Professional Documents
Culture Documents
4l PDF
4l PDF
'
Prerequisites Nyquist criterion, gain and phase margin, LQG state space
control
Version 5.1
1 raoul.herzog@heig–vd.ch
2 juerg.keller1@fhnw.ch
Advanced Control, An Overview on Robust Control MSE
trade–offs strictly depend on the plant, and cannot be overcome by any sophisticated
control [1].
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 state–of–the–art in robust control. Section 9 concludes
with some general remarks on robust control.
• the choice of norm used in the optimization problem associated with the con-
troller 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.
u1 (t)
u2 (t)
u(t) = .. . (2)
.
un (t)
ẋ = Ax + Bu
y = Cx + Du
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.
||y||2
||G||∞ = sup (5)
u∈L2 ||u||2
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 single–input single–
output systems ||G||∞ equals the peak magnitude in the Bode diagram of the transfer
function G(j ω):
||G||∞ = sup |G(j ω)| (6)
ω
Bode Diagram
2
Magnitude (abs) 10
1
10
0
10
−1
10
0
−45
Phase (deg)
−90
−135
−180
−2 −1 0
10 10 10
Frequency (Hz)
1
Figure 1: Example: The || · ||∞ norm of G(s) = s2 +0.1s+1
is ||G||∞ ≈ 10.
to shape given closed–loop 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 im-
portant concept of singular values which is beyond the scope of this document.
Waterbed effect
1
10
minimize
the peak !
magnitude will pop up
elsewhere !
0
10
magnitude of a closed−loop function
−1
10
−2
10
−3
10
−1 0
10 10
frequency Hz
loop gain
L(s)
It follows that the critical distance dcrit is the reciprocal of the sensitivity function
complex plane
0.5
imaginary part
−1 L(j ∞) L(j 0)
0
1/A
m
dcrit
−0.5
φ
m
−1
L(j ω)
Figure 4: Definition of the critical distance dcrit and the classical stability margins.
Here, the Nyquist plot is only drawn for positive frequencies ω > 0.
peak:
1 1
dcrit = , where S(jω) = . (8)
||S||∞ 1 + L(jω)
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.
The small gain theorem follows directly from the Nyquist criterion: if L(s) is stable
and ||L||∞ < 1, then the loop gain |L(jω)| is smaller than 1 for all frequencies ω,
and hence L(jω) does not encircle the critical point −1. It follows that the small
gain condition ||L||∞ < 1 is a sufficient but not necessary condition for closed–loop
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.
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 P0 +
-C
nominal plant +
P0 +
'a
controller
C -C
1+P0C
12
also called Popov criterion.
13
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||∞ = ∞.
−C −C
seen by ∆a is 1+P 0C
. The corresponding loop gain is ∆a · 1+P 0C
. Since both ∆a and
the nominal closed–loop 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 =⇒ ||∆a ||∞ < C . (10)
1 + P0 C ∞ 1+P C
0 ∞
A high complementary sensitivity peak value ||T ||∞ leads to a small tolerable multi-
plicative perturbation ∆m . We will now discuss an extremely fundamental relation-
ship, called Bode integral theorem:
Z ∞ X
ln |S(jω)| dω = π Re(p). (12)
0 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(jω) is conserved under the action of feedback.
If the open–loop 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(jω)| must become negative. The Bode in-
tegral 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 diffi-
cult 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].
10
L og Ma gnitude
1.0
0.1
0.0 0.5 1.0 1.5 2.0
Frequency
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
The example shows that it is not always clear how to classify the disturbances.
Therefore, it is not surprising that in µ–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.
tractable with a formalism which can easily be used in controller design. A simple
solution is found, if all dynamics, time–invariant 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):
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 unstruc-
tured 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)
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:
1
P∆ (s) = , τ ∈ [0.02 . . . 0.05]. (18)
(τ s + 1)2
P (s)
∆m (s) = − 1. (21)
P0 (s)
0.5
nominal plant 10
5
0
10
−0.5
−1
−1 0 1 2 3
10 10 10 10 10
−1.5
0
−2
−100
−2.5
−200
−3 −300
−1 0 1 2 3 4 5 −1 0 1 2 3
10 10 10 10 10
The magnitude of the perturbations ∆a (jω) and ∆m (jω) 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:
7.5s
W2a (s) = . (22)
(s + 1)(s + 15)
Multiplicative weighting:
0.997s(s + 170)
W2m (s) = . (23)
(s + 9)(s + 135)
0 1
10 10
approx. error bound approx. error bound
additive. errors multipl. errors
0
10
−1
10
−1
10
−2
10
−2
10
−3 −3
10 10
−1 0 1 2 3 −1 0 1 2 3
10 10 10 10 10 10 10 10 10 10
5
0.5 10
perturbed plant set(w)
0
0
10
−0.5
−5
−1 10
−1 0 1 2 3
10 10 10 10 10
−1.5 100
0
−2
−100
−2.5
−200
−3 −300
−1 0 1 2 3 4 5 6 −1 0 1 2 3
10 10 10 10 10
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 dead–time
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:
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 dead–time can be fit into a multiplicative uncertainty description
1
10
0
10
−1
10
−2
10
−3
10
−1 0 1 2 3
10 10 10 10 10
h
g
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 represen-
tation 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:
" #
0 I
P = . (30)
I P0
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.
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 τ in P∆ (s) as uncertain parameter lying in the range
of 0.02 . . . 0.05. Suppose that the nominal value τ0 corresponds to the mid–range
value τ0 = 0.035, and τ = τ0 + ∆τ , where ∆τ is a real parameter varying between
5 1
−0.015 . . . + 0.015. The uncertain system P (s) = P0 (s) · P∆ (s) = s+1 · (τ s+1) 2 can
be recasted with the following linear fractional transform in figure 14. Note that
Paug
w z
the uncertainty block ∆ in figure 14 is diagonal with the same diagonal element ∆τ
repeated twice. The reason is that the uncertain time constant τ 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 ((τ0 + ∆τ )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 :
5
-----
s + 1
Bode Diagram
50
0
Magnitude (dB)
−50
−100
−150
0
−90
Phase (deg)
−180
−270
−2 −1 0 1 2 3
10 10 10 10 10 10
Frequency (rad/sec)
cannot directly by solved in the structure of figure 16. Since performance specifi-
cations 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 perfor-
mance 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 ele-
gant representation. The structured uncertainty problem is solved with µ-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
(a) Robust Stability as norm condition (b) Performance as artificial plant uncer-
tainty
1
10
0
10
−2 −1 0 1 2
10 10 10 10 10
Multiplicative uncertainty:
" #
γW1 (I + P C)−1
. (37)
W2,m P C(I + P C)−1
∞
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 fol-
lowing problem has to be solved:
||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.
ẋ = Ax + B1 w + B2 u (40)
z = C1 x + D11 w + D12 u (41)
y = C2 x + D21 w + D22 u (42)
(43)
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:
..
A . B1 B2
··· ··· "· · ·#
.. 0
Gs = C . (45)
1 . 0
I
. h
C2 ..
i
0 I 0
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 transfor-
mation of D21 and D12 into the form with the identity matrix, requires that the two
matrices have full rank. This condition is usually violated, when weighting func-
tions are strictly proper, i.e. have a zero D–matrix. 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:
7 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.
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 prob-
lem 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 algo-
rithm 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.
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 eigen-
values 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 LMI’s 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 ẋ = A x is
stable if there exist a Lyapunov function V (x) = xT P x > 0 with a negative
derivative V̇ = 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 v1
convex non-convex
constraint constraint
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 non–exhaustive list of problems addressed by LMI techniques include the fol-
lowing:
3. robust H∞ control
4. multi–objective synthesis
It is fair to say the advent of LMI optimization has significantly influenced the
direction of research in robust control. A widely–accepted technique for numerically
solving robust control problems is to reduce them to LMI problems.
9 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 real–world17 problems.
In robust control, specific reasons for this are the following: Modeling is a chal-
lenging 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 real–world specifications may be very complex18
to such an extent that no direct synthesis procedure exist. No matter which con-
troller 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 trade–offs
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
17
= non–academic
18
e.g. mixed time domain and frequency domain specifications, low controller order, etc.
A.1 Examples
A.1.1 Example 1
H∞ –controller design will be demonstrated on the first order system with unstruc-
tured 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 band-
width.
The nominal plant is:
5
P0 (s) = . (48)
s+1
In the mixed sensitivity approach, the following design problem is solved:
" #
γW1 (s)(I + P C)−1
max
γ
W2,a (s)C(I + P C)−1
≤ 1, (49)
∞
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:
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
0.0075s(s + 1000)
W2a (s) = . (57)
(s + 1)(s + 15)
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 perfor-
mance measure W1 (s) ∗ S(s) to the H∞ –norm. Obviously, the performance measure
contributes the major part for ω < 5rad/s. Above that frequency the robustness
condition, shown in the lower diagram, is dominant. When γ is maximized, the
same behavior is observed.
The parameter γ is used to tune the controller. The larger, the more bandwidth
is achieved. If γ > 25 there exist no solution to the H∞ -problem. In the sequel, the
design for γ = 2.5 and γ = 5 will be compared and it will become appearant, that
it might not be optimal to completely maximize γ. 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 γ = 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
Bode Diagram
To: Out(1)
−50
−100
Magnitude (dB) ; Phase (deg)
180
To: Out(1)
90
0
To: Out(2)
−50
−100
360
To: Out(2)
180
0
−1 0 1 2 3 4
10 10 10 10 10 10
Frequency (rad/sec)
industrial control application, whereas the controller for γ = 2.5 has the shape of a
PI–controller with additional low pass filter. Therefore, no practical problems will
be expected. The control system with γ = 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 high-
gain freuquency range is similar to the design system.
Summary
The simple design example shows, that
• special attention has to be given to the fact, that the H∞ –problem is well
defined for all frequencies, especially at ω = 0 and ω = ∞. This can usually
be achieved with minor modifications which usually have a neglectable impact
on the final result.
0 10
0
−10
Magnitude (dB)
Magnitude (dB)
−10
−20
−20
−30
−30
−40 −40
0 0
Phase (deg)
Phase (deg)
−45 −45
−90 −90
−1 0 1 2 3 4 −1 0 1 2 3 4
10 10 10 10 10 10 10 10 10 10 10 10
Frequency (rad/sec) Frequency (rad/sec)
50 40
20
0
Magnitude (dB)
Magnitude (dB)
−50 −20
−40
−100
−60
−150 −80
−90 0
−45
Phase (deg)
Phase (deg)
−135 −90
−135
−180 −180
−1 0 1 2 3 4 5 −1 0 1 2 3 4 5
10 10 10 10 10 10 10 10 10 10 10 10 10 10
Frequency (rad/sec) Frequency (rad/sec)
30 35
30
20
Magnitude (dB)
Magnitude (dB)
25
10
20
0
15
−10
10
−20 5
0 90
45
Phase (deg)
Phase (deg)
−45 0
−45
−90 −90
−1 0 1 2 3 4 −1 0 1 2 3 4
10 10 10 10 10 10 10 10 10 10 10 10
Frequency (rad/sec) Frequency (rad/sec)
100 50
0
0
Magnitude (dB)
Magnitude (dB)
−50
−100 −100
−150
−200
−200
−300 −250
−90 −90
Phase (deg)
Phase (deg)
−180 −180
−270 −270
−360 −360
−1 0 1 2 3 4 5 −1 0 1 2 3 4 5
10 10 10 10 10 10 10 10 10 10 10 10 10 10
Frequency (rad/sec) Frequency (rad/sec)
A.1.2 Example 2
In this example, H∞ –controller design with structured uncertainty will be demon-
strated. 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, par-
ticularly if the controller is calculated with µ–synthesis. The following uncertainties
will be modeled:
1
i̇ = (u − Ri − Φω) (62)
L
September 19, 2011 Raoul Herzog, Jürg Keller 37
Advanced Control, An Overview on Robust Control MSE
1
ω˙1 = (−c1 ω1 − k(ϕ1 − ϕ2 ) + Φi) (63)
J1
1
ω˙2 = (k(ϕ1 − ϕ2 ) − c2 ω2 − d) (64)
J2
For solving the structured uncertainty design problem, the system must be trans-
formed 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 ω2 , or the inverse of the parameter is a factor on an
algebraic term, e.g. J12 (k(ϕ1 − ϕ2 ) − c2 ω2 − d). The idea to formalize the model
construction process is, to define new variables for the algebraic terms and to repre-
sent 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.
1
The term J2
is represented as an upper LF T in δJ2 as follows:
1 1
= ¯ (65)
J2 J2 (1 + ∆J2 δJ2 )
1 ∆J2
= ¯ − ¯ δJ2 (1 + ∆J2 δJ2 )−1
J2 J2
= F(MJ2 , δJ2 )
−∆J2 J¯12
" #
MJ 2 =
−∆J2 J¯12
1 1 1
i̇ = − pR − Φω1 + u
L L L
pR = F(MR , δR )i
ϕ̇1 = ω1
ϕ̇2 = ω2 (68)
1 1 1
ω˙1 = − c1 ω1 − kϕ1 + kϕ2 + Φi
J1 J1 J1
ω̇2 = p J2
p J2 = F(MJ2 )fJ2
fJ2 = k(ϕ1 − ϕ2 ) − pc2 − d
pc2 = F(Mc2 )ω2
0 − L1 Φ 1
0 − L1
i̇ 0 0 0 i L
0 0 u
ω̇1
Φ − J11 c1 0 − J11 k 1
J1
k
ω1 0
0 0 0 0
d
ω̇2 =
0 0 0 0 0
ω2 + 0
0 0 1 0
pr
ϕ̇1
0 1 0 0 0
ϕ1 0
0 0 0 0
p J2
ϕ̇2 0 0 1 0 0 ϕ2 0 0 0 0 0 pc2
ϕ1 0 0 0 1 0 i 0 0 0 0 0 u
ϕ2 0 0 0 0 1 ω1 0 0 0 0 0 d
ω2 =
0 0 1 0 0
ω2 +
0 0 0 0 0
pr
i
1 0 0 0 0
ϕ1
0 0 0 0 0
p J2
fJ2 1 0 0 k −k ϕ2 0 −1 0 0 −1 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
100 100
50 50
Magnitude (dB)
0 Magnitude (dB) 0
−50 −50
−100 −100
−150 −150
−200 −200
−90 0
−45
−180
Phase (deg)
Phase (deg)
−90
−270
−135
−360
−180
−450 −225
−3 −2 −1 0 1 2 3 −3 −2 −1 0 1 2 3
10 10 10 10 10 10 10 10 10 10 10 10 10 10
Frequency (rad/sec) Frequency (rad/sec)
(a) u to ϕ1 (b) u to ϕ2
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 ω = ∞.
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
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 closed-
loop 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)
We =
s + 10−5
0.000018 (s + 1)
Wd = (70)
s + 0.001
0.003 (s + 1)
W ϕ1 =
s + 100
W ϕ2 = W ϕ2
0.1s
Wu =
s + 100
(71)
%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];
W W
e d
60 −30
−40
40
−50
20 −60
0 −70
−80
−20
−90
−40 −100
−4 −2 0 2 4 −4 −2 0 2 4
10 10 10 10 10 10 10 10 10 10
−40
−60
−60
−70
−80
−80
−100
−90
−120
−100 −140
−4 −2 0 2 4 −4 −2 0 2 4
10 10 10 10 10 10 10 10 10 10
Bode Diagram
80
60
To: Out(1)
40
Magnitude (dB) ; Phase (deg)
20
−20
540
360
To: Out(1)
180
−180
−360
0 0
10 10
Frequency (rad/sec)
NOMINAL_DK = Ex2_Sys_LF;
NMEAS_DK = 2;
NCONT_DK = 1;
BLK_DK = [-1 1; -1 1; -1 1; 4 2]
NAME_DK = ’mds’;
See MATLAB manual for the robust-control toolbox for a description of the
parameters.
50 20
0
0
Magnitude (dB)
Magnitude (dB)
−20
−50 −40
−60
−100
−80
−150 −100
0 0
−90
−90
Phase (deg)
Phase (deg)
−180
−180
−270
−270
−360
−450 −360
−2 −1 0 1 2 −2 −1 0 1 2
10 10 10 10 10 10 10 10 10 10
Frequency (rad/sec) Frequency (rad/sec)
(a) r to ϕ1 (b) r to ϕ2
1.4 1.4
1.2 1.2
1 1
Amplitude
Amplitude
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
0 0
0 2 4 6 8 10 12 14 0 2 4 6 8 10 12 14
Time (sec) Time (sec)
(a) r to ϕ1 (b) r to ϕ2
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.
3. Determine the distance between the Nyquist plot of loop gain L(jω) 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
||S||∞ 1
Am > , φm > 2 arcsin
||S||∞ − 1 ||S||∞
x1 z = x2
position
measurement
control
force
w
m1 m2
spring
k
diagonal uncertainty
block
'
h g
augmented plant
w P z
1. Find a state space realization of the augmented plant P, where the structured
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.
[1] G. Stein. Respect the unstable. IEEE Control Systems Magazine, August 2003.
[2] J. Doyle, K. Glover, P.P. Khargonekar, and B.A. Francis. State–space 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.
[7] M. Green and D. Limebeer. Linear Robust Control. Pearson Education, Inc.,
2002.
[10] LMITOOL: A Package for LMI Optimization in Scilab User’s Guide, 1995.
[11] D. Gu, P. Petkov, and M. Konstantinov. Robust Control Design with MATLAB.
Springer, 2005.