Robust Control of MIMO Systems

Robust control of MIMO systems
Olivier Sename
Grenoble INP / GIPSA-lab
September 2015
Olivier Sename (Grenoble INP / GIPSA-lab)
Robust control
September 2015
1 / 138
1. The H norm and related definitions

System definitions
H norm as a measure of the system
gain ?
LTI systems and signals norms
2. Performance analysis
Definition of the sensitivity functions
Frequency-domain analysis
Stability issues
3. Performance specifications for analysis
and/or design
Introduction
SISO case
MIMO case
Some performance limitations
4. Introduction to Linear Matrix Inequalities
Background in Optimisation
LMI in control
O. Sename [GIPSA-lab]
Some useful lemmas

5. H control design
About the control structure
Solving the H control problem
The Riccati approach
The LMI approach for H control design
Example
6. Uncertainty modelling and robustness
analysis
Introduction
Representation of uncertainties
Definition of Robustness analysis
Robustness analysis: the unstructured
case
Robustness analysis: the structured case
Towards Robust control design
7. Introduction to LPV systems and control
LPV modelling
LPV Control
2/138
Reference books
To be studied during the course
S. Skogestad and I. Postlethwaite, Multivariable Feedback Control: analysis and design, John
Wiley and Sons, 2005.

www.nt.ntnu.no/users/skoge/book, chap 1 to 3 available
K. Zhou, Essentials of Robust Control, Prentice Hall, New Jersey, 1998.
www.ece.lsu.edu/kemin, book slides available

J.C. Doyle, B.A. Francis, and A.R. Tannenbaum, Feedback control theory, Macmillan
Publishing Company, New York, 1992.

https://sites.google.com/site/brucefranciscontact/Home/publications, ,
book available
Carsten Scherers courses
http://www.dcsc.tudelft.nl/ cscherer/. , Lecture slides available (MSc Course

"Robust Control", MSc Course "Linear Matrix Inequalities in Control")
+ all the MATLAB demo, examples and documentation on the Robust Control toolbox
(mathworks.com/products/robust)
Other references (some in french)
G.C. Goodwin, S.F. Graebe, and M.E. Salgado, Control System Design, Prentice Hall, New Jersey, 2001.
csd.newcastle.edu.au
G. Duc et S. Font, Commande Hinf et -analyse: des outils pour la robustesse, Herms, France, 1999.
D. Alazard, C. Cumer, P. Apkarian, M. Gauvrit, et G. Ferreres, Robustesse et commande optimale,
Cpadues Editions, 1999.
3/138
Robust control in 1 slide ?

y =
1
1+G(s)K(s)
(GKr)
1
Sensitivity function S(s) =
1+L(s)
Complementary Sensitivity function
T (s) =
Why S is the key function in control ?

S allows to characterize many things:
L(s)
1+L(s)
Example: an uncertain mass-spring-damper system

controlled by a proportional gain.
S = 1 T S = ry .
r
For performance analysis: (S( = 0) =

steady-state error, bandwidth )
if output disturbance dy then, y = S.

dy
S to be minimized !
distance from -1 to Nyquist plot =
inf | 1 L(j) |=

1
1
sup | 1+L(j)
|
Robustness w.r.t model uncertainties
Robust control: Find K s.t S satisfies all

requirements
Robust stability condition : | S(j) |<|

4/138
1
W (j)
|,
1 The H norm and related definitions
System definitions
Nonlinear systems
In control theory, dynamical systems are mostly modeled and analyzed thanks to the use of a set
of Ordinary Differential Equations (ODE)1 .
Nonlinear dynamical system modeling is the "most" representative model of a given system.
Generally this model is derived thanks to system knowledge, physical equations etc. Nonlinear
dynamical systems are described by nonlinear ODEs.
Definition (Nonlinear dynamical system)
For given functions f : Rn Rnw 7 Rn and g : Rn Rnw 7 Rnz , a nonlinear dynamical
system (N L ) can be described as:

x = f (x(t), w(t))
N L :
z = g(x(t), w(t))
(1)
where x(t) is the state which takes values in a state space X Rn , w(t) is the input taking values
in the input space W Rnw and z(t) is the output that belongs to the output space Z Rnz .
1
Partial Differential Equations (PDEs) may be used for some applications such as irrigation channels, traffic flow, etc. but
methodologies involved are more complex compared to the ones for ODEs.
5/138
System definitions
Comments
The main advantage of nonlinear dynamical modeling is that (if it is correctly described) it
catches most of the real system phenomena.

On the other side, the main drawback is that there is a lack of mathematical and
methodological tools; e.g. parameter identification, control and observation synthesis and
analysis are complex and non systematic (especially for MIMO systems).
In this field, notions of robustness, observability, controllability, closed loop performance etc.
are not so obvious. In particular, complex nonlinear problems often need to be reduced in
order to become tractable for nonlinear theory, or to apply input-output linearization
approaches (e.g. in robot control applications).
As nonlinear modeling seems to lead to complex problems, especially for MIMO systems
control, the LTI dynamical modeling is often adopted for control and observation purposes.
The LTI dynamical modeling consists in describing the system through linear ODEs.
According to the previous nonlinear dynamical system definition, LTI modeling leads to a local
description of the nonlinear behavior (e.g. it locally describes, around a linearizing point, the
real system behavior).
From now on only Linear Time-Invariant (LTI) systems are considered.
6/138
System definitions
Definition of LTI systems

Definition (LTI dynamical system)
Given matrices A Rnn , B Rnnw , C Rnz n and D Rnz nw , a Linear Time Invariant
(LTI) dynamical system (LT I ) can be described as:

x(t)
= Ax(t) + Bw(t)
(2)
LT I :
z(t) = Cx(t) + Dw(t)
where x(t) is the state which takes values in a state space X Rn , w(t) is the input taking values
in the input space W Rnw and z(t) is the output that belongs to the output space Z Rnz .
The LTI system locally describes the real system under consideration and the linearization
procedure allows to treat a linear problem instead of a nonlinear one. For this class of problem,
many mathematical and control theory tools can be applied like closed loop stability, controllability,
observability, performance, robust analysis, etc. for both SISO and MIMO systems. However, the
main restriction is that LTI models only describe the system locally, then, compared to nonlinear
models, they lack of information and, as a consequence, are incomplete and may not provide
global stabilization.
7/138
Towards H norm
Illustration of the gain computation

For a SISO system, y=Gd, the gain at a given frequency is simply
|G(j)d()|
|y()|
=
= |G(j)|
|d()|
|d()|
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:
ky()k2
kG(j)d()k2
=
= kG(j)k2
kd()k2
kd()k2
Which is independent of the input magnitude. But this is not a correct definition. Indeed the
input direction is of great importance
8/138
Towards H norm
A first bad approach

Let consider

G=
5
3
4
2
How to define and evaluate its gain ?

Consider
inputs:
fivedifferent

1
0
0.707
0.707
0.6
u=
u=
u=
u=
u=
0
1
0.707
0.707
0.8
Input magnitude :
ku1 k2 = ku2 k2 = ku3 k2 = ku4 k2 = ku5 k2 = 1
But thecorresponding

outputs
are

5
4
6.3630
y=
y=
y=
3
2
3.5350
the ratio are kyk2 /kuk2
5.8310
4.4721
7.2790
0.9998
0.2828

y=
0.7070
0.7070

y=
0.2
0.2
9/138

and
Towards H norm
Towards SVD
MAXIMUM SINGULAR VALUE
max
d6=0
kGdk2
=
(G)
kdk2
MINIMUM SINGULAR VALUE

min
d6=0
kGdk2
= (G)
kdk2
We see that, depending on the ratio

d20/d10, the gain varies between 0.27
and 7.34 .
10/138
Towards H norm
What about eigenvalues ?

Eigenvalues are a poor measure of gain. Let

0
0
0
1
G=
100
0
Eigenvalues are 0 and 0.

But an input vector

leads to an output vector

100
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.
11/138
Towards H norm
For MIMO systems

Finally, in the case of a transfer matrix G(s) : (m inputs, p outputs) u vector of inputs, y vector of
outputs.
(G(j))
ky()k2

(G(j))
ku()k2
Example of A two-mass/spring/damper system

2 inputs: F1 and F2 2 outputs: x1 and x2
Singular Values
40
largest singular
value
Hinf norm 11.4664 = 21.18 dB
20
Singular Values (dB)
-20
-40
smallest singular
value
-60
-80
-100
-1
10
10
10
10
Frequency (rad/sec)
G=ss(A,B,C,D): LTI system

normhinf(G): Compute Hinf norm
norm(G,inf): Compute Hinf norm
sigma(G): plot max and min SV
12/138
Signal norms
Reader is also invited to refer to the famous book of Zhou et al., 1996, where all the following
definitions and additional information are given.
All the following definitions are given assuming signals x(t) C, then they will involve the
conjugate (denoted as x (t)). When signals are real (i.e. x(t) R), x (t) = xT (t).
Definition (Norm and Normed vector space)
Let V be a finite dimension space. Then p 1, the application ||.||p is a norm, defined as,
||v||p =
|vi |p
1/p
(3)
Let V be a vector space over C (or R) and let k.k be a norm defined on V . Then V is a
normed space.
13/138
L? norms
Definition (L1 , L2 , L norms)
The 1-Norm of a function x(t) is given by,
+
Z
kx(t)k1 =
|x(t)|dt
(4)
The 2-Norm (that introduces the energy norm) is given by,
kx(t)k2
=
=
qR
q
+
x (t)x(t)dt
0
R +
1
X (j)X(j)d
2
(5)
The second equality is obtained by using the Parseval identity.

The -Norm is given by,
kx(t)k = sup |x(t)|
(6)
kXk =
sup
Re(s)0
kX(s)k = sup kX(j)k
(7)
if the signals that admit the Laplace transform, analytic in Re(s) 0 (i.e. H ).
14/138
About vector spaces

Definition (Banach, Hilbert, Hardy and Lp spaces)
A Banach space is a (real or complex) complete (i.e. all Cauchy sequences, of points in K
have a limit that is also in K) normed vector space B (with norm k.kp ).
A Hilbert space is a (real or complex) vector space H with an inner product < ., . > that is
complete under the norm defined by the inner product. The norm of f H is then defined by,
p
kf k = < f, f >
(8)
Every Hilbert space is a Banach since a Hilbert space is complete with respect to the norm
associated with its inner product.
The Hardy spaces (Hp ) are certain spaces of holomorphic functions (functions defined on an
open subset of the complex number plane C with values in C that are complex-differentiable
at every point) on the unit disk or upper half plane.
The Lp space are spaces of p-power integrable functions (function whose integral exists,
generally called Lebesgue integral), and corresponding sequence spaces.

For example, Rn and Cn with the usual spatial p-norm, k.kp for 1 p < , are Banach spaces.
This means that a Banach space is a vector space B over the real or complex numbers with a
norm k.kp such that every Cauchy sequence (with respect to the metric d(x, y) = ||x y||) in B
has a limit in B.
15/138
L and H spaces
Definition (L space)
L is the space of piecewise continuous bounded functions. It is a Banach space of
matrix-valued (or scalar-valued) functions on C and consists of all complex bounded matrix
functions f (j), R, such that,
sup [f (j)] <
(9)
Definition (H and RH spaces)

H is a (closed) subspace in L with matrix functions f (j), R, analytic in Re(s) > 0
(open right-half plane). The real rational subspace of H which consists of all proper and real
rational stable transfer matrices, is denoted by RH .
Example
In control theory
s+1
(s+10)(s+6)
s+1
(s10)(s+6)
s+1
(s+10)
RH
6 RH
RH
(10)
16/138
H norm
Definition (H norm)
The H norm of a proper LTI system defined as on (2) from input w(t) to output z(t) and which
belongs to RH , is the induced energy-to-energy gain (L2 to L2 norm) defined as,
kG(j)k
=
=
supR (G(j))
kz(s)k
supw(s)H2 kw(s)k2
maxw(t)L2
kzk2
kwk2
(11)
In MATLAB, use norm(sys,inf)

Remark
H physical interpretations
This norm represents the maximal gain of the frequency response of the system. It is also
called the worst case attenuation level in the sense that it measures the maximum
amplification that the system can deliver on the whole frequency set.
For SISO (resp. MIMO) systems, it represents the maximal peak value on the Bode
magnitude (resp. singular value) plot of G(j), in other words, it is the largest gain if the
system is fed by harmonic input signal.
Unlike H2 , the H norm cannot be computed analytically. Only numerical solutions can be
obtained (e.g. Bisection algorithm, or LMI resolution).

17/138
Recalls on Singular Value Definition

Let A Rmn , There exists unitary matrices:
U = [u1 , u2 , . . . , um ] and V = [v1 , v2 , . . . , vn ]
such that
A = U V T , =
1
0
0
0
, 1 =
1
0
..
.
0
0
2
..
.
0
...
...
...
0
0
...
p
with 1 2 . . . p 0 , p = min{m, n}.

Singular values are good measures of the size of a matrix Singular vectors are good indications of
str, ong/weak input or output directions. Note that:
Avi = i ui and AT ui = i vi . Then
AAT ui = i2 ui
= max (A) = 1 = the largest singular value of A.
and
= min (A) = p = the smallest singular value of A.
18/138
L2 and H2 spaces
Definition (L2 space)
L2 is the space of piecewise continuous square integrable functions. It is a Hilbert space of
matrix-valued (or scalar-valued) functions on C and consists of all complex matrix functions f (j),
R, such that,
s
Z +
1
Tr[f (j)f (j)]d <
(12)
||f ||2 =
2
Definition (H2 and RH2 spaces)
H2 is a subspace (Hardy space) of L2 with matrix functions f (j), R, analytic in Re(s) > 0
(functions that are locally given by a convergent power series and differentiable on each point of its
definition set). In particular, the real rational subspace of H2 , which consists of all strictly proper
and real rational stable transfer matrices, is denoted by RH2 .
Example
In control theory
s+1
(s+10)(s+6)
s+1
(s10)(s+6)
s+1
(s+10)
RH2
6 RH2
6 RH2
(13)
19/138
H2 norm
Definition (H2 norm)
The H2 norm of a strictly proper LTI system defined as on (2) from input w(t) to output z(t) and
which belongs to RH2 , is the energy (L2 norm) of the impulse response g(t) defined as,
qR
+
kG(j)k2 =
g (t)g(t)dt
q
R +
1
(14)
=
Tr[G (j)G(j)]d
2
supw(s)H2
||z(s)||
||u(s)||2
The norm H2 is finite if and only if G(s) is strictly proper (i.e. G(s) RH2 ).
In MATLAB, use norm(sys,2)
Remark
H2 physical interpretations and remarks
For SISO systems, it represents the area located below the so called Bode diagram.
For MIMO systems, the H2 norm is the impulse-to-energy gain of z(t) in response to a white
noise input w(t) (satisfying W (j)W (j) = I, i.e. uniform spectral density).
The H2 norm can be computed analytically (through the use of the controllability and
observability Grammians) or numerically (through LMIs).

20/138
2 Performance analysis
Outline
System definitions
H norm as a measure of the system gain ?
Stability issues
3. Performance specifications for analysis and/or design
Introduction
SISO case
MIMO case
LMI in control
Some useful lemmas
5. H control design
Example
6. Uncertainty modelling and robustness analysis
Introduction
Robustness analysis: the unstructured case
LPV modelling
LPV Control
21/138
Introduction
Objectives of any control system [Skogestad and Postlethwaite(1996)]
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.
This is formulated as:
Nominal stability (NS): The system is stable with the nominal model (no model uncertainty)
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 the real plant)
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).
22/138
The control structure
Figure: Complete control scheme
In the SISO case, it leads to:

y =
u =
1
1+G(s)K(s)
1
1+K(s)G(s)
(GKr + dy GKn + Gdi )

(Kr Kdy Kn KGdi )
Loop transfer function

Sensitivity function
L = G(s)K(s)
1
S(s) = 1+L(s)
Complementary Sensitivity function
T (s) =
L(s)
1+L(s)
N.B. S is often referred to as the Output Sensitivity.

23/138
Stability and robustness margins ...

Classical definitions:
Stability | L(j ) |< 1, where
is the phase crossover frequency

defined by (L(j )) = .
Gain Margin: indicates the additional
gain that would take the closed loop

to the critical stability condition
(GM (dB) = [|L(j |]dB )
Phase margin: quantifies the pure
phase delay that should be added to

achieve the same critical stability
condition
(M = 180o + arg[L(jc )], where
|L(jc |) = 1(0dB))
Delay margin: quantifies the
maximal delay that should be added

in the loop to achieve the same
critical stability condition, P M/c
24/138
Robustness margins ...

The MODULE MARGIN is a robustness
It is important to consider the module

margin that quantifies the minimal
distance between the curve and the
critical point (-1,0j): this is a robustness
margin.
margin. Indeed, the influence of plant

modelling errors on the CL transfer function:
T =
K(s)G(s)
1 + K(s)G(s)
is given by:
1
G
G
T
=
= S.
T
1 + K(s)G(s) G
G
A good module margin implies good gain and
phase margins:
GM
1
MS
andP M
MS 1
MS
For MS = 2, then GM > 2 and P M > 30

Last:
M = 1/M
S
MS = max |S(j)| = kSk
MT = max |T (j)|
A good value : MT < 1.5(3.5dB)
Good value MS < 2 (6dB)

25/138
Input/Output performances
Defining two new sensitivity functions:
Plant Sensitivity: SG = S(s).G(s) (often referred to as the Input Sensitivity, e.g in Matlab)
Controller Sensitivity: KS = K(s).S(s)
26/138
Performance analysis: answer to ....

Analysis of S:
Steady state error in tracking and output disturbance rejection : S( = 0) = 0 ?
Maximum peak criterion (Module Margin): kSk < 2 ?
Analysis of T :
Steady state error in tracking : T ( = 0) = 1 ?.
Attenuation of measurement noise: |T (j)| small when ?
Maximum of T , kT k < 1.5 ?
Analysis of KS:
Input saturation: |u(t)| < |umax | ? (where |umax | < kKSk |rmax |).
Attenuation of measurement noise: |KS(j)| small when ?
Analysis of SG:
Steady state error in input disturbance rejection : SG( = 0) = 0 ?
Attenuation of input disturbance effet: |SG(j)| small in the frequency range of interest ? .
Plus: all the transient behaviors (bandwidths), as seen later.
27/138
Definition of the sensitivity functions- The MIMO case
Figure: Complete control scheme
The output & the control input satisfy the following equations :
(Ip + G(s)K(s))y(s) = (GKr + dy GKn + Gdi )
(Im + K(s)G(s))u(s) = (Kr Kdy Kn KGdi )
BUT : K(s)G(s) 6= G(s)K(s) !!
28/138
Using Matlab
% 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)
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
SG=S*G;
poleSG=pole(SG)
KS=Khinf*S;
poleKS=pole(KS)
%%%%
w=logspace(-3,3,500); %% to be adjusted
subplot(2,2,1), sigma(S,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,w), title('Controller*Sensitivity')
29/138
Definition of the sensitivity functions- The MIMO case

Definitions
Output and Output complementary sensitivity functions:
Sy = (Ip + GK)1 ,
Ty = (Ip + GK)1 GK,
Sy + Ty = Ip
Input and Input complementary sensitivity functions:

Su = (Im + KG)1 ,
Tu = KG(Im + KG)1 , Su + Tu = Im
Properties
Ty = GK(Ip + GK)1
Tu = (Im + KG)1 KG
Su K = KSy
30/138
MIMO Input/Output performances

Defining two new sensitivity functions:
Plant Sensitivity: Sy G = Sy (s).G(s)
Controller Sensitivity: KSy = K(s).Sy (s)
31/138
Frequency-domain analysis- Some classical analysis criteria (1)

The transfer function KSy (s) should be
upper bounded so that u(t) does not

reach the physical constraints, even for
a large reference r(t)
The effect of the measurement noise
n(t) on the plant input u(t) can be

made small by making the
sensitivity function KSu (s) small (in
High Frequencies)
The effect of the input disturbance di (t)
on the plant input u(t) + di (t) (actuator)

can be made small by making the
sensitivity function Su (s) small (take
care to not trying to minimize Tu which
is not possible)
32/138
Frequency-domain analysis- Some classical analysis criteria (2)

The plant output y(t) can track the
reference r(t) by making the

complementary sensitivity function
Ty (s) equal to 1. (servo pb)
The effect of the output disturbance
dy (t) (resp. input disturbance di (t) ) on

the plant output y(t) can be made
small by making the sensitivity
function Sy (s) (resp. Sy G(s) ) small
The effect of the measurement noise
n(t) on the plant output y(t) can be

made small by making the
complementary sensitivity function
Ty (s) small
33/138
Trade-offs
But
S? + T? = I
Some trade-offs are to be looked for...
These trade-offs can be reached if one aims :
to reject the disturbance effects in low frequencies
to minimize the noise effects in high frequencies
It remains to require:
Sy and Sy G to be small in low frequencies to reduce the load (output and input) disturbance
effects on the controlled output

Ty and KSy to be small in high frequencies to reduce the effects of measurement noises on
the controlled output and on the control input (actuator efforts)
34/138
Frequency-domain analysis- Dynamical behavior

As mentioned in [Skogestad and Postlethwaite(1996)]:
The concept of bandwidth is very important in understanding the benefits and trade-offs involved
when applying feedback control. Above we considered peaks of closed-loop transfer functions,
which are related to the quality of the 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.
Definition
Loosely speaking, bandwidth may be defined as the frequency range [1 , 2 ] over which control is
effective. In most cases we require tight control at steady-state so 1 = 0, and we then simply call
2 the bandwidth. The word "effective" may be interpreted in different ways : globally it means
benefit in terms of performance.
35/138
Frequency-domain analysis- Bandwidth definitions

Definition (S )
The (closed-loop) bandwidth, S , is the frequency where |S(j)| crosses 3dB (1/sqrt2) from
below.
Remark: |S| < 0.707, frequency zone, where e/r = S is reasonably small
Definition (T )
The bandwidth (in term of T ), T , is the frequency where |T (j)| crosses 3dB (1/sqrt2) from
above.
Definition (c )
The bandwidth (crossover frequency), c , is the frequency where |L(j)| crosses 1 (0dB), for the
first time, from above.
36/138
Some remarks
Remark
Usually S < c < T
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 S , |S| is less than 0.7, and
control is effective in terms of improving performance. In the frequency range [S , T ] control still
affects the response, but does not improve performance. Finally, at frequencies higher than T ,
we have S ' 1 and control has no significant effect on the response.
Remark
Usually S < c < T
Finally the following relation is very useful to evaluate the rise time:
tr =
2.3
T
37/138
Examples
38/138
Stability issues
Well-posedness
Consider
G=
s1
,
s+2
Therefore the control input is non proper:

u=
s+2
s1
(r n dy ) +
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 + 1 (1) = 0 Note that if G is strictly proper, this always holds.
39/138
K=1
Stability issues
Internal stability
DEF: A system is internally stable if all the transfer functions of the closed-loop system are stable
Figure: Control scheme
y
u

=
(I + GK)1 GK
K(I + GK)1
(I + GK)1 G
K(I + GK)1 G

r
di
For instance :
1
G=
,
s1
s1
K=
,
s+1
y
u

=
1
s+2
s1
s+2
s+1
(s1)(s+2)
1
s+2
!
r
di
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!!).
40/138
Stability issues
Small Gain theorem

Consider the so called M loop.
Figure: M form
Theorem
Suppose M (s) in RH and a positive scalar. Then the system is well-posed and internally
stable for all (s) in RH such that kk 1/ if and only if
kM k <
41/138
Stability issues
Input-Output Stability
Definition (BIBO stability)
A system G (x = Ax + Bu; y = Cx) is BIBO stable if a bounded input u(.) (kuk < ) maps a
bounded output y(.) (kyk < ).
Now, the quantification (for BIBO stable systems) of the signal amplification (gain) is evaluated as:
peak =
sup
0<kuk <
kyk
kuk
and is referred to as the PEAK TO PEAK Gain.

Definition (L2 stability)
A system G (x = Ax + Bu; y = Cx) is L2 stable if kuk2 < implies kyk2 < .
Now, the quantification of the signal amplification (gain) is evaluated as:
energy =
sup
0<kuk2 <
kyk2
kuk2
and is referred to as the ENERGY Gain, and is such that:

energy = sup kG(j)k := kGk
For a linear system, these stability definitions are equivalent (but not the quantification criteria).
42/138
Stability issues
Synthesis
Provide a clear and detailed frequency-domain performance analysis using the sensitivity
functions in order to explain the trade-off performance/robustness/actuator constraints.
Qualitative analysis
Use of Sy , Ty , KSy , Sy G to:
predict the behavior of the output w.r.t different external inputs (reference, disturbance, noise)
predict the behavior of the control input w.r.t different external inputs (reference, disturbance,
noise)
Quantitative analysis
Use of Sy , Ty , KSy , Sy G to:
Stability analysis and margins.
Compute the steady-state errors in tracking, output and input disturbance attenuations.
Give the maximum of the input/output gains to analyze the transient behaviors of the output
and control input (incl. saturation).

Give all the bandwidths of the sensitivity functions
Evaluate the rise time in tracking
Evaluate the robustness margins
43/138
3 Performance specifications
Outline
System definitions
Stability issues
Introduction
SISO case
MIMO case
LMI in control
Some useful lemmas
5. H control design
Example
Introduction
LPV modelling
LPV Control
44/138
Introduction
Framework... templates
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.
Sketch of the method:
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
Mathematically, these specifications may be captured by an upper bound, on the magnitude of a

sensitivity function, given by another transfer function, as for S:
|S(j)|
1
,
|We (j)|
kWe Sk 1
where We (s) is a WEIGHT selected by the designer.
45/138
SISO case
Template on the sensitivity function - Weighted sensitivity

Typical specifications in terms of S include:
1
Minimum bandwidth frequency S
Maximum tracking error at selected frequencies.
System type, or the maximum steady-state tracking error 0
Shape of S over selected frequency ranges.
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.
46/138
SISO case
Template on the sensitivity function S
S(s) =
1
1 + K(s)G(s)
s + b
1
=
We (s)
s/MS + b
Generally ' 0 is considered, MS < 2
(6dB) or (3dB - cautious) to ensure
sufficient module margin.
b influences the CL bandwidth : b
faster rejection of the disturbance
10
0
MS = 2 (6dB)
10
better robustness w.r.t. parametric
20
faster CL tracking response
uncertainties
Template on the sensitivity function S

Template on the Sensitivity function S
b s.t. |1/We | = 0dB
30
40
50
60 4
10
= 1e 3
10
10 Frequency
10(rad/sec) 10
0
47/138
10
SISO case
Template on the function KS
KS(s) =
K(s)
1 + K(s)G(s)
1 s + bc
1
=
Wu (s)
s + bc /Mu
Mu chosen according to LF behavior of
the process (actuator constraints:
saturations)
bc influences the CL bandwidth : bc
Template on the Controller*Sensitivity KS

10
0
noises
roll-off starting from bc to reduce
modeling errors effects
better limitation of measurement
= 2
10
20
30
bc f o r |1 / W u | = 0 d B
c = 1 e-3
40
50
60
10
Frequency (rad/sec)
48/138
10
SISO case
Template on the function T
T (s) =
K(s)G(s)
1 + K(s)G(s)
T s + bt
1
=
WT (s)
s + bt /MT
Generally T ' 0 is considered,
MT < 1.5 (3dB) to limit the overshoot.
bt influences the bandwidth hence the
transient behavior of the disturbance
rejection properties: bt
10
Template on the Complementary sensitivity function T
0
10
better filtering of HF modelling errors
20
better noise effects rejection
MT =1.5
bT s.t. |1/WT | = 0 dB
30
40
T =1e-3
50
60 4
10
10
10 Frequency
10(rad/sec) 10
0
49/138
10
SISO case
Template on the function SG

G(s)
1 + K(s)G(s)
s + SG SG
1
=
WSG (s)
s/MSG + SG
MSG allows to limit the overshoot in the
response to input disturbances.
Generally SG ' 0 is considered,
SG influences the CL bandwidth :
SG = faster rejection of the
disturbance.
20
Template on the Sensitivity*Plant SG
10
SG(s) =
MSG = 10
10
20
sg s.t. |1/WSG | = 0 dB
SG =1e-2
30
40 4
10
10
10 Frequency
10(rad/sec) 10
0
50/138
10
MIMO case
A first insight into the MIMO case

The direct extension of the performances objectives to MIMO systems could be formulated as
follows:
1
Disturbance attenuation/closed-loop performances:
(Sy ) <
1
|W1 (j)|
with |W1 (j)| > 1 for < b

2
Actuator constraints:
(KSy ) <
1
|W2 (j)|
with |W2 (j)| > 1 for > h

3
Robustness to multiplicative uncertainties:
(Ty ) <
1
|W3 (j)|
with |W3 (j)| > 1 for > t

However these objectives do not consider the specific MIMO structure of the system, i.e. the
input-output relationship between actuators and sensors.
It is then better to define the objectives accordingly with the system inputs and outputs.
51/138
MIMO case
Towards MIMO systems

Let us consider a system with 2 inputs and 1 output and define:

K1
G = G1 G2 , K =
K2
Therefore

GK = G1 K1 + G2 K2 ,
KG =
K1 G1
K2 G1
K 1 G2
K2 G2
and the sensitivity functions are:

Sy =
1
, KSy =
1 + G1 K1 + G2 K2
K1
1+G1 K1 +G2 K2
K2
1+G1 K1 +G2 K2
While a single template We is convenient for Sy it is straightforward that the following diagonal
template should be used for KSy :

Wu1 (s)
0
Wu (s) =
2
0
Wu (s)
where Wu1 and Wu2 are chosen in order to account for each actuator specificity (constraint).
52/138
MIMO case
The MIMO general case

Let us consider G with m inputs and p outputs.
In the MIMO case the simplest way is to defined the templates as diagonal transfer matrices,
i.e. using (MSi , bi , i )

In that case, a weighting function should be dedicated for each input, and for each output.
These weighting functions may of course be different if the specifications on each actuator
(e.g. saturation), and on each sensor (e.g. noise), are different.

In addition, during the performance analysis step, take care to plot, in addition to the MIMO
sensitivity functions, the individual ones related to each input/output to check if the individual
specification is met. Hence, in the simplest case:
1
If the specifications are identical then it is sufficient to plot:

(Sy (j)) and |W 1(j)| , for all
e

(KSy (j)) and |W 1(j)| , for all
u
If the specifications are different, one should plot

1
i | , for
|We
and 1k ,
|Wu |

(Sy (i, :)) and
all , i = 1, . . . , p

(KSy (k, :))
for all , k = 1, . . . , m.
i.e. p plots for all output behaviors and m plots for the input ones.
3
In a very general case, plot

(Sy ) with
(1/We )
53/138
MIMO case
More on weighting functions

When tighter (harder) objectives are to be met .....
the templates can be defined more accurately by transfer functions of order greater than 1, as

We (s) =
s/MS + b
s + b
k
,
if a roll-off of 20 k dB per decade is required.

Take care to the choice of the parameters (MS , b , ) to avoid incoherent objectives!
$1/W_u$ and $1/Wu_^2$ with parameter modification

20
Original weight (specications) Mu =2, 1 = 0.01, b c=100 for Wu
Wu and Wu square without parameter modification

10
20
Mu = 2, 1 = 0.01,
b c=200 for 1/Wu2 case 2
Mu=2, wbc=100, epsi1=0.01
10
0
0
10
20
30
10
Mu = 2, 1 = 0.01,
20
40
Mu = 2, 1 = 0.01,
30
50
60
40
70
80
O. Sename
2 [GIPSA-lab]
3
50
54/138
MIMO case
Final objectives
In terms of control synthesis, all these specifications can be tackled in the following problem: find
K(s) s.t.

We S

Wu KS

WT T

WSG SG
kWe Sk 1
kWu KSk 1
kWSG SGk 1
kWT T k 1
Often, the simpler following one (referred to as the mixed sensitivity problem) is studied:

Find K s.t.

We S
1
Wu KS
since the latter allows to consider the closed-loop output performance as well as the actuator
constraints.
55/138
Introduction
Framework
Main extracts of this part: see 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 as
understanding what is possible!
Example of structural constraints
S + T = 1,
We then cannot have, for any frequency 0 , |S(j0 )| < 1 and |T (j0 )| < 1. This implies that,
disturbance and noise rejection cannot be achieved in the same frequency range.
Bodes Sensitivity Integral for open-loop stable systems
It is known that, for open loop stable plant:
Z
log|S(j)|d = 0
0
Then the frequency range where |S(j)| < 1 is balanced by the frenquencies where |S(j)| > 1
56/138
Bode sensitivity
Nice interpretation of the balance between reduction and magnification of the sensitivity.
For unstable system we have stronger constraints:
log|det(S(j))|d =
0
Np
X
Re(pi ),
i=1
where pi design the Np RHP poles. Therefore, in the presence of RHP poles, the control effort
necessary to stabilize the system is paid in terms of amplification of the sensitivity magnitude.
57/138
The interesting case of systems with RHP zeros

Theorem
Let G(s) a MIMO plant with one RHP zero at s = z, and Wp (s) be a scalar weight. Then,
closed-loop stability is ensured only if:
k Wp (s)S(s) k |Wp (s = z)|
To illustrate the use of that theorem, if Wp is chosen as:

Wp (s) =

s/MS + b
,
s + b
and , if the controller meets the requirements, then

k Wp (s)S(s) k 1
Therefore a necessary condition is:
|
z/MS + b
| 1
z + b
To conclude, if z is real, and if the performance specifications are such that : MS = 2 and = 0,
then a necessary condition to meet the performance requirements is :
z
b
2
58/138
4 Introduction to LMIs
Outline
System definitions
Stability issues
Introduction
SISO case
MIMO case
LMI in control
Some useful lemmas
5. H control design
Example
Introduction
LPV modelling
LPV Control
59/138
Brief on optimisation
Definition (Convex function)
A function f : Rm R is convex if and only if for all x, y Rm and [0 1],
f (x + (1 )y) f (x) + (1 )f (y)
(15)
Equivalently, f is convex if and only if its epigraph,

epi(f ) = {(x, )|f (x) }
(16)
is convex.
Definition ((Strict) LMI constraint)
A Linear Matrix Inequality constraint on a vector x Rm is defined as,
F (x) = F0 +
m
X
Fi xi 0( 0)
(17)
i=1
where F0 = F0T and Fi = FiT Rnn are given, and symbol F 0( 0) means that F is
symmetric and positive semi-definite ( 0) or positive definite ( 0), i.e. {u|uT F u(>) 0}.
60/138
Convex to LMIs
Example
Lyapunov equation. A very famous LMI constraint is the Lyapunov inequality of an autonomous
system x = Ax. Then the stability LMI associated is given by,
xT P x
xT (AT P + P A)x
>
<
0
0
(18)
which is equivalent to,

F (P ) =
P
0
0
AT P + P A

0
(19)
where P = P T is the decision variable. Then, the inequality F (P ) 0 is linear in P .

LMI constraints F (x) 0 are convex in x, i.e. the set {x|F (x) 0} is convex. Then LMI based
optimization falls in the convex optimization. This property is fundamental because it guarantees
that the global (or optimal) solution x of the the minimization problem under LMI constraints can
be found efficiently, in a polynomial time (by optimization algorithms like e.g. Ellipsoid, Interior
Point methods).
61/138
LMI problem
Two kind of problems can be handled
Feasibility: The question whether or not there exist elements x X such that F (x) < 0 is
called a feasibility problem. The LMI F (x) < 0 is called feasible if such x exists,
otherwise it is said to be infeasible.
Optimization: Let an objective function f : S R where S = {x|F (x) < 0}. The problem to
determine
Vopt = infxS f (x)
is called an optimization problem with an LMI constraint. This problem involves
the determination of Vopt , the calculation of an almost optimal solution x (i.e., for
arbitrary > 0 the calculation of an x S such that Vopt f (x) Vopt + , or
the calculation of a optimal solutions xopt (elements xopt S such that
Vopt = f (xopt )).
62/138
Examples of LMI problem

Stability analysis is a feasability problem.
LQ control is an optimization problem, formulated as:
LQ control
Consider
a controllable system x = Ax + Bu. Find a state feedback u(t) = Kx(t) s.t
R
J = 0 (xT Qx + uT Ru)dt is minimum (given Q > 0 and R > 0) is an optimisation problem
whose solution is obtained solving the Riccati equation:
F ind P > 0, s.t. AT P + P A P BR1 B T P + Q = 0
and then the state feedback is given by:
u(t) = R1 B T P x(t)
which is equivalent to: find P > 0 s.t
T
A P + PA + Q
BT P
PB
R

>0
63/138
Semi-Definite Programming (SDP) Problem

LMI programming is a generalization of the Linear Programming (LP) to cone positive
semi-definite matrices, which is defined as the set of all symmetric positive semi-definite matrices
of particular dimension.
Definition (SDP problem)
A SDP problem is defined as,
min
under constraint F (x)
cT x
0
(20)
where F (x) is an affine symmetric matrix function of x Rm (e.g. LMI) and c Rm is a given
real vector, that defines the problem objective.
SDP problems are theoretically tractable and practically:
They have a polynomial complexity, i.e. there exists an algorithm able to find the global
minimum (for a given a priori fixed precision) in a time polynomial in the size of the problem
(given by m, the number of variables and n, the size of the LMI).
SDP can be practically and efficiently solved for LMIs of size up to 100 100 and m 1000
see ElGhaoui, 97. Note that today, due to extensive developments in this area, it may be even
larger.
64/138
LMI in control
The state feedback design problem

Stabilisation
Let us consider a controllable system x = Ax + Bu. The problem is to find a state feedback
u(t) = Kx(t) s.t the closed-loop system is stable.
Using the Lyapunov theorem, this amounts at finding P = PT > 0 s.t:
(A BK)T P + P(A BK) < 0
AT P
+ PA K T B T P P BK < 0
which is obviously not linear...

Solution; use of change of variables
First, left and right multiplication by P1 leads to
P1 AT + AP1 P1 K T B T BKP1 < 0
QAT + AQ + YT B T + BY < 0
with Q = P1 and Y = KP1 .

The problem to be solved is therefore formulated as an LMI ! and without any conservatism !
65/138
LMI in control
The Bounded Real Lemma

The L2 -norm of the output z of a system LT I is uniformly bounded by times the L2 -norm of
the input w (initial condition x(0) = 0).
A dynamical system G = (A, B, C, D) is internally stable and with an ||G|| < if and only if
there exists a positive definite symmetric matrix P (i.e P = PT > 0 s.t
AT P + P A
BT P
C
PB
I
D
CT
T
< 0,
D
I
P > 0.
(21)
The Bounded Real Lemma (BRL), can also be written as follows (see Scherer)
I
A
0
C
T
0
0
B
P
I 0
D
0
P
0
0
0
0
0
2 I
0
0
I
0
A
0 0
C
I
0
B
0
I
D
(22)
Note that the BRL is an LMI if the only unknown (decision variables) are P and (or 2 ).
66/138
LMI in control
Quadratic stability
This concept is very useful for the stability analysis of uncertain systems.
Let us consider an uncertain system
x = A()x
where is an parameter vector that belongs to an uncertainty set .
Theorem
The considered system is said to be quadratically stable for all uncertainties if there exists a
(single) "Lyapunov function" P = PT > 0 s.t
A()T P + PA() < 0, for all
(23)
This is a sufficient condition for ROBUST Stability which is obtained when A() is stable for all
.
67/138
Some useful lemmas
Schur lemma
Lemma
Let Q = QT and R = RT be affine matrices of compatible size, then the condition

Q
S
0
T
S
R
(24)
is equivalent to
R
Q SR1 S T

0
0
(25)
The Schur lemma allows to a convert a quadratic constraint (ellipsoidal constraint) into an LMI
constraint.
68/138
Some useful lemmas
Kalman-Yakubovich-Popov lemma
Lemma
For any triple of matrices A Rnn , B Rnm , M R(n+m)(n+m) =
M11
M21
M12
M22

, the
following assessments are equivalent:

1
There exists a symmetric K = K T 0 s.t.

I
A
0
B
T
0
K
K
0

I
A
0
B

+M <0
M22 < 0 and for all R and complex vectors col(x, w) 6= 0

A jI
x
w

=0
x
w
T

M
x
w

<0
T

I
0
0 K
I
0
then the second statement is equivalent to
A B
K
0
A B
the condition that, for all R with det(jI A) 6= 0,

Q
S
I
I
>0
C(jI A)1 B + D
C(jI A)1 B + D
ST R

If M =
This lemma is used to convert frequency inequalities into Linear Matrix Inequalities.
69/138
Some useful lemmas
Projection Lemma
Lemma
For given matrices W = W T , M and N , of appropriate size, there exists a real matrix K = K T
such that,
W + M KN T + N K T M T 0
(26)
if and only if there exist matrices U and V such that,

W + MU + UT MT
W + NV + V T NT
TWM
M
TWN
N
0
0
0
0
(27)
or, equivalently, if and only if,
(28)
T M = 0).
where M and N are the orthogonal complements of M , N respectively (i.e. M
The projection lemma is also widely used in control theory. It allows to eliminate variable by a
change of basis (projection in the kernel basis). It is involved in one of the H solutions
[Doyle et al.(1989)Doyle, Glover, Khargonekar, and Francis]see e.g..
70/138
Some useful lemmas
Completion Lemma
Lemma
Let X = X T , Y = Y T Rnn such that X > 0 and Y > 0. The three following statements are
equivalent:
1
There exist matrices X2 , Y2 Rnr and X3 , Y3 Rrr such that,

2

3
X
I
X
I
I
Y
I
Y
X
X2T
X2
X3
X
I
0 and rank

0 and
I
Y
X
X2T
X2
X3
1

=
Y
Y2T
Y2
Y3

(29)

n+r
0 and rank [XY I] r
This lemma is useful for solving LMIs. It allow to simplify the number of variables when a matrix
and its inverse enter in a LMI.
71/138
Some useful lemmas
Finslers lemma
This Lemma allows the elimination of matrix variables.
Lemma
The following statement are equivalent
xT Ax < 0 for all x 6= 0 s:t: Bx = 0
T AB
< 0 where B B
=0
B
A + B T B < 0 for some scalar
A + XB + B T X T < 0 for some matrix X
T
B AB < 0
72/138
Some useful lemmas
Interest of LMIs
LMIs allow to formulate complex optimization problems into "Linear" ones, allowing the use of
convex optimization tools.
Usually it requires the use of different transformations, changes of variables ... in order to linearize
the considered problmems: Congruence, Schur complement, projection lemma, Elimination
lemma, S-procedure, Finslers lemme ...
Examples of handled criteria
stability
H , H2 , H2 /H performances
robustness analysis: Small gain theorem, Polytopic uncertianties, LFT representations...
Robust control and/or observer design
pole placement
stability, stabilization with input constraints
Passivity constraints
Time-delay systems
73/138
5 H control design
Outline
System definitions
Stability issues
Introduction
SISO case
MIMO case
LMI in control
Some useful lemmas
5. H control design
Example
Introduction
LPV modelling
LPV Control
74/138
5 H control design
The General Control Configuration

This approach has been introduced by Doyle (1983). The formulation makes use of the general
control configuration.
P is the generalized plant (contains the plant, the weights, the uncertainties if any) ; K is the
controller. The closed-loop transfer function is given by:
Tew (s) = Fl (P, K) = P11 + P12 K(I P22 K)1 P21
where Fl (P, K) is referred to as a lower Linear Fractional Transformation.
75/138
5 H control design
H control design- Problem definition

The overall control objective is to minimize some norm of the transfer function from w to e , for
example, the H norm.
Definition (H optimal control problem)
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.
Definition (H suboptimal control problem)
Given a pre-specified attenuation level, a H sub-optimal control problem is to design a
stabilizing controller that ensures :
kTew (s)k
max
(Tew (j))
The optimal problem aims at finding min (done using hinfsyn in MATLAB).
76/138
5 H control design
A first case: the state feedback control problem

Let consider the system:
x(t)
Ax(t) + B1 w(t) + B2 u(t)
y(t)
Cx(t)
(30)
The objective is to find a state feedback control law u = Kx s.t:

kTyw (s)k
The method consists in applying the Bounded Real Lemma to the closed-loop system, and then
try to obtain some convex solutions (LMI formulation).
77/138
5 H control design
H control design- Problem formulation

We consider here the case of Dynamic Output Feedback controllers K(s) defined as

K(s) :
x K (t)
u(t)
=
=
AK xK (t) + BK y(t),
CK xK (t) + DK y(t).
It will be shown how to formulate such a control problem using "classical" control tools. The
procedure will be 2-steps:
Get P : Build a control scheme s.t. the closed-loop system matrix does correspond to the
tackled H problem (for instance the mixed sensitivity problem). Use of
Matlab, sysic tool.
Compute K: Use an optimisation algorithm that finds the controller K solution of the
considered problem.

We S
Illustration on
Wu KS

We S

In that case the closed-loop system must be kTew (s)k =
Wu KS i.e a system with
1 input and 2 outputs. Since S = ry

and KS = u
, the control scheme needs only one external
r
r
input r.
78/138
5 H control design
How to consider performance specification in H control ?

Note that: 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.
Control objectives:
y = Gu = GK(r y) tracking error : = Sr
u = K(r y) = K(r Gu) actuator force : u = KSr
To cope with that control objectives the following control scheme is considered:
Objective w.r.t sensitivity functions: kWe Sk 1,

Idea: define 2 new virtual controlled outputs:
kWu KSk 1.
e1 = We Sr
e2 = Wu KSr
79/138
5 H control design
About the control structure- Problem definition

The performance specifications on the tracking error & on the actuator, given as some weights on
the controlled output, then leads to the new control scheme:
The associated general control configuration is :
80/138
5 H control design
About the control structure- Problem definition

The corresponding
control problem is therefore to find a controller K(s) such that
H suboptimal

We S

with
kTew (s)k =
Wu KS
Tew (s)
= Fl (P,K)= P11 + P12 K(I P22 K)1 P21
We
We G
=
+
K(I + GK)1 I
0
Wu

We S
=
Wu KS
in Matlab
81/138
5 H control design
What about disturbance attenuation ?

To account for input disturbance rejection, the control scheme must include di :
This corresponds to the closed-loop system.

We Sy
Tew =
Wu KSy
We Sy G
Wu Tu
The new H control problem therefore includes the input disturbance rejection objective, thanks
to Sy G that should satisfy the same template as S, i.e an high-pass filter!
Remarks: Note that Wu Tu is an additional constraint that may lead to an increase of the
attenuation level since it is not part of the objectives. Hopefully Tu is low pass, and Wu as well.
The input weight has to be on u not u + di which would lead to an unsolvable problem.
82/138
5 H control design
Improve the disturbance attenuation

The previous problem, allows to ensure the input disturbance rejection, but does not provide any
additional d-o-f to improve it (without impacting the tracking performance). In order to decouple
both performance objectives, the idea is to add a disturbance model that indeed changes the
disturbance rejection properties.
Let then consider : di (t) = Wd .d. In that case the closed-loop system is This corresponds to the
closed-loop system.

We Sy
We Sy GWd
Tew =
Wu KSy
Wu Tu Wd
and the template expected for Sy G is now W 1.W .
e
d
First interest: improve the disturbance weight as Wd = 100... but this has a price (see Fig. below
for an example)
Sensitivity*Plant SG
Controller*Sensitivity KS
20
40
30
with Wd=1
with Wd=100
20
Magnitude (dB)
-20
-40
-60
with Wd=100
10
0
-10
-20
-80
with Wd=1
-30
-100
-40
-120 -2
10
O. 10
Sename [GIPSA-lab]
10
-1
10
10
10
-50 -2
10
10
-1
10
10
83/13810
10
5 H control design
More generally...
To include multiple objectives in a SINGLE H control problem, there are 2 ways:
1
add some external inputs (reference, noise, disturbance, uncertainties ...)
add new controlled outputs
Of course both ways increase the dimension of the problem to be solved....thus the complexity as
well. Moreover additional constraints appear that are not part of the objectives ....
General rule: first think simple !!
84/138
5 H control design

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.
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
[Zhou et al.(1996)Zhou, Doyle, and Glover] [Skogestad and Postlethwaite(1996)].
Notations:
A
B1
B2
x = Ax + B1 w + B2 u
e = C1 x + D11 w + D12 u
P
P = C1
D11 D12
y = C2 x + D21 w + D22 u
C2
D21 D22
with x Rn , x Rnw , u Rnu , z Rnz et y Rny
85/138
5 H control design
Problem formulation
Let K(s) be a dynamic output feedback LTI controller defined as

x K (t)
u(t)
K(s) :
=
=
AK xK (t) + BK y(t),
CK xK (t) + DK y(t).
where xK Rn , and AK , BK , CK and DK are matrices of appropriate dimensions.

Remark. The controller will be considered here of the same order (same number of state variables)
n than the generalized plant, which here, in the H framework, the order of the optimal controller.
With P (s) and K(s), the closed-loop system N (s) is:

N (s) :
x cl (t)
z(t)
=
=
A xcl (t) + B w(t),

C xcl (t) + D w(t),
(31)
T

T
where xT
cl (t) = x (t) xK (t) and

A + B2 DK C2 B2 CK
A
=
,
BK C2

AK
B1 + B2 DK D21
B=
,
BK D21
C
+
D
D
C
,
D
C
=
,
1
12
2
12 CK
K
D = B1 + B2 DK D21 .
The aim is of course to find matrices AK , BK , CK and DK s.t. the H norm of the closed-loop
system (31) is as small as possible, i.e. opt = min s.t. ||N (s)|| < .
86/138
5 H control design
Asuumptions for the Riccati method

A1: (A, B2 ) stabilizable and (C2 , A) detectable: necessary for the existence of stabilizing
controllers
A2: rank(D12 ) = nu and rank(D21 ) = ny : Sufficient to ensure the controllers are proper, hence
realizable

A jIn
B2
A3: R, rank
= n + nu
C1
D12

A jIn
B1
A4: R, rank
= n + ny Both ensure that the optimal controller does
C2
D21
not try to cancel poles or zeros on the imaginary axis which would result in CL instability
A5:
D11 = 0,
D22 = 0,

B1
D21
D12 T [ C1

D21 T =
D12 ] =

Inu
I ny
not necessary but simplify the solution (does correspond to the given theorem next but can be
easily relaxed)
87/138
5 H control design
The problem solvability

The first step is to check whether a solution does exist of not, to the optimal control problem.
Theorem (1)
Under the assumptions A1 to A5, there exists a dynamic output feedback controller
u(t) = K(.) y(t) such that the closed-loop system is internally stable and the H norm of the
closed-loop system from the exogenous inputs w(t) to the controlled outputs z(t) is less than , if
and only if

A
2 B1 B1T B2 B2T
i the Hamiltonian H =
has no eigenvalues on the
T
T
C1 C1
A
imaginary axis.
ii there exists X 0 t.q. AT X + X A + X ( 2 B1 B1T B2 B2T ) X + C1T C1 = 0,

AT
2 C1T C1 C2T C2
iii the Hamiltonian J =
has no eigenvalues on the
B1 B1T
A
imaginary axis.
iv there exists Y 0 t.q. A Y + Y AT + Y ( 2 C1T C1 C2T C2 ) Y + B1 B1T = 0,
v the spectral radius (X Y ) 2 .
88/138
5 H control design
Controller reconstruction
Theorem (2)
If the necessary and sufficient conditions of the Theorem 1 are satisfied, then the so-called central
controller is given by the state space representation

Ksub (s) =
A
F
Z L
0
with
= A + 2 B1 B T X + B2 F + Z L C2
A
1
F = B2T X , L = Y C2T
1
Z = I 2 Y X
The Controller structure is indeed an observer-based state feedback control law, with
u2 (t) = B2T X x
(t),
where x
(t) is the observer state vector

x
(t) = A x
(t) + B1 w(t)
+ B2 u(t) + Z L C2 x
(t) y(t) .
and w(t)
is defined as
89/138
(32)
5 H control design
The LMI approach for H control design- Solvability

In this case only A1 is necessary. The solution is base on the use of the Bounded Real Lemma,
and some relaxations that leads to an LMI problem to be solved [Scherer(1990)].
when we refer to the H control problem, we mean: Find a controller C for system M such that,
given ,
||Fl (P, K)|| <
(33)
and is called the optimal H

The minimum of this norm is denoted as
gain. Hence, it comes,
min
(AK ,BK ,CK ,DK )s.t.AC
kTzw (s)k
(34)
As presented in the previous sections, this condition is fulfilled thanks to the BRL. As a matter of
fact, the system is internally stable and meets the quadratic H performances iff. P = P T 0
such that,
T
A P + PA
PB
CT
(35)
BT P
22 I DT < 0
C
D
I
where A, B, C, D are given in (31). Since this inequality is not an LMI and not tractable for SDP
solver, relaxations have to be performed (indeed it is a BMI), as proposed in
[Scherer et al.(1997b)Scherer, Gahinet, and Chilali].
90/138
5 H control design
The LMI approach for H control design- Problem solution

Theorem (LTI/H solution [Scherer et al.(1997a)Scherer, Gahinet, and Chilali])

Ac Bc
A dynamical output feedback controller of the form C(s) =
that solves the H
Cc Dc
e and D),
e B,
e C
e while
control problem, is obtained by solving the following LMIs in (X, Y, A,
minimizing ,
M11 ()T ()T ()T

M21 M22 ()T ()T
M31 M32 M33 ()T 0

(36)
M41 M42 M43 M44
X In
0
In
Y
where,
T
e+C
e BT
M11 = AX + XAT + B2 C
2
T
e
eT
M22 = YA + A Y + BC2 + C T B
e + AT + C T D
e BT
M21 = A
2
2
T T
T
T
e
M31 = B + D D B
T B
eT
M32 = B1T Y + D21
e
M41 = C1 X + D12 C
e 21
M43 = D11 + D12 DD
M33 = Inu
e 2
M42 = C1 + D12 DC
M44 = Iny
21
(37)
91/138
5 H control design
Controller reconstruction
Once A, B, C, D, X and Y have been obtained, the reconstruction procedure consists in finding
non singular matrices M and N s.t. M N T = I X Y and the controller K is obtained as follows
Dc
Cc
B
c
Ac
=
=
=
=
e
D
e Dc C2 X)M T
(C
e YB2 Dc )
N 1 (B
e YAX YB2 Dc C2 X N Bc C2 X YB2 Cc M T )M T
N 1 (A
(38)
where M and N are defined such that M N T = In XY (that can be solved through a singular
value decomposition plus a Cholesky factorization).
Remark. Note that other relaxation methods can be used to solve this problem, as suggested by
[Gahinet(1994)].
92/138
6 Robust analysis
Outline
System definitions
Stability issues
Introduction
SISO case
MIMO case
LMI in control
Some useful lemmas
5. H control design
Example
Introduction
LPV modelling
LPV Control
93/138
6 Robust analysis
Introduction
Introduction
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.

The approach
1
2
3
determine the uncertainty set: mathematical representation

check Robust Stability
check Robust Performance
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
Measurement imperfections (due to sensor)
At high frequencies, even the structure and the model order is unknown (100
Choice of simpler models for control synthesis
Controller implementation
Two classes: parametric uncertainties / neglected or unmodelled dynamics

94/138
6 Robust analysis
Example 1: uncertainties [?]
G(s)
=
k
esh ,
1 + s
2 k, h, 3
Let us choose the nominal parameters as, k = h = = 2.5 and G the according nominal model.
We can define the relative uncertainty, which is actually referred as a MULTIPLICATIVE
UNCERTAINTY, as
G(s)
= G(s)(I + Wm (s)(s))
with Wm (s) = 3.5s+0.25
s+1
and kk 1
95/138
6 Robust analysis
Example 2: unmodelled dynamcis

Let us consider the system:
G(s)
= G(s)
1
, max
1 + s
This can be modelled as:
G(s)
= G0 (s)(I + Wm (s)(s)), Wm (s) =
max j
1 + max j
with kk 1
which can be represented as
with
N (s) =

N11 (s)
N21 (s)

N12 (s)
0
=
N22 (s)
G0 Wm (s)
I
G0 (s)
96/138
6 Robust analysis
Example 3: parametric uncertainties

Consider the first order system:
G(s) =
1
,
s+a
a0 b < a < a0 + b
Define now a = a0 + .b with || < 1. Then it leads:

1
1
1
.b 1
=
=
(1 +
)
s+a
s + a0 + .b
s + a0
s + a0
This can then be represented as a Multiplicative Inverse Uncertainty:
with
z = y =
1
(w
s+a0
bu )
97/138
6 Robust analysis
Example 4: parametric uncertainties in state space equations

Let us consider the following uncertain system:
x 1 = (2 + 1 )x1 + (3 + 2 )x2
x 2 = (1 + 3 )x2 + u
G:
y
= x1
(39)
In order to use an LFT, let us define the uncertain inputs:

u1 = 1 x1 , u2 = 2 x2 , u3 = 3 x2
Then the previous system can be rewritten in the following LFR:
where and y are given as:
1
0
0
x1
0 , y =
x2
= 0 2
0
0 3
x2
and N given by the state space representation:
=
2x1 3x2 + u1 + u
x 1
x 2
=
x2 + u + u3
N :
y
=
x1
98/138
6 Robust analysis
Towards LFR (LFT)

The previous computations are in fact the first step towards an unified representation of the
uncertainties: the Linear Fractional Representation (LFR).
Indeed the previous schemes can be rewritten in the following general representation as:
Figure: N structure
This LFR gives then the transfer matrix from w to z, and is referred to as the upper Linear
Fractional Transformation (LFT) :
Fu (N, ) = N22 + N21 (I N11 )1 N12
This LFT exists and is well-posed if (I N11 )1 is invertible
99/138
6 Robust analysis
LFT definition
In this representation N is known and (s) collects all the uncertainties taken into account for the
stability analysis of the uncertain closed-loop system.
(s) shall have the following structure:
(s) = diag {1 (s), , q (s), 1 Ir1 , , r Irr , 1 Ic1 , , c Icc }
k k
with i (s) RHi i , i R and i C.

Remark: (s) includes
q full block transfer matrices,
r real diagonal blocks referred to as repeated scalars (indeed each block includes a real
parameter i repeated ri times),

c complex scalars i repeated ci times.
Constraints: The uncertainties must be normalized, i.e such that:

kk 1, |i | 1, |i | 1
100/138
6 Robust analysis
Uncertainty types
We have seen in the previous examples the two important classes of uncertainties, namely:
UNSTRUCTURED UNCERTAINTIES: we ignore the structure of , considered as a full
complex perturbation matrix, such that kk 1.

We then look at the maximal admissible norm for , to get Robust Stability and Performance.
This will give a global sufficient condition on the robustness of the control scheme.
This may lead to conservative results since all uncertainties are collected into a single matrix
ignoring the specific role of each uncertain parameter/block.
STRUCTURED UNCERTAINTIES: we take into account the structure of , (always such that
kk 1).
The robust analysis will then be carried out for each uncertain parameter/block.
This needs to introduce a new tool: the Structured Singular Value. We then can obtain more
fine results but using more complex tools.
The analysis is provided in what follows for both cases.
In Matlabthis analysis is provided in the tools robuststab and robustperf.
101/138
6 Robust analysis
Robustness analysis: problem formulation

Since the analysis will be carried you for a closed-loop system, N should be defined as the
connection of the plant and the controller. Therefore, in the framework of the H control, the
following extended General Control Configuration is considered:
Figure: P K structure
and N is such that

N = Fl (P, K)
102/138
6 Robust analysis
Robust analysis: problem definition

In the global P K General Control Configuration, the transfer matrix from w to z (i.e the
closed-loop uncertain system) is given by:
z = Fu (N, )w,
with Fu (N, ) = N22 + N21 (I N11 )1 N12 .
and the objectives are then formulated as follows:
Nominal stability (NS): N is internally stable
Nominal Performance (NP): kN22 k < 1 and NS
Robust stability (RS): Fu (N, ) is stable , kk < 1 and NS
Robust performance (RP): kFu (N, )k < 1 , kk < 1 and NS
103/138
6 Robust analysis
Towards Robust stability analysis

Robust Stability= with a given controller K, we determine wether the system remains stable for all
plants in the uncertainty set.
According to the definition of the previous upper LFT, when N is stable, the instability may only
come from (I N11 ). Then it is equivalent to study the M structure, given as:
Figure: M structure
This leads to the definition of the Small Gain Theorem

Theorem (Small Gain Theorem)
Suppose M RH . Then the closed-loop system in Fig. 7 is well-posed and internally stable
for all RH such that :
kk (resp. < 1)
if and only if
kM (s)k < 1/(resp. kM (s)k 1)

104/138
6 Robust analysis
Definition of the uncertainty types
Figure: 6 uncertainty representations

105/138
6 Robust analysis
Robust stability analysis: additive and output multiplicative case

Objective: applying the Small Gain Theorem to these unstructured uncertainty representations.
Let us consider the following simple control
scheme as:
The objective is to obtain:
Additive case:
G(s)
= G(s) + WA (s)A (s).
Computing the N form gives

WA KSy WA KSy
N (s) =
Sy
Ty
Output Multiplicative uncertainties:
G(s)
= (I + WO (s)O (s))G(s).
Then it leads

WO Ty WO Ty
N (s) =
Sy
Ty
106/138
6 Robust analysis
General results
Theorem (Small Gain Theorem)
Consider the different uncertainty types, and assume that NS is achieved, i.e M RH for each
type. Then the closed-loop system is robustly stable, i.e. internally stable for all k RH (for
k =A, 0, I, iO, iI) with kk 1, if and only if:
Additive :
kWA KSy k 1
Additive Inverse:
kWiA Sy Gk 1
Output Multiplicative:
kWO Ty k 1
Input Multiplicative:
kWI Tu k 1
Output Inverse Multiplicative: kWiO Sy k 1
Input Inverse Multiplicative:
kWiI Su k 1
This gives some robustness templates for the sensitivity functions. However this may be
conservative.
107/138
6 Robust analysis
Illustration on the SISO case

Here Robust Stability is analyzed through the Nyquist plot. For illustration, let us consider the case
of Multiplicative uncertainties (Input and Output case are identical for SISO systems), i.e
= G(I + Wm m )
G
Then the loop transfer function is given as:
= GK
= GK(I + Wm m ) = L + Wm Lm ;
L
According to the Nyquist theorem, RS is

achieved the the closed-loop system is
should not encircle, i.e L
stable for any L

should not encircle -1 for all uncertainties.
According to the figure, a sufficient condition
is then:
|Wm L|
< |1 + L| ,
mL
< 1,
W
1+L
|Wm T | < 1
108/138
6 Robust analysis
A first insight in Robust Performance

Objective: applying the Small Gain Theorem to these unstructured uncertainty representations.
Let us consider the following simple control
scheme as:
Case of Output Multiplicative uncertainties:
G(s)
= (I + WO (s)O (s))G(s).
Computing the N form gives
N (s)
=
=
We wish to get:

N11 (s) N12 (s)
N21 (s) N22 (s)

WO Ty WO Ty
We Sy
We Sy
The objectives are then formulated as

follows:
NS: N is internally stable
NP: kWe Sy k < 1 and NS
RS: kWO Ty k < 1 and NS
RP: kFu (N, )k < 1 , kk < 1,
Sufficient condition: NS and
(WO Ty ) +
(We Sy ) < 1,
109/138
6 Robust analysis
Illustration on the SISO case

Here Robust Performance is analyzed through the Nyquist plot. For illustration, let us consider the
case of Multiplicative uncertainties (Input and Output case are identical for SISO systems), i.e
= G(I + Wm m )
G
Then the loop transfer function is given as:
= GK
= GK(I + Wm m ) = L + Wm Lm ;
L
First NP is achieved when:

|We S| < 1 , |We | <
|1 + L| , .
Therefore RP is achieved if

We S < 1, S,

, L,

|We | < 1 + L

|1 + L| |Wm Lm |, a
Since 1 + L
sufficient condition is actually:
|We | + |Wm L| < |1 + L| ,

|We S| + |Wm T | < 1,
110/138
6 Robust analysis
The structured case
= {diag{1 , , q , 1 Ir1 , , r Irr , 1 Ic1 , , c Icc } Ckk }

with i Cki ki , i R, i C,
where i (s), i = 1, . . . , q, represent full block complex uncertainties, i (s), i = 1, . . . , r, real
parametric uncertainties, and i (s), i = 1, . . . , c, complex parametric uncertainties.
Taking into account the uncertainties leads to the following General Control Configuration,
(s)
vr
zr
- P (s)
v
-e
y
K(s)
Figure: General control configuration with uncertainties
where .
111/138
(41)
6 Robust analysis
The structured singular value

To handle parametric uncertainties, we need to introduce , the structured singular value, defined
as:
Definition ()
For M Cnn , the structure singular value is defined as:
(M ) :=
1
min{() : , det(I M ) 6= 0}
In other words, it allows to find the smallest structured which makes det(I M ) = 0.
Theorem (The structured Small Gain Theorem)
Let M (s) be a MIMO LTI stable system and (s) a LTI uncertain stable matrix, (i.e. RH ).
The system in Fig. 7 (M ) is stable for all (s) in (41) if and only if:
R
(M (j)) 1, with M (s) := Nzv (s)
More generally both following statements are equivalent

For R, N (s) and (s) belong to RH , and
R,
(M (j))
the system represented in figure 7 (M ) is stable for any uncertainty (s) of the form (41)
such that :
112/138
6 Robust analysis
Build the whole control scheme
113/138
6 Robust analysis
Introduction of a fictive block

Usually only real parametric uncertainties (given in r ) are considered for RS analysis. RP
analysis also needs a fictive full block complex uncertainty, as below,
(s)
f

r
vr
z r
N (s)
Figure: N

where N (s) =
N11 (s)
N21 (s)
N12 (s)
N22 (s)

, and the closed-loop transfer matrix is:
Tew (s) = N22 (s) + N21 (s)(s)(I N11 (s))1 N12 (s)
(42)
114/138
6 Robust analysis
Robust analysis theorem

For RS, we shall determine how large (in the sense of H ) can be without destabilizing the
feedback system. From (42), the feedback system becomes unstable if det(I N11 (s) = 0 for
some s C, <(s) 0. The result is then the following.
Theorem
Assume that the nominal system New and the perturbations are stable. Then the feedback
system is stable for all allowed perturbations such that ||(s)| | < 1/ if and only if
R, (N11 (j)) .
Assuming nominal stability, RS and RP analysis for structured uncertainties are therefore such
that:
NP
(N22 ) = f (N22 ) 1,
RS
r (N11 ) < 1,
RP
(N ) < 1, , =
f
0
0
r
Finally, let us remark that the structured singular value cannot be explicitly determined, so that the
method consists in calculating an upper bound and a lower bound, as closed as possible to .
115/138
6 Robust analysis
Summary
The steps to be followed in the RS/RP analysis for structured uncertainties are then:
Definition of the real uncertainties r and of the weighting functions
Evaluation of (N22 ) , (N11 ) and (N)
f
Computation of the admissible intervals for each parameter
Remark: The Robust Performance analysis is quite conservative and requires a tight definition of
the weighting functions that do represent the performance objectives to be satisfied by the
uncertain closed-loop system. Therefore it is necessary to distinguish the weighting functions
used for the nominal design from the ones used for RP analysis.
116/138
6 Robust analysis
Robust control design
Brief overview on robust control design

In order to design a robust control, i.e. a controller for which the synthesis actually accounts for
uncertainties, some of the methods are:
Unstructured uncertainties: Consider an uncertainty weight (unstructured form), and
include the Small Gain Condition through a new controlled output. For example, robustness
face to Ouptut Multiplicative Uncertainties can be considered into the design procedure
adding the controlled output ey = WO y, which, when tracking performance is expected, leads
to the condition k WO Ty k 1.
Structured uncertainties: the design of a robust controller in the presence of such
uncertainties is the synthesis. It is handled through an interactive procedure, referred to

as the DK iteration. This procedure is much more involved than a "simple" H control
design and often leads to an increase of the order of the controller (which is already the sum
of the order of the plant and of the weighting functions).
Use other mathematical representation of parametric uncertainties,
[Scherer and Wieland(2004)], as for instance the polytopic model. In that case the set of
uncertain parameters is assumed to be a polytope (i.e. a convex) set. The stability issue in
that framework is referred to as the Quadratic stability i.e find a single Lyapunov function for
the uncertainty set. While in the general case this is an unbounded problem, in the polytopic
case (or in the affine case), the stability is to be analyzed only at the vertices of the polytope,
which is a finite dimensional problem.
This approach can then be applied to find a single controller, valid over the potyopic set. Note
that this approach gives rise to the LPV design for polytopic systems, as described next.
117/138
7 LPV systems
Outline
System definitions
Stability issues
Introduction
SISO case
MIMO case
LMI in control
Some useful lemmas
5. H control design
Example
Introduction
LPV modelling
LPV Control
118/138
7 LPV systems
LPV modelling
The LPV approach

Definition of an Linear Parameter Varying system

x
A()
() : z = C1 ()
y
C2 ()
B1 ()
D11 ()
D21 ()
B2 ()
x
D12 () w
D22 ()
u
x(t) Rn , ...., = (1 (t), 2 (t), . . . , N (t)) , is a vector of time-varying parameters (

convex set)
Example
(Scherer, ACC Tutorial 2012)
Dampened mass-spring system:

Exogeneous
Inputs
System()
z
y
ematical Systems Theory
Controlled
outputs
Measured
outputs
Control
Inputs
p + c p_ + k(t) p = u; y = x
First-order state-space representation:
0 1
0
10 1
d @pA
dt p_
y
@

0 1
0 1 A @ p A + @ 0 A u;
k (t ) c
p_
1
= 10
0 1
@pA
p_
119/138
7 LPV systems
LPV modelling
Example
Let consider the previous systems with : c = and k [1, 3]. The frozen Bode plots
120/138
7 LPV systems
LPV modelling
Different Models
According to the dependency on the parameter set, we may have several classes of models:
1
Affine parameter dependency: A() = A0 + A1 1 + ... + AN N
Polynomial dependency: A() = A0 + A1 + A2 2 + ... + AS S
Rational
dependency:A() = [An0 + An1 n1 + ... + AnN nN ][I + Ad1 d1 + ... + AdN dN ]1
If = (x(t), t) then the system is referred to as quasi-LPV.

For instance:
x(t)
= x2 (t) = (t)x(t)
with = x.
Affine to polytopic
An affine LPV system can be transformed into a polytopic representation (explained below) which
is easier to handle from a convex optimisation point of view.
Let us denote N the number of parameters.

The parameter are assumed to be bounded : i i i .
The vector of parameters evolves inside a polytope represented by Z = 2N vertices i , as
Co{1 , . . . , Z }
(43)
121/138
7 LPV systems
LPV modelling
Polytopic models
It is then written as the convex combination:
=
Z
X
i i , i 0,
i=1
Z
X
(44)
i = 1
i=1
where the vertices are defined by a vector i = [i1 , . . . , iN ] where ij equals j or j .

A system is then represented as
() =
Z
X

k ()
k=1
Ak
Ck
Bk
Dk

,
with
2
X
k () = 1 , k () > 0
k=1

Ak Bk
is a LTI system corresponding to a vertex k.
Ck Dk
ATTENTION: Do remind that the LPV system is TIME-VARYING so a polytopic LPV system
is not an uncertain polytopic system (in the latter case k is constant even if unknown)
where
122/138
7 LPV systems
LPV modelling
Poytopic models
h
i
For a LPV system with 2 parameters, boundend 1,2 1,2 , the corresponding polytope
onws 4 vertices as:
n
o
P = (1 , 2 ), (1 , 2 ), (1 , 2 ), (1 , 2 )
(45)
The polytopic coordinates are (i ) are obtained as:
1 =
1 1
1 1
1 = ( , ),
2 =
1 1
1 1
2 = ( , 2 ),
1 1
3 = (1 , ),
3 =
4 = (1 , 2 ),
4 =
1 1
1 1
1 1
2 2
2 2
2 2
2 2
2 2
2 2
2 2
2 2
(k)
where 1 and 2 are the instantaneous values of the parameters (i

step).
The LPV system is then rewritten under the polytopic representation:

A(1,2 )
C(1,2 )
(46)
in the implementation

B(1,2 )
A(1 ) B(1 )
A(2 ) B(2 )
= 1
+ 2
D(1,2 )
C(1 ) D(1 )
C(2 ) D(2 )

A(3 ) B(3 )
A(4 ) B(4 )
+ 3
+ 4
C(3 ) D(3 )
C(4 ) D(4 )
123/138
(47)
7 LPV systems
LPV modelling
Some recent books

R.Toth, Modeling and identification of linear parameter-varying systems, Springer 2010
Control of Linear Parameter Varying Systems with Applications, Mohammadpour, Javad,
Scherer, Carsten W. , Springer-Verlag New York, 2012 (Eds.)

Robust Control and Linear Parameter Varying Approaches: Application to Vehicle Dynamics,
Olivier Sename, Peter Gaspar, Jozsef Bokor (Eds), Springer, 2013

Due to the time-varuing nature of the parameters, observability and controllability properties are
different from LTI systems.
As for non linear systems (and time-delay systems), (initial) observability and reconstructibility
notions must be distinguished (the latter is considered for observer design).
124/138
7 LPV systems
LPV modelling
LFT models
The equation of the system under LFR representation is as

follows:
x
A
B
B1
B2
x
z C D D1 D2 w
z = C1
D1
D11
D12 w
y
C2
D2
D21
D22
u
Figure: System under LFT form
125/138
7 LPV systems
LPV modelling
Towards a LFT/LPV model of an AUV

Considered the NL model with 2 state variables (the pitch angle and velocity q):
=cos()q sin()r
M q = p r(Ix Iz ) m[Zg (q w r v)] (Zq m Zf V )gsin()
(Xq m Xf V )gcos()cos() + Mwq w|q| + Mqq q|q| + Ff ins
where M interia matrix, m AUV mass, V volume and mass density. The other parameters
(Ix , Iz , Zg , Zq , Zf , Xq , Xf , Mwq ) appear in the dynamical and hydrodynamical functions.
Ff ins : forces and moments due to the fins.
Tangent linearization around Xeq = [0 ueq 0 0 0 0 0 0 eq 0 0 0]:
= q
= [(Zg m Zf V )gcos(eq ) + (Xg m Xf V )gsin(eq )] + Ff ins
M q
eq
where and q are the variations of and q.
126/138
(48)
7 LPV systems
LPV modelling
A LFR model
An LFR model of the vehicle described in the form:
()
w
wk
uk
P (z)
zk
yk
The block contains the varying part of the model, which depends on the linearization point (eq ).
The LFR form is defined by:

u
xk+1 = Axk + B B1
u

(49)
y C
u
D
0
=
xk +
C1
0
D1
y
u
127/138
7 LPV systems
LPV modelling
A LFR model
with

A=

B1 =

D =
0
0
1
0
0
(Zq m Zf V )g

0
1 1
1
, C1 =
, C =
Ff ins
0 0
0

0 0
0 0
, D1 =
0 0
0 0
, B =
z =

; u = z ; = 1
0
(Xq m Xf V )g

0
1
(50)
0
2

(51)
with:

1 = cos(eq )
2 = sin(eq )
(52)
128/138
7 LPV systems
LPV Control
Towards LPV control

The "gain scheduling" approach
External
parameters
Adaptation
strategy
References
Controller()
Measured orestimated
Parameters
Control
Inputs
Outputs
System()
Some references
Modelling, identification : (Bruzelius, Bamieh, Lovera, Toth)
Control (Shamma, Apkarian & Gahinet, Adams, Packard, Beker ...)
Stability, stabilization (Scherer, Wu, Blanchini ...)
Geometric analysis (Bokor & Balas)
129/138
7 LPV systems
LPV Control
The H /LP V control problem

Definition
kzk
Find a LPV controller C() s.t the closed-loop system is stable and for > 0, sup kwk2 < ,
2
Unbounded set of LMIs (Linear Matrix Inequalities) to be solved ( )

Some approaches: polytopic, LFT, gridding. See Arzelier [HDR, 2005], Bruzelius [Thesis,
2004], Apkarian et al. [TAC, 1995]...

A solution: The "polytopic" approach [C. Scherer et al. 1997]
Problem solved off line for each vertex of a polytope (convex optimisation) (using here a
single Lyapunov function i.e. quadratic stabilization).

On-line the controller is computed as the convex combination of local linear controllers
2
C() =
2N
X
k=1

k ()
Ac (k )
Cc (k )
X
2N
Bc (k )
,
k () = 1 , k () > 0
Dc (k )
k=1
C(2 )
C(4 )
C()
C(1 )
1
C(3 )
- 1
Performance and stability ensured for all parameter variations and .... easy implementation !!
130/138
7 LPV systems
LPV Control
LPV control design

CL ()
w
Generalized
plant
()
w: exagenous input
u: control input
z: output to minimize
y: measurement
Controller
S ()
Problem on standard form
Dynamical LPV generalized plant:
A()
x
() : z = C1 ()
y
C2 ()
B1 ()
D11 ()
D21 ()
B2 ()
x
D12 () w
u
D22 ()
(53)
LPV controller structure:

S() :
xc
u

=
Ac ()
Cc ()
Bc ()
Dc ()

B()
D()

xc
y
(54)
LPV closed-loop system:

CL() :

=
A()
C()

(55)
131/138
7 LPV systems
LPV Control
LPV control design

H criteria Apkarian et al. [TAC, 1995]
Stabilize system CL() (find K > 0) while minimizing .
A()T K + KA() KB () C ()T

T
2
T
B () K
I
D () < 0
C ()
D ()
I
Infinite set of LMIs to solve ( ) ( is convex)

LPV control designs Arzelier [HDR, 2005], Bruzelius [Thesis, 2004]
LFT, Gridding, Polytopic
132/138
7 LPV systems
LPV Control
LPV control design

Polytopic approach
Solve the LMIs at each vertex of the polytope formed by the extremum values of each varying
parameter, with a common K Lyapunov function.
N
C() =
2
X

k ()
k=1
Ac (k )
Cc (k )
Bc (k )
Dc (k )
where,
QN
k () =
c
j=1 |j C (k )j |
QN
j=1 (j j )
where C c (k )j = {j if (k )j = j or j } otherwise.
N
2
X
k () = 1 , k () > 0
k=1
2
2
C(2 )
C(1 )
C(4 )
C()
C(3 )
133/138
7 LPV systems
LPV Control
LPV/H control synthesis

Proposition - feasibility (brief) Scherer et al. (1997)
Solve the following problem at each vertices of the parametrized points (illustration with 2
parameters):
= min
e 1 , 2 ) + (?)T
AX + B2 C(
e 1 , 2 ) + AT
A(
B1T
e 1 , 2 )
C1 X + D12 C(
s.t.
s.t.
s.t.
s.t.
(57) |1 ,2
(57) |1 ,2
(57) |1 ,2
(57) |1 ,2
(?)T
e 1 , 2 )C2 + (?)T
YA + B(
T B(
e 1 , 2 )T
B1T Y + D21
C

1
X I
0
I Y
(56)
(?)T
(?)T
I
D11
(?)T
(?)T
(?)T
I
134/138
(57)
7 LPV systems
LPV Control
LPV/H control synthesis

Proposition - reconstruction (brief) Scherer et al. (1997)
Reconstruct the controllers as,
solve
Cc (1 , 2 )
Bc (1 , 2 )
Ac (1 , 2 )
=
=
=
(59) |1 ,2
(59) |1 ,2
(59) |1 ,2
(59) |1 ,2
(58)
e 1 , 2 )M T
C(
e 1 , 2 )
N 1 B(
1
e 1 , 2 ) Y AX N Bc (1 , 2 )C2 X
N
A(

Y B2 Cc (1 , 2 )M T M T
where M and N are defined such that M N T = I XY which may be chosen by applying a
singular value decomposition and a Cholesky factorization.
135/138
(59)
7 LPV systems
LPV Control
Interest of the LPV approach

LPV is a key tool to the control of complex systems.
Some examples :
Modelling of complex systems (non linear)
Use of a quasi-LPV representation to include non linearities in a linear state space model
(even delays)
Transformation of constraints (e.g. saturation) into an external parameter
Modelling of LTV, hybrid (e.g. switching control)
leads to Polynomial, LFT or poltytopic representations

BUT :
A q-LPV system is not equivalent to the non linear one:
stability: = (x(t), t) is assumed to be bounded... so are the state trajectories
controllability: some non controllable modes of a non linear system may vanish according to
the LPV representation
136/138
7 LPV systems
LPV Control
Interest of the LPV approach

Some of works using LPV approaches - former PhD students
Gain-scheduled control
Account for various operating conditions using a variable "equilibrium point": (Gauthier 2007)
Control with real-time performance adaptation using parameter dependent weighting
functions from endogenous or exogenous parameters (Poussot 2008, Do 2011)

Analysis and control of LPV Time-Delay Systems: delay-scheduled control Briat 2008
Control under computation constraints: H variable sampling rate controller with sampling
dependent performances (Robert 2007, Roche 2011, Robert et al., IEEE TCST 2010))
Coordination of several actuators for MIMO systems
An LPV structure for control allocation Poussot et al. (CEP 2011)
Selection of a specific parameter for the control activation Poussot et al. (VSD 2011)
137/138
7 LPV systems
LPV Control
Some PhD students on robust and/or LPV control
Soheib Fergani, "H /LPV robust MIMO control of vehicle dynamics", PhD, GIPSA-lab, Universit Grenoble Alpes, 2014.
Commande robuste pour une gestion nergtique fonction de l?tat de sant de la batterie au sein des vhicules hybrides
Caroline Ngo, "Surveillance du sysyme de post-traitement essence et contrle de chane dair compress", PhD, GIPSA-lab / RENAULT, Grenoble
INP, 2014.
Tinghong Wang, "Robust control approach to battery health accommodation for Energy Management Systems in hybrid vehicles ", PhD, GIPSA-lab,
Universit Grenoble Alpes, 2013.
Maria Rivas, "Modeling and Control of a Spark Ignited Engine for Euro 6 European Normative", PhD, GIPSA-lab / RENAULT, Grenoble INP, 2012.
David Hernandez, "Robust control of hybrid electro-chemical generators", PhD, GIPSA-lab / G2Elab, Grenoble INP, 2011.
Sbastien Aubouet, "Semi-active SOBEN suspensions modelling and control", PhD, GIPSA-lab / SOBEN, INP Grenoble, 2010.
David Robert, "Contribution linteraction commande/ordonnacement", PhD, LAG, Grenoble INP, 2007.
Alessandro ZIN, "Sur la commande robuste de suspensions automobiles en vue du contrle global de chssis", PhD, LAG / Grenoble INP, 2005.
Ahn-Lam Do, "LPV Approach for Semi-active Suspension Control & Joint Improvement of Comfort and Security", PhD, GIPSA-lab, Grenoble INP,
2011.
Emilie Roche, "Commande Linaire Paramtres Variants discrte chantillonnage variable : application un sous-marin autonome", PhD,
GIPSA-lab, Grenoble INP, 2011.
Charles Poussot-Vassal, "Robust LPV Multivariable Global Chassis Control", PhD , GIPSA-lab, INP Grenoble, 2008.
Corentin Briat, "Robust control and observation of LPV time-delay systems", PhD, GIPSA-lab, INP Grenoble, 2008.
Christophe Gauthier, "Commande multivariable de la pression dinjection dans un moteur Diesel Common Rail", PhD, LAG / DELPHI, Grenoble INP,
2007.
Marc Houbedine, "Contribution pour lamlioration de la robustesse et du bruit de phase des synthtiseurs de frquences" , PhD, LAG / ST
Microlectronics, Grenoble INP, 2006.
Julien Brely, " Rgulation multivariable de filires de production de fibre de verre", PhD, LAG / ST Gobain Vetrotex, Grenoble INP, 2003.
Giampaolo Filardi, "Robust Control design strategies applied to a DVD-video player", PhD, LAG / ST Microlectronics, Grenoble INP, 2003.
Damien Sammier, "Modlisation et commmande de vhicules automobiles: application aux lments suspension", PhD, LAG / Grenoble INP, 2001.
Anas Fattouh, "Observabilit et commande des systmes linaires retards", PhD, LAG / Grenoble INP, 2000.
138/138
7 LPV systems
LPV Control
J.C. Doyle, K. Glover, P. Khargonekar, and B. Francis.

State space solution to standard H2 and H control problems.
IEEE Transaction on Automatic Control, 34(8):831847, december 1989.
P. Gahinet.
A linear matrix inequality approach to H control.
International Journal of Robust and Nonlinear Control, 4:421448, 1994.
C. Scherer.
The riccati inequality and state-space H -optimal control.
Ph.D. dissertation, University Wurzburg, Germany, 1990.
C. Scherer and S. Wieland.
LMI in control (lecture support, DELFT University).
2004.
C. Scherer, P. Gahinet, and M. Chilali.
Multiobjective output-feedback control via LMI optimization.
IEEE Transaction on Automatic Control, 42(7):896911, july 1997a.
C. Scherer, P. Gahinet, and M. Chilali.
Multiobjective output-feedback control via lmi optimization.
IEEE Transaction on Automatic Control, 42(7):896911, 1997b.
S. Skogestad and I. Postlethwaite.
Multivariable Feedback Control. Analysis and Design.
John Wiley and Sons, Chichester, 1996.
138/138
7 LPV systems
LPV Control
K. Zhou, J. Doyle, and K. Glover.

Robust and Optimal Control.
New Jersey, 1996.
138/138

Robust Control of MIMO Systems

Hochgeladen von

Dokumentinformationen

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

Robust Control of MIMO Systems

Hochgeladen von

Copyright:

Verfügbare Formate

Robust control of MIMO systems

Olivier Sename (Grenoble INP / GIPSA-lab)

1. The H norm and related definitions

Some useful lemmas

Wiley and Sons, 2005.

www.ece.lsu.edu/kemin, book slides available

Publishing Company, New York, 1992.

http://www.dcsc.tudelft.nl/ cscherer/. , Lecture slides available (MSc Course

Robust control in 1 slide ?

Complementary Sensitivity function

Why S is the key function in control ?

Example: an uncertain mass-spring-damper system

For performance analysis: (S( = 0) =

if output disturbance dy then, y = S.

Robustness w.r.t model uncertainties

Robust control: Find K s.t S satisfies all

Robust stability condition : | S(j) |<|

1 The H norm and related definitions

1 The H norm and related definitions

catches most of the real system phenomena.

1 The H norm and related definitions

Definition of LTI systems

1 The H norm and related definitions

Illustration of the gain computation

1 The H norm and related definitions

A first bad approach

How to define and evaluate its gain ?

1 The H norm and related definitions

MAXIMUM SINGULAR VALUE

MINIMUM SINGULAR VALUE

We see that, depending on the ratio

1 The H norm and related definitions

What about eigenvalues ?

Eigenvalues are 0 and 0.

leads to an output vector

Clearly the gain is not zero.

1 The H norm and related definitions

For MIMO systems

Example of A two-mass/spring/damper system

Hinf norm 11.4664 = 21.18 dB

Singular Values (dB)

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

1 The H norm and related definitions

LTI systems and signals norms

1 The H norm and related definitions

LTI systems and signals norms

The 2-Norm (that introduces the energy norm) is given by,

The second equality is obtained by using the Parseval identity.

kx(t)k = sup |x(t)|

kX(s)k = sup kX(j)k

1 The H norm and related definitions

LTI systems and signals norms

About vector spaces

generally called Lebesgue integral), and corresponding sequence spaces.

1 The H norm and related definitions

LTI systems and signals norms

Definition (H and RH spaces)

1 The H norm and related definitions

LTI systems and signals norms

In MATLAB, use norm(sys,inf)

obtained (e.g. Bisection algorithm, or LMI resolution).

1 The H norm and related definitions

LTI systems and signals norms

Recalls on Singular Value Definition

System type, or the maximum steady-state tracking error 0