- Email: [email protected]
Lessons Learned from the Mobile Offshore Base Project
IFAC
Copyright" IFAC Mechatronic Systems, California, USA, 2002
c::
0
[>
Publications www.elsevier.comllocate/ifac
LESSONS LEARNED FROM THE MOBILE OFFSHORE BASE PROJECT Anouck R. Girard' , Daniel M. Empey, Stephen C. Spry and J. Karl Hedrick4 The University of California at Berkeley Ocean Engineering Graduate Group 230 Bechtel Engineering Center #1708 Berkeley, CA 9472~1708 IOcean Engineering Graduate Group. [email protected] lCalijornia PATH Experimental Group, [email protected] 3Departm ent of Mechanical Engineering, [email protected] 4Department of Mechanical Engineering, [email protected]
Abstract: The Mobile Offshore Base (MOB) is a large, self-propelled, floating, pre-positioned ocean structure formed of three to five modules and reaching up to 1,500 meters in length. The alignment of the modules is maintained through the use of thrusters and/or connectors. A simulation framework has been used to provide evaluation of different control strategies. We have conducted experiments in which three I: 150 scale modules are kept aligned by pivoting thrusters and form a miniature runway. This paper Copyright © addresses lessons learned in the course of the MOB project.
2002IFAC Keywords: Sliding Mode Control, Variable Structure Control, Ship Control, Marine Vehicles, Dynamic Positioning I.
interest to logistic fleets), to thruster assisted mooring and automated docking. In some concepts, the MOB is formed of multiple independent modules, which remain tightly aligned through the use of rotating thrusters. The DP system must maintain overall orientation and relative position between modules.
INTRODUCTION
The concept of a floating, at -sea base stems from the necessity for the United States to be able to !iage military and/or humanitarian operations in any part of the world[4] . The Mobile Offshore Base is a very large floating ocean structure meant to provide the same capabilities as an on-land army base. It must accommodate the landing and take-off of C-17 conventional aircraft, host 3000 troops, carry 10 million gallons of fuel and provide 3 million square feet of internal configurable storage [3]. The modules forming the MOB must be able to perform long-term station keeping at sea, in the presence of wavlS, winds and currents. This is usually referred to as Dynamic Positioning (DP) control [1,2]. Dynamic positioning is further complicated in the presence of multiple ships whose operation has to be coordinated, thus extending the scope of DP with maneuvers under tight constraints. The breadth of potential applications has significantly increased the interest in this mult~disciplinary technology. Applications range from the Mobile Offshore Base (MOB), to cargo transfer between ships (with special
A simulation framework has been implemented in SHIFT, a specification language for hybrid systems [5]. Libraries of mission scenarios, control techniques and strategies, platform and actuator models and disturbance models can be combined to create custom simulations. A laboratory testbed consisting of three scaled (J: 150) models of MOB platforms has been developed and is used to evaluate concepts and validate simulation results. This paper summarizes lessons learned during the MOB project at the University of California, Berkeley.
581
2. LESSONS LEARNED FROM DYNAMIC POSITIONING CONCEPTS AND THE CONTROL ARCHITECTURE
MANEUVER
'f
1:1 J: FAIL THRUSTER
GOTO(1D,J.,y.IKadu.&)
(be.t~kl.U.rll"cr )d)
Summary of Major Accomplishments Studied the problem domain. Split the motions and motion coordination ofplatforms into maneuvers. Associated a controller with each maneuver. Determined a layered structure for the controllers. Considered the problem of trajectory generation fo r each maneuver. Developed new thruster allocation algorithms. Studied coordinated motion strategies. Compared various controllers.
[i} ...-:,..... JOIN (IDI.1D2)
,-.IW
1!2iI '.
SEPARATE(I~
ID
Iiii} ALL MOVE (IDI, ID2, l,y, p.l)
IiiilfiwJ .--.fiiiJ1iiiiI ROTAT£(lOl, p.1)
il§Yli9@~ EiJLLl: MOB maneuvers.
'.' ',.
. .
;
The highest level o{ !he control system, hence, IS I.n charge of keeping track of which state the system IS in, and supervises the switching from one mode into another. Therefore the highest level of control has been coined the supervisory control system. The supervisory control system takes input commands from the operator/commander of the MOB, and translates these commands so that lower levels of control can take charge of communication between modules and stability and control of the modules themselves.
2.1 New. challenging problem The problem is much more difficult than just the traditional DP problem. Multiple platforms create a lot of new, difficult problems, involving the coordination of the different problems, such as assembly and separation of various platforms at sea, re-orientation of the assembled string of modules into the wind etc .. . In fact, the coordinated maneuvering of several ships/platforms/structures at sea has not been solved in general by the industry at this point in time.
Of particular importance in this project was a feature of the modeling called dynamical networks of hybrid automata. In systems formed of several networked vehicles, there are some problems that recur, such as maintaining the vehicles in a formation, tracking changes of the configuration with time, and joining/separating from the formation. In these cases, the software needs to be sufficiently flexible to accommodate for these changes. The most important thing is the ability to create and destroy, at run time, software components that control and model the vehicles being added/removed form the formation. One also wants to be able to create/destroy communication channels between components.
2.2 Maneuvers and the control architecture The difficulty of the problem we considered led us to the partitioning of the full coordinated DP problem into a number of simpler maneuvers. This allows us to make the problem tractable, and has advantages form the point of view of clarity, maintainability etc. .. However, the decoupling of problems into simpler, more manageable sub -problems does cause us to loose global optimality. The decomposition of the problem into maneuvers is organized in the control architecture. The maneuvers considered for the MOB are presented in figure2 .1.
As an example, consider the following scenario:
With the problem partitioned into several modes, it becomes useful to model the controllers in the framework of hybrid systems. Hybrid systems have gained widespread acceptance in the control systems community as a relevant formalism for the modeling of networked vehicle systems. They are formed of systems with both discrete and continuous variables. They are particularly well suited for systems that have several modes of operation (the maneuvers, such as assemble modules, rotate assembly into wind etc ... ) with discrete switches, or transitions, between modes.
EIIJ~_
2
3
--~ 2
8 Fig. 2.2: Dynamic MOB reconfiguration.
582
neighbors. At this point the stability of the formation is considered, as well be addressed further down in this section. The dynamic positioning control layer outputs a desired force and moment system for each module to meet the position setpoint determined by the setpoint generation layer. Further below, a layer termed thruster allocation logic deals with the cptimal use of the thrusters to meet the force and moment system chosen by the DP control layer. This layer directly considers the physics of the thrusters, as well as their orientation and thrust levels at each iteration, and optimally allocates the individ ual thrusters to deal with issues such as fuel consumption.
Initially, modules I, 2 and 3 are forming a MOB. Module 3 is separated to move to a separate, far away location. Once module 3 (shown in red) is operating autonomously at a safe distance, the control system for the MOB can delete the software model for module 3 and considers only modules I and 2 as forming the MOB. The communication links to module 3 are also destroyed as it becomes out of reach. At a later time, module 8 (shown in blue) joins the MOB. A software component for it is created, and it is used by the join and coordinated DP maneuvers. Communication links to this module are also dynamically added as the module is in range. Classical software mode ling environments, such as Matlab, do not have built in facilities for this kind of on line addition and destruction of com ponents. In this project, we used languages that had these features built in, which greatly simplified the software design and the modeling of the systems. The mode ling formalism used is termed dynamic networks of hybrid automata, and the software tools used to implement these are the SHIFT and TEJA software packages, as described in previous sections of this report.
2.4 Trajectory controller
generation
for
each
maneuver
In our framework, each maneuver controller has a trajectory generator, which decides on the best position and orientation for each module. This problem has not been considered at this point by the offshore industry that we know of. It was solved individually for each maneuver as presented in the report. The trajectory generators represent higher levels of control, and feed positions into the DP controllers, which are described in the following section.
2.3 Coordinated dynamic positioning is a layered problem
2.5 DP Controller organization
Coordinated DP can be conveniently addressed as a hierarchical problem.
We have organized our lower-level DP controller in a fairly standard fashion. The flow of information among elements is presented in the following figure:
Fig. 2.3: Hierarchical layers. . . . wiodforce
A possible decomposition into hierarchical layers is presented in the figure above. This only represents a subset of the layers considered at UCB during this project, in particular the supervision layers (dealing with changes in the configuration of the MOB) are not depicted, and neither are the lower level control layers dealing with the individual thruster commands. Nonetheless the figure is a proper abstraction if for now we restrict ourselves to the intermediated layers.
current force wave force
Fig. 2.4: Information flow. A detailed description of each element can be found in previous sections of this report. Here we recapitulate over one of the more pertinent developments used in this project, the thruster allocation logic. 2.6 Thruster allocation logic (TAL)
The set point generation layer concerns itself with the best way to maintain the alignment of the modules. Several strategies have been considered to compute the setpoints for each module, including a standard minimax strategy of minimizing the motion of all modules to form the alignment. One step below, the dynamic positioning control layer deals with the control of individual modules in relation to their
The thruster allocation problem consists of allocating commanded thrust and azimuth to multiple thrusters (more than two) in order to best achieve the desired force system on a ship or platform. The approach must allow for consideration of many of the realworld characteristics of thrusters, such as maximum
583
MIDDL&AS-LEADER
thrust capability, maximum slew rate of the thrusters etc . ..
--1-9"'~ ,~. 9...J---.. .
The problem is formulated as a linear programming problem, and the existence of a global minimum is guaranteed. Small variations are assumed so that the problem can be linearized. The formulation enables the consideration of constraints on the slew and thrust rates, which are not present in previous work. The objective function is a linear combination of force error, moment error and power consumptIOn. This development can also take advantage of a receding-horizon control scheme without significant modifications.
~:
Middl6-as-leader strategy.
2.7c Leader less strategy LEADERLESS ".cb
n~n~n~n~n
2.7 Concepts related to the coordinated DP of several ocean vehicles/structures
",.do,
INCb I"nrlfll
I .' .
INcb ~
rl/lrl"cr r, • .
>,«i.,
,pod.,
_u.r
'rlfdi .
;""ri.1
rl/"'''( I
Strings of automatically controlled vehicles can exhibit "string instabilities", i.e., disturbances In the front of the string are amplified as they are propagated upstream.
trwlu
1,,~rtI.1
;"~rrilll
INch ;" ,,,..1
rr/'""(I
rr/rrrru:r
Nlnn.«
Ei.&....L1: Leaderless strategy.
It was shown in [6) through linear transfer function analysis that these instabilities could be eliminated by the introduction of a common reference trajectory for all of the vehicles in the case of platoons of cars .. If all of the vehicles in the platoon had knowledge of the lead vehicle ' s absolute velocity then "weak string stability" could be achieved, i.e., no disturbance would ever be amplified as it traveled upstream in the platoon . Also, if all of the vehicles in the platoon had knowledge of the relative position error between themselves and the lead vehicle, then "strong string stability" could be achieved, i.e., all downstream disturbances could be geometrically attenuated as they traveled upstream in the platoon. The lead vehicle information needs to be communicated to all of the vehicles. These results can be generalized to 2 and 3 dimensional problems. We tested several different string control strategies in simulation for the MOB.
2.7a First as leader strategy
This approach does not suffer from string i~stability problems. A gain can be selected that modIfies the weight of the two different position references, inertial and relative. With high relative gains and low inertial gains, the assembly is more prone to drifting with disturbances. With high inertial and low relattve gains, the inertial pOSition. is maintain~d more closely, at the cost of potentIally poorer ahgnment. So far this gain has been determined emptrtcally. Further work could involve the determination of a fuel optimal value for this gain. 2.8 Comparison between different controllers Comparisons were made both in simulation and on the experiment of the behavior of different DP controllers. Several different trajectory generation schemes and several dynamic positioning control algorithms were tested, including model predictive control (MPC), linear quadratic control (LQ), dynamic surface sliding mode control (the Berkeley Backup Controller, or BBC), and a coordinated PID controller. Conceptually, the major differences between the controllers are as follows:
FlRST-AS-LEADER
~;;;Q
InIeU
Table I: Controller Differences.
' Q' Q. 9,..lm___
_ ..... s
Fig. 2.5: First-lIi-leader strategy.
Ex....tecl
PerformaBce
PID
LQ
SMC
MPC
+
++
+++
+++
* OP
*
-
0 ••
State of lIIe Art laduatry I.J:per\eIIce
This strategy suffers from the potential for string instabilities.
"No.-UMar
Aa __tate/,
DOIl-Iiaeari~
2.7b Middle as leader strategy
0pentIaa PoIDla MuItI-V.rlabJe
In this approach, the potential for string instabilities is half that of the first-as-Ieader approach . ..
"P...lble to . " .coIDpaaTAL . . .11....
584
OP
** ** * *
Other Fields Other Fields
several
several
* * *
3. LESSONS LEARNED FROM SIMULATION WORK AND THE SHIFT IMPLEMENTATION
Summary of Major Accomplishments • Decided on software architecture for simulation, • Implemented nonlinear 6DOF platform models, Developed and implemented disturbance models, including wind, waves, currents and solitons. Developed and implemented hierarchical control schemes, Compared various controllers in different disturbance scenarios.
Fig. 3.2: SHIFT language models. 3.3 Dynamic modeling ofplatforms Hydrodynamic coefficients for the platform models were obtained through the software package WAMIT. Difficulties that arose in this part of the project are related to the difficulty to obtain adequate documentation for the software tools. In particular, units and non-dimensionalization were non-trivial.
3.1 Use of dynamic network of hybrid automata model
3.4 Disturbance modeling
We developed a simulation environment for Mobile Offshore Base control concepts. The environment is complete with disturbance models, dynamic platform models and hierarchical controllers.
A large part of the effort invested in building MOBSHIFT went into the disturbance modeling. The first difficulty consisted of selecting meaningful models. Little data was available at the beginning of the MOB project for disturbances on areas pertinent to a structure such as a MOB. The statistical description of the ocean makes the programming tedious and error-prone. The early development of test cases for each disturbance would have saved some time. 4. LESSONS LEARNED FROM THE EXPERIMENTAL WORK AND REALTIME IMPLEMENTATION
EilL1.l:
Simulation environment. Summary ofMajor Accomplishments Designed a generic MOB experimental testbed, including o The design ofscaled thruster o The selection ofproper sensors o The design of the scaled models themselves o The design of the power system for the modules o The refurbishing of the pool at the Richmond Field Station facility Assem bled and wired the platforms • Developed a control system implementation, including o Choosing the number of processors and their speed o Setting up the inputs/outputs and connecting the computers on a network • Developing and testing the software, hardware and their interactions.
A main motivation for the development of our simulation environment, dubbed MOB-SHIFT, is that to adequately compare control concepts, things must be tested in a uniform environment. It makes no sense to compare data obtained by different teams if the teams are using different models to generate the data - this would in effect amount to comparing apples and oranges. Our simulation was developed in the SHIFT language. It contains numerous models, which are summarized in figure 3.2. 3.2 SHIFT Formalism SHIFT is a programming language for describing dynamic networks of hybrid automata. Such systems consist of components that can be created, interconnected and destroyed as the system evolves. Components exhibit hybrid behavior, consisting of continuous-time phases separated by discrete-event transitions. Components may evolve independently, or they may interact through their inputs, outputs and exported events. The interaction network itself may evolve.
585
4. I Implementation difficulties
5. LESSONS LEARNED FROM THE EV ALUA TION FRAMEWORK
The implementation of the hardware was rendered difficult by the fact that various elements were provided by different manufacturers, and the integration of these elements was a non-trivial task.
Summary ofMajor Accomplishments Decided on measures ofperformance for the MOB DP control schemes, Set up the MOJJSHlFT Evaluation framework, • Produced a test plan in the form of environment specification, including wind, waves, currents and soli tons. Applied the test plan various controllers in different disturbance scenarios.
4.2 Sensors: Laser (Global) and Ultrasonic (Relative) Positioning Systems The laser positioning system caused tremendous difficulties during the project. The sensors were selected for their simplicity and cost-effectiveness. Unfortunately, the system was subject to large variations in performance. It is difficult to say at this time whether an alternate sensor system such as GPS would have worked better, since we did not test one. The ultrasonic sensors gave accurate measurements. The initial setup was somewhat difficult and timeconsuming due to interference problems between the supporting electronics and other components on the modules. Serial communications from the ultrasonic sensors on the modules to the central computer were unreliable over long cable lengths, which somewhat limited the umbilical cord length.
Preliminary results from MOB-SHIFT served to compare DP contrQI concepts in a uniform fashion. If power (or fuel) oor1~umption is to be lSed as the principal measure of Pl;:rformance for MOB control concepts, it would be ~i:lequate to improve on the existing power models in MOB-SHIFT. Also, it seems the principal control difficulty may be in ramping/building up seas, not in steady-state sea states. Further testing along these lines would most certainly produce interesting results. 6. CONCLUDING REMARKS
4.3 Thrusters This project provided interesting and challenging issues that were framed in the context of real-world problems. The work presented here only s::ratches the surface of the issues surrounding the control of such multi-body systems, and much additional work remains to be conducted to develop optimal controllers for the system in question as well as for more generalized systems of mUltiple vehicles. The simulation environment and physical models developed for this system represent an important asset for the development of controllers for this type of system, and the authors feel that additional research using these tools is warranted.
Generally the thrusters functioned well. The motors, amplifiers and stepper motors were reliable and relatively trouble free . Thruster indicators were very useful in establis hing the behavior of the thrusters from the side of the pool or the overhead bridge. The use of a regular testing routine in software allowed the repeatable testing of the thrusters with good results and allowed the operators to diagnose thruster misbehavior, if occurring, from the start of experiments. This allowed big time savings.
4.4 Software Development and Testing Exhaustive testing of the software was done using a principled approach in which the software was incrementally developed and tested. This allowed for easier debugging. The amount of testing to be done in the tank remained relatively high due to the high number of sensors and actuators, which needed to work simultaneously. It is unclear at this point whether it is possible to make meaningful comparisons between different DP controller designs using the experiment, since the accuracy of the DP controllers seems to be limited by the accuracy of the sensor system. However, the motions observed in the tank mirror motions obtained in MOB-SHIFT simulations, and hence give more weight to controller comparison tests run in simulation. The main difference seems to be an increase in drag forces in the experiment compared to simulation. This could be a consequence of hull-bottom interactions.
7. ACKNOWLEDGEMENTS The material is based upon work supported by the U.S. Office of Naval Research's MOB Program under grant NOOOI4-98-1~744 and the Link Foundation. Many thanks to Joao Sousa who fIrSt suggested writing this paper.
8. REFERENCES [1] MJ. Morgan, M, Dynamic Positioning of Mobile Offshore Platforms. The Petroleum Publishing Company. Tuisa, Oklahoma,
1978. (2) T.!. Fossen, Guidance and Control of Ocean Vehicles. John Wiley and Sons, Inc., New York, 1994. (3) G. Remmers, R. Taylor, P. Palo and R Brackett, "Mobile Offshore Base: A Seabasing Option", Keynote Address, in Proc. of the Third International Workshop on Very Large Floating Structures, VLSF'99, pp. 1-7. (4) R. Zueck, P. Palo and R. Taylor, "Mobile Offshore Base: Research Spin-Offs", in Proc. of the 1999 ISOPE Conference, Brest, France, June 1999, pp 10-16. (5) www.path.berkeley.edulshift! (6) D. Swaroop and J.K. Hedrick, 1996, "String Stability of Interconnected Systems", IEEE Transactions on Automatic Control. Vo!. 41, No. 3, pp. 349-35.
586
Copyright" IFAC Mechatronic Systems, California, USA, 2002
c::
0
[>
Publications www.elsevier.comllocate/ifac
LESSONS LEARNED FROM THE MOBILE OFFSHORE BASE PROJECT Anouck R. Girard' , Daniel M. Empey, Stephen C. Spry and J. Karl Hedrick4 The University of California at Berkeley Ocean Engineering Graduate Group 230 Bechtel Engineering Center #1708 Berkeley, CA 9472~1708 IOcean Engineering Graduate Group. [email protected] lCalijornia PATH Experimental Group, [email protected] 3Departm ent of Mechanical Engineering, [email protected] 4Department of Mechanical Engineering, [email protected]
Abstract: The Mobile Offshore Base (MOB) is a large, self-propelled, floating, pre-positioned ocean structure formed of three to five modules and reaching up to 1,500 meters in length. The alignment of the modules is maintained through the use of thrusters and/or connectors. A simulation framework has been used to provide evaluation of different control strategies. We have conducted experiments in which three I: 150 scale modules are kept aligned by pivoting thrusters and form a miniature runway. This paper Copyright © addresses lessons learned in the course of the MOB project.
2002IFAC Keywords: Sliding Mode Control, Variable Structure Control, Ship Control, Marine Vehicles, Dynamic Positioning I.
interest to logistic fleets), to thruster assisted mooring and automated docking. In some concepts, the MOB is formed of multiple independent modules, which remain tightly aligned through the use of rotating thrusters. The DP system must maintain overall orientation and relative position between modules.
INTRODUCTION
The concept of a floating, at -sea base stems from the necessity for the United States to be able to !iage military and/or humanitarian operations in any part of the world[4] . The Mobile Offshore Base is a very large floating ocean structure meant to provide the same capabilities as an on-land army base. It must accommodate the landing and take-off of C-17 conventional aircraft, host 3000 troops, carry 10 million gallons of fuel and provide 3 million square feet of internal configurable storage [3]. The modules forming the MOB must be able to perform long-term station keeping at sea, in the presence of wavlS, winds and currents. This is usually referred to as Dynamic Positioning (DP) control [1,2]. Dynamic positioning is further complicated in the presence of multiple ships whose operation has to be coordinated, thus extending the scope of DP with maneuvers under tight constraints. The breadth of potential applications has significantly increased the interest in this mult~disciplinary technology. Applications range from the Mobile Offshore Base (MOB), to cargo transfer between ships (with special
A simulation framework has been implemented in SHIFT, a specification language for hybrid systems [5]. Libraries of mission scenarios, control techniques and strategies, platform and actuator models and disturbance models can be combined to create custom simulations. A laboratory testbed consisting of three scaled (J: 150) models of MOB platforms has been developed and is used to evaluate concepts and validate simulation results. This paper summarizes lessons learned during the MOB project at the University of California, Berkeley.
581
2. LESSONS LEARNED FROM DYNAMIC POSITIONING CONCEPTS AND THE CONTROL ARCHITECTURE
MANEUVER
'f
1:1 J: FAIL THRUSTER
GOTO(1D,J.,y.IKadu.&)
(be.t~kl.U.rll"cr )d)
Summary of Major Accomplishments Studied the problem domain. Split the motions and motion coordination ofplatforms into maneuvers. Associated a controller with each maneuver. Determined a layered structure for the controllers. Considered the problem of trajectory generation fo r each maneuver. Developed new thruster allocation algorithms. Studied coordinated motion strategies. Compared various controllers.
[i} ...-:,..... JOIN (IDI.1D2)
,-.IW
1!2iI '.
SEPARATE(I~
ID
Iiii} ALL MOVE (IDI, ID2, l,y, p.l)
IiiilfiwJ .--.fiiiJ1iiiiI ROTAT£(lOl, p.1)
il§Yli9@~ EiJLLl: MOB maneuvers.
'.' ',.
. .
;
The highest level o{ !he control system, hence, IS I.n charge of keeping track of which state the system IS in, and supervises the switching from one mode into another. Therefore the highest level of control has been coined the supervisory control system. The supervisory control system takes input commands from the operator/commander of the MOB, and translates these commands so that lower levels of control can take charge of communication between modules and stability and control of the modules themselves.
2.1 New. challenging problem The problem is much more difficult than just the traditional DP problem. Multiple platforms create a lot of new, difficult problems, involving the coordination of the different problems, such as assembly and separation of various platforms at sea, re-orientation of the assembled string of modules into the wind etc .. . In fact, the coordinated maneuvering of several ships/platforms/structures at sea has not been solved in general by the industry at this point in time.
Of particular importance in this project was a feature of the modeling called dynamical networks of hybrid automata. In systems formed of several networked vehicles, there are some problems that recur, such as maintaining the vehicles in a formation, tracking changes of the configuration with time, and joining/separating from the formation. In these cases, the software needs to be sufficiently flexible to accommodate for these changes. The most important thing is the ability to create and destroy, at run time, software components that control and model the vehicles being added/removed form the formation. One also wants to be able to create/destroy communication channels between components.
2.2 Maneuvers and the control architecture The difficulty of the problem we considered led us to the partitioning of the full coordinated DP problem into a number of simpler maneuvers. This allows us to make the problem tractable, and has advantages form the point of view of clarity, maintainability etc. .. However, the decoupling of problems into simpler, more manageable sub -problems does cause us to loose global optimality. The decomposition of the problem into maneuvers is organized in the control architecture. The maneuvers considered for the MOB are presented in figure2 .1.
As an example, consider the following scenario:
With the problem partitioned into several modes, it becomes useful to model the controllers in the framework of hybrid systems. Hybrid systems have gained widespread acceptance in the control systems community as a relevant formalism for the modeling of networked vehicle systems. They are formed of systems with both discrete and continuous variables. They are particularly well suited for systems that have several modes of operation (the maneuvers, such as assemble modules, rotate assembly into wind etc ... ) with discrete switches, or transitions, between modes.
EIIJ~_
2
3
--~ 2
8 Fig. 2.2: Dynamic MOB reconfiguration.
582
neighbors. At this point the stability of the formation is considered, as well be addressed further down in this section. The dynamic positioning control layer outputs a desired force and moment system for each module to meet the position setpoint determined by the setpoint generation layer. Further below, a layer termed thruster allocation logic deals with the cptimal use of the thrusters to meet the force and moment system chosen by the DP control layer. This layer directly considers the physics of the thrusters, as well as their orientation and thrust levels at each iteration, and optimally allocates the individ ual thrusters to deal with issues such as fuel consumption.
Initially, modules I, 2 and 3 are forming a MOB. Module 3 is separated to move to a separate, far away location. Once module 3 (shown in red) is operating autonomously at a safe distance, the control system for the MOB can delete the software model for module 3 and considers only modules I and 2 as forming the MOB. The communication links to module 3 are also destroyed as it becomes out of reach. At a later time, module 8 (shown in blue) joins the MOB. A software component for it is created, and it is used by the join and coordinated DP maneuvers. Communication links to this module are also dynamically added as the module is in range. Classical software mode ling environments, such as Matlab, do not have built in facilities for this kind of on line addition and destruction of com ponents. In this project, we used languages that had these features built in, which greatly simplified the software design and the modeling of the systems. The mode ling formalism used is termed dynamic networks of hybrid automata, and the software tools used to implement these are the SHIFT and TEJA software packages, as described in previous sections of this report.
2.4 Trajectory controller
generation
for
each
maneuver
In our framework, each maneuver controller has a trajectory generator, which decides on the best position and orientation for each module. This problem has not been considered at this point by the offshore industry that we know of. It was solved individually for each maneuver as presented in the report. The trajectory generators represent higher levels of control, and feed positions into the DP controllers, which are described in the following section.
2.3 Coordinated dynamic positioning is a layered problem
2.5 DP Controller organization
Coordinated DP can be conveniently addressed as a hierarchical problem.
We have organized our lower-level DP controller in a fairly standard fashion. The flow of information among elements is presented in the following figure:
Fig. 2.3: Hierarchical layers. . . . wiodforce
A possible decomposition into hierarchical layers is presented in the figure above. This only represents a subset of the layers considered at UCB during this project, in particular the supervision layers (dealing with changes in the configuration of the MOB) are not depicted, and neither are the lower level control layers dealing with the individual thruster commands. Nonetheless the figure is a proper abstraction if for now we restrict ourselves to the intermediated layers.
current force wave force
Fig. 2.4: Information flow. A detailed description of each element can be found in previous sections of this report. Here we recapitulate over one of the more pertinent developments used in this project, the thruster allocation logic. 2.6 Thruster allocation logic (TAL)
The set point generation layer concerns itself with the best way to maintain the alignment of the modules. Several strategies have been considered to compute the setpoints for each module, including a standard minimax strategy of minimizing the motion of all modules to form the alignment. One step below, the dynamic positioning control layer deals with the control of individual modules in relation to their
The thruster allocation problem consists of allocating commanded thrust and azimuth to multiple thrusters (more than two) in order to best achieve the desired force system on a ship or platform. The approach must allow for consideration of many of the realworld characteristics of thrusters, such as maximum
583
MIDDL&AS-LEADER
thrust capability, maximum slew rate of the thrusters etc . ..
--1-9"'~ ,~. 9...J---.. .
The problem is formulated as a linear programming problem, and the existence of a global minimum is guaranteed. Small variations are assumed so that the problem can be linearized. The formulation enables the consideration of constraints on the slew and thrust rates, which are not present in previous work. The objective function is a linear combination of force error, moment error and power consumptIOn. This development can also take advantage of a receding-horizon control scheme without significant modifications.
~:
Middl6-as-leader strategy.
2.7c Leader less strategy LEADERLESS ".cb
n~n~n~n~n
2.7 Concepts related to the coordinated DP of several ocean vehicles/structures
",.do,
INCb I"nrlfll
I .' .
INcb ~
rl/lrl"cr r, • .
>,«i.,
,pod.,
_u.r
'rlfdi .
;""ri.1
rl/"'''( I
Strings of automatically controlled vehicles can exhibit "string instabilities", i.e., disturbances In the front of the string are amplified as they are propagated upstream.
trwlu
1,,~rtI.1
;"~rrilll
INch ;" ,,,..1
rr/'""(I
rr/rrrru:r
Nlnn.«
Ei.&....L1: Leaderless strategy.
It was shown in [6) through linear transfer function analysis that these instabilities could be eliminated by the introduction of a common reference trajectory for all of the vehicles in the case of platoons of cars .. If all of the vehicles in the platoon had knowledge of the lead vehicle ' s absolute velocity then "weak string stability" could be achieved, i.e., no disturbance would ever be amplified as it traveled upstream in the platoon . Also, if all of the vehicles in the platoon had knowledge of the relative position error between themselves and the lead vehicle, then "strong string stability" could be achieved, i.e., all downstream disturbances could be geometrically attenuated as they traveled upstream in the platoon. The lead vehicle information needs to be communicated to all of the vehicles. These results can be generalized to 2 and 3 dimensional problems. We tested several different string control strategies in simulation for the MOB.
2.7a First as leader strategy
This approach does not suffer from string i~stability problems. A gain can be selected that modIfies the weight of the two different position references, inertial and relative. With high relative gains and low inertial gains, the assembly is more prone to drifting with disturbances. With high inertial and low relattve gains, the inertial pOSition. is maintain~d more closely, at the cost of potentIally poorer ahgnment. So far this gain has been determined emptrtcally. Further work could involve the determination of a fuel optimal value for this gain. 2.8 Comparison between different controllers Comparisons were made both in simulation and on the experiment of the behavior of different DP controllers. Several different trajectory generation schemes and several dynamic positioning control algorithms were tested, including model predictive control (MPC), linear quadratic control (LQ), dynamic surface sliding mode control (the Berkeley Backup Controller, or BBC), and a coordinated PID controller. Conceptually, the major differences between the controllers are as follows:
FlRST-AS-LEADER
~;;;Q
InIeU
Table I: Controller Differences.
' Q' Q. 9,..lm___
_ ..... s
Fig. 2.5: First-lIi-leader strategy.
Ex....tecl
PerformaBce
PID
LQ
SMC
MPC
+
++
+++
+++
* OP
*
-
0 ••
State of lIIe Art laduatry I.J:per\eIIce
This strategy suffers from the potential for string instabilities.
"No.-UMar
Aa __tate/,
DOIl-Iiaeari~
2.7b Middle as leader strategy
0pentIaa PoIDla MuItI-V.rlabJe
In this approach, the potential for string instabilities is half that of the first-as-Ieader approach . ..
"P...lble to . " .coIDpaaTAL . . .11....
584
OP
** ** * *
Other Fields Other Fields
several
several
* * *
3. LESSONS LEARNED FROM SIMULATION WORK AND THE SHIFT IMPLEMENTATION
Summary of Major Accomplishments • Decided on software architecture for simulation, • Implemented nonlinear 6DOF platform models, Developed and implemented disturbance models, including wind, waves, currents and solitons. Developed and implemented hierarchical control schemes, Compared various controllers in different disturbance scenarios.
Fig. 3.2: SHIFT language models. 3.3 Dynamic modeling ofplatforms Hydrodynamic coefficients for the platform models were obtained through the software package WAMIT. Difficulties that arose in this part of the project are related to the difficulty to obtain adequate documentation for the software tools. In particular, units and non-dimensionalization were non-trivial.
3.1 Use of dynamic network of hybrid automata model
3.4 Disturbance modeling
We developed a simulation environment for Mobile Offshore Base control concepts. The environment is complete with disturbance models, dynamic platform models and hierarchical controllers.
A large part of the effort invested in building MOBSHIFT went into the disturbance modeling. The first difficulty consisted of selecting meaningful models. Little data was available at the beginning of the MOB project for disturbances on areas pertinent to a structure such as a MOB. The statistical description of the ocean makes the programming tedious and error-prone. The early development of test cases for each disturbance would have saved some time. 4. LESSONS LEARNED FROM THE EXPERIMENTAL WORK AND REALTIME IMPLEMENTATION
EilL1.l:
Simulation environment. Summary ofMajor Accomplishments Designed a generic MOB experimental testbed, including o The design ofscaled thruster o The selection ofproper sensors o The design of the scaled models themselves o The design of the power system for the modules o The refurbishing of the pool at the Richmond Field Station facility Assem bled and wired the platforms • Developed a control system implementation, including o Choosing the number of processors and their speed o Setting up the inputs/outputs and connecting the computers on a network • Developing and testing the software, hardware and their interactions.
A main motivation for the development of our simulation environment, dubbed MOB-SHIFT, is that to adequately compare control concepts, things must be tested in a uniform environment. It makes no sense to compare data obtained by different teams if the teams are using different models to generate the data - this would in effect amount to comparing apples and oranges. Our simulation was developed in the SHIFT language. It contains numerous models, which are summarized in figure 3.2. 3.2 SHIFT Formalism SHIFT is a programming language for describing dynamic networks of hybrid automata. Such systems consist of components that can be created, interconnected and destroyed as the system evolves. Components exhibit hybrid behavior, consisting of continuous-time phases separated by discrete-event transitions. Components may evolve independently, or they may interact through their inputs, outputs and exported events. The interaction network itself may evolve.
585
4. I Implementation difficulties
5. LESSONS LEARNED FROM THE EV ALUA TION FRAMEWORK
The implementation of the hardware was rendered difficult by the fact that various elements were provided by different manufacturers, and the integration of these elements was a non-trivial task.
Summary ofMajor Accomplishments Decided on measures ofperformance for the MOB DP control schemes, Set up the MOJJSHlFT Evaluation framework, • Produced a test plan in the form of environment specification, including wind, waves, currents and soli tons. Applied the test plan various controllers in different disturbance scenarios.
4.2 Sensors: Laser (Global) and Ultrasonic (Relative) Positioning Systems The laser positioning system caused tremendous difficulties during the project. The sensors were selected for their simplicity and cost-effectiveness. Unfortunately, the system was subject to large variations in performance. It is difficult to say at this time whether an alternate sensor system such as GPS would have worked better, since we did not test one. The ultrasonic sensors gave accurate measurements. The initial setup was somewhat difficult and timeconsuming due to interference problems between the supporting electronics and other components on the modules. Serial communications from the ultrasonic sensors on the modules to the central computer were unreliable over long cable lengths, which somewhat limited the umbilical cord length.
Preliminary results from MOB-SHIFT served to compare DP contrQI concepts in a uniform fashion. If power (or fuel) oor1~umption is to be lSed as the principal measure of Pl;:rformance for MOB control concepts, it would be ~i:lequate to improve on the existing power models in MOB-SHIFT. Also, it seems the principal control difficulty may be in ramping/building up seas, not in steady-state sea states. Further testing along these lines would most certainly produce interesting results. 6. CONCLUDING REMARKS
4.3 Thrusters This project provided interesting and challenging issues that were framed in the context of real-world problems. The work presented here only s::ratches the surface of the issues surrounding the control of such multi-body systems, and much additional work remains to be conducted to develop optimal controllers for the system in question as well as for more generalized systems of mUltiple vehicles. The simulation environment and physical models developed for this system represent an important asset for the development of controllers for this type of system, and the authors feel that additional research using these tools is warranted.
Generally the thrusters functioned well. The motors, amplifiers and stepper motors were reliable and relatively trouble free . Thruster indicators were very useful in establis hing the behavior of the thrusters from the side of the pool or the overhead bridge. The use of a regular testing routine in software allowed the repeatable testing of the thrusters with good results and allowed the operators to diagnose thruster misbehavior, if occurring, from the start of experiments. This allowed big time savings.
4.4 Software Development and Testing Exhaustive testing of the software was done using a principled approach in which the software was incrementally developed and tested. This allowed for easier debugging. The amount of testing to be done in the tank remained relatively high due to the high number of sensors and actuators, which needed to work simultaneously. It is unclear at this point whether it is possible to make meaningful comparisons between different DP controller designs using the experiment, since the accuracy of the DP controllers seems to be limited by the accuracy of the sensor system. However, the motions observed in the tank mirror motions obtained in MOB-SHIFT simulations, and hence give more weight to controller comparison tests run in simulation. The main difference seems to be an increase in drag forces in the experiment compared to simulation. This could be a consequence of hull-bottom interactions.
7. ACKNOWLEDGEMENTS The material is based upon work supported by the U.S. Office of Naval Research's MOB Program under grant NOOOI4-98-1~744 and the Link Foundation. Many thanks to Joao Sousa who fIrSt suggested writing this paper.
8. REFERENCES [1] MJ. Morgan, M, Dynamic Positioning of Mobile Offshore Platforms. The Petroleum Publishing Company. Tuisa, Oklahoma,
1978. (2) T.!. Fossen, Guidance and Control of Ocean Vehicles. John Wiley and Sons, Inc., New York, 1994. (3) G. Remmers, R. Taylor, P. Palo and R Brackett, "Mobile Offshore Base: A Seabasing Option", Keynote Address, in Proc. of the Third International Workshop on Very Large Floating Structures, VLSF'99, pp. 1-7. (4) R. Zueck, P. Palo and R. Taylor, "Mobile Offshore Base: Research Spin-Offs", in Proc. of the 1999 ISOPE Conference, Brest, France, June 1999, pp 10-16. (5) www.path.berkeley.edulshift! (6) D. Swaroop and J.K. Hedrick, 1996, "String Stability of Interconnected Systems", IEEE Transactions on Automatic Control. Vo!. 41, No. 3, pp. 349-35.
586