440
Example 2
Controller
H
2
393 7 225
H
inf
180 5 132
Example 3
Controller
H
2
7720 723 1104
H
inf
5120 444 633
5 REMARKS
− The fully coupled, simulation model of the ship
with acceptable accuracy gives possibilities to
perform the identification trials instead of costs
and time consuming full-scale experiments. One
can the build the multidimensional linear model
and estimate the system uncertainties: their rang-
es and sources, based on the results from simula-
tion runs.
− The introduction of parametric uncertainties into
the plant model enables to cover the changes of
object characteristics (even nonlinear) in the all
range of assumed work conditions. On the other
hand it causes the increasing difficulty in the con-
troller synthesis.
− Very important advantage (or attribute) of both
regulators is its fixed structure and constant val-
ues of coefficients. It means that navigators do
not need to adjust any coefficients of these con-
trollers.
− The H
2
controller works worse than the robust
one. One can compare tables with results for con-
trol quality and steering effort. One of the main
reasons for such a steering can be the lack of the
robust properties of the regulator (see the H
inf
norm of this regulator).
− Both systems were tested in the presence of a
medium level of wind, in spite of fact that exter-
nal disturbances were not taken into account dur-
ing controllers synthesis processes. The robust
regulator still seems to be a better one in such
work conditions. The external disturbances one
can try to introduce into the controller synthesis
process but often no enough adequate regulator is
obtained (eg. without robust properties).
− As one can see in Fig. 12 - Fig.19, the steering is
almost de-coupling despite the full matrices B, C
and D in the controllers.
− The both closed-loop systems are stable under all
tested work conditions.
− The most important problems are related to yaw
steering (especially for H
2
controller). One of the
possible sources was the gyrocompass (with its
accuracy 0.2[deg]) and one was the fact that the
training ship is high weatherly.
− In general regulator calculated for one ship can
not be transferable to another one due to linear
object model specified for particular ship. It is a
similar situation like with PID controllers in
many industrial processes. But the possibility of
using a simulation model of the ship’s dynamics
instead a real ship for experiments for H
2
or H
inf
robust controller synthesis seems to be a great
advantage of described approach.
REFERENCES
Anderson, B. & Moore, J. (2005), Optimal Filtering, Dover
Publications, UK.
Balas, G., Doyle, J., Glover, K., Packard, A. & Smith, R.
(2001), μ-Analysis and Sythesis Toolbox. Ver.4, The
Mathworks Inc., Natick, USA.
Doyle, J. (1982), ‘Analysis of feedback systems with struc-
tural uncertainties’, IEE Proc., Part D 129(6), 242–250.
Doyle, J., Glover, K., Khargonekar, P. & Francis, B. (1989),
‘State-space solutions to standard H2 and H∞ control prob-
lems’, IEEE Trans. on Automatic Control 34(8), 831–847.
Fossen, T. (2002), Marine Control Systems, Marine Cybernet-
ics, Trondheim, Norway.
Gierusz, W. (2001), Simulation model of the ship handling
training boat ,,Blue Lady”, in ‘Int. IFAC Conference Con-
trol Applications in Marine Systems CAMS’01’, Glasgow,
Scotland.
Gierusz, W. (2006), ‘The steering of the ship motion - a μ-
synthesis approach’, Archives of Control Sciences 16/1, 5–
27.
Gierusz, W. & Tomera, M. (2006), ‘Logic thrust allocation ap-
plied to multivariable control of the training ship’, Control
Engineering Practice 14, 511–524.
Niederlinski, A., Moscinski, J. & Ogonowski, Z. (1995), Adap-
tive control (in polish), PWN, Warszawa, Poland.
Redheffer, R. (1960), ‘On a certain linear fractional transfor-
mation’, Journal of Mathematical Physics 39, 269–286.
Skogestad, S. & Postlethwaite, I. (2003), Multivariable Feed-
back Control - Analysis and Design, John Wiley and Sons,
Chichester, UK.
Zhou, K. (1998), Essentials of Robust Control, Prentice Hall,
Upper Saddle River, USA.