Lectures Master Robust Control

Advanced Control
An Overview on Robust Control
'
Scope
allow the student to assess the potential of different methods in

robust control without entering deep into theory. Sensitize for
the necessity of robust feedback control.
Keywords
uncertainty representations, H , synthesis, LMI
Prerequisites
Nyquist criterion, gain and phase margin, LQG state space

control
Contact
Raoul Herzog1 , Jurg Keller2
Version
5.1
Date
October 4, 2010
1 raoul.herzog@heigvd.ch
2 juerg.keller1@fhnw.ch
Advanced Control, An Overview on Robust Control
Raoul Herzog, J
urg Keller
MSE
October 4, 2010
Contents
1
Introduction to Robust Control . . . . . . . . . . . . . . . . . . . . .

1.1
Motivation of Robust Control . . . . . . . . . . . . . . . . . .
1.2
An Attempt to define Robust Control . . . . . . . . . . . . . .
1.3
Structure of this Document . . . . . . . . . . . . . . . . . . .
2
Review of Norms for Signals and Systems . . . . . . . . . . . . . . . .
3
The Nyquist Criterion and the Small Gain Theorem . . . . . . . . . .
3.1
Review of the Nyquist Criterion and Classical Stability Margins
3.2
The Small Gain Theorem . . . . . . . . . . . . . . . . . . . .
3.3
Applications of the Small Gain Theorem to Robust Control .
4
Description of Model Uncertainty . . . . . . . . . . . . . . . . . . . .
4.1
Unstructured Uncertainty . . . . . . . . . . . . . . . . . . . .
4.2
Structured Uncertainty . . . . . . . . . . . . . . . . . . . . . .
5
Formulation of the Standard H Problem . . . . . . . . . . . . . . .
6
A Glimpse on the H State Space Solution . . . . . . . . . . . . . .
7
Limitation of H Methods . . . . . . . . . . . . . . . . . . . . . . . .
8
Outlook: Synthesis and LMI Methods . . . . . . . . . . . . . . . .
8.1
Structured Singular Values (SSV) and Synthesis . . . . . . .
8.2
Linear Matrix Inequalities (LMI) . . . . . . . . . . . . . . . .
9
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
A.1 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
A.1.1 Example 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
A.1.2 Example 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
A.2 Exercise 1: Linear Fractional Transformation . . . . . . . . . . . . . .
A.3 Exercise 2: Small gain theorem . . . . . . . . . . . . . . . . . . . . .
A.4 Exercise 3: Drawback of classical stability margins . . . . . . . . . . .
A.5 Exercise 4: Two cart problem . . . . . . . . . . . . . . . . . . . . . .
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
October 4, 2010
Raoul Herzog, J
urg Keller
4
4
5
5
6
8
8
10
11
12
13
19
22
25
26
27
27
27
29
30
30
37
47
47
48
49
51
MSE
Introduction to Robust Control
1.1
Motivation of Robust Control
System modeling is generally one of the most important tasks in engineering, and
in particular in control engineering. Modeling is always a delicate task, since the
physical reality is often complicated, and all attempt to mathematically describe a
real physical process involves simplifications and assumptions, e.g.:
dynamics are often consciously neglected1 to make the model tractable, e.g.
unmodeled sensor and actuator dynamics, higher order modes from large scale
structures modeled by finite elements (FEM),
nonlinearities are often either hard to model or too complicated,
parameters are often not exactly known, either because they are hard to measure precisely, or because of varying manufacturing conditions2 .
Furthermore, it is desirable to end up with simple plant models. Indeed, in modern
control, the controller is the output of an optimization problem, and the complexity of the controller is directly linked with the complexity of the plant model: a
complicated highorder plant will automatically lead to a complicated highorder
controller, which is undesirable.
Robust control deals with system analysis and control design for such imperfectly
known process models. One of the main goals of feedback control is to maintain
overall stability and system performance despite uncertainties in the plant.
In general, robustness does not come for free from a controller designed via
optimal control and estimation theory (observer design): a controller designed for a
nominal process model generally works fine for the nominal plant model, but may
fail3 for even a nearby plant model.
An important point of all feedback control synthesis methods is the control engineers awareness of inherent tradeoffs : increasing the robustness will generally
make the controller less aggressive, and will thereby decrease system performance.
Robust control allows to specify more or less directly the plant uncertainty, and
allows to predict the possible tradeoffs between robustness and closedloop performance.
Sometimes bachelorlevel courses in control may give the impression that everything is feasible with control, and that its only a matter of finding a good controller. But this impression is completely false: the plant itself implies inherent limitations4 , especially if the plant is unstable. The achievable performance/robustness
1
Its common to say: My model is precise up to . . . Hz.

For serial production of devices incorporating feedback control, individual tuning is not desirable. The robustness of the system should be sufficient to tolerate all uncertainties and tolerances
from the manufacturing processes. For mechatronical systems, the manufacturing uncertainties
are coming from the mechanical and electronical parts. Generally, there are no manufacturing
tolerances inside the digital controller because software is perfectly reproducable.
3
Instability or unacceptable performance degradations may occur.
4
e.g. by the Bode integral relation.
2
Raoul Herzog, J
urg Keller
October 4, 2010
MSE
tradeoffs strictly depend on the plant, and cannot be overcome by any sophisticated
control [1].
1.2
An Attempt to define Robust Control
A definition of robust control could be stated as:

Robust control aims at designing a fixed (nonadaptive) controller such that some
defined level of performance5 of the controlled system is guaranteed, irrespective of
changes in plant dynamics within a predefined class.
Robust control offers a collection of powerful mathematical tools and efficient
software algorithms able to partly6 answer the following key questions:
Describing and characterizing plant uncertainty: What is a good way to describe plant variations or plant uncertainty? Some descriptions attempt to
faithfully describe the real situation (e.g. probability distributions on physical
parameters), but unfortunately there may exist no efficient method to solve the
problem. Other uncertainty descriptions are less direct, but more convenient
for the theory (e.g. analytic solutions exist).
Robustness analysis problems: As an example consider a nominal plant and a
given stabilizing controller. The uncertainty in the plant model is defined by
a class of perturbations. There is no true model, we have to deal with a
given set of possible plant models. Now, we would like to know whether or not
the closedloop remains stable for the whole class of plants. This is a typical
robustness analysis problem.
Robustness synthesis problems: Find a controller which stabilizes a given class
of plants. Generally, synthesis problems are more difficult to solve than analysis problems.
1.3
Structure of this Document
The document is structured as follows: In section 2, norms for signals and systems
are reviewed for the single input single output case (SISO). Section 3 recapitulates
the Nyquist criterion and the small gain theorem, an important working horse in
robust control. Section 4 addresses the question how to describe model uncertainty.
Sections 5 introduces the setup of H control. It is not the goal of this document to
describe the underlying mathematical theory, which is quite demanding. Therefore,
5
e.g. closedloop stability, reference tracking performance, and disturbance rejection performance.
6
It should be noted that some simple problems in robust control (e.g. exact stability determination of a linear system in which several parameters vary over given ranges) have shown to be
NPhard, hence as difficult as other famous problems for which no efficient solutions are known to
exist, or likely to be found.
October 4, 2010
Raoul Herzog, J
urg Keller
MSE
section 6 only sketches the H solution from a user point of view. Section 7 discusses
some limitations and drawbacks of standard H methods. Finally, section 8 gives
an outlook to the actual stateoftheart in robust control. Section 9 concludes
with some general remarks on robust control.
Review of Norms for Signals and Systems
Norms are used in many places of engineering to quantify the magnitude of an object
(e.g. the amplitude of a signal), or to quantify the proximity of two objects (e.g.
the proximity of two systems). In robust control, norms play a crucial role:
the choice of metric used to quantify the amount of process uncertainty,
the choice of norm used in the optimization problem associated with the controller synthesis. The linear quadratic gaussian LQG control is based on the
optimization of a ||||2 norm, whereas H control is based on the optimization
of a || || norm.
The L2 (Euclidean) norm of a timedomain scalar signal u(t) is defined as:
||u||2 =
sZ
u2 (t) dt.
(1)
If this integral is finite, then the signal u is square integrable, denoted as u L2 .

For vectorvalued signals u(t),
u(t) =
u1 (t)
u2 (t)
..
.
un (t)
(2)
the 2norm is defined as

||u||2 =
sZ
||u(t)||22 dt =
sZ
uT (t) u(t) dt,
(3)
where the superscript T denotes transposition.

A Linear TimeInvariant (LTI) system G can be described either by a state space
realization
x = Ax + Bu
y = Cx + Du
or by its corresponding transfer function
G(s) = C(sI A)1 B + D.
6
Raoul Herzog, J
urg Keller
(4)
October 4, 2010
MSE
Similar to signals, we would like to define different norms for systems, respectively for
transfer functions. Two systems G1 and G2 are close if the norm of the difference
of their transfer functions ||G1 G2 || is small. Two systems might be close with
respect to one defined norm, but far with respect to another norm7 .
The H2 norm of a stable system G(s) is defined as the L2 norm of its impulse
response g(t). When dealing with stochastic signals, the H2 norm of a system
corresponds to the variance E[||y||2 ] of the system response y(t) when the system is
excited by a unit intensity white noise input u(t).
The H norm of a stable transfer function G(s) is defined as the maximum RMS
amplification over arbitrary square integrable input signals u 6= 0.
||G|| = sup
uL2
||y||2
||u||2
(5)
Note that the H norm for a system is induced by the L2 norm for signals. The
physical interpretation of the H norm corresponds simply to the maximum energy
amplification over all input signals. It can be shown that for singleinput single
output systems ||G|| equals the peak magnitude in the Bode diagram of the transfer
function G(j ):
||G|| = sup |G(j )|
(6)
In H control the performance to be optimized is defined in terms of minimizing

the H norm of closedloop transfer functions, e.g. the sensitivity function S(s)
and the complementary sensitivity function T (s). This type of optimization problem is also called minmax problem: H control seeks to minimize the worstcase
scenario, i.e. when the closedloop function has its peak. In contrast, classical LQG
control minimizes the closedloop behaviour for known input signals8 , whereas H
control works with unknown input signals and tries to optimize the worstcase scenario. Therefore, the solution of H control problems typically inhibits flat9 Bode
magnitude plots (no more peak). When minimizing the absolute peak in the Bode
magnitude diagram, automatically other local peaks pop up. This effect is called
waterbedeffect. Finally, the value of all local peaks join the value of the absolute
peak, and the response becomes flat. The theoretical background of this comes
from the Bode integral theorem, see equation (12) and figure 6 page 12.
Instead of flat Bode magnitude plots we can also impose a given shape. This
corresponds to a modern version of classical loopshaping. In this context, important
design parameters are frequency dependent weighting functions W (s). They allow
1
With the metric induced by the H norm, the two systems P1 (s) = s+
and P2 (s) = 1s are
infinitely far, no matter how small is. This may seem illogical because a reasonable controller
will stabilize both plants P1 and P2 simultaneously, and the resulting closedloop systems will be
nearly identical. A remedy consists in using the socalled Vinnicombe metric, which also allows to
treat marginally stable and unstable systems.
8
e.g. the energy of the resulting closedloop impulse response
9
allpass behaviour
7
October 4, 2010
Raoul Herzog, J
urg Keller
MSE
Bode Diagram
2
Magnitude (abs)
10
10
10
10
Phase (deg)
45
90
135
180
2
10
10
Frequency (Hz)
Figure 1: Example: The || || norm of G(s) =
10
1
s2 +0.1s+1
is ||G|| 10.
to shape given closedloop functions, e.g. the sensitivity S(s), by optimizing their
weighted norm ||W S|| .
The H norm of a multivariable (MIMO) transfer matrix G(s) needs the important concept of singular values which is beyond the scope of this document.
The Nyquist Criterion and the Small Gain Theorem
The Nyquist criterion is a cornerstone of classical control, and is of fundamental

importance in robust control. The following chapter recapitulates the Nyquist criterion, the small gain theorem, and shows its application to robust control.
3.1
Review of the Nyquist Criterion and Classical Stability Margins
The Nyquist criterion allows to check closedloop stability based on the inspection
of the loop gain L(s) = C(s) P (s), without computing the closedloop poles, i.e.
the roots of 1 + L(s) = 0. The Nyquist criterion is based on Cauchys argument,
and says:
The closedloop system with loop gain L(s) and a negative feedback polarity is
stable if and only if the complete10 Nyquist plot of L(j) encircles the critical point
scrit = 1 exactly NP anticlockwise times in the complex plane, where NP is the
number of unstable (right halfplane) poles of L(s).
As a special case, if L(s) already is a stable open loop transfer function, NP is
zero, and the Nyquist plot of L(j) must not encircle the critical point 1 in order
10
For the complete Nyquist plot, runs from to +.
Raoul Herzog, J
urg Keller
October 4, 2010
MSE

Waterbed effect
10
minimize
the peak !
magnitude will pop up
elsewhere !
magnitude of a closedloop function
10
10
10
10
10
10
frequency Hz
Figure 2: Illustrating the waterbed effect.
loop gain

L(s)
feedback with minus polarity
Figure 3: Elementary feedback system with loop gain L(s)
to ensure closedloop stability. Figure 4 recapitulates the definition of the classical

gain margin Am and phase margin m . For example a gain margin of Am = 2 means
that the closed-loop stays stable even if the loop gain doubles. The phase margin
m indicates the amount of additional delay the feedback loop can tolerate before
becoming unstable. Robust control does not work with the classical stability margins
Am and m for two reasons: 1) there exist no analytical optimization techniques for
these margins, and 2) there are cases where the classical margins indicate a good
robustness against individual gain and phase tolerances, whereas the feedback loop
is not at all robust against simultaneous variations of gain and phase, see exercice
2, page 48. For these reasons, robust control prefers as margin the critical distance
dcrit between the Nyquist plot of L(j) and the critical point scrit = 1:
dcrit = min
(dist(L(j), scrit )) = min
|L(j) scrit | = min
|1 + L(j)|.
(7)
It follows that the critical distance dcrit is the reciprocal of the sensitivity function
October 4, 2010
Raoul Herzog, J
urg Keller
MSE
Nyquist plot of L(s)
complex plane
1
imaginary part
0.5
L(j )
L(j 0)
1/A
dcrit
0.5
1
L(j )
1.5
0.5
0.5
1.5
real part
Figure 4: Definition of the critical distance dcrit and the classical stability margins.
Here, the Nyquist plot is only drawn for positive frequencies > 0.
peak:
dcrit =
1
1
.
, where S(j) =
||S||
1 + L(j)
(8)
Maximizing the critical distance dcrit corresponds to minimizing the || || norm of

the sensitivity function. Therefore, one of the natural objectives11 of robust control
consists of minimizing the || || norm of the sensitivity function.
3.2
The Small Gain Theorem
The small gain theorem follows directly from the Nyquist criterion: if L(s) is stable
and ||L|| < 1, then the loop gain |L(j)| is smaller than 1 for all frequencies ,
and hence L(j) does not encircle the critical point 1. It follows that the small
gain condition ||L|| < 1 is a sufficient but not necessary condition for closedloop
11
It is important to see that sensitivity peak minimization is not the only objective in real world
problems. In fact, minimizing only the || || norm of the sensitivity function turns out to be an
ill-posed problem, because the gain of the resulting controller would be infinite.
10
Raoul Herzog, J
urg Keller
October 4, 2010
MSE
stability. The small gain condition corresponds to an infinite phase margin m = .

If ||L|| < 1, the closed loop stays stable even if the feedback polarity is wrong!
The small gain theorem is not limited to linear feedback control: it can even
be generalized to nonlinear feedback control. In this case, L becomes a nonlinear
operator in time domain, and the || || norm condition on L(s) must be replaced
by an operator norm condition. This may sound abstract, but a simple application
is the feedback of a linear dynamic system in cascade with a static nonlinearity. In
this case, the small gain theorem yields a sector bounded12 nonlinearity as a sufficient
condition for closedloop stability.
It makes no sense to apply the small gain theorem directly for controller synthesis: for good tracking performance a high loop gain is needed within the control
bandwidth13 . However, the small gain theorem has a great utility for analyzing
feedback loops with unstructured uncertainty.
3.3
Applications of the Small Gain Theorem to Robust Control
Suppose that P0 (s) denotes the nominal plant, and C(s) a stabilizing controller.
Now consider an additive perturbation which yields a cloud of plants P (s) =
P0 (s)+a (s), where a (s) denotes an unknown stable transfer function representing
the modeling uncertainty of P0 . The question arises how much uncertainty the
closed loop may tolerate before becoming unstable. Figure 5 shows that the feedback
perturbed plant P
'a
feedback
seen by '
additive uncertainty
'a
nominal plant +
P0
P0
-C
'a
controller
-C
1+P0C
Figure 5: Application of the Small Gain Theorem to additive plant uncertainty.

12
also called Popov criterion.

If the controller includes an integral action (e.g. the I part of PID), the loop gain is infinite at
0 Hz, and hence ||L|| = .
13
October 4, 2010
Raoul Herzog, J
urg Keller
11
MSE
C
C
. The corresponding loop gain is a 1+P
. Since both a and
seen by a is 1+P
0C
0C
the nominal closedloop are stable we can apply the small gain theorem for the
feedback loop seen by a . A sufficient condition for the stability of the perturbed
system is therefore:

C
a
< 1.
(9)

1 + P0 C
Using the submultiplicative property of norms ||AB|| ||A|| ||B||, the sufficient
stability condition becomes:

C
< 1
||a ||
1 + P0 C

1
||a || < C .
1+P C
0
(10)
Equation (10) is a conservative bound indicating the amount of tolerated additive

uncertainty which preserves stability.
In case of multiplicative unstructured uncertainty, the set of perturbed plants is
described as P (s) = P0 (s) (1 + m (s)). Using the small gain theorem, the following
stability bound can be found:
1
1
||m || < P C =
.
(11)
0
||T ||

1+P0 C
A high complementary sensitivity peak value ||T || leads to a small tolerable multiplicative perturbation m . We will now discuss an extremely fundamental relationship, called Bode integral theorem:
Z
ln |S(j)| d =
Re(p).
(12)
unstable poles p of L(s)
This law can be seen as a conservation law: the integrated value of the log of the
magnitude of sensitivity function S(j) is conserved under the action of feedback.
If the openloop L(s) is stable, then the integral becomes zero. At low frequencies,
in order to have good tracking performance, the sensitivity must be much smaller
than 1 (negative dB levels), i.e. ln |S(j)| must become negative. The Bode integral theorem states that the average sensitivity improvement at low frequencies
is compensated by the average sensitivity deterioration at high frequencies. This is
illustrated by figure 6. If the plant is unstable, the situation is becoming worse since
the right hand side of equation (12) is now positive. This means that the average
sensitivity deterioration is always larger than the improvement. The more unstable
the plant is, the more positive the real part of the unstable poles, the more difficult the situation becomes. This applies to every controller, no matter how it was
designed. Unstable plants are inherently more difficult to control than stable plants
[1].
Description of Model Uncertainty
It is worthwhile to clarify what is meant by model uncertainty. In a control system

there are two categories of uncertainties: disturbance signals and perturbations in
12
Raoul Herzog, J
urg Keller
October 4, 2010
MSE
L og Ma gnitude
10
1.0
0.1
0.0
0.5
1.0
Frequency
1.5
2.0
Figure 6: Loopshaping constraints: Sensitivity reduction at low frequencies inevitably leads to sensitivity increase at higher frequencies. Picture taken from [1].
the plant dynamics. Disturbances are external stochastic inputs which are not under
control. Examples of disturbance signals are sensor and actuator noise or changes
of the environment. Dynamic perturbations are model uncertainties caused by not
exactly known or slowly changing plant parameters, and unmodeled or approximated
system dynamics. In a house temperature control, disturbances and dynamic plant
perturbations would be :
Disturbances:
changing outdoor temperature, wind speed, open windows and doors, heating
due to electrical equipment or human bodies, sun radiation
Dynamic plant perturbations:
not exactly known isolation coefficients, unknown and changing heat capacities, uncertain efficiency of the heating device
The example shows that it is not always clear how to classify the disturbances.
Therefore, it is not surprising that in synthesis the two categories of disturbances
are handled within the same formalism. For H controller design plant uncertainty
results from approximating the real system with a mathematical model of tractable
complexity. Additionally its physical parameters are not exactly known.
As explained in the motivation section, the aim of H based controller design is
to include uncertainty into controller design. To achieve this it is necessary to have
a mathematical description of model uncertainty. In this section, two different types
of uncertainty descriptions will be introduced. The first is unstructured uncertainty,
whereas the second is structured uncertainty.
4.1
Unstructured Uncertainty
A description of model uncertainty has to meet the following goals: it should be

simpler than a physical model of the neglected system dynamics, and it should be
October 4, 2010
Raoul Herzog, J
urg Keller
13
MSE
tractable with a formalism which can easily be used in controller design. A simple
solution is found, if all dynamics, timeinvariant perturbations that may occur in
different parts of the system, are represented with a single perturbation transfer
function (s). There are many possibilities to include (s) into a control system,
but only the two most commonly used will be presented in the following. These are
shown in figures 7 and 8. The focus is on SISO systems. Figure 7 shows a (s) as
an additive perturbation of the plant transfer function P (s):
P (s) = Po (s) + a (s),
(13)
where P (s) is the actual, perturbed plant, and P0 (s) is the nominal plant.
The transfer function a (s) is used to describe a frequency dependent unstructured uncertainty as follows:
a (s)
a (s) = W2 (s)
(14)
a (s) is any stable transfer function with:

where the normalized perturbation
a || 1
||
(15)
This uncertainity description is closely related to a plant transfer function repre a || = 1 defines a circle,
sentation in C, the Nyquist plot, as shown in figure 9a). ||
14
whose radius is scaled with W2 (s).
a
Figure 7: Additive uncertainty

Dynamic perturbations can also be described with multiplicative uncertainty.
The corresponding block diagram is shown in figure 8.
P (s) = P0 (s) (1 + m (s))
(16)
In multivariable systems, transfer functions are matrices. It is known that matrix

multiplication is not commutative, so for matrixvalued P0 and m , P0 (s) (I +
m (s)) is generally different from (I + m (s)) P0 (s). As a consequence, input and
output multiplicative perturbations have to be distinguished. Input multiplicative
perturbations are able to model actuator uncertainty, whereas output multiplicative
perturbations are used to describe sensor related uncertainties.
14
14
Index 2 for W is commonly used for uncertainty weighting functions.
Raoul Herzog, J
urg Keller
October 4, 2010
MSE
Figure 8: Multiplicative uncertainty at plant input
The capabilities of additive or multiplicative uncertainty representation are illustrated with an example. For comparison purposes the unmodeled high frequency
dynamic is specified. It will be investigated how this a priori known model error can
be represented with an additive or multiplicative uncertainty.
Example:
Many plants can roughly be approximated by a first order system, e.g. PT1. Such
a model usually neglects the high frequency dynamic of the system. As it is well
known, a control system with a PT1-plant can achieve any required closed loop
performance. Consequently, we are interested in an uncertainty description which
incorporates possible unmodeled phase loss into controller design and prevents a
controller design with unrealistic high bandwidth.
The nominal plant model is
5
P0 (s) =
.
(17)
s+1
The unmodeled dynamic is represented with a set of transfer functions with the
following elements:
P (s) =
1
,
( s + 1)2
[0.02 . . . 0.05].
(18)
The plants to be controlled are:

P (s) = Po (s) P (s).
(19)
For comparison purposes the unmodeled dynamic described above is represented

as an additive and multiplicative perturbation. Additive uncertainties are best represented in the complex plane C, i.e. in the Nyquist diagram of figure (9a), whereas
multiplicative uncertainties can easily be plotted in a Bode diagram of figure (9b).
The figures show the plants frequency response P (j) for different values of . Next,
a single description of uncertainty has to be found for the specified range of values
for the uncertain parameter . To accomplish this, the additive and multiplicative
(j) are determined for a set of values of according to the following formulas:
October 4, 2010
Raoul Herzog, J
urg Keller
15
MSE
Additive unstructured perturbation:

a (s) = P (s) P0 (s).
(20)
Multiplicative unstructured perturbation:

m (s) =
P (s)
1.
P0 (s)
(21)
0.5
nominal plant
perturbed plant set
10
0
0
10
0.5
1
1
10
10
10
10
10
1.5
0
100
2.5
3
1
200
300
1
10
(a) Additive uncertainty
10
10
10
10
(b) Multiplicative uncertainty
Figure 9: Plant uncertainty

The magnitude of the perturbations a (j) and m (j) are shown in figure 10.
As defined in (14), frequency dependency is specified with the weighting transfer
function W2 (s). The solid line in the figures is an envelope of all error frequency
responses and (s) = W2 (s) (s) is therefore an upper bound for the uncertainty.
Additive weighting:
W2a (s) =
7.5s
.
(s + 1)(s + 15)
(22)
W2m (s) =
0.997s(s + 170)
.
(s + 9)(s + 135)
(23)
Multiplicative weighting:
In figure 11 the resulting uncertainty regions are plotted in the corresponding

plots.
Compare the resulting regions with the set of plants in figure 9! Obviously, the
uncertainty region is very large due to the fact, that the nominal plant is not in the
center of the plant set. Since the unnecessary uncertainty regions are mainly opposite
of the critical point, it might not have a large impact on controller design. After
= 10 rad/s the 0gain point lies within the uncertainty set. As a consequence, a
16
Raoul Herzog, J
urg Keller
October 4, 2010
MSE

1
10
10
approx. error bound

additive. errors
approx. error bound

multipl. errors
10
1
10
10
10
10
10
10
10
10
10
10
10
10
(a) Additive uncertainty
10
10
10
10
(b) Multiplicative uncertainty
Figure 10: Uncertainty descriptions

5
0.5
10
perturbed plant set(w)
0
0
10
0.5
10
10
1.5
10
10
10
10
100
0
100
2.5
200
3
1
(a) Additive bounds
300
1
10
10
10
10
10
(b) Multiplicative bounds
Figure 11: Uncertainty bounds

plant phase cannot be determined anymore. In the Bode diagram of figure phase
bounds are set to 270 and 90 degrees, in order to get a nice bode plot.
In a similar way, static nonlinearities (sector criterion) or uncertain deadtime
can be described with unstructured uncertainties.
There arises the question, which of the representations of uncertainty should be
used. Since the optimal H controller has the order of the plant plus the orders of
all the weighting functions, it is preferable to choose a representation which leads
to a minimal order weighting. In the SISO case, additive uncertainty can be recast
into multiplicative uncertainty with simple algebraic operations:
P0 (s) m (s) = a (s)
(24)
Since the additive representation includes the plant transfer function P0 (s) it is
expected, that an additive representation usually is of higher degree. To illustrate
this: uncertain deadtime can be fit into a multiplicative uncertainty description
October 4, 2010
Raoul Herzog, J
urg Keller
17
MSE
which is independent of the the plant:

P (s) = Po (s) esT
(25)
m (s) = esT 1
(26)
This function is shown in the amplitude plot of figure 12. It can be bounded with
a DT1 transfer function.
1
10
10
10
10
10
10
10
10
10
10
Figure 12: Deadtime uncertainty
The chosen uncertainty representation determines which closedloop transfer

function has to be considered in H controller design in order to guarantee robustness with respect to stability and performance.
Additive and multiplicative uncertainty are special cases of a more general framework which will be explained below. First, notice that we have to distinguish two
different feedback loops: the feedback loop formed by the controller and the perturbed plant, and the feedback loop in which the uncertainty resides. A general
solution for all different uncertainty descriptions can be obtained, if the control
system is described in the feedback structure shown in figure 13. The plant P is
partitioned according to the dimensions of .
P =
"
P11 P12
P21 P22
(27)
With simple algebraic calculations it follows:

z = [P22 + P21 (I P11 )1 P12 ] w
18
Raoul Herzog, J
urg Keller
(28)
October 4, 2010
MSE
Figure 13: Standard P configuration
if (I P11 )1 exists.
Define:
F(P, ) := P22 + P21 (I P11 )1 P12
(29)
F(P, ) is called a linear fractional transformation LFT of P and . In the single

input single output case the LFT corresponds to a bilinear transform. This representation is also used for more sophisticated uncertainty representation (next section)
and also to formulate the general H controller design problem.
Additive and multiplicative uncertainty representations are special cases of a
linear fractional transformation. For additive perturbations, the matrix P becomes:
P =
"
0 I
I P0
(30)
For multiplicative perturbations at the plant output, the matrix P becomes:

P =
"
0 P0
I P0
(31)
The example before shows that the conservative error bounds on the plant can be
large, compared to the real plant perturbation. An unstructured uncertainty model
is also not suited for perturbations which only affects a part of an interconnected
system. This motivates a more general representation of uncertainty which will be
introduced in the next section.
4.2
Structured Uncertainty
Structured uncertainty means uncertainty (tolerances) of concrete physical parameters. Some examples are listed below:
electrical components like resistors or capacitors are always affected with tolerances, e.g. 2% for a standard resistor.
In a magnetic bearing system, the nominal air gap is an important parameter
which is affected by manufacturing tolerances and by thermal growth when
the machine is running.
October 4, 2010
Raoul Herzog, J
urg Keller
19
MSE
The relative permeability r of a magnetic material is usually not very precisely

known
The internal structure of a linear plant can be represented as a block diagram with
integrators, summators, and gains. The physical parameters are gains in the block
diagram. It can be shown that the plant transfer function is always an LFT (linear
fractional transform) with respect to every physical parameter p1 , . . . , pn .
As an example, consider the uncertain plant equation (17), (18) and (19) page 15.
We will treat the time constant in P (s) as uncertain parameter lying in the range
of 0.02 . . . 0.05. Suppose that the nominal value 0 corresponds to the midrange
value 0 = 0.035, and = 0 + , where is a real parameter varying between
5
1
0.015 . . . + 0.015. The uncertain system P (s) = P0 (s) P (s) = s+1
( s+1)
2 can
be recasted with the following linear fractional transform in figure 14. Note that
diagonal uncertainty block
'
'W
'W
h
Paug
Figure 14: Structured uncertainty with an unknown time constant appearing

twice.
the uncertainty block in figure 14 is diagonal with the same diagonal element
repeated twice. The reason is that the uncertain time constant appears twice in
the second order term P (s). The augmented plant Paug (s) will be of third order
with 3 inputs and 3 outputs.
Exercise : Calculate Paug (s) and show that the associated linear fractional transform
yields
5
1
F(Paug , ) =
(32)
s + 1 ((0 + )s + 1)2
There are available software tools capable to directly define and handle systems
with structured or unstructured uncertainty. Using the Robust Control toolbox of
Matlab the example above can be entered as follows :
>> P0 = tf(5, [1, 1])
Transfer function:
20
Raoul Herzog, J
urg Keller
October 4, 2010
MSE
5
----s + 1
>> tau = ureal(tau, 0.035, range, [0.02, 0.05])
Uncertain Real Parameter: Name tau, NominalValue 0.035, Range [0.02
>> Pdelta = tf(1, [tau, 1])*tf(1, [tau, 1])
USS: 2 States, 1 Output, 1 Input, Continuous System
tau: real, nominal = 0.035, range = [0.02 0.05], 2 occurrences
>> P = P0 * Pdelta
USS: 3 States, 1 Output, 1 Input, Continuous System
tau: real, nominal = 0.035, range = [0.02 0.05], 2 occurrences
>> bode(P)
0.05]
which gives the family of Bode plot in figure 15.

Bode Diagram
50
Magnitude (dB)
50
100
Phase (deg)
150
0
90
180
270
2
10
10
10
10
Frequency (rad/sec)
10
10
Figure 15: Example : structured uncertainty with an unknown time constant .
In contrast to unstructured uncertainty, see figure 13, structured uncertainty

leads to a feedback configuration figure 15 with a static diagonal block . In addition, the diagonal terms are realvalued, whereas in the unstructured case they are
mostly complexvalued normbounded transfer functions (s).
It is obvious that structured uncertainty representations are often closer to physical specifications. However, the algorithms needed to tackle robust control analysis
or synthesis problems with structured uncertainty are much more complicated, see
outlook section page 27.
October 4, 2010
Raoul Herzog, J
urg Keller
21
MSE
We disadvise from recasting highly structured parameter uncertainty as additive

or multiplicative unstructured uncertainty, because this implies a high degree of conservativeness. Indeed, the class of perturbations may become excessively large, and
the resulting controller, if there is any complying with the robustness specifications
is likely to show poor performance.
Formulation of the Standard H Problem
In this section the controller design problem is formulated in the H formalism.

The objective is to design a controller, which does not only have nominal stability
and performance, but has this properties with all the plants represented by some
uncertainty description. H controller design was specially developped, to solve
this problem in a systematic manner. As system which is stable with all the plants
of the uncertainty set is called robustly stable. It has a robust performance, if the
performance specifications are met for all the admissible plant behaviors.
The control system with controller and uncertainty model is shown in figure 16.
In H controller design, the H norm of the mapping from w to z can be minimized. As a consequence the control problem has to be formulated in this formalism.
The inputs w are typically reference or disturbance signals, whereas the outputs z
can be the control error or the controller output. The H controller design problem
Figure 16: Control system with uncertainty
cannot directly by solved in the structure of figure 16. Since performance specifications and stability conditions can both be expressed as H norm conditions on
some transfer functions, there are two ways to simplify the control system of figure
16. Either the stability conditions are formulated in the same formalism as performance specifications, resulting in a structure as in figure 17(a), or the performance
specification is formulated similarly to the uncertainty description p (see figure
17(b)). For unstructured uncertainties the structure of figure 17(a) is easier to go,
but for structured uncertainties, only the structure in figure 17(b) leads to an elegant representation. The structured uncertainty problem is solved with -synthesis,
which is beyond the scope of this introduction. Consequently, the first approach will
be followed in the sequel.
As a result from the small gain section, it is clear that a system is robustly stable,
if
22
Raoul Herzog, J
urg Keller
October 4, 2010
MSE
(a) Robust Stability as norm condition
(b) Performance as artificial plant uncertainty
Figure 17: Possible system representations

2
10
10
10
2
10
10
10
10
10
Figure 18: Performance Weighting Function
for additive uncertainty:

||W2 C(I + P C)1 || 1
(33)
for multiplicative uncertainty:

||W2 P C(I + P C)1 || 1
(34)
In a control system as shown in figure 19 the robust stability condition corresponds to the H norm of the mapping from r to u for the additive case and from
r to y for the multiplicative case.
System performance is usually specified by conditions on disturbance attenuation, i.e. on the H norm of the mapping from d to y, or for reference tracking on
the H norm of r e. For both mappings an acceptable performance is achieved,
if the following condition on the sensitivity function S = (I + P C)1 is met:
||W1 (I + P C)1 || 1
(35)
1
.
|W1 (j)|
Consequently, a typical weighting

This condition is fulfilled, if S(j)
function W1 (s) has a shape as shown in figure 18. An integrator in W1 (s) forces
October 4, 2010
Raoul Herzog, J
urg Keller
23
MSE
the sensitivity function to be zero at = 0. The parameter < 1 is used as an

optimization parameter. The larger the better the disturbance attenuation.
Performance and robust stability can therefore be achieved, if the two conditions
are combined to the following condition (mixed sensitivity approach):
Additive uncertainty:
"
#

W1 (I + P C)1

W2,a C(I + P C)1
(36)
Multiplicative uncertainty:
"
#

W1 (I + P C)1

W2,m P C(I + P C)1
(37)
Figure 19: Control System for mixed Sensitivity Optimization

The block diagram representation of the mixed sensitivity problem is shown in
19. Here it becomes obvious, that the robust stability condition can be viewed as a
mapping of signals.
A controller can be found by solving the following optimization problem:
Additive uncertainty:
"
#

W1 (s)(I + P C)1

max
W2,a (s)C(I + P C)1
= 1,
(38)
(Similarly for the multiplicative uncertainty)
Figure 20: LFT for H optimization
24
Raoul Herzog, J
urg Keller
October 4, 2010
MSE
With some matrix algebra, the equation (38) can be transformed into a lower
linear fractional transformation F(P, C) as shown in figure 20. Therefore, the following problem has to be solved:
Find a stabilizing controller C, for which:
||F(P, C)|| := ||P11 + P12 C(I P22 C)1 P21 || = 1
(39)
Nowadays, various solutions to this problem are known. The two most important
are state space methods based on the solution of two Riccati equation and the
solution based on linear matrix inequalities (LMI). Special attention has to be payed
on additional assumptions, that are imposed for the algorithms. Some of them will
be treated in the next chapter.
A Glimpse on the H State Space Solution
H controller design problems can be formulated in different ways, but finally, the
problem has to be brought to a representation as shown in figure 20. Its state space
representation is:
x = Ax + B1 w + B2 u
z = C1 x + D11 w + D12 u
y = C2 x + D21 w + D22 u
(40)
(41)
(42)
(43)
The system matrices are usually represented as follows:

..
B2
A . B1

Gs =
..
C1 . D11 D12
.
C2 .. D21 D22
(44)
A closed optimal solution for the above general system is not yet published. For
the H problem a closed solution was first developped in [2] for the following special
case:
..
B1
B2
A .

" #
..
0
Gs =
(45)
C
.
0
1 .
I
i
. h
0 I
0
C2 ..
In a design tool the general state space description of equation (44) is transformed
into the representation of equation (45) by means of loop shifting [3]. The transformation of D21 and D12 into the form with the identity matrix, requires that the two
October 4, 2010
Raoul Herzog, J
urg Keller
25
MSE
matrices have full rank. This condition is usually violated, when weighting functions are strictly proper, i.e. have a zero Dmatrix. To circumvent this problem,
the weighting D-matrix can usually be set to a small value, without influencing the
controller design.
The conditions for existence of a solution are:
(A, B2 ) is stabilisable and (C2 , A) is detectable
D12 and D21 must be of full rank
"
A jI B2
C1
D12
has ful column rank for all
"
A jI B1
C2
D21
has ful column rank for all
Another important aspect is that the H problem must have a solution at

= . An unappropriate weighting at = can lead to unsatisfactory controller designs. For a closed loop system with plant as in equation (44) the direct
feedthrough term (gain at = ) is:
Dcl = D11 + D12 (I Dc D22 )1 Dc D21
(46)
where Dc is the controller D-matrix.

The term D11 is modified with the controller Dc -matrix. This can only be done
to a limited extent, which is depending on the sizes of D12 and D21 . Bounds are
documented in [3].
Limitation of H Methods
In this section some of the limitations of H controller design will be summarized.

The H controller design methods offers powerful tools to solve various design
problems. Nevertheless H performance measures are not always adequate
for the investigated design problem and although robustness and performance
measures are combined in the design procedure, it does not really guarantee,
that performance is robust with respect to plant uncertainty. Consequently
it was proposed to combine H robustness measures with other performance
measures, for example measures similar to the LQR-design, the mixed H2 /H
design. Also LMI (see chapter outlook) offers many possibilities for mixed
design problems.
H controller design leads to controllers of high order. A high order controller
needs more resources for its implementation and is suspect to numerical problems. Therefore it is favorable to have low order controllers. The optimal
solution for a system as described by equation ((44)) has the same order as
26
Raoul Herzog, J
urg Keller
October 4, 2010
MSE
the system itself (size of the A-matrix). The order of the system is the order
of the plant to be controlled and the sum of the orders of all the weighting
functions, see figure 19. As already mentioned in the section on uncertainty
description, the weighting functions have to be of minimal order. The problem of high order controllers can be diminished with order reduction methods.
There are several order reduction methods available. These can be applied
either on plant before H controller design or after design on the controller
itself. The designer should be cautious not to spoil the controller design with a
radical order reduction. There are also suboptimal H controller design algorithm which allow a direct design of a reduced order controller, but these are
not currently available in the commercially available controller design tools,
although a direct low order controller design is doubtless the way to go.
As it can be seen from figures 11 the uncertainty description is not tight in
the sense, that it incorporates a much larger set of plants than necessary. This
leads to a conservative controller design, which might result in very poor closed
loop performance. This problem is reduced, when a structured uncertainty
description is used in conjuction with synthesis (see Chapter Outlook).
8
8.1
Outlook: Synthesis and LMI Methods

Structured Singular Values (SSV) and Synthesis
The standard H minimizes an H norm between input signal w and output signal
z. Often these signals are vectorvalued (MIMO), as it is already the case for z in
the mixedsensitivity approach. In practice we are interested to individual transfer
functions between a scalar component wk and a scalar component zl . A large amount
of conservativeness is introduced in the design when optimizing an overall H norm.
The framework allows a more selective optimization related to structured uncertainty. A detailed description is out of the scope of this document. Basically the
framework introduces a new norm based on structured singular values which is the
new quantity to be minimized. The solution procedure is iterative (DK iteration)
and involves a sequence of minimizations, first over the controller K (holding the
scaling variable D associated with the ), and then optimizing over the scaling D
(holding the controller K fixed). The DK procedure is not guaranteed to converge
neither to the global minimum nor to a local minimum of the value, but often
works well in practice. It has been applied to a number of realworld applications
with success. These applications include vibration suppression for flexible structures,
flight control, chemical process control, and acoustic reverberation suppression in
enclosures.
8.2
Linear Matrix Inequalities (LMI)
Linear Matrix Inequalities (LMI) have emerged as powerful numerical design tools
in areas ranging from robust control (e.g. H ) to system identification.
October 4, 2010
Raoul Herzog, J
urg Keller
27
MSE
The canonical form of linear matrix inequality is any constraint of the form
M (v) = M0 + v1 M1 + v2 M2 + . . . + vn Mn < 0
(47)
where v = [v1 , . . . , vn ] is a vector of unknowns called decision variables, and M0 , . . . , Mn

are given symmetric matrices. M (v) < 0 stands for negative definite, i.e. all eigenvalues of M (v) are negative.
The scalar decision variables v1 , . . . , vn are often related to unknown matrices in
control problems, e.g. quadratic Lyapunov functions V (x) = xT P x with P > 0.
Here the entries of the symmetric matrix P correspond to decision variables. It is
common to express LMIs in matrix form instead of the scalar formulation (47).
There are three generic LMI problems:
Feasibility problem: Find a solution v to the LMI system M (v) < 0. The set
of all solutions is also called feasibility set. A simple example for an LMI
feasibility problem is the determination of stability. The system x = A x is
stable if there exist a Lyapunov function V (x) = xT P x > 0 with a negative
derivative V = AT P + P A < 0. This turns out to be an LMI problem with
an unknown matrix P . Another frequent example is a state space system with
bounded infinity norm. This turns out to be an LMI problem.
It is easy to show that the feasibility set of equation (47) is convex. Convexity
has an important consequence: even though the optimization problem has no
analytical solution in general, it can be solved numerically with a guaranteed
convergence to the solution when one exists. This is in sharp contrast to
general nonlinear optimization algorithms which may not converge towards
the global optimum.
v2
v2
v1
convex
constraint
v1
non-convex
constraint
Figure 21: Convexity
Linear objective minimization problem: Minimize the linear function cT v subject to the LMI constraint M (v) < 0.
Generalized eigenvalue minimization problem: Minimize subject to M (v) <
B(v), B(v) > 0, and C(v) < 0.
Efficient15 interiorpoint algorithms [4] are available to solve these three generic
15
However the complexity of LMI computations can grow quickly with the problem order. For
example, the number of operations required to solve a standard Riccati equation is o(n3 ) where n
is the state dimension. Solving an equivalent LMI Riccati inequality needs o(n6 ) operations!
28
Raoul Herzog, J
urg Keller
October 4, 2010
MSE
LMI problems. Many control problems and general design specifications can be
formulated in terms of LMI. The main strength of LMI formulations is the ability to
combine various convex design constraints or objectives in a numerically tractable
manner [5] [6].
A nonexhaustive list of problems addressed by LMI techniques include the following:
1. robust pole placement
2. optimal LQG control
3. robust H control
4. multiobjective synthesis
5. robust stability of systems with structured parameter uncertainty (analysis)
It is fair to say the advent of LMI optimization has significantly influenced the
direction of research in robust control. A widelyaccepted technique for numerically
solving robust control problems is to reduce them to LMI problems.
Conclusion
Robust control emerged around 1980, and the progress in theory and numerical
algorithms during the last 25 years has been enormous. Efficient commercial and
public domain software tools are available nowadays. It is possible to successfully
use these tools without understanding the deepest details of the underlying theory.
However, a minimum of theoretical knowledge is necessary, and as it is the case with
any complex scientific software tools16 , a lot of practical experience is needed before
being able to solve realworld17 problems.
In robust control, specific reasons for this are the following: Modeling is a challenging engineering task, and finding an estimation the modeling uncertainty needs
some experience. Additionally, either control specifications are not entirely known
in the beginning, i.e. incomplete, or realworld specifications may be very complex18
to such an extent that no direct synthesis procedure exist. No matter which controller synthesis method is used it remains an iterative process which also needs some
experience. In robust control, the user often needs to play around with frequency
weighting functions in order to understand the performance/robustness tradeoffs
of his specific problem.
One of the main achievements in robust control might be the consciousness about
the industrial importance of robustness. Generally speaking, system failure is not
accepted in our society, and robustness considerations will remain very important.
16
e.g. tools for finite element modeling, or any other complex scientific tool
= nonacademic
18
e.g. mixed time domain and frequency domain specifications, low controller order, etc.
17
October 4, 2010
Raoul Herzog, J
urg Keller
29
A.1
Examples
A.1.1
Example 1
MSE
H controller design will be demonstrated on the first order system with unstructured additive uncertainty, already introduced in 4.1. The nominal plant model is a
first order system, for which any closed-loop performance is achievable, if there were
no model uncertainty. The uncertainty description is given by equation (22) and
for controller design the mixed sensitivity approach will be used (see (33))19 . The
performance requirements are: zero steady-state error and maximum closed loop
bandwidth. We will investigate, how model uncertainty limits closed loop bandwidth.
The nominal plant is:
5
.
(48)
P0 (s) =
s+1
In the mixed sensitivity approach, the following design problem is solved:
"
#

W1 (s)(I + P C)1

max

W2,a (s)C(I + P C)1
1,
(49)
For additive uncertainty, the weighting function is:

W2a (s) =
7.5s
.
(s + 1)(s + 15)
(50)
What is an appropriate performance weighing W1 (s)? To answer that question,

one has to know, how the frequency response of a desirable sensitivity function
S(s) = (I + P C)1 looks like. If the above norm condition of is met, there is also
h
i

W1 (s)(I + P C)1
1,
(51)
For scalar system it is easily seen, that
h
h
i
i

(I + P C)1 (W1 (j))1 ,
(52)
The sensitivity function is bounded by the inverse of the weighting function

W1 (s). Zero steady-state error is attained if W1 (s) has a pole at = 0, i.e. if it has
an integrator.
1
(53)
W1 (s) =
s
An unsuitable selection of W1 (s) may lead to an unsolvable H controller design
problem. Remember that the sensitivity function has to meet the Bode integral
theorem (12). Consequently, also W1 (s)1 has to be in compliance with the integral
equation. The weighting (53) has the drawback, that there is now limit to possible
resonance peaks of the sensitivity function. With an additional zero, which should be
larger than the expected closed loop bandwidth, this drawback could be eliminated.
19
30
The multiplicative case can similarly be solved as exercise
Raoul Herzog, J
urg Keller
October 4, 2010
MSE
Furthermore, the H problem has to be solvable at = and the technical

conditions for solving the design problem, summarized in (5) have to be fulfilled.
To be able to solve the H controller design problem, the mixed sensitivity
problem has to be formulated as an LFT as shown in 22(a) and 22(b).
(a) System blockdiagram
(b) LFT
Figure 22: System Blockdiagram

This can be defined in MATLAB with the following m-file:
gam=2.5;
P=nd2sys(5,[1 1],1);
W1=nd2sys(1,[1 0],gam);
W1a=nd2sys([1 0],[1 16 15],7.5);
systemnames = P W1 W1a;
inputvar = [w; u];
outputvar = [W1; W1a; P];
input_to_P = [u];
input_to_W1 = [w-P];
input_to_W1a = [u];
sysoutname = add_sys_LF;
cleanupsysic = yes;
sysic
% now calculate the hinf controller
n_ctr=1;n_meas=1;gmin=1;gmax=10;tol=0.001;
[K_hin,clp] = hinfsyn(add_sys_LF,n_meas,n_ctr,gmin,gmax,tol);
After executing the m-file, which tries also to calculate the H optimal controller,
the following error message is obtained:
[a b1;c2 d21] does not have full row rank at s=0}
Despite the very convenient controller design tools, it is now necessary to derive an
algebraic expression for the system, shown in 22(b), to be able to understand the
October 4, 2010
Raoul Herzog, J
urg Keller
31
MSE
reason for the error message.
z1
z 2a)
(
W1
..
. P W1
..
. W2 a
..
.
P
"
#
(54)
We will see, that the message shows that, when setting the weighting functions,
we didnt concern about solvability of the H -Problem. Therefore, some more
problems have to be expected!
First, let us have a closer look at the system matrix add sys LF, which represents
the state space model of the LFT-plant:
..
..
. 2.5
.
0
0
0
0
1
.
.
.. 1.1 ..
2.2 0
0
0
0
..
..
0
0 15 7.7 .
0
. 3.0
..
..
0
0
1 .
0
. 0.8
Gs =

..
..
0
0.9
0
0
.
0
.
0
..
..
0 2.4 0.6 .
0
.
0
..
..
2.0
0
0
0
.
1
.
0
(55)
The pair (A, B2 ) must be stabilisable. The only unstable pole, due to the integrator in the weighting function W1 is not stabilisable, because the matrix B2 has
a zero in the corresponding row. Equivalently, it can be seen in the block diagram
(22(a)), that there is no mean to stabilize a possibly unstable transfer function W1 .
Furthermore, D21 and D12 have to be of full rank. This is fullfilled for D21 but not
for D12 . Finally, as indicated in the error message, one of the two rank conditions
is not met.
How can the design be improved, so that an H solution exists? As it can be
seen from (54) the above quantities are determined by the weighting functions W1
and W2a . The weighting functions W1 and W2a can be slightly modified, so that the
conditions are met. For the performance weight W1 , the integrator is replaced with
a stable pole close to the origin, i.e.
W1 (s) =
1
s + 105
(56)
This causes no perfect integral action, but this can be achieved by a subsequent
controller approximation.
With the chosen scaling W2a also the upper right block in (54), has a singular Dmatrix, i.e. D12 . There are two potential remedies to fix the problem: either the
32
Raoul Herzog, J
urg Keller
October 4, 2010
MSE
plant transfer function or the model uncertainty description is modified to be strictly

proper. Its reasonable to modify the plant uncertainty description by adding a small
zero to the transfer function. This causes only a neglectable effect to the uncertainty
description.
W2a (s) =
0.0075s(s + 1000)
.
(s + 1)(s + 15)
(57)
The resulting m-file is:

gam=2.5;
P=nd2sys(5,[1 1],1);
% gamma is defined in workspace
W1=nd2sys(1,[1 0.00001],gam);
W1a=nd2sys(conv([1 0],[1 1000]),[1 16 15],7.5/1000);
systemnames = P W1 W1a;
inputvar = [w; u];
outputvar = [W1; W1a; P+w];
input_to_P = [u];
input_to_W1 = [-w-P];
input_to_W1a = [u];
sysoutname = add_sys_LF;
cleanupsysic = yes;
sysic
% now calculate the hinf controller
n_ctr=1;n_meas=1;gmin=0.5;gmax=1;tol=0.001;
[K_hin,clp] = hinfsyn(add_sys_LF,n_meas,n_ctr,gmin,gmax,tol);
With this m-file, a controller is successfully calculated. The achieved value of the
H norm is 0.5. The system clp is the closed loop system. Its frequency response is
shown in Figure 23. The upper Bode-diagram shows the contribution of the performance measure W1 (s) S(s) to the H norm. Obviously, the performance measure
contributes the major part for < 5rad/s. Above that frequency the robustness
condition, shown in the lower diagram, is dominant. When is maximized, the
same behavior is observed.
The parameter is used to tune the controller. The larger, the more bandwidth
is achieved. If > 25 there exist no solution to the H -problem. In the sequel, the
design for = 2.5 and = 5 will be compared and it will become appearant, that
it might not be optimal to completely maximize . We will investigate the resulting
sensitivity function, gain and phase margin and the resulting controller frequency
response.
The comparison of the two design show the following: with = 5 the closed loop
bandwidth is increased. The cost of this is the resonance peak in the sensitivity
function. The is due to the bode integral relation, see also Figure 6. The larger
bandwidth is achieved with a controller that show double differential action, as it
can be seen in Figure 26. Such a controller will most likely not be successful in an
October 4, 2010
Raoul Herzog, J
urg Keller
33
MSE
Bode Diagram
50
To: Out(1)
100
180
90
0
To: Out(2)
Magnitude (dB) ; Phase (deg)
To: Out(1)
50
To: Out(2)
100
360
180
0
1
10
10
10
10
Frequency (rad/sec)
10
10
Figure 23: H solution
industrial control application, whereas the controller for = 2.5 has the shape of a
PIcontroller with additional low pass filter. Therefore, no practical problems will
be expected. The control system with = 5 has another drawback: the open loop
has quite a small roll-off after crossing over.
How does the control system operate for the worst case disturbance? To answer that
question the open-loop with worst case disturbance can be analyzed, see Figure 27.
For both designs, phase margin is sufficiently large and the open-loop in the highgain freuquency range is similar to the design system.
Summary
The simple design example shows, that
the unstructured uncertainty description limits the closedloop bandwidth to
the frequency range, where the plant is sufficiently well known.
stable weighting functions are powerful instruments to achieve the required
closed loop properties
special attention has to be given to the fact, that the H problem is well
defined for all frequencies, especially at = 0 and = . This can usually
be achieved with minor modifications which usually have a neglectable impact
on the final result.
34
Raoul Herzog, J
urg Keller
October 4, 2010
MSE
Bode Diagram
Bode Diagram
10
0
10
Magnitude (dB)
Magnitude (dB)
20
10
20
30
30
40
0
Phase (deg)
Phase (deg)
40
0
45
45
90
90
1
10
10
10
10
Frequency (rad/sec)
10
10
10
10
(a) = 2.5
10
10
Frequency (rad/sec)
10
10
(b) = 5
Figure 24: Sensitivity function S(s)

Bode Diagram
Gm = Inf dB (at Inf rad/sec) , Pm = 89.8 deg (at 5.26 rad/sec)
Bode Diagram
Gm = Inf dB (at Inf rad/sec) , Pm = 125 deg (at 11 rad/sec)
50
40
Magnitude (dB)
Magnitude (dB)
20
0
50
0
20
40
100
60
150
90
80
0
Phase (deg)
Phase (deg)
45
135
90
135
180
180
1
10
10
10
10
Frequency (rad/sec)
10
10
10
10
10
10
(a) = 2.5
10
Frequency (rad/sec)
10
10
10
(b) = 5
Figure 25: Margins
Bode Diagram
Bode Diagram
30
35
30
Magnitude (dB)
Magnitude (dB)
20
10
0
10
25
20
15
10
20
0
5
90
Phase (deg)
Phase (deg)
45
45
0
45
90
90
1
10
10
10
10
Frequency (rad/sec)
10
10
10
(a) = 2.5
10
10
10
Frequency (rad/sec)
10
10
(b) = 5
Figure 26: Controller

October 4, 2010
Raoul Herzog, J
urg Keller
35
MSE
Bode Diagram
Gm = 17.3 dB (at 19.6 rad/sec) , Pm = 61.9 deg (at 4.95 rad/sec)
Bode Diagram
Gm = 34 dB (at 107 rad/sec) , Pm = 72.1 deg (at 8.64 rad/sec)
100
50
Magnitude (dB)
Magnitude (dB)
0
0
100
50
100
150
200
200
250
90
Phase (deg)
Phase (deg)
300
90
180
270
360
180
270
360
1
10
10
10
10
Frequency (rad/sec)
10
10
10
(a) = 2.5
10
10
10
10
Frequency (rad/sec)
10
10
10
(b) = 5
Figure 27: Margins for worst case disturbance
36
Raoul Herzog, J
urg Keller
October 4, 2010
MSE
A.1.2
Example 2
In this example, H controller design with structured uncertainty will be demonstrated. The system shown in Figure 28 is a typical example for an industrial
rotational control problem. A DC-motor with inertia J1 drives a load inertia J2 .
The positions of the two inertias are measured and the second inertia is controlled
to a required position. The two inertia are coupled with an ideal spring, rotation is
damped with viscous friction in the bearings. The electric circuit of the DC-motor
is modeled as an RL-Element. Model uncertainty arises from changing load inertia,
varying friction and changing resistance due to temperature changes in the motor.
It is not sensible to assume that all model parameters are uncertain although this is
the reality. Every uncertain parameter adds complexity to the controller design, particularly if the controller is calculated with synthesis. The following uncertainties
will be modeled:
+ RR )
J2 = J2 (1 + J2 J2 ), c2 = c2 (1 + c2 c2 ), R = R(1
(58)
J2 , p2 and R represent the possible, relative perturbations on the corresponding

parameters. Their range is [1, 1].
Figure 28: Model for the rotational system
Modeling the system leads to the following differential equations:

Li = u Ri 1
(59)
i = J1 1 + c1 1 + k(1 2 )
(60)
k(1 2 ) = J2 2 + c2 2 + d
(61)
d is a disturbance torque acting upon the load inertia. With some algebraic
manipulations the equations are reordered as follows:
1
i = (u Ri )
L
October 4, 2010
Raoul Herzog, J
urg Keller
(62)
37
1 =
MSE
1
(c1 1 k(1 2 ) + i)
J1
(63)
1
(k(1 2 ) c2 2 d)
(64)
J2
For solving the structured uncertainty design problem, the system must be transformed into the form shown in Figure 13. This can be a tedious task if the procedure
is not formalized in such a way, that it can be done with a suitable tool. A closer
look to the above equation reveals, that the uncertain parameters are either factors
on variables, e.g. Ri and c2 2 , or the inverse of the parameter is a factor on an
algebraic term, e.g. J12 (k(1 2 ) c2 2 d). The idea to formalize the model
construction process is, to define new variables for the algebraic terms and to represent the uncertain parameter as LFT models. Afterwards a CACSD-Tool (CACSD:
Computer aided control system design) can interconnect the model an build its state
space representation.
2 =
The term
1
J2
is represented as an upper LF T in J2 as follows:

1
1
=
J2 (1 + J2 J2 )
J2
1
J2
= J2 (1 + J2 J2 )1
J2
J2
= F(MJ2 , J2 )
"
#
J2 J12
MJ 2 =
J2 J12
(65)
The term c2 = c2 (1 + c2 p2 ) is represented as an upper LF T in p2 as follows:

c2 = F(Mc2 , c2 ), Mc2 =
"
0 c2
c2 c2
(66)
+ RR ) is represented as an upper LF T in R as follows:

and R = R(1
R = F(MR , R ), MR =
"
0 R
R R
(67)
The LFTs block diagram are shown in Figure 29. In the figure also input variables
are defined, if necessary. All together, the system model is:
1
1
1
i = pR 1 + u
L
L
L
pR = F(MR , R )i
1 = 1
2 = 2
38
Raoul Herzog, J
urg Keller
(68)
October 4, 2010
MSE
Figure 29: LFT of uncertain parameters
1 =
2
p J2
f J2
pc2
=
=
=
=
1
1
1
c1 1 k1 + k2 + i
J1
J1
J1
p J2
F(MJ2 )fJ2
k(1 2 ) pc2 d
F(Mc2 )2
Its state space representation is:
i
1
2
1
2
1
2
2
i
f J2
0 L1
J11 c1
0
0
0
1
0
0
0
0
0
1
1
0
0
0
0
0
0
0
1
0
0
0
0
0 J11 k
0
0
0
0
1
0
0
1
k
J1
0
0
0
1 0
0 1
0 0
0 0
k k
i
1
2
1
2
i
1
2
1
2

0

+ 0

0
0 0
0 0
0 0
0 0
0 1
0 L1
0 0
0 0
0 0
0 0
1
L
0
0
0
0
0
0 0
0 0
0 0
0 0
0 1
0
0
1
0
0
0
0
0
0
0
u
d
pr
p J2
pc2
u
d
pr
p J2
pc2
(69)
Together with the system equations (69) the plant is now completely defined. We
let MATLAB calculate the interconnections of all the subsystems. In Figure 31 the
build process is visualized. In the variable systemnames all block are specified, the
system output names are listed in the variables inputvar and outputvar and the
interconnection are defined with input to . . . variables. The numerical parameter
values are documented in table 1
The resulting set of plant frequency responses are drawn in Figure 30.
For optimizing performance the approach shown in Figure 17(b) is chosen, i.e.
performance is formulated with a fictitious uncertainty block p , also called the
performance uncertainty block. Consequently, performance has to be expressed as
an input-output mapping, where performance is guaranteed when its norm is less
than 1. Figure 32 shows the block diagram of the system to be optimized and its
representation with a fictitious uncertainty block is drawn in 33 The controller inputs
are the position error of the load inertia and the position of the drive inertia. Both
October 4, 2010
Raoul Herzog, J
urg Keller
39
name
J1
J2
c1
c2
k
R
L
MSE
nominal
0.1 kg m2
1.5 kg m2
0.002 N ms/rad
0.003 N ms/rad
100 N m/rad
18
2.1 103 H
0.1 N m/A
variance
0
50%
0
10%
0
10%
0
0
Table 1: Parameter values

set of plants
100
50
50
Magnitude (dB)
Magnitude (dB)
set of plants
100
0
50
100
0
50
100
150
150
200
90
200
0
45
Phase (deg)
Phase (deg)
180
270
360
90
135
180
450
225
3
10
10
10
10
Frequency (rad/sec)
10
10
10
(a) u to 1
10
10
10
10
Frequency (rad/sec)
10
10
10
(b) u to 2
Figure 30: Closed-loop frequency response

measurements are corrupted with measurement noise and the plant is disturbed with
torque changes d on the second inertia. Performance outputs are the control error
and controller output u, both weighted with a suitable transfer function.
Again, one has to be very careful to define a sensible H optimization problem,
i.e. the problem has to be well defined for all frequencies, especially at = .
Technically, this follows from the full rank condition on the matrices D12 and D21 .
Consequently, the mapping from any input to the two controller input must have a
direct feed-through term of rank two. Both, the perturbation inputs and the torque
disturbance d have zero direct feed-through terms to the position measurement. The
reference signal r and the measurement noise n1 and n2 can contribute to a full rank
matrix D12 , if the measurement noise weightings are chosen to be strictly propper.
This becomes obvious, if the control system is drawn as an LFT as shown in Figure
33. A full rank matrix D21 , which is the direct feed-through term from the control
signal u to the system outputs, is guaranteed, if the weighting Wu is strictly proper.
Beside its technical need, weighting functions provide useful tuning knobs for the
controller design. Often, the plant is a good low pass filter, and consequently the
spectra of the measurements lack of high frequency content. This might lead to a
controller with very large gain at high frequency. Applied to the true plant with
40
Raoul Herzog, J
urg Keller
October 4, 2010
MSE
Figure 31: System model
inevitable measurement noise, the actuator will be heavily stressed and will not have
a large life-time. As a remedy, the measurement noise weighting function offer the
possibility to add high frequency content to the measurements spectra. The spectra
of load disturbances are modeled with the weighting Wd . This physically motivated
weighting introduces low frequency disturbances into the system.
Disturbance attenuation and tracking error are tuned with the control error
weight We . The function will be similarly to W1 of example 1. With Wu closedloop bandwidth can be influenced, because it can be used to limit controller gain in
around the crossing-over frequency.
The weight are chosen as follows:
0.085 (s + 10)
s + 105
0.000018 (s + 1)
=
s + 0.001
We =
Wd
October 4, 2010
Raoul Herzog, J
urg Keller
(70)
41
MSE
Figure 32: Control structure
0.003 (s + 1)
s + 100
= W 2
0.1s
=
s + 100
W 1 =
W 2
Wu
(71)
The frequency responses are shown in Figure 34.
Now, all is defined for the controller design. In MATLAB, a system can be build
according to Figure 33 and an H controller can be calculated. If the block diagonal
structure of the uncertainty matrix is not taken into account, the minimal value
of the H -Norm is 15.9 and the resulting control system is unstable. This shows,
that for structured uncertainties, the simple H approach is not a good choice. To
circumvent the problem -synthesis was developed. As described in 8.1 this involves
optimization of the H -Norm over all possible frequency dependent scalings of the
structured uncertainty . Using D-K-Iteration, the controller, described in the sequel is obtained.
The controller frequency response is shown in Figure 35. The mapping of the control
error r 2 to the input u is shown on the left. On the right the frequency response
of 1 to u drawn. The controller is of order 22 and exhibits large gain at frequency
above 1 rad/s. The frequency responses cannot be easily be approximated with a
simple PIDT1-controller. The closed-loop frequency responses from the setpoint r
to 1 and 2 can be analyzed in Figure 36 and the corresponding step responses are
in Figure 37.
Analysis of the closed-loop performance shows, that the attained bandwidth is
not very large. This is due to the large controller gain, necessary to achieve a crossing over frequency above 2 rad/s. Due to the control signal weight Wu this can not
become arbitrarily large. In order to achieved larger bandwidth, the noise weightings Wn or Wu have to be relaxed.
M-File for plant
42
Raoul Herzog, J
urg Keller
October 4, 2010
MSE
Figure 33: Control system as LFT
%Parameters
omega=logspace(-1,2,100);
J1=0.1;J2=1.5;DJ2=0.5;c1=0.01;c2=0.02;Dc2=0.1;k=100;R=18;DR=0.1;L=0.0021;Phi=0.1;
Ap=[0
-Phi/L 0 0
0;
Phi -c1/J1 0 -k/J1 k/J1;
0
0
0 0
0;
0
1
0 0
0;
0
0
1 0
0];
Bp=[1/L 0 -1/L 0 0;
0
0 0 0 0;
0
0 0 1 0;
0
0 0 0 0;
0
0 0 0 0];
Cp=[0 0 0 1 0;
0 0 0 0 1;
0 0 1 0 0;
1 0 0 0 0;
1 0 0 k -k];
Dp=[0 0 0 0 0;
0 0 0 0 0;
0 0 0 0 0;
0 0 0 0 0;
0 -1 0 0 -1];
P=pck(Ap,Bp,Cp,Dp);
mat_R=[0 R;DR R];
mat_J2=[-DJ2 1/J2;-DJ2 1/J2];
October 4, 2010
Raoul Herzog, J
urg Keller
43
MSE
60
30
40
40
50
20
60
70
80
20
40
4
10
90
2
10
10
10
10
100
4
10
10
Wphi1 and Wphi2
10
10
10
Wu
50
20
40
60
60
70
80
80
100
90
100
4
10
120
2
10
10
10
10
140
4
10
10
10
10
10
Figure 34: Weighting functions
mat_c2=[0 c2;Dc2 c2];

gam=2.5;
systemnames = P mat_R mat_J2 mat_c2;
inputvar = [uR; uJ2; uc2; d; u];
input_to_P = [u; d; mat_R(2); mat_J2(2); mat_c2(2)];
input_to_mat_R = [uR; P(4)]; %P(4) is current i
input_to_mat_J2 = [uJ2; P(5)]; %P(5) is fJ2
input_to_mat_c2 = [uc2; P(3)]; %P(3) is omega_2
outputvar = [mat_R(1); mat_J2(1); mat_c2(1); P(1); P(2)];
%Phi2 for performance, all disturbance outputs, phi1 and phi2 for control
sysoutname = Ex2_Plant_LF;
cleanupsysic = yes;
sysic}
M-File to define control system with weghtings
% now calculate the hinf controller system
gam = 0.085;
W_e=nd2sys([1 10],[1 0.00001],gam);
W_d=nd2sys(0.0018*[0.01 0.01],[1 0.001]);
W_phi1=nd2sys(0.003*[1 1],[1 100]);
W_phi2=W_phi1;
W_u=nd2sys([0.1 0],[1 100]);
systemnames = Ex2_Plant_LF W_e W_d W_phi1 W_phi2 W_u;
inputvar = [uPert{3}; dw; n1; n2; r; u];
input_to_Ex2_Plant_LF = [uPert; W_d; u];
input_to_W_e = [r-Ex2_Plant_LF(5)-W_phi2];
input_to_W_d = [dw];
input_to_W_phi1 = [n1];
input_to_W_phi2 = [n2];
input_to_W_u = [u];
outputvar = [Ex2_Plant_LF(1:3); W_u; W_e; r-Ex2_Plant_LF(5)-W_phi2; Ex2_Plant_LF(4)+W_phi1];
sysoutname = Ex2_Sys_LF;
cleanupsysic = yes;
sysic
M-File for -controller design
44
Raoul Herzog, J
urg Keller
October 4, 2010
MSE

Bode Diagram
From: In(1)
From: In(2)
100
To: Out(1)
60
40
20
0
20
540
360
To: Out(1)
Magnitude (dB) ; Phase (deg)
80
180
0
180
360
0
10
10
Frequency (rad/sec)
Figure 35: Controller frequency response
% now calculate the hinf controller system

DK_DEF_NAME = dk_mds;
dkit
K_mu = k_dk3mds;
% Analysis with nominal plant
Ex2_Plant_nom=sel(Ex2_Plant_LF, 4:5, 4:5);
figure(2), clf, bode_dm(K_mu,omega)
systemnames = Ex2_Plant_nom K_mu;
inputvar = [dw; n1; n2; r];
input_to_Ex2_Plant_nom = [dw; K_mu];
input_to_K_mu = [r-Ex2_Plant_nom(2)-n2; Ex2_Plant_nom(1)+n1];
outputvar = [Ex2_Plant_nom(1:2); K_mu];
sysoutname = Ex2_Closed_Loop;
cleanupsysic = yes;
sysic
M-File with parameter for -controller design
NOMINAL_DK = Ex2_Sys_LF;
NMEAS_DK = 2;
NCONT_DK = 1;
BLK_DK = [-1 1; -1 1; -1 1; 4 2]
OMEGA_DK = logspace(-2, 3,100);
AUTOINFO_DK = [1 3 1 4*ones(1, size(BLK_DK,1))];
NAME_DK = mds;
See MATLAB manual for the robust-control toolbox for a description of the
parameters.
October 4, 2010
Raoul Herzog, J
urg Keller
45
MSE
set of closed loop systems
set of closed loop systems
50
20
Magnitude (dB)
Magnitude (dB)
0
0
50
20
40
60
100
80
150
0
100
0
90
Phase (deg)
Phase (deg)
90
180
270
180
270
360
450
360
2
10
10
10
Frequency (rad/sec)
10
10
10
10
10
Frequency (rad/sec)
(a) r to 1
10
10
(b) r to 2
Figure 36: Closed-loop frequency response
Step Response
1.4
1.2
1.2
Amplitude
Amplitude
Step Response
1.4
0.8
0.8
0.6
0.6
0.4
0.4
0.2
0.2
6
8
Time (sec)
10
12
14
(a) r to 1
6
8
Time (sec)
10
12
14
(b) r to 2
Figure 37: Closed-loop step response
46
Raoul Herzog, J
urg Keller
October 4, 2010
MSE
A.2
Exercise 1: Linear Fractional Transformation
Transforming a system into a LFT is a frequent task in controller design. Find an

LFT representation (Figure 39) for the plant with a 2 degree of freedom controller
as shown in Figure 38
Figure 38: 2 Degree of Freedom control system
Figure 39: 2 Degree of Freedom control system
A.3
Exercise 2: Small gain theorem
A high order controller C(s) is approximated with a reduced order controller Cred (s).
The control loop with the reduced-order controller remains unchanged, if the original
controller C(s) is added and subtracted in parallel to the reduced order controller.
This is shown in Figure 40. With this simple block diagram manipulation, we now
have the original control loop, which is perturbed with (s) = Cred (s) C(s).
1. Closed-loop stability should be maintained. Use the small-gain theorem to
derive a condition on the norm of in order to guarantee closed loop stability
for the system with the reduced-order controller.
2. In which frequency range the controller approximation should be best? Use
the previously derived condition to analyse control systems for plants with or
without integrator.
October 4, 2010
Raoul Herzog, J
urg Keller
47
MSE
Figure 40: Controller approximation
A.4
Exercise 3: Drawback of classical stability margins
Consider a feedback system with the following loop gain

L(s) =
0.38(s2 + 0.1s + 0.55)

.
s(s + 1)(s2 + 0.06s + 0.5)
The feedback system exhibits very good gain and phase margins, but is not at all
robust with respect to simultaneous uncertainties of gain and phase.
1. Calculate the gain margin Am and the phase margin m .
2. Determine the peak of the sensitivity ||S|| .
3. Determine the distance between the Nyquist plot of loop gain L(j) and the
critical point 1.
This example shows that it is much better to assess robustness using ||S|| instead
of using traditional gain and phase margins. A low value of ||S|| , e.g. implies good
gain and phase margins while the converse is not true. It follows from elementary
trigonometric relations that
Am >
||S||
1
, m > 2 arcsin
||S|| 1
||S||
For example ||S|| = 2 gives a garanteed gain margin of Am > 2, corresponding to

a worst case gain margin of 6 dB, and a phase margin m > 60 .
48
Raoul Herzog, J
urg Keller
October 4, 2010
MSE
A.5
Exercise 4: Two cart problem
Consider the following mechanical plant: two carts are moving on a horizontal plane
with no friction. The control force w is acting on the left cart with mass m1 , and the
position of the right cart z = x2 is measured. The carts are connected by a linear
spring k. The nominal parameters are m1 = 1 kg, m2 = 1 kg, and k = 1 N/m.
All parameters are affected by independent tolerances of 10% for the masses, and
20% for the spring.
z = x2
x1
position
measurement
control
force
w
m1
m2
spring
k
Figure 41: Two cart ACC benchmark problem
diagonal uncertainty
block
'
h
g
augmented plant
Figure 42: Structured uncertainty setup
1. Find a state space realization of the augmented plant P, where the structured
October 4, 2010
Raoul Herzog, J
urg Keller
49
MSE
parameter uncertainty is represented as a diagonal feedback block
1 0 0
= 0 2 0
.
0 0 3
The uncertainty 1 should correspond to the uncertainty of m1 , and a variation
of 1 < 1 < +1 should correspond to a variation of 10% of m1 around its
nominal value. Similarly, 2 , 3 correspond to the uncertainties of m2 , k, and
both should also be normalized to a range of 1.
2. Why is it inappropriate to formulate the uncertainties of this system as additive
or multiplicative uncertainties?
3. In the plant model, no damping is considered. This leads to a pole pair on the
imaginary axis. Explain, why this leads to severe controller design problems,
if the pole location changes due to plant uncertainty.
50
Raoul Herzog, J
urg Keller
October 4, 2010
Bibliography
[1] G. Stein. Respect the unstable. IEEE Control Systems Magazine, August 2003.
[2] J. Doyle, K. Glover, P.P. Khargonekar, and B.A. Francis. Statespace solutions
to standard H2 and H control problems. IEEE Trans. Aut. Control, 34(8),
August 1989.
[3] M. Safonov and D. Limebeer. Simplifying H theory via loop shifting. Proc.
27th Conference on Decision and Control, December 1988.
[4] Y. Nesterov and A. Nemirowski. Interior Point Polynomial Methods in Convex Programming: Theory and Applications. Studies in Applied Mathematics,
SIAM books Philadelphia, 1994.
[5] S. Boyd, L. El Ghaoui, E. Feron, and V. Balakrishnan. Linear Matrix Inequalities in Systems and Control Theory. Studies in Applied Mathematics, SIAM
books Philadelphia, vol. 15, 1994.
[6] P. Gahinet and P. Apkarian. A linear matrix inequality approach to H . Int.
J. Robust and Nonlinear Control, 4, 1994.
[7] M. Green and D. Limebeer. Linear Robust Control. Pearson Education, Inc.,
2002.
[8] J. Doyle, B. Francis, and A. Tannenbaum. Feedback Control Theory. McMillan
Publishing Co., 1990.
[9] Robust Control Toolbox 3, Users Guide, The Mathworks, 2008.
[10] LMITOOL: A Package for LMI Optimization in Scilab Users Guide, 1995.
[11] D. Gu, P. Petkov, and M. Konstantinov. Robust Control Design with MATLAB.
Springer, 2005.
October 4, 2010
Raoul Herzog, J
urg Keller
51

Lectures Master Robust Control

Enviado por

Dados do documento

Direitos autorais

Formatos disponíveis

Compartilhar este documento

Compartilhar ou incorporar documento

Opções de compartilhamento

Você considera este documento útil?

Este conteúdo é inapropriado?

Direitos autorais:

Formatos disponíveis

Lectures Master Robust Control

Enviado por

Direitos autorais:

Formatos disponíveis

Advanced Control

An Overview on Robust Control

allow the student to assess the potential of different methods in

uncertainty representations, H , synthesis, LMI

Nyquist criterion, gain and phase margin, LQG state space

Raoul Herzog1 , Jurg Keller2

Advanced Control, An Overview on Robust Control

Introduction to Robust Control . . . . . . . . . . . . . . . . . . . . .

Advanced Control, An Overview on Robust Control

Introduction to Robust Control

Motivation of Robust Control

Its common to say: My model is precise up to . . . Hz.

Advanced Control, An Overview on Robust Control

An Attempt to define Robust Control

A definition of robust control could be stated as:

Structure of this Document

Advanced Control, An Overview on Robust Control

Review of Norms for Signals and Systems

If this integral is finite, then the signal u is square integrable, denoted as u L2 .

the 2norm is defined as

uT (t) u(t) dt,

where the superscript T denotes transposition.

Advanced Control, An Overview on Robust Control

In H control the performance to be optimized is defined in terms of minimizing

Advanced Control, An Overview on Robust Control

Figure 1: Example: The || || norm of G(s) =

The Nyquist Criterion and the Small Gain Theorem

The Nyquist criterion is a cornerstone of classical control, and is of fundamental

Review of the Nyquist Criterion and Classical Stability Margins

For the complete Nyquist plot, runs from to +.

Advanced Control, An Overview on Robust Control

magnitude of a closedloop function

Figure 2: Illustrating the waterbed effect.

feedback with minus polarity

Figure 3: Elementary feedback system with loop gain L(s)

to ensure closedloop stability. Figure 4 recapitulates the definition of the classical

Advanced Control, An Overview on Robust Control

Nyquist plot of L(s)

Maximizing the critical distance dcrit corresponds to minimizing the || || norm of

The Small Gain Theorem

Advanced Control, An Overview on Robust Control

stability. The small gain condition corresponds to an infinite phase margin m = .

Applications of the Small Gain Theorem to Robust Control

Figure 5: Application of the Small Gain Theorem to additive plant uncertainty.

also called Popov criterion.

Advanced Control, An Overview on Robust Control

Equation (10) is a conservative bound indicating the amount of tolerated additive

unstable poles p of L(s)

Description of Model Uncertainty

It is worthwhile to clarify what is meant by model uncertainty. In a control system

Advanced Control, An Overview on Robust Control

A description of model uncertainty has to meet the following goals: it should be

Advanced Control, An Overview on Robust Control

a (s) is any stable transfer function with:

Figure 7: Additive uncertainty

In multivariable systems, transfer functions are matrices. It is known that matrix

Index 2 for W is commonly used for uncertainty weighting functions.

Advanced Control, An Overview on Robust Control

Figure 8: Multiplicative uncertainty at plant input

The plants to be controlled are:

For comparison purposes the unmodeled dynamic described above is represented

Advanced Control, An Overview on Robust Control

Additive unstructured perturbation:

Multiplicative unstructured perturbation: