Professional Documents
Culture Documents
All Righls Reserved. No part of this publication may be reproduced, stored in a retrieval system or transmitted in any form or by any means:
electronic. electrostatic, magnetic tape, mechanical, photocopying, recording or otherwise, without permission in writing from the copyrighl
holders.
A catalogue record for this book is available from the Library of Congress
ISBN 0-08-043236 0
ISSN 1474-6670
These proceedings were reproduced from manuscripts supplied by the authors, therefore the reproduction is not completely uniform but neither
the format nor the language hove been changed in the interests of rapid publication. Whilst every effort is made by the publishers to see that
no inaccurate or misleading data, opinion or statement appears in this publication, they wish to moke it clear that the data and opinions
appearing in the articles herein are the sole responsibility of the contributor concerned. Accordingly, the publisher, editors andtheir respective
employers, officers and agents accept no responsibility or liability whatsoever for the onsequences of any such inaccurate or misleading data.
opinion or statement.
Elsevier Science welcomes enquiries concerning publishing proposals: books, journal special issues, conference proceedings, etc.
All formats and media can be considered. Should you have a publishing proposal you wish to discuss, please contact; without
obligation, the publisher responsible for Elsevier's industrial and control engineering publishing programme:
Dr Martin Ruck
Publishing Editor
Elsevier Science Ltd
The Boulevard, Langford Lane Phone: +44 1865843230
Kidlington, Oxford Fax: +44 1865843920
OX5 1GB, UK E.mail: m.ruck@e1sevier.co.uk
General enquiries, including placing orders, should be directed to Elsevier's Regional Sales Offices - please access the
Elsevier homepage for full contact details (homepage details at the top of this page).
IFAC CONFERENCE ON CONTROL APPLICATIONS IN
MARINE SYSTEMS 2001
Sponsored by
International Federation of Automatic Control (IF AC)
Co-sponsored by
Technical Committees on:
- Transportation Systems
- Intelligent Autonomous Vehic1es
Advanced Control Technology Club
Organized by
Industrial Control Centre, Department of Electronic and Electrical Engineering, University of
Strathclyde, Scotland
The IFAC Conference on Control Applications in Marine Systems (CAMS 2001) was successfully
held on I8-20th of July 2001 at the University of Strathclyde in Glasgow, Scotland. More than 120
delegates from over 20 countries attended the Conference and discussed and exchanged their research
results in marine control systems design and applications. CAMS 2001 attracted the highest number of
papers and delegates compared to the previous ones. This shows the importance and the increasing
trend in the control design and application in marine systems. The papers were presented in the areas
of Autopilot Systems, Dynamic Positioning Systems, Tracking Systems, Roll Stabilization, High
Speed Crafts, Underwater Vehicles, Robotic Off-shore Systems, Ship Navigation Systems, ship
propulsion, Traffic Guidance and Control Systems, Engine and Machinery Control Systems, Safety
and Fault Tolerant Control, and Training and Vehicle Simulation.
The three plenary sessions were well attended and presented by Mogens Blanke, Danish Technical
University, who reviewed recent advances on maritime safety using fault tolerant control systems. Ken
Cartledge, Rolls Royce, UK, presented the second plenary on 'In pursuit of lean manning: ship
automation and the value of simulation' and discussed the importance of simulation in reducing
manning in ship automation from an industrial viewpoint. The third plenary was presented by John
Moscopoulos, US Navy on Ship automation from the US Navy perspective and summarized the
requirement for new areas of research in marine control. The three semi-plenary papers were presented
by Ikuo Yamamoto, Mitsubishi Heavy Industry, Japan, who discussed the result of their research on
flexible fin propulsion and robotic fish. Gregory Provan, Rockwell Automation, USA, presented a
motivating talk on Dynamically reconfigurable and scalable distributed shipboard automation system
for improving sustainability and survivability and A.J. Haywood, Maritime Dynamics Inc., presented a
talk on various aspects of ride control systems that have been developed for high-speed ferries over the
last decade.
There were also three invited sessions on underwater vehicle control and high speed ship control
which gave an excellent review of the current work in these areas.
The papers presented in CAMS200I Conference gave an excellent view both of the state-of-the-art as
well as insight into the future of control in marine systems.
Dr Reza Katebi
Industrial Control Center
Dept of Electronic and Electrical Engineering
University of Strathclyde
Glasgow, Scotland
September 2001
Email: r.Katebi@eee.strath.ac.uk
CONTENTS
PLENARY PAPER I
An Improved Numerical Model for the Study of Controlled Ship Motions in Extreme Following and Quartering Seas 27
Z. AYAZ, K.J. SPYROU, D. VASSALOS
v
Assisted Dynamic Positioning System for a FPSO Based on Minimization of a Cost Function 107
E.A. TANNURI, C.P. PESCE, D.C. DONHA
SEMI-PLENARY PAPER I
Research on Flexible Oscillating Fin Propulsion System and Robotic Fish 119
!. YAMAMOTO, Y. TERADA
Observer-Based FDI for Gain Fault Detection in Ship Propulsion Benchmark - A Geometric Approach 125
T.F. LOOTSMA, R. IZADI-ZAMANABADI, H. NIJMEUER
A Nonlinear State-Space Model of Diesel Propulsion Plant Operation Using Neural Nets 131
N.!. XIROS, N.P. KYRTATOS
Marine Engine Room Monitoring and Control System for Simulating Real Processes on a Ship 137
R. SMIERZCHALSKI
A 2-D Visual Servoing Technique for Underwater Vehicle Station Keeping 143
J.-F. LOTS, D.M. LANE, E. TRUCCO, F. CHAUMETTE
AUVS' Dynamics Modeling, Position Control, and Path Planning Using Neural Networks 167
H. SAYYAADI, T. URA
vi
On the Design of Alerts for Future Platfonn Management Systems 203
M.A. TAINSH
PLENARY PAPER II
In Pursuit of Lean Manning: Ship Automation and the Value of Simulation 221
KJ. CARTLEDGE
Genetic Method of Optimization of Evacuation Ways in Cases the Fire Growth at Ferryboat. 233
D. LOZOWICKA, P. NIKONCZUK
A Genetic Algorithm for Autonomous Underwater Vehicle Route Planning in Ocean Environments with Complex
Space-Time Variability 237
A. ALVAREZ, A. CAITI
On the Design of an IGV-Based Servo System for Air Cushion Vessels 249
S.M. SAVARESI, D. BERTIN, S. BITT ANTI
Experimental Study of Controlled Flaps and T-Foil for Comfort Improvement of a Fast Ferry 261
J.M. GIRON-SIERRA, S. ESTEBAN, B. De ANDRES, J.M. DIAZ, J.M. RIOLA
Decreasing of the Motion Sickness Incidence by a Multivariable Classic Control for a High Speed Ferry 273
J. ARANDA, J.M. DiAz, P. RUIPEREZ, T.M. RUEDA, E. LOPEZ
Control Code Generator Used for Control Experiments in Ship Scale Model 279
OR POLO, S. ESTEBAN, A. MARON, L. GRAU, J.M. De la CRUZ
A Sensor Fusion Scheme for Autonomous Underwater Vehicle Localization in Cable Following 291
A. BALASURIY A, T. URA
VII
Control of Limit Cycles for Underwater Vehicle Systems Using Describing Functions 297
Y.J. HUANG, Y.-J. WANG
Combined Trajectory Tracking and Path Following for Underwater Vehicles 303
P. ENCARNA<;::AO, A. PASCOAL
02CA2: A New Hybrid Control Architecture for a Low Cost AUV 311
P. RIDAO, M. CARRERAS, J. BATLLE, J. AMAT
SEMI-PLENARY PAPER II
Shipboard System Diagnostics and Reconfiguration Using Model-Based Autonomous Cooperative Agents 323
S. CHill, G. PROVAN, Y.-L. CHEN, F. MATURANA, S. BALASUBRAMANIAN, R. STARON, D. VASKO
Wave Spectrum Correction with the Ship's Speed and the Incidence Angle 331
V. NICOLAU, E. CEANGA
"
On Board Artificial Stable Platform System Free from Roll and Pitch Motions 337
J. KAY ANO, K. OHTSU
High Performance PI(D) Controller on Ship Steering with a Conditional Reset Windup 363
R. FERREIRO GARCIA, C.A.F. AMEAL
Design ofPID Controllers for Ship Course Keeping Using Approximations to H_-Solutions 369
S. RUMY ANTZEV, A. MIROSHNIKOV, E. POPOV A
The Study of H_ Control theory on Ship Lift Feedback Fin Stabilizer 401
ZHANG XIAOYU, JIN HONGZHANG, LI GUOBIN, II MING
viii
Relay-Based Process Identification in the Frequency Domain 505
Y.G. WANG, W.J. CAI
Algorithms for the Ship Trajectory Planning and Its Tracking in the Collision Avoidance Process 523
Z. ZWIERZEWICZ
x
Copyright © IF AC Control Applications in Marine Systems,
Glasgow, Scotland, UK, 2001
Mogens Blanke *
1
The effects of faults critical to ships' safety are in the plant and in the automation system to make
required assessed by legislation of all countries "intelligent" software that monitors behavior of
within the International Maritime Organization components and function blocks. Faults are iso-
(IMO) since the start of this millennium. This lated, and appropriate remedial action is taken to
means there is available information to support prevent that faults develop into critical failures.
analysis of faults and their consequences. This The overall FTC strategy is to maintain plant
paper contributes with an overview of methods availability high and accept reduced performance
that can help design automation systems to diag- when critical faults occur.
nose and handle faults and contribute to enhanced
Fault-tolerance can be obtained as follows. A fault
safety.
diagnosis schemes operates on-line. A discrete
The paper focus on diagnosis and on-line handling event signal to a supervisor-agent is generated
of faults whereas diagnosis for maintenance is not when a fault is detected. This in turn activates
within the scope. An overview is first presented of fault handling, which can be predetermined for
tools to analyze structure and other fundamental each type of critical fault, or can be obtained
properties of an automated system in normal from real-time analysis and optimization. Figure
and in faulty conditions. The aim is to find 1 illustrates a plant as a set of constraints, e.g.
inherent redundancy and utilize this to maintain differential equations, the structure of which is de-
availability if faults occur. Methods for diagnosis termined by the state of an automaton. The logic
of faults are subsequently discussed, emphasizing states can change as a consequence of changes in
essential properties for hard real-time use. Two continuous state or of the occurrence of faults.
examples illustrate the methodology, one is sensor A fault-tolerant control system employs diagno-
fusion for navigation, the other is main engine sis and autonomous supervision for handling of
propulsion control. Bibliographic references are faults. Reconfiguration (Blanke et at., 2000) will
included in a final section of the paper. change the use of interface between plant and
control system as a consequence of a new (faulty)
condition that has been diagnosed. Accommoda-
2. REQUIREMENTS AND SAFETY tion would change controller parameters but not
the interface with the plant.
An essential requirement in manoeuvring of a ship
or operation of a machinery is that the transient
state that might follow due to a fault should not
exceed a specified limit. Such requirements are
vessel and operation dependent, extremes being
a fast ferry at 50 kt operating in narrow waters,
an off-shore supply vessel dynamically positioned
close to a rig, a container vessel at ocean passage.
Requirements specify the allowed erroneous ship
motion: a fault in the navigation or automatic
steering control should never cause a heading de-
viation larger than 5 degrees; a fault in equipment
servicing or controlling the main engine should
never cause an over speed shut-dow of the engine;
ships' speed should not exceed the desired by
more than 4 kt.; ships' position should remain
within a virtual box of dimension 0.7 (ship length,
beam); manoeuvrability could be required intact Fig. 1. A plant is a hybrid entity. Changes in
or recovered within 15 seconds, as an example. state or faults can cause changes in structure
or in parameters of the constraints that de-
scribe the plant. Diagnosis and autonomous
3. FAULT-TOLERANT CONTROL supervisor functions implement appropriate
handling when a fault occurs.
Automation for safety-critical applications, where
neither function failure nor performance degrada-
tion could be tolerated, normally requires redun- 3.1 Detailed Requirements
dant hardware to facilitate fault recovery. Fail-
operational systems are insensitive to any single Requirements to limit ship erroneous motion are
point component failure. Fail-safe systems make closely related to the effect of the fault and the
controlled shut-down to a safe state when a sensor structure of the control loop. The impact factor
measurement indicates a critical fault. In contrast, of the fault is a function of control loop structure
fault-tolerant control systems, employ redundancy and properties of the faulty system, magnitude of
2
the fault, time to detect and time to reconfigure. failure of parts of a system is a classical disci-
An important parameter is the time to detect that pline and the Failure Mode and Effects Analysis
a fault has occurred, and the impact the fault has (FMEA) is widely used and appreciated in the
during this time. marine industry. The traditional FMEA does not
support the analysis of fault handling, only their
• If a component fault is abrupt and the di-
propagation. In automated systems, 'the goal of
agnosis method is used for operator assisted
fault-tolerance is to continue operation, if this is at
change of operational mode, the detection
all possible. An extended method for Fault Prop-
time can be seconds to minutes.
agation Analysis (FPA) can be employed using
• If abrupt faults in setpoint values to a closed
an algebraic approach for propagation analysis.
loop control are used by a controller for au-
The aim of the FP A is to show the end effects
tomatic accommodation, the time to detect
of faults. It can also assist fault tolerant design
the fault can be within five to ten samples.
so that severe consequences are avoided (if the
• If abrupt faults in a feedback element in a
system structure makes this possible). If the FPA
closed loop control occur, time to detect and
analysis finds that serious failure can occur, due
reconfigure is one to two samples.
to certain faults, these can be included in a list
The categorization above indicated there is a need of fault effects to be detected. Whether this is
to distinguish between the time-wise development possible is disclosed in a later analysis of struc-
of a fault (abrupt or incipient) and the way in ture, that shows which redundant information is
which the information is used (from maintenance available in the system. The final step is to de-
to autonomous accommodation). A total fault termine actions, preferably within the software of
management system which may use robust design the controller, that can accommodate the fault
methods provides fault tolerant control. Process and prevent the serious failures from occurring. To
faults are only considered here as their effects determine whether this will be possible is a part of
propagate through the control system and the the analysis procedures for the quality measures
ways in which the control system can alter their used to judge the possible recovery of the system.
propagation.
3
Abstract
This paper discusses a controller architecture that allows continuous
improvement, upgrading and expansion of complex control systems. Its success
has been proven in practice with the development of a DpDt system for
dredgers. The paper discusses the merits of the approach both for Industry and
for University. Copyright © 2001 IFAC
21
Delft University of Technology and the Royal those gains). If such a relation exists, it will be
Netherlands Navy) from scratch to sea acceptance possible for a designer to quickly add modifications
test took 9 years (Van der Klugt, 1991). Nevertheless, or even replace this part by some alternative.
as shown by the development of a DpDt system for Otherwise, too much time will be lost to verify the
hopper dredgers, it can be done, conditionally. impact of the change on the overall design. Also, it
One of these conditions is the focus of this paper: the allows him to start simple, say with a PID control
selection of an appropriate controller architecture. algorithm for controlling a specific motion, to be
replaced with a better alternative as soon as that is
This paper presents a controller architecture that has convenient. Finally, it enables co-designing; i.e. parts
been proven to be flexible enough are supplied by a third party and merely integrated.
• to realise diverse ship control systems such as an An often-overlooked problem with complex control
RRS autopilot and a DpDt (Dynamic positioning / systems is, that if there is a problem encountered in
Dynamic tracking) system for different types of practice, it will be difficult to find. Having
ships such as frigates, LPDs and hopper dredgers, recognisable parts with familiar signals that can be
• to allow embedding of parts that are supplied by monitored (such as rudder, heading and filtered
third parties, heading to verify the impact of the rudder on the
• to reduce the development time of complex ship heading or the heading on the filtered heading) is a
motion control systems substantially and great help for finding the root of the problem.
• to realis~ a performance that will be difficult to • The architecture is built from parts that have
match wIth other approaches. simple alternatives in today's technology.
22
In this diagram, the following holds: The filter / estimator structure resembles a Kalman
• It comprises a number of sensors (heading, roll, filter structure. It shows model-based estimation of
rudder angle, etc.) providing the data required to the ship motions using 'known' actuator and sensor
calculate (filtering / estimation) the signals upon inputs. From these ship motion estimates, sensor
which a controller has to act. estimates are derived. The difference between
• The block 'control' calculates actions that are to estimate and actual value is used to update the filter
be distributed over the available actuators. states.
• An allocation algorithm distributes the control When regarding the above diagram, one has to be
outputs over the available actuators aware of some simplifications that would otherwise
• Filters, controllers and allocation algorithms are clutter the diagram:
adjuste? according to operator requirements given • The consequence of having coloured noise on
the avai!able sensors en actuators. sensors is not shown neither is the colouring of
• Data i~ ~xchanged by means of some system noise other than a constant disturbance. In
commumcatiOn layer . practice, these add 'states' to the system as well as
• Alarm management provides data about the state coefficients that describe the colouring.
of the system and its components to the operator • Some sensors (dashed text) need coordinate
and for co~figuration ~anagement p~oses. transformation to bring them to the ship
• The blo~k configuratiOn management selects the coordinates. In theory, it is irrelevant whether a
appr~pnate sens~rs, actuato~s an~ control comparison between transformed sensor and
algonthms dependmg on what is possible (for a estimate or between sensor and transformed
part indicated by Alarm Management) and what is estimate is used. In practice, one solution may be
requested (by the operator). more convenient.
• The 'adjust' blocks can be very simple (such as
changing the course control accuracy upon The word :optimal' cannot really be used with
operator request) or may comprise complex respect to this filter. The force model may have non-
adaptation schemes (such as to fmd optimal gains linear coefficients that prohibit the calculation of
with changing weather conditions. The same 'optimal' update gains. In practice this is not a
applies for the other blocks; they can be very serious drawback as the variances of the system noise
simple or so complex that they have to be built and measurement noise are generally not known.
from a network of building blocks themselves. Satisfactory settings can be found by following some
rules of thumb from control theory with respect to the
The actual implementation is such that the actuators relation between time constants and the underlying
and sensors can be either the actual systems on board process. However, as indicated in Fig. I, the structure
a ship or simulated alternatives. That way, most does allow provisions for optimising such gains
testing can be executed in a (design friendly) office (block 'adjust filter / estimator').
environment.
The state-space domain provides filter and controller The above structure has been used in several
solutions that match naturally the indicated realisations ofImtech's DpDt control system:
architecture. This is illustrated with the filter design • As part of the RRS autopilot for the M-class
as used within the baseline DpDt system of Imtech frigate, the structure provides estimated roll (-
Marine & Industry. rate), heading (-rate) and waterspeed. The force
23
Fig. 3 shows a simplified model of the impact of The allocation algorithm will automatically adjust
rudder and stabiliser setpoints on the roll moment and the actuator setpoints if heading control is done
the yaw moment. It is valid for a ship with one rudder by the operator. If access to the rudder is denied,
and one stabiliser set (i.e. two fin stabilisers). The it will use the rudder position to calculate the
gains kop ... koo-2 are considered to be known stabiliser setpoints. Or, if that is allowed by the
coefficients (part of 'force model' shown in Fig. 2) operator, it may add a stabilisation component to
that may be non-linear or may change in time. the rudder setpoint.
The above model describes a part of the overall 6. Other types of manual control are possible
problem of controlling roll and heading of a ship. It is Joystick control, in which the operator gives a
also the part that causes the greatest difficulties in force setpoint in the ship's X-Y plane, is a typical
calculating an 'optimal' controller (i.e. the example; an allocation algorithm is responsible
interference between roll and heading control). for positioning the available actuators.
Furthermore, it is unconventional in the sense that it
regards the two fin stabilisers independently (usually, 2.3 Practice
both stabilisers operate in tandem). But this part is Just before the turn of the century, the first ships
well suited to illustrate the practical benefits of sailed with motion control systems that are based on
regarding 'allocation' as an independent part of a the control architecture described above. Besides the
controller design (the appropriate allocation short time span to develop appropriate algorithms, a
algorithm for this example goes beyond the scope of number of other constraints had to be met
this paper): • In all cases, actuator capability is not guaranteed
1. Making changes in the actuator configuration to be sufficient under all operational conditions.
does not affect the controller design This implies that the control system has to be able
Adding a bow-thruster or adding stabilisation to operate continuously with limited actuators.
tanks merely extends the freedom of an allocation • The performance requirements were close to what
algorithm to find the optimal set of actuator is possible in theory. In case of the hopper
setpoints. If an actuator fails, the allocation dredgers, that would result In a substantial
algorithm will automatically introduce a different improvement for track keeping In dredging
set (if possible). In principle, redesign of the conditions.
controller is not necessary. • The performance had to be demonstrated within
2. It IS relatively easy to coop with actuator the time span of the sea acceptance trials. Hardly
limitations such as rate limit and angle limit. any time would be available for testing or
Upon reaching a limitation, an appropriate modifying algorithms.
allocation algorithm will automatically try to • Actual systems had to be supplied with user-
compensate by adjusting the setpoints of the other friendly human interfaces that encourage the use
actuators. of such complex systems.
3. It is relatively easy to introduce priority schemes
An allocation algorithm for the above example
• Extensive functionality, especially in case of the
hopper dredgers. This is illustrated in Fig. 4
could be realised such that rudder motion for roll giving an overview of the baseline architecture of
reduction will be introduced only when the the actual system. All different operational modes
stabilisers reach one of their limitations. It is even of a hopper dredger had to be considered resulting
possible to change the priority scheme on line, for in a mix of manual and automatic control with
instance to let the stabilisers compensate for different requirements to be executed with
rudder induced roll during a turning manoeuvre different sets of actuators and different control
only. This does not affect the design of the schemes.
controller or the filter/estimator.
4. A conflict of interest between more than one
process output to be controlled can be clearly
recognised.
The allocation algorithm will recognise when it
reaches the boundaries of the system's capability.
For example, the difference between desired and
possible roll moment can be used as input to some
adaptation scheme (' adjust control' in Fig. 1) to
find a better compromise between controlling the
heading and the roll. This information cannot be
obtained as easily from comparing the difference
between actuator setpoint and actuator position.
5. It is possible to allow combinations of automatic
and manual control
25
Applying the new controller architecture may be addresses rudders, bow thruster(s) and main thrusters.
considered as one of the prime reasons why these It is expected that in future this will expand with
projects have been concluded successfully. It was more types of actuators and control of the remaining
possible to work with a team where each team ship motions (pitch, vertical acceleration, etc.).
member was responsible for a priori defined parts of
the overall system. A simulation environment was A substantial improvement of the development time
created including models of ship, disturbances, of complex control systems is possible!
actuators and sensors for testing purposes. Although not the only driving factor, the introduction
Subsequently, while executing tests with their of the controller architecture above (and the
individual parts, the team members automatically consequential parallellisation of the tasks that had to
validated the contributions of each other. Thus, the be carried out) has been a key element in reducing the
different design problems could be tackled in development time ofImtech's baseline DpDt system.
parallel. The other key element has been the embedding of a
simulation architecture within the same control
system. The combination was crucial for developing
3. CONCLUSIONS the actual systems for a number of hopper dredgers
Universities have to be wary that a too strong and other ship~ ~imultaneously and in time.
commitment to long time research will increase the The replaceabIhty property enabled the developers to
gap with companies. This is particular true in those add quickly basic filter, control and allocation
areas where companies face tough competition and algorithms (often re-use of existing technology) as a
where a short development time of new products has starting point in a simulation environment prior to
become essential. In case of control systems, focusing defining the fin~l algorithms. Thus, in a v~ry early
on algorithms that have not a clearly recognisable ~hase of t~e desI~, the customer could be Involve.d
relation with the systems to be controlled is merely In the ~roJect. D:sIgners cou!d take adv~tage of hIS
increasing this problem. extensIve operatIOnal expenence and Increase the
The paper introduces a way to look at control systems chance of acceptance of the final design.
that may be used as a way out of this growing ..
problem. It is a way that has its roots in Object Contmuous Improvement of complex control systems
Oriented Engineering and it demands adhering to a is feasible! ..
certain controller architecture. It has been proven to The .controller archIte~ture allows Improvement of
be beneficial for industry, but it seems to have merits relatively small parts In a way that does not affect
for Universities as well: other parts. Within Imtech's baseline, examples are
• A company that uses such a architecture will be already many: othe~ control alg~rithms, ~dding a
more lenient towards letting Universities research stem thruster, replacing the allocation algonthm by a
potential improvements. Modifications that more adv~nced s~h~me, etc: .
adhere to this architecture can be easily embedded CommercIally, It IS a sIgnIficant advantage that
within the existing system and therefore brought changes are confined and that resource consequences
quickly to the market. Upgrading complex control (co~ts & m~npower) c.an be overseen. ~xtras are
systems can be done gradually over time. eaSIer to estimate causing development nsks to be
• A University that adheres to this architecture reduced.
sacrifices some of its freedom to a more organised
form of research. But it will gain better facilities 4. REFERENCES
to work in changing teams at complex control Amerongen, J. van, H.R. van Nauta Lemke and
problems. It will be more easy to compare P.G.M. van der Klugt, Application of Adaptive
different solutions (and thus finding their merits) Control to Rudder Roll Stabilization of Ships,
without having to resort to simplifying the Proceedings ACC, 1986.
processes to be controlled. And, most important, Grimble, MJ., S.A. Carr and M.R. Katebi, Integrated
it will enhance their success in finding companies Ship Control using Hoc Robust Design Techniques,
interested in their work. Proceedings IEE, 1994.
Klugt, P.G.M. van der, Rudder Rolls Stabilization in
A general-architecture for controlling all ship Hr. Ms. Karel Doorman, J. Nav. Eng. 33(2), 1991.
motions is feasible. Klugt, P.G.M. van der and S.A. de Meijer,
The available structure includes the means to control Autopilots: the next step, Proceedings 12th SCSS,
all relevant ship motions of a growing number of ship 1999.
designs. At this moment in time, Dynamic
Positioning, Track Control, Speed control and RRS
are examples of modem control options that are
available. Compensation is possible for wind, current
and dredging forces. The allocation algorithm
26
allowing for a meaningful combination of manoeuvring
and seakeeping motions in the prediction of extreme
ship motions. Furthermore, convolution terms (so
called "memory effects") are being incorporated that
would improve the prediction of the behaviour of the
vessel in non-zero encounter frequencies.
2. EQUATIONS OF MOTIONS
Abstract: This paper describes a methodology to implement and operate an onboard expert
system to be applied on real time ship stability computation. The system comprises a data
acquisition system associated to a software based tool on a PC computer. Such instruments
measure roll motions and process the acquired data to give information on actual ship stability,
such as an alarm when the actual stability deviation exceed a predetermined limit value: Ship
stability is computed by means of an FFT algorithm associated to further filtering based in the
frequency domain analysis of ship motions associated to sea state parameters. Copyright ©
20011FAC
.
Keywords: Ship stability, Roll motion, Fast Fourier Transform, Low pass filter,
The basic methodology for most of the Some methods to compute seagoing stability
methods to compute ship stability in any operating have been developed. In (Brown and Witz, 1996)
condition, is the designer's information supplied as roll motion records have been used to implement the
"stability and model loading manual". The stability computation procedure by means of roll
information serves to define the stability decrement method. In (Haddara, et aI., 1994),
characteristics in typical and unspecified cases of validation of the random decrement technique with
ship loading based on the data of masses and static experimental data was also performed to extract roll
moments for the deadweight components with regard natural frequencies for a static stability assessment in
to the influence of liquid cargo free surfaces. waves. Some other methods like solution, the
autocorrelation equation, the modulation function
As is well known, the actual weights of many technique, (Haddara and Zhang, 1994), (Nechaev,
categories of standard cargo units, like containers or 1995) or other using signal processing, (Zeng, et al.
trailers may differ significantly from those specified 1999).
in the terminal documentation.
1.1. The Basic Motion Model
Also important is the error of the basic data
on the cargo centre of gravity co-ordinates, that Modelling the motion of a ship in a seaway is
together with errors in the necessary amount of extremely complex and the full solution to the
ballast contribute to show that the calculation problem has yet to be established, see (Okhusu,
method of stability does not satisfy with permissible 1996).The difficulties lie m the non-linear
39
interaction between the rigid body structure of the motion which can be most realistic treated in
ship and the surrounding random fluid field.This isolation, related with the other modes of motion,
Issue IS further complicated if the structural (Okhusu, 1996).
deformation of the vessel is taken into account when
considering its response, changing from a six degree To obtain de initial stability characteristics of
of freedom model to an infinite modes of motion the ship, like GM, or the KG, the height of the
elastic body. gravity centre over the keel of the ship it is necessary
to estimate, first the normal or natural frequency or
The motion considered in the ship is mainly roll period of the ship, through a relation between
due to the wave disturbance, which in effect, is the roll motion, wave excitation, and other quantities
input to the system considered. like ship roll inertia and roll damping moment, all of
them strongly dependent of the ship draft.
For a rigid ship travelling with constant speed
v at a arbitrary angle to sea waves, the resultant
motions in the six degrees of freedom are governed 2. THE SIMULAnON MODEL
by a set of second order differential equation, that for
all linear motions of the ship, has the following The problem for calculate the initial stability
standard form, (Okhusu, 1996). based on instrumentation data , i.e. true KG of the
ship demands the realisation of three different
alternative tasks:
(m + a)q(t)+ bq(t)+ cq(t) = Q(t) (1)
40
THE ANALYSIS OF A SHIP FUZZY DOMAIN IN A RESTRICTED AREA
Zbigniew Pietrzykowski
Abstract: The paper presents an analysis of the fuzzy domain of a ship in a restricted area. The
notion of the ship fuzzy domain has been defined. Also, a method of its determination has been
presented. Boundaries and areas of the ship fuzzy domain have been analysed for various levels
of the navigational safety. Copyright ©2001 IFAC
45
AUTOMATIC SHIP BERTHING USING PARALLEL NEURAL CONTROLLER
Email: im@naoe.eng.osaka-u.ac.jp
Abstract: In this paper a parallel ANN(artificial neural networks) for the automatic berthing will
be discussed. This controller has a separated hidden layer each control an engine and a rudder
respectively. Using this controller simulations were carried out where the initial conditions such
as ship's positions and heading angle are different from teaching data provided. Finally
comparison of separated hidden layer and united hidden layer will be described. Copyright ©
20011FAC
51
d2 is the remaining distance to the goal point. These
factors can explain the fact that navigators usually 3.2 Procedure of training
make imaginary line to goal point under berthing
work. Even if ; ,1] can explain the location Popular method of training, back propagation, is
information, they are not enough to explain the ship's adopted. The Neural Network Toolbox from the
location information. MATLAB package has been used to train the model.
Just variable learning rate method is used to minimize
time taken. Generally learning rate is held constant
3. PROCEDURE FOR LEARNING AND throughout training. If the learning rate is set too high,
MAKING TEACHING DATA the algorithm may oscillate and become unstable. If
the learning rate is set too small, the algorithm will
3.1 Preparing for the teaching data take too large time to converge. So variable learning
rate is adopted here. Since a neural network in this
This paper focuses mainly on how a parallel neural paper have separated hidden layers, training is needed
network will work effectively. So authors didn't try twice to make one set of weight and bias which will
the automatic berthing problem using a tug or side produce one set of output units. The group for rudder
thruster. This problem will be a next challenge. and the group for engine are trained separately. One
Automatic berthing mentioned here means that a ship example of these training is presented in Fig. 5. The
stops near the goal point within O.2m1sof speed and figure explain very well the before and after training,
between 250-270deg of heading angle. Basically 6 where the circle shapes indicate the original data,
sets of teaching data were adopted as like the Fig. 4-1 teaching data, triangle means before training data and
where disturbance is not considered. 4 sets of square means after training data respectively. It is
teaching data are also adopted to compare their easy to understand how much training is conducted
effectiveness with 6 sets of teaching data. Fig. 4-2 well by this figure.
shows 4 sets of teaching data.
shown in Fig. 7.
55
5. CONCLUSIONS AND DISCUSSIONS
K. Kose et al., "On a Computer Aided Maneuvering
In this paper, a parallel ANN for ship berthing was System in Harbors", Journal of the Society of Naval
discussed. In this paper, two groups of input units are Architects of Japan, Vol.160, Dec. 1986, p.l 03-11 0
considered to compose a parallel NN where hidden (in Japanese)
layers are split into two. The first group consider all
of input units and anther group include only two K. Kose et al., "On a Mathematical Model of
input units such as remaining distance, d2, and a ship Meuvering Motions of Ships in Low Speeds", JSNA
speed, u. But it can be said that additional research is of Japan, Vol. 155, June 1984, p 132-138 (in
needed to determine how input units should be Japanese)
separated. Conclusions of this paper can be drawn as
follows T. Koyama and Y. Jin, "A Systematic Study On
1) Newly designed ANN was used in automatic Automatic Berthing Control(151 Report)", Journal of
berthing problem as a controller. the Society of Naval Architects of Japan, Vol. 162,
2) A parallel ANN has good control ability December 1987, p.201-21O (in Japanese)
comparing with normal ANN of united hidden
layer.
3) The funnel area suggests the possibility that
automatic berth from every direction can be
accomplished.
4) Successful berthing has been accomplished even if
under different initial condition and different
starting point with teaching data provided.
6. REFERENCES
57
AN ELECTRONICAL AID FOR MANOEUVRING SHIPS IN RESTRICED WATERS
Abstract: Manoeuvre procedures which use rudder, propeller and thrusters are very
complex in which the crucial criteria for the feasibility of a manoeuvre is the navigational
space. Currently used automatic track systems configure successful way-point determined
routes although no technical support is given if a mix of rudder and engine manoeuvres are
necessary in harbour waters. The main project goal is to extend the function of the
automatic track systems. The first investigation involves developing a useful algorithm
which is carried out in the project. Examples of newly applied simulated automatic tracks in
harbour waters are given and discussed. The following paper describes the principles and
methods for developing an assistance system for the aid of navigation in harbour waters.
The project is supported by the Federal Ministry for Education, Science, Research,
Technology and Building. Copyright © 2001 IFAC
59
4. PRACTICAL TESTS
5. CONCLUSION
63
AUTOPILOT DESIGNED WITH FUZZY SET THEORY
I Departmentof Information and Systems, University of Pavia, Pavia, Italy Via Ferrata I, 1-27100 Pavia, Italy.
Also: Institute of Ship Automation C.N.R Via de Marini 6,1-16149 Genoa, Italy.
2Mechatronics Research Centre, University of Wales College, Newport Allt-yr-yn Campus, P.O. Box 180,
Newport, NP205XR, UK
3Department of Mechanical and Marine Engineering, University of Plymouth, Drake Circus, Plymouth, PIA
8AA, UK.
Abstract: In this paper, the design of course-keeping autopilot, is based on the cognitive modelling
approach and fuzzy set theory. Using this modelling approach, the aim is to design a control
system based on a model of the expert, who is able to specify the general properties of the system.
rather than on a model of the system to be controlled. After an heuristic description of the desired
manoeuvre characteristics, fuzzy set theory is used for the synthesis of the autopilot. Simulations
on a non-linear model of a containership are presented for course-keeping manoeuvres in the
presence of wave disturbances.
71
Based on this stored knowledge, the actual situation is
evaluated in order to infer the appropriate control
action. The deduced control action, performed by the
so-called inference machine, is based on fuzzy logic
where uncertainties are easily handled.
2. FUZZY AUTOPILOT
T K Chang and D L Yu
Abstract: A fault-tolerant control (FTC) approach based on neural network (NN) inversion method
is proposed for multi-input, multi-output (MIMO) unknown non-linear dynamic system. An
adaptive multi-layer perceptron network (MLPN) on-line learns the system non-linear dynamic,
including behaviour of system with actuator or component faults. This MLPN is inverted based on
the Extended Kalman-filter (EKF) to estimate the appropriate control action to the non-linear
system. The stability of the NN inversion is proved with Lyapunov method. The results of FTC
application to a continuous stirred tank reactor (CSTR) process simulation show that the controlled
faulty system maintains the control performance and stability. Copyright © 2001 IFAC
Keywords: Fault-tolerant systems, adaptive control, non-linear systems, model inversion, Extended
Kalman filters.
77
NEURAL NETWORKS FOR SYSTEM IDENTIFICATION OF COUPLED SHIP DYNAMICS
83
The above Figures, 12, 13, and 14, demonstrate the
effective generalisation by the network for the final
third, after 3470 seconds. Similarly high quality
results have been obtained using another eight data
sets, but only individually. It was originally hoped
that a single network could be trained using each of
the data sets in turn, in order to generate a network
capable of providing an approximation to each of the
sets. However, this has not been the case, as re-
training a network with another set tends to degrade
the performance of the network acting on the
previous set.
Full-scale measurement data have revealed that slowly varying fluctuations in the
course of survey vessels performing marine seismic data acquisition with towed
recording (streamer) cables are capable of forcing the in-sea hardware in longitudinal
and transverse directions. This may give rise to undesired low-frequency oscillations in
the depth of streamer cables. The interaction between the transverse and longitudinal
motions takes place at the location of the depth-control devices that are mounted along
the streamers. A quasi-static model that describes the low-frequency dynamics of
streamer cables is proposed for the purpose of controller design and analysis. The
simulation study showed that the overall depth-control performance is sensitive to
appropriate tuning of controller gains, selection of sampling period and application of
dead bands. A successful depth control will rely on the number and lift capability of the
. mounted depth-control devices. Copyright ©2001 IFAC
89
2. FULL-SCALE MEASUREMENTS
The success of marine seismic data acquisition The principal result presented in this paper is a quasi-
utilising towed streamer cables is completely static model that describes low-frequency dynamics
dependent upon the ability of the streamers to keep of towed marine seismic streamer cables. The effect
an assigned target depth during all surveying of varying controller settings of the depth-controllers
conditions. However, problems with maintaining the that are spaced along a seismic cable is analysed.
depth of streamers do occasionally arise. In such
cases, the acoustic reflections received may be The simulation results indicated that the proposed
subject to errors and uncertainties that affect the quasi-static cable model is adequate for simulating
reliability and accuracy of the final post-processed slowly varying oscillations of streamer cables. It is
images of the geological structures. Thus, slowly shown that the depth-controllers may be able to
varying depth-oscillations may have a significant compensate the undesired cable motions if sufficient
negative impact on the operational efficiency of this lift (control) forces are provided in combination with
kind of nautical operation. minimum dead bands and low sampling rate.
94
NON-LINEAR SEMI-SUBMERSIBLE POSITIONING
SYSTEM DESIGN USING AN Hoo CONTROLLER
Abstract: This paper describes the development of a Dynamic Positioning System for a
large production semisubmersibleplatform. The main purpose of the work is to present
the development of a full non-linear mathematical model for motion simulations and
the synthesis of a controller for a Dynamic Positioning System using the Hoo approach.
The non-linear model used for simulations and the linear model derived for control
synthesis include representations of all the important environmental disturbance
sources and reference dynamics. Robustness and performance are treated specifically.
Results are presented and analysed. Copyright ©200J IFAC
95
very simple linear time invariant models, which
only give good performance near nominal con-
ditions. Several adaptive schemes were tried but
never implemented probably due to the intrin-
sic complexity (Katebi et al., 1985), (Donha and
Brinati,1994) .
This paper presents the design of an fico con-
troller for DP applications where the objective
is to improve robustness of stability and perfor- where M is the platform mass; Xu, Yv and Nt are
mance. To achieve these goals, a control system the added mass along the surge (x) and sway (y)
which couples the three degree of freedom of the directions and the added yaw (around z) moment
horizontal plane (surge, sway and yaw) was devel- of inertia, respectively; u, v and r are the surge,
oped. To assist in the performance and stability sway and yaw speed in the body coordinates; Xn,
robustness assessment of the system, simulations Yh and N h are the speed dependent surge and
are carried out using a non-linear model. sway hydrodynamic forces and yaw hydrodynamic
moment, respectively; Xw, Yw and Nw are the
The work is organised as follows. In section 2 and wind induced surge and sway forces and yaw
3, respectively, brief descriptions of the mathe- moment, respectively; Xc, Yc and Nc are the
matical models for simulation purposes and for control forces and yaw moments produced by the
control synthesis are discussed. Section 4 sum- thrusters and lzz is the moment of inertia about
marises the control strategy. Results are presented the vertical axis z; X, Y and IJ1 are, respectively,
and analysed in section 5 and conclusions are the surge, sway and yaw motion in the inertial
drawn in section 6. system and 'IjJ is the yaw motion in the moving
reference system.
The hydrodynamic derivatives for the platform
RS-35 were estimated by tests in a conventional
towing tank (Donha and Brinati, 1990).
4. CONTROL DESIGN
Abstract: The article discusses sea route planning and optimisation methods which take
into account minimisation of the time needed to pass a so called time-minimum route at
dynamically changing weather conditions. The above task is a kind of dynamic
optimisation problem in which dynamic changes in weather conditions are defmed as
moving constraints. The effect of hydro-meteorological conditions on changes in the
parameters of ship motion and route was determined. The obtained solutions and
authors' experience gained during studying the presented methods have made the basis
for developing new evolutionary algorithms which were not used so far in planning
ship routes. Copyright ©200J IFAC
101
simplifications, one can defme the tenn, existing in the area occupied by the tropical cyclone which
the literature, of a "typical" trajectory of the affect the ship, and may, for instance, result in
tropical cyclone. This trajectory consists of three damage or incorrect operation of navigational
sections: instruments, thus affecting the safety of sailing.
• Equatorial section - along which the cyclone
moves west, with small north component 3.3 Modelling of the area of potential danger
(south on the Southern Hemisphere). The
speed of the translator motion of the cyclone Warnings about tropical cyclones most often
along this section is, as a rule, low, within 5 to include the infonnation on predicted direction and
12 knots (8-9 knots, as average). The cyclone speed of the cyclone motion, frequently
very often happens to alternately accelerate complemented by predicted positions of the centre
and decelerate its speed, to move north only, of the cyclone system. It is difficult, however, to
or to increase the north component. evaluate both the exact position of the cyclone eye
• Meridional section - begins when the north at the time of analysis and the evaluation error
(south) component starts dominating over the itself. Therefore, to make the manoeuvre of
west component, and ends when the east avoiding cyclone effective, the area of potential
component starts dominating over the north danger should be evaluated, i.e. the area which can
(south) component. This section includes a so be reached by the eye of the cyclone moving in the
called point of recurvature. The cyclone speed prescribed manner in the time of interest.
along this section increases to 12-20 knots.
• Polar section - the main direction of the
cyclone motion is north-east (south-east on the
Southern Hemisphere), the cyclone speeds up
to 20-30, or even 50 knots.
6 CONCLUSIONS
105
planning problem can be reduced then to the multi- Banachowicz A. (1995) Environment and sea area
criterion dynamic optimisation task with static and parameters having an influence on
dynamic constraints. restriction of ship's moving. Gdynia
Further works on the subject will present an Maritime Academy, Navigational Faculty.
evolutionary method used for solving the problem The work under Andrzej Banachowicz
of ship route planning in the presence of changing leadership. Gdynia.
weather perturbations. Taking into account Jurdzinski M. (1989) Sea passage planning. Marine
continuously increased speed of computers it seems Publisher, Gdansk.
possible to use an interactive and multi-criterion Jurdzinski M. (1994) Influence of technical,
approach in the future to support navigator's constructional and operational parameters
decision, where the operator would choose a on ship's moving in optimalization
solution from a number of multi-criterion effective aspects. Gdynia Maritime Academy,
solutions prepared by the decision support system. Navigational Faculty. The work under
Miroslaw Jurdziilski leadership. Gdynia.
ACKNOWLEDGEMENTS M~droszkiewicz K. (1974) Sea passage
optimisation taking into account
The research reported in this paper was supported economical criterion. Publications of
by the Polish State Committee for Scientific Gdansk Polytechnic no. 220, Electrician
Research. XXXIV. Gdansk.
Marsz A., Styszynska A.: (1992) Exercises in
REFERENCES meteorology and oceanography. Part II,
Tropical cyclones. Gdynia Maritime
Aligne F., Papageorgion M., Walter E.: (1998) .. Academy, Gdynia ..
Incorporation power variations into NOWICkI A.: (19:8) Knowledge of maneuvermg
weather routing and why it may lead to ocean-gomg vessels. Problems of theory
better results. IFAC CAMS Fukuoka and practice. Marine Publisher, Gdansk.
Japan. ' Wisniewski .B.: (1986) Ship's route optimali:~tion
Chotkowski W., Madroszkiewicz K. (1974) takl~ account .weather. condltl~ns.
Analysis of methods used to control of .. M~tJtJme Academy m Szczecm, Szczecm ..
ship's route with minimum sailing time. Wismewski B., Holec M.: (1983) OceanographyIn
Publications of Gdansk Polytechnic no. bro~d outline. Part /J, Dy~amic of sea.
220, Electrician XXXIV. Gdansk. PolIsh Navy Academy, GdynIa.
Wrobel F.: (1992) Vademecum of navigator,
Marine Publisher, Gdansk.
ASSISTED DYNAMIC POSITIONING SYSTEM FOR A FPSO BASED ON
MINIMIZATION OF A COST FUNCTION
107
POSITIONAL GAME CONTROL OF SHIP IN COLLISION SITUATIONS
Jozer Lisowski
Abstract: The paper concerns the application of elements of the game theory for the purpose
of optimal control of some dynamic continuous processes. Using as an example the process
of safe ship's control, the paper presents the problem of applying a positional non-
cooperative game of "j" objects for the description of the process considered as well as for
the synthesis of optimal strategies. A mathematical model of differential game is
formulated and its approximation in the form of linear programming problem is used for the
synthesis of safe ship's trajectory as a multistage process decision. The considerations have
been illustrated an example of a computer simulation the POSGAME programme to
determine the safe ship's trajectory in situation of passing a big number of the objects
encountered, recorded on the ship's radar screen in real navigational situation at sea.
Copyright ©2001 IFAC
Keywords: differential games, positional games, dual linear programming, safety analysis,
risk, ship control, computer simulation.
113
RESEARCH ON
FLEXIBLE OSCILLATING FIN PROPULSION SYSTEM
AND ROBOTIC FISH
Abstract: The purpose of this paper is to describe the feasibility research of marine vehicles
with oscillating fin propulsion control system. The oscillating fm propulsion system was
designed and constructed to be combined with a ship model. Tank tests using the ship
model have confmned the system's feasibility. As a result, several advantages of the
oscillating fin system have been found out. A neural network was successfully applied for
an identification of the ship model dynamics with the oscillating fin, and its effectiveness
was confmned. Robotic fish, intended as an amusement attraction for aquariums, using the
oscillating fm propulsion system have been developed. Its capability of untethered 3-
dimensional movement was confmned. Copyright ©20011FAC
Keywords: Flexible oscillating fin, Marine system, Propulsion device, Robotic Fish
1. INTRODUCTION Isiki et al. (1983), have reported that the flexible fin is
superior to the rigid fin. Theoretical studies, see Isiki et
It is well known that marine creatures such as fish al. (1983), on flexible oscillating fins have been partly
swim using small power even at high speeds conducted, while studies on the flexible fin including
(dolphin: 60 kmIh, swordfish: 80 kmIh), see Yoshida the control system have not yet been performed. In this
(1976), and sweetfish, etc. are superior in their research after the control system for a flexible
position keeping characteristics. These characteristics oscillating fm propulsion device and the oscillating fm
as creatures have been of interest as science from the driving device were designed and manufactured, a
old times and much research has been conducted, see cruising test was performed first by a numerical
Hertel, H. (1966), however; it is rare to study these simulation and then with a model ship, and the
characteristics from the viewpoint of a engineering fundamental performance has been grasped and
subject, see Isiki et al. (1983). prospects of putting them to practical use have been
.... obtained. And robotic fish for amusement in aquariums
One such research IS on a flexIble osclllatmg fm
. etc. has been deveI0ped as an applied product.
control system whIch could be used for the '
propulsion of marine vehicles by positively making
the. most of the ~haracteristi~s of the flexibl~ p~ 2. OUTLINE OF THE SYSTEM
ThIs method obtams a propulsIon force by osclllatmg
fins equipped to vehicles on the analogy of the . '
.. In many cases the kineUc parameters 0f the oscillatmg
moUon of manne creatures. . ·
fin cannot be directly detected m controI 0f an oscillatlng
The linear theory analysis has been applied to the fm and there are problems choosing and identifying
rigid oscillating fin which regards the oscillating fm parameters to be used for control, and a control system
as a rigid part, and past experiments and research, see able to cope with such problemsshould be architected.
119
A NONLINEAR ST A TE-SP ACE MODEL OF
DIESEL PROPULSION PLANT OPERATION USING NEURAL NETS
131
Current generation unmanned underwater vehicles, equipped with robotic manipulators, are teleoperated and consequently
place a large workload burden on the human operator. A greater degree of automation could improve the efficiency and
accuracy with which underwater tasks are carried out. These tasks can involve manipulator motion that is both
unconstrained and/or constrained. For unconstrained motion, where a trajectory requires following, a prerequisite is good
joint angle control. An adaptive self-tuning pole-placement controller is used for joint angle control. Practical results show
the benefits compared to conventional fixed-gain control. For constrained motion, often simultaneous control of position
and force is required. An adaptive hybrid position/force controller is proposed and compared to a fixed-gain version.
Simulation and practical results illustrate the merits and drawbacks of each scheme. Copyright © 2001 IF AC
1. Introduction
Unmanned Underwater Vehicles (UUVs) are commonly manipulator dynamics. A self-tuning approach, using
used in offshore, marine science, defence and salvage pole-placement control, was chosen.
activities for inspection, maintenance and repair.
Workclass vehicles are equipped with one or two There have also been many proposed control strategies to
manipulators, and are controlled using tele-operation by achieve simultaneous position and force control. Two of
an operator on the surface. The manipulators are generally the most popular are hybrid position/force control [2] and
hydraulic, with no tactile or contact sensing for impedance control [3]. A hybrid position/force control
monitoring or controlling contact forces during task scheme decomposes the task space into two orthogonal
execution. Consequently, damage to manipulator and sub-spaces, one controlling the manipulator
environment are common during typical operations. The position/orientation in certain Cartesian axes, the other
use of sensed force and torque data would greatly assist in controlling forces/torques in the remaining directions.
execution of many tasks [1]. Such an ability is central to This enables the various tasks encountered during subsea
activities such as weld inspection and grinding. In intervention, such as weld inspection and mating of
addition for unconstrained manipulator motion, possibly connectors, to be readily accomplished. Often, underwater
involving complex reference trajectories, accurate joint tasks require specific forces to be applied to unknown
angle control is essential. In this paper, we consider the environments and this is difficult to guarantee with
problems of joint angle control and simultaneous control impedance control schemes. We have therefore chosen a
of manipulator position and applied force using an hybrid position/force control approach.
underwater hydraulic arm with a fixed base.
During subsea intervention the manipulator operates in an
There has been an abundance of joint angle control unstructured and time varying environment, with wide
schemes proposed over the last two decades e.g. changes in operating conditions. These variations must be
fuzzy/neural, model reference, computed torque, variable accommodated by the control scheme used. It has been
structure, optimal and self-tuning. Any successful control shown that a fixed gain hybrid position/force control
scheme must be able to cope with the following problems. scheme which performs well for one particular set of
The highly nonlinear dynamics of both the manipulator operating conditions, degrades considerably away from
and actuator, including inertia, gravitational, Coriolis and this operating point [4]. In this paper we consider a self-
centrifugal effects, friction, mechanical flexibility, tuning adaptive hybrid scheme to cope with any
backlash and actuator geometry. Accurate control is unmodelled disturbances and variations in the robot and
required over a wide range of operating conditions. There environment dynamics.
is cross-coupling between neighbouring inputs and
outputs of the system and the system dynamic parameters The paper is organised as follows. Section 2 describes the
are time-varying. An adaptive control scheme was self-tuning adaptive joint angle control scheme. The
deemed most suitable for this application, as it can following section shows the practical results of the
automatically accommodate the wide variations in adaptive scheme and compares it to a fixed-gain
149
During the tests, some buoyancy and power problems
were exposed. After some modifications, the tests
were repeated and the vehicle successfully docking
onto the submerged station.
159
Abstract: Accurate identification of nonlinear time variant MIMO systems, especially in
case of AUVs is essential for implementation of control algorithms and navigation
purposes. Control problems of AUVs have also difficulties due to the nonlinear
dynamics behaviors of vehicles and also unpredictable effects come from the
surrounded water mass. These nonlinear effects are so complicated that bring
difficulties for dynamics modeling and position control descriptions while using
conventional methods. The proposed method here uses neural networks as a general
idea for dynamics modeling and position control of any six-degree of freedom rigid
body and are applied to an AUV, named Twin Burger 2, as an example. Supervised
Learning and Unsupervised Learning are used for adjusting the neural networks'
synaptic weights and the results are illustrated. Path planning of AUVs using neural
network is also addressed here as of a complicated control scheme and Reinforcement
Learning is used for adjusting the neural network parameters of the path planning
module via some obstacle avoidance examples. Copyright ©2001IFAC
167
Fujii is its consumption time while on line
adaptation. Imaginary Training algorithm proposed
by Ishii et ai. (1995) modifies the structure proposed
by Fujii by dividing it into two different parts called
Imaginary World. and Real World. Real World part
is used for on line implementation of the control task
and Imaginary World part is used for off line
adaptation of the controller. The problem regarding
to the network proposed by Ishii is that the Forward
Model and Controller Network are just applied to
SISO system and have not enough generality to be
applied to MIMO ones. In this paper it is going to
overcome to this problem by extension of his idea to
dynamics modeling. and position control problems of
six degree of freedom rigid bodies. Twin Burger 2 is
considered as an example and some results are
illustrated (Sayyaadi et aI.. 2000).
168
complex and large that cannot be trained easily in a
single step of training. As a result it has to be trained
gradually in some sequences of training procedures.
Dynamics modeling of MIMO systems is called here
as Coupled Model Neural Network Identifier
(CMNNI), and position control is called as Couple
Model Neural Network Controller (CMNNC). Each
one is decomposed to a numbers of SISO systems,
called Single Degree of Freedom Neural Network
Identifiers (SDFNNI), and Single Degree of Freedom
Neural Network Controllers (SDFNNC). After
training of all of these SISO identifiers and
controllers, additional interconnecting synapses Every neural network needs some kind of data sets or
compose MIMO identifier and controller and then teachers to be used as of teaching data. Using
these new synaptic weights are adjusted through Supervis~ Le~ing carries out training of the neural
additional training processes. network Identifiers (dynamics modeling). It means
that input/output behaviors of the system are sampled
Identifier and controller networks have three layers, through some experiments and then these sampled
which are input layer, hidden layer, and output layer. data sets are used for adjusting the parameters of the
To simulate the motion of the vehicle by giving input neura~ net~o~ks. The proposed identification
force and only initial condition of velocity, the output techmque IS Impl~mented to a four degrees of
of the first integration layer has to be fed back to the freedom AUV, Twm Burger 2, which is shown in
input layer through RC-l loop. The recurrent Fig. 3. This AUV has four motion modes called:
connection from the hidden layer to the input layer, Surg~ (longitudinal), Sway (lateral), Heave
called RC-2, allows the network to express the (Vertical), and Yaw (rotational). Doppler Sonar
dynamics behaviors with reduced number of input Sensor detects state variables of the vehicle. Number
state variables. The RC-3 loop makes the controller of neurons in hidden layer of every SDFNNI and
able to provide control signal (force/moment) by S?FNNC are 6 neurons and as a result this number
inputting only its initial value at the beginning of the will be 24 neurons for CMNNI and CMNNC.
control task. Inputs to the controller network are past ...
time step of control signaVsignals (force/moment), ~rammg algonthm of neural network identifiers in
position error/errors, and velocity error/errors. smgle degree of freedom modes and coupled mode is
Output of the controller network is control carried out in 4 consequent steps. Error Back
signaVsignals of the next time step. Inputs to the Propagation rule is used while training processes. In
identifier network are control signaVsignals, and past the first step of training, inputs to the identifier
time step velocity/velocities. networks are control signaVsignals and
velocity/velocities come from the sampled data sets.
Evaluation error is based on the discrepancies
between simulated acceleration/accelerations and
actual one/ones. In the second step, inputs are
control signaVsignals come from data pack and
simulated velocity/velocities. Evaluation function is
the same as step one. In the third and fourth steps,
inputs are the same as step two. Evaluation functions
are made from discrepancies between actual
velocity/velocities and simulated one/ones in step
three, and actual position/positions and simulate
one/ones in step four. All of the synaptic weights of
the SDFNNIs are set to some small random values
when training starts and then these parameters are
adjusted gradually. After getting proper convergence
in the evaluation function of each SDFNNI, the
CMNNI is made using additional interconnecting
synaptic weights. These weights are also set to some
small random values initially and then will be
modified while training algorithm.
170
3. PATH PLANNING OBSTACLE AVOIDANCE
MODULE FOR AUVS
179
Abstract: Due to more largely becoming ship units its manoeuvring freedom in
enveloped sea areas is limited. Therefore a more exact knowledge of the sea area
conditions and variable disturbances in the zone of manoeuvre than with smaller ship
units is necessary. The on-board sensor technology can determine only the body-related
variable disturbances. By using of an external acoustic doppler current profiler (ADCP)
in connection with a telemetry mechanism and an up-to-date identified ship's model the
calculation results of the manoeuvre predictor should be improved. The harbour
approach Puttgarden will be the example area for the investigations. Copyright © 2001
IFAC
185
which form these data are available in other
navigation systems. The data of the enumerating I to
4 were measured directly as analogue voltage signal
of ±10 V. These are supplied to an analogue input
modules with high impedance of a field bus node.
The field bus node converts these voltage signals into
appropriate NMEA 183 strings and put it out byaRS
232 interface module. Here it is mentioned that the
NMEA 183 standard is no longer sufficient with
dynamically relevant data, because for applications
of real times a data inquiry of 0.1 seconds is
necessary.
2.2 Structure of the sea area-fixed system component The known theories of the description of the motion
of ships suppose the motion on the undisturbed
Enveloped sea area specific information are surface of the sea. In consequence influences of
necessary to increase the prediction quality, because current in modem simulators are realised by a
all area information is not available always from on constant drift in the inertial reference system'
board sensors and data bases, e.g. ECDIS. The because a constant impulse is transmitted to the ship
approach of the ferry harbour Puttgarden was (Hilten, 2000). This assumption is always ingenious
determined as an example area for the project in such cases, if the ship moves in the free sea area in
MAPSYS in order to examine the influence of the which the change of gradient of current is very slight
187
Abstract: In this paper the problem of ship trajectory tracking with £-accuracy is
considered. Results presented are based on a three-degrees-of-freedom model with full
dynamic interaction between motions in roll, sway and yaw. The aim of control is to
minimise the influence of wave disturbances on roll motion and to track with £-accuracy
a ship trajectory. For this purpose a linearised and a non-linear ship model are used. The
Matlab-Simulink simulations confirm the viability of the proposed method for the high
precision control of ship track-keeping. Copyright © 2001 IFAC
191
ON THE DESIGN OF ALERTS FOR
FUTURE PLATFORM MANAGEMENT SYSTEMS
Michael A. Tainsh, F.Erg.S., Eur.Erg.
1. INTRODUCTION
The control of a ship and its services is carried out The Defence Evaluation and Research Agency
within the Platform Management System (PMS). (DERA) Centre for Human Sciences (CHS) was
There is sound evidence that there are problems approached by the MoD/Ship Support Agency (SSA)
within current Royal Navy (RN) PMSs. The status to study the handling of alerts in future RN ships to
information, warnings and alarms are displayed, support procurement programmes in line with current
acknowledged and controlled in ways that do not practice (References 1, 2 and 3). It was agreed that
match user characteristics. In fact, the general design the study should employ a number of techniques,
concept that appears to have been applied in many including:
cases seems to be more associated with the design of
electromechanical dials. In this case, the physical (a) Top-down analysis of known assumptions,
measure from a sensor (status) is available plus high-level issues, or system characteristics in
additional information on whether it is slightly order to derive likely human factors issues;
different from a desired range (a warning) or (b) Observations of lessons learned in the past
substantially different from that range (an alarm). on similar projects and general concepts of best
This is an 'equipment centred' approach to displaying practice;
information and takes little account of the users' (c) Exercises in early-prototyping, trials and the
characteristics. In the past, this may have been the use of Subject Matter Experts (SMEs).
only information that could have been provided in
non-computer based systems. However, in current In this case, all three were used to investigate design
and future systems this can be judged a very limited options and mitigate risks. The work described here
approach. addresses the issues associated with future ship PMSs
and aimed for entirely novel design solutions.
203
2. DEFINITIONS quite different in the future ftom any ships in current
operation.
The initial high level issue stems ftom the established
set of categories for describing alerting information: (c) An understanding was gained of the alternative
status, warning and alann. The established set was means of on-line control of the plant, equipment and
seen as 'equipment centred' based on what electro- services likely to be available to the PMS, including
mechanical technology could offer. These are known automation and support. The control of the PMS is
to be associated with displays that are poorly used likely to depend on more sensors and more processed
and disliked on some RN ships. In this study, a infonnation. In particular, the use of automation to
provisional definition of an alert was adopted: the reduce the need for the user to execute manual
mechanism by which a user is infonned of a change control functions was seen as central to the
of information. This general definition provided a consideration of the users' roles.
starting point for an empirical study. This definition
of alerts is both arbitrary and general. However, it (d) A concept of the Human-Computer Interaction
was believed that this defmition would ensure that (HCI) for a future PMS was develo~ed. The
there would be a broad opportunity for a wide range implications for the HCI are profound. There may be
of possible design solutions to arise ftom the a central workstation but equally the user may act
investigation. It was hoped that this would be ftom a remote position.
achieved by adopting the general definition that
makes no a priori distinctions between the categories (e) The generation of implications of the HCI concept
of alerting information that might be provided to a for the provision of alerts. It was inferred that the
user. scope of alert infonnation should cover three main
categories:
204
5. TRIAL ONE The tasks were:
213
ENHANCED NEURAL NETWORK MODELLING FOR PROCESS FAULT DIAGNOSIS
Abstract: A neural network (NN) based fault detection and isolation (FDI) approach for unknown
non-linear system is proposed. This FDI scheme is able to detect both actuator and sensor faults. An
enhanced parallel (independent) NN model is trained to represent the process, and used to generate
residual. A mean-weight strategy is introduced to overcome the unmodelled noise and disturbance
problem. An information pre-processor is implemented to convert the quantitative residual to
qualitative form and applied to a NN fault classifier to isolate different faults. The developed
techniques are demonstrated with a multi-variable non-linear tank process. Copyright © 20011F AC
Keywords: Fault detection, fault isolation, neural networks, non-linear systems, process
identification.
215
small and sensitive to modelling error. The fault
classifier will suffer from this small residual. Ideally,
a parallel NN should be used to generate residual.
However, an efficient algorithm for a parallel NN is
not available, due to its recurrent operation.
In both naval and commercial marine sectors, ship owners and operators are faced with the need to
operate sophisticated vessels with lean-manning levels. This is graphically illustrated by the goals of
the DD21 program in the US, where the target complement of95 is one of the main drivers behind a
70% reduction in operating costs. Such lean-manning policies are one motivator towards increasingly
sophisticated data-driven systems being fitted on-board. The design of such systems is no longer the
remit of purely control engineers, but relies heavily on both on human science and simulation.
This paper outlines the changing customer requirements in advanced marine applications and the
impact of these trends on a major supplier of marine systems. The development of advanced Platform
Management Systems in support of cost reduction and lean manning is then reviewed in more depth,
with particular focus on the role of human factors and simulation.
221
complement is a complex issue; it is a function command, electrical power, propulsion,
of the vessels operational role, battle damage damage control. Level 4 examples would be an
policy, platform design, career structure and integrated bridge system or PMS which
other issues. Pressures on operational budgets integrates together Level 3 functions. Level 5
are driving down manning levels. Crew covers Aftermarket and Fleet Management
reductions may be achieved by a number of applications, which are increasingly web-
means, including (but not limited to) further enabled.
automation. This requires trade-offs to be
carried out between Non-Recurring Costs Shipbuilders, owners and end-users have
(NRC), Unit Production Cost (UPC) and recognised that there are tangible benefits if
Through-Life Cost (TLC), leading to the different control systems throughout the
automation as a "spend to save" opportunity. vessel can be specified and procured in a co-
ordinated manner. This gives opportunity to
2.2 Risk Management reduce the number of different control systems
(and suppliers) involved, hence reducing
Availability of suitable technology is not interfaces and risks. A reduction in the
generally seen as a major barrier in the number of different hardware platforms offers
development of marine automation schemes. savings in carried-on-board and carried-at-base
Advances in shore-based industrial automation spares inventory. Where security and
systems have provided a plethora of availability considerations allow it, there is
commercial off the shelf (COTS) equipment, scope for integrating control functionality from
which in general can be configured for use in a various levels, on a common network and
maritime environment. However, it is the operating platform. This offers potential for
deployment of this technology in a fighting reducing passive inter-connections, cabling
vessel which brings with it technical and social and bulkhead penetrations, and allows operator
issues which are unique, and require special interfaces to be presented in a consistent,
considerations. The Royal Navy, the Defence integrated manner.
Procurement Agency (DPA) and the Prime
Contractors wish to reap the benefit of this
technology, whilst managing the attendant
problems. This is highlighted within the UK
MoD, who have recently placed a five-year
contract aimed at de-risking the application of
PMS to future Naval Platforms. An industrial
team, comprising BAE Systems, Vosper
Thornycroft Controls and Rolls-Royce, will
execute a program of studies and possibly
trials, focussed on mitigating and avoiding the
risk associated with the application of
advanced technology in Naval controls &
surveillance systems. These studies will
consider the functional and non-functional
requirements including safety, human factors
and supportability. The work will support the Figure 1 Control System Framework
Future Surface Combatant, Future Carrier
(CVF), Future Attack Submarine (FASM) and This is of course the territory of Integrated
other ship programmes. Platform Management Systems, which offer
the potential to provide integration in two
2.3. Controls Integration orthogonal directions. Horizontal integration
can be achieved using a resilient ship-wide
Control schemes can be classified within a network to allow control and information to be
simple five-layer model, which is given in distributed around the vessel. Vertical
Figure 1. The simplest controller (perhaps integration between operational controls,
locally controlling a winch) would correspond health monitoring, communications, CCTV
to Levell. A controller for a collection of and information management applications are
local equipment comprising a sub-system, also possible, with data sharing and fusion at
would fall into level 2 (examples of this all levels. This is analogous to the industrial
would be a gas turbine controller or an approach of sharing data from sensor to
automatic stabilising system). Level 3 boardroom.
integrates entire ship systems such as
222
With the operator effectively out of the loop
for much of the time, and working in a
supervisory role, situational awareness may be
impaired. For example, the designers of
autopilots and Dynamic Position controllers
are conscious to avoid the potential for
auto/manual mode confusion errors.
224
~ SMCTD Following task and workload analysis, various
A number of years ago, Rolls-Royce automation schemes have been identified
developed the Submarine Machinery Controls which allow the designers and future operators
Technology Demonstrator (SMCTD). This to evaluate the design prior to full
facility comprises a dynamic simulation of a implementation. Moreover, as the Platform
full-scale control console, connected to a Management System is fully developed, the
model of a generic nuclear submarine SMCTD models can be used as a dynamic test
propulsion system and electrical system. The facility, allowing the system to be exercised in
facility (shown in Figure 3) has been used a comprehensive range of operational
primarily to evaluate advanced control scenarios, including emergencies. This also
technologies, examples of which are SUR VEX represents a useful training experience for
(an expert surveillance system offering current and future ship's staff.
advanced operator diagnostics) and FEED (a
fuzzy-logic based boiler feed water control 6. CONCLUSION
system).
Marine controls engineers and hardware/
software "box" suppliers are enabling the
introduction ofPMS into Naval service. There
are significant technical challenges facing the
specifier, integrator and user of the advanced
ship automation systems of the future, for
example: supportability, obsolescence,
wireless interconnectivity, survivability,
software safety justification to name a few.
However, the successful introduction of
Platform Management Systems cannot be
carried out effectively by controls engineers
alone. This is an immensely wide-ranging and
broad subject area, and in the author's opinion,
no one person should pretend to be an
"expert".
Figure 3 Submarine Controls Technology
Demonstrator What is clear is that the design of successful
Platform Management Systems in the future
will be significantly affected by human factors
In recent years the SMCTD facilities had considerations. The contractual acceptance of
become difficult to support, and like many of operability will be a key element in the
the other Manoeuvring Room Simulators used handover of such systems. Human trust in the
by the RN, there came a need to replace the technology, acceptance of reduced manning
computer systems. In order to retain the levels and appropriate career structures and
considerable investment made in generating training arrangements are greater risk
the model library, and to bring the facility up considerations than technological risk alone.
to date, the models were re-hosted on a PC-
platform, using automatic code conversion Obviously some of these factors are outside
tools, developed originally for Simulator the influence of marine suppliers. Industry is
lifetime extension. Hence SMCTD II was turning to simulation as a route to the leaner
created, and is currently being used to support manning challenge: a challenge, which is
development of the propulsion and power being faced universally as a method of
management systems for the Astute Class of reducing operational costs.
SSN's (under contract from CAE Electronics
Ltd).
225
Copyright © IF AC Control Applications in Marine Systems,
Glasgow, Scotland, UK, 2001
Keywords: simulation, ship control, marine systems, transportation control, MIMO and
mathematical models.
267
When an experiment finishes, all data obtained along
time are there, as computer files, together with other
information pertaining to a qualification of the
results. This information is in the form of numerical
indexes, to measure the general effects of the control
along the complete experiment. For instance, the MSI
(motion sickness incidence) index
(O'Hanlon, MacCawley, 1974), and other related to
the reduction of vertical accelerations. Therefore we
have the means to compare different control
solutions.
Once the modelling stage of vertical dynamics of a Vertical dynamic of a high speed ferry can be
high speed ferry (De la Cruz et aI., 1998; Aranda et described (De la Cruz et aI., 1998; Aranda et aI.,
aI., 2000) and actuators (Esteban et aI., 2000) is 2000) by four transfer functions:
finished then the following stage is to design a pitch
273
1. INTRODUCTION We decided to employ an industrial PC to test
the control algorithms with the scaled down
This work is part of the investigation about replica. It was important to check the different
alleviation of fast ferry vertical accelerations control strategies, and to be able to do changes
using appendages. Before it, the following quickly in the CEHIP AR installations. This was
stages of the project have been completed: the motivated because the trials had to be
experiments with the scale model without performed in a short time without interrupts.
actuators had been done in CEHIP AR (Canal de
Experiencias Hidrodinamicas de El Pardo, At present, our group is working in the
Madrid; in English: El Pardo Model Basin) development of a software tool for control code
[CEHIP AR, 2000], vertical dynamic models of automatic generation. This kind of tools are
the ship had been obtained from them, suitable to solve the design of control software
SIMULINK simulation environment had been for reactive systems like the control experiments
tested and our work group had started to design with the replica. The tool runs on MS-Windows
the control algorithms. and generates C++ code that is supported by a
real time operating system called RTKemel.
The next logical step was to apply the control The results, during our experiments, have
on the scaled down replica with appendages. satisfied the expectations. We have proved the
279
efficiency of the tool to redesign quickly the 3. THE ENVIRONMENT FOR FAST
control software "in situ". Moreover we always CREATION OF REAL TIME CONTROL
could work with a stable control program that SYSTEMS
met the specifications.
As a result of the investigation performed by
This tool can be used in other experiments our group, a CASE tool called EdROOM has
related with ship control. Because of it we been development. This tool is capable to
consider interesting its presentation. This paper generate automatically real time control code
begins with a introduction to ROOM starting from a ROOM model of the system. It
methodology, after that we explain the tool includes a graphic editor to define the actor
developed and its utilization to solve the control structure of the model and the behaviour of its
experiments with the replica. Finally we present actors. The figure 1 shows part of the structure
the conclusions of this work of a model, made with EdROOM, that includes
5 actors and their interconnections. The figure 2
shows the behaviour of one of this actors edited
2. ROOM METHODOLOGY with EdROOM. The tool also lets define the
communication protocols, and the petitions to
ROOM (Real-Time Object Oriented Modelling) the communication, timing and scheduling
was introduced by Selic in 1994 [Selic, et aI., servIces.
1994 ]. It is a formalism for modelling real time
systems using the object oriented paradigm.
This paradigm was introduced in the beginnings
of SIMULA and SMALTALK languages and it
is characterized to confer behaviour to the
objects of computation. In a reactive system, the
object of the computation will be actors whose
behaviours cooperate to solve the system.
280
detenninism for working in a real time maximum rotation speed of the motors is 67.5 0/
conditions. Ms-DOS is not a real time system seg. This value corresponds with a 13.5 0/ seg
and it does not offer multitasking therefore we speed in the real ship.
avoid it.
The sensors located in the replica measure the
After we have constructed the graphic model of following variables: heave, pitch, the height of
the system and we have defined a few functions the arriving wave, the drag forces ( starboard
integrated with it, we can generate C++ control and port) and the accelerations in several points
code using a specific utility of EdROOM. This of the replica.
source code is compiled and linked with a
library called mv_rtk.lib which implements the The replica is moved by a carriage which has a
communication, timing and scheduling services complete installation of data acquisition devices
in RTKemel. The result is a compact executable and video cameras. The control is perfonned by
file which can run on a any PC with Ms-DOS. the industrial PC fixed to the carriage and
The implementation of mv_rtk.lib in other real located near the replica. The PC includes the
time operating systems lets us construct portable Advantech PCL812PG data acquisition card
and highly reusable code using EdROOM. connected to its bus. Six of its sixteen analog
inputs channels are used by the control program
to perfonn the periodic sampling and four of its
4. REPLICA CONTROL HARDWARE sixteen digital outputs manage the control of the
motors. Depends on the control algorithm it is
The Replica constructed by CEHIPAR is 4.5 necessary to sample a different set of signals.
meters long. It has incorporated a T-foil near the For example the first control studies have been
bow and two transom flaps. There is a step done using the WVA acceleration.
motor to move the T-foil wings and other to
move the Flaps. We have used an industrial PC The industrial PC has also a console in which
to control them. Figure 3 shows a picture of the we can observe the values of the main variables.
replica with its appendages. Figure 4 shows a This has been very useful in the calibration
picture of one of the step motors. tasks and to complete the basic control tests
(moving by hand the replica and checking if the
value of the motor rotation signs truly
counteract the wave effects).
8. ACKNOWLEDGEMENTS
9. REFERENCES
284
Abstract: This paper is about a research on the use of active appendages to smooth the
motions of fast ferries. According to BAZAN specifications, the research focused on a
fast ferry with a T-foil and transom flaps. A collaboration of three research groups, with
the experimental support of CEHIP AR was established to accomplish the objectives.
The research was scheduled as two main steps. A first step of control-oriented
modelling has successfully been achieved. The second step is dedicated to control
design and experimental evaluation, seeking for the best solution. First experimental
results confirm good expectations with the use of the active controlled appendages. This
paper describes the main aspects of the research: the problem to be solved, the
methodology and fulfilment of the research project, and the most relevant results
obtained. Copyright © 2001 IFA C
285
The T-foil and the flaps can move, to counteract the
effect of each incident wave. By means of a control
system, these appendages should move in the most
effective way: that means a problem of control
design. The objective of the research is to solve this
problem, following some important steps:
4. RESULTS
5. CONCLUSIONS
292
plane and that region is transfonned into the Hough
plane. Also using the position data of the AUV, the
interested region in the Hough plane is predicted and
the features of the cable image are extracted from the
Hough plane. The size of the region varies from frame
to frame and as a result, the image processing time will Two modes of operation are carried out to mltlal1ze the
not be a constant. Therefore, in order to compensate position at the cross point of the mark shown in Fig. 7.
the delays caused by image processing, a predictor The initial dive of the AUV is done by keeping the
based on the dynamics ofthe AUV which is explained heading direction, determined prior to the mission, along
in detail in (Balasuriya et at., 1998) is used as the delay the Y axis. Once the defined height of the AUV with
compensator. Once the instantaneous features of the respect to the bottom is reached the AUV switch into
cable is detennined, these features are verified to check the INITIAL SWAYING mode. Here, using visual
whether they represent the cable. The properties of the feedback the AUV is navigated in the direction parallel
cable features are; 1) there are no abrupt changes, and to X ax~s only with sway actuator. At this instant heading
2) number of pixels on the line feature should be greater and height references are kept constant. This mode is
than a defined threshold. If the features extracted by can:ied out until the image of the cable appears at the
image processing represents the properties of the cable honzontal centre of the image. In this task the camera is
then those 2D features are transfonned into the vehicle kept in a forward-down looking position. The camera
coordinate system for detennining the navigational axis is parallel to the centre axis of the AUV. Once the
references. Else the model line features are used for cable is at the centre ofthe image the X-axis position of
detennining the navigational references. The generation the AUV is made zero.
of navigational references are discussed in (Balasuriya
etal., 1997,1998).
311
compared with the actual sensor readings. If there is a discrepancy, the diagnostic engine is invoked to determine the
most likely cause of this discrepancy.
For diagnostics, the actual sensor values are propagated through the causal-network model to determine the faulty
component and its operational mode that would produce the abnormal sensor values. Conversely, for system
reconfiguration, the desired machine or process output values are propagated through the causal network to
determine the equipment settings necessary to achieve the desired output (subject to the constraint that certain
components are unavailable or are in failure mode). In this way, the same causal-network model can be used for
both diagnostics and for assisting control reconfiguration. When multiple diagnostic solutions or reconfiguration
solutions are found in the model, the best solution can be determined by evaluating probabilities or costs associated
with the various component modes or settings. Details on the model-based methodology for diagnosis and control
reconfiguration can be found in Provan and Chen (1999).
We can create causal-network models from a library of component models by interconnecting the components. For
example, the pump loop shown in Fig. 5 is composed of 6 component types: motor/pump, pipe, valve, tank, pressure
sensor and flowmeter. Rockwell Science Center has developed software tools for defming component models,
interconnecting the components to form causal-network models, and generating embeddable C code from the causal-
network models.