Professional Documents
Culture Documents
79
© 1999 Kluwer Academic Publishers. Printed in the Netherlands.
JOHN N. LYGOURAS
Electronics and Digital Systems Laboratory, Department of Electrical and Computer Engineering,
67100 Xanthi, Greece
Key words: remotely operated vehicle (ROV), underwater technology, DC thrusters, integral wind-
up compensation, position control.
1. Introduction
In the past few years, the use of Remotely Operated Vehicles (ROV) has rapidly
increased due to the development of these vehicles to perform operations in deeper
and riskier areas where human divers cannot reach. Applications of ROVs include
ocean surveying, maintenance and construction of underwater structures, mainte-
nance of nuclear plants, water pollution measurements, etc. [1–6]. The design of
a high performance position control system for an ROV is of interest both from
the view of motion stabilisation as well as manoeuvring and tracking performance.
The dynamic response provided by such controllers should satisfy a set of strict
specifications in terms of speed, precision, overshoot, and interactions among the
controlled and/or uncontrolled coordinates. The controller should also be able to
cope with environmental disturbances, such as sea current and turbulence near
subsea structures, normally acting on the ROV.
Several ROV controlling methods have been proposed in the literature. It is
well known that conventional controllers with fixed gains do not guarantee high
quality response of the overall system when significant changes occur in the vehicle
dynamics. Most of the controlling methods are designed on the basis to reduce
the inherent coupling between the vehicle response modes that naturally exist in
ROVs. The approach leads to a set of separate designs for the steering, diving, and
speed control systems. Simulations are then executed based on the dynamics of the
vehicle to illustrate the robustness and validity of the concept.
Among the variety of possible control strategies proposed in the relevant papers,
two of them, the conventional linear P-PI controller is preferable for its simplicity
and has been shown to give good results in position control, at least when the vehi-
cle parameters are constant and no external disturbance exists. On the other hand,
the state-feedback Variable Structure Control (VSC) algorithm gives better results
since precise modelling of the ROV is not needed and unmodelled perturbations
can be effectively rejected [7, 8]. Recently, neural network-based controllers for
Autonomous Underwater Vehicles (AUV’s) have been proposed [9], and fuzzy
logic has been adopted to control the ballast of Unmanned Undersea Vehicles
(UUV’s) [10]. Most of the above control schemes, however, are not implemented
in real ROV models and their performance is verified only by simulation.
Although conventional controllers have some disadvantages as it was mentioned
above, they are preferable in many cases due to the simplicity of implementation. In
this paper, the P-PI controller with an integral antiwind-up compensator has been
used for position control of an ROV for every degree of freedom. It consists of an
inner velocity control loop and an outer positioncontrol loop. Another advantage of
the P-PI controller is its ability to reject step disturbances, i.e., to eliminate offset
provoked by step disturbance by means of the integral action of the inner loop [7].
An experimental model, 1.4 m long underwater ROV using DC thrusters (com-
bination of DC motor with a propeller) has been constructed. The model is a
remotely controlled submersible vehicle, rectangular in shape and composed of
a Ferrum-Aluminum framework which encloses and supports all components. The
configuration of the vehicle is of an open-frame type with buoyant material on its
top. The water flow around the vehicle of this type is complex, and it is very difficult
to estimate its hydrodynamic characteristics by computation. So, hydrodynamic
tests provide the only way to determine the derivatives which are necessary for the
evaluations of manoeuvrability.
and wv are the linear and angular velocity of the vehicle, respectively, and ur , wr
are the linear and angular velocity with respect to the fluid, respectively.
This is the general model for the ROV system. In fact the model is similar in
structure to the models for objects which are not underwater. The main difference
is caused by the presence of the additional mass and the drag forces for underwater
vehicles. The external force is given as the sum of the hydrodynamic force Ff , grav-
ity force Fg , thrust Ft and the tension Fc of the tether cable at the cable termination
point. Thus:
F = Ff + Fg + Ft + Fc = Ff + Fb + Fw + 6Fti + Fc , (2)
where Fb is the buoyancy, Fw is the weight force of the vehicle, and Fti is thrust
of the ith thruster. The general ROV model is simplified by many authors not
considering, for example, the cable tension Fc (considering the influence of the
cable on ROV as disturbances) since the forces and the torques arising from the
cable are difficult to specify. However, if the external forces due to the cable are to
be considered, a number of force models can be found in the literature [11].
When the dynamics of the ROV and the thrusters are considered, a complex
nonlinear system is obtained, even when the craft dynamics is assumed to be linear
(for instance, hover), due to the nonlinear features of the thrusters at low speed op-
eration. Thus, when a control system for the thrusters of an ROV is to be designed
one must take into account all the existing nonlinearities.
To apply linear control theory, it is necessary to obtain a linear model for the
ROV. The nonlinear model describing an ROV can be written as follows [12, 13]:
ẋ = f (x, u), (3)
where x is a state vector, and u is the input, which are forces and moments pro-
duced by the thrusters. Given an operating point: x0 = (u0 , v0 , w0 , p0 , q0 , r0 )T
with inputs: Th = (t10 , t20 , t30 , t40 , t50 , t60 )T , a small variation around this can be
approximated by a linear model usually written as:
δ ẋ = (∂f/∂x) x=x δx + (∂f/∂u) x=x δu (4)
0 0
u=u0 u=u0
r denote the three components of the absolute angular velocity W in the Oxyz
system and ϑ, ϕ and φ are the Euler angles.
Considering an input of the form:
u = u0 + δu (7)
where u0 is the nominal input, and δu is the difference between the actual and
nominal, we can evaluate x0 = f (x0 , u0 ). Then, nominal input and nominal state
should satisfy the ROV nonlinear model, namely
so x0 can be obtained through the above equation. In this case, the range in which
the linear model coincides with the nonlinear one can be investigated. Based on the
simulation results of linear velocities of the ROV given in [11], this range can be
limited as follows: if U10 , W10 , q10 and Un , Wn , qn , the velocity u, w and q from the
linear model with zero inputs as the nominal one and the ones from the nonlinear
model, respectively, the ROV nonlinear model can be approximated by a linear one
with satisfactory precision, i.e., U10 ∼= Un and q10 ∼ = qn if
Un < 0.5 m/s, wn < 0.02 m/s and qn < 0.03 rad/s
and W10 ∼
= Wn and q10 ∼
= qn if
Un < 0.01 m/s, wn < 0.2 m/s and qn < 0.012 rad/s
if the above conditions are satisfied, the nonlinear model can be approximated by
the linear one with satisfactory precision.
There are practical reasons why the dynamic models of DC motors cannot
be applied directly to model the motors of the DC thrusters. Although many of
the characteristic parameters are provided by the motor manufacturer, there are
parameters (e.g., the moment of inertia of the load, frictional torque, the dumping
constant, the drag forces, the gravity and buoyancy forces and moments) that must
be obtained experimentally after the motors are built into the vehicle.
Underwater vehicles are generally equipped with thrusters which consist of a
propeller driven by a torque source τ . In normal operation, the rotational velocity
of the propeller n and the advance velocity of the vehicle, VA , are both positive (first
quadrant operation). The load torque Q from the propeller and the thrust force T
are then usually written as:
characteristics are quite similar to those in the first quadrant. In general, however,
the thrust and torque coefficients are not equal in these two quadrants (i.e., sym-
metrical), since positive thrust is usually greater than negative thrust. It is assumed
that an algebraic relation, although complicated, exists between the thrust Ti of
propeller i and the physical input. Therefore, the thrust will be chosen as input in
the system:
ui = Ti . (11)
This means that the system will be linear in the inputs. In order to adjust the output
torque of each thruster the motor’s speed must be controlled.
The control system for the ROV thrusters has to consider two aspects. The
first one is related to the nonlinear character of its dynamics, as it is described by
Equations (9) and (10). The influence of the nonlinear term |n|n is reflected in time
responses of the thrusters to different amplitude step functions. For larger inputs
the response is more rapid so that at hover, where the nominal force required is
minimum, the time response of the thruster is higher. The second aspect is related
to the operating condition of the thrusters at hover.
One major problem that dc motors introduce, when driven by low input voltages
directly or using a PWM driver circuit, is that of dead-band nonlinearity. Consider-
ing as static this nonlinearity, it is cancelled in our application by specially designed
logarithmic networks. These circuits are designed in a way to present a nonlinear
transfer characteristic, symmetric to the axis y = x, to the static nonlinearity of
the dc motor. When such a circuit is connected in series with the dc motor driver
circuit, the dead-band nonlinearity of the motor is cancelled [14].
(a)
(b)
Figure 1. (a) The THETIS ROV propeller placement and (b) the Position Error Transforma-
tion (PET) technique to drive the system.
vehicle and to represent the scanned profile on the controlling computer screen.
The scanning ultrasonic tranceiver can be stopped in any desired direction if an
obstacle has been detected. The actual position in the 3-D space of the ROV can be
derived using combined information of the above sensors.
In Figure 1(a) the THETIS ROV considered is equipped with four propellers
(p1 , p2 , p3 , p4 ) which allow us to control the vehicle on the horizontal plane
(xh , yh ), the depth zh , and the heading φ. The remaining degrees of freedom (pitch
θ and roll ϕ angles) cannot be directly controlled. However, the system uses passive
stability keeping the metacentric height of the vehicle sufficiently large resulting in
small pitch and roll angles.
The ROV control inputs, denoted as uj (j = x, y, z, φ) are the signals that
are applied to the command unit of the ROV propellers. In manual control, those
inputs are produced by the tele-operator through a joystick. Using the Position
Error Transformation (PET) technique to drive the system, (Figure 1(b)) the posi-
tion error and the ROV velocity measured in inertial coordinates are transformed
to body coordinates and the controller gives directly the ROV control inputs uj .
Assuming that θ and ϕ are zero, the position error in body coordinates is given by
the following transformation:
[ex ey ez ]T = Tz (φ)(Pref − P), eφ = φref − φ, (12)
" #
cos φ sin φ 0
Tz (φ) = − sin φ cos φ 0 . (13)
0 0 1
A block diagram of the digital position control system is shown in Figure 2. It is
composed of the power circuit, the controlled thruster, the controlling DSP, suitable
sensors and the appropriate interface. Using the ultrasonic scanning system the
actual distance of the vehicle from the surrounding environment is displayed. If a
stable target is detected, the vehicle is capable of keeping a constant distance xd
from it. The respective thrusters are then actuated in a direction so that to make the
position error xd −xr zero. In the same way the heading φr is read from a gyroscope
and the respective thrusters P1 and P2 can be actuated to turn the vehicle about the
z axis.
Figure 2. Ultrasonics, depth sensor, and a gyrocompass-based ROV position control system.
Type DC motor
The power circuit that allows us to control the vehicle on the horizontal plane
(xh , yh ) consists of two dc thrusters P1 and P2 which can be operated together
or separately driven by four quadrant choppers and using the PWM technique.
Four-quadrant operation is necessary for accelerating and braking the motors in
both directions. The choppers operate at a fixed frequency of 10 KHz with a vari-
able duty cycle controlled by an input voltage (−10 V < Vin < + 10 V). Specially
designed D/A converters are used to convert the desired control word into the
corresponding input voltage for the chopper. The characteristics of the thrusters
used in our experiments are listed in Table I.
We have made the following assumptions on the position control system:
(a)
(b)
Figure 3. The implemented position control schemes: (a) position and velocity feedback and
(b) integral antiwind-up compensator.
is simple to include mechanisms for the exact adjustments of the integral term into
digital controllers.
To overcome the integral wind-up problem of the speed controller with PI-
action, integral antiwind-up compensation is used and extensively tested. Accord-
ing to this technique, the controller employs the PI-action when the magnitude of
the velocity error is smaller than a prescribed value. Otherwise, the controller gen-
erates the maximum permissible control signal with the integrator being forced to
reset. The design, which is experimentally verified, proves that the speed controller
of this type provides a fast settling time property with small overshoots regardless
of the magnitude of the reference input signal. In this way, the integral wind-up is
prevented and also the maximum ratings of hardware components are utilized.
The purpose of the speed control loop is to reduce the effects of hydrodynamic
and drag forces and also to compensate for the load carried by the thruster. The
vehicle’s actual speed is obtained by numerical differentiation of the translational
displacement along the axis during a fixed time interval. The required position
and velocity accuracy depend on the specific application. Although this method
does not provide so accurate results, since position measurements can be noisy,
it is the simplest way since it is difficult to derive the linear and angular speed
of the vehicle in all directions. Moreover, the accuracy achieved by the ultrasonic
scanning system (1 cm) is enough for our application. The number representing the
speed for the ith sampling period is given by:
[u(i) − u(i − 1)]
v(i) = KuF , (14)
1t
where KuF is the feedback gain of the speed.
The implementation of the PI-controllers for the speed control and current con-
trol loops is achieved by using the incremental trapezoidal integration algorithm
which computes the integrals of the errors with an increased accuracy:
vu = vu−1 + 1vu (15)
T T
1vu = Kx eu 1 + − eu−1 1 − (16)
2Ti 2Ti
The gain of the outer position controller has been estimated using the Zigler–
Nichols technique operating the system only with the position control loop. The
selection of time constants of the PI controllers is based on the conventional analy-
sis.
The transfer function of this type of controller is given by the following equa-
tion:
Ku (eu (t) + (1/Tu )I ), for eu (t) 6 k,
u(t) = Emax and I = 0 for eu (t) > k, (17)
−Emax and I = 0 for eu (t) < −k,
where
The above algorithm describes the bang-bang control action which in a block
diagram is shown in Figure 3(b). The desired speed here is compared with the
actual speed resulting from differentiation of the distance signal derived from the
ultrasonic scanning system. If the absolute value of the error is less than or equal
to the prescribed value k then the error is given via the PI controller to the motor.
When the error is greater than k, then the maximum allowable voltage is given
to the motor while keeping the integrator reset. From the implementation of the
above control algorithm two significant advantages are resulting. The first one is
the minimization of response time of the system while keeping the overshoot low,
and the second is avoidance of the so-called Integral Wind-up phenomenon. The
threshold k used in the PI speed controller is experimentally determined as in the
other relevant applications.
5. Experimental Results
The modelling, design and control of an open-frame underwater remotely operated
vehicle have been described in this paper. Under the assumption that the linear
velocity of the vehicle along the axis of motion is low, the ROV model can be
considered as linear. This assumption is not far away from reality since the velocity
of this kind of vehicles is rarely higher than 0.5 m/s. The flow round the vehicle
of this type is complex, and it is difficult to estimate its hydrodynamic charac-
teristics by simulations. So, hydrodynamic tests were conducted to determine the
derivatives which are necessary for the evaluations of manoeuvrability. Since the
thruster parameters are unknown, experimental data are acquired to identify the
continuous-time model. The order of the model is chosen to ensure acceptable ac-
curacy, without increasing the complexity of the controller. Velocity step response
of the dc thruster was made to identify the transfer function from pulse width to
velocity. Velocity measurements are taken every 5 ms. The dominant time constant
of the motor is found to be 120 ms. The motor exhibits nonlinear saturation at the
maximum velocity (4000 rpm in no load conditions, 1500 rpm inside the water)
and a frictional dead zone at small command inputs (|Vin | < 2.0 V).
The above described control schemes in Figure 3 have been experimentally im-
plemented and extensively tested. The ROV system has been tested in a water-pool,
in the laboratory of Hydraulics. Artificial water-pool bed profile on the controlling
computer screen, using the ultrasonic scanning system, is illustrated in Figure 4.
This image enables the user to know the distances of the objects surrounding the
Figure 4. Artificial water-pool bed profile on the controlling computer screen, using the
ultrasonic scanning system.
vehicle on a plane including the X-axis and rotated at an angle β relatively to the
x-z plane. The scanning transceiver can be rotated round the x axis and in this way
it can scan the space around the vehicle over almost a hemisphere. The two-way
scanning time for the ultrasonic transducer is about 1.0 s.
Step input response is taken moving the vehicle longitudinally along the x axis.
Nonlinearities in the control system (motor saturation and position quantization)
have been neglected in the controller design. The nonlinearity caused by friction
and moment of inertia of the rotor (dead-band nonlinearity) is cancelled using the
circuit described in reference [14] with very satisfactory results. Controller gains
have been selected to satisfy the design goals of minimum overshoot and settling
time. Experimental results showing the response of the P-PI controller to step
inputs in the x direction are shown in Figure 5. Obviously, the last response is
a better one since it has the minimum response time and smaller overshoot. The
satisfying system response is achieved with small overshoot while no disturbances
exist.
6. Conclusions
The position controller using PI plus integral antiwind-up compensation has been
implemented and successfully tested for control of the thrusters of the constructed
ROV. Initial experimental results have been presented in this paper. This control
law is simple and can be very easily implemented. Even though the presented ROV
is essentially an experimental, test-bed vehicle, it is still capable of performing
(a)
(b)
Figure 5. Step input response of the three tested configurations: (a) large signal step response
for the case where position and velocity feedback is used with P and PI controller, respectively.
Here, Kx = 13, Ku = 1.1 and Kx = 21.5, Tu = 20; (b) position, velocity and current
feedback with P, PI and PI controllers, respectively. Here Kx = 10, Ku = 1.1, Tu = 30,
Kc = 1, Tc = 70 and Kx = 12, Ku = 1.0, Tu = 25, Kc = 1.1, Tc = 60 and (c) position and
speed feedback with P, PI controllers, respectively, with integral antiwind-up compensation.
Here, Kx = 15, Ku = 0.8, Tu = 5 and n = 0.55 V.
various shallow water tasks. It is equipped with a B/W video-camera and other
suitable sensors and its primary use, at the present stage of development, is to
perform water pollution measurements. It is a low-cost, efficient and reliable ve-
hicle which can prove a good subsea support system for underwater site surveys
(wrecks, coral, shells, pipelines, etc.), in heavily polluted water up to 100 m deep.
It can also be used for surface monitored underwater tasks as well as diver support,
(c)
(d)
(e)
Figure 5. (Continued.)
thus improving diver efficiency and safety. A photograph of the constructed ROV
is shown in Figure 6.
References
1. Choi S. K., Yuh J., and Takashige G. Y.: Development of the omni-directional intelligent
navigator, IEEE Robotics Automat. Magazine (March 1995), 44–52.
2. Blidberg D. R.: Autonomous underwater vehicles: A tool for the ocean, Unmanned Systems
(1991), 10–15.
3. Yuh J.: Underwater Robotic Vehicles: Design and Control, TSI Press, Albuquerque, NM
(1995).
4. von Alt C., Allen B., Austin T., and Stokey R.: Remote environmental measuring units, IEEE
Oceanic Engineering Society Newsletter (1995), 9–14.
5. Leonard N. E.: Control synthesis and adaptation for an underactuated autonomous underwater
vehicle, IEEE J. Oceanic Engrg. 20(3) (1995), 211–220.
6. von Alt C. J. and Grassle J. E. LEO-15: An unmanned long term environmental observatory,
in: Proc. of the Oceans ’92, Vol. II, Fall, 1992, pp. 849–854.
7. Cunha J. P. V. S., Costa R. R., and Hsu L.: Design of a high performance variable structure
position control of ROV’s, IEEE J. Oceanic Engrg. 20(1) (1995), 42–54.
8. Healey A. J. and Lienard D.: Multivariable sliding mode control for autonomous diving and
steering of unmanned underwater vehicles, IEEE J. Oceanic Engrg., 18(3) (1993), 327–338.
9. Ishii K., Fujii T., and Ura T.: An on-line adaptation method in a neural network-based control
system for AUV’s, IEEE J. Oceanic Engrg., 20(3) (1995), 221–228.
10. DeBitetto P.A.: Fuzzy logic for depth control of unmanned undersea vehicles, IEEE J. Oceanic
Engrg. 20(3) (1995), 327–338.
11. Qiao H.: On the simulation of a ROV system, Industrial Control Unit, ICU/ 333, University of
Strathclyde, 1991, pp. 1–51.
12. Liceaga-Castro E.: A model for remotely operated vehicles, Industrial Control Unit., ICU/325,
University of Strathclyde, 1991, pp. 1–36.
13. Liceaga-Castro E.: A strategy for rov-robot control systems, Industrial Control Unit., Rep. 332,
University of Strathclyde, 1991, pp. 1–20.
14. Lygouras J. N., Tarchanidis K. N., Tsalides P. G., and Dimitriadis C. M.: Non-linear circuit
cancels the nonlinearity of a DC motor, Internat. J. Electronics 84(2) (1998), 147–156.
15. Warwick K. and Rees D.: Industrial Digital Control Systems, IEEE Control Engineering Series,
Vol. 37, UK, 1988.
16. Umeno T., Kaneko T., and Hori Y.: Robust servosystem design with two degrees of freedom and
its application to novel motion control of robot manipulators, IEEE Trans. Industr. Electronics
40(5) (1993), 473–485.
17. Caroll J. J., Dawson M., Jr.: Integrator backstepping techniques for the tracking control of
permanent magnet brush DC motors, IEEE Trans. Industry Appl. 31(2) (1995), 248–255.
18. Lygouras J. N., Dimitriadis C. M., Tsortanidis M. C., Bakos G. C., and Tsalides P. G.: Digital
ultrasonic scanning system for positioning underwater remotely operated vehicles, Internat. J.
Electronics 76(3) (1994), 541–550.