

17
Robust control : analysis and design» © Olivier Sename 2010

MIMO GAIN

A=[0 0 1 0;
0 0 0 1;
-k1/m1 k1/m1 -b1/m1 b1/m1;
k1/m2 -(k1+k2)/m2 b1/m2 -(b1+b2)/m2];
B=[0 0;0 0;1/m1 0;0 1/m2];
Singular Values
C=[1
C [ 0 0 0;0
; 1 0 0];
];
30
D=[0 0;0 0]; Hinf norm: 21.1885 dB
20

largest singular value


10
Control system toolbox
Singular Values (dB)

G1=ss(A,B,C,D) : LTI system 0

Tf(G1) : transfer function -10

normhinf(G1) smallest singular value


-20
sigma(G1)
-30

Mu-analysis toolbox -40

G2=pck(A,B,C,D) -50
-1
10 100 101
hinfnorm(G2) Frequency (rad/sec)
>> norm between 11.4704 and 11.4819
18
Robust control : analysis and design» © Olivier Sename 2010

9
MIMO GAIN

Mathematical backgrounds

19
Robust control : analysis and design» © Olivier Sename 2010

PERFORMANCE ANALYSIS

Well-posedness
dy(t
di(t) ) s −1
r(t)
+
e(t)
K(s) u(t) + G(s) + y(t) G=− , K =1
+ + s+2
-
+
+ n(t)

s+2 s −1
Therefore the control input is non proper: u= (r − n − d y ) + di
3 3

DEF: A closed-loop system is well-posed if all the transfer functions are proper

⇔ I + K ( ∞ )G ( ∞ ) is invertible

In the example 1+1x(-1)=0


Note that if G is strictly proper, this always holds.

20
Robust control : analysis and design» © Olivier Sename 2010

10
PERFORMANCE ANALYSIS

More important : Internal Stability


DEF: A system is internally stable if all the transfer functions of the closed-
loop system are stable

di(t)
r(t) ε(t) u(t) + ⎛ y ⎞ ⎛ ( I + GK ) −1 GK ( I + GK ) −1 G ⎞⎛ r ⎞
+ K(s) G(s) y(t)
⎜⎜ ⎟⎟ = ⎜⎜ ⎟⎜ ⎟
- +
⎝ u ⎠ ⎝ K ( I + GK )
−1
− K ( I + GK ) −1 G ⎟⎠⎜⎝ d i ⎟⎠

⎛ 1 s +1 ⎞
1 s −1 ⎛ y ⎞ ⎜⎜ s + 2 ⎟
( s − 1)( s + 2) ⎟⎛⎜ r ⎞⎟
For instance : G= , K= , ⎜⎜ ⎟⎟ =
s −1 s +1 ⎝ u ⎠ ⎜⎜ s − 1 1 ⎟⎝⎜ d i ⎟⎠
− ⎟
⎝s+2 s+2 ⎠

There is one RHP pole (1), which means that this system is not internally
stable.
This is due here to the pole/zero cancellation (forbidden!!).

21
Robust control : analysis and design» © Olivier Sename 2010

OUTLINE

‰ Motivation

‰ H∞ norm, stability

‰ Performance analysis/specifications
Sensitivity functions
Some criteria and performances quantifiers
Example
MIMO case
A first robustness criteria
‰ H∞ control design Performance specification

‰ Uncertainties and robustness


‰ Performances limitations

22
Robust control : analysis and design» © Olivier Sename 2010

11
PERFORMANCE ANALYSIS

Objectives of any control system :

shape the response of the system to a given reference and


get (or keep) a stable system in closed-loop, with desired
performances, while minimising the effects of disturbances
and measurement noises, and avoiding actuators
saturation, this despite of modelling uncertainties,
parameter changes or change of operating point.

23
Robust control : analysis and design» © Olivier Sename 2010

PERFORMANCE ANALYSIS

Objectives of any control system

Nominal stability (NS): The system is stable with the nominal model (no
d l uncertainty)
model t i t )

Nominal Performance (NP): The system satisfies the performance


specifications with the nominal model (no model uncertainty)

Robust stability (RS): The system is stable for all perturbed plants about
the nominal model, up to the worst-case model uncertainty
((including
g the real plant)
p )

Robust performance (RP): The system satisfies the performance


specifications for all perturbed plants about the nominal model, up to the
worst-case model uncertainty (including the real plant).

24
Robust control : analysis and design» © Olivier Sename 2010

12
FEEDBACK STRUCTURE

In the following:
SISO (Single Input Single Output) and MIMO systems
((Multi Input
p Multi Output)
p ) are considered

SOME CONTROL STRUCTURES

« Classical » one degree-of-freedom structure

Two degree-of-freedom structure

RST structure

25
Robust control : analysis and design» © Olivier Sename 2010

FEEDBACK STRUCTURE

Classical one degree-of-freedom structure

Input Output
disturbance disturbance
reference
f
di(t) dy(t) Output
r(t) + u(t) + uP(t) + y(t)
K(s) G(s) +
Control + Plant
- Input Input

+
+ n(t)
PLANT = G(s) Measurement
CONTROLLER = K(s) FEEDBACK noise

26
Robust control : analysis and design» © Olivier Sename 2010

13
FEEDBACK STRUCTURE

Two degree-of-freedom structure

di(t) dy(t)
r(t) + + + y(t)
Kp(s) G(s) +
+
-
+
+
K(s)
n(t)
FEEDFORWARD
FEEDBACK
Improves tracking performance

27
Robust control : analysis and design» © Olivier Sename 2010

FEEDBACK STRUCTURE

RST structure
di(t) dy(t)
r(t)
(t) u(t) + + y(t)
+
T 1/S G +
+
-
+
+
R
n(t)
A two DOF structure with:
POLYNOMIALS Kp= T/S and K=R/S

28
Robust control : analysis and design» © Olivier Sename 2010

14
FEEDBACK STRUCTURE

GENERAL CONTROL CONFIGURATION

Disturbance
and reference w e Controlled Output
p
P
Control Input u y Measured output

P is the generalized plant (contains the plant, the weights,


the uncertainties if any) ; K is the controller

29
Robust control : analysis and design» © Olivier Sename 2010

SENSITIVITY FUNCTIONS

Input Output
disturbance disturbance
reference dy(t)
di(t)
r(t)
() e(t)
() u(t) + + y(t) Output
+ K(s) G(s)
+ +
-
+
+ Measurement
n(t) noise
Firstly, SISO case

The output & the control input satisfy the following equations :
1
y( s) = (GKr + d y − GKn + Gd i )
1 + G (s) K (s)
1
u (s) = ( Kr − Kd y − Kn − KGd i )
1 + K ( s )G ( s )

30
Robust control : analysis and design» © Olivier Sename 2010

15
SENSITIVITY FUNCTIONS

1
y( s) = (GKr + Gd i + d y − GKn)
1 + G ( s) K ( s)
1
u (s) = ( Kr − KGd i − Kd y − Kn)
1 + K ( s )G ( s)
Let us define the well known sensitivity functions:
1
Sensitivity S (s) =
1 + G (s) K ( s)

Complementary T (s) =
G (s) K (s)
Sensitivity 1 + G (s) K ( s)

Loop transfer function L( s ) = K ( s )G ( s)

31
Robust control : analysis and design» © Olivier Sename 2010

SENSITIVITY FUNCTIONS

Input and Output Performance analysis


using the Sensitivity functions

Input Output
disturbance disturbance
reference
f dy(t)
di(t)
r(t) e(t) u(t) + + y(t) Output
+ K(s) G(s)
+ +
-
+
+ Measurement
n(t) noise

The output & the control input performances can be


studied through 4 « sensitivity » functions only.

32
Robust control : analysis and design» © Olivier Sename 2010

16
SENSITIVITY FUNCTIONS

1
y (s) = (GKr + d y − GKn + Gd i )
1 + G (s) K ( s)
1
u ( s) = ( Kr − Kd y − Kn − KGd i )
r(t) 1 + K ( s )G ( s )
KS(s)
() r(t)
T(s)
di(t)
-T(s) di(t)
u(t) SG(s)
dy(t) Σ y(t)
dy(t) Σ
-KS(s)
S(s)
n(t)
-KS(s) n(t)
-T(s)

Input performance Output performance


33
Robust control : analysis and design» © Olivier Sename 2010

SENSITIVITY FUNCTIONS

1
u (s) = ( Kr − KGd i − Kd y − Kn)
1 + K ( s)G ( s )

r(t) The transfer function KS(s) should be upper bounded


KS(s)
() so that u(t) does not reach the physical constraints,
even for a large reference r(t)
di(t)
-T(s)
u(t)
dy(t) Σ
-KS(s) The effect of the input disturbance di(t) on the plant
input u(t)+ di(t) (actuator) can be made « small » by
n(t) making
ki the
th sensitivity
iti it function
f ti S(s)S( ) smallll
-KS(s)
The effect of the measurement noise n(t) on the plant
input u(t) can be made « small » by making the
sensitivity function KS(s) small (in High Frequencies)
Input performance

34
Robust control : analysis and design» © Olivier Sename 2010

17
SENSITIVITY FUNCTIONS

1
y( s) = (GKr + Gd i + d y − GKn)
1 + G( s) K (s)
The plant output y(t) can track the reference r(t) by
making the complementary sensitivity function T(s)
equal
q to 1. (servo
( pb)
p ) r(t)
T(s)
The effect of the output disturbance dy(t) (resp. input
disturbance di(t) ) on the plant output y(t) can be made di(t)
« small » by making the sensitivity function S(s) (resp. SG(s)
y(t)
SG(s) ) « small »
dy(t) Σ
The effect of the measurement noise n(t) on the plant S(s)
output y(t) can be made « small » by making the
complementary sensitivity function T(s) « small » n(t)
()
-T(s)
BUT S(s) + T(s) = 1
Output performance
Some trade-offs are to be looked for

35
Robust control : analysis and design» © Olivier Sename 2010

SENSITIVITY FUNCTIONS
r(t)
These trade-offs can be reached if one aims : KS(s)
• to reject the disturbance effects in low frequency di(t)
• to minimize the noise effects in high frequency -T(s)
u(t
dy(t) Σ

We will require: -KS(s)

• S and SG to be small in low frequencies to reduce n(t)


the load (output and input) disturbance effects on -KS(s)
the controlled output
r(t)
T(s)
• T and KS to be small in high frequencies to
reduce the effects of measurement noises on the di(t)
SG(s)
()
controlled
t ll d output
t t andd on the
th control
t l input
i t (actuator
( t t y(t)
()
Σ
efforts) dy(t)
S(s)

n(t)
-T(s)

36
Robust control : analysis and design» © Olivier Sename 2010

18
PERFORMANCE ANALYSIS: CRITERIA

Time domain performances


Classical performance indices
Rise time : the time, usually required to be small, that it takes for the output to
first reach 90 % of its final value

Settling time : the time after which the output remains within 5 % of its final value,
which is also usually required to be small.

Overshoot: the peak value divided by the final value : should typically be 1.2 (20
%) or less

Decay ratio: the ratio between the second and first peaks, which should typically
be 0.3 or less

Steady-state offset: the difference between the final value and the desired final value,
this offset is usually required to be small.

37
Robust control : analysis and design» © Olivier Sename 2010

PERFORMANCE ANALYSIS: CRITERIA

Time domain performances : other criteria



ISE (Integral Square Error) = ∫ e(t ) dt ; e = r − y
2
J ISE
0


ITAE (Integral Time
weighted Absolute Error)
J ITAE = ∫ t e(t ) dt
0
; e=r−y

A better and more advisable index should


include the control input effect

∫ (Q e(t ) )

= + R u (t ) dt
2 2
J eu
0

38
Robust control : analysis and design» © Olivier Sename 2010

19
PERFORMANCE ANALYSIS: CRITERIA

Frequency domain performances: criteria

GAIN, PHASE, DELAY and MODULE MARGINS


The g
gain margin
g indicates the additional g gain that would Good value for
take the closed loop to the critical stability condition Gm : >6dB

The phase margin quantifies the pure phase delay that Good value for
should be added to achieve the same critical stability Φm : > 30/40°
condition
The delay margin quantifies the maximal delay that should ΦM
Δτ =
be added in the loop to achieve the same critical stability ωΦ
condition M

The module margin quantifies the minimal distance


between the curve and the critical point (-1,0j): this is a
robustness margin

39
Robust control : analysis and design» © Olivier Sename 2010

PERFORMANCE ANALYSIS: CRITERIA

Frequency domain performances: criteria

DELAY MARGIN :
acceptable
p pure
p time-
delay before instability

ΦM
Δτ =
ωΦ M

40
Robust control : analysis and design» © Olivier Sename 2010

20
PERFORMANCE ANALYSIS: CRITERIA

MODULE MARGIN

i 1 + GK ( jω )
ΔM = min
ω

Good value ΔM > 0.5

41
Robust control : analysis and design» © Olivier Sename 2010

PERFORMANCE ANALYSIS: CRITERIA

Bandwidth
The concept of bandwidth is very important in understanding the benefits and
trade-offs involved when applying feedback control. Above we considered peaks
off closed-loop
l dl transfer
f functions,
f i which
hi h are related
l d to the
h quality
li off the
h response.
However, for performance we must also consider the speed of the response, and
this leads to considering the bandwidth frequency of the system.

In general, a large bandwidth corresponds to a faster rise time, since high


frequency signals are more easily passed on to the outputs. A high bandwidth also
indicates a system which is sensitive to noise and to parameter variations.
Conversely, if the bandwidth is small, the time response will generally be slow,
and the system will usually be more robust.

42
Robust control : analysis and design» © Olivier Sename 2010

21
PERFORMANCE ANALYSIS: CRITERIA

Bandwith :
Loosely speaking, bandwidth may be defined as the frequency range [w1,
w2] over which control is effective. In most cases we require tight control
att steady-state
t d t t so w1=0,
0 and
d we then
th simply
i l callll w2 the
th bbandwidth.
d idth
The word “effective” may be interpreted in different ways : globally it
means benefit in terms of performance.

Definition1: The (closed-loop) bandwidth, wS, is the frequency where


|S(jw)| crosses –3dB (1/√2) from below.

Remark: |S|<0.707, frequency zone, where e/r = -S is reasonably small

43
Robust control : analysis and design» © Olivier Sename 2010

PERFORMANCE ANALYSIS: CRITERIA

Another interpretation: when it changes the output response.


As y=Tr, T must be sufficiently large

Definition2: The bandwidth (in term of T), wT, is the frequency where
√ from
|T(jw)| crosses –3dB (1/√2) f above.

Remark: In most cases, the two definitions in terms of S and T yield


similar values for the bandwidth. In other cases, the situation is
generally as follows. Up to the frequency wS, |S| is less than 0.7, and
control is effective in terms of improving performance. In the frequency
range [wS, wT] control still affects the response
response, but does not improve
performance. Finally, at frequencies higher than wT, we have S ≅ 1 and
control has no significant effect on the response.

44
Robust control : analysis and design» © Olivier Sename 2010

22
PERFORMANCE ANALYSIS: CRITERIA

The gain crossover frequency :

Definition3: The bandwidth (crossover frequency), wC, is the frequency


where |L(jw)| crosses 1 (0dB), for the first time, from above.

Remark:It is easy to compute and usually gives

wS < wC < wT

Note that the rise time can often be evaluated as :

t r = 2 .3
ωT

45
Robust control : analysis and design» © Olivier Sename 2010

PERFORMANCE ANALYSIS: example

Position control of a DC motor, using internal speed feedback

Ω x Vs
Ve + +
A 1/kv 1/(τp+1) 1/(np) U0/(2Π)
-
-

RCp/(RCp+1) 1/kv 1/kv

46
Robust control : analysis and design» © Olivier Sename 2010

23
PERFORMANCE ANALYSIS: example

S T

Good disturbance rejection


20 5

0 0
ar Values (dB)

ar Values (dB)
-20 -5
Singula

Singula
-40
40 -10
10

-60 -15

-80
10-2 10-1 100 101 102
Good noise rejection
-20
10-2 10-1 100 101 102

Frequency (rad/sec)WB=15.3 rad/s Frequency (rad/sec)


WBT=27.1 rad/s
KS SG

0 10

5
alues (dB)

alues (dB)
-20

0
Singular Va

Singular Va
-40
-5

Bad: control input sensitive


-60

Bad: Input disturbance (di)


-10

to noise -80
10-2 10-1 100 101 102
are not rejected
-15
10-2 10-1 100 101 102

Frequency (rad/sec) Frequency (rad/sec)

47
Robust control : analysis and design» © Olivier Sename 2010

PERFORMANCE ANALYSIS: example

2.5

Input disturbance di

2
Output disturbance dy

1.5

05
0.5

reference step
0
0 1 2 3 4 5 6 7 8 9 10

48
Robust control : analysis and design» © Olivier Sename 2010

24
PERFORMANCE ANALYSIS: example

Bandwith : And it holds : >>[Gm,Pm,W180,Wc]=margin(sys)


WS=15.3 rad/s >> MT=hinfnorm(T)MS=hinfnorm(S);
wS < wC < wT
WT=27.1 rad/s Phase margin= 72.4 deg
Wc=21 rad/s Gain margin = inf
Module margin
g < 1.5 db,, MT=0.5db
Bode Diagrams
From: U(1)
100
Nichols Charts
From: U(1)
100 50

deg); Magnitude (dB)


0

50

-50
n-Loop Gain (dB)

0 -100
To: Y(1)

-80

-50
50
Open

Phase (d

-100
100

-120
To: Y(1)

-100
ωc=27 rad/s
-140

-160

-150
-180 -170 -160 -150 -140 -130 -120 -110 -100 -90 -180
100 101 102
Open-Loop Phase (deg)
Frequency (rad/sec)

49
Robust control : analysis and design» © Olivier Sename 2010

PERFORMANCE ANALYSIS: example

Bandwith : 1.4

WS=15.3 rad/s
1.2
WT=27.1 rad/s
1

Wc=21 rad/s 0.8

It holds : 0.6

wS < wC < wT 0.4


rise time

0.2
and :
0
1 1.05 1.1 1.15 1.2 1.25 1.3 1.35 1.4 1.45 1.5

t r = 2.3 = 85ms
ωT

50
Robust control : analysis and design» © Olivier Sename 2010

25
PERFORMANCE ANALYSIS : MIMO case

Sensitivity functions: MIMO case


Input Output
disturbance disturbance
reference dy(t)
di(t)
r(t) ε(t) u(t) + + y(t) p outputs
+ K(s) G(s)
+
m control +
-
inputs +
+ Measurement
n(t) noise

The output & the control input satisfy the following equations :
( I p + G ( s ) K ( s )) y ( s ) = (GKr + d y − GKn + Gd i )
( I m + K ( s )G ( s ))u( s ) = ( Kr − Kd y − Kn − KGd i )

BUT : K(s)G(s) ≠ G(s)K(s)

51
Robust control : analysis and design» © Olivier Sename 2010

PERFORMANCE ANALYSIS : MIMO case

Sensitivity functions
( I p + G ( s) K ( s)) y ( s) = (GKr + d y − GKn + Gd i )
( I m + K ( s)G ( s))u( s) = ( Kr − Kd y − Kn − KGd i )

Output and Output complementary sensitivity functions:


S y = ( I p + GK ) −1 , T y = ( I p + GK ) −1 GK , S y + Ty = I p

Input and Input complementary sensitivity functions:


S u = ( I m + KG ) −1 , Tu = KG ( I m + KG ) −1 , S u + Tu = I m

Properties Ty = GK ( I p + GK ) −1
Tu = ( I m + KG ) −1 KG
S u K = KS y

52
Robust control : analysis and design» © Olivier Sename 2010

26
PERFORMANCE ANALYSIS : MIMO case

r(t) r(t)
KSy(s) Ty(s)
di(t) di(t)
-Tu(s) SyG(s)
()
u(t) y(t)
(t)
dy(t) Σ Σ
dy(t)
-KSy(s) Sy(s)

n(t) n(t)
-KSy(s) -Ty(s)

Input performance Output performance


f

( I p + G ( s ) K ( s )) y ( s ) = (GKr + d y − GKn + Gd i )
( I m + K ( s)G ( s ))u( s ) = ( Kr − Kd y − Kn − KGd i )

53
Robust control : analysis and design» © Olivier Sename 2010

PERFORMANCE ANALYSIS : MIMO case

As MIMO framework is concerned, MIMO gain of the sensitivity


functions is considered through Frequency-domain plots of the singular
values (sigma using the Control toolbox, and using the Mu toolbox:

‰ The analysis of SISO systems can then be extended, except for


stability margins.
‰ 5 sensitivity functions have to be studied (Sy, Ty, Tu, Ksy, SyG)
‰ The robustness margins are the maximum peak of S and T. These
may not be sufficient to ensure robustness properties and should be
completed by a robust stability analysis

54
Robust control : analysis and design» © Olivier Sename 2010

27
A first approach to ROBUSTNESS ANALYSIS

Introduction: Skegestad & Postlewaite


A control system is robust if it is insensitive to differences between the
actual system and the model of the system which was used to design the
controller
How to take into account the difference between the actual system and
the model ?

A solution: using a model set BUT : very large problem and not
exact yet
A method: these differences are referred as model uncertainty.
Th approach
The h
‰ determine the uncertainty set: mathematical representation
‰ check Robust Stability
‰ check Robust Performance
55
Robust control : analysis and design» © Olivier Sename 2010

A first approach to ROBUSTNESS ANALYSIS

56
Robust control : analysis and design» © Olivier Sename 2010

28
A first approach to ROBUSTNESS ANALYSIS

MODULE MARGIN/
Maximum Peak criteria

i 1 + GK ( jω )
ΔM = min
ω
Also :

ΔM = 1
MS
M S = max S ( jω ) = S ∞
ω

Good value MS < 2 (6 dB)

57
Robust control : analysis and design» © Olivier Sename 2010

PERFORMANCE ANALYSIS

The MODULE MARGIN is a robustness margin.

ΔM = 1
MS

Indeed, the sensitivity function allows to qualify the robustness of the


control system, as
K ( s )G ( s )
Closed-loop transfer function TBF =
1 + K ( s )G ( s )
IInfluence
fl off plant
l t modelling
d lli errors on the
th ΔTBF 1 ΔG
CL transfer function
=
TBF 1 + K ( s )G ( s ) G
Sensitivity function

58
Robust control : analysis and design» © Olivier Sename 2010

29
A first approach to ROBUSTNESS ANALYSIS

Advantage: good module margin implies


good gain and phase margins

MS 1
GM ≥ and PM ≥
M S −1 MS

For MS=2, then GM>2 and PM>30°

Last one : M T = max T ( jω )


ω

Good value MT < 1.5 (3.5 dB)

59
Robust control : analysis and design» © Olivier Sename 2010

A first approach to ROBUSTNESS ANALYSIS

Robust Stability analysis: SISO case


yΔ uΔ
Let us consider the case : ΔI(s) +
u z
+ G(s)

The loop transfer function is then: L p = G p K = GK ( I + wI Δ I ) = L + wI LΔ I ;


Therefore RS ⇔ System stable ∀ Lp. ⇔ Lp should not encircle the point -1
Im
-1
Re

⇔ wI L < 1 + L , ∀ω
1+ L RS
wI L
⇔ < 1, ∀ω
1+ L
L ( jω )
⇔ wI T < 1 ∀ω
wI L

60
Robust control : analysis and design» © Olivier Sename 2010

30
PERFORMANCE SPECIFICATION

Objective : good performance specifications are


important to ensure better control system

mean : give some templates on the sensitivity


functions

For simplicity, presentation for SISO systems first.

61
Robust control : analysis and design» © Olivier Sename 2010

PERFORMANCE SPECIFICATION

Templates on the sensitivity functions


Robustness and performances in regulation can be specified by imposing
frequential templates on the sensitivity functions.

If the sensitivity functions stay within these templates, the control


objectives are met.

These templates can be used for analysis and/or design. In the latter they
are considered as weights on the sensitivity functions

The shapes of typical templates on the sensitivity functions are given in the
following slides

62
Robust control : analysis and design» © Olivier Sename 2010

31
PERFORMANCE SPECIFICATION

Template on the sensitivity function - Weighted sensitivity

Typical specifications in terms of S include:


1. Minimum bandwidth frequency wS (defined as the frequency where
|S(jw)| crosses 0.707 from below).
2. Maximum tracking error at selected frequencies.
3. System type, or the maximum steady-state tracking error, ε
4. Shape of S over selected frequency ranges.
5. Maximum peak magnitude of S, ||S||∞ < MS.

The peak specification prevents amplification of noise at high


frequencies, and also introduces a margin of robustness; typically we
select MS=2.

63
Robust control : analysis and design» © Olivier Sename 2010

PERFORMANCE SPECIFICATION

Template on the sensitivity function


Weighted sensitivity

Mathematically, these specifications may be captured by an


Mathematically
upper bound, on the magnitude of S, given by another
transfer function :

1
S ( jω ) ≤ , ∀ω ⇔ We S ≤1
W e ( jω ) ∞

where We(s) is a weight selected by the designer.

64
Robust control : analysis and design» © Olivier Sename 2010

32
PERFORMANCE SPECIFICATION

Template on the sensitivity function S ( s) =


1
1 + K ( s )G ( s )
1 s + ω bε
=
We ( s ) s / MS + ωb
Generally ε = 0 is considered,
MS <2 (6dB) or (3dB - cautious) to
ensure sufficient module margin

ωb ≥ ωS required
ωS obtained ≥ ωb fixed
ωb influences the CL bandwidth : ωb ↑
¾ faster rejection of the disturbance
¾ faster CL tracking response
¾ better robustness w.r.t. parametric uncertainties

65
Robust control : analysis and design» © Olivier Sename 2010

PERFORMANCE SPECIFICATION

K ( s )G ( s )
Template on the complementary sensitivity function T (s) =
1 + K ( s )G ( s )
ε T s + ω BT
s + ω BT / M T
Generally εT = 0 is considered,
MT <2 (6dB) or (3dB - cautious) to
ensure sufficient module margin

wBT influences robustness : wBT ↓


¾ better noise effects rejection
¾better filtering of HF modelling errors

66
Robust control : analysis and design» © Olivier Sename 2010

33
PERFORMANCE SPECIFICATION

K (s)
Template on the sensitivity function KS(s) KS ( s ) =
1 + K ( s )G ( s )
1 ε s + ω BC
= 1
Wu ( s ) s + ω BC / Mu
Mu chosen according to LF
behavior of the process
(actuator constraints:
saturations)

wBC influences robustness : wBC ↓


¾ better limitation of measurement noises
and roll-off starting from wBC to reduce modeling errors effects

67
Robust control : analysis and design» © Olivier Sename 2010

PERFORMANCE SPECIFICATION

G ( s)
Template on the sensitivity function SG(s) SG ( s ) =
1 + K ( s )G ( s )
s MSG
s + ωSG MSG
ωSG
Limitation of input disturbance effects
on the output by the choice of wSG
zero static error for constant input
disturbance

68
Robust control : analysis and design» © Olivier Sename 2010

34
PERFORMANCE SPECIFICATION

The templates can be defined more accurately by transfer


functions of order greater than 1, as

k
⎛ s / MS + ωb ⎞
We ( s ) = ⎜⎜ ⎟⎟ ,
⎝ s + ω b ε ⎠
if we require a roll-of of –20*k dB per
decade is required

In the MIMO case the simplest way is to defined the templates as


diagonal transfer matrices, i.e. using (Msi, ωbi, εI, ….)

69
Robust control : analysis and design» © Olivier Sename 2010

THE MIXED SENSITIVITY PROBLEM

In terms of control synthesis, all these specifications can be tackled in


the following problem: find K(s) s.t.

We S WT T
≤1
Wu KS WSG SG ∞

which is called a mixed sensitivity problem. Often, the simpler following


one is studied:
We S
≤1
Wu KS ∞

The latter allows to consider the closed-loop output performance as well


as the actuator constraints.

70
Robust control : analysis and design» © Olivier Sename 2010

35
OUTLINE

‰ Motivation Industrial examples

‰ H∞ norm, stability

‰ Performance analysis/specifications Performances quantifiers


A first robustness criteria

‰ H∞ control design Mixed sensitivity problem

‰ Uncertainties and robustness Representing uncertainties


Robust stability
stability, Robust performance
Robust control design

‰ Performances limitations Bode and Poisson sensitivity integral

71
Robust control : analysis and design» © Olivier Sename 2010

H∞ CONTROL APPROACH

H∞ control is devoted to SISO systems as well as MIMO ones

Main advantage: the plant modelling errors as well as


the disturbance and noise effects can be specified in the
frequency domain.

H∞ control is strongly linked to the weighted sensitivity


functions.

P f
Performance specification
ifi ti iis then
th off greatt iimportance
t iin
H∞ control approach.

72
Robust control : analysis and design» © Olivier Sename 2010

36
H∞ CONTROL APPROACH

This approach has been introduced by Doyle (1983). The formulation


makes use of the general control configuration.

GENERAL CONTROL CONFIGURATION


Disturbance
w ⎛P P12 ⎞ e Controlled Output
and reference
P = ⎜⎜ 11 ⎟
⎝ P21 P22 ⎟⎠
Control Input u y Measured output

P is the g
generalized plant
p (contains
( the plant,
p , the weights,
g , the
uncertainties if any) ; K is the controller. The closed-loop transfer
function is:

Tew ( s ) = Fl ( P, K ) = P11 + P12 K ( I − P22 K ) −1 P21

73
Robust control : analysis and design» © Olivier Sename 2010

H∞ CONTROL APPROACH

based on the GENERAL CONTROL CONFIGURATION

Disturbance
and reference w ⎛P P12 ⎞ e Controlled Output
P = ⎜⎜ 11 ⎟
⎝ P21 P22 ⎟⎠
Control Input u y Measured output

P is the generalized plant (contains the plant, the weights, the uncertainties if
any) ; K is the controller. The closed-loop transfer function is:
Tew ( s ) = P11 + P12 K ( I − P22 K ) −1 P21

H∞ suboptimal control problem : Given γ a pre-specified attenuation level, a H∞


sub-optimal control problem is to design a stabilizing controller that insures :

Tew ( s ) ∞
= max σ (Tew ( jω )) ≤ γ
ω

74
Robust control : analysis and design» © Olivier Sename 2010

37
THE MIXED SENSITIVITY PROBLEM

Robust Control toolbox (MATLAB R2009)

% Generalized plant P is found with function sysic%


systemnames = 'G We Wu';
i
inputvar = '[ r(1);u(1)]';
(1) (1)]'
outputvar = '[We; Wu; r-G]';
input_to_G = '[u]';
input_to_We = '[r-G]';
input_to_Wu = '[u]';
sysoutname = 'P';
cleanupsysic = 'yes';
sysic;
%% Find H-infinity
H infinity optimal controller%
nmeas=1; nu=1;
[K,CL,GAM,INFO] = hinfsyn(P,nmeas,nu,'DISPLAY','ON');
gopt

75
Robust control : analysis and design» © Olivier Sename 2010

H∞ CONTROL APPROACH

The overall control objective is to minimize some norm of the transfer


function from w to e , for example, the H∞ norm.

H∞ control problem: Find a controller K(s) which based on the information


in y, generates a control signal u which counteracts the influence of w on e,
thereby minimizing the closed-loop norm from w to e.

H∞ suboptimal control problem : Given γ a pre-specified attenuation level,


a H∞ sub-optimal
b ti l control
t l problem
bl is
i to
t design
d i a stabilizing
t bili i controller
t ll that
th t
insures :
Tew ( s ) ∞
= max σ (Tew ( jω )) ≤ γ
ω

76
Robust control : analysis and design» © Olivier Sename 2010

38
THE MIXED SENSITIVITY PROBLEM

How to consider performance specification in H∞ control ?


In practice the performance specification concerns at least two
sensitivity functions (S and KS) in order to take into account the
tracking objective as well as the actuator constraints.

A simple example :

r(t) + ε(t) u(t) y(t)


K(s) G(s)
-

y = Gu = GK (r − y ) ⇒ tracking error : ε = Sr
u = K (r − y ) = K (r − Gu ) ⇒ actuator force : u = KSr

77
Robust control : analysis and design» © Olivier Sename 2010

THE MIXED SENSITIVITY PROBLEM

Objective w.r.t sensitivity functions: We S ∞


≤ 1, Wu KS ∞
≤1

e1 = We Sr
New controlled outputs :
e2 = Wu KSr

The performance specifications on the tracking error & on the actuator can
be given as some weights on the controlled output as follows :
e1(t)
We(s)
e2(t)
Wu((s))

r(t) + y(t)
K(s) G(s)
ε(t) u(t)
-

78
Robust control : analysis and design» © Olivier Sename 2010

39
THE MIXED SENSITIVITY PROBLEM

e1(t)
We(s)
e2(t)
Wu(s)

r(t) + y(t)
K(s) G(s)
ε(t) u(t)
-

The associated general control configuration is :


External ⎡We − We G ⎤
w=r
inputs ⎢0 Wu ⎥⎥
e=(e1, e2 )T Controlled
⎢ Outputs
Control Inputs u ⎣⎢ I − G ⎥⎦
r-y
Measured outputs
K

79
Robust control : analysis and design» © Olivier Sename 2010

THE MIXED SENSITIVITY PROBLEM

The corresponding H∞ suboptimal control problem is therefore to find a


controller K(s) such that :

We S
Tew ( s ) ∞
= ≤γ
Wu KS ∞

where Tew ( s ) = Fl ( P, K ) = P11 + P12 K ( I − P22 K ) −1 P21


⎡W ⎤ ⎡− We G ⎤
= ⎢ e⎥ + ⎢ ⎥ K ( I + GK ) −1 I
⎣ 0 ⎦ ⎣ Wu ⎦
⎡We S ⎤
=⎢ ⎥
⎣Wu KS ⎦

80
Robust control : analysis and design» © Olivier Sename 2010

40
H∞ CONTROL APPROACH

The solution of the H∞ control problem is based on a state


space representation of P, the generalized plant, that
includes the plant model and the performance weights.
⎧ x& = Ax + B1w + B2u

P ⎨e = C1 x + D11w + D12u
⎪y = C x + D w + D u
⎩ 2 21 22

The calculation of the controller, solution of the H∞


control problem , can then be done using the Riccati
approach or the LMI approach of the H∞ control
problem (see Zhou 98, Skogestad&Postlewaite 96).

81
Robust control : analysis and design» © Olivier Sename 2010

H∞ CONTROL SOLUTION

⎧ x& = Ax + B1 w + B2 u ⎡A B1 B2 ⎤

P ⎨e = C1 x + D11 w + D12 u ⇒ P = ⎢⎢ C1 D11 D12 ⎥⎥
⎪y = C x + D w + D u ⎢⎣C 2 D22 ⎥⎦
⎩ 2 21 22 D21
Assumptions:
(A1) (A,B
(A B2) stabilizable
t bili bl and
d (C2,A)
A) detectable
d t t bl
Necessary for the existence of stabilizing controllers
(A2) D12 and D21 have full rank (resp. m2 and p2)
Sufficient to ensure the controllers are proper, hence realizable
(A3) [A-j ωI, B2; C1 D12] has full column rank n+m2 for all ω
(A4) [A-j ωI, B1; C2 D21] has full row rank n+p2 for all ω
B th ensure th
Both thatt th
the optimal
ti l controller
t ll does
d nott ttry tto cancell
poles or zeros on the imaginary axis which would result in CL instability
⎡B ⎤ T ⎡ 0 ⎤
D11 = 0, D22 = 0, D12 [C1 D12 ] = [0 I m 2 ], ⎢ 1 ⎥ D21 = ⎢ ⎥
T
(A5)
⎣ D21 ⎦ ⎣ I p2 ⎦
not necessary but simplify the solution (can be relaxed)
82
Robust control : analysis and design» © Olivier Sename 2010

41
H∞ CONTROL SOLUTION

Theorem1 : Under the previous assumptions, there exists a


controller K such that ||Tew||∞<γ if and only if

((i)) ⎡⎢ A γ −2 B1 B1T − B2 B2T ⎤ has no eigenvalues on the



⎣− C1 C1 − AT ⎦ imaginary axis
T

(ii) ∃ X∞>0, X ∞ A + AT X ∞ + X ∞ (γ −2 B1 B1T − B2 B2T )X ∞ + C1T C1 = 0

(iii) ⎡ A γ −2C1T C1 − C 2T C 2 ⎤ has no eigenvalues on the


T

⎢ ⎥
⎣− B1 B1 −A ⎦ imaginary axis
T

(iv) ∃ Y∞>0, (
AY∞ + Y∞ AT + Y∞ γ −2C1T C1 − C 2T C 2 Y∞ + B1 B1T = 0 )
(v) ρ ( X ∞ Y∞ ) < γ 2

83
Robust control : analysis and design» © Olivier Sename 2010

H∞ CONTROL SOLUTION

Theorem2 : If the necessary and sufficient conditions of theorem 1 are


satisfied, then the so-called central controller is:
Aˆ = A + γ −2 B B T X + B F + Z L C
⎡ Aˆ − Z ∞ L∞ ⎤ ∞ 1 1 ∞ 2 ∞ ∞ ∞ 2
K sub ( s ) = ⎢ ∞ ⎥ F∞ = − B2T X ∞ , L∞ = −Y∞ C 2T
⎣ F∞ 0 ⎦
(
Z ∞ = I − γ − 2Y∞ X ∞ )−1

Controller structure = observer-based state feedback control law


xˆ& = Axˆ + B1 wˆ + B2 u + Z ∞ L∞ (C 2 xˆ − y )
u = F∞ xˆ , wˆ = γ − 2 B1T X ∞ xˆ

1) ŵ is an extimation of the worst case disturbance


2) Z∞ L∞ is the filter gain for the OE problem of estimating F∞ xˆ
in the presence of the worst case disturbance

84
Robust control : analysis and design» © Olivier Sename 2010

42
H∞ CONTROL APPROACH

Advantage : if We and Wu are chosen according to the previous


templates on the sensitivity functions, and if a solution does exist, the
control objectives are met.

This procedure can be easily performed by solving two Riccati equations or


two LMIs.
Using the Riccati formulation, the minimal value of γ can be approached by
g-iteration (dichotomy).
Using LMI formulation, the minimal value of γ is solved as an optimization
problem

It iis also
l available
il bl in
i classical
l i l control
t l software,
ft e.g.
MATLAB®.

This can be completed by robust analysis and/or design according to


model uncertainties.

85
Robust control : analysis and design» © Olivier Sename 2010

H∞ CONTROL APPROACH

LMI solution :
Linear Matrix Inequalities (LMIs) are powerful design tools in control
engineering, system identification and structural design.

Main advantages :
• Many design specifications and constraints can be expressed as LMIs.
• Once formulated in terms of LMIs, a problem can be solved exactly by
efficient convex optimization algorithms.
• While most problems with multiple constraints or objectives lack
analytical solutions in terms of matrix equations, they often remain
tractable in the LMI framework.

86
Robust control : analysis and design» © Olivier Sename 2010

43
H∞ CONTROL APPROACH

The Bounded Real Lemma (Scherer) :


Let M(s)=C(sI-A)-1B with A stable. Then ||M||∞<1 holds iff

AT X + XA + XBB T X + C T C = 0
has a stabilizing solution. This is equivalent to the fact that

AT X + XA + XBB T X + C T C < 0

has a solution.
This is rewritten as a Linear Matrix Inequality (Schur complement)
⎡ AT X + XA XB C T ⎤
⎢ ⎥
⎢ * −I 0 ⎥<0
⎢ * * − I ⎥⎦

87
Robust control : analysis and design» © Olivier Sename 2010

H∞ CONTROL APPROACH

Design Method:
Only assumption A1) is required
Application of the BRL on the closed-loop system (with P and K)
⎡ AclT X + XAcl XBcl C clT ⎤
⎢ ⎥
⎢ * − γI DclT ⎥<0
⎢ * * − γI ⎥⎦

where Acl, Bcl, Ccl and Dcl are the state space matrices of the closed-
loop system.
It leads to a non convex problem as it includes the state space
matrices of the controller and some matrices X to be determined
(coupling)
Then we have to use of some linearization method to get LMIs: for
instance the projection lemma. It leads to a set of LMIs.

Therefore it can be solved using convex optimization algorithms.

88
Robust control : analysis and design» © Olivier Sename 2010

44
OUTLINE

‰ Motivation Industrial examples

‰ H∞ norm, stability

‰ Performance analysis/specifications Performances quantifiers


A first robustness criteria

‰ H∞ control design Mixed sensitivity problem

‰ Uncertainties and robustness Representing uncertainties


Robust stability
stability, Robust performance
Robust control design

‰ Performances limitations Bode and Poisson sensitivity integral

89
Robust control : analysis and design» © Olivier Sename 2010

UNCERTAINTY AND ROBUSTNESS

Introduction: Skegestad & Postlewaite


A control system is robust if it is insensitive to differences between the
actual system and the model of the system which was used to design the
controller
How to take into account the difference between the actual system and
the model ?

A solution: using a model set BUT : very large problem and not
exact yet
A method: these differences are referred as model uncertainty.
Th approach
The h
‰ determine the uncertainty set: mathematical representation
‰ check Robust Stability
‰ check Robust Performance
90
Robust control : analysis and design» © Olivier Sename 2010

45
UNCERTAINTY AND ROBUSTNESS

Lots of forms can be derived according to both our knowledge of the


physical mechanism that cause the uncertainties and our ability to
represent these mechanisms in a way that facilitates convenient
manipulation.
Several origins :
‰ Approximate knowledge and variations of some parameters
‰ Measruement imperfections (due to sensor)
‰ At hign frequencies, even the structure and the model order is
unknown (100% is possible)
‰ Choice of simpler models for control synthesis
‰ Controller implementation

Two classes: parametric uncertainties / neglected or unmodelled


dynamics

91
Robust control : analysis and design» © Olivier Sename 2010

UNCERTAINTY AND ROBUSTNESS

Example (Skogestad-Postlewaite, 96):

k
G p (s) = e − sh , 2 ≤ k , h, τ ≤ 3
1 + τs
Nominal model: parameters h=k=τ=2.5

3.5s + 0.25
G p ( s) = G ( s )( I + Wm Δ ); Wm ( s ) =
s +1
Multiplicative uncertainties :

92
Robust control : analysis and design» © Olivier Sename 2010

46
UNCERTAINTY AND ROBUSTNESS

Relative uncertainties (Gp-G)/G

Wm(s)

0
10
Magnitude

Parametric
variations

-1
10
-2 -1 0 1
10 10 10 10
Frequency

93
Robust control : analysis and design» © Olivier Sename 2010

UNCERTAINTY AND ROBUSTNESS

94
Robust control : analysis and design» © Olivier Sename 2010

47
UNCERTAINTY AND ROBUSTNESS

A simple example of unmodelled dynamcis

G ( s ) = G0 ( s )
1
, τ ≤ τ max G ( s) τ max jω
Then : −1 ≤ ; ∀ω
1 + τs G0 ( s ) 1 + τ max jω
The neglected dynamics can therefore be modelled as :

G(s) τ max s
− 1 = w( s )Δ( s ) avec w(s) = et Δ <1
G0 ( s ) 1 + τ max s ∞

This corresponds to multiplicative uncertainty

yΔ uΔ
Δ( )
Δ(s) w(s)
( )
+ z
w
G0(s)
+
N

95
Robust control : analysis and design» © Olivier Sename 2010

UNCERTAINTY AND ROBUSTNESS

From the previous example, let us define the


Linear Fractional Transformation (LFT)
Which is a general representation for uncertainty analysis

Δ
Δ includes all possible

uncertainties, and is uΔ
assumed to be
External N z Controlled
normalized : Δ <1 inputs
w
ouputs

NΔ structure
The upper LFT is then the transfer matrix from w to z
Fu ( N , Δ) = N 22 + N 21Δ( I − N11 Δ) −1 N 12
This LFT exists and is well- (I--N11Δ)-1 is invertible
well-posed if (I

96
Robust control : analysis and design» © Olivier Sename 2010

48
UNCERTAINTY AND ROBUSTNESS

A simple example of parametric uncertaities


1
G ( s) =
, a0 − b < a < a0 + b
s+a
Let a=a0+δb, 1<δ<1.
b -1< <1 Then 1 1 1 δb −1
= = (1 + )
s + a s + a 0 + δb s + a 0 s + a0

This corresponds to multiplicative inverse uncertainty


uΔ yΔ
δ Δ=δ
b yΔ

- ⎛ −b 1 ⎞
z ⎜ ⎟
w + G0(s) N =⎜
s + a0 s + a0 ⎟ z
⎜ −b 1 ⎟
w ⎜s+a
1 ⎝ s + a0 ⎟⎠
z = yΔ = ( w − bu Δ )
0

s + a0
97
Robust control : analysis and design» © Olivier Sename 2010

UNCERTAINTY AND ROBUSTNESS

As seen before, in a general way, the uncertainty matrix will have the
following diagonal form

{
Δ ( s ) = diag Δ 1 ( s ) L Δ q ( s ) δ 1 I n1 L δ r I nr }
Where Δi is a weight function and
δi is a real parametric uncertainty

Such that: Δi ∞
≤ 1; δi ≤1

98
Robust control : analysis and design» © Olivier Sename 2010

49
UNCERTAINTY AND ROBUSTNESS

Now there is two ways to tackle the control problem under


uncertainties:

UNSTRUCTURED UNCERTAINTIES: we ignore the structure of Δ,


considered as a full complex perturbation matrix, such that ||Δ||∞<1.
We then look at the maximal admissible norm for Δ, to have Robust
Stability and Performance. This may lead to conservative results.

STRUCTURED UNCERTAINTIES: we take into account the structure of


Δ (always such that ||Δ||∞<1). This leads to introduce a new tool: the
structured
t t d singular
i l value.
l We
W then
th can obtain
bt i more fine
fi results
lt but
b t
using more complex tools.

99
Robust control : analysis and design» © Olivier Sename 2010

UNCERTAINTY AND ROBUSTNESS

For control design the following general control configuration is used:

Δ
uΔ yΔ

External w e
inputs P
Controlled outputs

u y Measured outputs
Control input
K
N

And N is such that N = Fl ( P, K )


100
Robust control : analysis and design» © Olivier Sename 2010

50
UNCERTAINTY AND ROBUSTNESS

Now the transfer matrix from w to z

z = Fu ( N , Δ) w,
where Fu ( N , Δ) = N 22 + N 21Δ( I − N11Δ) −1 N12

NS ⇔ N is internally stable
NP ⇔ N 22 ∞
< 1; and NS
Objectives:
RS ⇔ Fu ( N , Δ) is stable ∀Δ, Δ ∞
< 1; and NS
RP ⇔ Fu ( N , Δ) ∞
< 1 ∀Δ, Δ ∞
< 1; andd NS

101
Robust control : analysis and design» © Olivier Sename 2010

UNCERTAINTY AND ROBUSTNESS

Robust Stability analysis

Robust Stability: with a given controller K, we determine wether the system


remains stable for all plants in the uncertainty set

Δ
It is equivalent to study the uΔ
MΔ structure yΔ
M=N11

Robust Performance: If RS is satisfied we determine how « large


g » the
transfer function from exogeneous inputs w to outputs z may be for all
plants in the uncertainty set

For Robust Performance analysis it is necessary to consider the NΔ structure

102
Robust control : analysis and design» © Olivier Sename 2010

51
UNCERTAINTY AND ROBUSTNESS

ROBUST STABILITY :
SMALL GAIN THEOREM



M=N11

Small Gain Theorem: Suppose


pp M∈ RH ∞ . Then the closed-loopp system
y is
well-posed and internally stable for all Δ ∈ RH ∞ such that :
a) Δ ∞
≤δ if and only if M ( s) ∞
< 1/ δ
b) Δ ∞
<δ if and only if M ( s) ∞ ≤ 1 / δ

Robust control : analysis and design» © Olivier Sename 2010 103

UNCERTAINTY AND ROBUSTNESS


6 uncertainty types
Additive uncertainty Additive inverse uncertainty
yΔ uΔ uΔ yΔ
ΔA(s) + ΔiA(s)
z - z
u
G(s) + G( )
G(s)
+

Output Multiplicative uncertainty Input Multiplicative uncertainty


yΔ uΔ
yΔ uΔ
+
ΔO(s) ΔI(s) +
+ z
u z u G(s)
G(s) + +

Output Inverse Multiplicative case Input Inverse Multiplicative case


uΔ yΔ uΔ yΔ
ΔiO(s) - ΔiI(s)
u - z u z
G(s) + G(s)
+

104
Robust control : analysis and design» © Olivier Sename 2010

52
UNCERTAINTY AND ROBUSTNESS
Robust Stability analysis for unstructured uncertainties

Application of the small gain theorem to the different uncertainty types.

G p = G + w A Δ A ; ∀Δ A s.t . ΔA ∞
≤ 1 ; CNS : w A KS y ≤1

G p = ( I + wO Δ O )G; ∀Δ O s.t . ΔO ∞
≤ 1 ; CNS : wOT y ≤1

G p = G ( I + w I Δ I ); ∀Δ I s.t . ΔI ∞
≤ 1 ; CNS : w I Tu ∞
≤1
G p = ( I + w iO Δ iO ) −1 G; ∀Δ iO s.t . Δ iO ∞
≤ 1 ; CNS : w iO S y ≤1

G p = G ( I + w iI Δ iI ) −1 ; ∀Δ iI s.t . Δ iI ∞
≤ 1 ; CNS : w iI S u ∞
≤1

This gives some robustness templates for the sensitivity functions.


However this may be conservative.

105
Robust control : analysis and design» © Olivier Sename 2010

UNCERTAINTY AND ROBUSTNESS


Robust Stability analysis: SISO case
yΔ uΔ
Let us consider the case : ΔI(s) +
u z
+ G(s)

The loop transfer function is then: L p = G p K = GK ( I + wI Δ I ) = L + wI LΔ I ;


Therefore RS ⇔ System stable ∀ Lp. ⇔ Lp should not encircle the point -1
Im
-1
Re

⇔ wI L < 1 + L , ∀ω
1+ L RS
wI L
⇔ < 1, ∀ω
1+ L
L ( jω )
⇔ wI T < 1 ∀ω
wI L

106
Robust control : analysis and design» © Olivier Sename 2010

53
UNCERTAINTY AND ROBUSTNESS
Robust Performance analysis for unstructured uncertainties

107
Robust control : analysis and design» © Olivier Sename 2010

UNCERTAINTY AND ROBUSTNESS


Robust Performance analysis: SISO case
yΔ uΔ
ΔI(s) + L p = G p K = GK ( I + wI Δ I ) = L + wI LΔ I ;
u z
+ G(s)

First : NP ⇔ we S < 1 ∀ω , ⇔ we < 1 + L , ∀ω


RP if NP is true for all plants:.
Im
we ( jω )
-1 RP ⇔ we S p < 1, ∀S p , ∀ω
Re
⇔ we < 1 + L p , ∀L p , ∀ω
⇐ we + wI L < 1 + L , ∀ω
1+ L
L( jω ) ⇐ maxω ( we S + wI T ) < 1
wI L See previous slide (Zhou)

108
Robust control : analysis and design» © Olivier Sename 2010

54
UNCERTAINTY AND ROBUSTNESS

Often, for SISO systems, when we have NP and RS we get RP. This is not
a « big issue » for SISO systems.
For MIMO, systems, this approach may lead to very conservative results.
Further analysis will require the use of structured uncertainties
uncertainties, which
needs to consider the structured singular value. It is defined as :
Find the smallest structured Δ which makes det(I-MΔ)=0.

109
Robust control : analysis and design» © Olivier Sename 2010

UNCERTAINTY AND ROBUSTNESS

Robust Stability analysis:


structured uncertainties case

110
Robust control : analysis and design» © Olivier Sename 2010

55
6. ROBUSTNESS ANALYSIS

Fictive uncertainties: full


complex matrix representing
the H∞ norm specifications uncertainties
Δ
Δf Real
uncertainties:
Δr block diagonal
matrix

Disturbances
w G e
& references Wi Wo
Controlled outputs
p
P y Measured output
Control input u K

111
Robust control : analysis and design» © Olivier Sename 2010

6. ROBUSTNESS ANALYSIS

Δ
Δf
Δr
Uncertainties inputs v z Uncertainties outputs
⎡N N zw ⎤
N= ⎢ zv
Disturbances w ⎣ N ev N ew ⎥⎦ e Controlled outputs
& references

1. Definition of the real uncertainties Δr and of the transfer templates


e/w thanks to Wi et Wo

2. Evaluation of μ (New ) Δ f , μ (N zv ) Δ r and μ (N) Δ

3. Computation of the admissible intervals for each parameter

112
Robust control : analysis and design» © Olivier Sename 2010

56
6. ROBUSTNESS ANALYSIS

„ μ-analysis principle : Stable system for all uncertainties Δ

Δ 1
such that : Δ <

μ
M μ (M ) Δ ≤ μ
If and only if :

⎧⎪Δ = diag {Δ1 ,..., Δq , δ1I r 1 ,..., δ1I r 1 , ε1I c 1 ,..., ε1I c 1 }⎫⎪
Δ=⎨ ⎬
⎪⎩Δ i ∈ C i i , δ i ∈ R , ε i ∈ C
k ×k
⎭⎪

Robust control : analysis and design» © Olivier Sename 2010 113

6. ROBUSTNESS ANALYSIS

Δ
Δf
Δr
Uncertainties inputs v z Uncertainties outputs
⎡N N zw ⎤
Disturbances w N= ⎢ zv
⎣ N ev N ew ⎥⎦ e Controlled outputs
& references

Nominal Stability. (NS) : N stable

Nom. Performances (NP) : NS and σ (New ) = μ Δf (New ) ≤ 1, ∀ω


Robust Stability (RS): NS and μ Δr (Nzv ) ≤ 1, ∀ω

Robust Performance (RP):


⎡Δ 0⎤
NS and μ Δ (N) ≤ 1, ∀ω , Δ = ⎢ f
⎣0 Δ r ⎥⎦

114
Robust control : analysis and design» © Olivier Sename 2010

57
OUTLINE

‰ Motivation Industrial examples

‰ H∞ norm, stability

‰ Performance analysis/specifications Performances quantifiers


A first robustness criteria

‰ H∞ control design Mixed sensitivity problem

‰ Uncertainties and robustness Representing uncertainties


Robust stability
stability, Robust performance
Robust control design

‰ Performances limitations Bode and Poisson sensitivity integral

115
Robust control : analysis and design» © Olivier Sename 2010

PERFORMANCE LIMITATIONS

Main extracts of this part : Goodwin et al 2001.

“Performance limitations in control are not only inherently interesting,


but also have a major impact on real world problems.”

Objective : take into account the limitations inherent to the system


or due to actuators constraints, before designing the controller

Understanding what is not possible is as


important
p as understanding
g what is p
possible !

116
Robust control : analysis and design» © Olivier Sename 2010

58
PERFORMANCE LIMITATIONS

Mean : sensitivity functions analysis

Previous templates are partly obtained through the performance


specifications
ifi ti off the
th control
t l system.
t

They derive in fact from some trades-off including:


• sensors
• actuators
• uncertainties
• communication network (real-time scheduling)
•aarchitecture
c tectu e
• structural issues, as time-delays, RHP poles/zeroes….

117
Robust control : analysis and design» © Olivier Sename 2010

PERFORMANCE LIMITATIONS

CONSTRAINTS ON THE CONTROL SYSTEMS

Structural: S + T = 1 and
(KS = T/G)

|S|+|T|≥1 We cannot have, for any frequency w0,


|S(jw0)|<1 and |T(jw0)|<1

This implies that, disturbance and


noise rejection cannot be achieved in
the same frequency
q y range.
g

frequency splitting of specifications


• disturbances rejection (|S|<1) in LF
• noise reduction (|T|<1) in HF

118
Robust control : analysis and design» © Olivier Sename 2010

59
PERFORMANCE LIMITATIONS

Interpolation constraints

If p is an RHP pole of L(s), then:


T(p)=1 and S(p)=0

If z is an RHP zero of L(s), then:


T(z)=0 and S(z)=1

119
Robust control : analysis and design» © Olivier Sename 2010

PERFORMANCE LIMITATIONS Goodwin et al

SENSITIVITY INTEGRALS

120
Robust control : analysis and design» © Olivier Sename 2010

60
PERFORMANCE LIMITATIONS Goodwin et al

121
Robust control : analysis and design» © Olivier Sename 2010

PERFORMANCE LIMITATIONS Goodwin et al

122
Robust control : analysis and design» © Olivier Sename 2010

61
PERFORMANCE LIMITATIONS Goodwin et al

The
Waterbed
effect

123
Robust control : analysis and design» © Olivier Sename 2010

PERFORMANCE LIMITATIONS Goodwin et al

Comment: to a reduction zone corresponds an enlargment zone


• peaks on |S| and |T| (robustness ↓)

124
Robust control : analysis and design» © Olivier Sename 2010

62
PERFORMANCE LIMITATIONS

Bode sensitivity integral (the case of RHP poles)


Consider a plant with Np RHP poles pi. Then :
∞ Np

∫ ln det S ( jω ) dω = Π ∑ Re( p )
0 i =1
i

Therefore,, in the presence


p of RHP poles,
p , the control effort necessaryy
to stabilize the system is paid in terms of amplification of the
sensitivity magnitude.

125
Robust control : analysis and design» © Olivier Sename 2010

PERFORMANCE LIMITATIONS

Peak sensitivity theorems


Theorem1 : Let G(s) a MIMO plant with one RHP zero at s=z,
and Wp(s) be a scalar weight.
Then closed
Then, closed-loop
loop stability is ensured only if:

W p ( s) S ( s) ≥ W p ( z)

Theorem2 : Let G(s) a MIMO plant with one RHP pole at s=p,
and WT(s) be a scalar weight.
Then,, closed-loop
p stabilityy is ensured onlyy if:

WT ( s )T ( s ) ∞
≥ WT ( p )

126
Robust control : analysis and design» © Olivier Sename 2010

63
PERFORMANCE LIMITATIONS

Peak sensitivity theorems (SISO case)


Theorem3 : Let G(s) a SISO plant with Np RHP pole pi and
Nz RHP zero zj.
Then, closed
closed-loop
loop stability is ensured only if:

Np
z j + pi
W p (s) S ( s) ≥ c1 j W p ( z j ) , c1 j = ∏ ≥1

i =1 z j − pi
Nz z j + pi
WT ( s )T ( s ) ∞ ≥ c2i WT ( pi ) , c2i = ∏ ≥1
j =1 z j − pi

127
Robust control : analysis and design» © Olivier Sename 2010

PERFORMANCE LIMITATIONS

Peak sensitivity theorems: illustration


Let G(s) a MIMO plant with one RHP zero at s=z
s / MS + ω b
Let us choose Wp(s) as a performance specification weight WP ( s ) =
s + ω bε

For the design problem, if the controller meets the


requirements, then: W p ( s ) S ( s ) ≤ 1

z / MS + ω b
Therefore a necessary condition is : W p ( s ) S ( s ) ≥ W p ( z ) , i.e ≤1
∞ z + ω bε

If z is real, and if the performance specifications are such that : Ms=2


and ε=0, then a necessary condition is : z
ωb ≤
2
128
Robust control : analysis and design» © Olivier Sename 2010

64
PERFORMANCE LIMITATIONS

It will turn out that to avoid large frequency domain sensitivity


peaks, it is necessary to limit the range of sensitivity
reduction to be :

1. Less than any right half plane open loop zero

2. Greater than anyy right


g half plane
p open
p loop
p pole
p

129
Robust control : analysis and design» © Olivier Sename 2010

EXAMPLE using MATLAB

System definition
%Sample MATLAB program to synthesize H-infinity controller
%USE OF ROBUST CONTROL TOOLBOX FOR MATLAB R2009
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% plant
l t is
i G
N=[-1 20];
D=conv([1 0.01],[1 20]);
G=ss(tf(N,D)); % plant is G

%%%%%%%
% definition of the weighting function
%%%%%%%
Ms=2;wb=3;epsi=0 01;
Ms=2;wb=3;epsi=0.01;
Mu=3;wbc=50;epsi1=0.01;
We=ss(tf([1/Ms wb],[1 wb*epsi]));
Wu=ss(tf([1 wbc/Mu],[epsi1 wbc]));

130
Robust control : analysis and design» © Olivier Sename 2010

65
EXAMPLE using MATLAB

Problem solution
% Generalized plant P is found with function sysic%
systemnames = 'G We Wu';
inputvar
p = '[
[ r(1);
( ) d; u(1)]';
( )]
outputvar = '[We; Wu; r-G]';
input_to_G = '[u+d]';
input_to_We = '[r-G]';
input_to_Wu = '[u]';
sysoutname = 'P';
cleanupsysic = 'yes';
sysic;
%% Fi
Find
d H
H-infinity
i fi it optimal
ti l controller%
t ll %
%
nmeas=1; ncon=1;
[Khinf,CL,GAM,INFO] = hinfsyn(P,nmeas,ncon,'DISPLAY','on')

131
Robust control : analysis and design» © Olivier Sename 2010

EXAMPLE using MATLAB Plots

% Determination of the sensitivity fucntions


L=series(G,Khinf) % Loop transfer function L=GK
S=inv(1+L); % S= 1/(1+L)
poleS=pole(S)
T= feedback(L,1)
poleT=pole(T)
p p ( )
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%% Essayer avec
%Tverif = G*Khinf/(1+G*Khinf)
% et comparer l'ordre des fonctions de transfert obtenues.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
SG=S*G;;
poleSG=pole(SG)
KS=Khinf*S;
poleKS=pole(KS)
%%%%
w=logspace(-2,2,500);
subplot(2,2,1), sigma(S,1/We,w), title('Sensitivity function')
subplot(2,2,2), sigma(T,w), title('Complementary sensitivity function')
subplot(2,2,3), sigma(SG,w), title('Sensitivity*Plant')
subplot(2,2,4), sigma(KS,1/Wu,w), title('Controller*Sensitivity')

132
Robust control : analysis and design» © Olivier Sename 2010

66
EXAMPLE : First test

Resetting value of Gamma min based on D_11, D_12, D_21 terms

Test bounds: 0.5000 < gamma <= 2.6337

gamma
g hamx_eig
g xinf_eig
g hamyy_eig
g yinf
y _eig g nrho_xyy p/
p/f
2.765 2.2e+000 8.0e-007 3.0e-002 0.0e+000 0.1708 p
1.633 2.1e+000 8.6e-007 3.0e-002 0.0e+000 1.1068# f
2.199 2.1e+000 8.2e-007 3.0e-002 0.0e+000 0.3240 p
1.916 2.1e+000 8.3e-007 3.0e-002 -9.5e-013 0.5193 p
1.774 2.1e+000 8.4e-007 3.0e-002 -9.5e-013 0.7152 p
1.703 2.1e+000 8.5e-007 3.0e-002 0.0e+000 0.8719 p
1.668 2.1e+000 8.5e-007 3.0e-002 0.0e+000 0.9763 p
1.650 2.1e+000 8.6e-007 3.0e-002 0.0e+000 1.0378# f
1.659 2.1e+000 8.5e-007 3.0e-002 0.0e+000 1.0062# f

Gamma value achieved: 1.6681

133
Robust control : analysis and design» © Olivier Sename 2010

EXAMPLE

Sensitivity f unction Complementary sensitivity function


10
0
0
We
-10 -10

-20
ar Values (dB)

ar Values (dB)

-20
-30
Complementary
-40 Sensitivityy -30
Sensitivity
Singula

Singula

-50
function function
-40
S T
-60
-50
-70

-80 -60
-2 -1 0 1 2 -2 -1 0 1 2
10 10 10 10 10 10 10 10 10 10
Frequency (rad/sec) Frequency (rad/sec)

Sensitivity*Plant Controller*Sensitivity
-5 15

10
-10
5
-15
0

Wu
Singular Values (dB)

Singular Values (dB)

-20 -5

-10
-25
-15
Sensitivity Sensitivity
-30 -20
function function
-35 SG -25
KS
-30
-40
-35

-45 -40
-2 -1 0 1 2 -2 -1 0 1 2
10 10 10 10 10 10 10 10 10 10
Frequency (rad/sec) Frequency (rad/sec)

134
Robust control : analysis and design» © Olivier Sename 2010

67

Das könnte Ihnen auch gefallen