Types of uncertainty
Parametric (real) uncertainty.
Dynamic (frequency-dependent) uncertainty
Chap 8: Robust control
Part I: Uncertainty and robustness
In parametric uncertainty the model structure and order are
known, but (some) parameter values are uncertain.
In dynamic uncertainty in the model there exists (some)
erroneous or missing dynamics. Usually unmodeled dynamics
is in high frequencies.
Parametric uncertainty: each parameter is bounded in some
region
. The parameter sets are then
with a mean parameter value
and
The scalar uncertainty term satisfies
For dynamic uncertainty representations in frequency domain
with complex perturbations becomes adequate.
(SISO case)
The term
is any stable transfer function, which at each
frequency is less than or equal to 1 in magnitude. For example,
Inverse multiplicative uncertainty
E.g. Multiplicative uncertainty
allows uncertainty near unstable poles.
Parametric uncertainty is called structured; dynamic uncertainty
is called unstructured.
Example: Consider a plant with parametric uncertainty
Uncertainty regions: Example
Disc-shaped regions can be approximated by circles.
That results in a complex additive uncertainty description.
Note: even a real uncertainty
can be approximated
by a dynamical uncertainty
.
Write
which gives
which is in the inverse multiplicative form.
Obtaining the weights:
Additive uncertainty:
Note:
is usually a rational transfer function (stable).
It is possible to present the additive
uncertainty by a multiplicative one
-Select the nominal model G
-Additive uncertainty: for all possible plants
choose
and
-Multiplicative uncertainty:
Multiplicative uncertainty description
is considered to be more informative.
When
the uncertainty exceeds
100%, and the Nyquist curve may pass
through the origin.
Solid
line
SISO Robust stability and Robust performance
structure
RS with multiplicative uncertainty
The system can be represented by
Note: possible minus sign in
the diagram does not have an
effect, when the small gain
theorem is applied.
Robust stability:
Robust performance
where
. According
to the Small Gain Theorem a sufficient
condition for stability is then
But that implies
in the below figure
The weight function
given by the designer
implies nominal performance
when
where
That has an easy geometrical interpretation.
Next, consider again a system with multiplicative uncertainty. The
system possesses robust performance, if for all plants
By geometrical inspection the condition
for robust performance becomes
General problem formulation (SISO and MIMO)
Weights included
Controlled system
Controlled system
Uncertainty included
The generalized plant from
to
is
That can be partitioned as
such that
where
Closing the loop
Here
denotes the (lower) linear fractional
transformation (LFT) of P with K as parameter.
In the control system of the previous slide
In Mixed Sensitivity Design the control goal is to minimize the
norm from w to z. That corresponds to setting weights such
that the norm of
is minimized (note that the minus signs do not have any
effect in the norm).
Stability analysis in frequency domain
Loop gain
and return difference
If L has a state-space realization
Assume that there are no unstable pole-zero cancellations.
Then, internal stability of the closed-loop system is equivalent
to the stability of
. The system matrix of S is
given by (assume that
is invertible)
which can be deduced from the equations
or
the poles of L are the roots of the open-loop characteristic
polynomial
Now for the determinant of the return difference
The closed-loop characteristic polynomial becomes
Generalized (MIMO) Nyquist theorem:
Let Pol denote the number of open-loop unstable poles of L(s).
The closed-loop system with loop transfer function L(s) and
negative feedback is stable if and only if the Nyquist plot of
By applying Schurs formula
makes Pol anti-clockwise encirclements of the origin, and the
curve does not pass through the origin.
with
we obtain
Example: SISO case
The denominator is
which is in accordance with the general result
Structured and unstructured uncertainty
General control and uncertainty models
(Block) diagonal uncertainty
Linear fractional transformations
Nyquist contour (no poles)
Typical Nyquist diagram
Structuring the system with
perturbations
To analyze robust stability of F we can rearrange the system
into an
structure, where M = N11 is the transfer function
from the output to input of the perturbations.
Each individual perturbation is assumed to be stable and
normalized
Result: for
Uncertainty structures
Robust stability and robust performance
a: additive uncertainty
b: multiplicative input
uncertainty
c: multiplicative output
uncertainty
d: inverse additive u.
e: inverse multiplicative
input u.
f: inverse multiplicative
output u.
Robust stability of the
1. Robust stability (RS): with a given controller K determine
whether the system remains stable for all plants in the
uncertainty set.
2. Robust performance (RP): if RS is satisfied, how large the
transfer function from exogenous inputs w to outputs z may
be for all plants in the uncertainty set.
NS: N is internally stable
NP: NS and
RS: NS and
is stable
RP: NS and
structure:
Result: The
structure is stable for all allowed
perturbations (RS) if and only if
Suppose that N and
are both
stable. The only cause of instability
can come from the term
But this is a positive feedback
system. Apply the generelized
Nyquist theorem!
the Nyquist plot of
the origin for any
does not encircle
